(a) Absorber unit cell (b) Unit cells array in WR 430

#### **2.2 Experimental results**

waveguide

(c) Experiment set up

Fig. 7. Photographs of the fabricated prototype absorber unit cell and unit cells array.

Fig. 8. Measured S-parameters of the planar arrayed (13 ⨉ 3) unit cells.

Fig. 9. Measured absorbance curve.

In order to verify a new type of the backplane-less absorber was designed without the resistive sheet. We fabricate a proto-type unit cell on a FR-4 substrate (εr = 4.4, tan δ = 0.025,

waveguide

Fig. 8. Measured S-parameters of the planar arrayed (13 ⨉ 3) unit cells.

In order to verify a new type of the backplane-less absorber was designed without the resistive sheet. We fabricate a proto-type unit cell on a FR-4 substrate (εr = 4.4, tan δ = 0.025,

Fig. 9. Measured absorbance curve.

Fig. 7. Photographs of the fabricated prototype absorber unit cell and unit cells array.

(c) Experiment set up

**2.2 Experimental results** 

(a) Absorber unit cell (b) Unit cells array in WR 430

and thickness t = 0.8 mm) using standard photolithography technique, and an experiment was carried out. The photographs of the fabricated two-layer metallization MTM absorber sample are shown in Fig. 7 (a). In order to verify the effectiveness of the double-negative MTM absorber cells without a metallic backing plate, a planar array of absorber unit cells (13 ⨉ 3) was mounted on a polystyrene foam substrate with a relative permittivity of 1.02. The inter-element spacing between two unit cells was set to 6 mm. As shown in Fig. 7 (b), this planar absorber structure was placed inside a rectangular waveguide (WR430), and measurements were taken with the waveguide over the frequency range from 1.6 GHz to 2.8 GHz. At these frequencies, the EM wave propagation was confined to the TE10 mode. The measured magnitudes of S11 and S21 parameters for the planar arrayed (13 ⨉ 3) unit cells are plotted in Fig 8. This results show that the backplane-less planar absorber structure is able to achieve the well matched broadband absorber. However, the experimental transmission minimum frequency of 2.43 GHz is shifted approximately 90 MHz lower compared to the simulated transmission minimum frequency. By using the measured S11 and S22 parameters, the calculated corresponding absorbance A (= 1- ∣S11∣2- ∣S21∣2) was plotted in Fig. 9. It exhibits a maximum absorbance of 90% at 2.34 GHz, with a full width at half maximum (FWHM) of 120 MHz over the frequency range from 2.29 to 2.41 GHz. This broader bandwidth performance is likely due to a mutual coupling effect between the cells. Although, the field distribution of the TE10 mode wave is different compared to the plane wave, the proposed planar absorber structure shows the possibility of the planar-type absorber without metallic backing plates or a resistive sheet.

A new idea for a backplane-less planar-type microwave absorber based on the arrayed double-negative MTM unit cells has been presented. By properly arrayed a single unit cell with a SRR and two OCSRRs configuration, has been shown to effectively absorb most of the impinging power. The two main advantages of the proposed absorber are: 1) the absence of a metallic back plate or a resistive sheet and 2) the reduced thickness (close to 0.06λ0) in the EM wave propagation direction. The total miniaturized MTM absorber unit cell size was 7.4 mm ⨉ 7 mm ⨉ 0.8 mm. The proposed layout can be easily extended to work for more compact and thin backplane-less planar absorber designs.

#### **3. Design of a metamaterial absorber with an enhanced operating band-width**

#### **3.1 Choice of the resonant inclusion on the edge-coupled SRRs**

We present the employment of the modified microwave meta-material absorber structure in order to achieve a further enhancement of the operating bandwidth. The SRRs can be used as electrically resonant particles which exhibit a strong resonant permittivity, and can also be used as the magnetic resonant particles which exhibit a strong resonant permeability. Fig. 10 shows the various types of the edge-coupled SRRs with different electromagnetic orientations. The electric field polarization is kept along the x-axis, and magnetic field polarization is kept along the y-axis. In principle, when the SRRs excited purely by magnetic field normal to the SRRs plane with electric field perpendicular to the gap (Fig. 10(b)), it exhibits a magnetic response only. SRRs excited by electric field parallel to the gap and the magnetic field lying completely in the SRRs plane (Fig. 10(d)), it exhibits an electric response only. Furthermore, the SRRs can exhibit both a purely magnetic and electric response to different electromagnetic orientation, as shown in Fig. 10(a). This results in a complicated bianisotropic EM behaviour. When the SRRs excited by in-plane magnetic field with electric field perpendicular to the gap (Fig. 10(c)), there is no magnetic or electric response.

Fig. 10. The various types of the edge-coupled SRRs with different electromagnetic orientations.

#### **3.2 Double negative unit cell design**

The two type of the proposed absorber unit cells consisted of distinct metallic elements are shown in Fig. 11 (type 1) and Fig. 12 (type 2). In the absorber unit cell type 1, the purely electric and magnetic responses were provided by the two symmetrically placed OCSRRs and an SRR, respectively, as shown in Fig 11. In the absorber unit cell type 2, the purely electric responses were provided by the two OCSRRs and the electromagnetic responses were provided by the 900 rotated SRR, as shown in Fig 12.

Fig. 11. Geometry of the absorber unit cell (type 1).

We created electromagnetic responses by combining the OCSRRs with an SRR in a parallel plane separated by a lossy dielectric substrate. The absorber unit cells are made of a FR-4 substrate whose relative dielectric constant is εr = 4.4, and loss angle tangent tan δ = 0.025, and thickness t = 1.8 mm. The metal for metallic patterns is a copper whose conductivity is σ = 5.8 ⨉ 107S/m. By changing the geometry and the separation between the OCSRRs and SRR, the electromagnetic responses are tuned to match the impedance to free space and minimize the transmission at the aimed design frequency. The geometrical dimensions are: a = 7.4 mm, b = 7 mm, w = 3.2 mm, h = 6.4 mm l = 4.6 mm, and g = 0.4 mm. Computer simulations for one unit cell are carried out using the commercial finite-difference time

Fig. 12. Geometry of the absorber unit cell (type 2).

Fig. 10. The various types of the edge-coupled SRRs with different electromagnetic

The two type of the proposed absorber unit cells consisted of distinct metallic elements are shown in Fig. 11 (type 1) and Fig. 12 (type 2). In the absorber unit cell type 1, the purely electric and magnetic responses were provided by the two symmetrically placed OCSRRs and an SRR, respectively, as shown in Fig 11. In the absorber unit cell type 2, the purely electric responses were provided by the two OCSRRs and the electromagnetic responses

(a) OCSRR (bottom view) (b) SRR (top view)

We created electromagnetic responses by combining the OCSRRs with an SRR in a parallel plane separated by a lossy dielectric substrate. The absorber unit cells are made of a FR-4 substrate whose relative dielectric constant is εr = 4.4, and loss angle tangent tan δ = 0.025, and thickness t = 1.8 mm. The metal for metallic patterns is a copper whose conductivity is σ = 5.8 ⨉ 107S/m. By changing the geometry and the separation between the OCSRRs and SRR, the electromagnetic responses are tuned to match the impedance to free space and minimize the transmission at the aimed design frequency. The geometrical dimensions are: a = 7.4 mm, b = 7 mm, w = 3.2 mm, h = 6.4 mm l = 4.6 mm, and g = 0.4 mm. Computer simulations for one unit cell are carried out using the commercial finite-difference time

were provided by the 900 rotated SRR, as shown in Fig 12.

Fig. 11. Geometry of the absorber unit cell (type 1).

**3.2 Double negative unit cell design** 

orientations.

domain solver Microwave Studio by CST. The scattering parameters of theses MTM unit cells were then simulated, and the absorbance was calculated using the equation A = 1- ∣S11∣2- ∣S21∣2. The calculated absorbance is plotted in Fig. 13. In case of type 1, the maximum absorbance peak is 93% at 2.47 GHz, there is a weak secondary absorbance peak 40% at 2.98 Hz. In case of type 2, the maximum absorbance peak is 84% at 2.6 GHz, there is a strong secondary absorbance peak 86% at 2.97 Hz. When three cells are arrayed in vertical direction with a configuration of type 2, it exhibits near same absorbance characteristics. In order to understand the nature of this absorbance, the simulated surface current densities in the top resonator structure of a SRR and the lower resonator structure of OCSRRs are shown in Fig. 13 and Fig. 14, respectively.

Fig. 13. Calculated absorbance of the absorber unit cells with a different configuration.

Fig. 14. Simulated surface current densities (type 1).

Fig. 15. Simulated surface current densities (type 2).

For the 2.47 GHz resonance in type 1, we observe that the counter-circulating currents flow on the SRR provide magnetic resonance, and the stronger current density takes place in the shorted-end of the left-side OCSRR, which provide electric resonance. In contrast, the 2.98 GHz resonance is determined by the magnetic response associated with a circulating current flowing on the SRR and electric resonance associated with the shorted-end of the right-side OCSRR. For the 2.6 GHz resonance in type 2, we observe that a dipole resonance currents and weak circulating flow on the SRR, and the stronger current density takes place in the shorted-end of the right-side OCSRR. The 2.97 GHz resonance is determined by the magnetic response associated with a circulating current flowing on the SRR and electric resonance associated with the shorted-end of the left -side OCSRR.

Fig. 16 shows the sketch of the planar arrayed (13 ⨉ 4) structures of MTM absorber unit cells of type 2. The unit cells are vertically arrayed without a gap length and the horizontally arrayed with a spacing lengths g = 6 mm. A vertically polarized transverse electromagnetic (TEM) wave impinges upon this structure with the incident angle from 00 to 450 and the scattering parameters were then simulated. The simulated S-parameters are plotted in Fig. 17. We observe that the planar arrayed unit cell shows the first and secondary absorbance peak at 2.93, and 3.27 GHz, respectively. Due to a mutual coupling effect between the cells, the two maximum absorbance frequencies are shifted to a higher frequency band.

Fig. 16. The planar arrayed (39 ⨉ 39) unit cells of type 2.

Fig. 17. The simulated S-parameters.

#### **3.3 Experimental results**

72 Micromachining Techniques for Fabrication of Micro and Nano Structures

(a) 2.47 GHz (b) 2.98 GHz

(a) 2.6 GHz (b) 2.97 GHz

resonance associated with the shorted-end of the left -side OCSRR.

For the 2.47 GHz resonance in type 1, we observe that the counter-circulating currents flow on the SRR provide magnetic resonance, and the stronger current density takes place in the shorted-end of the left-side OCSRR, which provide electric resonance. In contrast, the 2.98 GHz resonance is determined by the magnetic response associated with a circulating current flowing on the SRR and electric resonance associated with the shorted-end of the right-side OCSRR. For the 2.6 GHz resonance in type 2, we observe that a dipole resonance currents and weak circulating flow on the SRR, and the stronger current density takes place in the shorted-end of the right-side OCSRR. The 2.97 GHz resonance is determined by the magnetic response associated with a circulating current flowing on the SRR and electric

Fig. 16 shows the sketch of the planar arrayed (13 ⨉ 4) structures of MTM absorber unit cells of type 2. The unit cells are vertically arrayed without a gap length and the horizontally arrayed with a spacing lengths g = 6 mm. A vertically polarized transverse electromagnetic (TEM) wave impinges upon this structure with the incident angle from 00 to 450 and the scattering parameters were then simulated. The simulated S-parameters are plotted in Fig. 17. We observe that the planar arrayed unit cell shows the first and secondary absorbance peak at 2.93, and 3.27 GHz, respectively. Due to a mutual coupling effect between the cells,

the two maximum absorbance frequencies are shifted to a higher frequency band.

Fig. 14. Simulated surface current densities (type 1).

Fig. 15. Simulated surface current densities (type 2).

In order to verify a new type of the backplane-less absorber was designed without the resistive sheet. We fabricate a proto-type unit cell on a FR-4 substrate (εr = 4.4, tan δ = 0.025, and thickness t = 1.8 mm) using standard photolithography technique, and an experiment was carried out. The photographs of the fabricated two-layer metallization MTM absorber sample are shown in Fig. 18 (a). In order to verify the effectiveness of the double-negative MTM absorber cells without a metallic backing plate, a planar array of absorber unit cells (39 ⨉ 39) was mounted on acryl test-zig, as shown in Fig. 18(b). The inter-element spacing between two unit cells was set to 7 mm and the total size of the planar absorber was 300 m ⨉ 300 mm, and measurements were taken over the frequency range from 2.6 to 3.4 GHz. The calculating absorbance by using the measured magnitudes of S11 and S21 parameters for the planar arrayed absorber cells are plotted in Fig. 18 (c). With the increasing incidence angle from 00 to 450 the absorbance curves have little change and the two absorbance peaks remain more than 90% for different incident angles. For normal incident waves, the experimental data exhibits a maximum absorbance of 93% at 2.93 GHz, with a full width at half maximum (FWHM) of 250 MHz, and a secondary peak absorbance 87% at 3.26 GHz with a FWHM of 260 MHz. By properly arrayed a single unit cell with a 900 rotated SRR and two OCSRRs configuration, has been shown to effectively absorb most of the impinging power. The two main advantages of the proposed absorber are: 1) the absence of a metallic back plate or a resistive sheet and 2) the enhanced absorbance operating band by using electromagnetic responses of the 900 rotated SRR. The thickness of the planar MTM absorber structure in the EM wave propagation direction was shown 7.4 mm (≈ λ0) at 2.6 GHz. The proposed layout can be easily extended to work for more compact and thin backplane-less planar absorber designs.

Fig. 18. Photographs of the fabricated absorber and the calculated results of the absorbance.

#### **4. Design of a meta-material absorber applications**

#### **4.1 Three-dimensional resonator for EM waves absorption**

In recent years, as the use of mobile phones has become widespread, the EM waves shielding technology is a hot issue for the intelligent building construction. In this work, a three-dimensional resonators made of cubic arrangements of the OCSRRs and a metal strips is proposed. By installing these spatial band-stop cube resonators inside the concrete walls, the incident electromagnetic waves into the concrete walls can be blocked effectively over the specific frequency band. For the practical applications it is desirable to have a less sensitive to the incident angle and polarization of the impinging waves. However, it is very difficult to implement in practices due to its size and surrounding environments of the concrete. In order to solve this problem, two OCSRRs were used as basic elements and a metal strip conductor was connected between these OCSSRs. As a result, a threedimensional resonators (7 mm ⨉ 7 mm ⨉ 7 mm) which acts individually interacting with

from 00 to 450 the absorbance curves have little change and the two absorbance peaks remain more than 90% for different incident angles. For normal incident waves, the experimental data exhibits a maximum absorbance of 93% at 2.93 GHz, with a full width at half maximum (FWHM) of 250 MHz, and a secondary peak absorbance 87% at 3.26 GHz with a FWHM of 260 MHz. By properly arrayed a single unit cell with a 900 rotated SRR and two OCSRRs configuration, has been shown to effectively absorb most of the impinging power. The two main advantages of the proposed absorber are: 1) the absence of a metallic back plate or a resistive sheet and 2) the enhanced absorbance operating band by using electromagnetic responses of the 900 rotated SRR. The thickness of the planar MTM absorber structure in the EM wave propagation direction was shown 7.4 mm (≈ λ0) at 2.6 GHz. The proposed layout can be easily extended to work for more compact and thin backplane-less planar absorber designs.

(a) One period of the arrayed unit cells.

(b) A planar arrayed MTM absorber (c) Measured absorbance

**4. Design of a meta-material absorber applications 4.1 Three-dimensional resonator for EM waves absorption** 

Fig. 18. Photographs of the fabricated absorber and the calculated results of the absorbance.

In recent years, as the use of mobile phones has become widespread, the EM waves shielding technology is a hot issue for the intelligent building construction. In this work, a three-dimensional resonators made of cubic arrangements of the OCSRRs and a metal strips is proposed. By installing these spatial band-stop cube resonators inside the concrete walls, the incident electromagnetic waves into the concrete walls can be blocked effectively over the specific frequency band. For the practical applications it is desirable to have a less sensitive to the incident angle and polarization of the impinging waves. However, it is very difficult to implement in practices due to its size and surrounding environments of the concrete. In order to solve this problem, two OCSRRs were used as basic elements and a metal strip conductor was connected between these OCSSRs. As a result, a threedimensional resonators (7 mm ⨉ 7 mm ⨉ 7 mm) which acts individually interacting with

(c) Surface current distribution (d) Simulated S-parameters of one unit cell

Fig. 19. Geometry of one unit cell and the simulated results.

the incoming electromagnetic waves at 2 GHz frequency band. The geometry of one unit element and the simulated results are shown in Fig. 19. The geometrical dimensions are: a = 7 mm, b = 3.8 mm, c = 2 mm, d = 0.85 mm, and e = 0.45 mm. When the unit element is placed within a rectangular waveguide (109 mm ⨉ 54.5 mm ⨉ 150 mm), the simulated transmission coefficient through with/without a resonator is shown in Fig. 19 (b). The resonance frequency of the proposed resonator shows 2.08 GHz and the stop-band width (≤ -10dB) is 19 MHz. The resonance frequency can be adjusted by the variation of the capacitance and inductance, which depend on the physical length of the slot width, ring spacing, ring radius, and the height of metal strip. The unit cell of the three-dimensional resonator (7 mm ⨉ 7 mm ⨉7 mm) is shown in Fig. 20. On each side of the six sides of the cube structures with a metallic patterns of a resonator and a lossy dielectric material (εr = 6.0) is filled inside of the cube.

Fig. 20. Geometry of tree-dimensional resonator.

Fig. 21. Transmission coefficient variation for various rotated angle.

The simulated transmission coefficient variation of a resonator by the various rotated angles (θ) of a cube is shown in Fig. 21. The transmission coefficient is degenerated by the cross polarized effect and the performance of a resonator is determined by the direction of incident waves. The cubic resonator shows stop-band characteristics for a selected set of rotated angles of 0o, 45o and 90o, but its resonance frequency is changed due to anisotropic nature of the structures. The photographs of a fabricated three-dimensional resonator cubic resonator and a concrete block sample are shown in Fig. 22. The resonator structures were fabricated with a surface micromachining process on flexible polyimide substrate. A 200 μm-thick copper was film was e-beam evaporated on the polyimide substrate to form the ground plane. For the metallic patters, direct laser micro-machining technology was used, and another 10 μm-thick polyimide was coated on the top of the metallic patterns. A 'T' shaped metallic patterns for covering the surfaces of a cubic resonator and a photograph of the metallic-zig structure for the concrete blocks fabrication are shown in Fig. 23. In order to check the stop-band characteristics of the proposed resonator, three concrete block samples with a different cubic resonator arrangement were loaded into the jig and tested.

In a measurement system, a Wiltron 360B vector network analyzer was connected to the measurement jig through an R-band waveguide to coaxial connector. A WR-430 standard waveguide, ranging from about 1.7 GHz to about 2.6 GHz, was used as a measurement jig. The WR-430 standard waveguide had a size of 10.922 cm × 5.461cm. Fig. 24 shows the measured transmission coefficient through the waveguide loaded with the cubic resonators array structures. We observed that the stop band-width becomes wider in comparison to the response of one cubic resonator. When nine cubic resonators are embedded into a concrete block (109 mm ⨉ 54.5 mm ⨉ 150 mm), the resonance frequency is about 2.05GHz, which is lower than simulated frequency and the stop-band width (≤ −10*dB*) is 150 MHz. The proposed concrete block structure can be applied to the intelligent building construction for EM waves shielding.

Fig. 22. Photographs of the fabricated three-dimensional resonator and a concrete block sample.

76 Micromachining Techniques for Fabrication of Micro and Nano Structures

Fig. 20. Geometry of tree-dimensional resonator.

Fig. 21. Transmission coefficient variation for various rotated angle.

The simulated transmission coefficient variation of a resonator by the various rotated angles (θ) of a cube is shown in Fig. 21. The transmission coefficient is degenerated by the cross polarized effect and the performance of a resonator is determined by the direction of incident waves. The cubic resonator shows stop-band characteristics for a selected set of rotated angles of 0o, 45o and 90o, but its resonance frequency is changed due to anisotropic nature of the structures. The photographs of a fabricated three-dimensional resonator cubic resonator and a concrete block sample are shown in Fig. 22. The resonator structures were fabricated with a surface micromachining process on flexible polyimide substrate. A 200 μm-thick copper was film was e-beam evaporated on the polyimide substrate to form the ground plane. For the metallic patters, direct laser micro-machining technology was used, and another 10 μm-thick polyimide was coated on the top of the metallic patterns. A 'T' shaped metallic patterns for covering the surfaces of a cubic resonator and a photograph of the metallic-zig structure for the concrete blocks fabrication are shown in Fig. 23. In order to check the stop-band characteristics of the proposed resonator, three concrete block samples

with a different cubic resonator arrangement were loaded into the jig and tested.

In a measurement system, a Wiltron 360B vector network analyzer was connected to the measurement jig through an R-band waveguide to coaxial connector. A WR-430 standard waveguide, ranging from about 1.7 GHz to about 2.6 GHz, was used as a measurement jig. The WR-430 standard waveguide had a size of 10.922 cm × 5.461cm. Fig. 24 shows the

(a) 'T'-shape film pattern (b) zig structure for the concrete blocks fabrication

Fig. 23. Geometry of **'**T'-shape film pattern and zig structure for fabricating the concrete blocks with cubic resonators.

Fig. 24. Measured transmission coefficient for the different cubic resonators array.

#### **4.2 Isolation improvement of two closely spaced loop antennas using double negative MTM absorber unit cells**

A standard method in improving isolation is to limit the radiation in the propagating direction and to increase the distance between the antennas. However, this method is not highly efficient because the area allocated to the antennas in mobile device is very small. In order to improve the isolation performance several studies have been done on spatial diversity and MIMO systems. A multiple slits etched on a single ground plane were used to isolate the feeding port of two closely-packed antennas. Another technology is to link the two antennas with a so-called neutralization line in order to increase the port-to-port isolation. However, the aforementioned techniques are only suitable for mobile diversity and MIMO antennas with a common ground plane. This study presents a new method to improve the isolation between two antennas without a shared ground plane. In order to increase the isolation between the two loop antennas, double negative MTM absorber unit cells were used. The MTMs are artificial composite structures that exhibit a homogeneous effective permittivity ε and permeability μ, which become negative over an operating frequency range. Due to the unique characteristics of the MTMs and their integration, a number of the potential MTM applications have been researched in many small antenna applications. A MTM absorber consists of a structure with negative permittivity and another structure with negative permeability, resulting in a negative refractive index. We use a double-layered structure with a SRR and two OCSRRs which are put on top of each other in order to make a miniaturized thin MTM absorber unit cell for the 2 GHz frequency band.

Fig. 25. Geometry of the proposed loop antenna.

**4.2 Isolation improvement of two closely spaced loop antennas using double negative** 

A standard method in improving isolation is to limit the radiation in the propagating direction and to increase the distance between the antennas. However, this method is not highly efficient because the area allocated to the antennas in mobile device is very small. In order to improve the isolation performance several studies have been done on spatial diversity and MIMO systems. A multiple slits etched on a single ground plane were used to isolate the feeding port of two closely-packed antennas. Another technology is to link the two antennas with a so-called neutralization line in order to increase the port-to-port isolation. However, the aforementioned techniques are only suitable for mobile diversity and MIMO antennas with a common ground plane. This study presents a new method to improve the isolation between two antennas without a shared ground plane. In order to increase the isolation between the two loop antennas, double negative MTM absorber unit cells were used. The MTMs are artificial composite structures that exhibit a homogeneous effective permittivity ε and permeability μ, which become negative over an operating frequency range. Due to the unique characteristics of the MTMs and their integration, a number of the potential MTM applications have been researched in many small antenna applications. A MTM absorber consists of a structure with negative permittivity and another structure with negative permeability, resulting in a negative refractive index. We use a double-layered structure with a SRR and two OCSRRs which are put on top of each other in order to make a miniaturized thin MTM absorber unit cell for the 2 GHz frequency band.

**MTM absorber unit cells** 

Fig. 25. Geometry of the proposed loop antenna.

The geometry of the proposed loop antenna is shown in Fig. 25. It was fed by a 50Ω coaxial connector and designed on a FR-4 dielectric substrate (relative permittivity = 4.6, thickness = 2 mm). The proposed antenna consists of a small square loop and a SRR shaped loop. The single loop antenna occupied 21 mm ⨉ 23.6 mm ⨉ 2 mm; the total length of the inner loop was 54 mm long (≈ 0.5 λg at 2.5 GHz). The input impedance of an electrically small loop antenna has a large inductive component. In order to match the input impedance, it needs a capacitive element. A small capacitive split ring resonator (SRR) structure was used as a matching element in this design. The two vertical loop arms have the same currents with a null in middle of the top loop section. The currents are seen to be equivalent to that of a pair of parallel monopole antennas driven in-phase with a spacing of approximately 1/8 λg. The geometry of the proposed MTM absorber unit cell is shown in Fig. 27. Two OCSRRs and a SRR structure were placed on both sides of the FR-4 dielectric substrate (relative permittivity = 4.6, thickness = 2 mm). Two OCSRRs are located with even symmetry on one side of the FR-4 substrate. We constructed an MTM unit cell using an OCSRR and SRR arrangement. The proposed absorber unit cell based on the double negative MTM structure was shown to reach a thickness on the order of 0.059 λ0 at the frequency of 2.54 GHz. The frequency characteristics of the unit cell structure were simulated by using a periodic boundary condition (PBC) method; the dimensions of the unit cell were 7.4 mm × 7 mm × 2 mm. The simulated scattering parameters for the proposed unit cell are plotted in Fig. 28. The performance of the proposed absorber shows a maximum absorbance of 93% at 2.54 GHz, and the full width at half maximum absorbance (≥ 50%) was at 200 MHz.

Fig. 26. The simulated surface current distributions.

Fig. 27. Geometry of the proposed MTM absorber unit cell.

Fig. 28. Simulated S-parameters for the proposed MTM absorber cell.

The geometry of the two proposed loop antennas with the three MTM absorber cells is shown in Fig. 29. It consists of two loop antennas and three MTM absorber cells; the total antenna size was 51 mm ⨉ 23.6 mm ⨉ 2 mm. When the inter-element spacing between two loop antennas is set to 13.5 mm (≈ λg/4), the maximum isolation of -23 dB between the two antennas was obtained at 2.54 GHz. When three cells were placed between the antennas, the resonant frequency changed slightly, but the port-to-port isolation ∣S21∣ is efficiently improved by about 10 dB within the operating bandwidth. The simulated magnetic field distribution for the antenna with the MTM absorber cells is shown in Fig. 30. When a signal with a frequency of 2.54 GHz was applied to port 1, most of the magnetic fields are induced onto the two vertical loop's arms near port 1. When the MTM absorber cells were placed in

(a) SRR (top view) (b) OCSRR (bottom view)

Fig. 27. Geometry of the proposed MTM absorber unit cell.

Fig. 28. Simulated S-parameters for the proposed MTM absorber cell.

The geometry of the two proposed loop antennas with the three MTM absorber cells is shown in Fig. 29. It consists of two loop antennas and three MTM absorber cells; the total antenna size was 51 mm ⨉ 23.6 mm ⨉ 2 mm. When the inter-element spacing between two loop antennas is set to 13.5 mm (≈ λg/4), the maximum isolation of -23 dB between the two antennas was obtained at 2.54 GHz. When three cells were placed between the antennas, the resonant frequency changed slightly, but the port-to-port isolation ∣S21∣ is efficiently improved by about 10 dB within the operating bandwidth. The simulated magnetic field distribution for the antenna with the MTM absorber cells is shown in Fig. 30. When a signal with a frequency of 2.54 GHz was applied to port 1, most of the magnetic fields are induced onto the two vertical loop's arms near port 1. When the MTM absorber cells were placed in the orthogonal direction against the magnetic fields, most of the magnetic fields were not coupled to the loop antenna at port 2. As a result, the port-to-port isolation between the two loop antennas was improved by using the MTM absorber cells between the two antennas. In order to verify the simulations, two loop antennas with three double negative MTM absorber cells were fabricated on a FR-4 substrate with the relative dielectric constant 4.6 and a thickness of 2 mm; two 50Ω SMA connecters were installed.

Fig. 29. Geometry of the proposed antenna with three absorber cells.

Fig. 30. The simulated magnetic field distributions of the antenna when a signal was fed to port 1(f = 2.54GHz).

(a) Antenna with cells (b) Antenna without cells

Fig. 31. Photographs of the fabricated antenna

Fig. 32. Comparison of the measured return loss and isolation for the proposed antenna with/without cells.

The photographs of the prototype fabricated antennas are shown in Fig. 31. Comparisons of the measured return loss and isolation results for the proposed antennas are shown in Fig. 32. The measured S-parameters were similar to the simulated results. When three absorber cells were placed between the antennas, the isolation ∣S21∣ shows a 10 dB improvement when compared to the isolation of the antennas without the absorber cells. A maximum isolation of -24 dB was achieved at 2.54 GHz. The operating bandwidth of the proposed antenna was slightly lower due to the absorber cells between the two loop antennas.

A new method to improve isolation between two closely spaced small antennas with ground plane is presented. The proposed antenna structures are composed of two small loop antennas and three double negative MTM absorber unit cells exhibiting negative permittivity and permeability at the aimed frequency of 2.54 GHz. The proposed absorber cell consists of a double-layered structure with a SRR and two OCSRRs. They are put on top of each other has shown to effectively absorb most of the impinging power. Both of them have same resonance frequency of 2.54 GHz. In order to obtain high isolation between the two closely spaced antennas, the proposed absorber cells can be effectively used. This technique can be applied for isolation improvement in multi antenna systems.

#### **5. Conclusions**

Much of the work in MTM has been focused on the real parts of the permittivity and permeability to enable the creation of a NRI material for various applications including electromagnetic wave cloaking, perfect lenses, and microwave components. By varying the dimensions of the electric and magnetic components, it is possible to adjust the permittivity and permeability independently. Additionally, by tuning the electric and magnetic resonances, an MTM can be impedance-matched to free space. As a result, 100 % absorbance is theoretically possible. By properly arrayed a single unit cell, it is possible to make the absorber's electrical size small enough. Since from the practical point of view, it would be rather difficult to employ a photo-etching technology for the fabrication of the absorber cells, operated at the higher microwave frequency and terahertz frequency regime.

In this work, the authors present the design and fabrication of the two-dimensional and three-dimensional microwave meta-material absorber unit cells by using a standard photolithography technique and a micro-machining technology, respectively. Moreover, a new type MTM absorber structure was proposed. By properly arrayed a single unit cell with a 900 rotated SRR and two OCSRRs configuration, has been shown to effectively absorb most of the impinging power. The two main advantages of the proposed absorber are: 1) the absence of a metallic back plate or a resistive sheet and 2) the enhanced absorbance operating band by using electromagnetic responses of the 900 rotated SRR. The thickness of the planar MTM absorber structure in the EM wave propagation direction was shown 7.4 mm (≈ λ0) at 2.6 GHz. The proposed layout can be easily extended to work for more compact and thin backplane-less planar absorber designs.

#### **6. Acknowledgment**

This research was supported Basic Science Research Program through National Research Foundation of Korea (NRF) funded by the Ministry of Education, Science and Technology (No. 2010-0011646)."

#### **7. References**

82 Micromachining Techniques for Fabrication of Micro and Nano Structures

Fig. 32. Comparison of the measured return loss and isolation for the proposed antenna

slightly lower due to the absorber cells between the two loop antennas.

technique can be applied for isolation improvement in multi antenna systems.

cells, operated at the higher microwave frequency and terahertz frequency regime.

The photographs of the prototype fabricated antennas are shown in Fig. 31. Comparisons of the measured return loss and isolation results for the proposed antennas are shown in Fig. 32. The measured S-parameters were similar to the simulated results. When three absorber cells were placed between the antennas, the isolation ∣S21∣ shows a 10 dB improvement when compared to the isolation of the antennas without the absorber cells. A maximum isolation of -24 dB was achieved at 2.54 GHz. The operating bandwidth of the proposed antenna was

A new method to improve isolation between two closely spaced small antennas with ground plane is presented. The proposed antenna structures are composed of two small loop antennas and three double negative MTM absorber unit cells exhibiting negative permittivity and permeability at the aimed frequency of 2.54 GHz. The proposed absorber cell consists of a double-layered structure with a SRR and two OCSRRs. They are put on top of each other has shown to effectively absorb most of the impinging power. Both of them have same resonance frequency of 2.54 GHz. In order to obtain high isolation between the two closely spaced antennas, the proposed absorber cells can be effectively used. This

Much of the work in MTM has been focused on the real parts of the permittivity and permeability to enable the creation of a NRI material for various applications including electromagnetic wave cloaking, perfect lenses, and microwave components. By varying the dimensions of the electric and magnetic components, it is possible to adjust the permittivity and permeability independently. Additionally, by tuning the electric and magnetic resonances, an MTM can be impedance-matched to free space. As a result, 100 % absorbance is theoretically possible. By properly arrayed a single unit cell, it is possible to make the absorber's electrical size small enough. Since from the practical point of view, it would be rather difficult to employ a photo-etching technology for the fabrication of the absorber

with/without cells.

**5. Conclusions** 


## **Laser Micromachining and Micro-Patterning with a Nanosecond UV Laser**

Xianghua Wang1, Giuseppe Yickhong Mak2 and Hoi Wai Choi2

*1Key Lab of Special Display Technology, Ministry of Education, National Engineering Lab of Special Display Technology, National Key Lab of Advanced Display Technology, Academy of Photoelectric Technology, Hefei University of Technology, Hefei, 2Department of Electrical and Electronic Engineering, The University of Hong Kong, 1China 2Hong Kong* 

#### **1. Introduction**

84 Micromachining Techniques for Fabrication of Micro and Nano Structures

R. A. Depine and A. Lakhtakia (2004). A new condition to identify isotropic dielectric-

J. Lee and S. Lim (2011). Bandwidth enhanced and polarization-insensitive metamaterial

M. A. Jensen J. W. Wallace (2004). A review of antennas and propagation for MIMO wireless communications, *IEEE Trans. Antennas Propagation,* vol. 52, pp. 2810-2824. C. Chiu, C. Chen, D. Murch, and R. Rowell (2007). Reduction of mutual coupling between

Y. Shin and S. Park (2007). Spatial diversity antenna for WLAN application, *Microwave* 

A. Diallo, C. Luxey, P. Le Thuc, and G. Kossiavas (2008). Enhanced two-antenna structures

A. Chebihi, C. Luxey, A. Diallo, P. Le Thuc, and R. Staraj (2008). A noval isolation technique

N. Engheta and R. W. Ziolkowski (2006). Metamaterials: physics and engineering

F. Falcone, T. Lopetegi, J. D. Baena, R. Marques, and M. Sorolla (2004). Effective negtive-ε

D. R. Smith, D. C. Vier, T. Koschny, and C. M. Soukoulis (2005). Electromgnetic parameter

*Microwave & Wireless Component Lett.,* vol. 14, pp. 280-282.

absorber using double resonance, *Electron Lett.,* vol. **47**, pp. 8-9.

*Technology Lett.,*vol. 41*,* pp. 315-316.

*Optical Technology Lett.,* vol. 49, pp. 1290-1294.

*Antennas and Propagation, IET*,vol. 2, pp. 93-101.

explorations, Wiley-Inter-sciencce, NJ.

1732-1738.

vol. 7, pp. 665-668.

10.

magnetic materials displaying negative phase velocity, *Microwave Optical* 

closely-packed antennas elements, *IEEE Trans. Antennas Propagation,* vol. 55, pp.

for universal mobile telecommunication system diversity terminals, *Microwaves,* 

for closely spaced PIFAs UMTS mobile phones, *IEEE Trans. Antennas Propagation,*

stop-band microstrip lines based on complementary split ring resonators, *IEEE* 

retrieval from inhomogeneous metamaterials, *Physics Review*, vol. E71, pp. 0316171-

Laser micromachining specifically refers to drilling and cutting with intensive laser beam usually in the form of pulse trains with energy far exceeding the ablation threshold of the target material. The ablation threshold, however, is dependent on material properties as well as the interacting laser characteristics such as laser wavelength and pulse width. Therefore, laser processes are flexible and can be highly selective. Laser wavelengths have been available in a wide spectrum range from the far infrared to the ultraviolet using a variety of medium material and frequency mixing techniques. Since late 1990s, ultra short laser beams (Gattass and Mazur 2008; Liu, et al. 1997; Molian, et al. 2009; Schaffer, et al. 2001; Stuart, et al. 1996; Varel, et al. 1997) from picosecond to femtosecond are intensively studied as to their interaction mechanism with transparent materials and applications in material processing. Although ultra short pulses within tens of picoseconds can be achieved by advanced techniques like Q-switching and mode-locking, the pulse energy is reduced by several orders of magnitude and the laser system is relatively bulky. Nanosecond lasers with intermediate pulse width are still predominant in industry due to their excellent flexibility and cost effectiveness. This chapter presents the work in laboratory on laser micromachining and micro-patterning of GaN/sapphire LED wafer with nanosecond UV laser for the purpose of chip-shaping and device isolation respectively.

#### **2. Laser micromachining**

#### **2.1 Laser-matter interaction**

Interaction of material with laser pulses generally involves the following fundamental processes that take place on different time scales after a pulse incidence (Zweig 1991). Firstly, light absorption takes place at a femtosecond time scale accompanied by ejection of excited electrons. Secondly, the excited electrons transfer their energy to the lattice in several tens of picoseconds through electron-phonon collision and induce melted zone in the bulk material. Following this, melted material evaporates on a nanosecond time scale and expands into the air in the form of atomic-sized particles in the ablation plume. Melt expulsion may also occur, termed hydrodynamic ablation, due to the recoil pressure of plume and result in large particles or clusters following the plume.

The specific ablation mechanism in laser micromachining and the resulting morphology of the ablated region such as aspect ratio and sidewall roughness varies with material property, laser wavelength, power intensity as well as pulse duration. The process of multiphoton excitation of electrons and collisional ionization is reported to play vital roles in semiconductor and dielectric laser ablation with a sharp threshold of power intensity observed. This mechanism become dominant when ultra-short laser pulses are employed for ablation, whereby a much increased portion of laser energy is transferred to electrons compared to the portion further transferred to the lattice (Stuart, et al. 1996). Under high intensity picosecond-pulsed laser fluence, the laser irradiated portion of n-type SiC material can be directly transformed into plasma on a few picoseconds time scale trough Coulomb explosion resulting in clean surfaces of the cleaves that outperform the results obtained with thermal ablation (Molian, et al. 2009).

#### **2.2 Optical setup of laser micromachining**

The laser micromachining system consists of a UV laser source, beam focusing optics and an x-y motorized translation stage. The laser source used in this experiment is a third harmonic neodymium-doped yttrium lithium fluoride (ND:YLF) diode-pumped solid-state (DPSS) laser manufactured by Spectra Physics. Being an actively Q-switched UV laser at 349 nm, the pulse repetition rate ranges from single pulse to 5 kHz. At the reference diode current of 3.2A, the pulse energy is 120 µJ with a pulse width of around 4 nanoseconds at 1 kHz repetition rate; however, this corresponds to very high intensity up to the order 1010 W/cm2 when focused to a 10-µm-diameter spot.

Fig. 1. Pulse energy versus diode current at different pulse repetition rates

excited electrons. Secondly, the excited electrons transfer their energy to the lattice in several tens of picoseconds through electron-phonon collision and induce melted zone in the bulk material. Following this, melted material evaporates on a nanosecond time scale and expands into the air in the form of atomic-sized particles in the ablation plume. Melt expulsion may also occur, termed hydrodynamic ablation, due to the recoil pressure of

The specific ablation mechanism in laser micromachining and the resulting morphology of the ablated region such as aspect ratio and sidewall roughness varies with material property, laser wavelength, power intensity as well as pulse duration. The process of multiphoton excitation of electrons and collisional ionization is reported to play vital roles in semiconductor and dielectric laser ablation with a sharp threshold of power intensity observed. This mechanism become dominant when ultra-short laser pulses are employed for ablation, whereby a much increased portion of laser energy is transferred to electrons compared to the portion further transferred to the lattice (Stuart, et al. 1996). Under high intensity picosecond-pulsed laser fluence, the laser irradiated portion of n-type SiC material can be directly transformed into plasma on a few picoseconds time scale trough Coulomb explosion resulting in clean surfaces of the cleaves that outperform the results obtained with

The laser micromachining system consists of a UV laser source, beam focusing optics and an x-y motorized translation stage. The laser source used in this experiment is a third harmonic neodymium-doped yttrium lithium fluoride (ND:YLF) diode-pumped solid-state (DPSS) laser manufactured by Spectra Physics. Being an actively Q-switched UV laser at 349 nm, the pulse repetition rate ranges from single pulse to 5 kHz. At the reference diode current of 3.2A, the pulse energy is 120 µJ with a pulse width of around 4 nanoseconds at 1 kHz repetition rate; however, this corresponds to very high intensity up to the order 1010 W/cm2

1.0 1.2 1.4 1.6 1.8 2.0 2.2 2.4 2.6 2.8 3.0

Current (A)

plume and result in large particles or clusters following the plume.

 100 Hz 500 Hz 1000 Hz 2000 Hz 5000 Hz

Fig. 1. Pulse energy versus diode current at different pulse repetition rates

thermal ablation (Molian, et al. 2009).

**2.2 Optical setup of laser micromachining** 

when focused to a 10-µm-diameter spot.

Pulse Energy (

J)

The beam intensity distribution is a Gaussian profile. There is a compromise between the pulse energy and the pulse repetition rate. According to the plots of pulse energy versus diode current recorded at a series of repetition frequencies, the pulse energy at lower repetition frequency goes higher until the repetition frequency is lower than 500 Hz as indicated in Figure 1.

As a smaller focus spot, which is diffraction limited, is desired for micro-scale machining, laser of shorter wavelength is preferred. The TEM00 beam in our laser source also allows for tight focusing, offering high spatial resolution. After beam expansion and collimation through a beam expander, the laser beam is reflected 90 by a dielectric laser line mirror and focused onto the horizontal machining plane to a very tiny spot several micrometers in diameter with a focusing triplet. The UV objective lens well suits the Gaussian beam profile of the laser. All optics are made of UV fused silica and anti-reflection coated. The size of the beam at the focal point is not only limited by the capability of the UV objective lens but is also sensitive to the coaxiality of the optics. Schematic diagram of the laser micromachining setup is drawn in Figure 2. A pair of aluminium mirrors is used to adjust the direction of the collimated laser beam to be coaxial with the UV objective lens.

Fig. 2. Schematic of laser micromachining setup

The additional feature of the set-up for tilted cutting, as illustrated in Figure 3, is the insertion of a UV mirror at an oblique angle within the optical path between the focusing optics and the machining plane, which serves to deflect the convergent beam to strike the sample at an oblique angle with respect to the horizontal working plane. With this modified set-up, it is relatively easy to optimize and monitor the beam through the tube lens imaged with a CCD camera. Once the optical setup is optimized before insertion of the tilting mirror, the mirror can be inserted without affecting the coaxiality of the laser beam.

The beam can be effectively applied for micro-sectioning with non-vertical sidewall profiles. The angle of incidence of the deflected laser beam on the wafer is 2, where as indicated in Figure 3 is the angle between the plane of the mirror and the normal. This angle is readily and precisely controlled by mounting the mirror onto a rotation stage; thus, the incident angle can be varied over a wide range. In this experiment, we have used a UV objective with a focal length of 75 mm, based on two considerations. Firstly, the focal length should be long enough to accommodate the mirror in the optical path. Secondly, an ideal tool for the fabrication of microstructures should have a very long penetration depth and negligible lateral dispersion. Nevertheless, an objective lens with a longer focal length also produces a larger focused beam spot. The diameter of the focused beam spot, *d*, is determined by the following equation:

$$d = \frac{4\lambda M^2 f}{\pi D} \tag{1}$$

where *M*2 is the beam quality factor, the wavelength of the laser beam, *f* the focal length and *D* the diameter of the incident beam.

Fig. 3. Schematic of the setup for tilted-cutting with a UV mirror inserted within the optical path

#### **2.3 Laser cutting of sapphire wafer**

88 Micromachining Techniques for Fabrication of Micro and Nano Structures

set-up, it is relatively easy to optimize and monitor the beam through the tube lens imaged with a CCD camera. Once the optical setup is optimized before insertion of the tilting

The beam can be effectively applied for micro-sectioning with non-vertical sidewall profiles.

Figure 3 is the angle between the plane of the mirror and the normal. This angle is readily and precisely controlled by mounting the mirror onto a rotation stage; thus, the incident angle can be varied over a wide range. In this experiment, we have used a UV objective with a focal length of 75 mm, based on two considerations. Firstly, the focal length should be long enough to accommodate the mirror in the optical path. Secondly, an ideal tool for the fabrication of microstructures should have a very long penetration depth and negligible lateral dispersion. Nevertheless, an objective lens with a longer focal length also produces a larger focused beam spot. The diameter of the focused beam spot, *d*, is determined by the

> <sup>2</sup> <sup>4</sup> *<sup>M</sup> <sup>f</sup> <sup>d</sup> D*

Laser line mirror

UV objective lens

Wafer

, where

(1)

the wavelength of the laser beam, *f* the focal length

as indicated in

mirror, the mirror can be inserted without affecting the coaxiality of the laser beam.

The angle of incidence of the deflected laser beam on the wafer is 2

following equation:

path

where *M*2 is the beam quality factor,

and *D* the diameter of the incident beam.

Tilting mirror

XY translation stage

Fig. 3. Schematic of the setup for tilted-cutting with a UV mirror inserted within the optical

Rotary

As our scheme of micromachining is targeted at die separation of GaN-based LEDs, sapphire wafers are used for testing the results, as it is the typical substrate for the metalorganic chemical vapour deposition (MOCVD) growth of GaN. The quality of the cleave can be quantified by the width, depth, linearity and sidewall roughness of the trench formed by the laser beam. Each of these parameters will be investigated. Since the focal length of the focusing lens (*f* = 75 mm) is much longer than the thickness of the sapphire wafer (*t* = 420 µm), the depth of the trench mainly depends on the number of micromachining cycles. The number of cycles is controlled by configuring the translation stage to repeat its linear path over a number of times. Since the position repeatability of the stage is better than 5 µm, increasing the number of cycles should not contribute significantly to the width of the feature. Figure 4 shows the cross-sectional optical image of a 420 µm thick sapphire wafer that has been micro-machined with an incident beam inclined at 45, with scan cycles ranging from 1 to 10. These incisions were carried out by setting the laser pulse energy to 54 µJ at a repetition rate of 2 kHz. The relationship between the inclined cutting depth and the number of passes of the beam are plotted in Figure 5. After the first pass of the beam, a narrow trench with a width of ~20 µm and a depth of ~220 µm was formed. Successive scans of the beam along the trench results in further deepening and widening, but the extent was increasing less. The depth of the trench depends on the effective penetration of the beam. From the second scan onwards, the beam has to pass through the narrow gap before reaching the bottom of the trench for further machining. The energy available at this point is attenuated, partly due to lateral machining of the channel (causing undesirable widening), absorption and diffraction effects. Therefore, the depth of the trench tends to saturate after multiple scans.

Fig. 4. Cross-sectional optical microphotograph of laser micro-machined micro trenches at an inclination angle of ~45 at a range of scan cycles of between 1 and 10 (left to right then down). (with permission for reproduction from American Institute of Physics)

Fig. 5. Depth of tilting micro-trenches as a function of scan cycles. (with permission for reproduction from American Institute of Physics Publishing)

After the chemical treatment, the surface morphologies of the micromachined samples are examined with atomic force microscopy (AFM). 3D images of the AFM scans are shown in Figure 6. The surface topography of the sapphire surface after 2 machining cycles exhibits a uniform roughness with an RMS value of ~150 nm. With more cycles, increasing densities and dimensions of granules are observed on the AFM image and the RMS roughness increased to ~218 nm after 5 cycles. The formation of the larger grains on the surface is a result of uneven aggregation and re-solidification of the melted material. The evacuation rate of the ablated species declines as the beam reaches deeper into the trench, and statistically the density of aggregation is more pronounced at these deeper sites.

Fig. 6. AFM morphology images of the inclined sapphire surfaces after laser micromachining for (a) 2 cycles and (b) 5 cycles, the corresponding RMS roughness are 150 nm and 218 nm respectively. (with permission for reproduction from American Institute of Physics)

Fig. 5. Depth of tilting micro-trenches as a function of scan cycles. (with permission for

statistically the density of aggregation is more pronounced at these deeper sites.

Fig. 6. AFM morphology images of the inclined sapphire surfaces after laser

Physics)

micromachining for (a) 2 cycles and (b) 5 cycles, the corresponding RMS roughness are 150 nm and 218 nm respectively. (with permission for reproduction from American Institute of

After the chemical treatment, the surface morphologies of the micromachined samples are examined with atomic force microscopy (AFM). 3D images of the AFM scans are shown in Figure 6. The surface topography of the sapphire surface after 2 machining cycles exhibits a uniform roughness with an RMS value of ~150 nm. With more cycles, increasing densities and dimensions of granules are observed on the AFM image and the RMS roughness increased to ~218 nm after 5 cycles. The formation of the larger grains on the surface is a result of uneven aggregation and re-solidification of the melted material. The evacuation rate of the ablated species declines as the beam reaches deeper into the trench, and

reproduction from American Institute of Physics Publishing)

#### **2.4 Front-side Laser micromachining of GaN/sapphire LED wafer**

In our laser micromachining setup, using 349 nm wavelength, a front-side machining scheme is employed to avoid damage to the active-layer as well as to achieve higher precision with beam alignment. A tilted incision with 60 (=30) tilting angle on the front side of the GaN/sapphire wafer is ablated after 5 successive scan cycles. The surface of the sidewall is exposed after laser micromachining for FE-SEM examination as shown in Figure 7. With the laser beam tightly focused, the kerf exposed at GaN layer shows a clear brim and the thickness of GaN estimated from the image is 4.5 µm. It is interesting to see a sharp interface between the sapphire substrate and the GaN layer and no heat affected zone (HAZ) is observed in the GaN layer after front-side machining. This finding may be attributed to the relatively low ablation threshold of GaN as it absorbs the 349 nm laser power. According to Figure 7, the sapphire substrate melts on the surface, while no melt is observed for the GaN layer. It is estimated that the surface temperature lies between the melting point of sapphire and GaN, which is in the range from 2040 °C to 2500 °C.

Fig. 7. FE-SEM image of a GaN/sapphire wafer after laser micromachining, the interface of GaN and sapphire and the brim of the 4.5 µm thick GaN layer is clear.

For comparison, surface morphology of backside micromachined LED wafer is illustrated in Figure 8 showing the feature of rugged sidewalls. The two images corresponds to a single scan of the laser beam at 50 µm/sec motion speed, with 30 µJ and 50 µJ pulse energy respectively, repeated at 1 kHz. This feature can be observed at varied pulse energies and scan cycles. With a high ablation threshold and optical transparency at the wavelength of 349 nm, sapphire is ablated with inferior surface quality. A large quantity of clusters is trapped within the groove which blocks light extraction from the sidewalls and also prevents heat dissipation via the sidewall surfaces. Improved quality of sapphire micromachining is possible by using a shorter wavelength or ultrashort pulse duration of the laser to suppress thermal effect during sapphire ablation. Laser ablation of the sapphire substrate with an absorptive wavelength to sapphire also avoids damaging on the epitaxial nitride layers.

Separation of some specially shaped LED such as a circular device after laser micromachining may be difficult if the wafer is cut insufficient in depth. The chips to be separated after machining are subject to uncontrollable fracture and crack whilst applying stress to the incision. In order to shape circular LEDs, the machining has to penetrate through the wafer to ensure separation in good shape. Although the penetration depth depends on laser power, it is important to adjust the focus position in the z direction in order to optimize the process condition.

Fig. 8. (a) SEM image of laser scribed lanes on the back-side sapphire substrate with 30 µJ pulses; (b) with 50 µJ pulses

Fig. 9. (a) Trends of surface ablation width, penetration depth and the aspect ratio with changing focusing levels during micromachining; (b) Relative position of the focus point with reference to the wafer; (c) The width of surface damage determined from front view of wafer, and penetration depth estimated from the back view (mirrored) after laser micromachining at 40 tilting angle. (with permission for reproduction from John Wiley and Sons)

depends on laser power, it is important to adjust the focus position in the z direction in

Fig. 8. (a) SEM image of laser scribed lanes on the back-side sapphire substrate with 30 µJ

Fig. 9. (a) Trends of surface ablation width, penetration depth and the aspect ratio with changing focusing levels during micromachining; (b) Relative position of the focus point with reference to the wafer; (c) The width of surface damage determined from front view of

wafer, and penetration depth estimated from the back view (mirrored) after laser micromachining at 40 tilting angle. (with permission for reproduction from John Wiley

order to optimize the process condition.

pulses; (b) with 50 µJ pulses

and Sons)

(a) (b)

Figure 9 shows the surface ablation widths and penetration depths of the incisions machined at 40 tilting angle from the vertical. The pulse energy is about 90 µJ with a pulsed repetition rate of 1 kHz. The beam is scanned over a round trip cycle at a constant scan speed of 50 µm/sec. Figure 9 (a) plots the measured dimensions with relation to a sequence of focal positions. The relative positions of the focal spot with respect to the wafer are depicted in Figure 9 (b). The surface ablation widths and penetration depths are estimated from the front and the back view optical microscopy images as shown in Figure 9 (c). The position where laser focal point coincides with the surface of GaN is recorded at the z coordinate of 9900 µm. The optimized region for micromachining spans over the range of [10000, 10500] as the surface ablation width is at minimum while the penetration depth and aspect ratio are at maximum values. It is also found that when the wafer deviate from the focus position there is a chance of beam deformation and induce additional scribing run parallel to the desired groove. This is observed at the coordinate of 11000 as shown on the leftmost in Figure 9 (c).

#### **2.5 Chip shaping of light-emitting diodes to improve light extraction**

Light extraction from GaN-based light-emitting diodes is seriously suppressed by total internal reflections within the semiconductor layers. With a high refractive index around 2.4, light extraction from the top surface is limited within a 23 emission cone as depicted in Figure 10 (c). One effective method to enhance light extraction is employing tilting sidewalls such as those in a truncated pyramid (TP) LED, where the conventionally confined light rays are extracted from the top surface via sidewall reflections that redirect the light ray into the top surface emission cone. Accordingly, top surface emission of a laser fabricated TP LED shown in Figure 10 (b) is particularly stronger compared to the conventional rectangular chip in (a). The overall light extraction can be enhanced by 85%. The improvement is attributed to the additional indirect light extraction from top surface via sidewall reflections.

Fig. 10. Optical micrographs of (a) conventional cuboid LED and that of (b)truncated pyramid LED with tilting sidewall,(c)shematic diagram of ehanced top surface light extraction via sidewall reflections. (d) SEM image of the truncated pyramid LED chip shaped by laser micromachining. (adapted from (Fu, et al. 2009) with permission for reproduction from IEEE)

Additional indirect light extraction also exists in a triangular LED, making it unique among polygonal LEDs. However, the mechanism is slightly different with that of a TP LED. In a triangular LED chip, enhanced light extraction is due to indirect light extraction from the sidewall via reflections on neighbouring sidewalls, while in the case of a rectangular chip or other polygons, the indirect extraction is trivial. Actual chip geometry from triangle to heptagon are fabricated with the laser micromachining system and shown in Figure 11.

Fig. 11. Optical Micrograph Polygonal LEDs as fabricated by laser beam (upper row) and biased at 2.5 V (lower row). (with permission for reproduction from American Institute of Physics)

#### **3. Device isolation on GaN-on-sapphire wafer via laser micro-patterning**

GaN is the major material for the fabrication of state-of-the-art blue light-emitting diodes. It is conventionally grown on sapphire substrates by metalorganic chemical vapour deposition (MOCVD), since sapphire is stable and can withstand the high temperature during the growth process. Although there are many issues involved with sapphire, such as lattice mismatch with GaN and poor heat conductivity, sapphire is still prevalent in the fabrication of low-power blue and white LEDs. In addition, being an electrical insulator, sapphire does not interfere with the current conduction in GaN. By selectively removing certain area of GaN, the GaN layer can be separated into multiple electrically isolated small-area LEDs. These LEDs can be connected together by metal interconnects at a later stage, allowing a variety of integrated optoelectronic circuits to be developed.

As GaN is highly resistant to wet etch, dry etch is the conventional technique for the partial or complete removal of GaN. Reactive ion etching (RIE) (Lee, et al. 1995) using CHF3/Ar and C2ClF5/Ar plasmas, for example, can achieve an etch rate between 60 and 470 angstrom/min (Liann-Be, et al. 2001). Inductively coupled plasma (ICP) etching using Cl2 and Ar, on the other hand, offers an attractive etch rate of up to 1 μm/min (Smith, et al. 1997). However, dry etch techniques require masking material to cover the regions not to be removed. Typically, with photoresist as an etch mask, the photoresist layer has to be at least as thick as the GaN layer to be etched (Liann-Be, et al. 2001), which is about 3-4 μm. Spincoating of photoresist layer of this thickness is often cumbersome (for example, edge bead effect may occur (Yang and Chang 2006)), coupled with the fact that thicker photoresists generally offer lower resolutions. Mask thickness can be reduced when hard masks such as SiO2 are used, but additional lithography and dry etch steps are needed for patterning.

Additional indirect light extraction also exists in a triangular LED, making it unique among polygonal LEDs. However, the mechanism is slightly different with that of a TP LED. In a triangular LED chip, enhanced light extraction is due to indirect light extraction from the sidewall via reflections on neighbouring sidewalls, while in the case of a rectangular chip or other polygons, the indirect extraction is trivial. Actual chip geometry from triangle to heptagon are fabricated with the laser micromachining system and shown in Figure 11.

Fig. 11. Optical Micrograph Polygonal LEDs as fabricated by laser beam (upper row) and biased at 2.5 V (lower row). (with permission for reproduction from American Institute of

**3. Device isolation on GaN-on-sapphire wafer via laser micro-patterning** 

variety of integrated optoelectronic circuits to be developed.

GaN is the major material for the fabrication of state-of-the-art blue light-emitting diodes. It is conventionally grown on sapphire substrates by metalorganic chemical vapour deposition (MOCVD), since sapphire is stable and can withstand the high temperature during the growth process. Although there are many issues involved with sapphire, such as lattice mismatch with GaN and poor heat conductivity, sapphire is still prevalent in the fabrication of low-power blue and white LEDs. In addition, being an electrical insulator, sapphire does not interfere with the current conduction in GaN. By selectively removing certain area of GaN, the GaN layer can be separated into multiple electrically isolated small-area LEDs. These LEDs can be connected together by metal interconnects at a later stage, allowing a

As GaN is highly resistant to wet etch, dry etch is the conventional technique for the partial or complete removal of GaN. Reactive ion etching (RIE) (Lee, et al. 1995) using CHF3/Ar and C2ClF5/Ar plasmas, for example, can achieve an etch rate between 60 and 470 angstrom/min (Liann-Be, et al. 2001). Inductively coupled plasma (ICP) etching using Cl2 and Ar, on the other hand, offers an attractive etch rate of up to 1 μm/min (Smith, et al. 1997). However, dry etch techniques require masking material to cover the regions not to be removed. Typically, with photoresist as an etch mask, the photoresist layer has to be at least as thick as the GaN layer to be etched (Liann-Be, et al. 2001), which is about 3-4 μm. Spincoating of photoresist layer of this thickness is often cumbersome (for example, edge bead effect may occur (Yang and Chang 2006)), coupled with the fact that thicker photoresists generally offer lower resolutions. Mask thickness can be reduced when hard masks such as SiO2 are used, but additional lithography and dry etch steps are needed for patterning.

Physics)

In this section, a maskless direct-write laser micromachining technique for device isolation on GaN-on-sapphire wafer is introduced. Unlike wafer dicing, where GaN and sapphire are to be ablated to complete separation, the laser ablation in our new technique automatically terminates at the GaN/sapphire interface. The principle lies on the large difference between: (1) the ablation thresholds (the minimum laser fluence to achieve ablation), and (2) the optical absorption coefficients at ultraviolet (UV) wavelength of GaN and sapphire, as shown in Table 1.


Table 1. Parameters of GaN and sapphire that facilitate selective laser ablation.

When the laser fluence is controlled between the two ablation thresholds, GaN is ablated while sapphire is left undamaged. A simple way to achieve this is by offsetting the wafer from the best focus plane and adjusting the laser spot size. As shown in Figure 12 (a), the laser energy is concentrated to a small spot in the vicinity of the best focus plane. GaN layer (comprising p-type GaN, InGaN/GaN multi-quantum well (MQW) and n-type GaN) and sapphire layer are cut through, which is the mode for die separation. When the focus offset increases, the laser spot is enlarged and the laser fluence is reduced. At a certain range of focus offset, the laser fluence is just high enough to ablate GaN but not sapphire. By scanning the laser across the wafer, a trench terminating at the GaN/sapphire interface is resulted. This is the desired mode for device isolation (Figure 12 (b)). If the focus offset is increased further, the laser fluence will not be sufficient to ablate GaN completely. Device isolation cannot be achieved (Figure 12 (c)).

Fig. 12. Control of laser fluence by focus offset. (with permission for reproduction from

#### American Institute of Physics)

A number of factors affect the quality of trenches. In our study, five laser parameters (focus offset, pulse energy, pulse repetition rate, scan speed and number of scan passes) and two ambient media (air and deionized water) were investigated. By the end of this section, two applications of this laser micromachining technique will also be discussed.

#### **3.1 Trench micromachining in air**

The laser micromachining experiment was first performed in ambient air at room temperature by using the setup shown in Figure 13 (schematic diagram shown in Figure 2). The laser source was a third-harmonic neodymium-doped yttrium lithium fluoride (Nd:YLF) diode-pumped solid-state (DPSS) laser, with center wavelength of 349 nm and pulse repetition rate of single pulse to 5 kHz. The full-width-at-half-maximum (FWHM) pulse width was 4 ns, while the pulse energy was varied by changing the diode pumping current. The expanded and collimated beam was guided by several laser mirrors and focused onto a piece of GaN-on-sapphire sample (emission wavelength = 470 nm, thickness of GaN = 3 μm and thickness of sapphire = 300 μm) on an XY motorized stage. The fusedsilica focusing triplet lens allowed UV and visible light to pass through and had a focal length of 19 mm. As the stage translated while keeping the laser spot stationary, trenches were scribed onto the sample. The scan speed was controlled by software with a precision up to 25 μm/s. The sample could be shifted away from the focus by manually adjusting the stage height. The accuracy of height adjustment was ±5 μm. A charge-coupled device (CCD) camera was installed confocal to the optical path for real-time observation of the micromachining process. Owing to the high temperature during laser ablation, sedimentary

Fig. 13. Experimental setup for laser micromachining in air. (with permission for reproduction from American Institute of Physics)

by-products were formed on the surface of GaN, such as Ga metal (Akane, et al. 1999b; Kelly, et al. 1996) and gallium oxide (Gu, et al. 2006). These substances were effectively removed by sonification of the sample in dilute hydrochloric acid (HCl) (18% by mass) for 15 min. The sample was then rinsed in DI water to remove the remaining acid. The morphology of the resulting trenches were observed by field-emission scanning electron microscopy (FE-SEM), identifying the effect of each laser parameter towards the trench quality.

#### **3.1.1 Focus offset**

96 Micromachining Techniques for Fabrication of Micro and Nano Structures

The laser micromachining experiment was first performed in ambient air at room temperature by using the setup shown in Figure 13 (schematic diagram shown in Figure 2). The laser source was a third-harmonic neodymium-doped yttrium lithium fluoride (Nd:YLF) diode-pumped solid-state (DPSS) laser, with center wavelength of 349 nm and pulse repetition rate of single pulse to 5 kHz. The full-width-at-half-maximum (FWHM) pulse width was 4 ns, while the pulse energy was varied by changing the diode pumping current. The expanded and collimated beam was guided by several laser mirrors and focused onto a piece of GaN-on-sapphire sample (emission wavelength = 470 nm, thickness of GaN = 3 μm and thickness of sapphire = 300 μm) on an XY motorized stage. The fusedsilica focusing triplet lens allowed UV and visible light to pass through and had a focal length of 19 mm. As the stage translated while keeping the laser spot stationary, trenches were scribed onto the sample. The scan speed was controlled by software with a precision up to 25 μm/s. The sample could be shifted away from the focus by manually adjusting the stage height. The accuracy of height adjustment was ±5 μm. A charge-coupled device (CCD) camera was installed confocal to the optical path for real-time observation of the micromachining process. Owing to the high temperature during laser ablation, sedimentary

Fig. 13. Experimental setup for laser micromachining in air. (with permission for

reproduction from American Institute of Physics)

**3.1 Trench micromachining in air** 

Figure 14 shows the micromachined trenches at three different focus offset levels while keeping the pulse energy, repetition rate, and scan speed constant. Upward focus offset is taken as positive. The results follow the principle introduced at the beginning of this section. In Figure 14 (a) where the sample is positioned near the best focal plane (300 μm above), the laser beam ablates both the GaN (lighter colour) and sapphire (darker colour). A V-shaped valley is formed in the sapphire layer due to the Gaussian beam shape. Although trenches like these serve the purpose of electrical isolation between adjacent devices, the deep Vshaped valley is not suitable for the conformal deposition of metal interconnect, since the interconnection will become discontinuous at the sharp corners of the valley. At the optimal focal offset plane (450 μm above), as shown in Figure 14 (b), the ablation terminates automatically at the GaN/sapphire interface, exposing a flat and smooth sapphire bottom surface. At a larger focus offset plane of 600 μm, the GaN layer is not completely removed, leaving a shallow and rugged trench on the surface (Figure 14 (c)).

Fig. 14. SEM images of trenches laser micromachined at different focus offset planes: (a) small offset of 300 μm; (b) optimal offset of 450 μm; (c) large offset of 600 μm. The pulse energy, pulse repetition rate, and scan speed were fixed at 23 μJ, 1 kHz, and 25 μm/s, respectively. (with permission for reproduction from American Institute of Physics)

#### **3.1.2 Pulse energy**

Pulse energy is another determining factor of trench quality. Figure 15 illustrates micromachined trenches processed at three different pulse energies between 7 and 45 μJ, while keeping all other parameters constant. The focus offset is kept at the optimal value of 450 μm, as determined from the previous set of experiment. When the pulse energy is set too high, the effect is similar to that of having a smaller focus offset, whereby the GaN as well as sapphire are ablated to form a V-shaped trench (Figure 15 (a)). Similar correspondence between low pulse energy and large focus offset can be observed in Figure 15 (c). Notice that the trench width also increases for higher pulse energy. This property will be further explored in laser micromachining in DI water.

Fig. 15. SEM images of trenches under different pulse energy: (a) higher pulse energy of 45 μJ; (b) optimal pulse energy of 23 μJ; (c) lower pulse energy of 7 μJ. The focus offset level, pulse repetition rate, and scan speed are fixed at 450 μm, 1 kHz, and 25 μm/s, respectively. (with permission for reproduction from American Institute of Physics)

#### **3.1.3 Pulse repetition rate**

Trenches that are laser-micromachined under an increasing pulse repetition rate are shown in Figure 16 (a)-(c); all other parameters are kept constant. When the pulse repetition rate increases from 1 to 5 kHz, the trench width remains more or less unchanged, but the sidewall and bottom surfaces become increasingly smooth. This observation can be understood in terms of heat accumulation effects and its consequence to the etch efficiency. As the repetition rate increases, cumulative heating by earlier pulses causes localized melting of the material (Schaffer, et al. 2003). This results in an increase in the average surface temperature and thus the removal rate of the ablated materials, minimizing redeposition of debris over the trench.

Fig. 16. SEM images of trenches under different pulse repetition rate: (a) 1 kHz; (b) 3 kHz; (c) 5 kHz. The focus offset, pulse energy, and scan speed were fixed at 450 μm, 23 μJ, and 25 μm/s, respectively. (with permission for reproduction from American Institute of Physics)

#### **3.1.4 Scan speed**

The rate at which the laser beam scans across the material is also investigated. From Figure 17, a faster translation rate does not result in a change in the trench width. However, it leads to degradation in the trench quality. At a faster translation speed, the exposure time to the laser light at each position becomes shorter. There is no enough time for temperature rise and/or photon-matter interaction. Stalagmite-like structures begin to appear around the sidewalls.

Fig. 17. SEM images of trenches under different scan speeds: (a) 25 μm/s; (b) 75 μm/s; (c) 125 μm/s. The focus offset, pulse energy, and repetition rate were kept at 450 μm, 23 μJ, and 5 kHz, respectively. (with permission for reproduction from American Institute of Physics)

#### **3.1.5 Number of scan passes**

98 Micromachining Techniques for Fabrication of Micro and Nano Structures

Fig. 15. SEM images of trenches under different pulse energy: (a) higher pulse energy of 45 μJ; (b) optimal pulse energy of 23 μJ; (c) lower pulse energy of 7 μJ. The focus offset level, pulse repetition rate, and scan speed are fixed at 450 μm, 1 kHz, and 25 μm/s, respectively.

Trenches that are laser-micromachined under an increasing pulse repetition rate are shown in Figure 16 (a)-(c); all other parameters are kept constant. When the pulse repetition rate increases from 1 to 5 kHz, the trench width remains more or less unchanged, but the sidewall and bottom surfaces become increasingly smooth. This observation can be understood in terms of heat accumulation effects and its consequence to the etch efficiency. As the repetition rate increases, cumulative heating by earlier pulses causes localized melting of the material (Schaffer, et al. 2003). This results in an increase in the average surface temperature and thus the removal rate of the ablated materials, minimizing

Fig. 16. SEM images of trenches under different pulse repetition rate: (a) 1 kHz; (b) 3 kHz; (c) 5 kHz. The focus offset, pulse energy, and scan speed were fixed at 450 μm, 23 μJ, and 25 μm/s, respectively. (with permission for reproduction from American Institute of Physics)

The rate at which the laser beam scans across the material is also investigated. From Figure 17, a faster translation rate does not result in a change in the trench width. However, it leads to degradation in the trench quality. At a faster translation speed, the exposure time to the laser light at each position becomes shorter. There is no enough time for temperature rise and/or photon-matter interaction. Stalagmite-like structures begin to appear around the

(with permission for reproduction from American Institute of Physics)

**3.1.3 Pulse repetition rate** 

**3.1.4 Scan speed** 

sidewalls.

redeposition of debris over the trench.

The remaining factor to consider is the number of scans. Similar to the effect of increasing scan speed, an increase in the number of scans does not alter the trench width. However, a narrow groove is formed at the center of the trenches for three and five passes, as illustrated in Figure 18 (b) and (c), respectively. There is also no remarkable improvement in the sidewall and bottom surface quality.

Fig. 18. SEM images of trenches under different number of scan passes: (a) single pass; (b) three passes; (c) five passes. The focus offset, pulse energy, repetition rate and scan speed were kept at 450 μm, 23 μJ, 5 kHz, and 25 μm/s, respectively. (with permission for reproduction from American Institute of Physics)

Through the above experiments, we can conclude that an optimal combination of focus offset and pulse energy, higher pulse repetition rate, slower scan speed and single pass of scan are essential for good trench quality. Nevertheless, substantial amount of redeposition and resolidification of ablated material still exists on the trench bottom surface and sidewall when the process is performed in air. This is the result of thermal ablation and photochemical ablation mechanisms of nanosecond lasers. Although the redeposition can be effectively removed by strong acids, this is not feasible when the underlying material also reacts with the acids. In order to reduce the heat load during ablation, laser micromachining with the sample immersed in a liquid is proposed. Criteria for the liquid include good thermal conductivity and high specific heat capacity. In addition, the attenuation of UV and visible light in that liquid should be low, so that laser energy can be transferred efficiently to the substrate and the micromachining process can be monitored concurrently. DI water would be a good choice to match these criteria. The mechanisms involved in the liquidimmersion laser micromachining of GaN will be investigated in the following subsection.

#### **3.2 Trench micromachining in DI water**

The characteristics of liquid-immersion laser micromachining for GaN were investigated experimentally using the optical setup shown in Figure 19. The setup was similar to that used for ambient air (Figure 13), except that the GaN-on-sapphire sample was immersed horizontally in a DI water bath with meniscus about 1 mm above the sample surface. Although thicker water layer can improve heat dissipation, attenuation of the laser beam will become more severe. On the other hand, the water layer cannot be too thin since the entire sample will not be wetted under the strong surface tension of water. The water bath was placed on the manual Z translation stage, mounted on the motorized XY translation stage to enable laser scanning. The laser fluence was again adjusted by offsetting the sample surface from the best focal plane. Besides observing the surface morphology by FE-SEM, the trench surface roughness was measured by atomic force microscopy (AFM). Elemental analysis of the trench surface was performed by the energy-dispersive X-ray spectroscopy (EDX) function offered by the FE-SEM.

Fig. 19. Experimental setup for laser micromachining in DI water. (with permission for reproduction from Springer)

#### **3.2.1 Trench quality as compared with ambient air**

Compared with laser micromachining in ambient air, DI water is capable of producing trenches with substantially smoother sidewalls and bottom surfaces. Figure 20 shows AFM scans of the trenches generated in both ambient media. By measuring the height values *<sup>i</sup> y*

The characteristics of liquid-immersion laser micromachining for GaN were investigated experimentally using the optical setup shown in Figure 19. The setup was similar to that used for ambient air (Figure 13), except that the GaN-on-sapphire sample was immersed horizontally in a DI water bath with meniscus about 1 mm above the sample surface. Although thicker water layer can improve heat dissipation, attenuation of the laser beam will become more severe. On the other hand, the water layer cannot be too thin since the entire sample will not be wetted under the strong surface tension of water. The water bath was placed on the manual Z translation stage, mounted on the motorized XY translation stage to enable laser scanning. The laser fluence was again adjusted by offsetting the sample surface from the best focal plane. Besides observing the surface morphology by FE-SEM, the trench surface roughness was measured by atomic force microscopy (AFM). Elemental analysis of the trench surface was performed by the energy-dispersive X-ray spectroscopy

Fig. 19. Experimental setup for laser micromachining in DI water. (with permission for

Compared with laser micromachining in ambient air, DI water is capable of producing trenches with substantially smoother sidewalls and bottom surfaces. Figure 20 shows AFM scans of the trenches generated in both ambient media. By measuring the height values *<sup>i</sup> y*

**3.2 Trench micromachining in DI water** 

(EDX) function offered by the FE-SEM.

reproduction from Springer)

**3.2.1 Trench quality as compared with ambient air** 

along the trench edge, the sidewall roughness *Ra* is determined by taking the arithmetic average of the absolute height deviation from the mean height. The trench micromachined in air has an *Ra* of 312 nm, contrasting sharply with 27.65 nm for the trench produced in DI water. The rms roughness of the bottom surfaces also reveals the superiority of liquid immersion (87.49 nm for air vs. 13.42 nm for DI water). As a comparison with inorganic material whose laser-matter interaction should be similar to that of GaN, we quoted that the *Ra* value of microchannels micromachined by femtosecond laser in aluminosilicate glass sheet in ambient air is approximately 40 nm (Zheng, et al. 2-006). This indicates that nanosecond laser micromachining in DI water can have comparable performance with femtosecond laser micromachining in air.

Fig. 20. AFM scans of trenches generated in: (a) and (c) air; (b) and (d) DI water. (c) and (d) are the zoomed-in images of the bottom surfaces. The focus offset, pulse energy, pulse repetition rate, scan speed were fixed at 400 μm, 25 μJ, 1 kHz and 25 μm/s. Only single pass of scan was performed. (with permission for reproduction from Springer)

Liquid-immersion laser micromachining also results in trenches of quality comparable to conventional lithographic and ICP deep etch processes. From the SEM images of sidewalls around ICP-etched regions in the literature (Ladroue, et al. 2010; Qiu, et al. 2011), we see that the smoothness is comparable to that of the laser-micromachined trench sidewalls. Besides, striations are observed over the sidewalls of ICP-etched regions when Ni hard mask is used. This is not observed in liquid-immersion laser micromachining, as shown in Figure 20 (b). Though SiO2 hard mask can be used to eliminate the striations in ICP, it comes with the price of sidewall steepness reduction.

As a demonstration of reduced redeposition, the EDX results of the trench bottom surface are shown in Table 2. Three elements were found: O, Al and Ga. Al and O are the constituent elements of sapphire (Al2O3), whereas Ga is a product from the thermal decomposition of GaN during laser ablation (Ambacher, et al. 1996; Choi, et al. 2002). It is found that there is a lower percentage of Ga over those trenches produced in DI water. It should be noted that while the sample for air had been sonicated in dilute HCl before EDX examination, that for DI water was not subjected the same treatment. The results indicate that liquid immersion is effective in ejecting the molten Ga (m.p. = 29.76 C, b.p. = 2204 C) from the irradiated site and preventing its resolidification around the trench.


Table 2. Atomic composition of trench bottom surface.

Besides improved surface quality, liquid-immersion laser micromachining also offers two advantages with respect to process control. The first is increased focus offset tolerance. Figure 21 (a)-(c) show three trenches micromachined in ambient air. When the sample position deviates from the optimum plane (Figure 21 (b)) by 150 μm, either damage to the sapphire layer (Figure 21 (a)) or incomplete trench (Figure 21 (c)) occurs. For micromachining in DI water, the trench quality is not compromised even with a deviation as large as 200 μm, albeit a slight decrease in the trench width (Figure 21 (d)-(f)). The second advantage of liquid immersion is the control over trench width by varying pulse energy. Owing to the focus offset tolerance, the variation of pulse energy does not significantly compromise the trench quality. Figure 22 shows an approximately linear relationship between pulse energy and trench width when performing micromachining in DI water at a fixed focus offset. For micromachining in ambient air, it is difficult to control the trench width just by altering the pulse energy. This is because the optimum focus offset depends strongly on the pulse energy. A new optimum focus offset needs to be found when the pulse energy is altered.

Fig. 21. SEM images of trenches laser-micromachined at different focus offset planes: (a)-(c) in air: (a) 300 μm, (b) 450 μm, (c) 600 μm, where 450 μm is the optimum; (d)-(f): in DI water: (d) 300 μm; (e) 500 μm; (f) 700 μm, where 500 μm is the optimum. The pulse energy, pulse repetition rate and scan speed were fixed at 25 μJ, 1 kHz and 25 μm/s. (with permission for reproduction from Springer)

Fig. 22. Trench width control by varying pulse energy. The focus offset, pulse repetition rate and scan speed were fixed at 550 μm, 1 kHz and 25 μm/s. (with permission for reproduction from Springer)

#### **3.3 Theoretical discussions**

The improved trench quality can be explained in terms of the following processes during the nanosecond-pulsed laser ablation of GaN: Heat transfer, plasma-induced recoil pressure, plasma shielding effect in water and collapse of cavitation bubbles.

#### **3.3.1 Heat transfer**

102 Micromachining Techniques for Fabrication of Micro and Nano Structures

O 64.67% 65.19% Al 34.81% 34.65% Ga 0.52% 0.16%

Besides improved surface quality, liquid-immersion laser micromachining also offers two advantages with respect to process control. The first is increased focus offset tolerance. Figure 21 (a)-(c) show three trenches micromachined in ambient air. When the sample position deviates from the optimum plane (Figure 21 (b)) by 150 μm, either damage to the sapphire layer (Figure 21 (a)) or incomplete trench (Figure 21 (c)) occurs. For micromachining in DI water, the trench quality is not compromised even with a deviation as large as 200 μm, albeit a slight decrease in the trench width (Figure 21 (d)-(f)). The second advantage of liquid immersion is the control over trench width by varying pulse energy. Owing to the focus offset tolerance, the variation of pulse energy does not significantly compromise the trench quality. Figure 22 shows an approximately linear relationship between pulse energy and trench width when performing micromachining in DI water at a fixed focus offset. For micromachining in ambient air, it is difficult to control the trench width just by altering the pulse energy. This is because the optimum focus offset depends strongly on the pulse energy. A new optimum focus offset needs to be found when the pulse

Fig. 21. SEM images of trenches laser-micromachined at different focus offset planes: (a)-(c) in air: (a) 300 μm, (b) 450 μm, (c) 600 μm, where 450 μm is the optimum; (d)-(f): in DI water: (d) 300 μm; (e) 500 μm; (f) 700 μm, where 500 μm is the optimum. The pulse energy, pulse repetition rate and scan speed were fixed at 25 μJ, 1 kHz and 25 μm/s. (with permission for

Table 2. Atomic composition of trench bottom surface.

energy is altered.

reproduction from Springer)

Air DI water

Referring to Table 3, DI water has a much higher specific heat capacity and thermal conductivity than air. Therefore DI water is expected to carry the excess heat away from the irradiated region faster than air. To verify this, a simulation on the heat conduction process for single-pulse irradiation was performed. The heat equation (in cylindrical coordinates) was solved by finite element method (FEM):

$$
\rho c \frac{\partial T}{\partial t} - \nabla \cdot \left(k \nabla T\right) = Q \tag{2}
$$

where , *c* and *k* are as defined in Table 3, *T* is the temperature distribution (K) and *Q* represents the heat source (W mm-3). *Q* originates from the Gaussian laser beam, thus both *Q* and the resulting *T* are assumed to be radially symmetric. The domain and boundary conditions are defined in Figure 23. Without going into the details, the simulation results are presented as follows. Figure 24 shows the variation of GaN surface temperature during the initial 100 ns at the center of the laser spot *r z* 0, 0 (solid curves) and near the trench edge *r z* 15 μm, 0 (dotted curves). The GaN surface temperature in air is found to be higher than that in DI water at both positions. In addition, sharper temperature peaks are found when the micromachining is performed in air. The maximum temperature is as high as 1000 C even near the trench edge. It is known that GaN begins to decompose into liquid Ga and N2 gas at a temperature of 900 C (Choi, et al. 2002). The rapid heating and cooling cycles in air can result in the increased generation and resolidification of molten Ga within each pulse period. The resolidified Ga droplets deposit around the sidewall, degrading the surface quality. Another consequence of rapid heating and cooling is the increased thermal stress incurred in the crystal structure of GaN. Cracks may be resulted.


Table 3. Density , specific heat capacity *c* and thermal conductivity *k* of the substances involved in the laser micromachining process.

Fig. 23. Simulation domain. (with permission for reproduction from Springer)

Fig. 24. Temporal variation of GaN surface temperature at two different positions: at the center of laser spot (solid); near the edge of the trench (dotted). (with permission for reproduction from Springer)

The two-dimensional temperature distribution also demonstrates the strong cooling effect of water. As shown in Figure 25, at *t* = 1 μs, the temperature of water immediately above the sample surface is lower than that of air. The slight decrease of water temperature below the room temperature can be attributed to the adiabatic expansion of water vapour as well as the vapour plume of ablated material (Gusarov, et al. 2000). At this instance, the GaN surface temperature in water is in general lower than that in air, indicating efficient heat extraction by water. The heat-affected zone (HAZ) is also seen to be smaller when the sample is immersed in DI water.

Fig. 25. Temperature distribution at t = 1 μs when the ambient medium is: (a) air; (b) DI water. (with permission for reproduction from Springer)

#### **3.3.2 Plasma-induced recoil pressure**

104 Micromachining Techniques for Fabrication of Micro and Nano Structures

GaN 6.15 0.49 1.3 Al2O3 4.025 0.75 0.35

(g cm-3) *c* (J g-1 oC-1) *k* (W cm-1 oC-1)

, specific heat capacity *c* and thermal conductivity *k* of the substances

C) 0.006

C ) 2.5 × 10-4

Table 3. Density

reproduction from Springer)

sample is immersed in DI water.

involved in the laser micromachining process.

H2O 1.0 4.18 (at 25

Air 1.184 × 10-3 1.012 (at 23

Fig. 23. Simulation domain. (with permission for reproduction from Springer)

Fig. 24. Temporal variation of GaN surface temperature at two different positions: at the center of laser spot (solid); near the edge of the trench (dotted). (with permission for

The two-dimensional temperature distribution also demonstrates the strong cooling effect of water. As shown in Figure 25, at *t* = 1 μs, the temperature of water immediately above the sample surface is lower than that of air. The slight decrease of water temperature below the room temperature can be attributed to the adiabatic expansion of water vapour as well as the vapour plume of ablated material (Gusarov, et al. 2000). At this instance, the GaN surface temperature in water is in general lower than that in air, indicating efficient heat extraction by water. The heat-affected zone (HAZ) is also seen to be smaller when the High-energy laser pulses are able to melt, vaporize and ionize the material being irradiated, resulting in the formation of plasma. When the plasma expands, a recoil pressure is exerted on the sample surface, causing material removal. If the ambient medium is water instead of air, the expansion of plasma will be confined by the underwater pressure. This leads to an increase in the plasma-induced recoil pressure and higher removal rate of material.

#### **3.3.3 Plasma shielding effect in water**

The plasma generated from laser ablation can absorb part of the incident laser energy and reduce the energy delivered to the sample. This is known as plasma shielding effect. The effect becomes stronger when the plasma has a larger size, longer duration and a starting time that overlaps with the laser pulse. Because of water confinement effects, the plasma size and duration are much reduced in water. The onset of plasma formation is also delayed by 5 ns, reducing the overlap time between the laser pulse and the plasma (Hong, et al. 2002). These factors weaken the plasma shielding effect, causing a stronger coupling of the laser beam with the material. Material removal is thus more efficient in water.

#### **3.3.4 Collapse of cavitation bubbles**

Formation of cavitation bubbles is a process that happens solely in liquids. In the experiment, the bubbles may originate from the dissolved gases in DI water, or the N2 gas from the laser ablation of GaN. When a bubble collapses near the sample surface, a highspeed liquid jet directed towards the surface will be generated. If there is no water between the bubble and sample surface, the liquid jet will produce a strong impulse towards the sample. This impulse can be 5.2-12.4 times that of the laser ablation impact in air (Lu, et al. 2004), which further helps the liquid-phase expulsion of Ga.

#### **3.4 Applications**

To close this section, two applications of liquid-immersion laser micromachining are introduced: alternating-current LED (ac-LED) and 5 × 7 dot-matrix microdisplay. Ac-LED is an LED chip that is directly powered by an ac voltage. The basic idea is to make use of the diode property of LEDs to construct an on-chip bridge rectifier. The rectified voltage is then used to power other on-chip LEDs. The plan view of the chip overlaid with the circuit diagram is shown in Figure 27 (a). The on-chip LEDs are isolated by the trenches from liquid-immersion laser micromachining. This chip is designed to work on 12 Vrms, but the design can be easily extended to higher voltages. Its full operation is shown in Figure 26 (b). The other application, dot-matrix microdisplay, also requires laser micromachining to separate rows of LED pixels (Figure 27 (a) and (b)). Narrow trenches are formed by low pulse energy, allowing pixels to be packed more closely together. By controlling the on-off sequence of pixels, alphanumeric characters can be formed, as shown in Figure 27 (c).

Fig. 26. Microphotograph of ac-LED, with circuit diagram superimposed. The directions of current flow during the positive and negative cycles of ac voltage are shown with red and blue arrows respectively; (b) Microphotograph of ac-LED operating under an ac voltage of 12 Vrms. (with permission for reproduction from American Institute of Physics)

Fig. 27. Circuit diagram of 5 × 7 dot-matrix microdisplay; (b) Microphotograph of the microdisplay chip; (c) Microdisplay in operation, showing the character "U".

#### **4. Conclusion**

Laser micromachining involving a nanosecond diode-pumped solid-state laser at 349 nm wavelength is developed for device fabrication of light-emitting diodes on sapphire wafer. The material ablation process can be readily controlled and optimized for varied micromachining purposes such as wafer cutting featuring high aspect ratios and precision etching for device isolation. Highest aspect ratio of the cleave on GaN/sapphire wafer is achieved by front side laser micromachining method with the beam focused at the GaN layer. With laser micromachining process optimized for chip shaping, novel chip geometry such as truncated pyramid and triangular LED is fabricated and the novel LED geometries effectively improve light extraction efficiency. Influence of ambient medium as well as effects of various process parameters are taken into account to develop a reliable laser micro-patterning process. The laser beam, focused with a shorter focal length focusing lens, is tuned to selectively etch conductive GaN layers on the insulating sapphire substrate generating isolation grooves with desired profile and pattern that facilitates further processing.

#### **5. References**

106 Micromachining Techniques for Fabrication of Micro and Nano Structures

used to power other on-chip LEDs. The plan view of the chip overlaid with the circuit diagram is shown in Figure 27 (a). The on-chip LEDs are isolated by the trenches from liquid-immersion laser micromachining. This chip is designed to work on 12 Vrms, but the design can be easily extended to higher voltages. Its full operation is shown in Figure 26 (b). The other application, dot-matrix microdisplay, also requires laser micromachining to separate rows of LED pixels (Figure 27 (a) and (b)). Narrow trenches are formed by low pulse energy, allowing pixels to be packed more closely together. By controlling the on-off sequence of pixels, alphanumeric characters can be formed, as shown in Figure 27 (c).

Fig. 26. Microphotograph of ac-LED, with circuit diagram superimposed. The directions of current flow during the positive and negative cycles of ac voltage are shown with red and blue arrows respectively; (b) Microphotograph of ac-LED operating under an ac voltage of

12 Vrms. (with permission for reproduction from American Institute of Physics)

Fig. 27. Circuit diagram of 5 × 7 dot-matrix microdisplay; (b) Microphotograph of the

Laser micromachining involving a nanosecond diode-pumped solid-state laser at 349 nm wavelength is developed for device fabrication of light-emitting diodes on sapphire wafer. The material ablation process can be readily controlled and optimized for varied

microdisplay chip; (c) Microdisplay in operation, showing the character "U".

**4. Conclusion** 


## **Laser Ablation for Polymer Waveguide Fabrication**

Shefiu S. Zakariyah *Advanced Technovation Ltd, Loughborough Innovation Centre, Loughborough, UK* 

#### **1. Introduction**

108 Micromachining Techniques for Fabrication of Micro and Nano Structures

Liu, Wei-Min, et al. 2002 Ablation of GaN using a femtosecond laser. Chinese Physics

Liu, X., D. Du, and G. Mourou 1997 Laser ablation and micromachining with ultrashort laser

Lu, J., et al. 2004 Mechanisms of laser drilling of metal plates underwater. Journal of Applied

Molian, P., B. Pecholt, and S. Gupta 2009 Picosecond pulsed laser ablation and micromachining of 4H-SiC wafers. Applied Surface Science 255(8):4515-4520. Muth, JF, et al. 1997 Absorption coefficient, energy gap, exciton binding energy, and

Patel, BS, and ZH Zaidi 1999 The suitability of sapphire for laser windows. Measurement

Qiu, Rongfu, et al. 2011 Optimization of inductively coupled plasma deep etching of GaN

Schaffer, C. B., et al. 2001 Micromachining bulk glass by use of femtosecond laser pulses

Schaffer, C. B., J. F. Garcia, and E. Mazur 2003 Bulk heating of transparent materials using a

Smith, SA, et al. 1997 High rate and selective etching of GaN, AlGaN, and AlN using an inductively coupled plasma. Applied Physics Letters 71(25):3631-3633. Stuart, B. C., et al. 1996 Nanosecond-to-femtosecond laser-induced breakdown in dielectrics.

Varel, H., et al. 1997 Micromachining of quartz with ultrashort laser pulses. Applied Physics

Yang, Y.K., and T.C. Chang 2006 Experimental analysis and optimization of a photo resist

Zheng, H. Y., et al. 2006 Ultrashort pulse laser micromachined microchannels and their

Zweig, A. D. 1991 A thermomechanical model for laser ablation. Journal of Applied Physics

coating process for photolithography in wafer fabrication. Microelectronics Journal

application in an optical switch. International Journal of Advanced Manufacturing

high-repetition-rate femtosecond laser. Applied Physics a-Materials Science &

recombination lifetime of GaN obtained from transmission measurements. Applied

pulses. Ieee Journal of Quantum Electronics 33(10):1706-1716.

and etching damage analysis. Applied Surface Science 257(7).

with nanojoule energy. Optics Letters 26(2):93-5.

Letters 19(11).

Physics 95(8):3890-3894.

Physics Letters 71:2572.

Processing 76(3):351-354.

37(8):746-751.

70(3):1684-1691.

Physical Review B 53(4):1749-1761.

Technology 27(9-10):925-929.

A (Materials Science Processing) 65(4-5).

Science and Technology 10:146.

An increase in interconnection density, a reduction in packaging sizes and the quest for lowcost product development strategy are some of the key challenges facing micro-optoelectronics design and manufacture. The influence of high-density, small-sized products has placed significant constraints on conventional electrical connections prompting various fabrication methods, e.g. photolithography, being introduced to meet these challenges and ameliorate the rapidly changing demand from consumers. While high-power solid state lasers are fundamental to large scale industrial production, excimer laser on the other hand has revolutionised the manufacturing industry with high precision, easy 3D structuring and less stringent production requirements. Micro-structuring using excimer laser, best known as laser ablation, is a non-contact micro- and nano-machining based on the projection of high-energy pulsed UV masked beam on to a material of interest such that pattern(s) on the mask is transferred to the substrate, often at a demagnified dimension with high resolution and precision. The use of mask with desired patterns and beam delivery system makes the fabrication in this case accurate, precise and easily controllable. The first part of this chapter introduces the fundamentals of laser technology and material processing. In the second part, optical interconnects as a solution to 'bottlenecked' conventional copper interconnections is introduced with emphasis on excimer laser ablation of polymer waveguides and integrated mirrors. Key research findings in the area of optical circuit boards using other techniques are also briefly covered.

#### **2. Introduction to laser technology**

The word 'laser' has been part of the lexis of the English language since its invention in 1960 and subsequent commercialisation few years later. It is an acronym that stands for Light Amplification by Stimulated Emission of Radiation, which is considered a modified version of its predecessor - 'maser' (Microwave Amplification by Stimulated Emission of Radiation); in other words, laser is an optical maser. The first laser, ruby, emitted red-coloured light at λ = 694.3 nm. Just over five decades later, laser (and laser technology) controls a remarkable market share in various applications ranging from research and medicine, to manufacturing and domestic applications. One of the sectors that have seen dramatic advancement with the advent of lasers is medical surgery (e.g. ophthalmology, cosmetic surgery and dentistry). Laser generation has been extensively covered in the literature, but essentially, but essentially there are three principles that must first take place: (i) stimulated emission to defeat spontaneous emission and absorption, (ii) population inversion to temporarily disturb normal distribution - these two processes require movement of species from a lower energy level to a higher one, and (iii) a feedback system to amplify the photon population.

#### **2.1 Laser micromachining (or material processing)**

Laser material processing is generally, though not technically, referred to as laser micromachining of engineering materials e.g. polymer, metals, glass and ceramics. This definition thus excludes applications of lasers to, for example, human tissues even though the mechanism is similar. The possible reason for this exclusive usage might be because early laser candidates found application in engineering sectors such as drilling and cutting of materials where high energies are needed. For laser micromachining, there are four key processes of importance (Figure 1).

Fig. 1. Schematic diagram showing key stages of a typical laser material processing.

#### **Beam generation**

This is the first stage and the backbone of any material processing; its output determines the components of the remaining stages. For example, if a ceramic material is to be processed then the output at this stage should be a high-powered laser. Furthermore, if the ceramic is to be processed with minimum thermal damage then the output beam should, for example, be a pulsed laser with short pulse duration to provide a minimum time interaction between the beam and the material.

#### **Beam delivery or propagation**

This involves transporting the output beam to the site of processing or workpiece. What constitutes the beam delivery system depends on the application. In general, the elements of the stage, whose number and arrangement varies, include various optical devices such as mirrors, lenses and attenuator among others. It is therefore imperative that careful combination is made to achieve optimum result without losing much power as a small fraction of beam energy is lost per element. Also to be considered is the length of the path between the laser chamber output window and the workpiece. This needs to be kept to a minimum in order to avoid beam profile distortion and divergence. Excimer laser usually has the longest beam path with the highest number of optical components while a CO2 laser employs the least.

#### **Laser beam monitoring**

110 Micromachining Techniques for Fabrication of Micro and Nano Structures

Laser generation has been extensively covered in the literature, but essentially, but essentially there are three principles that must first take place: (i) stimulated emission to defeat spontaneous emission and absorption, (ii) population inversion to temporarily disturb normal distribution - these two processes require movement of species from a lower energy level to a higher one, and (iii) a feedback system to amplify the photon population.

Laser material processing is generally, though not technically, referred to as laser micromachining of engineering materials e.g. polymer, metals, glass and ceramics. This definition thus excludes applications of lasers to, for example, human tissues even though the mechanism is similar. The possible reason for this exclusive usage might be because early laser candidates found application in engineering sectors such as drilling and cutting of materials where high energies are needed. For laser micromachining, there are four key

Fig. 1. Schematic diagram showing key stages of a typical laser material processing.

This is the first stage and the backbone of any material processing; its output determines the components of the remaining stages. For example, if a ceramic material is to be processed then the output at this stage should be a high-powered laser. Furthermore, if the ceramic is to be processed with minimum thermal damage then the output beam should, for example, be a pulsed laser with short pulse duration to provide a minimum time interaction between

This involves transporting the output beam to the site of processing or workpiece. What constitutes the beam delivery system depends on the application. In general, the elements of the stage, whose number and arrangement varies, include various optical devices such as mirrors, lenses and attenuator among others. It is therefore imperative that careful combination is made to achieve optimum result without losing much power as a small fraction of beam energy is lost per element. Also to be considered is the length of the path between the laser chamber output window and the workpiece. This needs to be kept to a minimum in order to avoid beam profile distortion and divergence. Excimer laser usually has the longest beam path with the highest number of optical components while a CO2 laser

**2.1 Laser micromachining (or material processing)** 

processes of importance (Figure 1).

**Beam generation** 

employs the least.

the beam and the material. **Beam delivery or propagation**  Many of the laser beam properties are essential for an optimum process. However, three of these - energy, beam diameter and beam profile – are highly important in micromachining. There are two methods of obtaining the beam energy. In the first approach, the beam is sampled during the processing; this provides an accurate account of beam energy utilised during a particular process. It is pertinent to note that this task is in some way difficult and risky. Three methods of beam sampling: static beam splitter, rotating chopper mirror and leaky resonator mirror are discussed in [Crafer & Oakley, 1993]. The second approach is by total beam measurement; this approach involves measuring the energy at the workpiece using a power meter. Although the method might not totally account for what happens during a process, it is easier than the sampling method [Crafer and Oakley, 1993]. A common way of examining both the beam diameter and profile is by using low energy to irradiate a suitable material; the etched sample is then analysed to measure the diameter and observe the profile. This is an indicative method especially when the process is thermal. Alternatively, beam profile and homogeneity is monitored using a beam profiler which shows the shape of the beam, in real-time, during a process.

#### **Laser-matter interaction (Laser processing)**

The wave-particle duality concept is quite useful in treating laser-matter interaction. For example, laser generation is better described using the quantum (or particle) approach while propagation and delivery is suitably described using the wave concept. For laser-matter interaction, it is appropriate to use quantum physics. Thus viewing the beam as a packet of photons hitting the matter with which it is interacting. When the laser beam strikes the material, the photon energy is transferred to the material and subsequently converted to other forms of energy depending on the material. With metals, this is transferred to the mobile electrons which results in the heat energy that can cause vaporisation and disintegration of the metal. However, with non-metals, the energy can either be converted to chemical energy required for bond-breaking or heat energy for vaporization. These two possibilities depend on the type of material, its bond energy and the wavelength of the laser or more precisely the photon energy. Essentially, there are two common mechanisms for laser material interactions, which can occur at varying degrees while processing a material.


The etch rate – the amount of material removed per pulse – is mainly a function of the photon energy and the material being processed. However, it is impractical to model lasermatter interactions based on the aforementioned two quantities as the mechanism is also influenced by numerous other factors (e.g. thermal diffusion, absorption saturation, surrounding medium, etc.) such that the measured ablation depths seldom agree with these predictions; this necessitates more complex 'models' often based on these two quantities [Tseng, et al., 2007]. Equations 1 & 2 provide two often referenced mathematical representations: Beer's law and the Srinivasan-Smrtic-Babu (SSB) model [Shin, et al., 2007], which are based on pure photochemical and combination of photochemical and photothermal mechanisms respectively. The two formulae are similar except that SSB's adds a photothermal part to Beer's model where *L*, *β*, f and *f*th are the etching depth per laser pulse, coefficient of absorption (cm-1), laser fluence per pulse (J/cm2) and threshold fluence (J/cm2) respectively.

$$L = \,^1\!/\_{\beta} \ln \left( ^f \!/ \_{f\_{\rm th}} \right) \, for \, f > f\_{\rm th} \tag{1}$$

$$L = \,^1\!/\_{\mathcal{B}} \ln \left( \,^{f} \slash\_{f\_{\text{th}}} \right) + \,^j\_{\text{phot}} \wedge \text{thermal} \; for \, f > f\_{\text{th}} \tag{2}$$

#### **2.1.1 Beam profile**

The most common laser beam profile is the Gaussian beam (TEM00 or fundamental mode) schematically shown in Figure 2a. Its beam intensity variation can be described according to equation 3, where *I*0 = *I*max = intensity at the centre of the profile, *I* is the intensity at any other point, and *r* is the radius of the beam taken at a point where the beam axis intensity has fallen to 1 �� ⁄ of its maximum. Although this Gaussian profile is better than and preferred to higher order modes, its intensity variation is still a source of concern in laser material processing and particularly in laser ablation. For this reason, a modified version which is thought to improve the tapering of the beam profile - is generated with uniform intensity across the entire profile similar, in principle, to that shown in Figure 2b. This is described as a 'top-hat' (or 'flat-top') profile perhaps due to the 'flatness' of the top of the profile. As shown in Figure 2c, a top-hat profile is obtained from its Gaussian counterpart by taking the energy from the weak intensity region, where beam intensity distribution is lower than 1 �� ⁄ (i.e. 13.5 %) of the centre and folding it back into the region within the beam waist. A point should be made here: saying that a laser operates in a single mode e.g. TEM00, simply means that this is the dominant mode of operation just like a given wavelength implies the fundamental (i.e. dominant) wavelength of operation.

$$I = \, \_0e^{-("/r\_0")^2} \tag{3}$$

Fig. 2. Typical laser bean profile (a) Gaussian beam profile, (b) overlapping of Gaussian profile to generate 'top-hat', and (c) 'Top-hat' beam profile.

#### **2.1.2 Ablation threshold**

112 Micromachining Techniques for Fabrication of Micro and Nano Structures

influenced by numerous other factors (e.g. thermal diffusion, absorption saturation, surrounding medium, etc.) such that the measured ablation depths seldom agree with these predictions; this necessitates more complex 'models' often based on these two quantities [Tseng, et al., 2007]. Equations 1 & 2 provide two often referenced mathematical representations: Beer's law and the Srinivasan-Smrtic-Babu (SSB) model [Shin, et al., 2007], which are based on pure photochemical and combination of photochemical and photothermal mechanisms respectively. The two formulae are similar except that SSB's adds a photothermal part to Beer's model where *L*, *β*, f and *f*th are the etching depth per laser pulse, coefficient of absorption (cm-1), laser fluence per pulse (J/cm2) and threshold fluence

� � 1

�� �� ��

implies the fundamental (i.e. dominant) wavelength of operation.

profile to generate 'top-hat', and (c) 'Top-hat' beam profile.

� � 1

�� �� ��

The most common laser beam profile is the Gaussian beam (TEM00 or fundamental mode) schematically shown in Figure 2a. Its beam intensity variation can be described according to equation 3, where *I*0 = *I*max = intensity at the centre of the profile, *I* is the intensity at any other point, and *r* is the radius of the beam taken at a point where the beam axis intensity has fallen to 1 �� ⁄ of its maximum. Although this Gaussian profile is better than and preferred to higher order modes, its intensity variation is still a source of concern in laser material processing and particularly in laser ablation. For this reason, a modified version which is thought to improve the tapering of the beam profile - is generated with uniform intensity across the entire profile similar, in principle, to that shown in Figure 2b. This is described as a 'top-hat' (or 'flat-top') profile perhaps due to the 'flatness' of the top of the profile. As shown in Figure 2c, a top-hat profile is obtained from its Gaussian counterpart by taking the energy from the weak intensity region, where beam intensity distribution is lower than 1 �� ⁄ (i.e. 13.5 %) of the centre and folding it back into the region within the beam waist. A point should be made here: saying that a laser operates in a single mode e.g. TEM00, simply means that this is the dominant mode of operation just like a given wavelength

� � ������ �� � )�

Fig. 2. Typical laser bean profile (a) Gaussian beam profile, (b) overlapping of Gaussian

��� � � ��� � � ��� (1)

 (3)

��� � � � ������������ ��� � � ��� (2)

(J/cm2) respectively.

**2.1.1 Beam profile** 

The ablation threshold is the point at which the applied energy density is enough to cause ablation either photolytic or pyrolytic. The value of this varies from polymer to polymer depending on the nature and strength of the bonds in the polymer and also on laser wavelength (Tables 1 & 2). An ablation threshold can be obtained from a plot of etch rate against a logarithmic scale of fluence at zero ablation rate [Jackson, et al., 1995; Tseng, et al., 2007]. Zakariyah (2010) obtained a threshold as the x-intercept value on a graph of ablation rate against incident fluence. Irrespective of the base of the logarithmic scale taken, the two approaches are found to produce the same value. Table 2 shows a list of common bonds in polymers with their respective bond energies, which need to be overcome during any laser ablation regardless of the nature of the mechanism. For photochemical ablation, the laser wavelength has to be carefully chosen such that the photon energy obtained from the laser is equal or greater than the bond energy of the polymer to be processed. When working below this threshold, no ablation is expected to occur, however, the chemical properties of the materials are subject to certain changes. Furthermore, operating at well above the threshold can cause or increase the heat-affected zone (HAZ) and debris deposition. The former is due to high energy while the latter is as a result of bombarding the ejected materials. It should be noted that intense bombardment of ejected particles above the ablation zone can retard the ablation rate. This is because the ejected materials might absorb fractions of the incoming beam thus reducing the effective fluence at the ablation zone. Wavelength is one of the factors that determine the thresholds of ablation. For example, the ablation threshold for PMMA (PolyMethyl MethAcrylate) is ~150 mJ/cm2 at 193 nm and ~500 mJ/cm2 at 248 nm – this is a 3-time increase in value between the two wavelengths. The rule-of-thumb for laser ablation of polymers is to have lower threshold fluences for ablation at shorter wavelengths [Pfleging, 2006].


Table 1. Ablation threshold fluence for some selected material [Chen, Y-T., et al., 2005, Jackson, et al., 1995; Meijer, 2004; Pfleging, 2006; Yung, et al., 2000; Zakariyah, 2010; Zeng, et al., 2003].

<sup>1</sup> A threshold of 33.8 mJ/cm2 is reported for PMMA at 193 nm by Chen, Y-T, et al. (2005)


Table 2. Table showing typical bonds in photopolymers and their respective bond energies [Basting, 2005; Crafer & Oakley, 1993; Meijer, 2004; Tseng, et al., 2007].

#### **2.2 Industrial laser – Excimer**

Lasers can be classified based on a number of factors e.g. active medium (solid, liquid and gas), output power (low, medium and high power lasers), excitation method (electrical, optical and chemical), operating mode (continuous wave, pulsed mode and Q-switched output mode), efficiency and applications. CO2, Nd:YAG and excimer lasers, with Ti-Saphire following suit, are the key lasers in material processing due to their relatively high power. These three form a complete laser assembly in PCB (printed circuit board) manufacturing processes. Excimer laser is described here as it is the prominent laser candidate for polymer waveguide fabrication; however, a UV Nd:YAG has recently been reported [Zaakriyah, et al., 2011] as a competitive alternative.

An excimer laser - a commonly used gas laser and the halide of noble gases – obtained its name from the contraction of the term 'EXCIted diMER'. Because a dimer strictly refers to a molecule composed of two similar subunits (ions, monomers, etc.), it is therefore more technical to refer to excimer as 'exciplex' meaning EXCIted comPLEX. The wavelengths of excimer lasers vary from about 190 nm (deep UV) to 350 nm (near UV)2 (Figure 3) but ArF, KrF and XeCl are the most commonly used. F2 (λ = 157 nm) laser is sometimes classified as a gas laser and sometimes as an excimer laser as implied in [Basting, et al., 2002; Tseng, et

Fig. 3. A graph of photon energy (eV) against excimer laser wavelengths.

<sup>2</sup> Basting, et al., (2002) put the range between 126 nm and 660 nm (visible region).

**Group Bond Energy (eV) Group Bond Energy (eV)** 


Table 2. Table showing typical bonds in photopolymers and their respective bond energies

Lasers can be classified based on a number of factors e.g. active medium (solid, liquid and gas), output power (low, medium and high power lasers), excitation method (electrical, optical and chemical), operating mode (continuous wave, pulsed mode and Q-switched output mode), efficiency and applications. CO2, Nd:YAG and excimer lasers, with Ti-Saphire following suit, are the key lasers in material processing due to their relatively high power. These three form a complete laser assembly in PCB (printed circuit board) manufacturing processes. Excimer laser is described here as it is the prominent laser candidate for polymer waveguide fabrication; however, a UV Nd:YAG has recently been

An excimer laser - a commonly used gas laser and the halide of noble gases – obtained its name from the contraction of the term 'EXCIted diMER'. Because a dimer strictly refers to a molecule composed of two similar subunits (ions, monomers, etc.), it is therefore more technical to refer to excimer as 'exciplex' meaning EXCIted comPLEX. The wavelengths of excimer lasers vary from about 190 nm (deep UV) to 350 nm (near UV)2 (Figure 3) but ArF, KrF and XeCl are the most commonly used. F2 (λ = 157 nm) laser is sometimes classified as a gas laser and sometimes as an excimer laser as implied in [Basting, et al., 2002; Tseng, et

222 (KrCl) 193 (ArF)

0 50 100 150 200 250 300 350 400

**Laser wavelength (nm)**

351 (XeF) 308 (XeCl)

282 (XeBr) 248 (KrF)

C = C 7.0 O-H 4.5 C = O 6.7, 4.2 H-H 4.6 Si-Si, Cl-Cl 1.8 – 3 O-O 5.1 C-H 3.5 C-C 6.2 C-N, C-C 3 – 3.5 C-O 11.2

[Basting, 2005; Crafer & Oakley, 1993; Meijer, 2004; Tseng, et al., 2007].

reported [Zaakriyah, et al., 2011] as a competitive alternative.

Fig. 3. A graph of photon energy (eV) against excimer laser wavelengths.

<sup>2</sup> Basting, et al., (2002) put the range between 126 nm and 660 nm (visible region).

**2.2 Industrial laser – Excimer** 

0

1

2

3

4

**Photon energy (eV)**

5

6

7

al., 2007]. The pulse duration and repetition rate are in the ranges of 5 – 50 ns and 1 – 100 Hz respectively.

Since its discovery and introduction into the market in 1970 and 1977 respectively, the excimer laser has turned out to be a multi-purpose, multi-featured laser with increasing market shares in industrial and medical applications. Its first commercially available product from Lamda Physik is called EMG 500 [Basting, et a1., 2002]. Although other lasers such as YAG and CO2 lasers are also extensively used in High Density Interconnection (HDI) technology, the excimer laser ablation is indispensable when it comes to 'fine' finish micro- and nano-fabrications. This is particularly true for hard and delicate materials. This is largely due to its wavelength, pulse duration, and of course its pulse energy allowing for what is generally termed as a 'cold ablation' process. The excimer laser also excels others in its ability to 'mask-project' patterns, using stencil or metal-on quartz masks [Tseng, et al., 2007], on to a sample with a minimal HAZ. The minimal HAZ is argued to be due to the short interaction between the laser beam and the material. In addition, the short pulse duration of the excimer is also a contributing factor. Nevertheless, picosecond and femtosecond lasers are now available today. These classes of lasers are designed to further reduce the HAZ. They are also characterized by higher etch rate, strong absorption by the material, improved surface roughness and lower ablation thresholds [Li, L., et al., 2011; Sugioka, et al., 2003].

These aforementioned features of the excimer laser have attracted and favoured its use not only for polymers [Wei & Yang, 2003] but also with other materials such as ceramics [Ihlemann, 1996], glasses [Tseng, et al., 2007] and silicon [Li, J. & Ananthasuresh, 2001] which are often hard to machine. Besides, excimer lasers are now used for surface modification of various materials. Pfleging, et al. (2006) have used excimer at fluences below the ablation threshold to fabricate single mode optical waveguides in PMMA similar to that employed using CO2 laser in [Ozcan, 200 8]. Thomas, et al. (1992) also used an excimer laser to effect changes to the chemical structures of materials (polymer and ceramic) with potential application in enhanced material adhesion and surface wettability among others.

#### **3. Polymer waveguide fabrication for optical interconnect on PCB**

#### **3.1 Optical Interconnects (OI)**

The miniaturisation in consumer electronics, dictated by the rise in demand for more features and the change in the manufacturing technology, has caused an increase in the data rate on the micro-levels such as backplane, board-to-board, and chip-to-chip. The bottleneck for copper transmission in PCB with high interconnection density and high-frequency is more pronounced at the 10 Gb/s limit where problems such as crosstalk, electromagnetic interference (EMI) and power dissipation, inter alia, cannot be tolerated [Holden, 2003; Offrein, 2008; Shioda, 2007]. To overcome this barrier, optical interconnect – as it has been successfully used for long haul communication - is being considered. The deployment suggested here is not to overhaul traditional copper technology but to create a hybrid electric-optical interconnect.

To address the bottleneck caused by the inherent problems in the copper transmission used in backplanes and boards, the last two decades have witnessed vigorous research input and output from researchers around the world to deploy OI on PCB. Japan, the EU and Asia-Pacific/North America, who led in the microvia technology, are also key figures in the OI deployment [Holden, 2003; Lau, 2000; Shioda, 2007]. Undoubtedly, the cost-effectiveness of OI is a major consideration if it is to be implemented [Huang, et al. 2003]. Hopkins & Pitwon (2007) asserted that at higher bandwidth for current and near future requirements for telecom and datacom systems, the application of OI at the backplane is unavoidable. It was argued that the cost of solving the bottleneck of copper transmission will surpass that of implementing OI at ~ 6.25 Gb/s (Fig. 4). Furthermore, the total power loss, commonly referred to as power budget, is also a consideration and is currently being investigated. It is written in [Uhlig & Robertson, 2005] that a ~20 dB would be an acceptable total loss for an optic link at the backplane; Dangel, et al. (2006) put this at 12 – 15 dB for board-to-board optical link of 30 – 100 cm. Uhlig and Robertson (2005, 2006) argued that at some point along the transmission, optical amplification would be needed for a realistic OI on PCB to be implemented. While optical loss is important, reliability (thermal cycling, athermal aging, high temperature reflow, environment, humidity tests, etc.) is another key characteristic and requirement for the deployment of the polymer waveguide [Dangel, 2006; Hwang, et al., 2010].

#### **Bandwidth**

Fig. 4. Relative cost of copper technologies as compared to optical technologies on PCB [Adapted from Hopkins & Pitwon, 2006].

The two OI approaches under consideration are either unguided or guided; both having their pros and cons. The latter can be further divided into fibre- and polymer-based technologies with silicon-based waveguides also gaining momentum (Figure 5). Current literature reports suggest that a polymer-waveguide is the favoured candidate. This is because: (i) polymers are relatively cheap, (ii) low acceptable loss is achievable with polymer, (iii) they are easily available, and (iv) most importantly, polymer waveguide

deployment [Holden, 2003; Lau, 2000; Shioda, 2007]. Undoubtedly, the cost-effectiveness of OI is a major consideration if it is to be implemented [Huang, et al. 2003]. Hopkins & Pitwon (2007) asserted that at higher bandwidth for current and near future requirements for telecom and datacom systems, the application of OI at the backplane is unavoidable. It was argued that the cost of solving the bottleneck of copper transmission will surpass that of implementing OI at ~ 6.25 Gb/s (Fig. 4). Furthermore, the total power loss, commonly referred to as power budget, is also a consideration and is currently being investigated. It is written in [Uhlig & Robertson, 2005] that a ~20 dB would be an acceptable total loss for an optic link at the backplane; Dangel, et al. (2006) put this at 12 – 15 dB for board-to-board optical link of 30 – 100 cm. Uhlig and Robertson (2005, 2006) argued that at some point along the transmission, optical amplification would be needed for a realistic OI on PCB to be implemented. While optical loss is important, reliability (thermal cycling, athermal aging, high temperature reflow, environment, humidity tests, etc.) is another key characteristic and requirement for the deployment of the polymer waveguide [Dangel, 2006; Hwang, et al.,

> Electrical Interconnection

Fig. 4. Relative cost of copper technologies as compared to optical technologies on PCB

**Bandwidth**

The two OI approaches under consideration are either unguided or guided; both having their pros and cons. The latter can be further divided into fibre- and polymer-based technologies with silicon-based waveguides also gaining momentum (Figure 5). Current literature reports suggest that a polymer-waveguide is the favoured candidate. This is because: (i) polymers are relatively cheap, (ii) low acceptable loss is achievable with polymer, (iii) they are easily available, and (iv) most importantly, polymer waveguide

6.25 Gb/s cross-over point

Optical Interconnection

[Adapted from Hopkins & Pitwon, 2006].

2010].

fabrication which is being considered, is compatible with the standard processes employed in PCB manufacturing such as soldering temperature, Coefficient of Thermal Expansion (CTE) matching, thermal stability and stress during lamination [Tooley, et al., 2001].

Fig. 5. Hierarchical classification of optical data communication system based on medium of transmission.

#### **3.2 Deposition of optical polymer**

The stages involved in laser ablation of a polymer waveguide are typified in Figures 6 and 7. In the first stage, liquid optical polymer is spun on FR4 substrate and subsequently UV cured to form both the lower cladding and the core layers. The samples were then dried in an oven (at 80 0C – 100 0C for about for about 60 minutes for Truemode™ acrylate polymer, ∆n ≈ 0.03 variable @ 850 nm) to ensure they were moisture-free. Laser ablation is carried out in the second stage to machine channels such that a ridge of polymer is left in-between the channels to form the waveguide. For one or more adjacent waveguides, the number of grooves required is equal to (*n+1*), where *n* is the number of adjacent waveguides. Finally, a layer of upper cladding is deposited using spin coating (or any other suitable coating technique) and then UV cured.

A single layer of waveguide fabrication is common as this is currently enough to provide the data rate requirements for OI, but a multilayer waveguide has also been demonstrated [Hendrickx, et al., 2007a, 2007b; Matsuoka, et al., 2010]. Multimode waveguides are also common; dimensions such as 20 µm × 20 µm, 30 µm × 30 µm, 35 µm × 35 µm, 45 µm × 45 µm, 50 µm × 50 µm, 50 µm × 20 µm, 70 µm × 70 µm, 75 µm × 75 µm, 85 µm × 100 µm have already been reported [Albrecht, et al., 2005; Bamiedakis, et al., 2007; Dangel, et al., 2004; Immonen, et al., 2005, 2007; Liang, et al., 2008; Tooley, et al., 2001; Van Steenberge, et al., 2004; Zakariyah, 2009, Zakariyah, et al., 2011]. Two or more adjacent waveguides with a pitch of 250 µm [Albrecht, et al., 2005; Horst, 2009; Hwang, et al., 2010; Kim, et al., 2007; Van Steenberge, et al., 2004] is preferred as it is the pitch used for Vertical Cavity Surface Emitting Lasers (VCSEL) and photodector arrays, but other pitch sizes such as 80 µm [Dangel, et al., 2007], 100 µm [Dangel, et al., 2004] and 125 µm [Matsuoka, et al., 2010; Van Steenberge, et al., 2006] have also been used. Since the optical link required for OI is relatively short, loss due to multimode is acceptable and that alignment between various optical components would be relaxed. However, single mode waveguides is much suitable with silicon-based waveguides due to their high refractive indices, though they still pose alignment challenges [Horst, 2009]. Papakonstantinou, et al. (2008) reported a low cost method of achieving high alignment accuracy.

Fig. 6. Schematic diagram (side view) of the three major stages in the fabrication of optical waveguides by laser ablation.

Fig. 7. (a) Flow diagram of the processes involved in patterning optical polymer waveguides using laser ablation, and (b) Schematic flow diagram showing procedure for depositing optical polymer on an FR4 substrate.

#### **3.3 Laser ablation of polymer waveguides**

Polymer waveguide fabrication for optical-PCB applications has been reported using a number of techniques, and more methods are still emerging. Selviah, et al. (2010) reported the use of four techniques - photolithography, laser direct writing, inkjet printing and laser

relatively short, loss due to multimode is acceptable and that alignment between various optical components would be relaxed. However, single mode waveguides is much suitable with silicon-based waveguides due to their high refractive indices, though they still pose alignment challenges [Horst, 2009]. Papakonstantinou, et al. (2008) reported a low cost

Fig. 6. Schematic diagram (side view) of the three major stages in the fabrication of optical

Fig. 7. (a) Flow diagram of the processes involved in patterning optical polymer waveguides using laser ablation, and (b) Schematic flow diagram showing procedure for depositing

Polymer waveguide fabrication for optical-PCB applications has been reported using a number of techniques, and more methods are still emerging. Selviah, et al. (2010) reported the use of four techniques - photolithography, laser direct writing, inkjet printing and laser

method of achieving high alignment accuracy.

waveguides by laser ablation.

optical polymer on an FR4 substrate.

**3.3 Laser ablation of polymer waveguides** 

ablation - in a flagship entitled '*Integrated Optical and Electronic Interconnect PCB Manufacture -* OPCB'. However, excimer laser ablation of optical waveguides is an emerging and competitive approach as it involves fewer steps when compared to others with great flexibility in pattern design. Furthermore, laser micromachining is currently being used for the drilling of vias for blind, buried and through holes in PCB manufacturing making it a more suitable choice when compatibility issues are taken into consideration. The key feature of this class of laser i.e. excimer is its wavelength and pulse duration. The latter reduces the degree of thermal diffusivity while the former is a key to high-energy intensity, high resolution and absorptivity of the laser beam not only in the polymer but also in tough materials such as glass [Tseng, et al., 2007]. The pulse duration of excimer laser is of significance when it comes to quality because shorter pulse width lasers give better machined quality though it is a costly task quality though it is a costly task [Chen, X. & Liu, 1999]; it also helps in reducing the ablation threshold [Ihlemann, 1996]. In fact most of the close competing lasers, for example YAGs and Ti-Sapphire, are found to operate in the UV regions and/or with very short pulse duration, thus intensifying competition.

The suitability of a UV laser (e.g. excimer) for a photochemical ablation over any other laser operating in the IR (or visible) region of wavelengths, such as CO2, could be demonstrated as follows. The photon energy is given by ܧ ൌ ݄ߥ*,* which is inversely proportional to its wavelength, thus a CO2 laser operating at 10.6 µm will produce an energy more than 40 times less than that produced by a 248 nm KrF laser. Obviously, this is not in the order of magnitude of the energies for chemical bond scission of typical polymers, usually between 3 – 8 eV [Tseng, et al., 2007]. Increasing the number of pulses to match the required bond energy will merely result in a cumulative heat effect on the polymer surface. It is thus clear that excimer lasers have the right order of photon energy to athermally ablate polymers, while on the other hand, IR laser sources have photon energies much lower than 3 eV causing the dominance of a thermal mechanism. Therefore, in principle using the aforementioned assertion, laser of a maximum wavelength of 414 nm is required in order to photochemically ablate a polymer material with a bond energy of 3 eV. There would be a shift in the dominance of the mechanism by changing the wavelength of the laser source. For example, a shorter wavelength e.g. 355 nm would guarantee or increase the dominance of a photochemical process. On the other hand, a longer wavelength e.g. 1064 nm in the IR would not only reduce the dominance of photochemical but also initiate thermal process for the same polymer.

Fig. 8. Samples machined at 30 Hz, 50 shots per point and 3.6 mm/min with different fluences of 80 mJ/cm2.

In Figure 8a above, a straight, shallow track is machined in an acrylate-based photopolymer while in Figure 8b above, two parallel tracks were etched leaving a ridge that constitutes a waveguide. In this case no upper cladding (as per stage 2, Figure 6) is applied. Sometimes, the ridge or waveguide may not be continuous. To examine this, light can be passed into one end of the guide for possible detection at the other end i.e. backlighting, as shown in Figure 9 where a single multimode waveguide of 50 µm × 35 µm and 60 mm long was illuminated from behind using a Flash™200 optical measuring device. The structure was made by ablating ~200 µm wide grooves in Truemode™ polyacrylate. Furthermore, waveguides can be 'crossed' at 90 degree (Figure 10) or other shapes may be desired. While excimer laser ablated waveguides is favoured, UV Nd:YAG (λ = 355 nm) [Van Steenberge, et al., 2004; Zakariyah, et al., 2011] and 10.6 µm CO2 [Zakariyah, 2010] have been demonstrated as promising candidates especially for mass production at a low-cost.

Fig. 10. Waveguides crossed over at 90 degree to each other machined at 100 mJ/cm2, 45 shots per point, 3.3 mm/min, 25 Hz and a single pass showing (a) a schematic diagram, and (b) an SEM image of an initial trial.

#### **3.4 Integrated mirror fabrication**

Optical signals on PCBs need to be routed to different parts of a device, such as between the boards of a backplane, if OI is to be fully utilised. Various proposals have been made on how to direct signals out of the plane of the board. These include 45-degree ended optical connection rods, microlens, 900-bent fibre connectors, 450-ended blocks, 450-ended I-shape

In Figure 8a above, a straight, shallow track is machined in an acrylate-based photopolymer while in Figure 8b above, two parallel tracks were etched leaving a ridge that constitutes a waveguide. In this case no upper cladding (as per stage 2, Figure 6) is applied. Sometimes, the ridge or waveguide may not be continuous. To examine this, light can be passed into one end of the guide for possible detection at the other end i.e. backlighting, as shown in Figure 9 where a single multimode waveguide of 50 µm × 35 µm and 60 mm long was illuminated from behind using a Flash™200 optical measuring device. The structure was made by ablating ~200 µm wide grooves in Truemode™ polyacrylate. Furthermore, waveguides can be 'crossed' at 90 degree (Figure 10) or other shapes may be desired. While excimer laser ablated waveguides is favoured, UV Nd:YAG (λ = 355 nm) [Van Steenberge, et al., 2004; Zakariyah, et al., 2011] and 10.6 µm CO2 [Zakariyah, 2010] have been demonstrated

Fig. 9. Excimer laser ablation of optical waveguide showing cross-section of a 50 µm x 35 µm

Fig. 10. Waveguides crossed over at 90 degree to each other machined at 100 mJ/cm2, 45 shots per point, 3.3 mm/min, 25 Hz and a single pass showing (a) a schematic diagram, and

Optical signals on PCBs need to be routed to different parts of a device, such as between the boards of a backplane, if OI is to be fully utilised. Various proposals have been made on how to direct signals out of the plane of the board. These include 45-degree ended optical connection rods, microlens, 900-bent fibre connectors, 450-ended blocks, 450-ended I-shape

as promising candidates especially for mass production at a low-cost.

multimode waveguide in Truemode™.

(b) an SEM image of an initial trial.

**3.4 Integrated mirror fabrication** 

waveguides, optical coupler and microprism. These aforementioned concepts of out-ofplane coupling utilises blade cutting, laser ablation, moulding, dicing and RIE among others with each having its benefits and limitations [Byung, et al., 2004; Cho, et al., 2005; Cho, 2005; Teck, et al., 2009; Van Steenberge, et al., 2006]. To improve the coupling efficiency, Glebov, et al.(2005) proposed a curved micro-mirror instead of the flat 45-degree commonly employed.

Coupling light in and out of the polymer waveguides could be achieved by relying on the air/vacuum refractive index which is capable of causing total internal reflection (TIR) (Figure 11a) at this interface as used in [Teck, et al., 2009], but this can be difficult in real application because: (i) a vacuum is not guaranteed in a typical electronics assembly, (ii) air content and temperature are subjects of the environmental conditions, and (iii) even if air refractive index is guaranteed to be constant, air reflectivity is not efficient for coupling. For these reasons, end facets of mirrors are coated with a metal to improve its reflectivity and for a good surface finish. The chosen deposition technique depends largely on the sample to be coated and adhesion adhesion inter alia. For example, the authors of [Glebov, et al., 2005] used sputtering to deposit a thin layer of gold on the mirror surface before filling the trench with upper cladding; similar process was used for laser ablated mirror [Van Steenberge, et al., 2006]. It should be noted that there is a potential of light scattering or reflection at the clad-core interface [Hendrickx, et al., 2007a, 2007b]. Furthermore, the inaccuracy of the fabricated mirror angle can cause a significant reduction in the amount of light emanating from the core-clad exit of the waveguide to that reaching the metallised mirror surface thus affecting the coupling efficiency; a short path with a minimum angle deviation can mitigate this challenge.

Fig. 11. Mirror fabrication schemes (a) TIR is used to deflect incoming signal out of the waveguide at the waveguide-air interface, and (b) light is coupled from a metal deposited at the surface of mirror trench which is the trench filled with cladding material.

While the out-of-plane coupling scheme is gaining impetus, there is no doubt that in-plane lateral routing of optical signals is also needed. A typical system architecture would require routing of signals not only from one layer to the other, but also within a layer; the latter would be extremely important if OI is extended to the board (and even chip) level as various roadmaps have laid down this possibility. Figure 12 is a schematic diagram of the in-plane mirror fabrication, which can be used to couple light between multiple components in the same layer. With this design, an effective turning angle of zero, 90-degree and multiples of 90-degrees are possible; a scheme demonstrated in [Glebov & Lee, M-G., 2006; Lamprecht, et al. 2009; Zakariyah, 2010]. Glebov & Lee, M-G. (2006) placed a vertical terminator at the end of the waveguide to form the mirror with a loss of 0.5 – 1.0 dB recorded for this approach; however, Zakariyah (2010) employed excimer laser ablation to manufacture the 450 lateral mirrors. It is argued [Zakariyah, 2010] that laser ablation is a more suitable fabrication technique as it allows for both the waveguide and the mirrors to be manufactured using a single process on the same system. The laser ablation approach was also used for out-of plane mirror coupling such as in [Teck, et al., 2009] for 3D out-of-plane coupling.

Fig. 12. Proposed 2D in-plane scheme showing (a) 45-degree in-plane coupling mirror design with 180-degree effective turning angle, and (b) 45-degree in-plane coupling mirror design with zero-degree effective turning angle.

#### **3.5 Loss measurement**

Signals launched at one end of an optical waveguide are not ideally identical in many cases to those arriving at the receiving end, either due to attenuation (change in amplitude) or distortion (change in waveform). These losses (propagation, coupling, angular misalignment, etc.) are quantified using a logarithmic unit called decibels (dB) using equation 4, where *P*1 and *P*2 represent the input and output power respectively. For a loss, it is a negative dB while a positive value indicates a power gain, usually obtained in amplifiers or amplification circuits. Sometimes the negative sign is omitted but replaced with 'loss' to mean attenuation in signal.

$$\text{Loss (dB)} = 10 \log\_{10} \frac{p\_2}{p\_1} \tag{4}$$

Reports have shown different values for the waveguide propagation loss depending mainly on the materials and the fabrication process used; Teck, et al. (2009) put the loss values in the range of 0.05 – 0.6 dB/cm, and the loss at a datacom (λ = 840 nm) in the range of 0.01 dB/cm – 0.8 dB/cm was given in [Holden, 2003]. Propagation loss of 0.24 dB/cm was recorded for a single mode waveguide in polyetherimide at 830 nm using laser ablation [Eldada, 2002]. A polymer waveguide manufactured by excimer laser ablation produced a propagation loss of 2 dB /cm at 1550 nm [Jiang, et al., 2004]; this high loss was attributed to the sidewall roughness of the guides. At 850 nm, propagation loss between 0.04 dB/cm and 0.2 dB/cm and 0.04 dB/cm and 0.18 dB/cm are reported for flexible and rigid waveguides respectively measured for different polymers [Shioda, 2007]. Table 3 is a list, though not exhaustive, of recent optical waveguide reports. While propagation loss is dependent on the waveguide characteristics, it is possible to reduce the insertion loss by reducing the coupling efficiency. One way of achieving this is through a good alignment between the coupling device and the waveguide. Jiang, et al. (2004) proposed an excimer laser ablation of the end facets for efficient coupling of light which in turn can reduce the loss.

mirrors. It is argued [Zakariyah, 2010] that laser ablation is a more suitable fabrication technique as it allows for both the waveguide and the mirrors to be manufactured using a single process on the same system. The laser ablation approach was also used for out-of

plane mirror coupling such as in [Teck, et al., 2009] for 3D out-of-plane coupling.

Fig. 12. Proposed 2D in-plane scheme showing (a) 45-degree in-plane coupling mirror design with 180-degree effective turning angle, and (b) 45-degree in-plane coupling mirror

Signals launched at one end of an optical waveguide are not ideally identical in many cases to those arriving at the receiving end, either due to attenuation (change in amplitude) or distortion (change in waveform). These losses (propagation, coupling, angular misalignment, etc.) are quantified using a logarithmic unit called decibels (dB) using equation 4, where *P*1 and *P*2 represent the input and output power respectively. For a loss, it is a negative dB while a positive value indicates a power gain, usually obtained in amplifiers or amplification circuits. Sometimes the negative sign is omitted but replaced with 'loss' to mean attenuation in signal.

�����(��) = 10 log��

facets for efficient coupling of light which in turn can reduce the loss.

Reports have shown different values for the waveguide propagation loss depending mainly on the materials and the fabrication process used; Teck, et al. (2009) put the loss values in the range of 0.05 – 0.6 dB/cm, and the loss at a datacom (λ = 840 nm) in the range of 0.01 dB/cm – 0.8 dB/cm was given in [Holden, 2003]. Propagation loss of 0.24 dB/cm was recorded for a single mode waveguide in polyetherimide at 830 nm using laser ablation [Eldada, 2002]. A polymer waveguide manufactured by excimer laser ablation produced a propagation loss of 2 dB /cm at 1550 nm [Jiang, et al., 2004]; this high loss was attributed to the sidewall roughness of the guides. At 850 nm, propagation loss between 0.04 dB/cm and 0.2 dB/cm and 0.04 dB/cm and 0.18 dB/cm are reported for flexible and rigid waveguides respectively measured for different polymers [Shioda, 2007]. Table 3 is a list, though not exhaustive, of recent optical waveguide reports. While propagation loss is dependent on the waveguide characteristics, it is possible to reduce the insertion loss by reducing the coupling efficiency. One way of achieving this is through a good alignment between the coupling device and the waveguide. Jiang, et al. (2004) proposed an excimer laser ablation of the end

�� ��

(4)

design with zero-degree effective turning angle.

**3.5 Loss measurement** 




#### **4. Conclusion**

In this chapter, the author presented the need for OI for both intra- and inter-board applications due to prevailing limitations with electrical interconnection on the PCB despite the various rectifying measures being considered. For successful implementation of OI, the

60 µm

30 µm × 30 µm

50 µm × 50 µm (250 µm pitch) & 35 µm × 35 (100 µm pitch)

≤ 50 µm × 10 µm multimode

45 µm × 45 µm

blade - 0.05 dB/cm

70 µm ×

50 µm

20 µm

40 µm × 40 µm

In this chapter, the author presented the need for OI for both intra- and inter-board applications due to prevailing limitations with electrical interconnection on the PCB despite the various rectifying measures being considered. For successful implementation of OI, the

**Dimension Loss Reference** 



0.5 – 0.8 dB at 850 nm


at 850 nm - Kopetz,

at 850 nm - Albrecht,

et al., 2004

et al., 2004

et al., 2006

et al., 2004.

et al., 2008

et al., 2005

et al., 2005

et al., 1998


0.3 – 0.7 dB Hikita,


Glebov, et al., 2005, 2007.

~ 0.1 dB/cm at 850 nm

0.05 dB/cm at 850 nm

0.035 – 0.05 dB/cm at 850 nm & 0.12 dB/cm at 990 nm

1.4 ± 0.5 dB/cm at 850 nm

70 µm - - Liang,

50 µm - - Chen, Y-M,

0.15 dB/cm

0.03 – 0.05 dB/cm at 850 nm

< 0.02 dB/cm at 830 nm

**Material Process Waveguide** 

(core) Hot-embossing 60 µm ×

(WGs) & Microdicing (mirrors)

acrylate polymer Photolithography

ORMOCER -

Fluorinated acrylate

SU-8 (NANOTM

Deuterated PMMA (core) & UV-cured epoxy resin (cladding)

**4. Conclusion** 

polymer

Truemode™ UV Nd:YAG

Polysiloxane Casting + Doctor

Photolithography

Laser Ablation

Soft molding (core) & spincoating (cladding)

Epoxy resin Spin-coating 50 µm ×

Siloxane polymer Photolithography 50 µm ×

SU-8-50) Photolithography 50 µm ×

& RIE

Spin coating, photolithography

Table 3. Optical polymer waveguide fabrication techniques

UV curable resins

Photopatternable

Photosensitive

polymer

 **Waveguide Mirror** 

following are needed: materials that would be compatible with PCB manufacturing procedures; fabrication techniques that would be easy, cost effective and efficient from the production point of view; and finally materials / waveguides that would satisfy the optical power budget requirement. A polymer-based waveguide is favoured for this technology primarily due to its low cost and compatibility. Multimode polymer waveguides with typical dimensions 50 ± 20 µm square are common as it relaxes alignment constrain thus lowering coupling. While various fabrication techniques have been reported with new still emerging procedures, laser ablation is a preferred approach since it is the technique currently being used for the drilling of µvias, which makes it a much compatible candidate. Furthermore, for the fabrication of integrated mirrors, either in-plane or out-of-plane, laser ablation using an excimer laser for example, is a much suitable option for this due to its excellent laser matter interaction, resulting in clean removal at micro-level scales. In addition, the mask projection available with excimer laser makes it possible for complex features to be easily defined. Although the cost and speed of excimer laser could be an issue from the production point of view at this stage of the deployment, other lasers such as UV Nd:YAG and CO2 can offer both prototyping and mass production opportunities as it has been demonstrated, thus making laser ablation an all-encompassing technique meeting required production speed, cost, efficiency and quantity. In light of this, the chapter also provides an overview of laser technology for material processing and in particular for polymer waveguide fabrication.

#### **5. Acknowledgment**

The authors wish to thank Khadijah Olaniyan, Abdul Lateef Balogun, Mayowa Kassim Aregbesola and Witold Kandulski for helpful discussions.

#### **6. References**


Chen, Y-T, Naessens, K., Baets, R., Liao, Y. & Tseng, A. (2005). Ablation of transparent

Chen, Y-M., Yang, C-L., Cheng, Y-L., Chen, H-H., Chen, Y-C., Chu, Y. & Hsieh, T-E. (2005).

*Electronic Components and Technology Conference*, 2005, Vo.2, pp. 1739-1743. Cho, M.H. (2005). High-coupling-efficiency optical interconnection using a 90-bent fiber

Crafer, R. & Oakley, P.J. (1993), *Laser processing in manufacturing*, Chapman and Hall, ISBN:

Dangel, R., Bapst, U., Berger, C., Beyeler, R., Dellmann, L., Horst, F., Offrein, B. & Bona, G-

Dangel, R., Beyeler, R., Horst, F., Offrein, B.J., Sicard, B., Moynihan, M., Knudsen, P. &

Do-Won Kim, In-Kui Cho, Seung Ho Ahn & Hyo-Hoon Park (2007). 5-Gb/s Chip-to-chip

Eldada, L. (2002). Polymer integrated optics: Promise vs. practicality, *Proceedings of SPIE* - The International Society for Optical Engineering, Vol. 4642, pp. 11-22. Glebov, A. L., Roman, J., Lee, M.G. & Yokouchi, K.(2005). Optical interconnect modules

Glebov, A.L. & Lee, M.G. (2006). 3D Routing on Optical Boards, *Proceedings of IEEE 19th* 

Glebov, A.L., Lee, M.G. & Yokouchi, K. (2007). Integration technologies for pluggable

*of Photo-optical Instrumentation Engineers*, vol. 46(1), pp. 15403-1 – 15403-10 Hendrickx, N., Van Erps, J., Van Steenberge, G., Thienpont, H. & Van Daele, P. (2007a).

Hikita, M., Yoshimura, R., Usui, M., Tomaru, S. & Imamura, S. (1998). Polymeric optical

Holden, H.T. (2003). The developing technologies of integrated optical waveguides in

*Electro-Optics - Pacific Rim*, 2007. CLEO/Pacific Rim 2007, pp. 1-2.

*Engineers Conference*, 2007. OFC/NFOEC 2007, pp 1-3.

*Physics*: Conference Series, Vol. 59, No.1, pp. 118-21.

printed circuits, *Circuit World*, Vol.29, No. 4, pp. 42-50.

No.6, pp. 427-441.

Vol.17, No.3, pp. 690-692.

0412415208, London.

2004, pp. 29 - 30.

1540-1542, Jul. 2005

308.

materials using excimer lasers for photonic applications, *Optical Review*, Vo.12,

10Gbps multi-mode waveguide for optical interconnect, *Proceedings of the 55th*

array connector in optical printed circuit boards, *IEEE Photonics Technology Letters*,

L. (2004). Development of a low-cost low-loss polymer waveguide technology for parallel optical interconnect applications, *Biophotonics/Optical Interconnects and VLSI Photonics/WBM Microcavities*, 2004 Digest of the LEOS Summer Topical Meetings,

Anzures, E. (2007). Waveguide Technology Development based on Temperatureand Humidity-Resistant Low-Loss silsesquioxane Polymer for Optical Interconnects, *Proceedings of Optical Fiber Communication and the National Fiber Optic* 

Optical Interconnection Using Polymeric Waveguides, *Proceedings of Lasers and* 

with fully integrated reflector mirrors, *IEEE Photonics Technology Letters*, vol. 17, pp.

*Annual Meeting on Lasers and Electro-Optics Society*, 2006. LEOS 2006. 2006, pp. 22-23.

backplane optical interconnect systems, *Optical engineering: the journal of the Society* 

Tolerance analysis for multilayer optical interconnections integrated on a printed circuit board, *IEEE Journal of Lightwave Technology*, Vol. 25, No. 9, pp. 2395-2401. Hendrickx, N., Van Steenberge, G., Geerinck, P. & Van Daele, P. (2007b). Laser ablation as

enabling technology for the structuring of optical multilayer structures, *Journal of* 

waveguides for optical interconnections, *Thin Solid Films*, Vol.331, No.1-2, pp. 303-


interconnections, *Optical Fiber Communication (OFC), collocated National Fiber Optic Engineers Conference, 2010 Conference on (OFC/NFOEC*), pp. 1-3.


Meijer, J. (2004). Laser beam machining (LBM), state of the art and new opportunities,

Offrein, B.J. (2008). Optical interconnects and nanophotonics, *Proceedings of IEEE/LEOS Internationall Conference Optical MEMs and Nanophotonics,* 2008, pp. 21-22. Ozcan, L.C. (2008). Fabrication of buried waveguides in planar silica films using a direct CW laser writing technique, *Journal of non-crystalline solids*, Vol.354, No.42, pp. 4833. Papakonstantinou, I., Selviah, D.R., Pitwon, R. & Milward, D. (2008). Low-Cost, Precision,

Pfleging, W. (2006). Excimer laser material processing – state of the art and new approaches

Pitwon, R., Hopkins, K. & Milward, D. (2006). An optical backplane connection system with

Selviah, D.R., Walker, A.C., Hutt, D.A., Wang, K., Mccarthy, A., Fernandez, F.A.,

Shin, B.S., Oh, J.Y. & Sohn, H. (2007). Theoretical and experimental investigations into laser

Shioda, T. (2007). Recent Progress and Potential Markets for Optical Circuit Boards*,* 

Sugioka, K., Obata, K., Midorikawa, K., Hong, M.H., Wu, D.J., Wong, L.L., Lu, Y.F. &

Teck, G.L., Ramana, P.V., Lee, B.S.P., Shioda, T., Kuruveettil, H., Li, J., Suzuki, K., Fujita, K.,

Thomas, D.W. (1992). Surface modification of polymers and ceramics induced by excimer laser-radiation, *Laser ablation of electronic materials*, Vol.4, pp. 221-228.

manufacturing research. *Circuit World*, Vol.36, No.2, pp. 5-19.

*Microelectronics and Photonics,* 2007, Polytronic 2007, pp. 167-169.

*Materials Processing Technology*, Vol.187-188, pp. 260-263.

*Multi-Conference on Wireless and Optical Communications*, pp. 297- 302. Hopkins, K. & Pitwon, R. (2007). Pluaggable Optical Backplane Technology, In: *White Paper,* 

*Engineers Conference, 2010 Conference on (OFC/NFOEC*), pp. 1-3.

Vol.31, No.3, pp. 502-511.

ptical\_Backplane\_2-0.pdf.

Vol.158, No.2-3, pp. 171-178.

No.2, pp. 509-16.

Optical Engineering*,* Vol. 6107.

*Journal of Materials Processing Technology*, Vol.149, No.1-3, pp. 2-17.

interconnections, *Optical Fiber Communication (OFC), collocated National Fiber Optic* 

Self-Alignment Technique for Coupling Laser and Photodiode Arrays to Polymer Waveguide Arrays on Multilayer PCBs, *IEEE Transactions on Advanced Packaging,* 

in microsystem technology, *Proceedings of SPIE -* The International Society for

pluggable active board interfaces, *Proceedings of the Sixth IASTED International* 

*Xyratex Technology*. Accessed : November 2011, Avaialble at : http://www.xyratex.com/pdfs/whitepapers/Xyratex\_white\_paper\_Pluggable\_O

Papakonstantinou, I., Baghsiahi, H., Suyal, H., Taghizadeh, M., Conway, P., Chappell, J., Zakariyah, S., S., Milward, D., Pitwon, R., Hopkins, K., Muggeridge, M., Rygate, J., Calver, J., Kandulski, W., Deshazer, D.J., Hueston, K., Ives, D.J., Ferguson, R., Harris, S., Hinde, G., Cole, M., White, H., Suyal, N., Rehman, H.U. & Bryson, C. (2010). Integrated optical and electronic interconnect PCB

ablation of polyimide and copper films with 355-nm Nd:YVO4 laser, *Journal of* 

*Proceedings of the* 6th International Conference on *Polymers and Adhesives in* 

Chong, T.C. (2003). Advanced materials processing based on interaction of laser beam and a medium, *Journal of Photochemistry and Photobiology A: Chemistry*,

Yamada, K., Pinjala, D. & Shing, J.L.H. (2009). Demonstration of direct coupled optical/electrical circuit board, *IEEE Transactions on Advanced Packaging*, Vol.32,


ablation using a 355 nm wavelength Nd:YAG laser, *IEEE Journal of Lightwave Technology*, 2011 (DOI : 10.1109/JLT.2011.2171318) (Manuscript No: JLT-13362- 2011).

Zeng, D.W., Yung, K.C. & Xie, C.S. (2003). UV Nd:YAG laser ablation of copper: chemical states in both crater and halo studied by XPS, *Applied Surface Science*, Vol. 217, No.1- 4, pp. 170-180.

## **Micro Eletro Discharge Milling for Microfabrication**

Mohammad Yeakub Ali, Reyad Mehfuz, Ahsan Ali Khan and Ahmad Faris Ismail *International Islamic University, Malaysia* 

#### **1. Introduction**

130 Micromachining Techniques for Fabrication of Micro and Nano Structures

Zeng, D.W., Yung, K.C. & Xie, C.S. (2003). UV Nd:YAG laser ablation of copper: chemical

2011).

4, pp. 170-180.

ablation using a 355 nm wavelength Nd:YAG laser, *IEEE Journal of Lightwave Technology*, 2011 (DOI : 10.1109/JLT.2011.2171318) (Manuscript No: JLT-13362-

states in both crater and halo studied by XPS, *Applied Surface Science*, Vol. 217, No.1-

Miniaturization of product is increasingly in demand for applications in numerous fields, such as aerospace, automotive, biomedical, healthcare, electronics, environmental, communications and consumer products. Researchers have been working on the microsystems that promise to enhance health care, quality of life and economic growth. Some examples are micro-channels for micro fuel cell, lab-on-chips, shape memory alloy 'stents', fluidic graphite channels for fuel cell applications, miniature actuators and sensors, medical devices, etc. (Madou, 2002; Hsu, 2002). Thus, miniaturization technologies are perceived as potential key technologies.

One bottleneck of product miniaturization is the lack of simpler and cheaper fabrication techniques. Currently the common techniques are based on silicon processing techniques, where silicon-based materials are processed through wet and dry chemical etching. These techniques are suitable for microelectronics, limited to few silicon-based materials and restricted to simple two dimensional (2D) or pseudo three dimensional (2.5D) planar geometries. Other fabrication processes, such as LiGA (lithography, electroforming and molding), laser, ultrasonic, focused ion beam (FIB), micro electro discharge machining (EDM), mechanical micromilling, etc. are expensive and required high capital investment. Moreover these processes are limited to selected materials and low throughput (Ehmann et al. 2002).

A less expensive and simpler microfabrication technique is sought to produce commercially viable microcomponents. Micro electro discharge (ED) milling, a new branch of EDM, has potential to fabricate functional microcomponents. The influences of micro ED milling process parameters on surface roughness, tool wear ratio and material removal rate are not fully identified yet. Therefore, modeling of ED milling process parameters for surface roughness, tool wear ratio and material removal rate are necessary.

#### **1.1 Micro electro discharge machining**

EDM has been successfully used for micromachining with high precision regardless of the hardness of work piece material. It uses the removal phenomenon of electrical discharges in a dielectric fluid. Two conductive electrodes, one being the tool and the other the workpiece, are immersed in a liquid dielectric. A series of voltage pulses are applied between the electrodes, which are separated by a small gap. A localized breakdown of the dielectric occurs and sparks are generated across the inter-electrode gap, usually at regions where the local electric field strength is highest. Plasma channels towards workpiece are formed during the discharge and high speed electrons come into collision with the workpiece. Each spark erodes a small amount of work material by melting and vaporizing from the surface of both the electrodes. The momentary local plasma column temperature can reach as high as 40,000 K (DiBitonto et al., 1989).

Fig. 1.1. Schematic of EDM principle (Kim et al., 2005)

This high temperature causes the melting and vaporization of the electrode materials; the molten metals are evacuated by a mechanical blast, resulting in small craters on both the tool electrode and work materials. It is understood that the shock waves, electromagnetic and electrostatic forces involved in the process are responsible for ejection of the molten part (debris) into the dielectric medium. The repetitive impulse together with the feed movement (by means of a servo mechanism) of the tool electrode towards the workpiece enables metal removal along the entire surface of the electrodes. Figure 1.1 shows the schematic of EDM principle.

Production of micro-features using EDM pays significant attention to the research community. This is due to its low set-up cost, high accuracy and large design freedom. Compared to lithography based miniaturization, micro EDM has the clear advantages in fabricating complex 3D shapes with higher aspect ratio (Lim et al., 2003). Moreover, all conductive materials regardless of hardness can be machined by EDM. Conventional EDM is especially useful to produce molds and dies. Micro EDM is now basically focused on the fabrication of miniaturized product, like molds and dies, with greater surface quality. In this endeavor new CNC systems and advanced spark generators are found as great assistance (Pham et al., 2004). Current micro EDM technology used for manufacturing micro-features can be categorized into four different types:


Precision micro holes required for gas and liquid orifices in aerospace and medical applications, pinholes for x-ray and nuclear fusion measurements, ink-jet printer nozzles and electron beam gun apertures can be fabricated with high precision accuracy and with

occurs and sparks are generated across the inter-electrode gap, usually at regions where the local electric field strength is highest. Plasma channels towards workpiece are formed during the discharge and high speed electrons come into collision with the workpiece. Each spark erodes a small amount of work material by melting and vaporizing from the surface of both the electrodes. The momentary local plasma column temperature can reach as high

This high temperature causes the melting and vaporization of the electrode materials; the molten metals are evacuated by a mechanical blast, resulting in small craters on both the tool electrode and work materials. It is understood that the shock waves, electromagnetic and electrostatic forces involved in the process are responsible for ejection of the molten part (debris) into the dielectric medium. The repetitive impulse together with the feed movement (by means of a servo mechanism) of the tool electrode towards the workpiece enables metal removal along the entire surface of the electrodes. Figure 1.1 shows the schematic of EDM

Production of micro-features using EDM pays significant attention to the research community. This is due to its low set-up cost, high accuracy and large design freedom. Compared to lithography based miniaturization, micro EDM has the clear advantages in fabricating complex 3D shapes with higher aspect ratio (Lim et al., 2003). Moreover, all conductive materials regardless of hardness can be machined by EDM. Conventional EDM is especially useful to produce molds and dies. Micro EDM is now basically focused on the fabrication of miniaturized product, like molds and dies, with greater surface quality. In this endeavor new CNC systems and advanced spark generators are found as great assistance (Pham et al., 2004). Current micro EDM technology used for manufacturing micro-features

a) Die-sinking micro-EDM, where an electrode with micro-features is employed to

b) Micro-ED drilling, where micro-electrodes (of diameters down to 5–10 µm) are used to

c) Micro-ED milling, where micro-electrodes (of diameters down to 5–10 µm) are employed to produce complex 3D cavities by adopting a movement strategy similar to

d) Micro-wire EDM, where a wire of diameter down to 20 µm is used to cut through a

Precision micro holes required for gas and liquid orifices in aerospace and medical applications, pinholes for x-ray and nuclear fusion measurements, ink-jet printer nozzles and electron beam gun apertures can be fabricated with high precision accuracy and with

as 40,000 K (DiBitonto et al., 1989).

principle.

Fig. 1.1. Schematic of EDM principle (Kim et al., 2005)

can be categorized into four different types:

'drill' micro-holes in the workpiece.

that in conventional milling.

conductive workpiece.

produce its mirror image in the workpiece.

surface roughness less than 0.1 µm using the micro EDM process (Dario et al., 1995). Although micro EDM plays an important role in the field of micromachining, it has disadvantages such as high tool-electrode wear ratio and low *MRR*. The wear of electrode must be compensated either by changing the electrode or by preparing longer electrode from the beginning or fabricating the electrode in situ for further machining (Asad et al., 2007).

#### **1.1.2 Micro electro discharge milling**

In micro ED milling the material is eroded by non-contact thermo-electrical process, where a series of discrete sparks occur between the workpiece and the rotating tool electrode. The workpiece and tool are immersed in a dielectric fluid. The dielectric fluid is continuously fed to the machining zone to convey the spark and flush away the eroded particles. The work feeding system of micro ED milling is similar to mechanical end milling process. Like mechanical end milling, here the workpiece is fed to the tool electrode while the tool is in rotation. The tool movement is controlled numerically to achieve the desired threedimensional shape with high accuracy. Figure 1.2 shows the schematic of micro ED milling.

Fig. 1.2. Schematic of micro ED milling (Murali and Yeo, 2004)


Table 1.1. Micro ED milling process parameters

#### **1.1.3 Process parameters**

Micro ED milling has a number of process parameters, which are influential on the machining performance. Depending upon the circuit type used in the machine, the importance of parameters is also varied. Table 1.1 shows different micro ED milling process parameters. Some researches were conducted to optimize different EDM process parameters for *Ra, Ry, TWR* and *MRR* (Puertas and Luis, 2004; Lin et al., 2006). But almost all of them were on conventional macro EDM. Because of the stochastic thermal nature of the EDM process, it is difficult to explain all of those effects fully. The full understanding of micro ED milling process parameters are yet to be developed because the process itself is a new branch of manufacturing. For electric discharge the discharge energy (E) generated in the discharge circuit can be expressed roughly as follows:

$$E = \frac{1}{2}(C+C^\cdot)V^2\tag{1.1}$$

Where,

C = capacitance of capacitors,

C' = total stray capacitance,

V = discharge voltage

According to the equation (1.1), there are two ways to reduce the discharge energy: by reducing the discharge voltage or by reducing the total capacitance (C + C' ). However, low voltage results in an unstable discharge. Therefore, the total capacitance should be controlled primarily. The total stray capacitance (C') in the circuit plays a significant role here. The following subsections discuss the influence of different micro ED milling parameters on surface roughness, tool wear and material removal rate.

#### **1.1.4 Surface roughness**

Variation of discharge energy affects surface roughness. It is experimentally identified that with the increase in discharge energy either by increasing voltage or capacitance, *Ra* increased in die sinking EDM, wire EDM (WEDM) and WEDG (Uhlmann et al., 2005; Han et al., 2006; Chiang, 2007). A large discharging energy usually causes violent sparks and results a deeper erosion crater on the surface. Accompanying the cooling process after the spilling of molten metal, residues remain at the periphery of the crater to form a rough surface (Liao et al., 2004). Theoretical model for *Ry* or *Rmax* was developed for transistorized pulse generator type EDM (Chen and Mahdivian, 2000). It showed that with the increase in current and pulse duration, *Ry* increased. Surface roughness of ceramic due to EDM was also investigated. EDM of boron carbide (B4C) showed that *Ra* and *Ry* increased with pulse duration but decreased with current (Puertas and Luis, 2004). On the other hand, EDM of high speed steel showed that with the increase of current, *Ra* increased and with the increase of pulse duration it decreased (Lin et al., 2006). It was suggested to use low conductive or higher resistive dielectric to ensure low *Ra* and *Ry* (Liao et al., 2004). Use of rotating electrode was helpful for high shape accuracy and higher *MRR* but it caused higher surface roughness (Her and Weng, 2001).

Formation of crack is a challenge in EDM. The presence of high crack density is not desirable because it leads early product failure. Investigation of EDM parameters showed that higher pulse-on duration increased both the average white layer thickness and the induced stress (Lee and Tai, 2003; Ekmekci et al., 2005). These two conditions tend to promote crack formation. When the pulse current was increased, the increase in material removal rate caused a high deviation of thickness of the white layer. Thus with the increase in pulse current, crack formation decreased. It was also verified that lower thermal conductivity of workmaterial caused higher crack formation. There was no analysis reported on the effect of feed rate on machined surface.

#### **1.1.5 Tool wear**

134 Micromachining Techniques for Fabrication of Micro and Nano Structures

Micro ED milling has a number of process parameters, which are influential on the machining performance. Depending upon the circuit type used in the machine, the importance of parameters is also varied. Table 1.1 shows different micro ED milling process parameters. Some researches were conducted to optimize different EDM process parameters for *Ra, Ry, TWR* and *MRR* (Puertas and Luis, 2004; Lin et al., 2006). But almost all of them were on conventional macro EDM. Because of the stochastic thermal nature of the EDM process, it is difficult to explain all of those effects fully. The full understanding of micro ED milling process parameters are yet to be developed because the process itself is a new branch of manufacturing. For electric discharge the discharge energy (E) generated in the

2')( <sup>2</sup>

According to the equation (1.1), there are two ways to reduce the discharge energy: by

voltage results in an unstable discharge. Therefore, the total capacitance should be controlled primarily. The total stray capacitance (C') in the circuit plays a significant role here. The following subsections discuss the influence of different micro ED milling

Variation of discharge energy affects surface roughness. It is experimentally identified that with the increase in discharge energy either by increasing voltage or capacitance, *Ra* increased in die sinking EDM, wire EDM (WEDM) and WEDG (Uhlmann et al., 2005; Han et al., 2006; Chiang, 2007). A large discharging energy usually causes violent sparks and results a deeper erosion crater on the surface. Accompanying the cooling process after the spilling of molten metal, residues remain at the periphery of the crater to form a rough surface (Liao et al., 2004). Theoretical model for *Ry* or *Rmax* was developed for transistorized pulse generator type EDM (Chen and Mahdivian, 2000). It showed that with the increase in current and pulse duration, *Ry* increased. Surface roughness of ceramic due to EDM was also investigated. EDM of boron carbide (B4C) showed that *Ra* and *Ry* increased with pulse duration but decreased with current (Puertas and Luis, 2004). On the other hand, EDM of high speed steel showed that with the increase of current, *Ra* increased and with the increase of pulse duration it decreased (Lin et al., 2006). It was suggested to use low conductive or higher resistive dielectric to ensure low *Ra* and *Ry* (Liao et al., 2004). Use of rotating electrode was helpful for high shape accuracy and higher *MRR* but it caused higher surface roughness

Formation of crack is a challenge in EDM. The presence of high crack density is not desirable because it leads early product failure. Investigation of EDM parameters showed that higher pulse-on duration increased both the average white layer thickness and the induced stress (Lee and Tai, 2003; Ekmekci et al., 2005). These two conditions tend to

(1.1)

). However, low

1 *VCCE*

reducing the discharge voltage or by reducing the total capacitance (C + C'

parameters on surface roughness, tool wear and material removal rate.

**1.1.3 Process parameters** 

C = capacitance of capacitors,

 = total stray capacitance, V = discharge voltage

**1.1.4 Surface roughness** 

(Her and Weng, 2001).

Where,

C'

discharge circuit can be expressed roughly as follows:

Tool wear is responsible for shape inaccuracy and dimensional instability. In micro ED milling at higher feed rate tool wear was found decreasing (Lim et al., 2003). It was also observed that tool wear was increasing with feed rate but after reaching the peak it started decreasing (Kim et al., 2005). High discharge energy usually caused high *TWR* (Uhlmann et al., 2005). In EDM of high speed steel and boron carbide (B4C), tool wear was found to be proportional to current and inversely proportional to pulse duration (Puertas and Luis, 2004; Lin et al., 2006). It was also found that tool wear showed an optimum peak value while pulse duration had increased (Ozgedik and Cogun, 2006). However, use of shorter pulse ontime was suggested to achieve lower *TWR* (Son et al., 2007).

The negative polarity of the tool gives a lower tool wear than that of positive polarity in the range of low to medium discharge current values. At high current settings, the polarity has no significant effect on the tool wear. A slight decrease in *TWR* was observed with increasing current in negative polarity. In the case of positive polarity, *TWR* decreased significantly with increasing current (Lee and Li, 2001).

The type of dielectric fluid is also responsible for tool wear rate. Kerosene is frequently use as the dielectric medium. In EDM, carbide layer formed on the workpiece surface with the use of kerosene, which has a higher melting temperature than the oxide layer formed with the use of distilled water. The carbide layer formed by kerosene needed higher pulse energy for melting and evaporation which caused high tool wear (Chen et al., 1999). The use of low resistive deionized water as a dielectric fluid reduced the tool wear as compared to kerosene (Chung et al., 2007).

#### **1.1.6 Material removal rate**

Experimental investigation in conventional EDM found that *MRR* increased with the discharge current. With the increase in pulse duration MRR showed a peak and then downfall (Chen and Mahdivian, 2000; Lin et al., 2006;). Increase in discharge energy by discharge voltage or capacitance, resulted in higher *MRR* (Kim et al., 2005). It was also observed that material removal depth was inversely proportional to feed rate (Lim et al., 2003). *MRR* can be increased in both positive and negative polarity by rotating the toolelectrode (Her and Weng, 2001). Higher *MRR* was obtained for low ultrasonic vibration frequency combined with tool rotation (Ghoreishi and Atkinson, 2002).

Dielectric flushing system is also responsible for the variation in *MRR*. The low material removal rate in static condition is mainly due to improper flushing of the molten and evaporated workpiece material from the machining gap (Ozgedik and Cogun, 2006). With the increase in flushing flow rate, *MRR* increased. Use of kerosene and tap water as the dielectric fluid showed higher *MRR* than distill water (Chen et al., 1999).

#### **1.1.7 Tool electrode**

EDM tool electrodes should be conductive. A list of different tool electrode materials is shown in Table 1.2. In micro range, the replacement of tool during machining is unacceptable because it leads to positional inaccuracy. For this, tool with low wear ratio are needed in micro EDM. Because of its low wear ratio, tungsten or tungsten carbide rods or tubes of diameters within the range 100-400 µm are preferred as tool electrodes for micro ED drilling and milling (Pham et al., 2004). The handling of tungsten and tungsten- carbide rods is difficult as they can be easily damaged. Therefore, sub-systems are incorporated into micro EDM machines for on-the-machine manufacture and holding of the required microelectrodes. The most common sub-systems are ceramic guides and dressing units such as WEDG.


Table 1.2. Features of different tool electrode materials

Fig. 1.3. Principle of WEDG with one wire guide (Fleischer et al., 2004)

unacceptable because it leads to positional inaccuracy. For this, tool with low wear ratio are needed in micro EDM. Because of its low wear ratio, tungsten or tungsten carbide rods or tubes of diameters within the range 100-400 µm are preferred as tool electrodes for micro ED drilling and milling (Pham et al., 2004). The handling of tungsten and tungsten- carbide rods is difficult as they can be easily damaged. Therefore, sub-systems are incorporated into micro EDM machines for on-the-machine manufacture and holding of the required microelectrodes. The most common sub-systems are ceramic guides and dressing units such as

Brass 1 : 1 Easy Low Holes, slits Copper 1 : 2 Easy Moderate Holes

Tungsten 1 : 8 Moderate High Large areas

Zinc 1 : 2 Easy Low Die cavities on steel

Tungsten 1 : 10 Poor High Small slots, holes

carbide 1 : 10 Poor High Small slots, holes

Steel 1 : 4 Easy Low Through holes into non-Fe

materials

(tool/workpiece) Machinability Cost Best application

WEDG.

Electrode material

Copper-

Tungsten

Volume wear ratio

Table 1.2. Features of different tool electrode materials

Fig. 1.3. Principle of WEDG with one wire guide (Fleischer et al., 2004)

The tool electrode can be dressed or prepared either by etching, grinding or lithography process. However, when the fabricated tool is chucked on the machine, an alignment error between tool center and spindle center and run-out error of tool which has an influence on the variation of the discharge gap occurs and cause serious problems in micro machining and assembly of micro parts. For this, on-machine tool dressing is always recommended. WEDG is an on-machine tool dressing method. The principle of WEDG was developed in the mid 1980s. The WEDG process is using a traveling wire between one or two guides. The wire is the tool electrode of 50 and 200 µm diameter. In Figure 1.3 the principle of the WEDG-unit with one wire guide is shown, where rotating workpiece is fed downwards and the discharge area is limited to the front edge of the wire. The straightness of the ground part depends only on the variation of the gap distance at the machining point. Minimum 5.5 m tool was fabricated by WEDG from experiment (Kim et al., 2005). But in WEDG it is difficult to control the shape, dimension and roughness of the ground electrode (Lim et al., 2003).

#### **1.1.8 Workpiece materials**

In general, all conductive materials can be machined by the process that follows EDM principle. Different work materials for EDM are as follows:


#### **2. Experimental design**

The experiment was designed based on 3-level full factorial statistical model, where the three micro ED milling process parameters were feed rate, capacitance and voltage. The experiment was designed to analyze the individual influences, interaction effects and quadratic effects of these parameters on *Ra*, *Ry*, *TWR* and *MRR*. Table 2.1 shows the experimental conditions. Total number of 32 experiments had been conducted to complete the analysis. Six experiments at the same combination were conducted to reduce the curvature effect. The response parameters were statistically analyzed by applying factorial ANOVA at probability value of 0.05. This experimental study has been briefly published before (Mehfuz and Ali, 2009)

The definition of response parameters are as follows:

*Ra (m)= The arithmetic average deviation from the mean line* 

*Ry (m)= Maximum peak-to-valley roughness height* 

> *Volume of tool wear TWR Volume of workpiece wear*

*Weight of material removed MRR (mg / min) Machining time*


Table 2.1. Experimental Conditions of micro ED milling

#### **2.1 Tool and workpiece materials**

The workpiece material was beryllium-copper (Be-Cu) alloy (Be = 0.4 %, Ni = 1.8 % Cu = 97.8 %), which is industrially known as Protherm. Some features of Be-Cu alloy are as follows:


The tool electrode material was tungsten (W). It has the following features:


#### **2.2 Experimental procedures and measurements**

In this subsection the equipments used to conduct the experiments and the experimental procedures are described. A commercial multi-purpose miniature machine tool (DT-110 Mikrotools, Singapore), shown in Figure 3.1, was used to conduct the micro ED milling experiments. The machine contained RC circuit to generate small but high frequency discharge energy. EDM-3 synthetic oil was used as the dielectric fluid. The following subsections discuss the different steps of the experiments.

#### **2.2.1 Sample preparation**

138 Micromachining Techniques for Fabrication of Micro and Nano Structures

Feed rate (µm/s) A 2.00 4.00 6.00 Capacitance (nF) B 0.10 1.00 10.00 Voltage (volts) C 80 100 120

Fixed Parameters Experimental conditions

The workpiece material was beryllium-copper (Be-Cu) alloy (Be = 0.4 %, Ni = 1.8 % Cu = 97.8 %), which is industrially known as Protherm. Some features of Be-Cu alloy are as

c) Excellent corrosion resistance, which enables it to become a suitable mold material.

In this subsection the equipments used to conduct the experiments and the experimental procedures are described. A commercial multi-purpose miniature machine tool (DT-110 Mikrotools, Singapore), shown in Figure 3.1, was used to conduct the micro ED milling experiments. The machine contained RC circuit to generate small but high frequency

The tool electrode material was tungsten (W). It has the following features:

Experimental conditions

*Level* 

Factors *1 2 3* 

Controlled Parameters

Tool electrode Tungsten workpiece Be-Cu alloy

Polarity Workpiece +

Dielectric fluid EDM-3 (synthetic oil)

Tool electrode dia (mm) 0.50 Spindle speed (rpm) 2000

Machining length (mm) 13.00 Machining depth (mm) 0.20

Table 2.1. Experimental Conditions of micro ED milling

a) High thermal conductivity (245.00 W/moC at 20OC).

**2.2 Experimental procedures and measurements** 

**2.1 Tool and workpiece materials** 

b) Resistance to high temperature.

b) Suitable for small holes or slot,

a) Low tool wear ratio,

d) Machinable by WEDG.

c) High cost,

d) Good machinability and polishability.

follows:

Before the experiments, the workpieces were prepared by grinding and polishing. Workpieces were ground by using (NAGASEI SGW 52, Japan) surface grinding machine. Next, the workpiece was grounded manually by using 240, 320, 400 and 1000 grade of sand papers respectively. Then the workpieces were polished by 3 µm diamond suspensions and lubricant. The polishing steps were needed to obtain mirror surface finish. After the final polishing, the workpiece were immersed in acetone and cleaned for 15 minutes by ultrasonic cleaning machine, (BRANSON 2510). Tungsten tool electrodes of 500 µm diameter were also cleaned by the same ultrasonic cleaning machine to clean. Ultrasonic cleaning is useful for the removal of micro sized dirt particle. Then both the workpieces and tool-electrodes were weighed on an electric balance (B204-S Mettler Toledo, Switzerland) with a resolution of 0.1 mg. The measured weights before and after machining were compared. The following shows the specifications of a sample workpiece and tool-electrode after the final polishing. The prepared samples were then used for micro ED milling as discussed in following subsection.

#### **Specification of the workpiece:**

Thickness = 5 mm Length = 50 mm Width = 15 mm

#### **Specification of the tool-electrode:**

Diameter = 500 µm Length = 50 mm

#### **2.2.2 Experiments by micro electro discharge milling**

At the beginning of the micro ED milling set-up, the tool electrode was clamped in the spindle, while the workpiece was clamped on the worktable. After clamping, the spindle rotation was fixed at a speed of 2000 rpm, the worktable was fed and dielectric fluids were flown to the sparking zone. When discharge occurred between the tool-electrode and the workpiece, a small portion of workpiece melted and evaporated due to high local temperature, creating a crater on the work surface. The workpiece cooled rapidly by the influence of dielectric fluid. Figure 2.1a schematically depicts the experimental setup. The experiment was done to fabricate a microchannel for the dimensions of 13 mm length, 0.5 mm width and 0.2 mm depth. Figure 2.1b shows the schematic of the microchannels after micro ED milling. Computer numerical coding was used to conduct the experiment. The machining time was obtained from the auto stopwatch of the integrated computer. After the experiment, both the tool-electrode and workpiece were cleaned again in ultrasonic cleaning bath and dried.

After this, the tool-electrode and workpiece were weighed to find out the weight loss due to machining. The differences of weight before and after machining were used to calculate the *TWR* and *MRR*. After weighing, the samples were put into a dry box to avoid the moisture contact. These samples were used for SEM investigation and measurement of *Ra* and *Ry* as discussed in the following subsections.

Fig. 2.1. Schematic of experiment: (a) micro ED milling process**,** (b) specification of the microchannel

#### **2.2.2.1 SEM Inspection**

140 Micromachining Techniques for Fabrication of Micro and Nano Structures

contact. These samples were used for SEM investigation and measurement of *Ra* and *Ry* as

a)

b)

Fig. 2.1. Schematic of experiment: (a) micro ED milling process**,** (b) specification of the

microchannel

discussed in the following subsections.

The work sample was cleaned by ultrasonic cleaning and sputter coated with carbon by Coater (SC 7640). Though the work material was conductive, it was coated to get high quality images. The machined surfaces were then inspected by scanning electron microscope (SEM) (JEOL JSM-5600). Figure 2.2 shows SEM micrographs of a microchannel produced by micro ED milling.

#### **2.2.2.2 Measurement of surface roughness**

Both the *Ra* and *Ry* were measured by using a precision surface profiler (Mitutoyo, Surftest SV-500, Japan). This profiler provided surface roughness along a line with a resolution of 10 nm. As the length of the microchannel was 13 mm long, the roughness was measured in three segments. Each segment was 3.5 mm long. Average of these three measurements was taken as the surface roughness value.

#### **3. Results and discussions**

The experimental results of *Ra, Ry, TWR* and *MRR*, are analyzed using variance approach (ANOVA) to check the adequacy of the developed statistical model. The analysis ultimately showed the main and interaction effects of the process variables on the responses. Main effect was the direct effect of an independent variable while interaction effect was the joint effect of two independent variables on the response.

#### **3.1 Calculation of analysis of variance**

ANOVA is a calculation procedure to allocate the amount of variation in a process and determine if it is significant or is caused by random noise. ANOVA tables are constructed for *Ra*, *Ry*, *TWR* and *MRR* respectively. The statistical significance of each effect by means of the comparison of the mean squares (MS) with an estimation of the experimental error was performed. In the table, SS represents sum of squares while DF represents the number of degrees of freedom. The column corresponding to MS is obtained simply by dividing SS by its corresponding DF. In contrast, column of F value is calculated as the quotient of each of the MS of the effects divided by the value of the MS corresponding to the residual. Column of


Table 3.1. Design matrix of the experiment and measured responses

Probability values gives the probability values associated with values that take the variable of a function of distribution F.

$$SS\_{total} = SS\_{error} + SS\_{traatmembs}$$

*SS SS SS total error treatments SS SS SS total error treatments MS mod el Model F value MS residual*

#### **3.1.1 Surface roughness**

142 Micromachining Techniques for Fabrication of Micro and Nano Structures

1 2.00 0.10 80.00 0.04 0.31 0.121 0.02 2 4.00 0.10 80.00 0.04 0.35 0.044 0.07 3 6.00 0.10 80.00 0.04 0.36 0.133 0.09 4 2.00 1.00 80.00 0.10 0.78 0.154 0.04 5 4.00 1.00 80.00 0.12 0.91 0.066 0.09 6 6.00 1.00 80.00 0.10 0.89 0.165 0.09 7 2.00 10.00 80.00 0.44 3.23 0.182 0.06 8 4.00 10.00 80.00 0.44 3.21 0.089 0.10 9 6.00 10.00 80.00 0.48 3.53 0.220 0.11 10 2.00 0.10 100.00 0.05 0.49 0.165 0.03 11 4.00 0.10 100.00 0.05 0.40 0.049 0.09 12 6.00 0.10 100.00 0.06 0.54 0.157 0.09 13 2.00 1.00 100.00 0.19 1.28 0.194 0.06 14 4.00 1.00 100.00 0.17 1.20 0.065 0.16 15 6.00 1.00 100.00 0.18 1.38 0.186 0.14 16 2.00 10.00 100.00 0.53 3.99 0.239 0.10 17 4.00 10.00 100.00 0.54 4.08 0.098 0.35 18 6.00 10.00 100.00 0.53 4.05 0.216 0.13 19 2.00 0.10 120.00 0.05 0.44 0.198 0.06 20 4.00 0.10 120.00 0.07 0.52 0.065 0.15 21 6.00 0.10 120.00 0.08 1.23 0.166 0.10 22 2.00 1.00 120.00 0.19 1.81 0.227 0.07 23 4.00 1.00 120.00 0.17 1.22 0.086 0.36 24 6.00 1.00 120.00 0.18 1.37 0.198 0.14 25 2.00 10.00 120.00 0.56 3.11 0.261 0.16 26 4.00 10.00 120.00 0.62 3.79 0.128 0.41 27 6.00 10.00 120.00 0.54 3.77 0.247 0.15 28 4.00 1.00 100.00 0.22 1.30 0.058 0.10 29 4.00 1.00 100.00 0.22 1.53 0.072 0.49 30 4.00 1.00 100.00 0.15 1.06 0.067 0.24 31 4.00 1.00 100.00 0.19 1.47 0.062 0.13 32 4.00 1.00 100.00 0.16 1.28 0.075 0.20

**Voltage (volt)** 

*Ra* **(µm)** *Ry*

**(µm)** *TWR MRR* 

**(mg/min)** 

**Capacitance (nF)** 

Table 3.1. Design matrix of the experiment and measured responses

of a function of distribution F.

Probability values gives the probability values associated with values that take the variable

*SS SS SS total error treatments*

**Exp No. Feed rate** 

**(µm/s)** 

Table 3.2 shows the analysis of variance for *Ra*. Using this analysis a second order quadratic model was developed, which is shown below in Equation (3.1). The model was developed for 95% level of confidence. The model F-value of 213.06 implies the model is significant. There is almost no influence, 0.01%, of noise on the model developed. By checking F value and P-value it is clearly seen that Factor B (capacitance), Factor C (voltage) and Factor B2 are most influential on *Ra*. The P-value of each of these factors indicates the confidence level is more than 99.00%, which shows their very strong influence. The P - value of interaction effects of BC shows the confidence level is above 95.0% and thus shows very good influence on *Ra*. The P-value of Factor A (feed rate) and interaction Factor AC has insignificant influence over *Ra* as it provides very high P values. The lack of fit F-value of 0.53 implies that the lack of fit is not significant compare to the pure error. The high P-value of lack of fit, 85.35%, indicates the model is fit, while the very low P-value of the model, 0.01%, indicates that the model is significant. The specific power transformation was chosen within the confidence level, which was suggested by the Design Expert software toolbox using Box-Cox plotting. In this case, natural log power transformation was suggested. Thus the developed statistical quadratic equation for *Ra* is:


$$\text{Ln(R}\_{a}\text{)} = -7.956 \text{ -} 0.044f + 1.423 \text{C} + 0.087 \text{V -} 0.111 \text{C}^2 - 0.0004 \text{V}^2 + 0.0006 \text{fV -} 0.0007 \text{C} \text{V} \tag{3.1}$$

Table 3.2. Analysis of variance for main and interaction effects on *Ra*

Fig. 3.1. Estimated response surface of *Ra* (µm): (a) *Ln(Ra)* vs. *f* and *V* when *C* = 1.0 nF, (b) *Ln(Ra)* vs. *C* and *V*, when *f* = 4.0 µm/s

(a)

(b) Fig. 3.1. Estimated response surface of *Ra* (µm): (a) *Ln(Ra)* vs. *f* and *V* when *C* = 1.0 nF, (b)

*Ln(Ra)* vs. *C* and *V*, when *f* = 4.0 µm/s

Figure 3.1 shows the effect of feed rate-voltage and capacitance-voltage on *Ra* respectively. Capacitance and voltage are strongly influential on *Ra*. *Ra* increases along with the increase in voltage, which is same as the conventional EDM. With the increase of capacitance from 0.1 to 5 nF the *Ra* value increases but for further increase of capacitance the *Ra* values reduce. As the capacitance increased, large energy dissipated which erodes more materials with stronger spark. This strong spark erodes materials with high amount of debris from both the tool electrode and workpiece creating uneven crater. As these debris are trapped in between the plasma channel, it causes unwanted spark. Thus, a high amount of discharge energy is employed to spark with debris, while work material is effectively removed by a small portion of discharge energy. Thus lower *Ra* is obtained.

Fig. 3.2. SEM micrograph of the surface texture of micro ED milled surface when feed rate = 4 µm/s and voltage = 100 volts, with varying capacitance of (a) 0.1 nF and (b) 1.0 nF.

Figure 3.2 shows SEM surface texture of two machined surfaces with varying capacitance. In the Figure 4.2a the surface texture was found very smooth (*Ra* = 0.05 µm) with 0.1 nF capacitance. Figure 4.2b shows the surface texture (*Ra* = 0.22 µm) with the capacitance of 1 nF. Here, the surface was found rougher with increased amount of unwashed debris.

Similarly the developed statistical quadratic equation for maximum peak-to-valley height roughness, *Ry* is:

$$\text{Ln(R}\_{y}\text{)} = -4.981 \text{--} 0.077 f + 1.264 \text{C} + 0.07 V \text{ -} 0.091 \text{C}^2 - 0.0003 V^2 + 0.0012 fV - 0.0014 \text{C} \text{V} \tag{3.2}$$

#### **3.1.2 Tool Wear Ratio**

A second order quadratic model was developed which is the mathematical expression of TWR:

$$\text{TWR} = 0.285 \cdot 0.203f + 0.031 \text{C} + 0.002V + 0.029f^2 \cdot 0.003 \text{C}^2 + \text{(3.0002fC - 0.0003fV + 0.0003fV)}\tag{3.3}$$

Figure 3.4 shows the effect of feed rate-capacitance and feed rate-voltage on *TWR*. Capacitance and voltage are found as the strong influencer of *TWR*, like Ra, along with the

Fig. 3.4. Estimated response surface of *TWR*: (a) *TWR* vs. *f* and *C* when *V* = 100 volts, (b) *TWR* vs. *f* and *V*, when C = 1.0 nF

(a)

(b)

Fig. 3.4. Estimated response surface of *TWR*: (a) *TWR* vs. *f* and *C* when *V* = 100 volts,

(b) *TWR* vs. *f* and *V*, when C = 1.0 nF

interaction effect of feed rate-voltage. *TWR* decreases with the increase in feed rate. For further increase in feed rate *TWR* starts rising. In the first phase, for a particular spark energy discharge if the feed rate is so small it will have high electron emission rate from the tool electrode, resulting high *TWR*. As the feed rate increase the spark energy is more involved in material erosion, which reduces *TWR* and reaches to minimum. In the second phase, for further increase of feed rate from the optimum, the unflashed eroded materials cause unwanted spark with the tool, which results more tool wear. Thus high *TWR* is obtained. With the increase in capacitance large energy dissipated which produces stronger spark resulting high work material erosion. Higher spark energy produces higher amount of debris. These debris sticking on the workpiece, trapped in and causes unwanted spark. The unwanted sparks erodes materials from the tool electrode, which results high tool wear. Thus, higher capacitance results higher *TWR*. As significant amounts of spark energy are employed in sparking with debris, effectively a lower amount of work material is eroded. *TWR* is a ratio of tool wear and workpiece wear, so decreases in material removal mean increase in *TWR*.

#### **3.1.3 Material Removal Rate**

The developed *MRR* statistical quadratic model is:

$$\frac{1}{\sqrt{\text{MRR}}} = 16.85 \text{--} 3.68f \text{--} 1.24 \text{C} \text{ -} 0.070V + 0.28f^2 + 0.09 \text{C}^2 + 0.04f \text{C} + 0.01fV \tag{3.4}$$

Figure 3.5 shows the effect of feed rate-capacitance and feed rate-voltage on *MRR* <sup>1</sup> .

Therefore, the decrease of *MRR* 1 with the increase of any of the factors means the increase

in *MRR*. *MRR* increases with the increase in feed rate. For further increase in feed rate *MRR* starts falling slightly. As the feed rate increase, the spark energy is more involved in material erosion, which increases *MRR* till reaches to the optimum. For further increase of feed rate from the optimum, the unflashed eroded materials cause unwanted spark with the tool, which changes the tool shape making it uneven and rough. The spark discharge from this uneven tool results less *MRR*. With the increase in capacitance large energy dissipated which erodes more work materials with stronger spark. With the material erosion, the unflashed debris sticking on the workpiece is trapped in between the machining zone, which causes unwanted spark with the tool-electrode. Thus a great portion of discharge energy occupies with unwanted sparking, while the remaining erodes the work material. Hence, effectively a lower amount of work material is eroded.

#### **3.2 Multiple response optimizations**

This subsection discusses the optimization of four output responses, *Ra, Ry, TWR* and *MRR*. The developed non-linear models Equations (3.1) to (3.4) are used for multiple response optimization. Desirability function approach was used for optimization. The method finds operating conditions, e.g. feed rate, capacitance and voltage that provide the "most desirable" values of the responses, e.g. *Ra, Ry, TWR* and *MRR*. The following subsection describes the desirability function approach.

Fig. 3.5. Estimated response surface of *MRR* (mg/min): (a) *MRR* 1 vs. *f* and *C* when *V* = 100 volts, (b) *MRR* 1 vs. *f* and *V*, when *C* = 1.0 nF

#### **3.2.1 Desirability function approach**

148 Micromachining Techniques for Fabrication of Micro and Nano Structures

(a)

(b)

1

vs. *f* and *C* when *V* = 100

Fig. 3.5. Estimated response surface of *MRR* (mg/min): (a) *MRR*

vs. *f* and *V*, when *C* = 1.0 nF

volts, (b) *MRR*

1

In the analysis the objective function, *D(Yi)*, called the desirability function, reflects the desirable ranges for each response *Yi*(*x*) (where, i = *Ra, Ry, TWR*, *MRR*). For each response, a desirability function *di*(*Yi*) assigns numbers between 0 and 1 to the possible values of *Yi*.

*di*(*Yi*) = 0 representing a completely undesirable value of *Yi* and

*di*(*Yi*) = 1 representing a completely desirable or ideal response value.

The individual desirabilities are then combined using the geometric mean, which gives the overall desirability *D*:

$$D = \sqrt[n]{(d\_1 \times d\_2 \times ... \times d\_n)} = (d\_1 \times d\_2 \times ... \times d\_n)^{n} \tag{3.5}$$

where n is the number of responses in the measure. From the equation (4.5) it can be noticed that if any response *Yi* is completely undesirable (*di*(*Yi*) = 0), then the overall desirability is zero. In this case, the geometric mean of overall desirability is as follows:

$$D = \left(d\_{R\_x} \times d\_{R\_y} \times d\_{TWR} \times d\_{MRR}\right)^{\frac{1}{3}}\tag{3.6}$$

Depending on whether a particular response *Yi* is to be maximized, minimized, or assigned a target value, different desirability functions *di*(*Yi*) can be used. In this case, *Ra, Ry* and *TWR* are needed to be minimized while *MRR* are needed to maximized. Following are the two desirability functions:

$$d\_l(\mathbf{Y}\_i) = \begin{cases} 0 \\ \frac{\mathbf{Y}\_i(\mathbf{x}) - L\_i}{T\_i - L\_i} \end{cases} \begin{cases} \text{if} & \mathbf{Y}\_i(\mathbf{x}) \le L\_i \\ \text{if} & L\_i \le \mathbf{Y}\_i(\mathbf{x}) \le T\_i \\ \text{if} & \mathbf{Y}\_i(\mathbf{x}) > T\_i \end{cases} \tag{3.7}$$

$$d\_i(Y\_i) = \left(\frac{Y\_i(\mathbf{x}) - \mathbf{U}\_i}{T\_i - \mathbf{U}\_i}\right)^s \begin{cases} \text{if} & Y\_i(\mathbf{x}) < T\_i \\ \text{if} & T\_i \le Y\_i(\mathbf{x}) \le \mathbf{U}\_i \\ \text{if} & Y\_i(\mathbf{x}) > \mathbf{U}\_i \end{cases} \tag{3.8}$$

where,

*Li* = Lower limit values

*Ui* = Upper limit values

*Ti* = Target values

*s =* weight (define the shape of desirability functions)


Table 3.3. Values of process parameters for the optimization of *Ra*, *Ry*, *TWR* and *MRR* 

Equation (3.7) is used when the goal is to maximize, while to minimize Equation (3.8) is needed. The value of s = 1 is chosen so that the desirability function increases linearly towards *Ti*. Table 3.3 shows the process parameters obtained after multiple response optimization. For the shown values of process parameters, it is 88.06% likely to get the *Ra* 0.04 µm, *Ry* 0.34 µm, *TWR* 0.044 and *MRR* 0.08 mg/min. Any other combination of the process parameters will either statistically less reliable or give poor results of at least one of the responses. The analysis was done by using computer software, Design Expert.

#### **3.3 Verification of optimized values**

Experiments were conducted to verify the result obtained from the multiple response optimization. The actual values obtained from the experiments are compared with the predicted values in Table 3.4. From the table it can be noticed that the predicted values of *Ra* shows no error with the actual, while TWR shows the maximum error. In 88.06% desirability, the percentages of error were found lesser for *TWR* and *MRR*. The bar charts of Figure 3.6 shows the comparison of predicted and actual values.


Table 3.4. Verification of multiple response optimization

Fig. 3.6. Comparison of predicted vs. actual responses: (a) at desirability of 88.06%

Equation (3.7) is used when the goal is to maximize, while to minimize Equation (3.8) is needed. The value of s = 1 is chosen so that the desirability function increases linearly towards *Ti*. Table 3.3 shows the process parameters obtained after multiple response optimization. For the shown values of process parameters, it is 88.06% likely to get the *Ra* 0.04 µm, *Ry* 0.34 µm, *TWR* 0.044 and *MRR* 0.08 mg/min. Any other combination of the process parameters will either statistically less reliable or give poor results of at least one of

Experiments were conducted to verify the result obtained from the multiple response optimization. The actual values obtained from the experiments are compared with the predicted values in Table 3.4. From the table it can be noticed that the predicted values of *Ra* shows no error with the actual, while TWR shows the maximum error. In 88.06% desirability, the percentages of error were found lesser for *TWR* and *MRR*. The bar charts of

Desirability Responses Predicted Actual % Error

Predicted vs. Actual (Desirability 88.06%)

Fig. 3.6. Comparison of predicted vs. actual responses: (a) at desirability of 88.06%

*Ra (µm)* 0.04 0.04 0.00

*Ry(µm)* 0.34 0.36 5.56

*TWR* 0.044 0.053 16.98

*MRR (mg/min)* 0.08 0.09 11.11

the responses. The analysis was done by using computer software, Design Expert.

Figure 3.6 shows the comparison of predicted and actual values.

Table 3.4. Verification of multiple response optimization

**3.3 Verification of optimized values** 

88.06%

#### **4. Application of micro ed milling: Micro swiss-roll combustor mold**

Micro swiss-roll combustor is a heat re-circulating combustor. It uses hydrocarbon fuels to generate high density energy. The advantage of a micro swiss-roll combustor is that it provides high density energy by reducing heat loss [Ahn and Ronney, 2005; Kim et al., 2007]. The generated heat inside the micro swiss-roll combustor is entrapped and recirculated. Thus, high density energy is obtained. One of the challenges in micro combustor design is to reduce the heat loss. To reduce the heat loss by reducing surface-to-volume ratio, wall thickness should be as small as possible [Ahn et al. 2004]. The application of micro swiss-roll combustor includes portable electronics, such as cell phone, laptop, space vehicles, military uses, telecommunication, etc.

Fig. 4.1. Proposed design of micro swiss-roll combustor mold cavity (a) top view and (b) isometric view

The proposed design of the micro swiss-roll combustor mold cavity is shown in Figure 5.1. Beryllium-copper alloy (Protherm) was selected as the mold material, because of its high thermal conductivity, high heat and corrosion resistance. The microchannel of the mold cavity was fabricated by using a tungsten tool-electrode of 100 µm diameter. The minimum gap between two microchannels was 380 µm. The preliminary drawing and the numerical code (NC) of the design was generated by using CATIA V.5 R14 computer aided drafting software.

#### **4.1 Fabrication of tool electrode by WEDG**

Commercially available 300 µm diameter cylindrical tungsten rod was first dressed to 100 µm diameter by WEDG. Later this rod was used as a tool-electrode in micro ED milling to fabricate microchannels. Figure 4.2a illustrates the mechanism of WEDG. Figure 4.2b is the picture taken during the experiment and Figure 4.2c illustrates the SEM image of fabricated tool electrode. Computer numerical coding was used to control the size and shape of the required tool-electrode. The process parameters used are shown in Table 4.1. The parameter values were selected after preliminary studies.


The dimensions of the proposed micro-swiss roll combustor mold are: *(length × width × depth*) = (4.5 mm × 4.5 mm × 1.0 mm)

Table 4.1. Experimental condition of WEDG for dressing of tool-electrode

Fig. 4.2. Schematic of tool-electrode dressing by WEDG

The proposed design of the micro swiss-roll combustor mold cavity is shown in Figure 5.1. Beryllium-copper alloy (Protherm) was selected as the mold material, because of its high thermal conductivity, high heat and corrosion resistance. The microchannel of the mold cavity was fabricated by using a tungsten tool-electrode of 100 µm diameter. The minimum gap between two microchannels was 380 µm. The preliminary drawing and the numerical code (NC) of the design was generated by using CATIA V.5 R14 computer aided drafting software.

Commercially available 300 µm diameter cylindrical tungsten rod was first dressed to 100 µm diameter by WEDG. Later this rod was used as a tool-electrode in micro ED milling to fabricate microchannels. Figure 4.2a illustrates the mechanism of WEDG. Figure 4.2b is the picture taken during the experiment and Figure 4.2c illustrates the SEM image of fabricated tool electrode. Computer numerical coding was used to control the size and shape of the required tool-electrode. The process parameters used are shown in Table 4.1. The parameter

Parameters Values Wire speed (mm/s) 20 Wire tension (%) 20 Capacitance (nF) 1 Voltage (volts) 100 Threshold (volts) 30 Polarity Wire –ve Spindle speed (rpm) 3000 Machining length (mm) 3

Di-electric medium EDM-3 synthetic oil

Table 4.1. Experimental condition of WEDG for dressing of tool-electrode

The dimensions of the proposed micro-swiss roll combustor mold are:

*(length × width × depth*) = (4.5 mm × 4.5 mm × 1.0 mm)

Fig. 4.2. Schematic of tool-electrode dressing by WEDG

**4.1 Fabrication of tool electrode by WEDG** 

values were selected after preliminary studies.

Fig. 4.3. Tool-electrode dressing by WEDG: (a) picture during WEDG and (b) SEM image of tool-electrode after dressing

#### **4.2 Fabrication of micro mold cavity**

The micro swiss-roll combustor mold cavity was fabricated by micro ED milling. Be-Cu alloy plate of 6 mm thickness was used as the workmaterial. The tool-electrode of 100 µm diameter was used, which produced microchannels of 120 µm width and 1 mm depth. Channel width comprises of the tool diameter and spark gap. Layer by layer approach was chosen to get better dimensional accuracy. The thickness of each layer was 200 µm. Figure 4.4 explains the layer by layer approach. The gap between two microchannels was 380 µm. After machining each 500 µm, the tool-electrode was dressed by WEDG to reduce the shape inaccuracy due to tool wear. The whole machining was done using computer numerical control. Figure 4.5a is the picture during experiments, Figure 4.5b shows the final product and Figure 4.5c shows the SEM micrographs of the window A in Figure 4.4b. The process parameters obtained from the multiple responses optimization were used in the microfabrication. The experimental condition is shown in Table 4.2.

Fig. 4.4. Layer by layer machining: (a) before machining, b) after machining


Table 4.2. Micro ED milling parameters for micro swiss-roll combustor mold

4.4 explains the layer by layer approach. The gap between two microchannels was 380 µm. After machining each 500 µm, the tool-electrode was dressed by WEDG to reduce the shape inaccuracy due to tool wear. The whole machining was done using computer numerical control. Figure 4.5a is the picture during experiments, Figure 4.5b shows the final product and Figure 4.5c shows the SEM micrographs of the window A in Figure 4.4b. The process parameters obtained from the multiple responses optimization were used in the

microfabrication. The experimental condition is shown in Table 4.2.

Fig. 4.4. Layer by layer machining: (a) before machining, b) after machining

Feed rate (µm/s) 4.79 Capacitance (nF) 0.1 Voltage (volts) 80 Threshold (volts) 30 Tool electrode dia (µm) 0.10 Spindle speed (rpm) 2000

Di-electric medium EDM-3 synthetic oil

Depth per pass (µm) 200 Machining length per tool dressing (µm) 500 Table 4.2. Micro ED milling parameters for micro swiss-roll combustor mold

Parameters Values

Fig. 4.5. Fabrication of micro swiss-roll combustor mold cavity by micro ED milling: (a) picture during micro ED milling, (b) fabricated micro swiss-roll combustor mold cavity, (c) SEM micrographs of window A in Figure 4.5 b.

#### **5. Conclusion**

Micro ED milling is shown as a potential fabrication technique for functional microcomponents. Influences of three micro ED milling parameters, feed rate, capacitance and voltage, were analyzed. Mathematical models were developed for output responses *Ra, Ry, TWR* and *MRR*. Analysis of multiple response optimization was done to get the best achievable response values. The micro ED milling process parameters obtained by the multiple response optimization were used in the fabrication of micro mold cavity. WEDG was used to dress the tool-electrode to a diameter of 100 µm. The final product was a micro swiss-roll combustor mold cavity. In brief, this research showed the followings:


#### **6. Acknowledgement**

This research was jointly funded by grant FRGS 0207-44 from Ministry of Higher Education, Malaysia and EDW B11-085-0563 from International Islamic University Malaysia.

#### **7. References**

Ahn, J., Ronney, P. D., (2005). *Plastic mesocombustor.* 4th Joint U.S. Sections Meeting, Combustion Institute, Philadelphia, PA. http://carambola.usc.edu/Research/MicroFIRE/PlasticMesoCombustors.pdf

Micro ED milling is shown as a potential fabrication technique for functional microcomponents. Influences of three micro ED milling parameters, feed rate, capacitance and voltage, were analyzed. Mathematical models were developed for output responses *Ra, Ry, TWR* and *MRR*. Analysis of multiple response optimization was done to get the best achievable response values. The micro ED milling process parameters obtained by the multiple response optimization were used in the fabrication of micro mold cavity. WEDG was used to dress the tool-electrode to a diameter of 100 µm. The final product was a micro

1. *C*apacitance and voltage have strong individual influence on both the *Ra* and *Ry*, while the interaction effect of capacitance and voltage also affects the roughness greatly. Ususally higher discharge energy results higher surface roughness. The unflushed debris sticking on the workpiece causes higher *Ra* and *Ry*. At very high discharge energy the entrapped debris inside the plasma channel creates unwanted spark with the tool-electrode. Thus only a small portion of discharge energy involves in material

2. *C*apacitance and voltage plays significant role on *TWR* along with the interaction effect of feed rate and voltage. At high discharge energy large amount of debris are produced,

4. Multiple response optimization shows 88.06% desirability for minimum achievable values of *Ra, Ry, TWR* and maximum achievable *MRR,* which are 0.04 µm, 0.34 µm, 0.044, 0.08 mg/min respectively when the feed rate, capacitance and voltage are 4.79 µm/s, 0.10 nF and 80.00 volts respectively. The achieved *Ra* and *Ry* values are in the

5. The result of multiple response optimization was verified by experiment. The percentages of errors for *Ra* (0.0%)*, Ry* (5.56%) at 88.06% desirability were found within the acceptable range. For *TWR* (16.98%) and *MRR* (11.11%), it was found relatively unsteady. Low resolution (0.1 mg) of electric balance could be a reason behind this. 6. A micro swiss-roll combustor mold cavity was fabricated by using the WEDG dressed tool. Optimized and verified micro ED milling process parameters were used for

7. Combination of micro ED milling and molding can be a suitable route for the mass

This research was jointly funded by grant FRGS 0207-44 from Ministry of Higher Education,

Ahn, J., Ronney, P. D., (2005). *Plastic mesocombustor.* 4th Joint U.S. Sections Meeting,

http://carambola.usc.edu/Research/MicroFIRE/PlasticMesoCombustors.pdf

which causes high *TWR* by generating unwanted sparks with the tool-electrode. 3. *F*eed rate, capacitance and voltage have strong individual and interaction effects on *MRR*. Usually, *MRR* is higher at high discharge energy. But the presence of high amount debris in the plasma channel often creates unwanted spark with the tool electrode. Thus only a

portion of energy involves in workmaterial removal, which reduces *MRR*.

fabrication. The final product has the channel dimension of 0.1 mm.

replication of miniaturized functional components at a lower cost.

Malaysia and EDW B11-085-0563 from International Islamic University Malaysia.

swiss-roll combustor mold cavity. In brief, this research showed the followings:

erosion process, which results low *Ra* and *Ry*.

acceptable range for many MEMS applications.

Combustion Institute, Philadelphia, PA.

**6. Acknowledgement** 

**7. References** 

**5. Conclusion** 


## **Mechanical Micromachining by Drilling, Milling and Slotting**

T. Gietzelt and L. Eichhorn

*Karlsruhe Institute of Technology, Campus Nord, Institute for Micro Process Engineering, Karlsruhe, Germany* 

#### **1. Introduction**

158 Micromachining Techniques for Fabrication of Micro and Nano Structures

Kim, Y. T., Park, S. J., Lee, S. J., (2005). *Micro/meso-scale shapes machining by micro EDM process*. International Journal of Precision Engineering and Manufacturing (6:2). King, J. A., Tucker, B. D., Vogt, D., (1999). *Electrically and thermally conductive nylon 6,6*.

Lee, H. T., Tai, T. Y., (2003). *Relationship between EDM parameters and surface crack formation*.

Lee, S. H., Li, X. P., (2001). *Study of the effect of machining parameters on the machining* 

Liao, Y. S., Huang, J. T., Chen, Y. H., (2004). *A study to achieve a fine surface finish in wire-*

Lim, H. S., Wong, Y. S., Rahman, M., Lee, M. K. E., (2003). *A study on the machining of high-*

Lin, Y. C., Cheng, C. H., Su, B. L., Hwang, L. R., (2006). *Machining characteristics and optimization of* 

Madou, M. J., (2002). *Fundamentals of micro fabrication*. The Science of Miniaturization. (2nd

Murali, M., and Yeo, H. S., (2004). *Rapid biocompatible micro device fabrication* by Micro Electro

Mehfuz, R., Ali, M. Y. (2009). *Investigation of Machining Parameters for the Multiple Response* 

Ronney, P. D., (2003). *Analysis of non-adiabatic heatrecirculating combustors*. Combustion and

Schoth, A., Fo¨rster, R., Menz, W., (2005). *Micro wire EDM for high aspect ratio 3D microstructuring of ceramics and metals*. Microsystem Technologies. (11) 250–253. Snakenborg, D., Klank, H., Kutter, J. P., (2004). *Microstructure fabrication with a CO2 laser system*. Journal of Micromechanics and Microengineering (14) 182-189. Son, S. M., Lim, H. S., Kumar, A. S. and Rahman, M., (2007). *Influences of pulsed power* 

Uhlmann, E., Piltz, S., Jerzembeck, S., (2005). *Micro-machining of cylindrical parts by electrical discharge grinding*. Journal of Materials Processing Technology (160) 15–23. Xingchao, Z., Yong, Z., Hao, T., Yong, L., Xiaohua, C., Guoliang, C., (2007). *Micro-electro-*

*discharge machining of bulk metallic glasses*. Proceedings of HDP'07.

*condition on the machining properties in micro EDM*. Journal of Materials Processing

Orloff, J. (1997), *Handbook of charged particles optic.* CRC Press, New York, USA, 319-360. Ozgedik, A. and Cogun, C., (2006). *An experimental investigation of tool wear in electric discharge machining*. International Journal of Advanced Manufacturing Technology (27) 488–500. Pham, D. T., Dimov, S. S., Bigot, S., Ivanov, A., Popov, K., (2004). *Micro EDM- recent developments and research issues*. Journal of Materials Processing Technology. (149) 50-57. Puertas, I., Luis, C. J., (2004). *A study of optimization of machining parameters for electrical discharge machining of boron carbide*. Materials and Manufacturing Processes, (19:6) 1041 – 1070. Rajurkar, K. P., Pandit, S. M., (1986). *Formation and ejection of EDM debris*. ASME Trans

*taguchi method*. Materials and Manufacturing Processes. (21:8) 922 – 929. Lin, Y., Matson, D. W., Kurath, D. E., Wen, J., Xiang, F., Bennett, W. D., Martin, P. *devices on* 

Discharge Machining. Biomedical Microdevices. (6:1) 41-45.

Advanced Manufacturing Technology, (43), 264-275.

Journal of Engineering Industry. (108) 22–26.

Flame. (135) 421-439

Technology. (190) 73–76.

*EDM*. Journal of Materials Processing Technology. (149) 165–171.

*characteristics in electrical discharge machining of tungsten carbide*. Journal of Materials

*aspect ratio micro-structures using micro-EDM*. Journal of Materials Processing

*machining parameters of SKH 57 high-speed steel using electrical-discharge machining based on* 

*polymer substrates for bioanalytical application* M., Smith, R. D., (1999). *Microfluidics*.

*Optimization of Micro Electro Discharge Milling*. The International Journal of

Journal of Materials Processing Technology (142) 676–683.

Polymer Composites, (20) No. 5.

Processing Technology (115) 344–358.

Pacific Northwest National Laboratory.

ed.) Florida, USA: CRC Press LLC.

Technology. (140) 318-325.

Micromachining is not only a simple miniaturization of processes using macroscopic tools. As a matter of fact, a lot of specific concerns have to be met for successful fabrication of microstructures. This chapter will be focussed on micromachining using geometrically determined cutting edges, namely on techniques like drilling, milling and slotting. These methods are very flexible. Compared to EDM, ECM or lithographic processes like LIGA, they can be applied to a wide range of materials, like polymers, metals and alloys as well as to some kinds of ceramics, possess a high material removal rate and allow a great degree of freedom concerning design. There are nearly no geometrical limitations and also 3Dstructures can be manufactured easily.

#### **2. Micromachining by geometrically determined cutting edges**

#### **2.1 Differences as compared to geometrically undefined cutting edges**

Micromachining techniques can be divided into two main categories: Processes working with undefined cutting edges e.g. grinding, honing, lapping, and processes are using defined cutting edges like drilling, milling and slotting.

Especially grinding works at high cutting rates. Most of the cutting energy is transferred into heat and absorbed by the work piece [Kön99, Fri08]. The properties of the work piece can be altered or decreased by surface cracks and internal stress due to external forces as well as by microstructural changes due to excessive heat.

Especially for micro grinding using small-diameter tools, extremely high numbers of revolutions are required to achieve a reasonable circumferential speed of up to more than 100 m/s. Compared to processes using defined cutting edges, the energy need is high and the material removal rate is comparably low. Nevertheless, especially for very hard materials like most ceramics where defined cutting edges do not work, grinding is a capable technique. However, since diamonds are used to machine the very hard ceramic materials, machining expenses can be a major cost factor for ceramic parts [War00].

When machining using geometrically determined cutting edges, the cutting energy is mostly used to overcome the cohesion forces of the machined material. The material removal rate is higher than for grinding and most of the heat is transferred to and removed with the chips. A good approximation for the removal of heat is, that 75% are transferred to the chips formed, 18% migrate to the tool and 7% to the work piece [Kön90]. Hence, the work piece and its microstructure are not as affected as in the case of grinding.

When rotating micro-sized tools, attention has to be paid in general to the response on external loads by deformation. The load case and the reaction of the tool are very important as regards the machining result. Especially in the case of rotating tools possessing two chip flutes, the cross section is reduced. Load cases can be distinguished for different machining processes:

In case of micro drilling, only a torsional moment acts on the tool. Depending on the length, bending and buckling may be an issue.

Slotting is an appropriate way if the desired trench width is smaller than that of commercially available end mills or for large aspect ratios where the stability of end mills can be problematic. An additional advantage of slotting is that the tool may not be axially symmetric. Hence, a better stability is accomplished and only bending acts on the tool. Tools with optimized shapes and angles can be made using precision grinding machines e. g. Ewag WS 11 [Ewa\_Ws] with worn hard metal end mills made of ultra-fine grain carbides. A disadvantage is the slow feed rate and, hence, a smaller material removal rate than in the case of micro milling.

In the case of micro milling both torsion and bending act on the tool. Predominantly, micro end mills are made of hard metal possessing two chip flutes. However, also tools made of monocrystalline diamonds with only one cutting edge are used.

Hence, dynamic fatigue due to cyclic bending or vibrations and irregular load may be a serious problem especially for two flute micro end mills. Characteristics like appropriate hard metal substrate, manufacturing process affecting roughness and cracks in the surface, coating technology and adapted tool shape will be discussed in Chapter 3 for micro end mills.

#### **2.2 Geometrical limits of tools**

Micro drills originate from conductor board manufacturing for contacting through multiple layers. Although the prepregs used consist of a cured resin and very abrasive glass fibres, uncoated hard metal drills are used with good success. Uncoated micro drills are available down to diameters of 20 µm [Ato\_Ad, Ham\_38]. Fig. 1 shows a 30 µm micro drill bit.

Fig. 1. 30 µm micro drill bit with detail of the cutting edge. Grooves from grinding with jagged edge due to the composite nature of hard metal can be seen.

the chips formed, 18% migrate to the tool and 7% to the work piece [Kön90]. Hence, the

When rotating micro-sized tools, attention has to be paid in general to the response on external loads by deformation. The load case and the reaction of the tool are very important as regards the machining result. Especially in the case of rotating tools possessing two chip flutes, the cross section is reduced. Load cases can be distinguished for different machining

In case of micro drilling, only a torsional moment acts on the tool. Depending on the length,

Slotting is an appropriate way if the desired trench width is smaller than that of commercially available end mills or for large aspect ratios where the stability of end mills can be problematic. An additional advantage of slotting is that the tool may not be axially symmetric. Hence, a better stability is accomplished and only bending acts on the tool. Tools with optimized shapes and angles can be made using precision grinding machines e. g. Ewag WS 11 [Ewa\_Ws] with worn hard metal end mills made of ultra-fine grain carbides. A disadvantage is the slow feed rate and, hence, a smaller material removal rate than in the

In the case of micro milling both torsion and bending act on the tool. Predominantly, micro end mills are made of hard metal possessing two chip flutes. However, also tools made of

Hence, dynamic fatigue due to cyclic bending or vibrations and irregular load may be a serious problem especially for two flute micro end mills. Characteristics like appropriate hard metal substrate, manufacturing process affecting roughness and cracks in the surface, coating

Micro drills originate from conductor board manufacturing for contacting through multiple layers. Although the prepregs used consist of a cured resin and very abrasive glass fibres, uncoated hard metal drills are used with good success. Uncoated micro drills are available

technology and adapted tool shape will be discussed in Chapter 3 for micro end mills.

down to diameters of 20 µm [Ato\_Ad, Ham\_38]. Fig. 1 shows a 30 µm micro drill bit.

Fig. 1. 30 µm micro drill bit with detail of the cutting edge. Grooves from grinding with

jagged edge due to the composite nature of hard metal can be seen.

monocrystalline diamonds with only one cutting edge are used.

work piece and its microstructure are not as affected as in the case of grinding.

processes:

case of micro milling.

**2.2 Geometrical limits of tools** 

bending and buckling may be an issue.

About five years ago, coating of micro end mills started only above 0.3 mm in diameter due to excessive rounding of the cutting edges by the coating layer. Up to this time, the gain of improved wear resistance due to the coating was less favourable than the increase of the cutting force due to the rounding of the cutting edge. Through improved coating process control, allowing thinner and more uniform layers, the relation was reversed. Today, coated micro mills down to 30 µm in diameter and with aspect ratios of 1.5 are commercially available [Hte\_Em], (Fig. 2).

Fig. 2. Left: Coated 30 µm end mill made by Hitachi. Right: Top view.

Fig. 3. Micro end mill 100 µm in diameter and 1 mm in length (AR=10).

Starting at a tool diameter of 100 µm, aspect ratios of up to ten are available now, as displayed in Fig. 3 [Nst-Em, Hte\_Ep2].

Further miniaturization of micro end mills made of hard metal seems to be useless regarding process yield and tool life. Furthermore, an isotropic mechanical behaviour cannot be achieved since hard metal is a composite material consisting of a hard material and a binder phase with very different mechanical properties.

For manufacturing and stability reasons, micro end mills made of monocrystalline diamond are no less than 50 µm in diameter (Fig. 4). Suppliers are mentioned in [Nst\_Di, Med, Möß, Con]. Diamonds are used for very hard and non-iron materials. In contact with iron, the carbon of the diamond would easily diffuse and destroy the tool. An exception occurs in the case of low cutting speeds e. g. during slotting and, hence, low temperatures avoiding diffusion. The advantage of monocrystalline diamond tools is that the cutting edge can be prepared to sharpness nearly at atomic level because diamond is a homogeneous and very hard material. There is much less burr formation on ductile work piece materials than in the case of hard metal tools.

Fig. 4. Left: End mill made of monocrystalline diamond. Right: Detail of the perfect cutting edge.

Micro slotting tools are much more stable than micro end mills because they are not rotationally symmetric and much more rigid. Grinding can be done according to individual requirements (Fig. 5).

Fig. 5. Left: Side view of a micro slotting tool. Right: Measurement of the cutting width (app. 14 µm).

Starting at a tool diameter of 100 µm, aspect ratios of up to ten are available now, as

Further miniaturization of micro end mills made of hard metal seems to be useless regarding process yield and tool life. Furthermore, an isotropic mechanical behaviour cannot be achieved since hard metal is a composite material consisting of a hard material

For manufacturing and stability reasons, micro end mills made of monocrystalline diamond are no less than 50 µm in diameter (Fig. 4). Suppliers are mentioned in [Nst\_Di, Med, Möß, Con]. Diamonds are used for very hard and non-iron materials. In contact with iron, the carbon of the diamond would easily diffuse and destroy the tool. An exception occurs in the case of low cutting speeds e. g. during slotting and, hence, low temperatures avoiding diffusion. The advantage of monocrystalline diamond tools is that the cutting edge can be prepared to sharpness nearly at atomic level because diamond is a homogeneous and very hard material. There is much less burr formation on ductile work piece materials than in the

Fig. 4. Left: End mill made of monocrystalline diamond. Right: Detail of the perfect cutting edge. Micro slotting tools are much more stable than micro end mills because they are not rotationally symmetric and much more rigid. Grinding can be done according to individual

Fig. 5. Left: Side view of a micro slotting tool. Right: Measurement of the cutting width (app.

displayed in Fig. 3 [Nst-Em, Hte\_Ep2].

case of hard metal tools.

requirements (Fig. 5).

14 µm).

and a binder phase with very different mechanical properties.

By micro slotting, minimum sizes of trenches can be reduced to about 15 µm in width at an aspect ratio of about ten (Fig. 6). Such dimensions cannot be achieved by micro milling (Fig. 7).

Fig. 6. PMMA trenches about 15 µm width, 150 µm in depth.

Fig. 7. Left: Mold insert for a cell chip made of brass with slotted trenches. Right: Trench at the base 60 µm in width, aperture angle 3° on both walls, 500 µm in depth.

The material removal rate for slotting is slow. Hence, also the cutting temperatures are low and monocrystalline tools can be used also for ferrous materials.

#### **2.3 Thermal aspects, lubrication and cooling**

Mechanical machining is connected with heat generation. Except in the case of dry machining, fluids are applied for cooling and lubrication to reduce the friction of the cutting edge with the work piece material and to decrease the thermal load of the cutting edge connected with increased wear and diffusion processes. As fluid, either water-based emulsions or oils are used. When using emulsions, bacterial contamination, aging and ecological aspects can involve issues of health and safety.

The fluid can be flushed or applied as mist. For lubrication by mist, a few milliliters of oil per hour are atomized by pressurized air. Oil mist has been preferred recently due to a number of advantages like reduced costs due to handling of smaller amounts of liquid, less storage and disposal costs, no hygienic problems due to bacterial contamination and less cleaning effort for liquid and the work piece. On the other hand, the right dosing of the oil in the air stream is essential especially in the case of micromachining to prevent the sticking of chips to the tool. The available apparatuses, however, lack in exact dosing systems. Sticking of chips leads to additional and fluctuating tool loads and can be an issue for tool failure. Additionally, the work piece surface quality is worse. For this reason, flush lubrication may be the better choice.

#### **3. Tooling aspects: The role of material substrate, coating technology and tool shape**

For micro tools, either hard metal or monocrystalline diamonds are used. Diamond tools are limited to nonferrous and non-carbide forming materials. A perfect cutting edge can be formed either by grinding or ion beam processing [Bor04]. When machining e. g. copper or brass, a very good edge quality without burrs can be achieved.

Hard metal, however, is a composite material. The cutting edge is always jagged (Fig. 8) causing burr formation on ductile materials like most metals.

Fig. 8. Left: Imperfect cutting edge of an uncoated hard metal tool with d=0.25 mm (by Dixi). Right: Cutting edge of a 30 µm-drill bit (by Atom).

#### **3.1 Influence of hard metal substrates**

In general, hard metals consist of a hard phase and a binder phase. For the hard phase, mainly tungsten carbide is used which is basically responsible for the wear resistance. However, also small amounts of tantalum carbide, niobium carbide, chromium carbide, vanadium carbide and titanium carbide are added. These act as grain growth inhibitors during transient liquid-phase sintering and improve the high-temperature properties [Yao\_Wc, Sad99]. Pure carbides cannot be sintered to full density because they would decompose at the necessary high temperatures. Furthermore, they are brittle, and crack propagation resistance is poor. Hence, already small defects in the surface would cause tool failure although a pure tungsten carbide would be desirable under the aspect of wear resistance. Instead, metals exhibiting a limited solubility for carbides at higher temperatures are used as binders. Mostly, cobalt (fcc structure) is used but also nickel and iron are possible. Fig. 9 shows the solubility of Co for WC.

number of advantages like reduced costs due to handling of smaller amounts of liquid, less storage and disposal costs, no hygienic problems due to bacterial contamination and less cleaning effort for liquid and the work piece. On the other hand, the right dosing of the oil in the air stream is essential especially in the case of micromachining to prevent the sticking of chips to the tool. The available apparatuses, however, lack in exact dosing systems. Sticking of chips leads to additional and fluctuating tool loads and can be an issue for tool failure. Additionally, the work piece surface quality is worse. For this reason, flush lubrication may be

**3. Tooling aspects: The role of material substrate, coating technology and** 

brass, a very good edge quality without burrs can be achieved.

causing burr formation on ductile materials like most metals.

Right: Cutting edge of a 30 µm-drill bit (by Atom).

possible. Fig. 9 shows the solubility of Co for WC.

**3.1 Influence of hard metal substrates** 

For micro tools, either hard metal or monocrystalline diamonds are used. Diamond tools are limited to nonferrous and non-carbide forming materials. A perfect cutting edge can be formed either by grinding or ion beam processing [Bor04]. When machining e. g. copper or

Hard metal, however, is a composite material. The cutting edge is always jagged (Fig. 8)

Fig. 8. Left: Imperfect cutting edge of an uncoated hard metal tool with d=0.25 mm (by Dixi).

In general, hard metals consist of a hard phase and a binder phase. For the hard phase, mainly tungsten carbide is used which is basically responsible for the wear resistance. However, also small amounts of tantalum carbide, niobium carbide, chromium carbide, vanadium carbide and titanium carbide are added. These act as grain growth inhibitors during transient liquid-phase sintering and improve the high-temperature properties [Yao\_Wc, Sad99]. Pure carbides cannot be sintered to full density because they would decompose at the necessary high temperatures. Furthermore, they are brittle, and crack propagation resistance is poor. Hence, already small defects in the surface would cause tool failure although a pure tungsten carbide would be desirable under the aspect of wear resistance. Instead, metals exhibiting a limited solubility for carbides at higher temperatures are used as binders. Mostly, cobalt (fcc structure) is used but also nickel and iron are

the better choice.

**tool shape** 

Fig. 9. Pseudo-binary phase diagram of WC-Co from [Upa98].

The mechanical properties of hard metals depend on the binder metal content, which affects mainly hardness and wear resistance, as well as on the average grain size, which is responsible for the flexural strength (Fig. 10).

Especially regarding micro milling, it is very important that the cross section of a tool consists of a sufficient number of hard particles to guarantee isotropic mechanical properties and long tool lifetime. Hence, submicron tungsten carbide powders with an average particle size of 0.2 µm were developed. It is obvious that for practical reasons the critical tool diameter depends on the micro structure of the substrate used. Tools made from submicron hard metal below 30 µm in diameter will not exhibit isotropic properties since a few dozen of hard particles should form the cross section at least.

Fig. 10. Left: Hardness versus grain size for hard metals depending on cobalt content [Wei96]. Right: Dependence of mechanical properties on cobalt content [Exn70].

The wear of the tools is mainly controlled by the binder phase content. The binder phase content is adapted to the type of application: For continuous cut low binder content is sufficient. For interrupted cut or fluctuating load, higher binder content is recommended.

#### **3.2 Coatings**

In the last five years, progress has been made in achieving a low, uniform thickness of wearresistant coatings. Previously, only tools larger than 0.3 mm in diameter were coated since the rounding of the cutting edges due to the coating thickness led to increased cutting forces which annihilated the gain of improved wear resistance (Fig. 11) [Klo05].

Fig. 11. Rounding of cutting edge by a DLC-coating at an end mill of d=0.4 mm by Karnasch.

Today, tools down to 30 µm in diameter are coated (Fig. 12). The coating is quite uniform and below 1 µm in thickness so rounding of the cutting edge can be neglected.

Fig. 10. Left: Hardness versus grain size for hard metals depending on cobalt content [Wei96]. Right: Dependence of mechanical properties on cobalt content [Exn70].

**Property/MPa** 

**4000** 

**6000** 

**2000** 

**Hardness/HV** 

The wear of the tools is mainly controlled by the binder phase content. The binder phase content is adapted to the type of application: For continuous cut low binder content is sufficient. For interrupted cut or fluctuating load, higher binder content is recommended.

In the last five years, progress has been made in achieving a low, uniform thickness of wearresistant coatings. Previously, only tools larger than 0.3 mm in diameter were coated since the rounding of the cutting edges due to the coating thickness led to increased cutting forces

which annihilated the gain of improved wear resistance (Fig. 11) [Klo05].

Fig. 11. Rounding of cutting edge by a DLC-coating at an end mill of d=0.4 mm by

and below 1 µm in thickness so rounding of the cutting edge can be neglected.

Today, tools down to 30 µm in diameter are coated (Fig. 12). The coating is quite uniform

**Hardness/HRA** 

**3.2 Coatings** 

**Grain size/µm** 

Karnasch.

Fig. 12. Coated cutting edge of a 30 µm end mill by Hitachi. Right: Cross section illustration coating thickness and hard metal microstructure.

However, the coating process seems not to be stable all the time. The reproducibility and the results may vary from batch to batch. The formation of droplets certainly must be avoided to prevent coating results having worse machining properties like the ones displayed in Fig. 13 [Klo05].

Fig. 13. Droplet formation on coated micro end mills.

Another issue is the adhesion of the coating. By SEM investigation, micro tools with flaking of coating layers were detected not only at the cutting edge but also in smooth substrate areas for different batches (Fig. 14). An appropriate surface processing is a prerequisite to prevent faults and varying quality of micro tools.

Obviously, an inspection of micro tools by SEM is advisable to guarantee machining results of a constant and good quality.

Different coatings influence the wear resistance of the tool, the rounding of the cutting edge, and the friction between work piece and tool. Monolithic, gradient or layered compositions of coatings are known.

Fig. 14. Faults of adhesion and uniformity of coatings.

It is obvious that the price for micro tools increases strongly with decreasing tool diameter. Apparently the yield of the manufacturing process by grinding decreases significantly for small-diameter tools. Sometimes, undetected cracks cause tool failure. Fig. 15 shows a new 30 µm end mill broken during ultrasonic cleaning for SEM. It seems that cracks and impurities (top of Fig. 15) are present in the cross section, probably originating from the manufacturing process and covered by the coating.

In general, the life time of micro tools is unpredictable and depends strongly on the material machined. Also, the approach of the micro tool to the work piece to get the zero level and the maintenance of a constant engagement across the surface can be an issue due to variation of the flatness of the work piece.

Fig. 15. Left: Cross section of a 30 µm end mill broken when sonicated for SEM analysis. Right: Detail.

#### **3.3 Adapted tool shape for micro milling**

During the past years, attention was paid to optimizing the shape of micro end mills to meet the specific demands of the micro cutting process. Especially for small-diameter end mills, bending, tool deflection and the avoidance of chatter marks on the work piece are of interest to improve the stability of the process.

It is obvious that the price for micro tools increases strongly with decreasing tool diameter. Apparently the yield of the manufacturing process by grinding decreases significantly for small-diameter tools. Sometimes, undetected cracks cause tool failure. Fig. 15 shows a new 30 µm end mill broken during ultrasonic cleaning for SEM. It seems that cracks and impurities (top of Fig. 15) are present in the cross section, probably originating from the

In general, the life time of micro tools is unpredictable and depends strongly on the material machined. Also, the approach of the micro tool to the work piece to get the zero level and the maintenance of a constant engagement across the surface can be an issue due to

Fig. 15. Left: Cross section of a 30 µm end mill broken when sonicated for SEM analysis.

During the past years, attention was paid to optimizing the shape of micro end mills to meet the specific demands of the micro cutting process. Especially for small-diameter end mills, bending, tool deflection and the avoidance of chatter marks on the work piece are of interest

Fig. 14. Faults of adhesion and uniformity of coatings.

manufacturing process and covered by the coating.

variation of the flatness of the work piece.

**3.3 Adapted tool shape for micro milling** 

to improve the stability of the process.

Right: Detail.

Not only was the fluted length reduced to increase the tool shaft cross section and stiffness. Also, the geometry at the intersection of the constant tool shaft diameter and the conical part where the bending moment is maximal was rounded to prevent crack initiation [Uhl06]. Some companies use a specially shaped fluted tip to eliminate chatter marks on the work piece (Fig. 16).

Fig. 16. Comparison of different tool shapes. Left: Conventional design. Right: Design adapted for micro milling [Hte\_Ep1].

#### **4. Machining strategies in respect of micro tool needs**

#### **4.1 Tolerance issues**

Dealing with features of less than 0.01 mm, attention should be paid to tool and machine manufacturing tolerances that are relevant to manufacturing expenses.

In micromachining, tools are often engaged with the full width but not to a certain degree that leads to high load promoting tool deflection. For large formats where a good surface quality of the superficies surfaces is essential, tool change following the depth of the microstructure or caused by tool wear should be avoided since an offset due to tool diameter variation or fluctuating run-out cannot be eliminated.

Quality control of micro features is mostly carried out by optical microscopy. The accuracy of the method should be kept in mind concerning optical resolution depending on magnification and numeric aperture as well as pixel size of the CCD camera used. Regarding the absolute feature size, it can be necessary to shift the microscope table or to stitch multiple pictures for measuring reasons. Specifying tolerances in a range where measuring accuracy or other reasons prevent proving is useless and may increase manufacturing expenses exponentially.

Mostly, quality control is carried out by optical microscopy only at the surface level by edge detection but not at a certain depth. Using tactile devices such as fiber probes [Wer], limitations according to their relevant dimensions must be taken into account.

Generally, tolerances should be one order of magnitude larger than the measuring accuracy and the achievable roughness. Mostly, roughness values for the arithmetic average Ra or highest and lowest peaks within a certain distance like Rt are specified or predefined. With mechanical micro structuring, Ra-values are in the range of 0.2 µm. Typically for micro milling, Rt is 7-10 times higher than Ra, namely in the range of 1-2 µm.

#### **4.2 CAM-software and machine controller issues**

Often, the CAM routines are not able to handle multiple structures according to the special needs of micromilling. For example, the tool path is not generated to meet sequential machining of multiple features but machining is often done in a randomized manner. As a consequence, pins or holes are machined irregularly as the tool moves over a certain area. Lift-off the tool and moving to the next spot take additional time and may cause deviations due to thermal drift when the machining time is very long (the structure displayed in Fig. 18 was machined in three frames, 8h each). Moreover, additional tool loads and bending occurs due to unnecessary sinking in at each new spot. It is obvious that dipping in has a strong impact on the wear and the lifetime of micro tools. In the case of the structure displayed in Fig. 17, sequential machining was forced by insertion of additional frames dividing the field into 19 fields with three lines of pins each.

Fig. 17. Multiple pin array of a fixed-bed reactor with 732 pins, diameter 0.8mm, height 0.8mm, distance in between 0.8 mm, machined in titanium grade 2 using a 0.6 mm micro end mill.

Mostly, quality control is carried out by optical microscopy only at the surface level by edge detection but not at a certain depth. Using tactile devices such as fiber probes [Wer],

Generally, tolerances should be one order of magnitude larger than the measuring accuracy and the achievable roughness. Mostly, roughness values for the arithmetic average Ra or highest and lowest peaks within a certain distance like Rt are specified or predefined. With mechanical micro structuring, Ra-values are in the range of 0.2 µm. Typically for micro

Often, the CAM routines are not able to handle multiple structures according to the special needs of micromilling. For example, the tool path is not generated to meet sequential machining of multiple features but machining is often done in a randomized manner. As a consequence, pins or holes are machined irregularly as the tool moves over a certain area. Lift-off the tool and moving to the next spot take additional time and may cause deviations due to thermal drift when the machining time is very long (the structure displayed in Fig. 18 was machined in three frames, 8h each). Moreover, additional tool loads and bending occurs due to unnecessary sinking in at each new spot. It is obvious that dipping in has a strong impact on the wear and the lifetime of micro tools. In the case of the structure displayed in Fig. 17, sequential machining was forced by insertion of additional frames dividing the field

Fig. 17. Multiple pin array of a fixed-bed reactor with 732 pins, diameter 0.8mm, height 0.8mm, distance in between 0.8 mm, machined in titanium grade 2 using a 0.6 mm micro

limitations according to their relevant dimensions must be taken into account.

milling, Rt is 7-10 times higher than Ra, namely in the range of 1-2 µm.

**4.2 CAM-software and machine controller issues** 

into 19 fields with three lines of pins each.

end mill.

Fig. 18. Top: Sputter mask with approximately 114.500 holes, 50 µm in diameter, made of lead-free brass with a thickness of 100 µm. Bottom: Detail views.

Also, the possibilities of defining machining strategies sometimes are not sufficient for micro milling. Using routines for simple 2D structures, it is not possible to combine a ramp for sinking in the tool and to approach to a contour tangentially to avoid a stop mark from bending of the tool and cutting clear when it stops for turnaround as can be seen in Fig. 19. A smooth tool movement without changes in the feed rate is required. Perpendicular approach of the tool to micro features must be avoided. Unfortunately, it is not easy to meet all these requirements at once. Especially for micro machining of prototypes it is often necessary to make a test piece for preliminary inspection.

The NC unit of the machine must be able to process sufficient numbers of instructions per second. A comparison of different machine control units ranging from 250 to 1000 cycles/s is given in [Wis\_Co]. Together with the definition of the accuracy (e. g. cycle 32 for Heidenhain, see [Hei]) requiring the machine to meet the exact NC data path, the drop of the feed rate caused by tiny details can be dramatic. Here, the influence of high axis acceleration becomes evident. Although already written some years ago, [Rie96] gives a good overview of the interaction of CAM data, data processing and NC-settings.

Fig. 19. Micro gearwheel top of teeth diameter: 800 µm, depth: 300 µm, diameter of column: 160 µm, height of the cone: 140 µm, smallest detail: 100 µm. Mark from clear cutting of the micro end mill at the perimeter of the pin caused by machining strategy and low tool stiffness.

#### **4.3 Machine issues**

#### **4.3.1 Thermal effects**

Especially for large numbers of microstructures, the thermal stability of machines is very important. A constant room temperature within 1 Kelvin and absence of direct solar irradiation are advised. Strict sequential machining of microstructures is a must to prevent irregularities. Often, this has to be forced by additional design work introducing multiple frames to prevent irregular machining.

The construction of the machine and the materials used also have an impact on thermal stability. For the machine bed, KERN uses polymer concrete with a low thermal coefficient of expansion of 10-20\*E-06/K [Epu\_Cr] and much better vibration damping properties than cast iron [Ker\_Ev]. Taking a closer look at the historic development of this class of machines, progress in spindle clamping is evident. Since the machine concept is similar to a c-shaperack and high-strength aluminium is used for the spindle clamping, the shape and fixing position of the clamping to the machine have a high impact on thermal drift due to the high thermal coefficient of expansion of 23\*E-06/K of aluminum. For this reason, we changed the original clamping of an older machine by one made of Invar (=1.7\*E-06/K). Other suppliers use granite and a portal architecture for their machines [Kug\_Mg, Ltu] for low thermal shift.

#### **4.3.2 Clamping and measurement of micro end mills**

The detection of tool length and tool diameter by laser [Blu\_Na] or mechanical dipping onto a force sensor [Blu\_Zp] is problematic for very small tool diameters. Laser measurement is normally only possible above 100 µm tool diameter. According to [Blu\_Ha], the limit was recently shifted down to 10 µm diameter using special laser diodes. Mechanical dipping ends at 50 µm tool diameter.

For such small tools, a very high true running accuracy is essential to make sure both cutting edges are engaged at the same load. Collet chucks must be closed applying a certain torque. Thermal shrinking is superior to mechanical clamping. True running accuracy for thermal shrinkage [Die\_Tg, Schun\_Ce] or hydro stretch chucks [Schun\_Tr] is about 3 µm, however, collet chucks are in the range of 5 to 10 µm only [Far, Ntt\_Er].

Finally, a number of interfaces from tool to the spindle are adding up. For minimization of the run-out it is favourable to use vector-controlled spindles to ensure the same orientation of the chuck inside the spindle.

#### **4.3.3 Spindle speed**

172 Micromachining Techniques for Fabrication of Micro and Nano Structures

Fig. 19. Micro gearwheel top of teeth diameter: 800 µm, depth: 300 µm, diameter of column: 160 µm, height of the cone: 140 µm, smallest detail: 100 µm. Mark from clear cutting of the micro end mill at the perimeter of the pin caused by machining strategy and low tool

Especially for large numbers of microstructures, the thermal stability of machines is very important. A constant room temperature within 1 Kelvin and absence of direct solar irradiation are advised. Strict sequential machining of microstructures is a must to prevent irregularities. Often, this has to be forced by additional design work introducing multiple

The construction of the machine and the materials used also have an impact on thermal stability. For the machine bed, KERN uses polymer concrete with a low thermal coefficient of expansion of 10-20\*E-06/K [Epu\_Cr] and much better vibration damping properties than cast iron [Ker\_Ev]. Taking a closer look at the historic development of this class of machines, progress in spindle clamping is evident. Since the machine concept is similar to a c-shape-

stiffness.

**4.3 Machine issues 4.3.1 Thermal effects** 

frames to prevent irregular machining.

Most machines on the market possess spindles with relatively low rotational speeds of 40- 60.000 rpm [Ker\_Ev, Mak\_22]. For micro machining, often very high numbers of revolution are necessary to achieve reasonable material removal rates. However, much more importance should be attached to questions like tool life, true running accuracy [Weu01, Bis06], the stability and the dynamic behaviour of the machine.

The stability and damping behaviour of the machine are important to avoid vibrations and chatter marks on the work piece surface as well as additional stress of the micro tool due to vibrations. Often, polymer concrete with a very good damping behaviour superior to that of grey cast iron is used for the machine base [Epu\_Fi].

Especially for micro features, the dynamic behaviour, namely the acceleration of the axes, the velocity to the NC-control unit and the maximum number of instructions per seconds are important to maintain a programmed feed rate. In this context, also the definition of how accurately the machine has to meet the calculated tool path is important. If the tolerance is very low, the servo-loop can cause an extreme breakdown of the feed rate. This leads to squeezing of the cutting edges, increased tool wear or even tool rupture. In the last decade, the acceleration could be improved from about 1.2 m/s² to more than 2 g (20m/s²) [Wis\_Ma] also by using hydrostatic drives [Ker\_Ac].

Especially high-frequency spindles lack sufficient torque at lower speed as well as an easyto-operate tool handling system. Mostly, three jaw chucks are used. Measurement of true running accuracy is a must in this case for ensuring a constant engagement of the normally two cutting edges of a micro end mill. Since the feed rate per tooth is far below 1 µm due to machine limitations and since the true running accuracy and cutting edge rounding are not taken into account, it is questionable if very high numbers of revolution in the range of 100.000 rpm and more that are stated e. g. in [Rus08] are appropriate. Instead, a minimal feed per tooth is required to obtain chip formation at all [Duc09].

Often, machining parameters like rotational speed and feed rate cannot be extrapolated. For instance, a speed of 15.000 rpm with a feed rate of 90 mm/min worked fine for micro drilling using a 50 µm drill bit for the sputter mask displayed in Fig. 18 but 40.000 rpm and 240 mm/min did not.

#### **4.4 Design rules**

Referring to the tool shapes with only a short fluted length as displayed in Fig. 3 and Fig. 16, new specific problems can occur. Whereas in Fig. 20 no shape distortion of the spinneret can be observed, a similar negative microstructure (Fig. 21) shows a strong distortion at a depth of 1 mm. Obviously, it is caused by insufficient chip removal from the narrow trenches. The chips are not conveyed by flutes up to the surface level and stick to the tool since oil mist instead of flushing was used for lubrication and cooling.

Fig. 20. Positive spinneret made of brass using Hitachi EPDRP-2002-2-09 with 1° slope, height 2.8 mm.

taken into account, it is questionable if very high numbers of revolution in the range of 100.000 rpm and more that are stated e. g. in [Rus08] are appropriate. Instead, a minimal

Often, machining parameters like rotational speed and feed rate cannot be extrapolated. For instance, a speed of 15.000 rpm with a feed rate of 90 mm/min worked fine for micro drilling using a 50 µm drill bit for the sputter mask displayed in Fig. 18 but 40.000 rpm and

Referring to the tool shapes with only a short fluted length as displayed in Fig. 3 and Fig. 16, new specific problems can occur. Whereas in Fig. 20 no shape distortion of the spinneret can be observed, a similar negative microstructure (Fig. 21) shows a strong distortion at a depth of 1 mm. Obviously, it is caused by insufficient chip removal from the narrow trenches. The chips are not conveyed by flutes up to the surface level and stick to the tool since oil mist

Fig. 20. Positive spinneret made of brass using Hitachi EPDRP-2002-2-09 with 1° slope,

feed per tooth is required to obtain chip formation at all [Duc09].

instead of flushing was used for lubrication and cooling.

240 mm/min did not.

**4.4 Design rules** 

height 2.8 mm.

Fig. 21. Left: Surface level of a negative spinneret made of brass with 1° slope, final depth 2.8 mm using Hitachi EPDRP-2002-2-09 and oil mist. Right: Distortion of the same microstructure at a level of -1 mm due to insufficient chip removal.

For serial production, all machining parameters can be optimized for a certain design to gain maximum output from the process but for prototype or small-scale production the effort exceeds the saving of machining time extremely.

#### **5. Material concerns in mechanical micro machining**

#### **5.1 Machinable materials**

Micro milling or slotting is a very variable process in terms of material classes possessing a high material removal rate. With some limitations on ceramic materials, all kinds of materials like metals, polymers and ceramics can be machined. However, the kind of material machined has a huge impact on machining time, tool wear, surface quality and burr formation.

For micro process devices, often highly corrosion-resistant materials are used. It is not possible to compare the machining behaviour of normal tool steels that are used e. g. for molds for injection molding with aluminum- and copper alloys, with tough materials like stainless steels, nickel base alloys, titanium and tantalum or with brittle materials like ceramics. Mostly, the recommendations given by the suppliers for infeed, lateral engagement, feed rate and number of revolutions depending on tool diameter and tool length are not appropriate for micro tools. Often, there is no defined engagement width but the tool is engage with its full diameter. Trial and error must be applied to find optimal parameters. Mostly it is a good idea to work with low infeed but higher feed rate instead of using the recommended infeed to keep the tool wear low, especially for tough materials.

Ductile materials tend to form burrs at the edges of micro structures. Depending on the resistance of a certain material against chipping and its strength, cold work hardening can be an issue. The machining strategy must be adapted to prevent deformation of very thin and high walls like displayed for stainless steel in Fig. 22. The structure was made of different materials, namely aluminum (Fig. 22), stainless steel (1.4301, Fig. 24) and MACOR (Fig. 25), a machinable ceramic consisting of about 45 % borosilicate glass and 55 % mica acting as micro crack propagators [Mac]. While MACOR and aluminum were easy to machine, stainless steel machining was very challenging. Machining of only a few trenches to the final depth led to cold work hardening. Subsequently, bending of narrow walls and tool deflection occurred (Fig. 23). Finally, the microstructure was machined successfully in stainless steel using three ball-nose tools made by Hitachi with lengths of 1, 2 and 3 mm and a diameter of 0.4 mm. For the first two tools, 36.000 rpm and a feed rate of 1800 mm/min were applied. The infeeds were 0.03 and 0.021 mm, respectively. For the 3 mm long tool the parameters were reduced to a speed of 32.000 rpm, a feed rate of 1600 mm/min and the infeed to 0.011 mm. With the first tool, all channels were machined with the same infeed to 0.6 mm depth followed by machining to a depth of 1.9 mm with the second and to the final depth with the third tool. Flushing with lubricant oil was applied. The wear of the tools was estimated not to be critical for any of the materials.

Fig. 22. Matrix heat exchanger made of aluminum, 14 in 15 comb-shaped interlaced micro channels, 23 mm long each. Channels are 0.4 mm in width; depth at beginning is 2.9 mm, ending at 0.6 mm, wall thickness 0.2 mm.

Fig. 23. Tests of the microstructure displayed in Fig. 22 made of stainless steel 1.4301 without optimization of the machining strategy using a radius end mill. Distortion of the thin walls and tool deflection can clearly be seen.

tool deflection occurred (Fig. 23). Finally, the microstructure was machined successfully in stainless steel using three ball-nose tools made by Hitachi with lengths of 1, 2 and 3 mm and a diameter of 0.4 mm. For the first two tools, 36.000 rpm and a feed rate of 1800 mm/min were applied. The infeeds were 0.03 and 0.021 mm, respectively. For the 3 mm long tool the parameters were reduced to a speed of 32.000 rpm, a feed rate of 1600 mm/min and the infeed to 0.011 mm. With the first tool, all channels were machined with the same infeed to 0.6 mm depth followed by machining to a depth of 1.9 mm with the second and to the final depth with the third tool. Flushing with lubricant oil was applied. The wear of the tools was

Fig. 22. Matrix heat exchanger made of aluminum, 14 in 15 comb-shaped interlaced micro channels, 23 mm long each. Channels are 0.4 mm in width; depth at beginning is 2.9 mm,

Fig. 23. Tests of the microstructure displayed in Fig. 22 made of stainless steel 1.4301 without optimization of the machining strategy using a radius end mill. Distortion of the

estimated not to be critical for any of the materials.

ending at 0.6 mm, wall thickness 0.2 mm.

thin walls and tool deflection can clearly be seen.

Fig. 24. Details of the final heat exchanger made of stainless steel 1.4301. No burr formation at the surface level but some lateral burrs.

Fig. 25. Microstructure of the matrix heat exchanger made of MACOR. Very good shape stability at the edges without flaws.

#### **5.2 Burr removal from ductile materials**

Micro milling of ductile materials is often accompanied by burr formation, especially at the edges of the microstructures. Burrs can be removed e. g. mechanically using small tools, preferably with sharp edges but consisting of a softer material. For steel e. g. spicular tools made of brass are suitable. For microstructures e. g. made of PMMA or PTFE, wood can be used. The disadvantage of this method is the high manual effort. Mostly, it is used only for single channels e. g. for microfluidic devices. For more complex designs of metallic parts, an electrochemical approach, namely electropolishing, is preferred. It can remove burrs from metals possessing a homogeneous microstructure like austenitic stainless steels, nickel and some copper base alloys. Homogeneity means that no precipitations at grain boundaries or a different second phase are present affecting the electrochemical behaviour and forming an electrochemical element in an electrolyte. For instance, in the case of brass, electropolishing works only for lead-free grades. For tool steels with a carbon content of more than 0.1 %, achievement of a good surface quality through electropolishing is not possible because the microstructure consists of a ferritic or martensitic matrix with embedded carbide particles of different chemical compositions. However, with a one order of magnitude smaller inhomogeneity, e. g. in the presence of small precipitations in the grains as in dispersionstrengthened alloys, electropolishing works very well (Fig. 26).

In the case of copper-based alloys, for example conventional alloyed Ampcoloy 940 and 944 [Amp] and dispersion-strengthened alloys like Glidecop or Discup [Dis\_1, Dis\_2], comparable mechanical strengths can be achieved. However, the microstructures are very different. Whereas Glidecop and Discup can be electropolished, Ampcoloy cannot.

Fig. 26. Micro milled structure made of a dispersion strengthened cooper alloy (Glidcop Al-60, [Gli]). Left: After micromilling. Right: After electropolishing.

Generally, electropolishing removes material according to the field line density. At the burrs and edges, the electric field has the highest density. For monitoring, electropolishing must be stopped and the microstructure evaluated by microscopy. After the burrs are removed, the process must be finished to avoid that edges are rounded. At spots without burrs, edges are eroded from beginning. That means, an uniform burr formation is preferred to only partial burrs. On flat surfaces ghost lines are flattened and roughness is decreased by electropolishing.

#### **5.3 Ceramic materials for micromachining**

Beside MACOR, most other ceramic materials like alumina, zirconia and so on can be machined in the CIP (cold isostatic pressed) or presintered state with acceptable tool wear (Fig. 27). At temperatures below normal sinter temperature sintering starts with neck formation between single powder particles. Depending on the residual porosity, the strength of the blanks and tool wear may vary in a wide range. However, the adhesion is much lower than at full density. After machining, the parts are sintered to full density assuming a certain shrinkage. The value of shrinkage must be known or determined by experiments and be taken into account to meet the exact dimensions. By doing so, accuracy within +/- 0.1 % can be achieved.

Another approach consists in using shrink free ceramics [Gre98, Hen99] e. g. based on intermetallic phases like ZrSi2 undergoing an internal oxidation into ZrSiO4 accompanied by an expansion compensating the shrinkage from pore densification. By adjusting the composition of the blend of low-loss binder, inert phase and ZrSi2, the final dimension can be controlled very exactly.

Generally, the material removal rate for ceramics is rather high since a higher infeed and feed rate can be applied. However, machines must be equipped for machining ceramics to protect guideways and scales from damage by abrasive particles.

Fig. 27. Microstructures made of shrink free ZrSi2O4 (left) and zirconia (right)

#### **6. Conclusion**

178 Micromachining Techniques for Fabrication of Micro and Nano Structures

different chemical compositions. However, with a one order of magnitude smaller inhomogeneity, e. g. in the presence of small precipitations in the grains as in dispersion-

In the case of copper-based alloys, for example conventional alloyed Ampcoloy 940 and 944 [Amp] and dispersion-strengthened alloys like Glidecop or Discup [Dis\_1, Dis\_2], comparable mechanical strengths can be achieved. However, the microstructures are very

Fig. 26. Micro milled structure made of a dispersion strengthened cooper alloy (Glidcop Al-

Generally, electropolishing removes material according to the field line density. At the burrs and edges, the electric field has the highest density. For monitoring, electropolishing must be stopped and the microstructure evaluated by microscopy. After the burrs are removed, the process must be finished to avoid that edges are rounded. At spots without burrs, edges are eroded from beginning. That means, an uniform burr formation is preferred to only partial burrs. On flat surfaces ghost lines are flattened and roughness is decreased by

Beside MACOR, most other ceramic materials like alumina, zirconia and so on can be machined in the CIP (cold isostatic pressed) or presintered state with acceptable tool wear (Fig. 27). At temperatures below normal sinter temperature sintering starts with neck formation between single powder particles. Depending on the residual porosity, the strength of the blanks and tool wear may vary in a wide range. However, the adhesion is much lower than at full density. After machining, the parts are sintered to full density assuming a certain shrinkage. The value of shrinkage must be known or determined by experiments and be taken into account to meet the exact dimensions. By doing so, accuracy

Another approach consists in using shrink free ceramics [Gre98, Hen99] e. g. based on intermetallic phases like ZrSi2 undergoing an internal oxidation into ZrSiO4 accompanied by an expansion compensating the shrinkage from pore densification. By adjusting the composition of the blend of low-loss binder, inert phase and ZrSi2, the final dimension can

different. Whereas Glidecop and Discup can be electropolished, Ampcoloy cannot.

strengthened alloys, electropolishing works very well (Fig. 26).

60, [Gli]). Left: After micromilling. Right: After electropolishing.

**5.3 Ceramic materials for micromachining** 

within +/- 0.1 % can be achieved.

be controlled very exactly.

electropolishing.

In this chapter, the recent developments in micromachining were outlined. Especially improvements of machine tool, spindles, clamping technology and tool production can be stated within the last five years, having a big impact on productivity.

In general, micromachining is a very flexible and cost efficient technique, not only for large scale series but also for prototyping and applicable for a wide range of materials.

Due to mechanical and material scientific reasons, further miniaturization of tools seems not to be promising in terms of stability and cost efficiency. Instead, attention should be paid to improvement of reliability of the cutting process and the adaption of machining routines to the specific requirements of sensitive micro tools.

Tolerances in micromachining should be always specified according the real practical demand, with respect to measuring accuracy as well as to achievable surface roughness values.

Especially for replication techniques like micro injection molding and hot embossing, burr formation can be an issue. For some ductile metallic materials the removal of burrs at microstructures can be achieved by electropolishing. Basically, the micro structure of the material has an impact on machinability and surface quality of microstructures after machining and electropolishing. Hence, a homogeneous microstructure is superior to heterogeneous materials.

#### **7. Acknowledgement**

All examples of microstructures displayed in this chapter were made by D. Scherhaufer, T. Wunsch and F. Messerschmidt. Only their professionalism and persistence enabled successful microstructuring of many different prototype designs made of a wide variety of materials.

We acknowledge support by Deutsche Forschungsgemeinschaft and Open Access Publishing Fund of Karlsruhe Institute of Technology.

#### **8. References**


http://www.atom21.co.jp/english/detail.php?act=view&page=0&id= adr&u=mm


http://www.contour-diamonds.com/HTML/ mircromilling.html


http://www.ecka-granules.com/en/ecka-granules/products/catalogue-request/


Mineralgusstechnik/Broschueren/EPUFILL/EPU\_ Epufill\_GB.pdf

Ewa\_Ws, Ewag AG, Industriestr. 4, 4554 Etziken, Switzerland, date of access: 26.09.2011, product information availabe from:

http://www.ewag.com/de/produkte/ schleifen/ws-11.html


Amp, data sheets of different Amcoloy-alloys, date of access: 16.08.2011, available from: http://www.ampcometal.com/common/datasheets/ampco\_alloy\_brochure.pdf Ato\_Ad, Atom Saito Seisakusho micro drill bit ADR-0002, date of access: 16.08.2011,

 http://www.atom21.co.jp/english/detail.php?act=view&page=0&id= adr&u=mm Bis06, G. Bissacco, H. N. Hansen, L. De Chiffre: "Size Effects on Surface Generation in Micro

Blu\_Ha L. Halder, answer be email from 13/07/2011: tools down to 10 µm in diameter with

Blu\_Na, Blum LaserControl Nano NT, date of access: 16.08.2011, available from: http://www.blum-novotest.de/de/messkomponenten/produkte/optische-

Bor04, S. Borek, K. Schauer, M. Füting, A. Heilmann: "Endbearbeitung von Mikrofräsern

Con, Contour, Fine Tooling Ltd., Wedgwood Way, Stevenage, Herts SG1 4QR, UK, date of

Dis\_1, presentation dispersion-strengthened Discup-alloy, date of access: 26.09.2011,

http://www.kupfer-institut.de/symposium/media/pdf/Cu-Werkstoffe durch

 http://www.ecka-granules.com/en/ecka-granules/products/catalogue-request/ Duc09, F. Ducobu, E. Filippi, E. Rivière-Lorphèvre: "Chip Formation and Minimum Chip

Epu\_Cr, overview about different types of polymer concrete possessing different thermal

Epu\_Fi, Epufill – Composite structures with mineral casting filling, p. 3, date of access:

Ewa\_Ws, Ewag AG, Industriestr. 4, 4554 Etziken, Switzerland, date of access: 26.09.2011,

Exn70, H. E. Exner, J. A Gurland: "A review of parameters influencing some mechanical

properties of tungsten carbide cobalt alloy", *Powder Metallurgy*, vol. 13, pp. 13-31,

coefficients of expansion, date of access: 15.09.2011, available from: http://www.epucret.de/en/epument-machine-beds-made-of-mineral-casting/

 http://www.epucret.de/fileadmin/rampf\_downloadcenter/EPUCRET-Mineralgusstechnik/Broschueren/EPUFILL/EPU\_ Epufill\_GB.pdf

http://www.ewag.com/de/produkte/ schleifen/ws-11.html

Far, Farion collet chucks, date of access: 26.09.2011, available from:

Thickness in Micro-milling", *Proc. of the 12th CIRP Conference on Modeling of* 

mittels Ionenstrahltechniken", *wt Werkstattechnik online*, vol. 94, no. 11/12, pp. 600-

Milling of Hardened Tool Steel", *CIRP Annals - Manufacturing Technology*, vol. 55,

**8. References** 

available from:

604, 2004

available from:

PM.pdf

materials/

1970

16.08.2011, available from:

product information availabe from:

Issue 1, pp. 593-596, 2006

special laser diode detectable

werkzeugmessung/lasercontrol-nt.html

access: 26.09.2011, available from:

http://www.diebold-hsk.de/pdfs/04.pdf

Dis\_2, information about Discup can be ordered under:

*Machining Operations*, pp 339-346, 2009

Blu\_Zp, Blum Probe Z-Pico, date of access: 16.08.2011, available from: http://www.blum-novotest.de/uploads/media/Z-Pico\_EN.pdf

 http://www.contour-diamonds.com/HTML/ mircromilling.html Die\_Tg, Diebold ThermoGrip, p. 6, date of access: 16.08.2011, available from:  http://shop.fahrion.de/index.php?id=17&mastersku=HF-A8+f%C3%BCr+CP& pgsn=060280030&sdzgrouping=FAHRION+450E+FM32&L=0&chash=cdf728&&L=1


http://www.ham-tools.com/fileadmin/templates\_ham/


http://www.kern-microtechnic.com/ upload/media/kern\_evo\_e.pdf


http://www.medidia-diamond-tools.com/page/?menu=220


http://www.ns-tool.com/cgi-bin/e/docat/dt.php?j=1&s=9


http://findarticles.com /p/articles/mi\_m3101/is\_n12\_v68/ai\_18389412/


## **Release Optimization of Suspended Membranes in MEMS**

Salvador Mendoza-Acevedo1, Mario Alfredo Reyes-Barranca1, Edgar Norman Vázquez-Acosta1, José Antonio Moreno-Cadenas1 and José Luis González-Vidal2 *1CINVESTAV-IPN, Electrical Engineering Department, 2UAEH, Computing Academic Area, México* 

#### **1. Introduction**

182 Micromachining Techniques for Fabrication of Micro and Nano Structures

Mak\_22, spindle speed of Makino V22, date of access: 15.09.2011, available from:

Med, Medidia GmbH, Alte Poststr. 23, 55743 Idar-Oberstein, date of access: 26.09.2011,

Möß, Mößner GmbH, Kelterstr. 82, 75179 Pforzheim, date of access: 16.08.2011, available

Nst\_Di, Monocrystalline diamond end mills by NS Tool Co. Ltd., Japan, date of access:

Nst\_Em, Coated 100 µm-end mills, AR=10 NS Tool Co. Ltd., Japan, date of access:

Ntt\_Er, ERC precision collet chuck, date of access: 15.09.2011, available from:

Rie96, Asko Riehn: "The Challenges of High Speed Cutting", published in *"Modern Machine* 

Rus08, Rusnaldy, T. J. Ko, H. S. Kim: "An experimental study on microcutting of silicon using a micromilling machine", *Int. J. Adv. Manuf. Technol*., vol. 39, pp. 85-91, 2008 Sad99, R. K. Sadangi et al.: "Grain growth inhibition in Liquid Phase Sintered of Nanophase WC/Co Alloys", *Int. J. of powder metallurgy*, No. 35, vol. 1, pp. 27-33, 1999 Schun\_Ce, Schunk, CELSIO SSF HSK-A 32, p. 78, date of access: 16.08.2011, available from:

Schun\_Tr, Schunk, HSK-E 25 Tribos Polygonal Toolholder, p. 130, date of access: 16.08.2011,

Uhl06, E. Uhlmann, D. Oberschmidt, K. Schauer: "Innovative Fräswerkzeuge für die Mikrozerspanung", *wt Werkstattstechnik online*, issue 96, vol. 1-2, pp. 2-5, 2006 Upa98, G. S. Upadhyaya: *"Cemented Tungsten Carbides – Production Properties and Testing"*,

War00, G. Warnecke, K. Eichgrün, L. Schäfer: "Zuverlässige Serienbauteile aus

Wei96, F. Weiland: "Cerametal S.à.r.l., eine Spitzenadresse nicht nur in Europa. Hartmetall

Weu01, H. Weule, V. Hüntrup, H. Tritschler: "Micro-Cutting of Steel to Meet New

Wis\_Co, Comparison of the velocity of different NC`s, date of access: 26.09.2011, available from: http://www.wissner-gmbh.de/Page/Technologie/ steuerungen.htm Wis\_Ma, acceleration of WISSNER-machines, date of access: 26.09.2011, available from:

Yao\_Wc, Z. Yao, J. J. Stiglich, T. S. Sudarshan: *"Nanograined Tungsten Carbide-Cobalt* 

http://www.wissner-gmbh.de/Page/GAMMA/gamma303.htm

*(WC/Co)"*, p. 8, date of access: 26.09.2011, available from:

http://www.matmod. com/Publications/armor\_1.pdf

und Hartmetallwerkzeuge auf und mit Welt-Niveau" *Revue Technique* 

Requirements in Miniaturization", *CIRP Annals - Manufacturing Technology*, vol. 50,

available from: http://www.schunk.com/schunk\_files/attachments/TRIBOS-

http://www.makino.com/machines/V22/Graphite/

http://www.medidia-diamond-tools.com/page/?menu=220

http://www.ns-tool.com/cgi-bin/e/docat/dt.php?j =1&s=73

http://www.ns-tool.com/cgi-bin/e/docat/dt.php?j=1&s=9

*Shop"*, May 1996, date of access: 15.09.2011, available from: http://findarticles.com /p/articles/mi\_m3101/is\_n12\_v68/ai\_18389412/

http://www.schunk.com/schunk\_files/attachments/WZH-

Noyes Publications, U.S., ISBN 0-8155-1417-4, p. 25, 1998

Wer, Werth fiber probe, date of access: 26.09.2011, available from:

http://www.werth.de/ index.php?id=76&L=1

Hochleistungskeramik", *Ingenieur-Werkstoffe*, Nr. 2, April 2000

http://www.nttooleurope.com/products/erc/

from: http://www.hamoedia.de/html/m-k-d-wz.html

available from:

26.09.2011, available from:

26.09.2011, available from:

Systeme\_\_HSK\_\_DE\_EN.pdf

*Luxembourgeoise*, vol. 2, 1996

Issue 1, pp. 61-64, 2001

RM\_Aufnahmeschaft\_HSK-E\_DE\_EN.pdf

Releasing part(s) of micro-fabricated devices using etching techniques is one of the fundamental post-processing steps in micro-machining and it is important to have a comprehensive concept on how it can be done, since the final result will significantly influence the electrical and mechanical performance of devices. As micro--electromechanical systems (MEMS) are three dimensional structures, they are obtained by eliminating materials commonly used in this technology, such as crystalline silicon, polycrystalline silicon, silicon dioxide and silicon nitride.

Micro-machining can be undertaken by etching the bulk of the substrate or sacrificial layers deposited at the surface of the wafer. Bulk etching removes great quantities of material and is usually applied to obtain thin membranes for pressure sensors, etching from the back surface of the substrate. On the other hand, surface micro-machining is based in the elimination of sacrificial layers deposited at the surface of a substrate underneath the layers that should be active. The present study will deal only with bulk etching.

Therefore, as several materials are used to obtain finally the typical sensors and actuators in MEMS devices, etching must be selective depending on the material that should be removed. There are available methods for material etching based either on gaseous or liquid etchants. The former employs complex and expensive equipment, but with good yield, and the latter is a low cost process, but care should be taken since in some cases it uses toxic or even corrosive solutions and facilities are needed to exhaust or protect from the vapours produced during etching.

However, both kinds of etching processes are widely used in MEMS and CMOS technology and this fact can be conveniently used to match both technologies, so complete MEMS devices can be integrated in the same substrate, reducing fabrication costs. In particular, wet etching can be either isotropic if material is removed uniformly in all crystallographic directions or anisotropic if etching is selective to a given crystallographic plane.

With regard to anisotropic etching, two solutions are commonly used like potassium hydroxide (KOH) and trimethyl ammonium hydroxide (TMAH). Both solutions are regularly used in MEMS technology to obtain structures, such as thin membranes and cantilevers. The present study gives emphasis to the properties of TMAH and how it acts in silicon upon different geometries designed to obtain a thin membrane used in semiconductor gas sensors (SGS) with the objective to reduce etching time so damage to materials different to silicon can be minimized.

As the convenience to keep compatibility between MEMS technology and CMOS integrated circuits technologies has been demonstrated, it is important to apply etching techniques without damaging the layers used as masks. Within the criteria that should be considered is the etching time since optimum processes must be applied so well defined tridimensional structures can be perfectly obtained at any time.

Based on crystallographic concepts and the way TMAH acts during silicon etching, the main purpose of the present study is to demonstrate the effect in etching time using a given geometry for suspended membranes. It will be shown that improvement can be achieved using a specific geometry outline, compared with other options. Knowledge of the influence of size and orientation of the geometric elements on the anisotropic etching made on wafers with (100) orientation can help to optimize the designing process, as will be shown. Simulations were made with the help of specialized software and experimentally confirmed.

#### **2. MEMS etching process compatible with CMOS technology**

Micro--electromechanical systems (MEMS) are devices designed for specific sensing of actuating functions based on tridimensional structures and mechanisms that can be fabricated with a set of known micro--machining steps being compatible with mature CMOS technologies used in micro--electronics for the fabrication of silicon integrated circuits.

Actually, those systems find wide application in diverse disciplines since their main functions as sensors and actuators at micro- and nano scale allow, for instance, size reduction of measurement instrumentation. Besides, the development followed by these devices has significantly influenced the creation of elements, such as optics for telecommunications, RF devices, analytic instrumentation, biomedics, optic systems for image processing, micro--fluidics, mechanical supports, etc.

It is clear that these tridimensional micro-structures (sensors and actuators) need an electronic circuit to operate properly as they interact with the environment to complete a desired function based on input or output signals. Configurations like analogue-to-digital converters, digital memories, artificial neural networks, temperature controllers, etc., are some of the circuits helping in tasks like signal reading or conditioning in domains like digital, analogue or mixed electronic circuits. Therefore, MEMS are considered systems that consist on several blocks with specific features integrated around to deliver or perform a particular function.

One way this integration can be done is to interface the sensors or actuators with separated micro--modules. Each of them (tridimensional structures and modules) are fabricated in a separate chip then connected and packaged together. The advantage of this alternative is the independence in the fabrication technology of each block, applying particular convenient steps to meet specific purposes, i.e. etching on one side and micro-electronics on the other.

Then, both are packaged and interconnected to configure the complete system. However, a great disadvantage that this alternative presents is that stray capacitance is added affecting the performance of the device. Also, packaging of the modules can be highly complex adding the chance of device failure and yield reduction as a consequence (Korvink & Paul, 2006).

cantilevers. The present study gives emphasis to the properties of TMAH and how it acts in silicon upon different geometries designed to obtain a thin membrane used in semiconductor gas sensors (SGS) with the objective to reduce etching time so damage to

As the convenience to keep compatibility between MEMS technology and CMOS integrated circuits technologies has been demonstrated, it is important to apply etching techniques without damaging the layers used as masks. Within the criteria that should be considered is the etching time since optimum processes must be applied so well defined tridimensional

Based on crystallographic concepts and the way TMAH acts during silicon etching, the main purpose of the present study is to demonstrate the effect in etching time using a given geometry for suspended membranes. It will be shown that improvement can be achieved using a specific geometry outline, compared with other options. Knowledge of the influence of size and orientation of the geometric elements on the anisotropic etching made on wafers with (100) orientation can help to optimize the designing process, as will be shown. Simulations were made with the help of specialized software and experimentally confirmed.

Micro--electromechanical systems (MEMS) are devices designed for specific sensing of actuating functions based on tridimensional structures and mechanisms that can be fabricated with a set of known micro--machining steps being compatible with mature CMOS technologies used in micro--electronics for the fabrication of silicon integrated circuits. Actually, those systems find wide application in diverse disciplines since their main functions as sensors and actuators at micro- and nano scale allow, for instance, size reduction of measurement instrumentation. Besides, the development followed by these devices has significantly influenced the creation of elements, such as optics for telecommunications, RF devices, analytic instrumentation, biomedics, optic systems for

It is clear that these tridimensional micro-structures (sensors and actuators) need an electronic circuit to operate properly as they interact with the environment to complete a desired function based on input or output signals. Configurations like analogue-to-digital converters, digital memories, artificial neural networks, temperature controllers, etc., are some of the circuits helping in tasks like signal reading or conditioning in domains like digital, analogue or mixed electronic circuits. Therefore, MEMS are considered systems that consist on several blocks with specific features integrated around to deliver or perform a

One way this integration can be done is to interface the sensors or actuators with separated micro--modules. Each of them (tridimensional structures and modules) are fabricated in a separate chip then connected and packaged together. The advantage of this alternative is the independence in the fabrication technology of each block, applying particular convenient steps to meet specific purposes, i.e. etching on one side and micro-electronics on the other. Then, both are packaged and interconnected to configure the complete system. However, a great disadvantage that this alternative presents is that stray capacitance is added affecting the performance of the device. Also, packaging of the modules can be highly complex adding the chance of device failure and yield reduction as a consequence (Korvink & Paul,

**2. MEMS etching process compatible with CMOS technology** 

image processing, micro--fluidics, mechanical supports, etc.

particular function.

2006).

materials different to silicon can be minimized.

structures can be perfectly obtained at any time.

Therefore, a good choice is to monolithically integrate the system in the same substrate, where sensors or actuators are placed next to the electronics in the same technological process where MEMS and circuitry are fabricated, eliminating extra interconnections, reducing the area of the system and increasing the yield. A clever design can give high compatibility among the different blocks used in the system and the packaging process.

Nevertheless, this apparent simplicity is true only for MEMS fabricated using compatible CMOS technologies. However, care should be taken to eliminate or reduce damage to layers and materials used as protection masks. So, the main goal is to process the chip without risk or damage to the masking layers when micro-machining the typical three dimensional structures needed as sensors or actuators.

Due to this limitation, it is important to optimize the geometric design of the structures in order to assure the physical integrity of the different layers used in the fabrication of an integrated circuit, as well as to keep the compatibility of MEMS fabrication with CMOS technologies (Baltes, 2005). One of the main advantages of this alternative is the reduction in production costs since a high number of devices can be fabricated in batch run.

It should be remembered that micro-machining is, in general, a set of techniques and tools used to obtain tridimensional elements and structures with high precision and good repeatability by adding or removing layers in a controlled way.

A basic technique in MEMS is volumetric wet etching, when an etching is made to a substrate thick enough with a solution prepared with certain elements that react with it, eliminating part of the substrate. The amount of material etched away depends on the kind and conditions of the solution used, like temperature, concentration, etching time, stirring and the crystallographic orientation of the substrate.

Materials used as protective masks also play an important role in the micro-machining process so the desired structure can be readily obtained. Hence it is important to keep in mind the type of substrate and layers that will be used in the fabrication of the integrated circuit that will contain MEMS, since this will indicate which solution must be used for micro-machining (Hsu, 2002).

Usually, volumetric wet etching is used with silicon substrates for the fabrication of structures like micro-cavities, thin membranes, through holes, beams and cantilevers, taking advantage of the structural layers included in CMOS integrated circuits' technology.

This kind of wet etching can be classified as isotropic and anisotropic. The first has a uniform etch rate to the substrate in all crystallographic directions of silicon, and the second is selective on the crystallographic direction, that is, the etch rate is higher on those directions whose atom density is not too high.

#### **3. Suspended membranes and applications**

The case presented here deals with membranes fabricated with anisotropic wet etching of a silicon substrate. These kinds of structures are also thin layers that can operate as sensors and mechanical support for the circuitry. Within the most common applications for these membranes, there are piezoresistive pressure sensors, micro-hotplates and pyroelectric sensors, among others (Barrettino et al., 2004a, 2004b; Capone et al., 2003; Chen et al., 2008; Gaitan et al., 1993; Tabata, 1995)

In general, these structures are fabricated etching the substrate from the back side of a silicon wafer, where no electronic devices are present. This method allows perfect protection of devices placed at the front of the wafer when the compatibility of the layers used is limited, using a mechanical mask or a protective film.

Likewise, this kind of process simplifies the design of the structures that are to be etched, since this step does not affect the geometric configuration of the circuitry at the other side of the wafer. However, a main disadvantage is the large area used for the definition of the membrane due to the characteristics of the anisotropic etching (Madou, 2001). Typical shapes obtained with anisotropic etching using trimethyl ammonium hydroxide and water (TMAHW) are shown in Fig. 1.

Fig. 1. Anisotropic etching of a silicon substrate.

Nevertheless, it is obvious that this technique takes a long time to complete the etching because the wafer is usually too thick and the etch rate of TMAHW is around 1 micron per minute. Also, this back side etching requires simultaneous alignment at the top and back of the wafer (double alignment) in order to perfectly define the needed structure. So this step introduces an extra restriction to achieve the thin membrane.

On the other hand, this membrane can also be obtained with an etching process made at the front surface of the wafer. This is called front bulk etching, a process also frequently used in MEMS. Although it is made at the front surface of the wafer, it is still considered bulk etching since no surface sacrificial layers are removed to obtain the thin membrane, whereas bulk silicon beneath the defined membrane area is etched away. As with the back etching, with the front silicon etching process it is also needed to protect those areas that will not be part of the tridimensional structure.

of devices placed at the front of the wafer when the compatibility of the layers used is

Likewise, this kind of process simplifies the design of the structures that are to be etched, since this step does not affect the geometric configuration of the circuitry at the other side of the wafer. However, a main disadvantage is the large area used for the definition of the membrane due to the characteristics of the anisotropic etching (Madou, 2001). Typical shapes obtained with anisotropic etching using trimethyl ammonium hydroxide and water

(a)

(b)

Nevertheless, it is obvious that this technique takes a long time to complete the etching because the wafer is usually too thick and the etch rate of TMAHW is around 1 micron per minute. Also, this back side etching requires simultaneous alignment at the top and back of the wafer (double alignment) in order to perfectly define the needed structure. So this step

On the other hand, this membrane can also be obtained with an etching process made at the front surface of the wafer. This is called front bulk etching, a process also frequently used in MEMS. Although it is made at the front surface of the wafer, it is still considered bulk etching since no surface sacrificial layers are removed to obtain the thin membrane, whereas bulk silicon beneath the defined membrane area is etched away. As with the back etching, with the front silicon etching process it is also needed to protect those areas that will not be

limited, using a mechanical mask or a protective film.

Fig. 1. Anisotropic etching of a silicon substrate.

part of the tridimensional structure.

introduces an extra restriction to achieve the thin membrane.

(TMAHW) are shown in Fig. 1.

Since the front part of the wafer is where the CMOS electronic devices are placed, metal layers are present as well, and they should be protected against the etching solution. Therefore, here the solution should be modified with additives so this solution can selectively etch only silicon when a etching CMOS post-process is applied to an integrated circuit chip. Some commercial products are available and used for protection purposes as an alternative. With this front etching, the design topology of the desired structures must be modified to expose only the silicon areas that must be etched away, taking care not to unprotect the remaining surface where the electronic devices are present.

Comparing back and front etching, it is obvious that the latter takes less time to complete the membrane since etching is carried out just a few microns down from the surface of the silicon wafer, not through almost all the bulk of the substrate. Besides, the area needed to obtain the complete structure is less than that needed with back etching, despite the characteristics of the anisotropic etching, since a shallower inverted truncated pyramid is obtained.

The membrane obtained with front etching is a suspended structure mechanically supported by two or more thin arms, with a central area as the active part of the membrane. Here it can contain circuitry or some other kind of devices having specific functions for the system's operation. The definition of the membrane's area is given with "etching windows" through which selected silicon areas are left exposed.

This can be achieved using appropriate layer layout to generate a CIF of GDS file used for the CMOS integrated circuit fabrication at the silicon foundry. Hence, the solution will only etch the exposed bulk silicon and the rest of the surface will remain protected with an overglass layer commonly used to isolate the integrated circuit from environment contamination before packaging. So compatibility is maintained to a high degree between the steps needed for MEMS fabrication and CMOS technology (Tabata, 1998; Tea et al., 1997).

As mentioned before, the etching solution used in this work for the delineation of the membranes was TMAHW, which is typical for anisotropic etching. This solution has different etch rates depending on the crystallographic orientation of the silicon substrate. Generally, rates for planes {100}, {110} and {111} are the most used in this kind of task, although other orientations could also be useful for etching purposes as high etch rates can be achieved.

Etch rates using TMAHW depend also on the temperature and concentration of the reactive. In particular, the study presented here was made with a concentration of 10% of TMAH and 90% of deionized water at 80°C, from which an etch rate of approximately 0.72 m/min was obtained for a (100) plane.

TMAH was used since it is highly selective for silicon etching allowing the use of SiO2 as the protective mask against etching. This is an important issue because SiO2 is one of the layers used through the fabrication of CMOS integrated circuits and there is no need to add extra layers that are not used in this technology.

Taking advantage of the photolithographic steps with an appropriate knowledge of the fabrication steps, the etching areas can be easily defined. Therefore, compatibility between etching and the layers used in the fabrication of CMOS integrated circuits is maintained.

It should be mentioned that the designed geometry in the mask will determine the final shape of the anisotropic etching. One of the most important features in this process is the way the etching proceeds in time with regard to the corners of the geometry i.e. concave or convex. It was found that with TMAH, when concave corners are aligned with a {110} plane, etching will stop at the moment when faces with {111} planes coincide, i.e. in the vertex formed by a {100} plane.

On the other hand, convex corners will generate {111} planes as well, but in the vertex of the adjacent planes, etching continues below the corner, etching away other planes and releasing the structure so defined. Fig. 2 shows a mask for a cantilever where the corresponding corners are indicated (Kovacs et al., 1998).

Fig. 2. Geometry of a cantilever illustrating convex and concave corners.

Furthermore, suppose there is a window with an irregular opening, such as the one shown in Fig. 3. The characteristic etching self-alignment with respect to the crystallographic planes due to the anisotropy will be evident.

Fig. 3. Etching resulting from an irregular open window.

On the other hand, convex corners will generate {111} planes as well, but in the vertex of the adjacent planes, etching continues below the corner, etching away other planes and releasing the structure so defined. Fig. 2 shows a mask for a cantilever where the

corresponding corners are indicated (Kovacs et al., 1998).

Fig. 2. Geometry of a cantilever illustrating convex and concave corners.

due to the anisotropy will be evident.

Fig. 3. Etching resulting from an irregular open window.

Furthermore, suppose there is a window with an irregular opening, such as the one shown in Fig. 3. The characteristic etching self-alignment with respect to the crystallographic planes Dealing with anisotropic etching, there is a feature that is important to consider. When the motifs are aligned with {100} planes, {100} walls will be obtained that are etched as the wafer surface.

#### **4. Geometry and optimization of the suspended membranes**

A micro-hotplate was designed to be used in a monolithic CMOS gas sensor which was later fabricated by MOSIS. Then, an anisotropic etching process was performed on the chip using TMAHW, following several formulations that increase the selectivity of the TMAH to avoid damage to the exposed aluminium on the chip caused by the etching solution (Fujitsuka et al., 2004; Sullivan et al, 2000; Yan et al, 2001).

The next figures show the fabricated chip after a TMAHW etching process.

Fig. 4. Fabricated chip after etching.

Fig. 5. Partially etched micro- hotplates.

It was found that the aluminium was sometimes still getting damaged by the solution in an unpredictable way and with a limited repeatability. The damage increased as the etching time was increased, so if the etching time can be reduced by a significant amount, the same applies to the damage of exposed aluminium.

Figure 6 shows photographs from before (left) and after etching, where the exposed aluminium is indicated. The damage can be seen.

Fig. 6. Comparison between before (left) and after etching.

This motivation is the main objective of this study, which comprises etching and mechanics simulations and the etching of the resulting designs. It should be noted that the designs presented are of micro-hotplates with general applications, as mentioned before.

The most common geometry used for micro-hotplates and suspended membranes are shown in Fig. 7. It can be seen in this figure that the central part of the structure is aligned to {110} planes of the substrate, while the supporting arms have an angle of 45° and 135° with respect to the horizontal reference, therefore aligned to <100> directions (Pierret, 1989). This slope allows other planes to be exposed to the etching solution, hence accelerating the etching process helping to the supporting arms' release. However, this process decelerates when the central part of the membrane is reached, as {111} planes are now exposed at this moment. As already indicated, these planes have the lowest etch rate and in this location, the etching proceeds as with convex corners.

From this moment on, etching takes a longer time until the structure is released. If these effects of the etching solution over the main planes exposed by this geometry are analyzed, alternatives can be found for geometries such that planes with a high etching rate can be readily exposed. For instance, if exposing {111} planes can be avoided or reduced; the consequence will be immediately reflected in a reduction in the etching time.

With this motivation in mind, a study of alternatives for the geometry of the micro-hotplate follows, directed to the reduction of the etching time and the corresponding effects. These two objectives were simulated previous to the experimental process with specialized software for anisotropic etching.

Fig. 7. Common suspended membrane geometry.

#### **4.1 Etching simulations**

190 Micromachining Techniques for Fabrication of Micro and Nano Structures

It was found that the aluminium was sometimes still getting damaged by the solution in an unpredictable way and with a limited repeatability. The damage increased as the etching time was increased, so if the etching time can be reduced by a significant amount, the same

Figure 6 shows photographs from before (left) and after etching, where the exposed

This motivation is the main objective of this study, which comprises etching and mechanics simulations and the etching of the resulting designs. It should be noted that the designs

The most common geometry used for micro-hotplates and suspended membranes are shown in Fig. 7. It can be seen in this figure that the central part of the structure is aligned to {110} planes of the substrate, while the supporting arms have an angle of 45° and 135° with respect to the horizontal reference, therefore aligned to <100> directions (Pierret, 1989). This slope allows other planes to be exposed to the etching solution, hence accelerating the etching process helping to the supporting arms' release. However, this process decelerates when the central part of the membrane is reached, as {111} planes are now exposed at this moment. As already indicated, these planes have the lowest etch rate and in this location,

From this moment on, etching takes a longer time until the structure is released. If these effects of the etching solution over the main planes exposed by this geometry are analyzed, alternatives can be found for geometries such that planes with a high etching rate can be readily exposed. For instance, if exposing {111} planes can be avoided or reduced; the

With this motivation in mind, a study of alternatives for the geometry of the micro-hotplate follows, directed to the reduction of the etching time and the corresponding effects. These

presented are of micro-hotplates with general applications, as mentioned before.

consequence will be immediately reflected in a reduction in the etching time.

applies to the damage of exposed aluminium.

aluminium is indicated. The damage can be seen.

Fig. 6. Comparison between before (left) and after etching.

the etching proceeds as with convex corners.

Features considered in this study for geometry optimization are: a) width of the membrane supporting arms; b) dimensions of the thin membrane; c) orientation of the thin membrane with respect to crystalline planes. Simulations with these considerations were first made with the AnisE software from Intellisuite. The base geometry (A) for the suspended membrane is shown in Fig. 8, having simple dimension ratios among the different elements of the membrane, such as supporting arms, etching windows and membrane area. During simulations, the bulk material considered was silicon and the masking material was exclusively silicon dioxide.

Fig. 8. Dimensions of the base membrane in µm. Geometry A.

First, if the width of supporting arms is increased, it was found that an overlap of the resulting etched areas must exist underneath the arms, proceeding from the exposed silicon windows. This allows for the membrane to be released, otherwise, only four rectangular and separated cavities will be obtained. The required etch overlap is shown in Fig. 9. Due to under etching – always present during the process – this overlap can be a minimum, enough for the supporting arms to be released.

Fig. 9. Geometry A. Etching areas (solid lines) and etching overlaps (shadowed).

A 102 min etching time for a complete membrane release was obtained after simulating with the geometry shown in Fig. 8 (Geometry A), with an etch pit depth of about 80m. It should be noted from this figure that the etch overlaps extend only across the supporting arms, such that when they are released the substrate under the thin membrane presents {111} plane faces to the etching solution, with the same dimensions as the membrane.

Therefore, after release of the supporting arms, the etch rate slows down taking a long time for releasing the thin membrane from the substrate. Then it can be concluded that planes generated at the corners below the supporting arms mainly contribute to the expected etching.

Considering this fact, another geometry (Geometry B) was tested including important overlaps, but that can also avoid features oriented parallel or perpendicular to <110> orientations that can generate {111} planes. It is expected a time reduction in the etching process with this modification, shown in Fig. 10.

As can be seen, the original geometry was rotated 45° with respect to the {110} plane reference, keeping the same area. The result obtained from the simulation of this new geometry was an 18% time reduction, that is, the membrane was completely released in 82 min.

One particularity of the geometry shown in Fig. 10 is the reduction of exposed {111} planes, since with this alternative, edges being parallel or perpendicular to {110} planes are avoided. This reduces both the bulk silicon to be etched away and the etching time.

Next, a new geometry (shown in Fig. 11a) was explored and will be identified as Geometry C. The difference with respect to geometries A and B, respectively, is that although the membrane is also rotated 45°, the supporting arms are aligned along the edges of the membrane. After simulation, a 27% etch time reduction compared to the results from Geometry A was obtained, since the thin membrane was released after 75 min.

First, if the width of supporting arms is increased, it was found that an overlap of the resulting etched areas must exist underneath the arms, proceeding from the exposed silicon windows. This allows for the membrane to be released, otherwise, only four rectangular and separated cavities will be obtained. The required etch overlap is shown in Fig. 9. Due to under etching – always present during the process – this overlap can be a minimum, enough

Fig. 9. Geometry A. Etching areas (solid lines) and etching overlaps (shadowed).

plane faces to the etching solution, with the same dimensions as the membrane.

process with this modification, shown in Fig. 10.

A 102 min etching time for a complete membrane release was obtained after simulating with the geometry shown in Fig. 8 (Geometry A), with an etch pit depth of about 80m. It should be noted from this figure that the etch overlaps extend only across the supporting arms, such that when they are released the substrate under the thin membrane presents {111}

Therefore, after release of the supporting arms, the etch rate slows down taking a long time for releasing the thin membrane from the substrate. Then it can be concluded that planes generated at the corners below the supporting arms mainly contribute to the expected

Considering this fact, another geometry (Geometry B) was tested including important overlaps, but that can also avoid features oriented parallel or perpendicular to <110> orientations that can generate {111} planes. It is expected a time reduction in the etching

As can be seen, the original geometry was rotated 45° with respect to the {110} plane reference, keeping the same area. The result obtained from the simulation of this new geometry was an 18% time reduction, that is, the membrane was completely released in 82

One particularity of the geometry shown in Fig. 10 is the reduction of exposed {111} planes, since with this alternative, edges being parallel or perpendicular to {110} planes are avoided.

Next, a new geometry (shown in Fig. 11a) was explored and will be identified as Geometry C. The difference with respect to geometries A and B, respectively, is that although the membrane is also rotated 45°, the supporting arms are aligned along the edges of the membrane. After simulation, a 27% etch time reduction compared to the results from

This reduces both the bulk silicon to be etched away and the etching time.

Geometry A was obtained, since the thin membrane was released after 75 min.

for the supporting arms to be released.

etching.

min.

Fig. 10. Geometry B. a) Membrane rotated 45° with respect to (110) plane reference; b) Etch overlap.

The reason for the efficiency increase for silicon etching is because with Geometry C there are less {111} planes generated at the perimeter of the thin membrane, allowing the underneath silicon to be etched from the beginning of the process, not after the supporting arms are first released.

According to the simulation, the etched pit is approximately 56µm deep. The difference between the etched depths obtained with geometries A and B can be attributed to the exposure of larger {110} planes, among others, which have a greater etch rate. This is illustrated with the overlaps shown in Fig. 11b.

Fig. 11. a) Geometry C; b) Etch overlap.

An alternative for this last geometry is presented in Fig. 12a, where additional supporting arms were added. This will be identified as Geometry D. The purpose for these extra supporting arms is to give mechanical support to the thin membrane so any damage can be prevented if an undesired vibration is suddenly present on the chip. After simulation, this modification showed no improvement in etching time, since the membrane was released also in 75 min with a depth of about 56µm for the etched pit. So, compared with Geometry C, it can be considered that the only advantage is the improvement in mechanical support. From Fig. 12b, the difference between the etch overlap areas of Geometry C and Geometry D can be clearly seen.

Fig. 12. a) Geometry D; b) Etching areas and etching overlaps.

Although there are no overlaps at the centre, a little substrate area is left (indicated as a thin cross outside the overlaps) that can be rapidly etched away due to its small cross section and the multiple planes present at the vertices of the membrane and the supporting arms.

#### **4.2 Mechanical simulations**

Based in a finite element analysis made with COMSOL, the behaviour of the suspended membranes was simulated with each of the geometries described before. Also, in this study it is important to know the weight that the membrane must support. As with restrictions indicated during the mechanical simulation, the extremes of the supporting arms and outer sides of the membrane were set as fixed; the remaining structure should have free movement. The main purpose of the present study was to determine the deformation and stress that exist in the alternative geometries, for comparison purposes.

Geometry A has been widely used and reported in literature and as so, it will be used as the reference geometry to be compared with other geometries. Variables, such as deformation and Von Mises stress, were obtained after simulation in order to evaluate all the membranes, so it can be determined if the proposed modifications introduce some mechanical failure. During simulation, a force equal to the corresponding weight of the

An alternative for this last geometry is presented in Fig. 12a, where additional supporting arms were added. This will be identified as Geometry D. The purpose for these extra supporting arms is to give mechanical support to the thin membrane so any damage can be prevented if an undesired vibration is suddenly present on the chip. After simulation, this modification showed no improvement in etching time, since the membrane was released also in 75 min with a depth of about 56µm for the etched pit. So, compared with Geometry C, it can be considered that the only advantage is the improvement in mechanical support. From Fig. 12b, the difference between the etch overlap areas of Geometry C and Geometry D

 (a) (b) Fig. 12. a) Geometry D; b) Etching areas and etching overlaps.

stress that exist in the alternative geometries, for comparison purposes.

Although there are no overlaps at the centre, a little substrate area is left (indicated as a thin cross outside the overlaps) that can be rapidly etched away due to its small cross section and

Based in a finite element analysis made with COMSOL, the behaviour of the suspended membranes was simulated with each of the geometries described before. Also, in this study it is important to know the weight that the membrane must support. As with restrictions indicated during the mechanical simulation, the extremes of the supporting arms and outer sides of the membrane were set as fixed; the remaining structure should have free movement. The main purpose of the present study was to determine the deformation and

Geometry A has been widely used and reported in literature and as so, it will be used as the reference geometry to be compared with other geometries. Variables, such as deformation and Von Mises stress, were obtained after simulation in order to evaluate all the membranes, so it can be determined if the proposed modifications introduce some mechanical failure. During simulation, a force equal to the corresponding weight of the

the multiple planes present at the vertices of the membrane and the supporting arms.

can be clearly seen.

**4.2 Mechanical simulations** 

membrane was applied considering also the material from which each membrane is made (SiO2) and its thickness (390nm).

For the case of Geometry A, the maximum deformation obtained was 6.357x10-15 m, with a maximum Von Mises stress of 1.229x10-3 MPa, that is significantly below the elastic limit for SiO2 (55 MPa). These results are illustrated in Fig. 13.

Fig. 13. FEM simulation for Geometry A.

On the other side, the maximum deformation and maximum Von Mises stress obtained in the case for Geometry B were 7.38x10-7 m and 1.523x10-5 MPa, respectively. This strain is also below the elastic limit for SiO2. Results are shown in Fig. 14.

Fig. 14. Deformation and stress for Geometry B.

Next, Geometry C showed a deformation of 2.952X10-5 µm with a maximum Von Mises stress of 161X10-3 MPa, showing also that it is a good design from the mechanical point of view. These results are shown in Fig. 15.

Fig. 15. Simulation results for Geometry C.

Now, Geometry D, having two extra supporting arms, shows a maximum deformation of 2.403X10-4 µm with a maximum Von Mises stress of 0.01X10-3 MPa located next to the arms' anchors. This is illustrated in Fig. 16.

As is demonstrated, Geometry D shows the highest deformation compared with Geometries A, B and C, but on the other hand, it resulted in the lowest Von Mises strain.

From these results it can be concluded that this geometry is better for the purposes of the present study and also, as will be demonstrated later, with this geometry the supporting arms are released in a considerably shorter etching time.

#### **4.3 Experimental results**

196 Micromachining Techniques for Fabrication of Micro and Nano Structures

Next, Geometry C showed a deformation of 2.952X10-5 µm with a maximum Von Mises stress of 161X10-3 MPa, showing also that it is a good design from the mechanical point of

Now, Geometry D, having two extra supporting arms, shows a maximum deformation of 2.403X10-4 µm with a maximum Von Mises stress of 0.01X10-3 MPa located next to the arms'

As is demonstrated, Geometry D shows the highest deformation compared with Geometries

From these results it can be concluded that this geometry is better for the purposes of the present study and also, as will be demonstrated later, with this geometry the supporting

A, B and C, but on the other hand, it resulted in the lowest Von Mises strain.

view. These results are shown in Fig. 15.

Fig. 15. Simulation results for Geometry C.

Fig. 16. Mechanical study results of Geometry D.

arms are released in a considerably shorter etching time.

anchors. This is illustrated in Fig. 16.

Silicon substrates were prepared with a thick silicon dioxide layer (390nm). Test geometries as those proposed above (A, C and D) were then defined with photolithography. Following, an etching with a 100 ml solution with 10% TMAHW at 80°C added with 1.36 gr of ammonium peroxidisulfate (APS), was done over 25, 50, 75 and 102 min. APS enhanced the sample finishing. This is a common formulation for etching solutions based on TMAHW.

After these times, the samples were checked with a microscope to verify the correct etching. Fig. 17 shows the advance of the etching process for Geometry A where the characteristic figure predicted during simulation is present at the centre of the membrane caused by the anisotropic attack (far left).

Fig. 17. Geometry A etching photographs.

For Geometry C, Fig. 18 shows the progress of the etching for 25, 50 and 75 min, where the distinctive planes are formed.

Fig. 18. Microphotographs of Geometry C.

In the same way, Geometry D was processed in TMAH and photographs were taken at the prescribed times. Fig. 19. shows how rapidly the flat bottom formed.

Fig. 19. Geometry D during etching at different times.

Next, results from the experimental etching processes applied are shown and discussed, supported with simulation (left) and SEM images (right). *Geometry A*.

*25 minutes*: Here it can be seen that after this time, the supporting arms are completely released, but the central bulk of the membrane is just starting to be etched at the corners.

*50 minutes*: A while later, {111} planes generated due to parallel or perpendicular lines to {110} planes are completely reduced, but there is still contact between the remaining silicon with the membrane.

Next, results from the experimental etching processes applied are shown and discussed,

*25 minutes*: Here it can be seen that after this time, the supporting arms are completely released, but the central bulk of the membrane is just starting to be etched at the corners.

*50 minutes*: A while later, {111} planes generated due to parallel or perpendicular lines to {110} planes are completely reduced, but there is still contact between the remaining silicon

Fig. 19. Geometry D during etching at different times.

*Geometry A*.

with the membrane.

supported with simulation (left) and SEM images (right).

*75 minutes*: After this time of etching, a square based pyramid shape is formed at the centre of the membrane, having planes from which the etching can continue thoroughly. At this time, there is a little pyramid still left.

*102 minutes*: Finally after this time the membrane has been completely released with a bottom cavity surface showing a smooth (100) plane. *Geometry C*.

*25 minutes*: With this geometry, initially the supporting arms are first released exposing {110} planes, that have, as commented before, a high etching rate.

*50 minutes*: Here it can be seen that a column with {110} facets is formed at the centre of the membrane, so etching can continue easily.

*75 minutes*: Finally, the membrane was completely released and the cavity has a smooth surface.

*Geometry D*.

*25 minutes*: After this initial etching time, the supporting arms were completely released, but a complex structure is still present having convex corners that can slow down the etching process.

*50 minutes*: Here it can be seen that a column with {110} facets is formed at the centre of the

*75 minutes*: Finally, the membrane was completely released and the cavity has a smooth

*25 minutes*: After this initial etching time, the supporting arms were completely released, but a complex structure is still present having convex corners that can slow down the etching

membrane, so etching can continue easily.

surface. *Geometry D*.

process.

*50 minutes*: After 25 extra minutes, the substrate of this geometry looks like that obtained after the same time with Geometry C, having also {110} planes with a high etching rate.

*75 minutes*: Finally, the etching process completely released the membrane also with a smooth cavity bottom.

Comparing the simulation figures and the SEM images from the experimental samples above, it is clear that they are nearly the same, both having a smooth bottom of the cavity, nevertheless, there is a difference in time prediction for the membrane release between theory and experiment.

It may seem clear that the time difference can be attributed to features not considered in AnisE [9] regarding the etching apparatus set, as temperature variations, pH level of the


Table 1. Suspended membrane etching comparison.

solution and saturation of the solution during the etching process. Despite this difference, the experimental anisotropic etching follows the same behaviour predicted by simulations. Table 1 summarizes these results.

#### **5. Proposed layout for a micro-hotplate**

From these results, a new layout for a micro- hotplate has been designed using Geometry D as described above. The micro-hotplate contains a micro-heater made with polysilicon, which will heat the structure when a voltage is applied to the heater terminals. There is a temperature sensor also made with polysilicon which will be connected to a control circuit to maintain the temperature of the micro-hotplate to a given value. Fig. 20 shows this design that must guarantee that the silicon substrate is exposed to the etching solution just where it is desired to accomplish the thin membrane (Marshall et al., 1992).

Fig. 20. Layout of micro-hotplate showing temperature sensor and micro-heater.

As mentioned before, this design is intended to be used in a CMOS semiconductor gas sensor that requires a heated thin film to perform the detection. Fabricating a micro-cavity below the heated zone using a MEMS etching process reduces the power needed to achieve the desired temperature and provides thermal isolation to the substrate and signal electronics (Suehle et al., 1993). This can be made following standard CMOS post-process etching steps, keeping compatibility between CMOS technology and MEMS micromachining.

Using the guidelines explained before, it is expected that this design will help reduce damage to the exposed aluminium pads during the fabrication process due to the reduction of etching time.

#### **6. Conclusions**

202 Micromachining Techniques for Fabrication of Micro and Nano Structures

solution and saturation of the solution during the etching process. Despite this difference, the experimental anisotropic etching follows the same behaviour predicted by simulations.

From these results, a new layout for a micro- hotplate has been designed using Geometry D as described above. The micro-hotplate contains a micro-heater made with polysilicon, which will heat the structure when a voltage is applied to the heater terminals. There is a temperature sensor also made with polysilicon which will be connected to a control circuit to maintain the temperature of the micro-hotplate to a given value. Fig. 20 shows this design that must guarantee that the silicon substrate is exposed to the etching solution just where it

Table 1 summarizes these results.

machining.

of etching time.

**5. Proposed layout for a micro-hotplate** 

is desired to accomplish the thin membrane (Marshall et al., 1992).

Fig. 20. Layout of micro-hotplate showing temperature sensor and micro-heater.

As mentioned before, this design is intended to be used in a CMOS semiconductor gas sensor that requires a heated thin film to perform the detection. Fabricating a micro-cavity below the heated zone using a MEMS etching process reduces the power needed to achieve the desired temperature and provides thermal isolation to the substrate and signal electronics (Suehle et al., 1993). This can be made following standard CMOS post-process etching steps, keeping compatibility between CMOS technology and MEMS micro-

Using the guidelines explained before, it is expected that this design will help reduce damage to the exposed aluminium pads during the fabrication process due to the reduction It was theoretically and experimentally demonstrated that the geometry of a micro-hotplate (as the one used in CMOS compatible micro-heaters for semiconductor gas sensors) can be conveniently modified to reduce the etching time with TMAHW, in the order of 20%. Decreasing the etching time is notably useful if these structures are fabricated with materials that can be damaged by the etching solution used. Therefore, selective solutions with repeatability results should be used to protect the integrity of those layers that have an electrical or structural function and need to be protected during the post-process. The purpose is to keep these devices in the etching solution for the least possible time, protecting them from chemical damage. The geometry analysis proposed, based on crystallographic concepts, is a useful strategy in MEMS design taking advantage of anisotropic etching properties. For the cases here presented and using TMAHW, it is concluded that {111} planes and when possible the <110> directions that generate these planes, should be avoided not only in micro-hotplate membranes, but also in structures with different purposes, such as cantilevers. So, there is an optimization of the etching process, especially if it is included as a post-process for integrated circuits fabricated with CMOS technology compatible with MEMS technology.

#### **7. Acknowledgements**

The authors are grateful with Edmundo Rodríguez for the preparation of the etching test masks for photolithography, Benito Nepomuceno for the oxidized silicon substrates preparation and Gaspar Casados for the SEM images of the structures.

#### **8. References**


Agent and Dissolved Si. *RD Review of Toyota CRDL*, Vol.39, No.2, pp. 34-40, ISSN 0385-1508.


## **Micro Abrasive-Waterjet Technology**

H.-T. Liu and E. Schubert *OMAX Corporation, USA* 

#### **1. Introduction**

204 Micromachining Techniques for Fabrication of Micro and Nano Structures

Gaitan, M.; Parameswaran, M.; Johonson, R.B. & Chung, R. (1993). Commercial CMOS

Korvink J. G. & Paul, O. (Eds.) (2006). *MEMS a practical guide to design, analysis and* 

Kovacs, G. T. A.; Maluf, N. I. & Petersen, K. E. (1998). Bulk Micromachining of Silicon,

Madou, M. J. (2001). *Fundamentals of Microfabrication*. CRC-Press, ISBN: 0-8493-9451-1, USA Marshall, J.C.; Parameswaran, M.; Zaghloul, M.E. & Gaitan, M. (1992). High-Level CAD

Tea, N. H.; Milanovic, V.; Zincke, C. A.; Gaitan, M.; Zaghloul, M. E. & Geis, J. (1997). Hybrid

Sullivan, P.; Offord, B. W. & Aklufi, M. E. (2000). Tetra-Methyl Ammonium Hydroxide

Tabata, O. (1995). pH-controlled TMAH Etchants for Silicon Micromachining, *Proceedings of* 

Tabata, O. (1998). Anisotropy and selectivity control of TMAH, *Proceedings of the 11th* 

Yan, G.; Chan, P.C. H.; Hsing, I. M.; Sharma, R. K. & Sin, J. K. O. (2001). An improved

*SPIE,* Vol.1969, doi:10.1117/12.154731 Orlando, FL, USA, April 1993. Hsu, T. R., (2002), *MEMS & Microsystems 1st Ed.,* Mc Graw Hill, ISBN 0-07-239391-2, New

1497-2 & 3-540-21117-9, NY, USA. & Heidelberg, Germany.

Vol.8, No.6, Nov 1992, pp. 10-17, ISSN 8755-3996.

*Device Letters*, Vol.14, No.3, pp. 118-120, ISSN: 07413106.

SPAWAR Systems Center San Diego, San Diego, USA.

*IX.* ISBN: 91-630-3473-5 Stockholm, Sweden, June 1995

4412-X Heidelberg , Germany, Jan. 1998.

*Actuators A: Physical.* Vol. 89, No. 1-2. Mar. 2001

0385-1508.

York, NY, USA, 2002.

Location, Aug 1998.

Agent and Dissolved Si. *RD Review of Toyota CRDL*, Vol.39, No.2, pp. 34-40, ISSN

Foundry Thermal Display for Dynamic Thermal Scene Simulation*, Proceedings of* 

*application*, William Andrew, Inc. & Springer-Verlag GmbH & Co. KG, ISBN 0-8155-

*Proceedings of the IEEE*, Vol.86, No.8, pp. 1536-1551, ISBN: 00189219, Conference

Melds Micromachined Devices with Foundries. *Circuits & Devices Magazine*, IEEE,

Postprocessing Etching for CMOS-Compatible MEMS. *Journal of Microelectromechanical Systems*, Vol.6, No.4, Dec. 1997, pp. 363-372, ISSN 1057-7157. Pierret, R. F. (1989). *Advanced Semiconductor Fundamentals 2nd Ed*., Addison-Wesley Longman Publishing Co., Inc., ISBN 0201053381, Boston, MA, USA. Suehle, J. S.; Cavicchi, R. E.; Gaitan, M. & Semancik, S. (1993). Tin Oxide Gas Sensor

Fabricated Using CMOS Micro-Hotplates and IN-Situ Processing. *IEEE Electron* 

(TMAH) Preferential Etching For Infrared Pixel Arrays. *Technical Document 3097*,

*the 8th International Conference on Solid State Sensors and Actuators, and Eurosensors* 

*International Workshop on Micro Electro Mechanical Systems, MEMS 98*, ISBN: 0-7803-

TMAH Si-etching solution without attacking exposed aluminum. *Sensors and* 

Waterjet technology has come a long way since its commercialization in the late 1970's.1 It has evolved from merely a rough cutting tool to a precision machine tool, competing on equal footing with established tools such as lasers, mechanical milling and routing tools, EDM, ultrasonics, photochemical etching, and various CNC tools.

The versatile waterjet technology inherently has advantages unmatched by most machine tools. Below is a short list of these advantages.



Figure 1 shows a typical abrasive-waterjet system along with a closeup of a nozzle and representative AWJ-cut parts made of various materials. The advantages and disadvantages of AWJs in comparison with lasers, EDM, plasma, flame cutting, and milling can be found in waterjets.org. 2 Waterjet technology has been adopted rapidly in the manufacturing industry to take advantage of its technological and manufacturing merits. 3 Waterjets can be used to machine noncritical components ready for assembly. For critical components requiring highprecision machining, waterjets have been used extensively as a net-shape tool, particularly for difficult-to-machine materials such as hardened steels, alloys, ceramics, and silicon carbide ceramic matrix composites. The versatility of waterjet technology has led to a broad

<sup>1</sup> Waterjet technology generally refers to the use of any of three jets: a water-only jet (WJ), an abrasivewaterjet (AWJ), or an abrasive slurry or suspension jet (ASJ) (Momber & Kovacevic, 1998, Fig. 1.1). For conciseness, the term waterjets is used to refer to all three types of jets unless specified otherwise. 2 http://waterjets.org/index.php?option=com\_content&task=category&sectionid=4&id=46&Itemid= 53#advantages\_of\_waterjet\_machining (8 August 2011).

<sup>3</sup> Waterjet machine tools emerged as the fastest growing segment of the overall machine tool industry in the last decade, and this trend is expected to continue (Frost and Sullivan – "The World Waterjet Cutting Tools Markets" Date Published: 30 Aug 2005 (www.frost.com).

range of capabilities from macro- to micromachining in most materials, which is unmatchable by most machine tools (Liu & McNiel, 2010).

Fig. 1. Typical AWJ system and parts cut with AWJ nozzle (Liu & McNiel, 2010)

The basic erosion mode of waterjets is consistent with micromachining processes.4 Waterjet machining is accomplished by erosion due to individual high-speed water droplets in a water-only jet or abrasives in an AWJ or ASJ impinging onto a workpiece. The size of machined features, such as the diameter of a hole or the kerf width of a slot, is proportional to the diameter of the jet stream in which the water droplets or abrasives are confined (Liu, 2010). Therefore, the key for developing micro abrasive-waterjet (µAWJ) technology is to reduce the diameter of the jet stream for meso-micro machining. Although the feasibility of applying waterjet technology for micromachining has been demonstrated in the mid-1990's, only limited progress has been made since then. The smallest features machined with commercial AWJs are generally larger than 200 µm. Several issues, such as the change in AWJ flow characteristics, poor flowability and clumping of fine abrasives, and nozzle clogging due to the wetting of abrasives caused by backsplash, have presented challenges for further downsizing AWJ nozzles. For example, micro WJs and low-pressure ASJs have been limited to machining relatively soft materials and to the singulation of SD chips for cellular phones, respectively (Jiang et al., 2005). AWJs have been predominantly used for machining applications.

Realizing the inherent advantages of AWJs and their potential for micromachining, considerable effort has been devoted to studying and seeking solutions to the above issues since the mid-2000's. Investigations have concentrated on understanding the physics of the supersonic/subsonic three-phase microfluidics of the abrasive slurry moving through smalldiameter mixing tubes and on issues related to the flow characteristics of fine particles. 5 These efforts have led to the development of novel approaches to resolving issues pertinent to AWJ micromachining. R&D efforts have developed miniature AWJ nozzles and ancillary

<sup>4</sup> Strictly speaking, micromachining is defined as machined features in the micron size range. Machined features around 100 μm are often referred to as meso-scale. In practice, however, features with sizes smaller than 150 μm are often loosely referred to as micro-scale. With that understanding, we chose the practical and rather loose definition of meso- and micro-scale for features between 100 μm and 250 µm and less than and equal to 100 μm, respectively.

<sup>5</sup> Supported by OMAX's R&D funds and NSF SBIR Grants #0944229 (Phase I) and #1058278 (Phase II).

range of capabilities from macro- to micromachining in most materials, which is

Fig. 1. Typical AWJ system and parts cut with AWJ nozzle (Liu & McNiel, 2010)

The basic erosion mode of waterjets is consistent with micromachining processes.4 Waterjet machining is accomplished by erosion due to individual high-speed water droplets in a water-only jet or abrasives in an AWJ or ASJ impinging onto a workpiece. The size of machined features, such as the diameter of a hole or the kerf width of a slot, is proportional to the diameter of the jet stream in which the water droplets or abrasives are confined (Liu, 2010). Therefore, the key for developing micro abrasive-waterjet (µAWJ) technology is to reduce the diameter of the jet stream for meso-micro machining. Although the feasibility of applying waterjet technology for micromachining has been demonstrated in the mid-1990's, only limited progress has been made since then. The smallest features machined with commercial AWJs are generally larger than 200 µm. Several issues, such as the change in AWJ flow characteristics, poor flowability and clumping of fine abrasives, and nozzle clogging due to the wetting of abrasives caused by backsplash, have presented challenges for further downsizing AWJ nozzles. For example, micro WJs and low-pressure ASJs have been limited to machining relatively soft materials and to the singulation of SD chips for cellular phones, respectively (Jiang et al., 2005). AWJs have been predominantly used for

Realizing the inherent advantages of AWJs and their potential for micromachining, considerable effort has been devoted to studying and seeking solutions to the above issues since the mid-2000's. Investigations have concentrated on understanding the physics of the supersonic/subsonic three-phase microfluidics of the abrasive slurry moving through smalldiameter mixing tubes and on issues related to the flow characteristics of fine particles. 5 These efforts have led to the development of novel approaches to resolving issues pertinent to AWJ micromachining. R&D efforts have developed miniature AWJ nozzles and ancillary

4 Strictly speaking, micromachining is defined as machined features in the micron size range. Machined features around 100 μm are often referred to as meso-scale. In practice, however, features with sizes smaller than 150 μm are often loosely referred to as micro-scale. With that understanding, we chose the practical and rather loose definition of meso- and micro-scale for features between 100 μm and 250 µm

5 Supported by OMAX's R&D funds and NSF SBIR Grants #0944229 (Phase I) and #1058278 (Phase II).

unmatchable by most machine tools (Liu & McNiel, 2010).

machining applications.

and less than and equal to 100 μm, respectively.

devices to machine features 100 µm and smaller. These AWJs have been used to machine miniature parts made from a wide range of materials. As a result, the technical feasibility of further downsizing AWJ nozzles has been demonstrated. This article reviews the current status of µAWJ technology developments.

#### **2. Waterjet technology**

As discussed above, waterjet technology generally refers to the use one of any one of three types of jets: a water-only jet (WJ), an abrasive-waterjet (AWJ), and an abrasive slurry or suspension jet (ASJ). A waterjet is generated by pumping high-pressure water through an orifice to achieve supersonic speeds based on Bernoulli's principle. The water is pressurized to 400 MPa and higher on the basis of Pascal's law of hydraulics. Although the basic features of modern waterjet technology were patented in 1939 by Leslie Tirrell (U. S. Patent No. 2,200,587), the lack of a water pump capable of producing the pressures needed for effective cutting limited the applications to surface cleaning and blasting.

Commercially viable 400-MPa pumping systems that took advantage of modern materials and design techniques were not introduced until the late 1970's. These high-pressure WJs, however, were limited to cutting only soft materials. In most cases, the waterjet-cut edges were often frayed with marginal edge quality at best. Abrasive-waterjets entraining abrasives into the WJ stream were developed by several independent R&D teams. 6,7 Abrasive-waterjets were subsequently commercialized in the mid-1980's. For the rest of the 1980's, in the absence of precision control software and hardware, AWJs were mainly used for the rough cutting of materials, particularly those that were difficult to cut using established tools. Since then, much of the research and development has focused on characterizing the emerging technology and realizing its technological and manufacturing merits, as described in Section 1. It was soon realized that, with further advances toward automation, precision, and ease of use, AWJ cutting could have a large market potential in many industrial applications.

Advances in personal computers and peripherals, pumping technology, and materials sciences have greatly benefited the advancement of waterjet technology. In the early 1990's, PC-based CAD/CAM software and controllers were developed and optimized to drive AWJ systems, leading to significant cost reductions for automation while improving the userfriendliness of AWJ machining. The patented "compute first - move later" control algorithm revolutionized AWJ machining (Olsen, 1995). Based on a cutting model (e.g., Zeng & Kim, 1995) and a set of parameters (including nozzle size and pressure; abrasive type, size, and flow rate; material type; path geometry; part thickness; and edge quality), the algorithm calculates the traverse speed of the nozzle along the entire cutting path derived from the CAD drawing of a part. Such a "smart" system greatly reduces the amount of operator input needed to optimize the cutting of a given geometry out of various materials. The path-linebased CAD program is then converted into an executable CAM module (Olsen, 1996). The entire part is machined simply by clicking a "Start" button on the display window without further intervention by the operator.

<sup>6</sup>As a team member of one of such R&D group, the senior author developed a tube catcher to dissipate and terminate the residual erosive power of spent abrasives.

<sup>7</sup> Since Tirrell's 1939 patent, several AWJ patents have been issued (refer to U. S. Patent No. 4,555,872).

One of the major improvements in recent years is the development of direct-drive pumps to replace the inefficient and bulky intensifier, resulting in increased efficiency from about 70% to about 90% along with reduced noise and pressure ripples. The simplicity of direct-drive pumps not only reduces operating and maintenance costs but also simplifies maintenance procedures. The availability of advanced materials together with design optimization has continuously extended the operating lives of pump components. Nowadays, the time between rebuilds for an advanced direct-drive pump has been extended to 1000 hours.8 Waterjets are generally amenable to 3D machining, although they have some limitations. Spent abrasives still possess considerable residual erosive power and could cause damage to the rest of the workpiece being cut, particularly if the waste stream comes in contact with other parts of the workpiece. For 3D parts with complex geometries, materials downstream and along the path of the cutting AWJ must be protected by a "catcher" to capture the spent abrasives, or by sacrificial pieces to absorb or dissipate the residual erosion power of the spent abrasives. For complex 3D parts, however, there is usually not enough room to place the catcher or sacrificial pieces where they are needed. Most commercial AWJ systems machine 2D and 2D+ parts, although there are commercial 5+-axis AWJ systems available for machining simple 3D parts. Alternatively, novel accessories for 2D AWJ systems have been developed that take into account the jet's geometry and imperfections in the raw materials to improve cutting accuracy, facilitating the fabrication of nearly 3D parts and multimode machining. Brief descriptions of representative accessories and their uses are given in subsequent sections.

#### **2.1 Fundamentals**

The PC-based CAD/CAM modules that incorporate a patented controller have been specifically designed to emphasize user friendliness. As a result, the underlying complex physical processes of AWJ flow phenomena are practically transparent to the operators. In this subsection, the fundamentals of AWJ flow phenomena are briefly described.

First, an AWJ is a supersonic, three-phase flow that involves fluid-fluid (air and water), fluid-solid (air, water, and abrasives), and solid-solid (abrasives and materials) interactions. During piercing and cutting, the complex flow is confined in a small space with rapidly changing boundary conditions. In addition, because water is slightly compressible, with a compressibility of approximately 15% at 400 MPa, numerical simulations of such complex flow phenomena have met with little success. CFD models have primarily been used for proof of concept (Liu et al., 1998). Fortunately, a majority of the research work was conducted in laboratory experiments at full scale.

The hydraulic power, *P*, required to generate a high-speed waterjet through an orifice is proportional to the product of the pressure, *p*, and the flow rate, *Q*, or

$$P = p \times Q \text{ / } \text{c} \tag{1}$$

where *c* is a constant that is equal to 60 when *P* is in kW, *p* is in MPa, and *Q* is in l/min, for example. The flow rate can then be estimated based on the Bernoulli principle that the water flow rate through an orifice with a cross-sectional area *A* is

$$Q = c\_d A \sqrt{2p \ / \ \rho} \tag{2}$$

<sup>8</sup> http://www.omax.com/enduromax.php (8 August 2011).

One of the major improvements in recent years is the development of direct-drive pumps to replace the inefficient and bulky intensifier, resulting in increased efficiency from about 70% to about 90% along with reduced noise and pressure ripples. The simplicity of direct-drive pumps not only reduces operating and maintenance costs but also simplifies maintenance procedures. The availability of advanced materials together with design optimization has continuously extended the operating lives of pump components. Nowadays, the time between rebuilds for an advanced direct-drive pump has been extended to 1000 hours.8 Waterjets are generally amenable to 3D machining, although they have some limitations. Spent abrasives still possess considerable residual erosive power and could cause damage to the rest of the workpiece being cut, particularly if the waste stream comes in contact with other parts of the workpiece. For 3D parts with complex geometries, materials downstream and along the path of the cutting AWJ must be protected by a "catcher" to capture the spent abrasives, or by sacrificial pieces to absorb or dissipate the residual erosion power of the spent abrasives. For complex 3D parts, however, there is usually not enough room to place the catcher or sacrificial pieces where they are needed. Most commercial AWJ systems machine 2D and 2D+ parts, although there are commercial 5+-axis AWJ systems available for machining simple 3D parts. Alternatively, novel accessories for 2D AWJ systems have been developed that take into account the jet's geometry and imperfections in the raw materials to improve cutting accuracy, facilitating the fabrication of nearly 3D parts and multimode machining. Brief descriptions of representative accessories and their uses are

The PC-based CAD/CAM modules that incorporate a patented controller have been specifically designed to emphasize user friendliness. As a result, the underlying complex physical processes of AWJ flow phenomena are practically transparent to the operators. In

First, an AWJ is a supersonic, three-phase flow that involves fluid-fluid (air and water), fluid-solid (air, water, and abrasives), and solid-solid (abrasives and materials) interactions. During piercing and cutting, the complex flow is confined in a small space with rapidly changing boundary conditions. In addition, because water is slightly compressible, with a compressibility of approximately 15% at 400 MPa, numerical simulations of such complex flow phenomena have met with little success. CFD models have primarily been used for proof of concept (Liu et al., 1998). Fortunately, a majority of the research work was

The hydraulic power, *P*, required to generate a high-speed waterjet through an orifice is

where *c* is a constant that is equal to 60 when *P* is in kW, *p* is in MPa, and *Q* is in l/min, for example. The flow rate can then be estimated based on the Bernoulli principle that the water

*P p Q/c* (1)

*Q cA 2 <sup>d</sup> p /* (2)

this subsection, the fundamentals of AWJ flow phenomena are briefly described.

given in subsequent sections.

conducted in laboratory experiments at full scale.

proportional to the product of the pressure, *p*, and the flow rate, *Q*, or

flow rate through an orifice with a cross-sectional area *A* is

8 http://www.omax.com/enduromax.php (8 August 2011).

**2.1 Fundamentals** 

where *A = d*2/4, *d* is the orifice diameter, *cd* is the discharge coefficient with a typical value of 0.65, *p* is the pressure, and *ρ* is the water density.

A normal diagram relating *P*, *Q*, d, and *p* as derived from Eqs. (1) and (2) with *cd* = 0.65 is shown in Fig. 2 for a variety of orifice diameters. Knowing any two of the four variables enables determination of the other two. For example, if a cutting pressure of 4000 bar is required using a 0.13 mm orifice, it will draw a flow rate of 0.43 l/min and the stream power will be 1.9 kW (dash-dotted lines). A motor larger than 1.9 kW must be used due to pump inefficiencies.

Fig. 2. Normal diagram of power, flow rate, and pressure

#### **2.2 Key components**

Abrasive-waterjet systems include both hardware and software components. They are integrated to maximize the cutting speed, user friendliness, and cost effectiveness.

#### **2.2.1 Hardware**

A typical AWJ system includes an AWJ nozzle, an abrasive feeding hopper, an X-Y traverse, a high-pressure pump, a motor, a PC, a catcher, and a support tank. Figure 1 illustrates an example of an AWJ system with several key components identified. Depending on the application, the catcher tank that also serves as the support for the X-Y traverse, which usually has a cutting area ranging from about 0.7 m x 0.7 m up to 14 m x 3 m or larger. The X-Y traverse, on which the AWJ nozzle, abrasive hopper, and other accessories may be mounted, has a position accuracy typically from 0.1 mm to 0.03 mm or better.

A high-speed waterjet is formed by using a high-pressure pump, either a hydraulic intensifier or a direct-drive pump, as illustrated in Fig. 3. Early high-pressure cutting systems used hydraulic intensifiers exclusively. At the time, the intensifier was the only pump capable of reliably creating pressures high enough for waterjet machining. A large motor drives a hydraulic pump (typically oil based) that in turn operates the intensifier. Inside the intensifier, hydraulic fluid pumped to about 21 MPa acts on a piston through a series of interconnecting hoses and piping and a bank of complex control valves. The piston pushes a plunger, with an area ratio of 20:1, to pressurize the water to 420 MPa. The intensifier typically uses a double-acting cylinder. The back-and-forth action of the intensifier piston produces a pulsating flow of water at a very high pressure. To help make the water flow more uniformly (thus resulting in a smoother cut), the intensifier pump is typically equipped with an "attenuator" cylinder, which acts as a high-pressure surge vessel. The direct-drive pump is based on the use of a mechanical crankshaft to move any number of individual pistons or plungers back and forth in a cylinder. Check valves in each cylinder allow water to enter the cylinder as the plunger retracts and then exit the cylinder into the outlet manifold as the plunger advances into the cylinder. Direct-drive pumps are inherently more efficient than intensifiers because they do not require a power-robbing hydraulic system. In addition, direct-drive pumps with three or more cylinders can be designed to provide a very uniform pressure output without the use of an attenuator system. Improvements in seal design and materials combined with the wide availability and reduced cost of ceramic valve components now make it possible to operate a crankshaft pump in the 280 to 414 MPa range with excellent reliability. This represents a major breakthrough in the use of such pumps for AWJ cutting. Nowadays, an increasing number of AWJ systems are being sold with the more efficient, quieter, and more easily maintained crankshaft-type pumps.

Abrasive-waterjet systems operating at 600 MPa using intensifier pumps were introduced in the mid-2000's based on the notion that increased pressure means faster cutting. However, such a notion ignores several factors and issues. Specifically, any increase in pressure, for a given pump power, must be matched by a decrease in the volume flow rate, which leads to a decrease in the entrainment and acceleration of abrasives (Fig. 2). In an AWJ cutting system, water is used to accelerate the abrasive particles that perform the cutting operation. It has been shown that the kinetic power of the particles and thus the cutting power of the system is proportional to the hydraulic power of the waterjet. An increase in pressure at the same abrasive load ratio therefore does not yield any gain in cutting performance. Furthermore, high pressure is the enemy of all system plumbing due to material fatigue. As the pressure increases from 400 to 600 MPa, material fatigue significantly reduces the operating lives of components such as high-pressure tubing, seals, and nozzles, leading to considerably higher operating and maintenance costs (Trieb, 2010).9 Finally, an intensifier pump is 28% less efficient than a direct-drive pump. When the above factors are taken into

<sup>9</sup> For example, the maximum von Misses stresses in traditional 3:1 (outside diameter to inside diameter) ratio components will be about 800 MPa to 1200 MPa, respectively. Based on data published in a NASA Technical Note (Smith et al., 1967), for hardened 304 stainless steel, the mean fatigue life will reduce from 35,000 cycles to 5,500 cycles, or a 6.4-fold reduction. As a result, high-pressure components are expected to reduce its life from several years to several months.

usually has a cutting area ranging from about 0.7 m x 0.7 m up to 14 m x 3 m or larger. The X-Y traverse, on which the AWJ nozzle, abrasive hopper, and other accessories may be

A high-speed waterjet is formed by using a high-pressure pump, either a hydraulic intensifier or a direct-drive pump, as illustrated in Fig. 3. Early high-pressure cutting systems used hydraulic intensifiers exclusively. At the time, the intensifier was the only pump capable of reliably creating pressures high enough for waterjet machining. A large motor drives a hydraulic pump (typically oil based) that in turn operates the intensifier. Inside the intensifier, hydraulic fluid pumped to about 21 MPa acts on a piston through a series of interconnecting hoses and piping and a bank of complex control valves. The piston pushes a plunger, with an area ratio of 20:1, to pressurize the water to 420 MPa. The intensifier typically uses a double-acting cylinder. The back-and-forth action of the intensifier piston produces a pulsating flow of water at a very high pressure. To help make the water flow more uniformly (thus resulting in a smoother cut), the intensifier pump is typically equipped with an "attenuator" cylinder, which acts as a high-pressure surge vessel. The direct-drive pump is based on the use of a mechanical crankshaft to move any number of individual pistons or plungers back and forth in a cylinder. Check valves in each cylinder allow water to enter the cylinder as the plunger retracts and then exit the cylinder into the outlet manifold as the plunger advances into the cylinder. Direct-drive pumps are inherently more efficient than intensifiers because they do not require a power-robbing hydraulic system. In addition, direct-drive pumps with three or more cylinders can be designed to provide a very uniform pressure output without the use of an attenuator system. Improvements in seal design and materials combined with the wide availability and reduced cost of ceramic valve components now make it possible to operate a crankshaft pump in the 280 to 414 MPa range with excellent reliability. This represents a major breakthrough in the use of such pumps for AWJ cutting. Nowadays, an increasing number of AWJ systems are being sold with the more

mounted, has a position accuracy typically from 0.1 mm to 0.03 mm or better.

efficient, quieter, and more easily maintained crankshaft-type pumps.

expected to reduce its life from several years to several months.

Abrasive-waterjet systems operating at 600 MPa using intensifier pumps were introduced in the mid-2000's based on the notion that increased pressure means faster cutting. However, such a notion ignores several factors and issues. Specifically, any increase in pressure, for a given pump power, must be matched by a decrease in the volume flow rate, which leads to a decrease in the entrainment and acceleration of abrasives (Fig. 2). In an AWJ cutting system, water is used to accelerate the abrasive particles that perform the cutting operation. It has been shown that the kinetic power of the particles and thus the cutting power of the system is proportional to the hydraulic power of the waterjet. An increase in pressure at the same abrasive load ratio therefore does not yield any gain in cutting performance. Furthermore, high pressure is the enemy of all system plumbing due to material fatigue. As the pressure increases from 400 to 600 MPa, material fatigue significantly reduces the operating lives of components such as high-pressure tubing, seals, and nozzles, leading to considerably higher operating and maintenance costs (Trieb, 2010).9 Finally, an intensifier pump is 28% less efficient than a direct-drive pump. When the above factors are taken into

9 For example, the maximum von Misses stresses in traditional 3:1 (outside diameter to inside diameter) ratio components will be about 800 MPa to 1200 MPa, respectively. Based on data published in a NASA Technical Note (Smith et al., 1967), for hardened 304 stainless steel, the mean fatigue life will reduce from 35,000 cycles to 5,500 cycles, or a 6.4-fold reduction. As a result, high-pressure components are consideration, the hydraulic power, rather than the pressure, is the main factor for cutting performance. Real-world experience has consistently demonstrated that the direct-drive 400- MPa pump outperforms the 600-MPa intensifier pump in material cutting tests and in actual operations under the same electrical power (Henning et al., 2011a).

Fig. 3. Two types of high-pressure pumping mechanisms: an intensifier pump (left) and a complete direct-drive pump system (right) (Liu et al., 2010b)

Unlike a rigid cutting tool where material removal is carried out at the contact surface of a fixed-dimension tool and the workpiece, the AWJ is a flexible stream that diverges with the distance travelled. Consequently, AWJ machining has anomalies that must be compensated for with dedicated hardware components together with software control. For example, AWJ-cut edges are tapered depending on the speed of cutting. On the other hand, the spent abrasives still possess considerable erosive power to remove material along their paths. As a result, a catcher or sacrificial pieces must be used to capture spent abrasives or to prevent them from causing collateral damage to the rest of the workpiece. Therefore, AWJs would not be applicable to machine certain complex 3D parts when the placement of the catcher or sacrificial piece to protect the workpiece exposed to spent abrasives becomes impractical or impossible unless controlled depth milling or etching is used to machine blind features. To broaden the performance of AWJ machining in terms of precision and 3D machining, a host of accessories have been developed. Representative accessories include:

 A Tilt-A-Jet® dynamically tilts the nozzle up to 9 degrees from its vertical position.10 It removes the taper from the part while leaving the taper in the scraps.

Fig. 4. Space Needle model machined with Rotary Axis (Liu & McNiel, 2010)

<sup>10</sup> http://www.omax.com/waterjet-cutting-accessories/Tilt-A-Jet/61 (8 August 2011)


By combining the Rotary Axis and the A-Jet, complex 3D features can readily be machined.

#### **2.2.2 Forms of waterjets**

Waterjets generally take one of three forms: a water-only jet (WJ), an abrasive-waterjet (AWJ), or an abrasive slurry or suspension jets (ASJ). Figure 5 shows drawings of these three jets. On the left is the WJ or the ASJ, depending upon whether the incoming fluid being forced through the small ID orifice is high-pressure water or abrasive slurry. On the right is the AWJ with gravity-fed abrasives entrained into the jet via the Venturi or jet pump effect. The abrasives are accelerated by the high-speed waterjet through the mixing tube.

Fig. 5. Three forms of waterjets (Liu, 2009)

11 http://www.omax.com/accessories-rotary-axis.php (8 August 2011)

<sup>12</sup> http://www.omax.com/waterjet-cutting-accessories/A-Jet/163 (8 August 2011)

For R&D and industrial applications, the majority of waterjet systems are AWJs. Water-only jets find only limited applications in the cutting of very soft materials. In principle, two-phase ASJs have a finer stream diameter, higher abrasive mass flow rate, and faster abrasive speed than do AWJs. As a result, the cutting power of ASJs is potentially up to 5 times greater than that of AWJs at the same operating pressure. Considerable R&D effort has been invested in developing ASJs. However, the high-pressure components, such as orifices, check valves, and seals, through which the high-speed abrasive slurry flows are subject to extremely high wear. The absence of affordable materials with high wear resistance has limited ASJs to pressures around 70 to 140 MPa for industrial applications (Jiang et al., 2005).

#### **2.2.3 Abrasives**

212 Micromachining Techniques for Fabrication of Micro and Nano Structures

 A Rotary Axis or indexer rotates a part (Fig. 4) during AWJ machining around what is commonly referred to as the 4th axis.11 It not only facilitates axisymmetric parts to be machined with AWJs but also enables multimode machining, including turning, facing,

An A-Jet™, or articulated jet, tilts the nozzle up to 60 degrees from its vertical

 A Collision Sensing Terrain Follower measures and adjusts the standoff between the nozzle tip and the workpiece to ensure that an accurate cut is maintained. Warped or randomly curved surfaces can be cut without the need to program in 3D. The collision sensing feature also protects components from becoming damaged if an obstruction is

By combining the Rotary Axis and the A-Jet, complex 3D features can readily be machined.

Waterjets generally take one of three forms: a water-only jet (WJ), an abrasive-waterjet (AWJ), or an abrasive slurry or suspension jets (ASJ). Figure 5 shows drawings of these three jets. On the left is the WJ or the ASJ, depending upon whether the incoming fluid being forced through the small ID orifice is high-pressure water or abrasive slurry. On the right is the AWJ with gravity-fed abrasives entrained into the jet via the Venturi or jet pump effect.

The abrasives are accelerated by the high-speed waterjet through the mixing tube.

position.12 It is capable of beveling, countersinking, and 3D machining.

parting, drilling, milling, grooving, etching, and roughing.

encountered during cutting.

Fig. 5. Three forms of waterjets (Liu, 2009)

11 http://www.omax.com/accessories-rotary-axis.php (8 August 2011)

12 http://www.omax.com/waterjet-cutting-accessories/A-Jet/163 (8 August 2011)

**2.2.2 Forms of waterjets** 

The most commonly used abrasive is garnet because of its optimum performance of cutting power versus cost and its lack of toxicity. It is also a good compromise between cutting power and wear on carbide mixing tubes. There are two types of garnet that are generally used: HPX® and HPA®, which are produced from crystalline and alluvial deposits, respectively.13 HPX garnet grains have a unique structure that causes them to fracture along crystal cleavage lines, producing very sharp edges that enable HPX to outperform its alluvial counterpart. There are other abrasives that are more or less aggressive than garnet.

#### **2.2.4 Speed of water droplets and abrasives**

When machining metals, glasses, and ceramics with AWJs, the material is primarily removed by the abrasives, which acquire high speeds through momentum transfer from the ultrahigh-speed waterjet. Therefore, knowing the speed of the abrasives in AWJs is essential for the performance optimization of AWJs. Several methods, such as laser Doppler anemometers or LDVs, laser transit anemometers or LTAs, dual rotating discs, and others, have been used to measure the speed of the waterjet and/or the abrasive particles to understand the mechanism of momentum transfer in the mixing tube in which the abrasives accelerate (Chen & Geskin, 1990; Roth et al., 2005; Stevenson & Hutchings, 1995; Swanson et al., 1987; Isobe et al., 1988). There is a large spread in the test results mainly due to the difficulty in distinguishing the speeds of the water droplets and of the abrasive particles using optical methods.

A dual-disc anemometer (DDA), based on the time-of-flight principle, was found to be most suitable for measuring the water-droplet and/or abrasive speed (Liu et al., 1999). Data discs made of Lexan and aluminum were successfully used to measure water-droplet speeds in WJs and abrasive particle speeds in AWJs. This was achieved by taking advantage of the large differences in the threshold speeds of water droplets and abrasive particles in eroding the two materials.

Figure 6 illustrates typical measurements of water-droplet speeds generated with an AWJ nozzle operating at several pressures from 207 to 345 MPa in the absence of abrasives. The solid curve and the solid circles correspond to the Bernoulli speed, *VB*, and the DDA measurements, *Vw*, with the abrasive feed port of the nozzle closed (i.e., no air entrainment), respectively. The Bernoulli speed is derived from Eq. (2). The close agreement between the two indicates that the WJ moves through the mixing tube with little touching of the

 13 http://www.barton.com/static.asp?htmltemplate=waterjet\_abrasives.html (8 August 2011)

sidewall. The open circles and dashed curve represent the abrasive speed, *Vwa*, with the feed port open and the corresponding best-fit values.

Fig. 6. Water-droplet speed in WJs exiting AWJ nozzle (Liu et al., 1999)

Measurements of abrasive speeds by entraining Barton 220-mesh garnet into the WJ are illustrated in Fig. 7. The measured and best-fit minimum, maximum, and average abrasive speeds are derived for a range of abrasive mass concentrations *Ca* = 0 to 1.08%.14 The average abrasive speed at *Ca* = 0.4% is 300 m/s, about 61% of the water-droplet speed. The decreasing trend in abrasive speed with *Ca* is evident. The DDA has subsequently been applied to characterize the performance of AWJs (Henning et al., 2011a; Henning et al., 2011b).

Fig. 7. Abrasive speed in AWJs, *p* = 345 MPa (Liu et al., 1999)

#### **2.2.5 Control system**

Historically, AWJ cutting systems have used traditional CNC control systems employing the familiar machine tool "G-code." G-code controllers were developed to move a rigid cutting

<sup>14</sup> *Ca* is defined as the percentage ratio of the abrasive master flow rate in pounds per minute to that of the water flow rate in gallons per minute.

sidewall. The open circles and dashed curve represent the abrasive speed, *Vwa*, with the feed

Measurements of abrasive speeds by entraining Barton 220-mesh garnet into the WJ are illustrated in Fig. 7. The measured and best-fit minimum, maximum, and average abrasive speeds are derived for a range of abrasive mass concentrations *Ca* = 0 to 1.08%.14 The average abrasive speed at *Ca* = 0.4% is 300 m/s, about 61% of the water-droplet speed. The decreasing trend in abrasive speed with *Ca* is evident. The DDA has subsequently been applied to

Historically, AWJ cutting systems have used traditional CNC control systems employing the familiar machine tool "G-code." G-code controllers were developed to move a rigid cutting

<sup>14</sup> *Ca* is defined as the percentage ratio of the abrasive master flow rate in pounds per minute to that of

characterize the performance of AWJs (Henning et al., 2011a; Henning et al., 2011b).

port open and the corresponding best-fit values.

Fig. 6. Water-droplet speed in WJs exiting AWJ nozzle (Liu et al., 1999)

Fig. 7. Abrasive speed in AWJs, *p* = 345 MPa (Liu et al., 1999)

**2.2.5 Control system** 

the water flow rate in gallons per minute.

tool, such as an end mill or mechanical cutter. The feed rate for these tools is generally held constant or varied only in discrete increments for corners and curves. Each time a change in the feed rate is desired, a programming entry must be made.

The AWJ definitely is not a rigid cutting tool; using a constant feed rate will result in severe undercutting or taper on corners and around curves. Moreover, making discrete step changes in the feed rate will also result in an uneven cut where the transition occurs. Changes in the feed rate for corners and curves must be made smoothly and gradually, with the rate of change determined by the type of material being cut, the thickness, the part geometry, and a host of nozzle parameters.

A patented control algorithm "compute first - move later" was developed to compute exactly how the feed rate should vary for a given geometry in a given material to make a precise part (Olsen, 1996). The algorithm actually determines desired variations in the feed rate in very small increments along the tool path to provide an extremely smooth feed rate profile and a very accurate part. Using G-code to convert this desired feed rate profile into actual control instructions for servomotors would require a tremendous amount of programming and controller memory. Instead, the power and memory of the modern PC is used to compute and store the entire tool path and feed rate profile and then directly drive the servomotors that control the X-Y motion. This results in a more precise part that is considerably easier to create than if G-code programming were used.

The advent of personal computing has led to the development of PC-based "smart" software programs for controlling the operations of most modern AWJ systems and a host of accessories for speeding up the cutting while maximizing the precision and quality of cuts. The flexibility of PC programming incorporates the versatility of waterjet technology very well, and the integration of modern PC-based software and hardware takes full advantage of the technological and manufacturing merits of waterjet technology.

One of the advanced software packages used for AWJ machining is the PC-based CAD/CAM.15 It was particularly designed with "ease of use" in mind to allow operators to focus on the work at hand rather than the intricacies of the AWJ's behavior. The software has a built-in cutting model for common engineering materials that assigns each material a machinability index, as illustrated in Fig. 8. Another important input parameter is the edge or surface finish quality, which is defined in levels from Q1 to Q5, with Q1 representing rough cutting and Q5 representing the best edge quality. Figure 9 illustrates a "five-finger" part to demonstrate the five quality levels as a function of cutting speed. Note that the length of the finger is proportional to the cutting speed or the length of cut. The curvature and amplitude of the striation pattern, which is made of grooves caused by jet fluctuations, increases with increases in the cutting speed.16 The amplitude of the striation is also proportional to the abrasive size.

To compensate for the AWJ as a flexible abrasive stream, the control algorithm optimally adjusts the cutting speed along various segments of the tool path. As soon as the cutting begins, the nozzle moves slowly along the lead path such that the piercing is complete at the

16 http://www.micromanufacturing.com/awj.htm (8 August 2011) or

<sup>15</sup> The description of the software package is based on OMAX's Intelli-MAX Software Suite. For detail, refer to (http://www.omax.com/waterjets/intelli-max-software-suite - (8 August 2011)

http://oir.omax.com/media/OMAX\_JetStream\_Simulator.mp4 (8 August 2011)

beginning of the tool path. The nozzle moves relatively fast along straight sections of the tool path and decelerates as corners are approached. Slowing down around corners ensures that there is minimal jet lag as the AWJ cuts the corner. Otherwise, there would be a

Fig. 8. Machinability of common engineering materials (Liu, 2009)

noticeable taper at the corner. The nozzle speeds up again after it passes the corner and accelerates to its maximum speed along straight segments. Figure 10 shows a color-coded diagram that illustrates the various cutting speeds used along a tool path.

The PC-based CAD is a built-in package that either works as a stand-alone program or allows drawings to be imported directly from other programs. It includes tools that are specific to AWJ machining such as automatic or manual lead in/out tools, tool path generation, collision prediction and correction, surface quality assignment tools, and many others. The PC-based CAM has many special features including the cutting model, six levels of cutting quality, taper compensation, estimate of time required to machine a part, report generation, creation and tracking of multiple home locations, rotating, scaling, flipping, and offsetting, among others. The CAM program also offers several special benefits such as part nesting, low-pressure piercing and cutting for brittle and delicate materials, the resizing of parts, and others.

beginning of the tool path. The nozzle moves relatively fast along straight sections of the tool path and decelerates as corners are approached. Slowing down around corners ensures that there is minimal jet lag as the AWJ cuts the corner. Otherwise, there would be a

Fig. 8. Machinability of common engineering materials (Liu, 2009)

parts, and others.

diagram that illustrates the various cutting speeds used along a tool path.

noticeable taper at the corner. The nozzle speeds up again after it passes the corner and accelerates to its maximum speed along straight segments. Figure 10 shows a color-coded

The PC-based CAD is a built-in package that either works as a stand-alone program or allows drawings to be imported directly from other programs. It includes tools that are specific to AWJ machining such as automatic or manual lead in/out tools, tool path generation, collision prediction and correction, surface quality assignment tools, and many others. The PC-based CAM has many special features including the cutting model, six levels of cutting quality, taper compensation, estimate of time required to machine a part, report generation, creation and tracking of multiple home locations, rotating, scaling, flipping, and offsetting, among others. The CAM program also offers several special benefits such as part nesting, low-pressure piercing and cutting for brittle and delicate materials, the resizing of

a) Fingers at qualities Q1 through Q5

b) Striation patterns for Q1 through Q5

Fig. 9. AWJ-machined five-finger part (Liu et al., 2009)

Fig. 10. Cutting speeds along tool path: white & light – fast; blue & dark – slow; green – traverse line (Olsen, 2009)

#### **2.3 Fatigue performance**

Current specifications require that AWJ-cut aluminum and titanium parts that will be used in fatigue-critical aerospace structures undergo subsequent processing to alleviate concerns of degradation in fatigue performance. It has been speculated that the striation patterns induced by AWJs (Fig. 9) could be a source of the initiation of micro-cracks under repeated loading. The requirement of a secondary process for AWJ-machined parts greatly negates the merits (cost effectiveness) of waterjet technology. An R&D program was initiated to revisit the fatigue performance of AWJ-machined aircraft aluminum and titanium parts for fatigue-critical applications by incorporating the most recent advances in waterjet technology (Liu et al., 2009a). 17 "Dog-bone" specimens were prepared by using AWJ and CNC machining. Several "low-cost" secondary processes, including dry-grit blasting with 180-grit aluminum oxide and sanding, were applied to remove the visual appearance of the striation patterns on AWJ-machined edges in an attempt to improve fatigue life. Fatigue tests of dog-bone specimens were conducted in the Fatigue and Fracture laboratory at the Pacific Northwest National Laboratory (PNNL).

Fig. 11. Fatigue life versus *Ra* of aircraft aluminum 2024 T3 (Liu et al., 2009b)

Figure 11 illustrates the results of fatigue tests for the aluminum dog-bone specimens. The abscissa and ordinate are the edge surface roughness, *Ra*, and the fatigue life, respectively. For the AWJ-cut specimens, *Ra* was measured near the bottom of the edge where the amplitude of the striation is at the maximum. The "error bars" in the figure represent the maximum and minimum fatigue life values from the measurements. Except for the dry-grit

<sup>17</sup> This work was a collaboration among OMAX Corporation, Boeing, Pacific Northwest National Laboratory (PNNL), and National Institute of Standards and Technology (NIST).

Current specifications require that AWJ-cut aluminum and titanium parts that will be used in fatigue-critical aerospace structures undergo subsequent processing to alleviate concerns of degradation in fatigue performance. It has been speculated that the striation patterns induced by AWJs (Fig. 9) could be a source of the initiation of micro-cracks under repeated loading. The requirement of a secondary process for AWJ-machined parts greatly negates the merits (cost effectiveness) of waterjet technology. An R&D program was initiated to revisit the fatigue performance of AWJ-machined aircraft aluminum and titanium parts for fatigue-critical applications by incorporating the most recent advances in waterjet technology (Liu et al., 2009a). 17 "Dog-bone" specimens were prepared by using AWJ and CNC machining. Several "low-cost" secondary processes, including dry-grit blasting with 180-grit aluminum oxide and sanding, were applied to remove the visual appearance of the striation patterns on AWJ-machined edges in an attempt to improve fatigue life. Fatigue tests of dog-bone specimens were conducted in the Fatigue and Fracture laboratory at the

**2.3 Fatigue performance** 

Pacific Northwest National Laboratory (PNNL).

Fig. 11. Fatigue life versus *Ra* of aircraft aluminum 2024 T3 (Liu et al., 2009b)

Laboratory (PNNL), and National Institute of Standards and Technology (NIST).

Figure 11 illustrates the results of fatigue tests for the aluminum dog-bone specimens. The abscissa and ordinate are the edge surface roughness, *Ra*, and the fatigue life, respectively. For the AWJ-cut specimens, *Ra* was measured near the bottom of the edge where the amplitude of the striation is at the maximum. The "error bars" in the figure represent the maximum and minimum fatigue life values from the measurements. Except for the dry-grit

17 This work was a collaboration among OMAX Corporation, Boeing, Pacific Northwest National

and AWJ-blasted specimens, Fig. 11 shows that the fatigue life depends mainly on *Ra* whether the dog-bone specimens were machined with AWJs (both as-cut samples and those with secondary sanding) or conventionally.

On the other hand, the grit-blasting process only reduces the *Ra* from 3.4 to 2.3 µm, although most specimens did not break at the gage. The question marks (?) in the figure correspond to the number of test cycles at which the test was terminated, whether the specimen broke at the gage or not even though it was not necessarily a failure (at the gage). The "average" fatigue life for the specimens prepared with the combined process is at least 3 or 4 times longer, respectively, than that for specimens machined by a conventional tool or by AWJs (Liu et al., 2009b). Fatigue tests were also conducted on the aircraft titanium dog-bone specimens, and a similar trend of improvement in the fatigue performance was observed (Liu et al., 2011a).

Subsequent measurements at the x-ray diffraction facility of NIST's Center for Neutron Research have demonstrated that the dry-grit and AWJ blasting processes induce residual compressive stresses on the AWJ-machined edges (Liu et al., 2009b). The residual compressive stresses induced by dry-grit and AWJ blasting were responsible for the fatigue performance improvement. The ability to improve fatigue performance further would have a significant impact on many micromachined parts (e.g., orthopedic implants).

#### **3. µAWJ technology**

Waterjet machining, a top-down manufacturing process, is accomplished by erosion as individual high-speed water droplets in a WJ or abrasive particles in an AWJ/ASJ impinge onto a workpiece. At the microscopic scale, this erosion process by individual water droplets or abrasive particles is consistent with a micromachining process, especially when the droplet or particle size is at micron and submicron scales. At the macroscopic scale, the size of a machined feature, such as the diameter of a hole or the kerf width of a slot, is proportional to the diameter of the jet stream in which the water droplets and abrasives are confined. Therefore, the waterjet stream diameter governs the size of machined features and must be downsized appropriately for meso-micro machining. Figure 12, modified from the drawing presented by Miller (2005), compares the stream diameters of various types of waterjets with the beam diameters of lasers. The solid and dotted portions of the lines signify the normal and outside ranges of stream/beam diameters, respectively. The outside range is either difficult or impractical to achieve for meso-micro machining.

Most of the advantages of waterjet technology discussed in Section 1 apply equally to AWJ meso-micro machining. Although the technical feasibility of applying waterjet technology for meso-micro machining has been demonstrated since the mid-1990's, µAWJ technology remains in the research and development stage (Miller et al., 1996). For cutting a limited set of soft materials, WJs using micron-size orifices have been applied reliably in production environments. In principle, the two-phase ASJs are more aggressive and inherently have smaller stream diameters than the three-phase AWJs. The lack of suitable engineering materials to fabricate highly wear-resistant check valves and other components exposed to the aggressive ASJ has limited its pressures to between about 70 and 140 MPa (Jiang et al., 2005). Most ASJ systems adopt batch feeding of the slurry just upstream of the orifice in order to isolate the abrasives from the high-pressure pump.

Fig. 12. Comparison of stream/beam diameters of waterjets and lasers (modified from Miller, 2005)

Abrasive-waterjets remain the mainstream of waterjet technology. Recent R&D efforts in further downsizing of AWJ nozzles have shown good promise, as described in Section 1. Most of the issues associated with downsizing AWJ nozzles beyond the current state of the art have been identified. Novel processes have been and are still being developed to meet the challenges to resolving these issues.

#### **3.1 Challenges**

Nowadays, the smallest features that can be machined with miniature AWJs are around 200 to 300 µm. Further downsizing of the AWJ presents considerable challenges and difficulties. The supersonic, three-phase AWJ is one of the most complex flow phenomena, particularly because it also involves fluid-fluid, fluid-solid, and solid-solid interactions in a rapidly changing spatial environment. When the ID of the mixing tube is reduced to sizes at which surface tension becomes important, the abrasive slurry has transitioned into a microfluidic flow, leading to additional complexities. In parallel, the size of the abrasive particles must be reduced proportionally with the size of the mixing tube. There are several concerns regarding gravity feeding of fine abrasives, as the flowability of the abrasives deteriorates with particle size distribution. In addition, fine abrasives tend to coagulate or clump together, causing difficulties in achieving consistent and uniform feeding of abrasives to the µAWJ nozzle. Several issues associated with AWJ machining and meso-micro machining are briefly discussed below.

#### **3.1.1 Microfabrication of µAWJ nozzle**

The µAWJ nozzle consists of three key components: the orifice, the mixing tube, and the nozzle body, in which the orifice and mixing tube are housed. The optimum ID ratio of the orifice and mixing tube is between 2 and 3. The optimum aspect ratio of the mixing tube (bore length to ID) is about 100 for production AWJ nozzles, which allows adequate acceleration of the entrained abrasives by the high-speed water droplets to produce a focused AWJ stream with minimal spread at the nozzle exit. Downsizing of the orifice and nozzle body is within the current capability of micromachining technology, as orifices made of sapphire or diamond with IDs of 10 µm and smaller are commercially available. At present, mixing tubes with IDs greater than 200 µm are fabricated by wire EDM. The challenge is to fabricate mixing tubes with IDs less than 200 µm and an adequate aspect ratio.

#### **3.1.2 Microfluidics**

220 Micromachining Techniques for Fabrication of Micro and Nano Structures

Fig. 12. Comparison of stream/beam diameters of waterjets and lasers (modified from

Abrasive-waterjets remain the mainstream of waterjet technology. Recent R&D efforts in further downsizing of AWJ nozzles have shown good promise, as described in Section 1. Most of the issues associated with downsizing AWJ nozzles beyond the current state of the art have been identified. Novel processes have been and are still being developed to meet

Nowadays, the smallest features that can be machined with miniature AWJs are around 200 to 300 µm. Further downsizing of the AWJ presents considerable challenges and difficulties. The supersonic, three-phase AWJ is one of the most complex flow phenomena, particularly because it also involves fluid-fluid, fluid-solid, and solid-solid interactions in a rapidly changing spatial environment. When the ID of the mixing tube is reduced to sizes at which surface tension becomes important, the abrasive slurry has transitioned into a microfluidic flow, leading to additional complexities. In parallel, the size of the abrasive particles must be reduced proportionally with the size of the mixing tube. There are several concerns regarding gravity feeding of fine abrasives, as the flowability of the abrasives deteriorates with particle size distribution. In addition, fine abrasives tend to coagulate or clump together, causing difficulties in achieving consistent and uniform feeding of abrasives to the µAWJ nozzle. Several issues associated with AWJ machining and meso-micro machining are

The µAWJ nozzle consists of three key components: the orifice, the mixing tube, and the nozzle body, in which the orifice and mixing tube are housed. The optimum ID ratio of the orifice and mixing tube is between 2 and 3. The optimum aspect ratio of the mixing tube (bore length to ID) is about 100 for production AWJ nozzles, which allows adequate acceleration of the entrained abrasives by the high-speed water droplets to produce a focused AWJ stream with

Miller, 2005)

**3.1 Challenges** 

briefly discussed below.

**3.1.1 Microfabrication of µAWJ nozzle** 

the challenges to resolving these issues.

As the mixing tube ID is decreased, the capillary effect becomes increasingly important together with the increase in the slurry flow resistance. A meniscus column of water supported by the capillary force will eventually fill the entire bore of the mixing tube. A backsplash of water and abrasives, produced by the front of the AWJ impacting the upper surface of the water column, could reach the interior surface of the nozzle body and the lower portion of the abrasive feed tube, leaving a layer of wet abrasives on these surfaces. After many on-off cycles, buildup of the wet abrasive layers on those surfaces would restrict and eventually block the dry abrasives flowing into the mixing tube. An insufficient quantity of abrasives adversely affects the cut quality. Therefore, optimum nozzle operations to maintain a desired cut quality require the mitigation of the accumulation of wet abrasives on all interior surfaces of the nozzle and the feed tube.

Remedies have been developed, with limited success, to minimize the degree of nozzle clogging by wet abrasives. One of the remedies is the use of vacuum assist and water flushing (Hashish, 2008). The incorporation of such a remedy into µAWJ nozzles is, however, not suitable due to the resultant increase in bulkiness and added complexity in process control.

#### **3.1.3 Feeding of fine abrasives**

As a rule of thumb, the optimum size distribution of abrasives for a given nozzle is such that the maximum abrasive size is no larger than one-third of the mixing tube ID to prevent clogging due to the bridging of two large abrasives. For example, for a mixing tube with a 100-µm ID, the abrasive size must be less than about 30 µm. Although fine abrasive or powder flow is a complex phenomenon, because it is affected by so many variables, it is generally accepted that flowability under gravity feed increases with the particle size (Yu et al., 2011; Liu et al., 2008b). Flowability in this sense simply means the ability of a powder to flow. As the abrasives become finer and finer, they eventually cease to flow through the feed tube when being gravity fed. In addition, fine abrasives tend to coagulate or clump together by static electricity or humidity. Abrasives that clump together would have difficulty flowing from the hopper to the mixing tube. As a result, the abrasive flow is unsteady at best, which leads to inconsistent abrasive feeding or even flow stoppage. Inconsistent abrasive flow would result in deterioration of the cut quality.

#### **3.1.4 Design optimization of traverse systems**

For an AWJ system with a flexible jet stream, the quality of an AWJ-machined part depends on additional parameters inherent to the characteristics of a waterjet. Note that the jet spreads with the distance it travels. Therefore, the profile of the jet, the abrasive size distribution, the abrasive flow rate, and the uniformity of abrasive feed will all affect the machining precision and cut quality. For example, the kerf width of a slot and/or the minimum diameter of a circle are usually used to define the machining precision. The striation pattern and edge taper are often used as qualifiers for the cut quality. For mesomicro machining, special attention must be devoted to optimization of both the design of the X-Y traverse and the characteristics of the µAWJ. Nowadays, linear traverses can be built with nanometer resolution and position accuracy, but their costs increase exponentially with the resolution. In designing a µAWJ system, one must take into consideration the ultimate machining precision achievable with a µAWJ nozzle. Overdesigning the traverse system for AWJ meso-micro machining would have no real benefit and would only inflate the cost of goods.

#### **3.2 Novel solutions under development**

Under support from OMAX's R&D funds and an NSF SBIR Phase I grant, a series of feasibility investigations were conducted regarding the development of µAWJ technology. The success of these investigations, together with the tremendous market potential of µAWJ technology, has led to the award of an NSF SBIR Phase II grant for the development of a µAWJ system prototype for meso-micro machining.

R&D efforts to meet the challenges in developing the µAWJ technology and to resolve various issues associated with nozzle downsizing have led to the development of several novel approaches. Each of the issues described in Section 3.1 has been investigated carefully, and practical solutions have been proposed. Additional solutions are currently being sought.


Success in the above research efforts would facilitate further downsizing of AWJ nozzles with the goal of achieving the capability to machine features around 100 to 50 µm. Another benefit of nozzle downsizing is that multiple small nozzles can be supported by a single pump designed for large nozzles. For the beta nozzles, a multi-nozzle platform has been developed and tested for machining up to four identical parts simultaneously.

Test cuts were made using these nozzles and ancillary devices to fabricate samples from a broad range of engineering materials to demonstrate the versatility of waterjet technology for meso-micro machining. The designs, drawings, and materials of many test samples were furnished by industrial and academic collaborators and customers with specific applications in mind. The finished parts were subsequently returned to the providers for inspection and evaluation. Based on the results of these evaluations, the performance of µAWJ technology for various applications was assessed.

#### **4. AWJ-machined samples and features**

In this section, selected machined samples are presented and discussed to demonstrate the versatility of µAWJ technology for meso-micro machining.

#### **4.1 Basic features**

222 Micromachining Techniques for Fabrication of Micro and Nano Structures

minimum diameter of a circle are usually used to define the machining precision. The striation pattern and edge taper are often used as qualifiers for the cut quality. For mesomicro machining, special attention must be devoted to optimization of both the design of the X-Y traverse and the characteristics of the µAWJ. Nowadays, linear traverses can be built with nanometer resolution and position accuracy, but their costs increase exponentially with the resolution. In designing a µAWJ system, one must take into consideration the ultimate machining precision achievable with a µAWJ nozzle. Overdesigning the traverse system for AWJ meso-micro machining would have no real benefit and would only inflate the cost of

Under support from OMAX's R&D funds and an NSF SBIR Phase I grant, a series of feasibility investigations were conducted regarding the development of µAWJ technology. The success of these investigations, together with the tremendous market potential of µAWJ technology, has led to the award of an NSF SBIR Phase II grant for the development of a

R&D efforts to meet the challenges in developing the µAWJ technology and to resolve various issues associated with nozzle downsizing have led to the development of several novel approaches. Each of the issues described in Section 3.1 has been investigated carefully, and practical solutions have been proposed. Additional solutions are currently being

 A better understanding of the relevant microfluidics has led to the development of novel processes to improve the flowability and uniform feeding of fine abrasives and,

System optimization is being made to develop a prototype of an efficient and cost-

R&D and beta miniature AWJ nozzles, with and without ancillary devices, have been

Success in the above research efforts would facilitate further downsizing of AWJ nozzles with the goal of achieving the capability to machine features around 100 to 50 µm. Another benefit of nozzle downsizing is that multiple small nozzles can be supported by a single pump designed for large nozzles. For the beta nozzles, a multi-nozzle platform has been

Test cuts were made using these nozzles and ancillary devices to fabricate samples from a broad range of engineering materials to demonstrate the versatility of waterjet technology for meso-micro machining. The designs, drawings, and materials of many test samples were furnished by industrial and academic collaborators and customers with specific applications in mind. The finished parts were subsequently returned to the providers for inspection and evaluation. Based on the results of these evaluations, the performance of µAWJ technology

In this section, selected machined samples are presented and discussed to demonstrate the

developed and tested for machining up to four identical parts simultaneously.

goods.

sought.

**3.2 Novel solutions under development** 

thus, mitigate nozzle clogging.

for various applications was assessed.

**4. AWJ-machined samples and features** 

versatility of µAWJ technology for meso-micro machining.

effective precision AWJ machine.

µAWJ system prototype for meso-micro machining.

Efforts are being made to reduce the tolerance stacking error.

developed for meso- and micro-scale test cutting.

The R&D nozzles and ancillary devices discussed above were used to pierce holes and machine slots to demonstrate their ability to machine very small features. Emphasis was made to determine the smallest features that could be machined with these nozzles with and without ancillary devices and also without perfecting the piercing and machining processes. Figure 13 illustrates 3D micrographs of a small hole pierced in a thin stainless steel shim 100 µm thick. The hole was pierced with a 380-m nozzle and a proprietary ancillary device to reduce the effective diameter of the AWJ.18 The diameter of the hole on both the entry and exit sides is around 100 µm. As a rule of thumb, the size of AWJ-cut features is slightly larger than the stream diameter of the AWJ or the mixing tube. In other words, the smallest hole that can be pierced with the 380-m nozzle is approximately 400 µm. The ancillary device has effectively reduced the stream diameter of the AWJ by a factor of 4. The advantage of using such a device is to enable relatively large AWJ nozzles and coarse abrasives to cut parts with features smaller than the diameter of the mixing tube. Note that Fig. 13 represents micrographs of the as-pierced hole. The circularity of the hole could be significantly improved by trepanning after piercing, which would increase the hole diameter slightly.

Fig. 13. Micrographs of holes pierced on 316 stainless steel shim - courtesy of Zygo Corp and Microproducts Breakthrough Institute (Liu et al., 2011b)

 18 The 380mm nozzle is one of the production nozzles. The nozzle size refers to the mixing tube ID which is twice the orifice ID.

Figure 14 illustrates three narrow slots machined in a stainless steel shim 0.25 mm thick. The kerf width of the slots was measured to be approximately 92 µm. These slots were machined with the 380-m nozzle together with another ancillary device to reduce the effective stream diameter of the nozzle. Note that the kerf width is smaller than the width of the fingerprints impressed on the surface of the shim.19

a. Three thin slots b. Micrograph of slot

#### Fig. 14. AWJ-machined slots using a novel ancillary device (Liu & Schubert, 2010**)**

#### **4.2 Components for green energy products**

At the Precision Engineering Research Group (PERG) of MIT, a novel concept was conceived to improve the efficiency of small motors/generators by means of surfacemounted armatures.20 Conventional armatures composed of multiple turns of round wires have a low compaction factor (ratio of volume filled by conducting wires to total allotted conductor volume) because there are gaps among wires resulting in relatively high armature resistance and reduced performance. The concept considered replacing wirewound armatures with slotted copper tubes of appropriate annular dimensions (Trimble, 2011 - patent pending). The smaller the kerf width of the cuts, the higher the compaction factor and better the performance of the armature become. Researchers at PERG ran into problems in machining narrow slots on copper tubes. Due to the high reflectivity of copper, laser cutting splattered. On the other hand, the tube geometry prohibited the use of wire EDM, while sinking EDMs were too slow and costly.

Abrasive-waterjets were applied cost-effectively to machine large-aspect-ratio microchannels for fuel cells on stainless steel and titanium shims (Liu et al., 2008a). Certain AWJ-cut slot/rib patterns on some materials were fabricated that were otherwise too difficult and/or costly to be machined by conventional tools. By mounting the copper tube on a rotary indexing tool, a 254-m nozzle successfully machined 16 narrow slots on the copper tube, as illustrated in Fig. 15.21 Preliminary results demonstrated that µAWJ technology could be an enabling tool for this method of surface armature manufacturing.

<sup>19</sup> Since the kerf width of a slot rather than the separation between slots is limited by the stream diameter of the AWJ, the minimum scale of micromachining is therefore defined by the kerf width. 20 http://pergatory.mit.edu/(8 August 2011)

<sup>21</sup> The 254-m nozzle is an R&D nozzle currently being beta tested.

Figure 14 illustrates three narrow slots machined in a stainless steel shim 0.25 mm thick. The kerf width of the slots was measured to be approximately 92 µm. These slots were machined with the 380-m nozzle together with another ancillary device to reduce the effective stream diameter of the nozzle. Note that the kerf width is smaller than the width of the fingerprints

> a. Three thin slots b. Micrograph of slot

At the Precision Engineering Research Group (PERG) of MIT, a novel concept was conceived to improve the efficiency of small motors/generators by means of surfacemounted armatures.20 Conventional armatures composed of multiple turns of round wires have a low compaction factor (ratio of volume filled by conducting wires to total allotted conductor volume) because there are gaps among wires resulting in relatively high armature resistance and reduced performance. The concept considered replacing wirewound armatures with slotted copper tubes of appropriate annular dimensions (Trimble, 2011 - patent pending). The smaller the kerf width of the cuts, the higher the compaction factor and better the performance of the armature become. Researchers at PERG ran into problems in machining narrow slots on copper tubes. Due to the high reflectivity of copper, laser cutting splattered. On the other hand, the tube geometry prohibited the use of wire

Abrasive-waterjets were applied cost-effectively to machine large-aspect-ratio microchannels for fuel cells on stainless steel and titanium shims (Liu et al., 2008a). Certain AWJ-cut slot/rib patterns on some materials were fabricated that were otherwise too difficult and/or costly to be machined by conventional tools. By mounting the copper tube on a rotary indexing tool, a 254-m nozzle successfully machined 16 narrow slots on the copper tube, as illustrated in Fig. 15.21 Preliminary results demonstrated that µAWJ technology could be an enabling tool for this method of surface armature manufacturing.

19 Since the kerf width of a slot rather than the separation between slots is limited by the stream diameter of the AWJ, the minimum scale of micromachining is therefore defined by the kerf width. 20 http://pergatory.mit.edu/(8 August 2011)

Fig. 14. AWJ-machined slots using a novel ancillary device (Liu & Schubert, 2010**)** 

impressed on the surface of the shim.19

**4.2 Components for green energy products** 

EDM, while sinking EDMs were too slow and costly.

21 The 254-m nozzle is an R&D nozzle currently being beta tested.

Fig. 15. AWJ-slotted copper tubes (design provided by MIT PERG) (Liu et al., 2011b)

#### **4.3 Biomedical components**

Advanced micromachining technology has been favorable for fabricating biomedical devices, which are continually becoming smaller and more intricate in terms of size, shape, and material. AWJ technology shows great potential for such applications based on the market size and current trends, the urgent need for cost reductions in healthcare, and the nature of biomedical components. For example, mini- and micro-plates for orthopedic implants to repair/reconstruct bone and skull fractures are one of the strong candidates for µAWJ technology (Haerle et al., 2009). Titanium is often used for these plates because of its biocompatibility. While conventional machine tools have difficulty in machining titanium, AWJs cut titanium 34% faster than stainless steel at considerably lower costs (Fig. 8). Note that material toughness does not play an important role in AWJ cutting. Therefore, materials that would normally be cut in an annealed condition may be cut in a hardened condition with insignificant loss of productivity. For emergency operations, in particular, a part could be machined by an AWJ from design to finish in minutes. Another potential benefit is improvements in the fatigue performance of implants via dry-grit blasting (Section 2.3).

Figure 16 illustrates several samples of AWJ-machined mini- and micro-plates made of titanium and stainless steel. The mesh-type mini-plates are made from titanium shim stock 0.34 mm thick. These plates are commonly used in facial and skull repair and reconstruction (Haerle et al., 2009). Machining was carried out at a pressure of 380 MPa with the use of the 254-m nozzle. The fine-mesh mini-plate (lower right in photo) took about 20 minutes to complete. Optimization of the nozzle performance is expected to reduce the machining time.

Fig. 16. Titanium and stainless steel orthopedic parts. Scale: mm (Liu et al., 2011b)

The same 254-m nozzle was also applied to cut a flexure to be used as a component of a medical device (Begg, 2011 - patent pending). The material was 6061 T3 aluminum with a thickness of 9.5 mm. The key feature is a narrow bridge, with a target width of 0.25 mm, between two connecting members of the flexure. Figure 17 shows a portion of an AWJ-cut flexure. The two lengths, L1 and L2, shown in the photo are the widths of one of the narrow bridges and one of the connecting members of the flexure. Note that the width of the bridge was measured at 0.31 mm.

Fig. 17. Small aluminum flexure (MIT PERG) (Liu et al., 2011b)

#### **4.4 Planetary gear set**

For machining miniature mechanical components, the 254-m nozzle was used to cut a set of planetary gears. The set consists of seven gears [a sun gear (9.68 mm OD), a ring gear (19.05 mm OD), and five small planetary gears (3.55 mm OD)], a gear mounting plate, and a gear carrier. Figure 18 illustrates the components of the gear set. Also included are the tool paths of the seven gears corresponding to the screen display of the PC-based CAD program

Fig. 18. Components and tool paths of planetary gear (Liu et al., 2011b)

The same 254-m nozzle was also applied to cut a flexure to be used as a component of a medical device (Begg, 2011 - patent pending). The material was 6061 T3 aluminum with a thickness of 9.5 mm. The key feature is a narrow bridge, with a target width of 0.25 mm, between two connecting members of the flexure. Figure 17 shows a portion of an AWJ-cut flexure. The two lengths, L1 and L2, shown in the photo are the widths of one of the narrow bridges and one of the connecting members of the flexure. Note that the width of the bridge

For machining miniature mechanical components, the 254-m nozzle was used to cut a set of planetary gears. The set consists of seven gears [a sun gear (9.68 mm OD), a ring gear (19.05 mm OD), and five small planetary gears (3.55 mm OD)], a gear mounting plate, and a gear carrier. Figure 18 illustrates the components of the gear set. Also included are the tool paths of the seven gears corresponding to the screen display of the PC-based CAD program

was measured at 0.31 mm.

**4.4 Planetary gear set** 

Fig. 17. Small aluminum flexure (MIT PERG) (Liu et al., 2011b)

Fig. 18. Components and tool paths of planetary gear (Liu et al., 2011b)

LAYOUT.22 Note that tool paths with a magenta color represent an edge quality of 3 out of 5, with 5 as the best edge quality. The ring and sun gears were nested to save material. The LAYOUT diagram was then transferred to the CAM program MAKE to machine the planetary gear set from 0.62-mm-thick stainless steel plate using the 254-m nozzle**.23** Figure 19 illustrates the assembled planetary gear set driven by a micro spur gear head motor (Solarobotics, Model GM14a) powered by a single AAA battery.

Fig. 19. Gear assembly (Liu et al., 2011b)

#### **4.5 Near 3D parts**

The 380-m nozzle was set up together with the Rotary Axis to machine axisymmetric features.6 The Space Needle models shown in Fig. 4 are examples cut with that setup. Subsequently, a titanium tube with an OD of 6 mm and an ID of 0.6 mm was mounted on the Rotary Axis. Interlocking features were machined on the titanium tube by the nozzle with the Rotary Axis rotating. A steel rod was inserted into the titanium tube serving as a sacrificial

Fig. 20. Titanium interlocking link (Liu et al., 2011b)

<sup>22</sup> http://www.omax.com/waterjets/layout-software (8 August 2011)

<sup>23</sup> http://www.omax.com/waterjets/make-software (8 August 2011)

piece to protect the opposite wall of the tube from being damaged by the spent high-speed abrasives. Figure 20 illustrates a photograph of the interlocking link. Since there are no soldering joints on the tube, the link is quite strong as compared with similar ones that are welded together. Also shown in the figure is a magnified view of one of the individual links. Figure 21 illustrates an A-Jet-cut aluminum blisk (OD = 25.5 mm; thickness = 13.1 mm). With the combination of the 380-m nozzle and the A-Jet capable of tilting the AWJ up to ±60 degrees from the vertical, many complex 3D features can be readily machined.

Fig. 21. A-Jet-machined aluminum blisk. Scale: mm.

The Intelli-ETCH (patent pending) is an advanced utility of the OMAX Intelli-MAX® software that allows a user to recreate images in various materials. These images are created from standard bitmap files (JPG, TIFF, BMP, etc.).9 By taking the brightness levels of an image and converting those levels into machine speeds, 3D features of the images can be etched onto substrates. Figure 22 illustrates an example of an AWJ-etched lizard on an aluminum substrate. The Intelli-ETCH would have great market potential for controlling AWJ nozzles for use as versatile jewelry/craft making tools.

Fig. 22. AWJ-etched lizard on aluminum substrate (Webers et al., 2010)

#### **4.6 Non-metal samples**

228 Micromachining Techniques for Fabrication of Micro and Nano Structures

piece to protect the opposite wall of the tube from being damaged by the spent high-speed abrasives. Figure 20 illustrates a photograph of the interlocking link. Since there are no soldering joints on the tube, the link is quite strong as compared with similar ones that are welded together. Also shown in the figure is a magnified view of one of the individual links. Figure 21 illustrates an A-Jet-cut aluminum blisk (OD = 25.5 mm; thickness = 13.1 mm). With the combination of the 380-m nozzle and the A-Jet capable of tilting the AWJ up to

The Intelli-ETCH (patent pending) is an advanced utility of the OMAX Intelli-MAX® software that allows a user to recreate images in various materials. These images are created from standard bitmap files (JPG, TIFF, BMP, etc.).9 By taking the brightness levels of an image and converting those levels into machine speeds, 3D features of the images can be etched onto substrates. Figure 22 illustrates an example of an AWJ-etched lizard on an aluminum substrate. The Intelli-ETCH would have great market potential for controlling

±60 degrees from the vertical, many complex 3D features can be readily machined.

Fig. 21. A-Jet-machined aluminum blisk. Scale: mm.

AWJ nozzles for use as versatile jewelry/craft making tools.

Fig. 22. AWJ-etched lizard on aluminum substrate (Webers et al., 2010)

Several samples made of non-metal materials were machined with the beta and R&D nozzles to demonstrate the material independence of waterjet technology. The materials included various composites and ceramics with machinability indexes ranging from about 700 to 4 (refer to Fig. 8). Figure 23 shows miniature samples machined from various composites using the 254-m nozzle (Liu et al., 2010a). The material used for each sample is given in the figure subtitle, along with a number in parentheses that is the thickness of the part in millimeters. Details of small features on the order of 100 µm in size remain sharp and crisp. There is no delamination or chipping on the edges. The thickness of the wheel of the smallest bike is about 200 µm. The carbon fiber (dark) and the epoxy (translucent) layers on the wheels are clearly identifiable in Fig. 23e.

Fig. 23. AWJ-machined miniature composite parts. Numbers in parentheses are thickness of part in mm. Scale: 1 mm/div. (Liu et al., 2010a)

Figure 24 illustrates features machined with the 254-m nozzle in an alumina plate 0.64 mm thick (M ≈ 4). The sharp and crisp edges of all features are evident.

Fig. 24. Features machined with the 254-m nozzle in alumina thin plate (Liu, 2009)

#### **4.7 Multi-nozzle platform**

The downsizing of an AWJ nozzle results in a reduction in the flow rate of the waterjet. Depending on the size of the orifice, the number of nozzles that can be supported by a pump increases accordingly. From Fig. 2, a 22.4-kW pump that is capable of supporting one 360-m orifice operating at 380 MPa with a water flow rate of 3.4 l/min is capable of supporting four 254-m nozzles operating at the same pressure. A multi-nozzle platform on which four 254-m nozzles could be mounted was designed, assembled, and tested, as illustrated in Fig. 25. The platform was subsequently delivered for beta testing at a specialty jewelry manufacturing shop. With the nozzles operating in tandem, four identical parts can be machined simultaneously to boost productivity. Among the advantages of using the 254 m nozzle together with 320-mesh garnet are that the amplitude of the striation is small and the finished parts are nearly free of burrs.

Fig. 25. Four nozzles mounted on a platform for increased productivity (Liu et al., 2011b)

#### **5. Conclusion**

Waterjet technology has inherent technological and manufacturing merits that make it suitable for machining most materials from macro to micro scales. It has been established as one of the most versatile precision machining tools and has proven amenable to micromachining. This technology has emerged as the fastest growing segment of the overall machine tool industry in the last decade.3

The smallest features that can be machined with state-of-the-art commercial AWJ systems are limited to greater than 200 µm. Further downsizing of AWJ nozzles for machining features less than 200 µm has met with considerable challenges, as described in Section 3.1. These challenges, which are due to the complexity of the jet flow as the AWJ flow characteristics change into microfluidics, include nozzle clogging by accumulation of wet abrasives, difficulty in the fabrication of mixing tubes with exit orifices less than 200 µm, the degradation in the flowability of fine abrasives, and other relevant issues.

The downsizing of an AWJ nozzle results in a reduction in the flow rate of the waterjet. Depending on the size of the orifice, the number of nozzles that can be supported by a pump increases accordingly. From Fig. 2, a 22.4-kW pump that is capable of supporting one 360-m orifice operating at 380 MPa with a water flow rate of 3.4 l/min is capable of supporting four 254-m nozzles operating at the same pressure. A multi-nozzle platform on which four 254-m nozzles could be mounted was designed, assembled, and tested, as illustrated in Fig. 25. The platform was subsequently delivered for beta testing at a specialty jewelry manufacturing shop. With the nozzles operating in tandem, four identical parts can be machined simultaneously to boost productivity. Among the advantages of using the 254 m nozzle together with 320-mesh garnet are that the amplitude of the striation is small and

Fig. 25. Four nozzles mounted on a platform for increased productivity (Liu et al., 2011b)

Waterjet technology has inherent technological and manufacturing merits that make it suitable for machining most materials from macro to micro scales. It has been established as one of the most versatile precision machining tools and has proven amenable to micromachining. This technology has emerged as the fastest growing segment of the overall

The smallest features that can be machined with state-of-the-art commercial AWJ systems are limited to greater than 200 µm. Further downsizing of AWJ nozzles for machining features less than 200 µm has met with considerable challenges, as described in Section 3.1. These challenges, which are due to the complexity of the jet flow as the AWJ flow characteristics change into microfluidics, include nozzle clogging by accumulation of wet abrasives, difficulty in the fabrication of mixing tubes with exit orifices less than 200 µm, the

degradation in the flowability of fine abrasives, and other relevant issues.

**4.7 Multi-nozzle platform** 

**5. Conclusion** 

machine tool industry in the last decade.3

the finished parts are nearly free of burrs.

Novel manufacturing and operational processes and ancillary devices have been investigated and developed to meet the above challenges. Miniature beta and R&D nozzles, without the need for vacuum assist and water flushing, have been assembled and tested to machine miniature samples made of various materials for a broad range of applications. Many of the samples with basic features as small as 100 µm were machined to demonstrate the versatility of waterjet technology for low-cost micromanufacturing of components for medical implants/devices and microelectronics, for green energy production systems, and for the post-processing of various micro-nano products.

The advancement and refinement of µAWJ technology continue. Efforts are being made to further downsize µAWJ nozzles for machining features around 100 and 50 µm. The goal is to commercialize a µAWJ system by integrating µAWJ nozzles with a low-cost, low-power, high-pressure pump and a precision small-footprint X-Y traverse. A host of accessories are already available to be downsized for facilitating 3D meso-micro machining.

#### **6. Acknowledgment**

This work was supported by an OMAX R&D fund and NSF SBIR Phase I and II Grants #0944229 and #1058278. A part of the work was supported by U. S. Pacific Northwest National Laboratory (PNNL) under Technology Assistance Program (TAP) Agreements: 07- 29, 08-02, 09-02, and 10-02. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the NSF and PNNL. Contributions from research institutes and industrial partners by furnishing sample materials and part drawings and by evaluating AWJ-machined parts are acknowledged. Collaborators include but are not limited to Microproducts Breakthrough Institute (MBI), MIT Precision Engineering Research Group, Ryerson University, and several OMAX's customers and suppliers. The authors would like to thank their colleagues at OMAX for reviewing the article and providing us with constructive feedback.

#### **7. References**


Henning, A.; Miles, P., Stang, D., (2011a). Efficient Operation of Abrasive Waterjet Cutting

Henning, A.; Liu, H.-T., & Olsen, C. (2011b). Economic and Technical Efficiency of High

Jiang, S.; Popescu, R., Mihai, C., & Tan, K., (2005). High Precision and High Power ASJ

Liu, H.-T. (2010). Waterjet Technology for Machining Fine Features Pertaining to

Liu, H.-T. & McNiel, D. (2010). Versatility of Waterjet Technology: from Macro and Micro

Liu, H.-T. & Schubert, E. (2010). Piercing and/or Cutting Devices for Abrasive Waterjet

Liu, H.-T.; Hovanski, Y., & Dahl, M. E. (2011a). Machining of Aircraft Titanium with

Liu, H.-T.; Schubert E., and McNiel, D. (2011b). AWJ Technology for Meso-Micro

Liu, H.-T.; Miles, P., & Veenhuizen, S. D. (1998). CFD and Physical Modeling of UHP AWJ

Liu, H.-T.; Hovanski, Y., Dahl, M. E. & Zeng, J. (2009a). Applications of Abrasive-Waterjets

Liu, H.-T.; Gnäupel-Herold, T., Hovanski, Y., & Dahl, M. E. (2009b). Fatigue Performance

*Proceedings 2009 American WJTA Conference*, Houston, Texas, August 18-20. Liu, H.-T.; Hovanski, Y., Caldwell, D. D., & Williford, R. E. (2008a). Low-Cost

*Conference and exposition.,* Houston, Texas, August 21–23, Paper 1A-3. Lamb, H. (1993). *Hydrodynamics* (6th Ed.). Cambridge University Press, ISBN 9780521458689. Liu, H.-T. (1998). Near-Net Shaping of Optical Surfaces with Abrasive Suspension Jets,

Texas, September 19-21.

September 21–23, pp. 285-294.

(doi:10.1016/ j.jmapro.2010.01.002).

*Jetting*, October 20–22, Graz, Austria.

Belgium, September 21–23, pp. 15-24.

Nottingham, UK: October, 15-17.

*PVP2009 Conference*, Prague, Czech, July 26-30.

*Vessel & Piping*.

#612348007.US.

September 19-21.

*Pressure Vessel & Piping* 

in Industrial Applications, *Proceedings of 2011 WJTA-IMCA Conference &* , Houston,

Performance Abrasive Waterjet Cutting, to appear in *ASME Journal of Pressure* 

Singulations for Semiconductor Manufacturing, *Proceedings of 2005 WJTA* 

*Proceedings of 14th International Conference on Jetting Technology*, Brugge, Belgium,

MicroMachining, *Journal of Manufacturing Processes*, Vol.12, No.1, pp. 8-18.

Machining for Most Materials, *Proceedings of 20th International Conference on Water* 

Systems and Associated Systems and Methods, U. S. Provisional Patent Application

Abrasive-Waterjets for Fatigue Critical Applications, to appear in *ASME Journal of* 

Machining, *Proceedings of 2011 WJTA-IMCA Conference and Exposition.*, Houston,

Drilling, *Proceedings of 14th International Conference on Jetting Technology*, Brugge,

for Machining Fatigue-Critical Aerospace Aluminum Parts, *Proceedings of ASME*

Enhancement of AWJ-Machined Aircraft Aluminum with Dry-Grit Blasting,

Manufacturing of Flow Channels with Multi-Nozzle Abrasive-Waterjets: A Feasibility Investigation, *Proceedings of 19th International Conference on Water Jetting*,


## **Electrochemical Spark Micromachining Process**

Anjali Vishwas Kulkarni *Centre for Mechatronics, Indian Institute of Technology Kanpur, India* 

#### **1. Introduction**

234 Micromachining Techniques for Fabrication of Micro and Nano Structures

Webers, N.; Olsen, C., Miles, P. & Henning, A. (2010) Etching 3D Patterns with Abrasive

Zeng, J. & Kim, T. J. (1995), Machinability of Engineering Materials in Abrasive Water Jet

October 20–22, pp. 51-63.

Waterjets, *Proceedings of 20th International Conference on Water Jetting*, Graz, Austria,

Machining, *International Journal of Water Jetting Technology*, Vol.2, No.2, pp. 103-110.

Electrochemical spark micromachining process (ECSMM) is a process suitable for micromachining of electrically non-conducting materials. Besides the classic semiconductor technology, there are various methods and processes for micromachining such as Reactive Ion Etching (RIE) (Rodriguez et al., 2003), femto-second pulse laser radiation (Hantovsky et al., 2006), chemical etching and plasma-enhanced chemical vapor deposition (Claire, 2004)], spark assisted chemical engraving (Fasico and Wuthrich, 2004) and micro-stereolithography (Rajaraman, 2006) in practice. Use of photoresist as sacrificial layer to realize micro-channels in micro fluidic systems is discussed in (Coraci, 2005). All these methods are expensive as they need the vacuum, clean environment and mostly involve in between multi processing steps to arrive at the final microchannel machining results. There is a need of an innovative process which is cost effective and straight forward without employing intermediate processing steps. One such process thought of and being researched is electrochemical spark micromachining (ECSMM) process. The ECSMM process is a stand alone process unlike others and it does not demand on intermediate processing steps such as: masking, pattern transfer, passivation, sample preparation etc. The use of separate coolants is also not required in performing the micromachining by ECSMM.

Micromachining needs are forcing reconsideration of electrochemical techniques as a viable solution (Marc Madau, 1997). Another similar process termed as spark assisted chemical engraving (SACE) (Wuthrich et al., 1999) has been employed for the micromachining of glass. ECSMM is a strong candidate for microfabrication utilizing the best of electrochemical machining (ECM) and electro discharge machining (EDM) together. Applications of ECS for microfabrication can be in the field of aeronautics, mechanical, electrical engineering and similar others. It can successfully process silicon (Kulkarni et. al., 2010a), molybdenum (Kulkarni et. al., 2011c), tantalum (Kulkarni et. al., 2011a), quartz (Deepshikha, 2007; Kulkarni et. al., 2011a), glass ((Kulkarni et al., 2011a, 2011b); Wuthrich et al. 1999)), alumina (Jain et al., 1999), advanced ceramics (Sorkhel et al., 1996) and many other materials.

The chapter discusses the details of the experimental set-up developed in the next section. The procedure for micromachining using the developed set-up is outlined next. The experimental scheme to perform machining on glass pellets (cover slips used in biological applications) is presented. Discussion of the micro machined samples is presented. This discussion is based on various on line and post process measurements performed. The qualitative material removal mechanism is presented based on the results and discussions.

#### **2. Experimental set-up**

A functional set-up of the ECSMM process is designed, developed and fabricated as shown in Figure 1 (Kulkarni et. al., 2011b). The main components of the ECS set-up are as follows and are described in the following sub sections:


#### **2.1 Machining chamber**

The machining chamber houses X-Y table, Z axis assembly, tool feed and tool holder assembly and ECS cell. X, Y, Z and tool feed stages are motorized.

Fig. 1. Photograph of experimental set-up (Kulkarni et al., 2011b)

#### **2.1.1 X-Y table**

X-Y table has resolution of 2 μm in X and Y directions and traverse of 100 mm in X as well as Y directions. The guide ways use non-recirculating balls as rolling elements. The mechanical drive is a ground lead screw of 400 μm pitch made of aluminium alloy. Rotation to the X and Y screws is provided by separate stepper motors. The table is mounted on a chrome plated MS plate. Chrome plating protects the plate from corrosion. The MS plate has mounting tapped holes on a 25 mm grid to mount the ECS cell. Bellows are provided to protect the motors and lead screws from the electrolyte splashes and fumes produced.

#### **2.1.2 Z axis assembly**

The Z axis is automated to move up or down to maintain a constant work piece-tool gap. The worm and worm wheel with a gear ratio of 1:38 transmit the power to a lead screw of 200 μm pitch. All the parts are fabricated with stainless steel and brass to resist corrosion due to acidic environment. It has positioning accuracy of 50 μm and maximum vertical travel of 80 mm.

#### **2.1.3 Tool feed and tool holder assembly**

Tool feed assembly is mounted on Z axis assembly. A glass tool holder is designed and developed. This tool holder provided the tool insulation and hence reduction in the stray currents. This glass tool holder is used to hold the tool wire in place. A fixture made of Perspex material is designed and fabricated to hold the tool holder on Z assembly. Cu wire of 200 μm diameter is used as a cathode (tool).

#### **2.1.4 ECS cell**

236 Micromachining Techniques for Fabrication of Micro and Nano Structures

A functional set-up of the ECSMM process is designed, developed and fabricated as shown in Figure 1 (Kulkarni et. al., 2011b). The main components of the ECS set-up are as follows

The machining chamber houses X-Y table, Z axis assembly, tool feed and tool holder

Control PC

Exhaust System

Machining Chamber

Power Supplies

X-Y table has resolution of 2 μm in X and Y directions and traverse of 100 mm in X as well as Y directions. The guide ways use non-recirculating balls as rolling elements. The mechanical drive is a ground lead screw of 400 μm pitch made of aluminium alloy. Rotation to the X and Y screws is provided by separate stepper motors. The table is mounted on a chrome plated MS plate. Chrome plating protects the plate from corrosion. The MS plate has mounting tapped holes on a 25 mm grid to mount the ECS cell. Bellows are provided to protect the motors and lead screws from the electrolyte splashes and fumes produced.

The Z axis is automated to move up or down to maintain a constant work piece-tool gap. The worm and worm wheel with a gear ratio of 1:38 transmit the power to a lead screw of 200 μm pitch. All the parts are fabricated with stainless steel and brass to resist corrosion due to acidic environment. It has positioning accuracy of 50 μm and maximum vertical

assembly and ECS cell. X, Y, Z and tool feed stages are motorized.

Fig. 1. Photograph of experimental set-up (Kulkarni et al., 2011b)

**2. Experimental set-up** 

1. Machining Chamber 2. Power Supply System 3. Exhaust System 4. Control PC

**2.1 Machining chamber** 

**2.1.1 X-Y table** 

**2.1.2 Z axis assembly** 

travel of 80 mm.

and are described in the following sub sections:

It is a rectangular box of 10 cm x 8 cm x 6 cm dimensions made up of Perspex material. It is mounted on X-Y table. It houses separate fixture arrangement for graphite anode and work piece holder. It is filled with the electrolyte. The electrolyte level is maintained at 1mm above the flat surface of the work piece. Electrolyte used is NaOH in varied concentration in the range of 14-20 %.

#### **2.2 Power supply system**

DC regulated power supplies of different ratings are used for driving stepper motors, machining supply and control circuitry. Use of separate power supply ensures the noise free operation.

#### **2.3 Exhaust system**

Proper exhaust system is designed and provided to take away the electrolyte fumes generated during the spark process inside the machining chamber. A small DC operated fan is placed in the machining chamber where the fumes are generated. These are carried away by a hose pipe and thrown away from the room with an exhaust fan.

#### **2.4 Control PC**

Stepper motors used for driving X, Y, Z and tool feed are all interfaced to motion controller card installed in PC. Precise control and drive of the machine is achieved with NI 7834 PCI card and NI 7604 drive board interfaced to a computer. Contouring functions in LabVIEW platform are used to carve different shapes of the micro channels [Kulkarni et al., 2008].

#### **3. Experimental procedures**

The supply voltage, electrolyte concentration and table speed are the control parameters. Pilot experiments are performed to determine the optimum window of these operating parameters.

It is observed that sparking occurs at supply voltage of 30 V and above. Glass samples break above 50 V supply voltage. Hence the working supply voltage range chosen is 40 V – 50 V.

Use of base solution is preferred over the acidic electrolyte. It was observed that in the acidic environment the surface roughness increases. The fumes formed of acidic solutions during the electrochemical sparking process are harmful. During the pilot experiments it was observed that machining takes place in diluted sodium hydroxide (NaOH) solution as electrolyte. The concentration window was decided upon by performing many experiments to arrive at a permissible concentration range. It was observed that machining does not take place below 14% concentration of NaOH. Above 20 % concentration of NaOH, the machined surface roughness is notable. Hence 14% -20% concentration range for NaOH electrolyte is arrived at. Moreover use of low concentration of NaOH as electrolyte makes the ECSMM process as a 'green process'. Level of electrolyte is maintained at 1 mm above the work piece surface in the ECS cell.

The table speed is chosen ranging between 12.5 μm/s – 25 μm/s. It is such that the traverse is not too slow to dig the micro channel and not too fast to miss the micro machining in that region.

Micro channels are formed using the ECSMM process on microscopic glass pellets using platinum wire as a tool of 500 μm diameter. Pellets are of 180 μm thickness, 18 mm diameter circles in size. Length of the tool protruding out of the tool holder is 4 mm. The gap between the cathode tool electrode tip and the work piece surface is maintained at around 20 μm using the tool feed device mounted on Z-axis. The distance between the tool and the anode is 40 mm. Figure 2 shows the photograph of the electrolytic cell with the spark visible at tool tip and electrolyte interface. Graphite anode is seen in the cell. It is a non consuming electrode.

Fig. 2. Photograph of the ECSMM cell with graphite anode, tool and work piece. The spark is visible near the tool tip (Kulkarni et al., 2011b).

Experiments are conducted with Voltage, Electrolyte Concentration and Table Speed as the control variables. The experiments are conducted in accordance with the central composite design scheme developed by the software 'Design Expert 07' to study the response surface. The range of the control variables chosen is as shown below:


The design resulted in total of twenty one experiments, out of these twenty one experiments, six central experiments were performed at 45 V supply voltage, 17 % electrolyte concentration and 18.75 µm/s table speed as the values for the control variables.

The responses measured are: average process current (I), width of microchannel (W) and depth of microchannel (D) formed using ECSMM. The scheme of the experiments is as shown in Table 1. Columns 2-4 list Vs, EC, and TS respectively. Columns 5-7 give average current, width, and depth of the microchannels respectively as the responses measured post process.

arrived at. Moreover use of low concentration of NaOH as electrolyte makes the ECSMM process as a 'green process'. Level of electrolyte is maintained at 1 mm above the work piece

The table speed is chosen ranging between 12.5 μm/s – 25 μm/s. It is such that the traverse is not too slow to dig the micro channel and not too fast to miss the micro machining in that

Micro channels are formed using the ECSMM process on microscopic glass pellets using platinum wire as a tool of 500 μm diameter. Pellets are of 180 μm thickness, 18 mm diameter circles in size. Length of the tool protruding out of the tool holder is 4 mm. The gap between the cathode tool electrode tip and the work piece surface is maintained at around 20 μm using the tool feed device mounted on Z-axis. The distance between the tool and the anode is 40 mm. Figure 2 shows the photograph of the electrolytic cell with the spark visible at tool tip and electrolyte interface. Graphite anode is seen in the cell. It is a non consuming

Fig. 2. Photograph of the ECSMM cell with graphite anode, tool and work piece. The spark

Experiments are conducted with Voltage, Electrolyte Concentration and Table Speed as the control variables. The experiments are conducted in accordance with the central composite design scheme developed by the software 'Design Expert 07' to study the response surface.

The design resulted in total of twenty one experiments, out of these twenty one experiments, six central experiments were performed at 45 V supply voltage, 17 % electrolyte

The responses measured are: average process current (I), width of microchannel (W) and depth of microchannel (D) formed using ECSMM. The scheme of the experiments is as shown in Table 1. Columns 2-4 list Vs, EC, and TS respectively. Columns 5-7 give average current, width, and depth of the microchannels respectively as the responses measured post process.


concentration and 18.75 µm/s table speed as the values for the control variables.

is visible near the tool tip (Kulkarni et al., 2011b).

The range of the control variables chosen is as shown below: - Factor 1 (Vs): Supply voltage ranging between 40 V - 50 V

surface in the ECS cell.

region.

electrode.


(\* could not be measured)

Table 1. Experimental parameters and responses

#### **3.1 On line measurements**

The average process current is measured with the help of a digital multimeter. Besides this average current, the time varying process current is measured on line by digital storage oscilloscope. For this purpose the 'resistive shunt method' is used. In this a 1 Ω resistance is connected in series with cathode and ground of the power supply to the ECS cell. The time varying voltage across this resistance is the direct measure of the time varying process current. The wave forms are saved on the control PC via RS 232 connectivity module of the oscilloscope (Hameg 1008). The on-time, off-time and the frequency of the sparks occurring are measured. These parameters are otherwise theoretically estimated. The occurrences of these pulses directly indicate the correlation between the presences of the sparks during the process. The analysis of these current pulses will be helpful to devise the electrical model of the process.

#### **3.2 Post process measurements**

A set-up is developed to measure the depth of the microchannels at various points with the resolution of 10 µm. The dial gauge used for this purpose is mounted on the Z Axis of a standard machine to achieve these measurements.

To study the surface topography and width measurement, SEM analysis of the micro channels is performed. SEM at different and higher magnification is performed to get the insight into the surface topography due to this process. SEM at increasing magnification clearly shows the imprints and development of how the material is removed from the work piece surface.

The results based on the above studies are presented in the following section.

#### **4. Results and discussions**

Measurements of on line average current, and post measurements of width and depth of microchannels are presented in Table 1 in column 5-7. Theses are discussed in details in the following sub sections. To study the surface topography and width measurement, SEM analysis of the microchannels is performed. Following section describes the microstructure analysis of the microchannels.

#### **4.1 Microstructure Analysis by SEM**

Detailed SEM is performed for the samples of central experiments (at 45 V, 17% electrolyte concentration, and 18.75 µm/s table speed) to study the effect of sparking on the microstructure. SEM is performed at successive higher magnification to visualize the surface closely. Figure 3 shows the photograph of three microchannels (forming an inverted 'C' section) carved at 45 V, 17% electrolyte concentration and 18.75 µm/s table speed. These are formed using X traverse through 2500 pulses in carving channel 1, Y traverse in carving channel 2 and then negative X traverse of X-Y table in carving the third channel, i.e. channel 3. The length of each section in the C type micro channel is about 5000 µm. The average width of channel 2 is around 535 µm and the average depth is around 370 µm.

Fig. 3. USB Photograph of the channels carved at 45 V, 17% electrolyte concentration and 18.75 µm/s table speed. The approximate length of each channel is 5000 µm, average width is 535 µm, and depth is 370 µm.

Figure 4 gives the microstructure of the micromachined glass surface at 162 X magnification carved at 37 V, 17% electrolyte concentration and 18.75 µm/s table speed. From SEM picture it is obvious that a shallow microchannel is obtained. This may be due to the lower supply voltage of 37 V. The width at two different regions of microchannel is 267 µm and 410.3 µm. Thus, average width of microchannel is 338.65 µm.

Figure 5 gives the microstructure of the micromachined coverslip surface at 881X magnification.The microstructure clearly shows the removal of material along the path of tool movement. Valleys and ridges are clearly visible which are due to melting and the layer by layer material removal in the spark affected region. A piece of material is seen which got re solidified and remained there.

Measurements of on line average current, and post measurements of width and depth of microchannels are presented in Table 1 in column 5-7. Theses are discussed in details in the following sub sections. To study the surface topography and width measurement, SEM analysis of the microchannels is performed. Following section describes the microstructure

Detailed SEM is performed for the samples of central experiments (at 45 V, 17% electrolyte concentration, and 18.75 µm/s table speed) to study the effect of sparking on the microstructure. SEM is performed at successive higher magnification to visualize the surface closely. Figure 3 shows the photograph of three microchannels (forming an inverted 'C' section) carved at 45 V, 17% electrolyte concentration and 18.75 µm/s table speed. These are formed using X traverse through 2500 pulses in carving channel 1, Y traverse in carving channel 2 and then negative X traverse of X-Y table in carving the third channel, i.e. channel 3. The length of each section in the C type micro channel is about 5000 µm. The average

Channel 1

Channel 2

Fig. 3. USB Photograph of the channels carved at 45 V, 17% electrolyte concentration and 18.75 µm/s table speed. The approximate length of each channel is 5000 µm, average width

Channel 3

Figure 4 gives the microstructure of the micromachined glass surface at 162 X magnification carved at 37 V, 17% electrolyte concentration and 18.75 µm/s table speed. From SEM picture it is obvious that a shallow microchannel is obtained. This may be due to the lower supply voltage of 37 V. The width at two different regions of microchannel is 267 µm and 410.3 µm.

Figure 5 gives the microstructure of the micromachined coverslip surface at 881X magnification.The microstructure clearly shows the removal of material along the path of tool movement. Valleys and ridges are clearly visible which are due to melting and the layer by layer material removal in the spark affected region. A piece of material is seen which got

width of channel 2 is around 535 µm and the average depth is around 370 µm.

**4. Results and discussions** 

analysis of the microchannels.

is 535 µm, and depth is 370 µm.

re solidified and remained there.

Thus, average width of microchannel is 338.65 µm.

**4.1 Microstructure Analysis by SEM** 

Fig. 4. SEM image (162X) showing width at two places along the microchannel, the average width of the channel is around 338.65 µm at 37 V, 17% electrolyte concentration and 18.75 µm/s table speed.

Fig. 5. SEM image of pellet ( 881X) treated at 45 V, 12% electrolyte concentration and 18.75 µm/s table speed.

Figure 6 gives the microstructure of the microchannel at 4500X magnification. The tearing off of the material is seen. The region shows the melting and solidification of the workpiece material. The thickness of the smallest layer at the corner is around 7.8 µm.

Fig. 6. SEM image (4500X) giving thickness of layer at the corner around 7.8 µm at 45 V, 17% electrolyte concentration and 18.75 µm/s table speed.

#### **4.2 Current analysis**

#### **4.2.1 On line average process current**

Column 5 of Table 1 gives the values of the average process current measured on line. The electro chemical action causing the migration of ions, and electrons contributes to the average current. The average current value as seen in Table 1, column 5 is ranging from 0.0125 – 0.9 A. Occasionally it has shoot up to 2.493 A for R#17 for 53.4 V supply voltage.

The interesting process phenomenon is not obvious from only recording the average current. The processes' time varying nature can be only revealed by studying the transient current. The transient current waveforms reveal the process complexity and help in understanding the holistic and time changing phenomena during the single entire current cycle of ECSMM process, as explained in the next section. The actual machining is occurring during the very short time of the instantaneous current pulses carrying high energy density.

#### **4.2.2 On line, transient process current**

The time varying current is measured with the help of a digital storage oscilloscope as mentioned in section 3.1. The snap shots of the stored waveform are presented in Figures 7 a and b. In Figure 7a it can be noted that there are many spikes during a time of 10 ms duration corresponding to 1 division of oscilloscope window. Each pulse represents a spark occurrence. The average process current can be seen at a level of 0.1 A. A pulse of height of 0.3 A can be seen of time period greater than 20 ms. A second pulse of instantaneous current value more than 0.4 A can be seen after a period of around 15 ms. It's time period is about 6ms. Many short duration (<1ms) pulses can be seen in between these two remarkable pulses. These pulses show the stochastic nature of the spark formation process.

7.8 µm

Fig. 6. SEM image (4500X) giving thickness of layer at the corner around 7.8 µm at 45 V,

Column 5 of Table 1 gives the values of the average process current measured on line. The electro chemical action causing the migration of ions, and electrons contributes to the average current. The average current value as seen in Table 1, column 5 is ranging from 0.0125 – 0.9 A. Occasionally it has shoot up to 2.493 A for R#17 for 53.4 V supply voltage. The interesting process phenomenon is not obvious from only recording the average current. The processes' time varying nature can be only revealed by studying the transient current. The transient current waveforms reveal the process complexity and help in understanding the holistic and time changing phenomena during the single entire current cycle of ECSMM process, as explained in the next section. The actual machining is occurring during the very short time of the instantaneous current pulses carrying high

The time varying current is measured with the help of a digital storage oscilloscope as mentioned in section 3.1. The snap shots of the stored waveform are presented in Figures 7 a and b. In Figure 7a it can be noted that there are many spikes during a time of 10 ms duration corresponding to 1 division of oscilloscope window. Each pulse represents a spark occurrence. The average process current can be seen at a level of 0.1 A. A pulse of height of 0.3 A can be seen of time period greater than 20 ms. A second pulse of instantaneous current value more than 0.4 A can be seen after a period of around 15 ms. It's time period is about 6ms. Many short duration (<1ms) pulses can be seen in between these two remarkable

pulses. These pulses show the stochastic nature of the spark formation process.

17% electrolyte concentration and 18.75 µm/s table speed.

**4.2 Current analysis** 

energy density.

**4.2.1 On line average process current** 

**4.2.2 On line, transient process current** 

Fig. 7 a. Snap shot of online time varying ECSMM process current during glass pellet micromachining.

In another waveform, Figure 7 b, two complete current pulses and two halfway current pulses can be seen. These are of different time durations, ranging from 0.1 ms to 0.3 ms time period. Hence the resulting frequency is variable and is ranging from 2.5 kHz to 5 kHz. The sparking frequency depends on many factors such as size of bubble formed, bubble growth time, time of its survival, etc. The size of the spark or discharge depends on the instantaneous current value. It is clear from these time varying current pulses that sparks of different energy strike the work piece surface resulting in softening, melting, and / or vaporizing of the work piece material.

Fig. 7 b. Snap shot of online time varying ECSMM process current during glass pellet micromachining.

The spark energy can be estimated by taking Vs = 45 V, iinstantaneous = 0.2 A and time = 0.2 ms. The instantaneous spark energy with a striking area of diameter less than the tool diameter, i.e. 200 µm, is of the order of 500 kJ/m2.

#### **4.3 Width of microchannels**

Columns 6 of Table 1 gives post process measured values of the width of the microchannels using the dial gauge. The last column summarizes the type of the microchannel formed. It says whether a channel is a through channel or a blind microchannel achieved. The rows corresponding to the successfully achieved microchannels are shown in bold face. The depth of the microchannels in Run # 8 and # 11 could not be measured for some reasons. In case of the through channels the depth of the microchannel achieved is more than 180 µm. Either the higher machining time or the lower travel speed or the smaller gap due to local irregularities is responsible for through machining to occur.

The minimum width achieved is 421.5µm for Run # 7 for 36.6 V, 17 % electrolyte concentration and 18.75 µm/s table speed. The maximum width achieved is 1110 µm for Run # 9 for 50 V, 20 % electrolyte concentration and 12.5 µm/s table speed. That means for higher voltage, higher electrolyte concentration and lower table speed combination of parameters, the width achieved is higher.

#### **4.4 Depth of microchannels**

The minimum depth achieved is 77.25 µm for Run # 18 for 45 V, 11.9 % electrolyte concentration and 18.75 µm/sec table speed. The maximum depth achieved is 123.6 µm for Run # 20 for 45 V, 22 % electrolyte concentration and 18.75 µm/sec table speed. Higher electrolyte concentration results in higher depth. Microchannels of the width between 400 – 1100 µm are achieved. The depth achieved is 75 -120 µm.

For other experiments, through machining has been occurred where the depth of cut is more than the thickness of the work piece. This may be partly due to the gap adjustment between the tool and the work piece surface. It is a crucial operation to maintain the gap at or above 20 µm without the closed loop control. This calls for a close loop control for maintaining the gap between the tool and the work piece surface.

A novel technique to measure the depth of these microchannels is devised and discussed in Kulkarni et. al., (2010b) and Kulkarni et. al., (2010c).

Parametric models pertaining to the average current, width and depth of the microchannels' are presented elsewhere.

Section 5 describes the systematic description on understanding the ECSMM process mechanism in view of the transient current.

#### **5. Understanding the process mechanism**

The material removal mechanism in ECSMM is complex as it is revealed by the SEM and current pulses analysis in the previous sections. This is primarily due to the non-thermal nature of these sparks. In the existent literature the spark energy is considered to be of thermal nature and thermal analysis and material removal are considered to be due to this thermal source (Jain et al.,1999; Basak & Ghosh, 1992). Experimentally it has been found that the spark is a non thermal discharge. This has been confirmed (Kulkarni et al, 2009) while making an attempt to measure the spark temperature by a pyrometer. Pyrometer failed to

The spark energy can be estimated by taking Vs = 45 V, iinstantaneous = 0.2 A and time = 0.2 ms. The instantaneous spark energy with a striking area of diameter less than the tool diameter,

Columns 6 of Table 1 gives post process measured values of the width of the microchannels using the dial gauge. The last column summarizes the type of the microchannel formed. It says whether a channel is a through channel or a blind microchannel achieved. The rows corresponding to the successfully achieved microchannels are shown in bold face. The depth of the microchannels in Run # 8 and # 11 could not be measured for some reasons. In case of the through channels the depth of the microchannel achieved is more than 180 µm. Either the higher machining time or the lower travel speed or the smaller gap due to local

The minimum width achieved is 421.5µm for Run # 7 for 36.6 V, 17 % electrolyte concentration and 18.75 µm/s table speed. The maximum width achieved is 1110 µm for Run # 9 for 50 V, 20 % electrolyte concentration and 12.5 µm/s table speed. That means for higher voltage, higher electrolyte concentration and lower table speed combination of

The minimum depth achieved is 77.25 µm for Run # 18 for 45 V, 11.9 % electrolyte concentration and 18.75 µm/sec table speed. The maximum depth achieved is 123.6 µm for Run # 20 for 45 V, 22 % electrolyte concentration and 18.75 µm/sec table speed. Higher electrolyte concentration results in higher depth. Microchannels of the width between 400 –

For other experiments, through machining has been occurred where the depth of cut is more than the thickness of the work piece. This may be partly due to the gap adjustment between the tool and the work piece surface. It is a crucial operation to maintain the gap at or above 20 µm without the closed loop control. This calls for a close loop control for maintaining the

A novel technique to measure the depth of these microchannels is devised and discussed in

Parametric models pertaining to the average current, width and depth of the microchannels'

Section 5 describes the systematic description on understanding the ECSMM process

The material removal mechanism in ECSMM is complex as it is revealed by the SEM and current pulses analysis in the previous sections. This is primarily due to the non-thermal nature of these sparks. In the existent literature the spark energy is considered to be of thermal nature and thermal analysis and material removal are considered to be due to this thermal source (Jain et al.,1999; Basak & Ghosh, 1992). Experimentally it has been found that the spark is a non thermal discharge. This has been confirmed (Kulkarni et al, 2009) while making an attempt to measure the spark temperature by a pyrometer. Pyrometer failed to

i.e. 200 µm, is of the order of 500 kJ/m2.

parameters, the width achieved is higher.

**4.4 Depth of microchannels**

are presented elsewhere.

irregularities is responsible for through machining to occur.

1100 µm are achieved. The depth achieved is 75 -120 µm.

gap between the tool and the work piece surface.

mechanism in view of the transient current.

Kulkarni et. al., (2010b) and Kulkarni et. al., (2010c).

**5. Understanding the process mechanism** 

**4.3 Width of microchannels** 

measure the temperature as the radiation is a non thermal type. Instead it is a discharge process similar to that of the breakdown of the hydrogen gas bubble isolating the tool tip from the surrounding electrolyte.

Secondly, electro chemical systems are known to exhibit complex non-linear behavior. These nonlinearities arise due to electro hydro dynamism, ionic reactions, bubble generation, their growth and their breakdown phenomena. The overall process seems to be discrete in nature though the supply voltage is DC. Positive as well as negative spikes are also observed in the current waveforms. The electrochemical kinetics includes negative faradic impedance in the electrolyte solution. There are many intermittent, small amplitude current spikes, of smaller duration. These seem to be representing the partial sparks due to the break down of the small hydrogen bubbles. The partial discharge is due to the total isolation of single or many such bubbles completely isolating the electrolyte contact. On the other place the total or complete sparking is that occurring due to the complete isolation of the cathode tip from the electrolyte surface. This can be understood by the pictorial representation as in Figure 8 a and b. In Figure 8 a, there is a local isolation of the tool tip from the surrounding electrolyte due to a small hydrogen bubble. This causes an instantaneous sparking across the bubble, resulting in a small amplitude current spike. Where as, in Figure 8 b, the tool tip is surrounded by a single larger bubble. Many small sized bubbles coalesce in a single larger bubble resulting in complete isolation of the tool tip from the electrolyte. The sparking resulting due to this kind of total isolation will result in the intensive sparking manifesting the large amplitude current spikes. This kind of behavior is reflected in the nature of the on line time varying current pulses studied. It was observed that the frequency of sparking

Fig. 8a. Partial sparking due to local isolation by small bubble resulting in a low energy spark

Fig. 8b. Intensive sparking due to complete blanketing of the tool tip by large sized bubble resulting in high energy spark

(oscillations) varies with varying supply voltage (Kulkarni, 2000). The sparking frequency is high (in the tens of MHZ range) and it lowers (in the few hundreds of kHz range) for higher supply voltage. This supports the possibilities of the many partial sparks due to breakdown of the single isolated hydrogen gas bubbles. Theses partial sparks or discharges are of less current value and hence having less energy. These may not result in material removal from the workpiece. These may die out before reaching the workpiece surface.

#### **5.1 Intermediate processes and their interrelation**

Thus ECSMM process comprises of many intermediate processes such as electro chemical action causing the migration of ions, followed by the nucleate pool boiling of hydrogen gas bubble due to immense local heating of tool tip immersed in electrolyte. The gas bubble growth dynamics is a complicated phenomenon. It is changing the isolation between the cathode tip from the electrolyte and hence creating a varying electric field. This varying electric field in turn affects the bubble growth dynamics.

Fig. 9. Operational flow of ECSMM process showing intermediate processes.

(oscillations) varies with varying supply voltage (Kulkarni, 2000). The sparking frequency is high (in the tens of MHZ range) and it lowers (in the few hundreds of kHz range) for higher supply voltage. This supports the possibilities of the many partial sparks due to breakdown of the single isolated hydrogen gas bubbles. Theses partial sparks or discharges are of less current value and hence having less energy. These may not result in material removal from

Thus ECSMM process comprises of many intermediate processes such as electro chemical action causing the migration of ions, followed by the nucleate pool boiling of hydrogen gas bubble due to immense local heating of tool tip immersed in electrolyte. The gas bubble growth dynamics is a complicated phenomenon. It is changing the isolation between the cathode tip from the electrolyte and hence creating a varying electric field. This varying

the workpiece. These may die out before reaching the workpiece surface.

Fig. 9. Operational flow of ECSMM process showing intermediate processes.

**5.1 Intermediate processes and their interrelation** 

electric field in turn affects the bubble growth dynamics.

It starts with the electron generation, these in turn generating the secondary electrons and hence causing the electron avalanche. These energetic electrons get drifted away from cathode (tool) to the work piece very quickly due to the high potential gradient getting generated within the tool – work piece gap because of the hydrogen bubble isolating the electrolyte, as described. These drifted electrons bombard on the work piece surface. A large current spike is seen as a result of electron flow from cathode to work piece as actually seen during the transient current measurements. The bombardment of electrons on the work piece surface results in intense heating and hence metal removal takes place.

The overall material mechanism of the ECSM process can be understood in the light of the electrochemistry, heat transfer, ionization theory and electrical response of the system. The operational flow of the overall process is as shown in Figure 9. Each involved intermediate process and the cross relation with other sub processes is illustrated further in section 5.1.1- 5.1.4.

#### **5.1.1 Electrochemical process**

When the supply to the electrolyte cell is applied in the proper polarity, (i.e. positive terminal connected to anode and negative terminal to cathode) electrochemical action starts. electrons move from the cathode–electrolyte interface, and go to the solution. At the anode– electrolyte interface, equal number of electrons are discharged from the solution to the anode. Electrochemical reactions that occur at the electrode–electrolyte interface continuously supply electrons from cathode to solution and solution to anode. The type of reaction depends on the characteristics of electrodes, electrolyte and applied voltage. This is called as the 'migration' state of the ECSMM process.

#### **a. Reactions at anode and electrolyte interface:**

The electrochemical reactions at anode–electrolyte interface cause generation of oxygen gas. Dissolution of anode does not occur as the anode material used is graphite which is non consumable.

$$\text{4(OH)}\cdot \rightarrow 2\text{H}\_2\text{O} + \text{O}\_2 + 4\text{e}^-$$

#### **b. Reactions at cathode and electrolyte interface:**

Following electrochemical reactions take place at cathode–electrolyte interface, and cause evolution of hydrogen gas.

$$\begin{array}{c} \text{Cu}^{2+} + 2\text{e}^- \rightarrow \text{Cu} \\\\ 2\text{H}^+ + 2\text{e}^- \rightarrow \text{H}\_2\uparrow \\\\ \text{Na}^+ + \text{e}^- \rightarrow \text{Na} \\\\ 2\text{Na} + 2\text{H}\_2\text{O} \rightarrow 2\text{NaOH} + \text{H}\_2\uparrow \\\\ 2\text{H}\_2\text{O} + 2\text{e}^- \rightarrow \text{H}\_2\uparrow + 2\text{OH}^- \end{array}$$

Hydrogen gas evolves at the cathode, subsequently forming an isolating film, as depicted in Figure 8 a, or b, which leads to sparking across the bubbles between the cathode and electrolyte interface.

#### **c. Reduction of electrolyte in the bulk:**

It is given by:

$$\text{NaOH} \rightarrow \text{Na} + \text{OH} -$$

These liberated positive ions move towards cathode and negative ions move towards anode. In the external circuit, electrons move towards the cathode–electrolyte interface, and go to the solution. At the anode–electrolyte interface equal numbers of electrons are discharged from the solution to the anode. Electrochemical reactions that occur at the electrode–electrolyte interface continuously supply electrons from cathode to solution and solution to anode. This ionic and electronic current is the average current of the order of 100 – 200 mA.

#### **5.1.2 Nucleate pool boiling of hydrogen bubble**

The tip of the cathode gets heated up this causes nucleate pool boiling of hydrogen bubble that leads to development and formation of isolation vapor chamber of H2 gas. The heat transfer controlled growth model applies and the radius of the bubble as a function of time can be found by the corresponding equations. According to this model, the vapor bubble starts growing till it reaches its departure diameter, reaching which the bubble gets detached from the lower surface of the tool. An isolating film of hydrogen gas bubble covers the cathode tip portion in the electrolyte, abruptly a large dynamic resistance is present and the current through the circuit becomes almost zero. At the same moment, a high electric field of the order of 107 V/m gets applied. This high electric field causes the bubble discharge, sparking takes place. This leads to generation of energetic electros. These electrons generate secondary electrons. These get drifted towards the workpiece surface due to potential gradient.

#### **5.1.3 Sparking and electron avalanche**

The high electric field causes spark within the gas bubble isolating the tip. The spark should occur between the tip of the tool and the inner surface of the electrolyte. At the instant when spark occurs, an avalanche of electrons caused by ionization flow towards work piece kept around 20 µm distance away from the tool tip. This avalanche of electron is manifested as the current pulses of short duration and high amplitude as seen in Figure 7a,b. As the potential gradient after varied varied, these electrons drift through the sparking channel towards the work piece surface. Experimentally it is found that this time to reach the electron avalanche to the work piece depends on the separation distance. This time is longer for work piece kept at a 500 µm distance (Kulkarni et al., 2002) than that kept at 20 µm. This fact supports that it is a drifting phenomena of the electron avalanche.

#### **5.1.4 Material removal**

The bombardment of electrons on the glass work piece surface results in intense heating and hence material removal takes place. Atoms of the parent material get dislodged and material removal takes place. There are partial sparks occurring, theses may not be having enough energy to cause the material removal. These hamper the efficiency of the process and also affect the surface finish.

NaOH → Na+ + OH− These liberated positive ions move towards cathode and negative ions move towards anode. In the external circuit, electrons move towards the cathode–electrolyte interface, and go to the solution. At the anode–electrolyte interface equal numbers of electrons are discharged from the solution to the anode. Electrochemical reactions that occur at the electrode–electrolyte interface continuously supply electrons from cathode to solution and solution to anode. This ionic and electronic current is the average current of the order of

The tip of the cathode gets heated up this causes nucleate pool boiling of hydrogen bubble that leads to development and formation of isolation vapor chamber of H2 gas. The heat transfer controlled growth model applies and the radius of the bubble as a function of time can be found by the corresponding equations. According to this model, the vapor bubble starts growing till it reaches its departure diameter, reaching which the bubble gets detached from the lower surface of the tool. An isolating film of hydrogen gas bubble covers the cathode tip portion in the electrolyte, abruptly a large dynamic resistance is present and the current through the circuit becomes almost zero. At the same moment, a high electric field of the order of 107 V/m gets applied. This high electric field causes the bubble discharge, sparking takes place. This leads to generation of energetic electros. These electrons generate secondary electrons. These get drifted towards the workpiece surface due

The high electric field causes spark within the gas bubble isolating the tip. The spark should occur between the tip of the tool and the inner surface of the electrolyte. At the instant when spark occurs, an avalanche of electrons caused by ionization flow towards work piece kept around 20 µm distance away from the tool tip. This avalanche of electron is manifested as the current pulses of short duration and high amplitude as seen in Figure 7a,b. As the potential gradient after varied varied, these electrons drift through the sparking channel towards the work piece surface. Experimentally it is found that this time to reach the electron avalanche to the work piece depends on the separation distance. This time is longer for work piece kept at a 500 µm distance (Kulkarni et al., 2002) than that kept at 20 µm. This

The bombardment of electrons on the glass work piece surface results in intense heating and hence material removal takes place. Atoms of the parent material get dislodged and material removal takes place. There are partial sparks occurring, theses may not be having enough energy to cause the material removal. These hamper the efficiency of the process and also

fact supports that it is a drifting phenomena of the electron avalanche.

**c. Reduction of electrolyte in the bulk:** 

**5.1.2 Nucleate pool boiling of hydrogen bubble** 

It is given by:

100 – 200 mA.

to potential gradient.

**5.1.4 Material removal** 

affect the surface finish.

**5.1.3 Sparking and electron avalanche** 

At the same instant, the bubble geometry gets disturbed the contact between the tool and the electrolyte reestablishes. Electrochemical reaction takes over, bubble gets built up and the cycle keeps repeating itself. This makes the process discrete and repetitive.

All these intermediate processes described in sections 5.1 through 5.1.4 are correlated with the transient current pulses as observed in Figures 7 a and b. Figure 10 presents this correlation pictorially. The figure is self explanatory illustrating the time events during the ECSMM process w.r.t current.

T: Time between two sparks, i.e. time required for the bubble growth till isolation of tool tip from electrolyte (T ranges between few hundreds of µs to few tens of ms)

t: Time required to reach the electron avalanche to the work piece surface

(t ranges between tens of µs to few hundreds of µs)

Sparking frequency fsparking = 1/(T+t)

(fsparking ranges between few hundred hertz to few tens of kHz)

Fig. 10. Part of an entire transient, instantaneous current pulse illustrating various time events during the ECSMM process w.r.t current

#### **6. Concluding remarks**

ECSMM process is found to be suitable for production of micro channels on glass pellets. The width of the micro channels achieved is in the range of 400 – 1100 µm. The depth achieved is in the range of 75 -120 µm. The time required to form these micro channels of 5mm length is about 5000 µm. SEM analysis shows that the micro machined surface is produced by melting and vaporization. The current pulses show the stochastic nature of the spark formation process.

The material removal mechanism is complex. It involves various intermediate processes such as: electrochemical reactions followed by nucleate pool boiling, followed by breakdown of hydrogen bubbles, generating the electrons, these electrons drifting towards the workpiece and causing the material removal. The process starts all over again by electrochemical reactions once the bubbles are burst due to sparking. And re establishment of contact between tool electrode – electrolyte takes place.

Close control for gap adjustment is must. Research efforts must be made to reduce the low energy sparks due to partial isolation to enhance the efficiency of the process and surface finish.

#### **7. Acknowledgements**

I am indebted to Prof. V K Jain for his immense guidance and support throughout my academic life at IIT Kanpur. I am thankful to Prof. K A Misra for his guidance in carrying out the work. Financial support for this work from Department of Science and Technology, Government of India, New Delhi, is gratefully acknowledged (Grant no. SR/S3/MERC-079/2004). Thanks are due to the staff at Manufacturing Science Lab and Centre for Mechatronics, at IIT, Kanpur. Ms. Shivani Saxena and Mr. Ankur Bajpai, Research Associates in the project, helped in carrying out the experiments. Their help is duly acknowledged. Thanks are also due to the staff at Glass Blowing section of IIT, Kanpur.

#### **8. References**


Close control for gap adjustment is must. Research efforts must be made to reduce the low energy sparks due to partial isolation to enhance the efficiency of the process and surface

I am indebted to Prof. V K Jain for his immense guidance and support throughout my academic life at IIT Kanpur. I am thankful to Prof. K A Misra for his guidance in carrying out the work. Financial support for this work from Department of Science and Technology, Government of India, New Delhi, is gratefully acknowledged (Grant no. SR/S3/MERC-079/2004). Thanks are due to the staff at Manufacturing Science Lab and Centre for Mechatronics, at IIT, Kanpur. Ms. Shivani Saxena and Mr. Ankur Bajpai, Research Associates in the project, helped in carrying out the experiments. Their help is duly acknowledged. Thanks are also due to the staff at Glass Blowing section of IIT,

Basak, I. Ghosh, A. (1992). Mechanism of Material Removal in Electrochemical Discharge

Basak, I., and Ghosh, A. (1996). Mechanism of Spark Generation During Electrochemical

Bhattacharyya, B., Doloi, B. N., and Sorkhel, S. K. (1999). Experimental Investigation Into

Claire, L.C., Dumais, P., Blanchetiere, C., Ledderhof, C.J., and Noad, J.P., (2004). Micro

Coraci, A., Podarul, C., Maneal, E., Ciuciumis, A. and Corici, O. (2005). New technological

Crichton, I.M. and McGough, J.A. (1985). Studies of the discharge mechanisms in electrochemical arc machining', *J. of Appl. Electrochemistry*, Vol. l15, pp.113–119 Deepshikha, P. (2007). Generation of microchannels in ceramics (quartz) using

Fascio, V., Wüthrich, R. and Bleuler, H. (2004). Spark assisted chemical engraving in the

Han, M-S., Min, B-K. and Lee, S.J. (2008). Modeling gas film formation in electrochemical

discharge machining processes using a side-insulated electrode', *J. Micromech.* 

electrochemical spark machining, MTech thesis, IIT Kanpur

*Microeng*., doi: 10.1088/0960-1317/18/4/045019

light of electrochemistry, *Electrochimica Acta*, Vol. 49, pp.3997–4003

Machining: A Theoretic Model and Experimental Verification. J. Mater. Process.

Discharge Machining: A Theoretical Model and Eexperimental Investigation. Jr. of

Electrochemical Discharge Machining of Non conductive Ceramic Material. Journal

channel arrays in borophsphosilicate Glass for Photonic Device and optical sensor applications, *Tokyo konfarensu Koen Yoshishu* L1351C, 294authors name

surface micro fabrication methods used to obtain microchannels based systems onto various substrates', *Semiconductor Conference CAS 2005 Proceedings*, Vol. 1,

finish.

Kanpur.

**8. References** 

missing

pp.249–252

Technology, 71, 350–359

Materials Processing Technology, 62 46-53

of Materials Processing Technology, 95, 145-154

**7. Acknowledgements** 


Wuthrich, R., Fascio, V., Viquerat, D. and Langen, H. (1999) 'In situ measurement and micromachining of glass', *Int. Symposium on Micromechatronic and Human Science*, pp.185–191

## **Integrated MEMS: Opportunities & Challenges**

P.J. French and P.M. Sarro

*Delft University of Technology, The Netherlands* 

#### **1. Introduction**

252 Micromachining Techniques for Fabrication of Micro and Nano Structures

Wuthrich, R., Fascio, V., Viquerat, D. and Langen, H. (1999) 'In situ measurement and

pp.185–191

micromachining of glass', *Int. Symposium on Micromechatronic and Human Science*,

For almost 50 years, silicon sensors and actuators have been on the market. Early devices were simple stand-alone sensors and some had wide commercial success. There have been many examples of success stories for simple silicon sensors, such as the Hall plate and photo-diode. The development of micromachining techniques brought pressure sensors and accelerometers into the market and later the gyroscope. To achieve the mass market the devices had to be cheap and reliable. Integration can potentially reduce the cost of the system so long as the process yield is high enough and the devices can be packaged. The main approaches are; full integration (system-on-a-chip), hybrid (system-in-a-package) or in some cases separate sensors. The last can be the case when the environment is unsuitable for the electronics. The critical issues are reliability and packaging if these devices are to find the applications. This chapter examines the development of the technologies, some of the success stories and the opportunities for integrated Microsystems as well as the potential problems and applications where integration is not the best option.

The field of sensors can be traced back for thousands of years. From the moment that humans needed to augment their own sensors, the era of measurement and instrumentation was born. The Indus Valley civilisation (3000-1500 BC), which is now mainly in Pakistan, developed a standardisation of weight and measures, which led to further developments in instrumentation and sensors. The definition of units and knowing what we are measuring are essential components for sensors. Also if we are to calibrate, we need a reference on which everyone is agreed.

When we think of sensors, we think in terms of 6 signal domains, and in general converting the signal into the electrical domain. The electrical domain is also one of the 6 domains. The signal domain is not always direct, since some sensors use another domain to measure. A thermal flow sensor is such an example, and these devices are known as "tandem sensors". The signal domains are illustrated in Figure 1.

Over the centuries many discoveries led to the potential for sensor development. However, up to the 2nd half of the 20th century sensor technology did not use silicon. Also some effects in silicon were known, this had not led to silicon sensors. The piezoresistive effect was discovered by Kelvin in the 19th century and the effect of stress on crystals was widely studied in the 1930s, but the measurement of piezoresistive coefficients made by Smith in 1954, showed that silicon and germanium could be good options for stress/strain sensors (Smith, 1954). Many other examples can be found of effects which were discovered and a century later found to be applicable in silicon.

Fig. 1. The six signal domains

An important step towards The beginnings of integrated sensors go back to the first transistor, invented in 1947 by William Shockley, John Bardeen and Walter Brattain, while working at Bell Labs., which was fabricated in germanium. This quickly led to thoughts of integrating more devices into a single piece of semiconductor. In 1949 Werner Jacobi working at Siemens filed a patent for an integrated-circuit-like semiconductor amplifying device (Jacobi, 1949). In 1956 Geoffrey Dummer, in the UK, tried to make a full IC but this attempt was unsuccessful. In 1958 Jack Kilby, from Texas Instruments made the first working IC in germanium (Texas Instruments, 2008). This first device is illustrated in Figure 2. Six months later Robert Noyce, from Fairchild Semiconductor came up with hi own IC in silicon and manage to address a number of practical problems faced by Kilby. From these simple beginnings has come a major industry worth billions. John Bardeen, Walter H. Brattain and William B. Shockley won the Nobel Prize in 1956 and Jack Kilby in 2000.

Fig. 2. First working IC

The discovery of sensing effects in silicon and the development of electronic devices in silicon led to many new sensor developments. In the 1950s the idea of p-n junctions for photocells was first investigated (Chapin, 1954). Staying within the radiation domain groups

An important step towards The beginnings of integrated sensors go back to the first transistor, invented in 1947 by William Shockley, John Bardeen and Walter Brattain, while working at Bell Labs., which was fabricated in germanium. This quickly led to thoughts of integrating more devices into a single piece of semiconductor. In 1949 Werner Jacobi working at Siemens filed a patent for an integrated-circuit-like semiconductor amplifying device (Jacobi, 1949). In 1956 Geoffrey Dummer, in the UK, tried to make a full IC but this attempt was unsuccessful. In 1958 Jack Kilby, from Texas Instruments made the first working IC in germanium (Texas Instruments, 2008). This first device is illustrated in Figure 2. Six months later Robert Noyce, from Fairchild Semiconductor came up with hi own IC in silicon and manage to address a number of practical problems faced by Kilby. From these simple beginnings has come a major industry worth billions. John Bardeen, Walter H. Brattain and William B. Shockley won the Nobel Prize in 1956 and Jack Kilby in 2000.

The discovery of sensing effects in silicon and the development of electronic devices in silicon led to many new sensor developments. In the 1950s the idea of p-n junctions for photocells was first investigated (Chapin, 1954). Staying within the radiation domain groups

Fig. 1. The six signal domains

Fig. 2. First working IC

is Philips and Bell Labs. worked in parallel to develop the first CCD devices (Sangster, 1959 & Boyle, 1970).

At Philips, in the Netherlands, work had begun on a silicon pressure sensor and this early micromachined sensor is given in Figure 3 (Gieles, 1968 & 1969). The membrane was made using spark erosion and chemical etching, but the breakthrough was that the whole structure was in one material and therefore thermal mismatches were avoided.

Fig. 3. Early pressure sensor in the early 1960's

Silicon had now been shown to be a material with many effects interesting for sensor development. The work of Gieles showed that the material could be machined. Work from Bean (Bean, 1978) showed the greater opportunities etching silicon with anisotropic etchants, and Petersen (Petersen, 1982) showed the great mechanical properties of silicon. The early days of IC and sensor development were quite separate, but time has shown that these two fields can benefit from each other leading to new devices with greater functionality.

#### **2. Technology**

Many of the technologies used in silicon sensors were developed for the IC industry, although the development of micromachining led to a new range of technologies and opportunities for new devices. IC technology is basically a planar technology, whereas micromachining often requires working in 3 dimensions which has presented new challenges, in particular when the two technologies were combined to make smart devices.

#### **2.1 Planar IC technology**

The basis of planar technology was developed in the 1940s with the development of a pnjunction, although the major breakthrough was in 1958 with the first IC. This development enabled more and more devices to be integrated into a single piece of material. IC processing can be seen as a series of steps including; patterning, oxidation, doping, etching and deposition. These have been developed over the decades to optimise for the IC requirements and to advance the devices themselves. The following sections will give a brief description of the main steps.

#### **2.1.1 Lithography**

Lithography is a basic step carried out a number of times during a process. Basically a resist layer is spun on to the wafer and, after curing, exposed to UV light through a mask. If we use positive resist, this will soften through exposure and negative resist will harden. This can be done using a stepper (which projects the image onto each chip and steps over the wafer) or a contact aligner where the mask is a 1:1 image of the whole wafer. There are also techniques such as e-beam and laser direct write.

#### **2.1.2 Oxidation and deposition**

Silicon oxidises very easily. Simply left exposed at room temperature and oxide layer of 15- 20Å will be formed. For thicker oxides the wafer is exposed to an oxygen atmosphere at temperatures between 700-1200oC. For thick oxides, moisture is added (wet oxidation) to increase the growth rate.

A number of deposition steps are used in standard processing. The first of these is epitaxy. Epitaxy is the deposition, using chemical vapour deposition (CVD), of a thick silicon layer, usually single crystal, although polycrystalline material can also be deposited in an epireactor (Gennissen, 1997). The second group of depositions are low pressure CVD (LPCVD) and plasma enhanced CVD (PECVD). Some examples of LPCVD processes are given in Table 1. PECVD uses similar gasses, but the use of a plasma reduces the temperature at which the gasses break down, which is of particular interest with post-processing, where thermal budget is limited (Table 2). The temperatures for PECVD can be reduced through adjusting other process parameters. These are only examples and there are many other options.


Table 1. Examples of LPCVD processes.


Table 2. Examples of PECVD processes.

The last of the deposition processes is the metallisation, which is usually done by sputtering or evaporation, which is widely used for metals.

#### **2.1.3 Doping**

256 Micromachining Techniques for Fabrication of Micro and Nano Structures

can be done using a stepper (which projects the image onto each chip and steps over the wafer) or a contact aligner where the mask is a 1:1 image of the whole wafer. There are also

Silicon oxidises very easily. Simply left exposed at room temperature and oxide layer of 15- 20Å will be formed. For thicker oxides the wafer is exposed to an oxygen atmosphere at temperatures between 700-1200oC. For thick oxides, moisture is added (wet oxidation) to

A number of deposition steps are used in standard processing. The first of these is epitaxy. Epitaxy is the deposition, using chemical vapour deposition (CVD), of a thick silicon layer, usually single crystal, although polycrystalline material can also be deposited in an epireactor (Gennissen, 1997). The second group of depositions are low pressure CVD (LPCVD) and plasma enhanced CVD (PECVD). Some examples of LPCVD processes are given in Table 1. PECVD uses similar gasses, but the use of a plasma reduces the temperature at which the gasses break down, which is of particular interest with post-processing, where thermal budget is limited (Table 2). The temperatures for PECVD can be reduced through adjusting other process parameters. These are only examples and there are many other

Layer Gasses Temperature

SiH4

SiH2Cl2 + NH3 SiH4 + NH3 SiH4+O2 SiH4+O2 +PH3 SiH4+O2 +BCl3 SiH4+O2 +PH3 +BCl3 550oC-700oC 750oC-900oC 700oC-800oC 400oC-500oC 400oC-500oC 400oC-500oC 400oC-500oC 900oC-1050oC

400oC 400oC 400oC 350oC 400oC 400oC

400oC

SiH4+ CH4

SiH4 + NH3 +N2 SiH4+ N2 +N2O TEOS+O2

SiH4+ CH4

The last of the deposition processes is the metallisation, which is usually done by sputtering

SiH4+ N2 +N2O +NH3 SiH4+ N2 +N2O +PH3 +B2H6

Layer Gasses Temperature

SiH4

techniques such as e-beam and laser direct write.

**2.1.2 Oxidation and deposition** 

increase the growth rate.

options.

Polysilicon Silicon nitride

Silicon carbide

Silicon nitride

Oxynitride

Silicon carbide

Silicon dioxide undoped Silicon dioxide, (TEOS)

a-Si

Silicon dioxide undoped PSG (phosphorus doped) BSG (boron doped)

BPSG (phosphorus/boron doped)

Table 1. Examples of LPCVD processes.

BPSG (phosphorus/boron doped)

Table 2. Examples of PECVD processes.

or evaporation, which is widely used for metals.

An essential part of making devices is to be able to make p and n type regions. The main dopants are: As, P and Sb for n-type material and B for p-type material. The main techniques to dope silicon are diffusion and implantation. Diffusion is a process where the wafer is exposed to a gas containing the dopant atoms at high temperature. Implantation is a process in which the ions are accelerated towards the wafer at high speed to implant into the material.

#### **2.1.4 Etching**

Etching can be divided into two groups, wet and dry. Standard wet etching, in standard IC processing, is used for etching a wide range of materials. However, in recent years, dry etching is also widely used. Commonly used wet etchants are given in Table 3.


Table 3. Commonly used wet etchants.

#### **3. Micromachining technologies**

Micromachining technologies moved the planar technology for IC processing into the 3rd dimension. These technologies can be divided into two main groups, bulk micromachining and surface micromachining. In addition there is epi-micromachining which is a variation on the surface micromachining. The following sections give a brief outline of these technologies. A more detailed description can be found in the chapter on micromachining.

#### **3.1 Bulk micromachining**

Bulk micromachining can be divided into two main groups: wet and dry. There are also other techniques such as laser drilling and sand blasting. The first to be developed was wet etching. Most wet micromachining processes use anisotropic, such as KOH, TMAH, hydrazine or EDP. These etchants have an etch rate dependant upon the crystal orientation allowing well defined mechanical structures (Bean, 1978). The basic structures made with these etchants are given in Figure 4 with their properties in Table 4.

All of these processes are relatively low temperature and can therefore be used as postprocessing after IC processing, although care should be taken to protect the frontside of the wafer during etching.

Bulk micromachining can also be achieved through electrochemical etching in HF. For this etchant there are two distinct structures. The first is micro/nano porous which is usually an isotropic process, or macro-porous which is an anisotropic process. The micro/nano pore structure can be easily removed due to its large surface area to leave free-standing structures (Gennissen, 1995). Porous silicon/silicon carbide can then be used as a sensor material such as humidity or ammonia sensors (O' Halloran, 1998, Connolly 2002).

Fig. 4. Basic bulk micromachined structures using wet anisotropic etchants.


Table 4. Properties of main anisotropic etchants

The formation of macroporous silicon is usually done using n-type material and illumination from the backside to achieve deep holes with high aspect ratio. The idea was first proposed by (Lehmann 1996) and has been used to make large capacitors (Roozeboom, 2001) and micromachined structure (Ohji 1999). Both of these structures are illustrated in Figure 5.

The macro-porous process usually requires low n-doped material and illumination from the backside, which may not be compatible with the IC process. However, some macro-porous etching has been achieved in p-type material (Ohji, 2000), although the process is more difficult to control.

Deep reactive ion etching (DRIE), addressed some of the limitations of wet etching, although the process is more expensive. Two main processes are cryogenic (Craciun 2001) and Bosch processes (Laemer 1999). The cryogenic process works at about –100oC and uses oxygen to passivation of the sidewall during etching to maintain vertical etching. The Bosch process uses a switching between isotropic etching, passivation and ion bombardment. This results in a rippled sidewall, although recent developments allow faster switching without losing etchrate, thus significantly reducing the ripples. The etching can be performed from both front and back-side and can be combined with the electronics. In addition to DRIE being used for making 3-D mechanical structures, it has been applied to packaging (Roozeboom 2008).

Fig. 4. Basic bulk micromachined structures using wet anisotropic etchants.

(100/(111)

200:1

Hydrazine SiO2, SiN Metals 0.5-3 16:1 100 <<100 Toxic, potentially

The formation of macroporous silicon is usually done using n-type material and illumination from the backside to achieve deep holes with high aspect ratio. The idea was first proposed by (Lehmann 1996) and has been used to make large capacitors (Roozeboom, 2001) and micromachined structure (Ohji 1999). Both of these structures are illustrated in Figure 5. The macro-porous process usually requires low n-doped material and illumination from the backside, which may not be compatible with the IC process. However, some macro-porous etching has been achieved in p-type material (Ohji, 2000), although the process is more

Deep reactive ion etching (DRIE), addressed some of the limitations of wet etching, although the process is more expensive. Two main processes are cryogenic (Craciun 2001) and Bosch processes (Laemer 1999). The cryogenic process works at about –100oC and uses oxygen to passivation of the sidewall during etching to maintain vertical etching. The Bosch process uses a switching between isotropic etching, passivation and ion bombardment. This results in a rippled sidewall, although recent developments allow faster switching without losing etchrate, thus significantly reducing the ripples. The etching can be performed from both front and back-side and can be combined with the electronics. In addition to DRIE being used for making 3-D mechanical structures, it has been applied to packaging (Roozeboom 2008).

Etch rate

SiO2, SiN 0.3-1.5 120 60 Toxic

SiO2 [Å/h]

1700-

(100) m/min Comments

SiN [Å/h]

35:1 <100 <10 Expensive

<sup>3600</sup> <10 Not cleanroom

explosive

compatible

Etchant Mask

difficult to control.

EDP Au, Cr, Ag, Ta,

KOH SiN, Au 0.5-2, up to

TMAH+ IPA SiO2, SiN 0.2-1, up to

Table 4. Properties of main anisotropic etchants

Fig. 5. (left) vertical holes using macro-porous techniques (reproduced with kind permission Fred Roozeboom Philips), and (right) free standing structure (Ohji).

#### **3.2 Surface micromachining**

Surface micromachining is quite different from bulk micromachining both in terms of processing steps and dimensions. Basically, this involves the deposition of thin films and selective removal to yield free standing structures. The basic process is given in Figure 6, although this can be augmented with additional sacrificial and mechanical layers.

There are many possible combinations of sacrificial and mechanical layers and a few examples are given in Table 5. It is important with deposited layers to have good stress control, preferably low tensile stress with little or no stress profile (Guckel, 1988, French 1996, 1997, Pakula, 2001). A further important issue when fabricating surface micromachined structures is the release while avoiding stiction. Techniques to achieve this include freeze-drying, super critical drying. Alternatively, vapour-etching has been applied, or dry etching of a sacrificial polymer layer.

Fig. 6. Basic Surface micromachining process, (ia) deposition and patterning of sacrificial layer, (ib) deposition and patterning of mechanical layer and (ic) sacrificial etching, ii lateral view of typical structures.


Table 5. Examples of combinations of sacrificial and mechanical layers.

#### **3.3 EPI micromachining**

Epi-micromachining where the epitaxial layer is used as the mechanical layer. It can be seen as a variation on surface micromachining. There are a number of processes available which are described below.

#### **3.3.1 SIMPLE**

The SIMPLE process (Silicon Micromachining by Plasma Etching), forms micromachined structures using a single etch step (Li, 1995). This process makes use of a Cl2/BCl3 chemistry which etches low doped material anisotropically and n-type material above a threshold of about 8x1019cm-2, isotropically. The basic process sequence is shown in Figure 7. The first additional step, to the bipolar process, is a heavily doped buried layer since the bipolar buried layer has a doping level which is too low to be under-etched. This is followed by the formation of the standard bipolar buried layer and epitaxial layer (Figure 7a) followed by an additional deep diffusion where the mechanical structure will be formed (Figure 7b). After this a full standard bipolar process is performed (Figure 7c). The final structure is given in Figure 7d. This shows clearly how the vertical etching continues in the trench during the lateral etching of the buried layer.

Fig. 7. Main steps for the SIMPLE process.

#### **3.3.2 Silicon on Insulator (SOI)**

Silicon-on-insulator wafers are used in standard IC processing, which give a number of opportunities for making mechanical structures. Plasma etching through the epi layer followed by wet etching of the oxide layer is one option. This basic process is given in Figure 8 (Diem, 1995).

carbide HF

Resist, polymers Aluminium, silicon carbide Acetone, oxygen plasma

Epi-micromachining where the epitaxial layer is used as the mechanical layer. It can be seen as a variation on surface micromachining. There are a number of processes available which

The SIMPLE process (Silicon Micromachining by Plasma Etching), forms micromachined structures using a single etch step (Li, 1995). This process makes use of a Cl2/BCl3 chemistry which etches low doped material anisotropically and n-type material above a threshold of about 8x1019cm-2, isotropically. The basic process sequence is shown in Figure 7. The first additional step, to the bipolar process, is a heavily doped buried layer since the bipolar buried layer has a doping level which is too low to be under-etched. This is followed by the formation of the standard bipolar buried layer and epitaxial layer (Figure 7a) followed by an additional deep diffusion where the mechanical structure will be formed (Figure 7b). After this a full standard bipolar process is performed (Figure 7c). The final structure is given in Figure 7d. This shows clearly how the vertical etching continues in the trench during the

Silicon-on-insulator wafers are used in standard IC processing, which give a number of opportunities for making mechanical structures. Plasma etching through the epi layer followed by wet etching of the oxide layer is one option. This basic process is given in

Sacrificial layer Mechanical layer Sacrificial etchant

Silicon dioxide Aluminium Pad etch, 73% HF

Polysilicon Silicon nitride, silicon carbide KOH Polysilicon Silicon dioxide TMAH

Table 5. Examples of combinations of sacrificial and mechanical layers.

Silicon dioxide Polysilicon, silicon nitride, silicon

**3.3 EPI micromachining** 

lateral etching of the buried layer.

Fig. 7. Main steps for the SIMPLE process.

**3.3.2 Silicon on Insulator (SOI)** 

Figure 8 (Diem, 1995).

are described below.

**3.3.1 SIMPLE** 

An alternative use of SOI was developed in Twente, The Netherlands, (de Boer, 1995). This process uses different modes of plasma etching to manufacture the free standing structure as shown in Figure 9. First the epi is etched anisotropically, a CHF3 plasma etch is used to etch the underlying oxide and deposit a fluorocarbon (FC) on the sidewall which protects the sidewall during further etching and also has a low surface tension to reduce sticking. This is followed by a trench floor etch using SF6/O2/CHF3. Finally an isotropic RIE etch removes the silicon from under the upper silicon resulting in the free standing structure which is also isolated.

Fig. 9. Basic process steps for making free-standing SOI structures.

#### **3.3.3 Merged Epitaxial lateral Overgrowth (MELO)**

The MELO process (Merged Epitaxial lateral Overgrowth) is an extension of selective epitaxial growth SEG. Selective epitaxial growth uses HCl added to the dichlorosilane. The HCl etches the silicon. However, if there is a pattern of bare silicon and silicon dioxide, the silicon deposited on the oxide will have a rough grain like structure, with a large surface area, and will therefore be removed more quickly. Thus, although the growth rate will be lower than a normal deposition, a selective growth can be achieved. Once the silicon layer reaches the level of the oxide, both vertical and lateral growth will occur yielding lateral overgrowth. This basic process is illustrated in Figure 10 (Bartek, 1994).

Fig. 10. Basic SEG process extended to ELO

If two of these windows are close enough together they will merge giving the MELO process. As a result we have buried silicon dioxide islands. This lends itself well to micromachining as shown in Figure 11 (Kabir 1993). This process has the advantage of producing single crystal structures, but the disadvantages that beams must be orientated in the <100> direction due to growth mechanisms of silicon (Gennissen, 1999) and since growth continues both laterally and vertically, the lateral dimensions are limited.

Fig. 11. Basic MELO process.

#### **3.3.4 Sacrificial porous silicon**

A process which using silicon as both mechanical and sacrificial layer is the sacrificial porous silicon technique (Lang, 1995, Gennissen, 1995, 1999, Bell, 1996). This process makes use of the fact that, without illumination, p-type material is made porous selectively. The high surface area of the porous materials results in rapid etching, after porous formation, in KOH. This makes the material highly suitable as a sacrificial material. The porous silicon formation rate is highly dependent upon the current density, HF concentration, illumination and the doping in the substrate (Gennissen, 1999). The process sequence is given in Figure 12. In this case the selective etching of p-type material over the n-type epi is used. First a plasma etch is used to etch through the epi-layer to reveal the substrate. The porous layer is then formed and finally removed in KOH at room temperature.

Fig. 12. Basic process steps for sacrificial porous silicon based micromachining

The porous silicon technique is extremely simple and can be applied as a post processing step and it is therefore fully compatible with the electronic circuitry. The only remaining problem is to protect the areas of electronics and metallisation from the HF etchant. One disadvantage of this technique is the added process complexity introduced by the requirement of a backside electrical contact during etching.

#### **3.3.5 Epi-poly**

262 Micromachining Techniques for Fabrication of Micro and Nano Structures

If two of these windows are close enough together they will merge giving the MELO process. As a result we have buried silicon dioxide islands. This lends itself well to micromachining as shown in Figure 11 (Kabir 1993). This process has the advantage of producing single crystal structures, but the disadvantages that beams must be orientated in the <100> direction due to growth mechanisms of silicon (Gennissen, 1999) and since

A process which using silicon as both mechanical and sacrificial layer is the sacrificial porous silicon technique (Lang, 1995, Gennissen, 1995, 1999, Bell, 1996). This process makes use of the fact that, without illumination, p-type material is made porous selectively. The high surface area of the porous materials results in rapid etching, after porous formation, in KOH. This makes the material highly suitable as a sacrificial material. The porous silicon formation rate is highly dependent upon the current density, HF concentration, illumination and the doping in the substrate (Gennissen, 1999). The process sequence is given in Figure 12. In this case the selective etching of p-type material over the n-type epi is used. First a plasma etch is used to etch through the epi-layer to reveal the substrate. The porous layer is

then formed and finally removed in KOH at room temperature.

requirement of a backside electrical contact during etching.

Fig. 12. Basic process steps for sacrificial porous silicon based micromachining

The porous silicon technique is extremely simple and can be applied as a post processing step and it is therefore fully compatible with the electronic circuitry. The only remaining problem is to protect the areas of electronics and metallisation from the HF etchant. One disadvantage of this technique is the added process complexity introduced by the

growth continues both laterally and vertically, the lateral dimensions are limited.

Fig. 11. Basic MELO process.

**3.3.4 Sacrificial porous silicon** 

Epi-poly is a polysilicon layers grown in the epitaxial reactor. Although this technique departs from using single crystal silicon as a mechanical material it has greater flexibility in terms of lateral dimensions. Alternatively, the mechanical layers can be formed at the same time as the single crystal epi required for the electronics (Gennissen 1997). The basic process is shown in Figure 13. After the formation of the sacrificial oxide, a polysilicon seed is deposited. A standard epi growth will then form epi-poly on the seed and single crystal where the substrate is bare. The epi growth rate on the polysilicon seed is about 70% of that on the single crystal silicon. Therefore the total thickness of the sacrificial layer and seed can be adjusted to ensure a planar surface after epi growth. A nitride layer is used to stop the polysilicon oxidising during subsequent process, thus minimising intrinsic stress. The mechanical layer is then patterned and released through sacrificial etching as shown in Figure 13d.

Fig. 13. Basic epi-poly process.

The epi-poly process requires minimal additional processing before epitaxial deposition and no detrimental effect on the electronics' characteristics has been found. The process is therefore fully compatible with the electronics processing.

#### **3.4 Packaging**

Packaging for standard ICs has been standardised, but for sensor, many additional challenges can arise and should be considered at an early stage. One of the issues, for example, is that many sensors have to be exposed to the environment. Another issue can be the wire bonding.

#### **3.4.1 Wire bonding**

Standard IC processing usually uses wire bonding to make electrical contact from the chip to the outside world. On each chip, metal bond pads are made, and wires (usually gold or aluminium) and made from the pad to the package.

In standard IC, this is all sealed in the package, but if the surface has to be exposed, these wires may not be desired. For example, sensors for catheters may not want bond wires on the front of the chip. In this case through wafer interconnect may be the best option. These options are shown, with the example of a catheter in Figure 14.

Fig. 14. (a) multi-chip approach using wire bonding, (b) multichip approach using throughwafer-interconnect and a printed circuit board substrate.

#### **3.4.2 Flip-chip**

As the field of Microsystems grew, there was more interest in combining different chips in a single package. Wire bonding from chip-to-chip is an option. An alternative is flip-chip. Flip-chip is a process where chips can be mounted. For this process a solder-bump is made on the wafer, after which two chips can be aligned and soldered together. An example of the solder process is given in Figure 15. Adapted from (Fujitsu 2003)

Fig. 15. Example of the Fujitsu solder-bump process.

#### **3.4.3 Multi-chip package**

264 Micromachining Techniques for Fabrication of Micro and Nano Structures

Fig. 14. (a) multi-chip approach using wire bonding, (b) multichip approach using through-

As the field of Microsystems grew, there was more interest in combining different chips in a single package. Wire bonding from chip-to-chip is an option. An alternative is flip-chip. Flip-chip is a process where chips can be mounted. For this process a solder-bump is made on the wafer, after which two chips can be aligned and soldered together. An example of the

wafer-interconnect and a printed circuit board substrate.

solder process is given in Figure 15. Adapted from (Fujitsu 2003)

Fig. 15. Example of the Fujitsu solder-bump process.

**3.4.2 Flip-chip** 

There are a number of options for combining different chips in a single package through flip-chip, through wafer interconnect and simply wire bonding, some of these options , from NXP, The Netherlands, are given in Figure 16 (Roozeboom, 2008-2).

Fig. 16. Different approaches to systems-in-a-package, based on (Roozeboom, 2008-2).

#### **4. Process integration**

When deciding how to integrate it is important to choose how this is incorporated in the process. There are a number of options:


Post-processing (Bustillo, 1994, Pakula 2004)


With these options there are a number of issues concerning the integration of the sensor structures with the electronics. In many cases the best option is to separate the sensor from electronics and then combining into a single package. The issues for integration are discussed below.

COMPATIBILITY WITH THE CLEANROOM

Materials or processing steps for the sensor may not be compatible with the cleanroom and therefore cause contamination.


A major issue in industrial application is the yield. Standard IC processing has been developed to have a high yield. Any additional processing can potentially reduce yield, which will increase your overall cost.

APPLICATION ISSUES

High temperatures operations may not benefit from integration. In some applications, the integrated option may be too expensive. In other applications the environment where the sensor has to work, may not be suitable for the electronics. However, if the above issues can be addressed there can be great benefits from integration.

Both surface and bulk micromachining present challenges in terms of integration. The following sections describe a number of integration options.

#### **4.1 Bulk micromachining integration**

As shown above there are a number of process technologies, which can be combined with electronics. Bulk micromachining is, in general, low temperature and does not yield any thermal limitations to the electronics. In some DRIE processes charging of the gate oxide can occur, but adjustment of the process can avoid this. With wet etching the main issue is masking/protecting the front side.

Using wet anisotropic etching we require a masking layer on the back of the wafer, which can be a PECVD layer and etching is performed through the wafer to produce, for example, membranes for pressure sensors, or mass/spring structures for accelerometers. One of the issues is how to define the thickness of the membrane. This can be done is several ways (Palik 1982): time stop, p+ etchstop (Gianchandani, 1991), electrochemical etchstop (Kloeck, 1989) and galvanic etchstop (Ashruf 1998 and Connolly 2003).

#### **4.2 Surface micromachining integration**

Surface micromachining presents more challenges. Most LPCVD processes use to high a temperature to be suitable as post-processing. However, PECVD processes and metal layers do have a much lower temperature, and can therefore, in principle, be added after IC processing. These options are discussed further below.

#### **4.2.1 Pre-processing**

Although described as pre-processing there are usually some processing steps also required after the standard processing. In the case of pre-processing the major consideration is to ensure that the mechanical properties are not detrimentally affected by the standard processing. Polysilicon has been used for pre-processing and this basic structure is given in Figure 17 (Smith 1996, Gianchandani 1997).

With these options there are a number of issues concerning the integration of the sensor structures with the electronics. In many cases the best option is to separate the sensor from electronics and then combining into a single package. The issues for integration are

Materials or processing steps for the sensor may not be compatible with the cleanroom and

Both ICs and mechanical structures can be very sensitive to any additional thermal process.

A major issue in industrial application is the yield. Standard IC processing has been developed to have a high yield. Any additional processing can potentially reduce yield,

High temperatures operations may not benefit from integration. In some applications, the integrated option may be too expensive. In other applications the environment where the sensor has to work, may not be suitable for the electronics. However, if the above issues can

Both surface and bulk micromachining present challenges in terms of integration. The

As shown above there are a number of process technologies, which can be combined with electronics. Bulk micromachining is, in general, low temperature and does not yield any thermal limitations to the electronics. In some DRIE processes charging of the gate oxide can occur, but adjustment of the process can avoid this. With wet etching the main issue is

Using wet anisotropic etching we require a masking layer on the back of the wafer, which can be a PECVD layer and etching is performed through the wafer to produce, for example, membranes for pressure sensors, or mass/spring structures for accelerometers. One of the issues is how to define the thickness of the membrane. This can be done is several ways (Palik 1982): time stop, p+ etchstop (Gianchandani, 1991), electrochemical etchstop (Kloeck,

Surface micromachining presents more challenges. Most LPCVD processes use to high a temperature to be suitable as post-processing. However, PECVD processes and metal layers do have a much lower temperature, and can therefore, in principle, be added after IC

Although described as pre-processing there are usually some processing steps also required after the standard processing. In the case of pre-processing the major consideration is to ensure that the mechanical properties are not detrimentally affected by the standard processing. Polysilicon has been used for pre-processing and this basic structure is given in

discussed below.

YIELD

therefore cause contamination.

which will increase your overall cost.

**4.1 Bulk micromachining integration** 

masking/protecting the front side.

**4.2 Surface micromachining integration** 

Figure 17 (Smith 1996, Gianchandani 1997).

**4.2.1 Pre-processing** 

APPLICATION ISSUES

COMPATIBILITY WITH THE CLEANROOM

be addressed there can be great benefits from integration.

following sections describe a number of integration options.

1989) and galvanic etchstop (Ashruf 1998 and Connolly 2003).

processing. These options are discussed further below.

COMPATIBILITY OF THE PROCESSING

Fig. 17. Structure fabricated using pre-processing combined with CMOS.

#### **4.2.2 Integrated processing**

With the integrated process option the wafers are removed from the standard line and after the addition of micromachining steps return to the standard line. The position of the additional process steps is extremely important. In some cases the additional depositions are added after the main thermal processing but before the aluminium. Depending on the sensitivity of the electronics devices to thermal budget, a maximum thermal budget for the micromachining is determined. An example of an integrated process combining polysiliconbased surface micromachining with bipolar electronics is given in Figure 18 (van Drieenhuizen, 1994).

Fig. 18. Polysilicon based integrated surface micromachining process

A similar process has been developed by Fischer et. al. (Fischer, 1996) using an aluminium gate CMOS process. The basic process is shown in Figure 19. The interesting feature of this process is that the deposition of the micromechanical structures is performed before the gate oxidation.

Fig. 19. Compatible surface micromachining process: a) After completion of the CMOS process (before gate oxidation) and capping silicon nitride; b) formation of the sacrificial and mechanical layers followed by gate oxidation and aluminium deposition; c) formation of the resist protection mask and sacrificial etching.

#### **4.2.3 Post-processing**

In the post processing option wafers go through the complete standard process. After standard processing either existing layers can be used or additional layers can be added. Both approaches can be found in the literature. Two examples can be seen in Figure 20. The first, Figure 20a uses the gate poly as the mechanical material (Hierold 1996) and the second, Figure 20b, (Fedder, 1996) a combination of oxide and metal used in standard processes.

Fig. 20. Post processing micromachining with CMOS (a) using the gate poly and (b) using aluminium and oxide.

An alternative approach is to use one of the aluminium layers as a sacrificial layer, and protecting the remaining aluminium layers with oxide (Westberg, 1996). The resulting structure is shown in Figure 21. These approaches have the advantage of simplicity, but these layers are not optimised for their mechanical properties. The approach is through careful design.

Fig. 19. Compatible surface micromachining process: a) After completion of the CMOS process (before gate oxidation) and capping silicon nitride; b) formation of the sacrificial and mechanical layers followed by gate oxidation and aluminium deposition; c) formation of the

In the post processing option wafers go through the complete standard process. After standard processing either existing layers can be used or additional layers can be added. Both approaches can be found in the literature. Two examples can be seen in Figure 20. The first, Figure 20a uses the gate poly as the mechanical material (Hierold 1996) and the second, Figure 20b, (Fedder, 1996) a combination of oxide and metal used in standard processes.

Fig. 20. Post processing micromachining with CMOS (a) using the gate poly and (b) using

An alternative approach is to use one of the aluminium layers as a sacrificial layer, and protecting the remaining aluminium layers with oxide (Westberg, 1996). The resulting structure is shown in Figure 21. These approaches have the advantage of simplicity, but these layers are not optimised for their mechanical properties. The approach is through

resist protection mask and sacrificial etching.

**4.2.3 Post-processing** 

aluminium and oxide.

careful design.

Fig. 21. Micromachining integrated into a CMOS process

One of the limitations of post-processing is the thermal budget. In particular, if aluminium is used for metallization, the maximum temperature that can be used is about 400oC. An alternative is to use another metal. Figure 22 shows an example where tungsten is used as the metal (Bustillo, 1994).

Fig. 22. Post processing micromachining using Tungsten metallisation for the electronics.

Alternatively, low temperature deposition materials, such as PECVD layers can be used. One PECVD layer with good mechanical properties is SiC (Pakula, 2004). With these low temperatures layers such as polyimides (Bagolini, 2002) can be used as sacrificial layer.

#### **4.3 System-in-a-package**

In the development of IC the packaging options were wire bonding in a package. This is still used today, but there are many other options. In many applications, the best option is to combine the sensors and the electronics in a single package, but not on a single chip. In this case the sensor can be optimised, without compromising the electronics. Furthermore, the larger production volume of the electronics chip can help to keep costs down. An example of this is given in Figure 23.

Fig. 23. Tyre Pressure Sensing System developed by LV Sensors integrated ASIC, pressure sensor and 2-axis acceleration sensor (reproduced with kind permission Janusz Bryzek).

The package can then be built into a complete system, as shown in Figure 24.

Fig. 24. Complete system fitted with the valve and a wireless communication system (reproduced with kind permission Janusz Bryzek).

#### **4.4 Integrated sensor**

One of the problems in developing a simple electronic compass based on the Hall effect was the offset. This was solved by the development of the "Spinning" Hall plate (Munter, 1990, Bellakom 1994). This technique can reduce the offset by a factor of 1000 to make it practical to be used as a compass. However, in order to make full use of this advantage, the devices should be integrated with a low-offset amplifier. Since the Hall plate only requires a thin diffused layer, this can be integrated with both bipolar and CMOS. This has since become a commercial product developed by Xensor Integration, in Delft. A photograph of this device is given in Figure 25.

Fig. 23. Tyre Pressure Sensing System developed by LV Sensors integrated ASIC, pressure sensor and 2-axis acceleration sensor (reproduced with kind permission Janusz Bryzek).

The package can then be built into a complete system, as shown in Figure 24.

Fig. 24. Complete system fitted with the valve and a wireless communication system

One of the problems in developing a simple electronic compass based on the Hall effect was the offset. This was solved by the development of the "Spinning" Hall plate (Munter, 1990, Bellakom 1994). This technique can reduce the offset by a factor of 1000 to make it practical to be used as a compass. However, in order to make full use of this advantage, the devices should be integrated with a low-offset amplifier. Since the Hall plate only requires a thin diffused layer, this can be integrated with both bipolar and CMOS. This has since become a commercial product developed by Xensor Integration, in Delft. A photograph of this device

(reproduced with kind permission Janusz Bryzek).

**4.4 Integrated sensor** 

is given in Figure 25.

Fig. 25. Single Chip CMOS compatible Compass sensor based on hall technology. The chip incorporates a low-offset Hall sensor with ultra low-noise low offset amplifier, a sigma delta ADC and an SPI digital output. (Courtesy of Xensor Integration, The Netherlands).

Another sensor system which can benefit from integration is the temperature sensor. In this case the sensor can be a simple p-n junction or a transistor and the accuracy can be achieved through the read-out electronics. Furthermore, the electronics can be used to reduce the costs of calibration. A chip photograph of this device is given in Figure 26 (Bakker 1998, 2000, Pertijs 2005).

Fig. 26. Chip photograph and block diagram of a smart temperature sensor chip with bus interface.

A further example of sensor integrated is given in Figure 27. The basic sensor, was developed in the 1990s, and is shown on left. The thermopiles used for measuring temperature differences use polysilicon and aluminium. On the right is the integrated version, which contains all power management, read-out and A-D conversion. Although the chip is more expensive, the total system is much cheaper and more efficient.

Fig. 27. Two wind-sensors (left) non integrated and (right) fully integrated (Makinwa, 2002).

The examples given above are where using standard processing. There have been many successes by integrating micromachining with electronics on a single chip. An important breakthrough for silicon sensors was with the introduction of integrated silicon accelerometers. Such an example is given in Figure 28. These devices could incorporate selftest and read-out electronics to make a complete system.

Fig. 28. Analog Devices integrated accelerometer, (left) a 2-D accelerometer, (right) a 1-D device with indication of the different circuitry (Copyright Analog Devices, Inc. All rights reserved).

A further example of a fully integrated device is the TI mirror array which, known at the digital light processor (DLP). This device makes use of the multiple aluminium layers to create the mechanical structure and also the mirrors. This is an excellent example of integrated devices. For this application the reflectivity of the aluminium is essential (Hornbeck, 1996). This device uses three layers of aluminium-based films for the mirror and its suspension system. The build up of the device is illustrated in Figure 29.

Fig. 27. Two wind-sensors (left) non integrated and (right) fully integrated (Makinwa, 2002). The examples given above are where using standard processing. There have been many successes by integrating micromachining with electronics on a single chip. An important breakthrough for silicon sensors was with the introduction of integrated silicon accelerometers. Such an example is given in Figure 28. These devices could incorporate self-

Fig. 28. Analog Devices integrated accelerometer, (left) a 2-D accelerometer, (right) a 1-D device with indication of the different circuitry (Copyright Analog Devices, Inc. All rights

its suspension system. The build up of the device is illustrated in Figure 29.

A further example of a fully integrated device is the TI mirror array which, known at the digital light processor (DLP). This device makes use of the multiple aluminium layers to create the mechanical structure and also the mirrors. This is an excellent example of integrated devices. For this application the reflectivity of the aluminium is essential (Hornbeck, 1996). This device uses three layers of aluminium-based films for the mirror and

test and read-out electronics to make a complete system.

reserved).

Fig. 29. Schematic of the TI digital display device (from http://www.dlp.com/).

#### **5. Conclusions**

There is no simple answer whether or not to integrate, for all applications. In many applications, the best option is system-in-a-package, in others it is better to move the electronics further away to protect it from the environment (in particular with high temperature applications). Another issue which may work against integration is potential increase in costs, through the more complex processing. In these cases the multi-chip approach is used. There are, however, examples where great benefits can be gained from integrating the sensor with the electronics, and a number of examples are given in this chapter. There is no simple answer to all applications and each case needs to be examined in its own right to decide the best path.

#### **6. Acknowledgments**

The author would like to thank the many colleagues who have provided pictures included in this chapter. Also the Technology Foundation STW, in The Netherlands, for their funding of work in Delft and their general support for Applied Research in The Netherlands. Last, but not least, colleagues in DIMES for their collaboration over the years.

#### **7. References**


Bakker, A, (2000) *High-accuracy CMOS smart temperature sensors,* PhD thesis, TU Delft, The

Bartek, M. Gennissen, PTJ, Sarro, PM, French, PJ and Wolffenbuttel, RF, (1994) *An integrated* 

Bean, KE., (1978) *Anisotropic etching of silicon*, IEEE Trans Electron Devices, Vol ED-25, pp

T.E. Bell, TE, Gennissen, PTJ, de Munter, D and Kuhl, M (1996) *Porous silicon as a sacrificial* 

Bellekom, AA., and Munter, PJA, (1994) *Offset reduction in spnning-current Hall plates*, Sensors

Boyle, WS. & Smith, GE. (1970) *Charge coupled semiconductor devices*, Bell Syst. Techn. J., Vol

Bustillo, JM, Fedder, GK, Nguyen CT-C and Howe, RT, (1994) *Process technology for the* 

Chapin, DM., Fuller, CS. and Pearson, GL. (1954), *A new silicon p-n junction photocell for converting solar radiation into electrical power,* J. Appl. Phys. Vol 25, pp 676-677. Connolly, EJ, O'Halloran, GM, Pham, HTM, Sarro, PM, French,PJ (2002). *Comparison of* 

Connolly, EJ, Sakarya, S, French, PJ, Xia, XH and Kelly, JJ (2003) *A practical galvanic etch-stop* 

Craciun, G, Blauw, M, van der Drift E and French, PJ, (2001) *Aspect ratio and crystallographic* 

de Boer, M, Jansen, H and Elwenspoek, M, (1995), *Black silicon V: A study of the fabricating of* 

Diem, B, Rey, P, Renard, S, Viollet Bosson, S, Bono, H, Michel, F, Delaye, MT and Delapierre,

Fedder, GK, Santhanan, S, Read, ML, Eagle, SC, Guillou, DF, Lu, MS-C and Carley, LR,

Fischer, M, Nägele, M, Eichner, D, Schöllhorn, C and Strobel, R (1996*) Integration of surface* 

French, PJ, van Drieënhuizen, BP. Poenar, D, Goosen, JFL, Sarro, PM and Wolffenbuttel RF,

French, PJ, Sarro, PM, Mallée, R, Fakkeldij, EJM and Wolffenbuttel, RF, (1997), *Optimisation* 

Fujitsu (2003), *Fujitsu Achieves Breakthrough in Ultrafine-Pitch Solder Bumping*, http://www.fujitsu.com/global/news/pr/archives/month/2003/20031215-

sensors and actuators, Sensors and Actuators, VolA46-47, pp 8-26

Proceedings MEMS 96, San Diego, USA, Feb 1996, pp 13-18.

*processing*, IEEE JMEMS, Vol 5, (1996), pp 187-196.

*sensing applications*, Sensors and Actuators Vol A 99 pp 25-30.

Transducers 01, Munich Germany, June 2001,

*modular integration of CMOS and polysilicon microstructures*, Microsystem

*porous silicon, porous polysilicon and porous silicon carbide as materials for humidity* 

*in KOH using sodium hypochlorite*, Proceedings IEEE MEMS 2003, Kyoto, Japan,

*orientation dependence in deep dry silicon etching at cyrogenic temperatures,* Proceedings

*moveable structures for micro electromechanical systems*, Proceedings Transducers 95,

G, (1995) SOI 'SIMOX' from bulk to surface micromachining, a new age for silicon

(1996) *Laminated high-aspect ratio microstructures in a conventional CMOS process*,

*micromachined polysilicon mirrors and a standard CMOS process*, Sensors and

(1996), *The development of a low-stress polysilicon process compatible with standard device* 

*of a low stress silicon nitride process for surface micromachining applications*, Sensors and

*silicon colour sensor using selective epitaxial growth* Sensors and Actuators, Vol A41-42,

Netherlands, 2000, 132 p. ISBN: 90-901-3643-6.

*material*, J. Micromech. Microeng., Vol 6, pp 361-369.

pp 123-128.

1185-1193

49 pp 587- 593

Mater., Vol 5, pp 253-263.

Technology, Vol 1, pp 30-41.

January 2003, pp 566-569.

Stockholm, Sweden, pp 565-568.

Actuators, Vol A52, pp 140-144.

Actuators, Vol A58, pp 149-157.

02.html accessed August 2011


## **Modeling and Simulation of MEMS Components: Challenges and Possible Solutions**

Idris Ahmed Ali

*Dean, College of Engineering & Applied Sciences Al Ghurair University, Academic City, Dubai, UAE* 

#### **1. Introduction**

276 Micromachining Techniques for Fabrication of Micro and Nano Structures

Munter, PJA, (1990), *A low-offset spinning-current Hall plate*, Sensors & Actuators, Vol A21-

O'Halloran, GM, Groeneweg, J, Sarro, PM and French, PJ, *Porous silicon membrane for* 

Ohji, H, Trimp, PJ and French, PJ (1999) *Fabrication of free standing structures using a single step electrochemical etching in hydrofluoric acid*, Sensors and Actuators, . Vol A73 pp. 95-100 Ohji, H, French, PJ and Tsutsumi, K, (2000). *Fabrication of mechanical structures in p-type silicon using electrochemical etching*, Sensors and Actuators Vol A82 pp 254 – 258 Pakula, L, Pham HTM and French, PJ, (2001) *Novel material for low temperature post-processing micromachining*, Proceedings MME, Cork, Ireland, September 2001, pp 70-73 Pakula, LS, Yang, H, Pham, HTM, French, PJ and Sarro, PM, (2004) *Fabrication of a CMOS* 

Palik, ED, Faust, Jr., JW, Gray, HF and Greene, RF, (1982) *Study of the etch-stop mechanism in* 

Pertijs, MAP, (2005*). Precision Temperature Sensors in CMOS technology* promotor: prof. dr. ir. J.H. Huijsing, 330p. ISBN-10: 90-9020097-5; ISBN-13: 978-90-9020097-2 Petersen, KE., (1982) *Silicon as a mechanical material*, Poc. IEEE, VOL. 70, pp 420-457.

Roozeboom, F, Elfrink, R, Rijks, TGSM, Verhoeven, J, Kemmeren A, and van den

Roozeboom, F, Blauw, MA, Lamy, Y, van Grunsven, E, Dekkers, W, Verhoeven, JF, van den

Roozeboom, F, Klootwijk, J, Dekkers, W, Lamy, Y, van Grunsven, E and Kim, H, (2008-2)

Sangster, LFJ & Teer, K. (1969) *Bucket-brigade electronics - new possibilities for delay, time-axis conversion and scanning*, IEEE J. Solid-State Circuits, SC-4 pp 131-136. Smith, CS. (1954). *Piezoresistive effect in Germanium and Silicon*, Physical Review. Vol 94,

Smith, JH, Montague, S, Sniegowski, JJ, Murray, JR, Manginell, RP and McWhorter, PJ,

van Drieënhuizen, BP, Goosen, JFL, French, PJ, Li, XL, Poenar, D and Wolffenbuttel, RF,

Proceedings Eurosensors 94, Toulouse, France, September 1994, p 108 Westberg, D, Paul, O, Andersson, GI and Baltes, H (1996) *Surface micromachining by sacrificial* 

Texas Instruments. (2008) . *The chip that jack built,* HTML accessed 10 August 2011

*aluminium etching*, J. Micromech. Microeng., Vol 6, pp 376-384.

P. Ramm, eds.), Wiley-VCH Verlag, Weinheim, pp. 47-91.

*silicon*, J. Electrochem. Soc. 137 (1982) 2051-2059

Packaging, 24 (3) (2001) pp. 182-196

Technology, vol 51, No. 5

(1954) pp 42-49

2879, pp 306-314.

*humidity sensing applications*, Proceedings Eurosensors 98, Southampton, UK, 13-16

*compatible pressure sensor for harsh environments*, J. Micromech. Microeng., Vol 14, pp

Meerakker, J, (2001), Proc. 34th Int. Symp. on Microelectronics (IMAPS 2001), Baltimore, Oct. 9-11, 2001, pp. 477-483- Int. J. Microcircuits and Electronic

Heuvel, F, van der Drift, E, Kessels WMM and van de Sanden, MCM. (2008) *Deep Reactive Ion Etching of Through-Silicon Vias*, in 'Handbook of 3-D Integration: Technology and Applications of 3D Integrated Circuits', (P. Garrou, C. Bower and

*System-in-package integration of passives using 3D through-silicon vias*, Solid State

(1996), *Characterisation of the embedded micromachined device approach to the monolithic integration of MEMS with CMOS*, Proceedings SPIE Micromachining and Microfabrication Process Technology II, Austin, Texas, USA, October 1996, vol

(1994), *Surface micromachined module compatible with BiFET electronic processing*,

A23, pp 743-746.

1478-1483.

September 1998, pp 901-904.

Micro-electro-mechanical systems (MEMS) represent a very important class of systems having applications ranging from small embedded sensors and actuators, passive components in RF and microwave fields, and micro-mirrors in the optical range. The importance of MEMS stems from their many advantages, among which are, their small compact size amendable to integration with other components, low loss and parameter variability.

From structural point of view, each MEMS component is, by itself, a very small electromechanical system of heterogeneous structure composed of materials with different chemical composition (dielectric substrate, metal alloys and conducting wire) and different physical (electrical, thermal, mechanical) properties. Moreover, MEMS components may represent static systems or they may contain some moving parts, such as in variable capacitor, moving membranes and cantilevers. The dimensional scale of the different parts of MEMS components may vary from very small (microns or even nanometers) in one dimension, such as thickness of a plate, to comparatively large of few hundred microns in other dimensions, thus resulting in large aspect-ratios.

When MEMS components are put into oration, they constitute systems, in which electrical, thermal, mechanical, and other physical phenomena take place and interact with each other. From mathematical modeling and simulation point of view, this calls for multi-physics treatment, in which coupled systems of differential equations of different combinations of electromagnetic, mechanical, fluid, heat transfer and/or transport equations, are formulated then solved depending on the type of boundary conditions imposed by MEMS component under investigation.

Mathematical modeling and simulation has been used in all fields and disciplines of engineering for decades, for theoretical characterization of devices and systems before manufacturing, or even before prototyping, for a number of reasons among which are reduction in manufacturing cost and time. However, the heterogeneous nature of MEMS structures, coupled with multi-physics phenomena that take place during their operation, makes modeling and simulation of MEMS components, a complex and challenging task.

The main objectives of this chapter are to outline the nature of MEMS componets, from both the structural and physical points of view and identify the difficulties that these heterogeneous structures impose on modeling and simulation of these systems and to give a comprehensive account of different modern modeling methods and techniques used to overcome MEMS modeling and simulation difficulties, with illustrative examples. From here onwards, the word 'components' will be suppressed and refer to MEMS component just as MEMS for brevity.

The chapter starts with some basic definitions, then shed light on the use and advantages of mathematical modeling and simulation in MEMS design and manufacturing, investigates the nature of MEMS and the main challenges facing researchers in their modeling and simulation. It presents systematic approach to MEMS modeling and simulation, gives a survey coverage of different methods and techniques used for simulation of MEMS, such as finite-differences time-domain (FDTD), finite-element (FE) and their different variants proved to be successful in their applications to such microstructures. A detailed illustrative MEMS modeling and simulation example is given and other examples developed by other active research workers, are cited. The chapter concludes with future outlook and new trends in MEMS modeling and simulation.

#### **2. Background**

Over the last few years, extensive research work in modeling and simulation of MEMS took place, and a large number of techniques appeared in the literature to tackle this complicated modeling problem, each having its own advantages and limitations (Bushyager et al., 2003; Lynn Khine & Moorthi, 2006; Khine A. et. al., 2008; Yong Zhu & Espinosa, 2003; Shanmugavalli et. al. 2006; Heung-Shik Lee et. al., 2008; Fengyan and Vaughn, 2009; Chiao & Liwei, 2000).

Bushyager and his coworkers considered modeling electro-statically actuated RF passive components, such as parallel–plate capacitors and tuners (Bushyagers & Tenzeris, 2002; Bushyager et .al 2003; Bushyager, Mc Garvey & Tenzer 2000). They used finite-difference time-domain technique (FDTD) or its variant MRTD, with adaptive gridding to tackle the problem of large aspect ratio and moving boundary conditions (see section 4).

Fengyan and Vaughn (2009) developed an on-line tool for simulating micro scale electrothermal actuators (ETA), using both distributed and lumped analysis. The model allows the user to input the electrical Current of ETA to compute both temperature distribution and the displacement of the ETA. Finite–Element method was used to find temperature distribution due to joule heating and the average temperature across the beam is used to find the displacement by lumped analysis. The Model takes into account both properties' variation with temperature as well as radiation effects at high temperature. The main characteristics of their model is the use of computationally expensive distributed analysis for modeling electro thermal phenomenon and the computationally efficient lumped analysis for the modeling the thermo mechanical phenomenon thus improving the overall computational efficiency, accessibility and ease of use. The last two properties (accessibility and ease of use) are due to on-line availability of model for remote users.

Chiao & Liwei (2000) Considered self buckling behavior of micromechanical clampedclamed micro beams under resistive heating using both analytical and finite-difference technique. Their model consisted of electro-thermal part in which electric current flowing through the beam gives rise to Joule Heating effects, and thermo-mechanical part which deals with the mechanical buckling of the beam due to thermal expansion. Results was verified by measurements with good agreement. They considered the effected of residual stress and found that when compressive residual is considered in the analysis the critical current causing bulking of the beams, decreased where as tensile residual stress hinders the actuation of the beams. They also considered the effect of process variation such as the width and the thickness of the beam, and found that they affect the performance of the beams such as the current–deflection curves, however variation expansion and Young's modulus with temperature, were not considered and called for characterization of these properties' variation in order to improve the accuracy of MEMS Models.

Most, if not all, of these attempts relied on numerical (computational), rather than analytical techniques, due to ability of numerical methods to model structures with arbitrarily complex geometrical shapes. Despite the advantages of numerical techniques researchers were faced with a number of difficulties when trying to use commercial simulation packages and adapt traditional numerical methods, such as finite-difference, time-domain (FDTD) and finite-element (FE), used in modeling structures at macro-scale. The main difficulties and challenges are highlighted in section 4.

#### **3. Modeling & simulating definition & advantages**

#### **3.1 Definitions**

278 Micromachining Techniques for Fabrication of Micro and Nano Structures

heterogeneous structures impose on modeling and simulation of these systems and to give a comprehensive account of different modern modeling methods and techniques used to overcome MEMS modeling and simulation difficulties, with illustrative examples. From here onwards, the word 'components' will be suppressed and refer to MEMS component

The chapter starts with some basic definitions, then shed light on the use and advantages of mathematical modeling and simulation in MEMS design and manufacturing, investigates the nature of MEMS and the main challenges facing researchers in their modeling and simulation. It presents systematic approach to MEMS modeling and simulation, gives a survey coverage of different methods and techniques used for simulation of MEMS, such as finite-differences time-domain (FDTD), finite-element (FE) and their different variants proved to be successful in their applications to such microstructures. A detailed illustrative MEMS modeling and simulation example is given and other examples developed by other active research workers, are cited. The chapter concludes with future outlook and new

Over the last few years, extensive research work in modeling and simulation of MEMS took place, and a large number of techniques appeared in the literature to tackle this complicated modeling problem, each having its own advantages and limitations (Bushyager et al., 2003; Lynn Khine & Moorthi, 2006; Khine A. et. al., 2008; Yong Zhu & Espinosa, 2003; Shanmugavalli et. al. 2006; Heung-Shik Lee et. al., 2008; Fengyan and Vaughn, 2009; Chiao

Bushyager and his coworkers considered modeling electro-statically actuated RF passive components, such as parallel–plate capacitors and tuners (Bushyagers & Tenzeris, 2002; Bushyager et .al 2003; Bushyager, Mc Garvey & Tenzer 2000). They used finite-difference time-domain technique (FDTD) or its variant MRTD, with adaptive gridding to tackle the

Fengyan and Vaughn (2009) developed an on-line tool for simulating micro scale electrothermal actuators (ETA), using both distributed and lumped analysis. The model allows the user to input the electrical Current of ETA to compute both temperature distribution and the displacement of the ETA. Finite–Element method was used to find temperature distribution due to joule heating and the average temperature across the beam is used to find the displacement by lumped analysis. The Model takes into account both properties' variation with temperature as well as radiation effects at high temperature. The main characteristics of their model is the use of computationally expensive distributed analysis for modeling electro thermal phenomenon and the computationally efficient lumped analysis for the modeling the thermo mechanical phenomenon thus improving the overall computational efficiency, accessibility and ease of use. The last two properties (accessibility and ease of use)

Chiao & Liwei (2000) Considered self buckling behavior of micromechanical clampedclamed micro beams under resistive heating using both analytical and finite-difference technique. Their model consisted of electro-thermal part in which electric current flowing through the beam gives rise to Joule Heating effects, and thermo-mechanical part which deals with the mechanical buckling of the beam due to thermal expansion. Results was verified by measurements with good agreement. They considered the effected of residual

problem of large aspect ratio and moving boundary conditions (see section 4).

are due to on-line availability of model for remote users.

just as MEMS for brevity.

**2. Background** 

& Liwei, 2000).

trends in MEMS modeling and simulation.

Before we go into further details of this chapter definition of some of the main concepts is in order. In the following are given, definitions of the main keywords of this chapter; modeling and simulation.

**Modeling:** Modeling, and a model, has different meanings depending on the context or field of application. In engineering science and technology, Modeling refers to mathematical representation of a physical phenomenon, system or device. Usually mathematical models can take many forms such as dynamic system models, statistical models, game models, differential equation…etc. But for the purpose of this chapter, we will be mainly concerned with mathematical models that are represented by differential equations.

**Simulation:** Similar to modeling, simulation also has different meanings depending on the context and type of application and it can take many forms. However for the purpose of this chapter, we will concentrate on computer simulation or computational modeling, which is defined as: a computer program, or package, that attempts to simulate or imitate an abstract model of a particular system or device. Computer simulations have become very much related and integrated with mathematical modeling, and usually modeling and simulation are taken as being one discipline that can be used to explore and gain insight into the new technology and predict and estimate behavior of complex systems and devices that are too much complicated for analytical solutions.

Modeling and Simulation generally have iterative nature. A model is first developed then simulated to gain some understanding. The Model is then revised, and simulated again and this process goes on until an adequate level of understanding is developed for the system/device under consideration.

#### **3.2 Need for MEMS modeling and simulation**

There are many reasons why we need modeling and simulation for MEMS, among which are:

a. Due to small dimensions of MEMS, direct experimentation for determination of some physical properties of MEMS is difficult, and measurement errors occur when dealing with these micro-level systems.


#### **4. MEMS modeling challenges**

The difficulties and challenges that face MEMS modeling and simulation experts are due to the very nature of MEMS and can be related to the following MEMS features: Physical principle of operation, geometrical structure, miniaturization, packaging, manufacturing and processing of MEMS, and environmental conditions. These are detailed in the following:

a. Physical Principles:

As mentioned earlier, MEMS are characterized by interaction of many physics domains in a single device or system. This is pictorially represented in Fig. 2. Interaction of many physical phenomena in MEMS needs dealing with different types of equations, each governing a certain physical phenomenon e.g. electrical, mechanical or thermal. Moreover these governing equations are always coupled sometimes strongly and sometimes weakly, thus calling for solution of coupled system of equation a phenomenon called multi- physics approach. Depending on the number of physics domains involved, we can further classify these multidomain systems as: Double-Physics (interaction of two physics domain, Triple-Physics (interaction of three physics domains), Quadruple-Physics (interaction of four physics domains) and so on. In section 7.2 , we show specific modeling and simulation examples illustrating these different physics categories.

b. MEMS geometrical structure:

One of the main obstacles facing accurate modeling of micro-systems is proper definition (construction) of the system geometry, one of the main first requirements in any modeling and simulation process. This is due to different deformations and irregularities during MEMS operation, such as in the case of tuners and microbeams that contain moving parts. In order to tackle this problem, Peyrou David et.al, 2004, proposed a reverse engineering technique, whereby the model is first built using the real shape of the device, then a virtual model is made from the deformed shape, using different software packages. This method was applied for simulation of RF-MEMS capacitive switch and electrical contact resistance of RF MEMS, with satisfactory results (Peyrou David et.al, 2004).

c. Large Aspect Ratio:

Many MEMS, such as metallic sheets on top of large substrate (e.g. capacive stub tuners), have very small dimension in one direction (thickness) compared to relatively large dimension ( length) in other direction, thus leading to large aspect ratio. Large aspect ratio, in turn lead to creation of large number of computations cells thus leading to long computation time (Ali, I. Kabula M. & Hartnagel H. L., 2010).

Fig. 1. Multiphysics nature of MEMS

d. Geometry change due to moving parts:

Some MEMS components, such as electrostatic and thermal actuators, contain moving parts, which lead to moving boundary conditions.Commercially available traditional frequency domain simulators are not adequate to handle these features, due to many approximations that are more to simplify modeling and time-domain techniques, are more appropriate (Bushyagers & Tenzeris. 2002; Bushyager et. al., 2003). However even the well-known time– domain techniques such as FDTD, are not adequate as they stand and some sort of modification are needed to deal with large grid size and moving boundary conditions. These are solved by using adaptive gridding in both space and time, which are included in FDTD technique with variable gridding capability or the use of multi resolution time domain technique known as MRTD for short.

e. Miniaturization:

280 Micromachining Techniques for Fabrication of Micro and Nano Structures

b. Time reduction: Designers need simulation tools that allow them try "what If "experiments in hours instead of days and weeks, thus reducing time to market. c. Production cost reduction: Modeling and simulations are needed in order to study the behavior of the design under different experimental conditions and different level of

d. MEMS are usually embedded within other systems or packaged with other micromachined components and systems. Therefore modeling and simulation is needed at

e. Modeling and simulation can render fast design cycle that allows extensive scoping for

f. Modeling and simulation allows better understanding of the device/system operation

g. Modeling and simulation enable designers and system developers to see and further investigate systems behavior which could have not been discovered otherwise.

The difficulties and challenges that face MEMS modeling and simulation experts are due to the very nature of MEMS and can be related to the following MEMS features: Physical principle of operation, geometrical structure, miniaturization, packaging, manufacturing and processing of MEMS, and environmental conditions. These are detailed in the

As mentioned earlier, MEMS are characterized by interaction of many physics domains in a single device or system. This is pictorially represented in Fig. 2. Interaction of many physical phenomena in MEMS needs dealing with different types of equations, each governing a certain physical phenomenon e.g. electrical, mechanical or thermal. Moreover these governing equations are always coupled sometimes strongly and sometimes weakly, thus calling for solution of coupled system of equation a phenomenon called multi- physics approach. Depending on the number of physics domains involved, we can further classify these multidomain systems as: Double-Physics (interaction of two physics domain, Triple-Physics (interaction of three physics domains), Quadruple-Physics (interaction of four physics domains) and so on. In section 7.2 , we show specific modeling and simulation

One of the main obstacles facing accurate modeling of micro-systems is proper definition (construction) of the system geometry, one of the main first requirements in any modeling and simulation process. This is due to different deformations and irregularities during MEMS operation, such as in the case of tuners and microbeams that contain moving parts. In order to tackle this problem, Peyrou David et.al, 2004, proposed a reverse engineering technique, whereby the model is first built using the real shape of the device, then a virtual model is made from the deformed shape, using different software packages. This method was applied for simulation of RF-MEMS capacitive switch and electrical contact resistance

Many MEMS, such as metallic sheets on top of large substrate (e.g. capacive stub tuners), have very small dimension in one direction (thickness) compared to relatively large

parameters, before prototyping, thus reducing production cost.

macro or system level as well as micro-level.

and gives scope for optimization of its operation.

examples illustrating these different physics categories.

of RF MEMS, with satisfactory results (Peyrou David et.al, 2004).

b. MEMS geometrical structure:

c. Large Aspect Ratio:

more and accurate decision making.

**4. MEMS modeling challenges** 

following:

a. Physical Principles:

Due to the very small dimensions of MEMS many physical phenomena such as surface tension from humidity, become significant at micro-level. Downscaling of the dimensions of MEMS structures, which are usually three-dimensional makes the scale of these effects (force, displacement … etc.) also significantly small which, in turn lead to accumulation of numerical errors.

#### f. Packaging:

In many cases MEMS are assembled and packaged with other electronic components and devices, and this calls for proper modeling and simulation at systems level taking into consideration all input/output feedback etc.

g. Manufacturing and processing:

Manufacturing and process uncertainties which result from the fact that the dimensions and properties used in simulation of the MEMS cannot be exactly produced during fabrication.

h. Environmental Conditions:

MEMS are sometimes needed to work at different environmental conditions, such as sensors on the surface of the aircraft wings or sensors and actuators in nuclear power plants or under sea. Being tiny components at micro level, these systems are much affected by these external conditions. This calls for careful considerations of these conditions when modeling.

#### **5. MEMS modeling and simulation process, techniques and tools**

In this section, we show a systematic procedure for modeling and simulation process of MEMS, and give a survey of techniques and software tools proved to be successful in tackling many of the modeling and simulation challenges illustrated in the last section.

#### **5.1 Modeling and simulation process cycle**

The process of modeling and simulation goes through a defined number of steps, briefly described and depicted graphically in Fig.1 below:

Step 1: Description of the Physical Problem:

The first step in the modeling and simulation process is the physical description of the problem under investigation in order to understand its geometrical structure and the entire physical phenomenon involved in its operation.

Step 2: formulation of governing equations:

Depending on the physical principles governing operation of the device/system, all equations and mathematical expressions governing this operation are setup together with appropriate boundary and/or initial conditions.

Step 3: approximation of governing equations:

At this step, some approximations that facilitate solution with adequate effort and/or resources are made. If more than one equation is involved, the degree of coupling between different types of equations is determined at this step.

Steps 2 and 3 are usually very much interrelated, and the process of approximation and formation of governing equations are done iteratively in one step.

Step 4: Method(s) of solution:

Here all adequate methods of solutions of the governing equations are explored, and the most appropriate one identified. Appropriateness here calls for proper consideration of all available computation resources (both hardware and software), and selection of the most efficient one in terms of computational resources.

Step 5: Solution of the governing equation(s):

The method(s) identified in step 4 is applied to find the solution of the system of equations under specified boundary/initial conditions.

Step 6: Verification of Results:

After finding the general solution in step 5, some simplified standard cases which can be verified by experimental measurements or for which analytical or known solutions are

Manufacturing and process uncertainties which result from the fact that the dimensions and properties used in simulation of the MEMS cannot be exactly produced during fabrication.

MEMS are sometimes needed to work at different environmental conditions, such as sensors on the surface of the aircraft wings or sensors and actuators in nuclear power plants or under sea. Being tiny components at micro level, these systems are much affected by these external conditions. This calls for careful considerations of these conditions when modeling.

In this section, we show a systematic procedure for modeling and simulation process of MEMS, and give a survey of techniques and software tools proved to be successful in tackling many of the modeling and simulation challenges illustrated in the last section.

The process of modeling and simulation goes through a defined number of steps, briefly

The first step in the modeling and simulation process is the physical description of the problem under investigation in order to understand its geometrical structure and the entire

Depending on the physical principles governing operation of the device/system, all equations and mathematical expressions governing this operation are setup together with

At this step, some approximations that facilitate solution with adequate effort and/or resources are made. If more than one equation is involved, the degree of coupling between

Steps 2 and 3 are usually very much interrelated, and the process of approximation and

Here all adequate methods of solutions of the governing equations are explored, and the most appropriate one identified. Appropriateness here calls for proper consideration of all available computation resources (both hardware and software), and selection of the most

The method(s) identified in step 4 is applied to find the solution of the system of equations

After finding the general solution in step 5, some simplified standard cases which can be verified by experimental measurements or for which analytical or known solutions are

**5. MEMS modeling and simulation process, techniques and tools** 

g. Manufacturing and processing:

**5.1 Modeling and simulation process cycle** 

Step 1: Description of the Physical Problem:

described and depicted graphically in Fig.1 below:

physical phenomenon involved in its operation. Step 2: formulation of governing equations:

appropriate boundary and/or initial conditions. Step 3: approximation of governing equations:

Step 4: Method(s) of solution:

Step 6: Verification of Results:

different types of equations is determined at this step.

efficient one in terms of computational resources. Step 5: Solution of the governing equation(s):

under specified boundary/initial conditions.

formation of governing equations are done iteratively in one step.

h. Environmental Conditions:

Fig. 2. The main Steps of Modeling and Simulation process (details of each step are given in the main text)

identified for comparison reasons. If adequate solution that agree with measurements or known solution is obtained, move to the following step or otherwise go back to check method of solution or even the model itself. This checking process goes on until adequate solution is obtained (see decision block in the flowchart of Fig.2.

Step 7: Use model for exploration and design optimization:

This is the last step at which you make use of your efforts in modeling and simulation. Here the model is used for analysis of varied situation under different conditions i.e. change of input parameters and study of what-if conditions.

#### **5.2 MEMS modeling techniques**

There are a number of modeling techniques, with different facilities and capabilities, for modeling and simulation of MEMS at both component and system levels. However, we will concentrate on two mostly used techniques; the finite difference time domain (FDTD) and Finite Elements techniques and their variants.

#### **5.2.1 Finite – Difference – Time Domain (FDTD)**

FDTD technique is direct method of Solution of Maxwell's curt equation (Taflove, A. & Hagness, S., 2001):

$$\nabla \times \bar{\mathbf{E}} = \text{-} \mu \frac{\partial H}{\partial t} \tag{1}$$

$$
\nabla \times H = \sigma \dot{\mathbf{E}} + \epsilon \frac{\partial \mathbf{E}}{\partial t} \tag{2}
$$

where **E** and **H** are the vector electric and magnetic fields, ߤ, ߳ and ߪare the respectively, permeability, permittivity and conductivity of the medium.

FDTD method is a full-wave time stepping procedure that uses simple central—Difference approximation to evaluate the space and time derivatives. FDTD method has been successfully used for simulation of complex structure at macro scale, however one of the most constraints of FDTD is the requirement that cell size be equal to or smaller than the smallest feature of the device or system being simulated. This leads to large grid size when simulating structures consisting of several different elements or structure with large aspect ratios, such as MEMS circuits. The large grid size leads, in turn to large computing time.

In order to overcome this difficulty, a number of modification and /or addition have been proposed, among which are the method of sub gridding, and the use of non-uniform grid (Shanmugavalli et. al., 2006; Bushyager et. al., 2001). The first, subgridding method subdivides sections of the FDTD grid, thus creating finer grids in areas of large field variation or interfaces between areas of different properties. This method can be used when small complex devices are embedded in a coarse grid, such as small passive compounds ( compactors' or inductors) connected by micro strip lines or when MEMS are packaged. Here a subroutine is needed to interface grids of different sizes. In the non uniform grid technique, each direction of the structure under investigation is treated separately, thus allowing each axis to be divided into a number of sections each having a different cell size.

#### **5.2.2 Multiresolution Time Domain (MRTD) technique**

MRTD employs multiresolution principles to discretize Maxwell's equations in wavelet expansions, which provide a set of functions with adaptive resolution. Higher resolution functions can be added and subtracted during simulation. Accordingly MRTD has built-in gridding capability in both time and time and space


When MRTD is applied to Maxwell's Curl equations (1 and 2 above), the orthogonality of the wavelets provides an efficient discretization that results in an explicit time marching scheme similar to that of FDTD technique. Due to this property and the characteristics of MRTD stated in 1-3 above, MRTD has been successfully applied by Bushyager and his coworkers to a number of modeling and simulation of RF MEMS structures, such as parallelplate capacitor (Bushyager & Tentzeris 2001), MEMS tunable capacitors (Bushyager et. al. 2002), analysis of MEMS embedded components in multilayer packages (Tentzeria et. al. 2003), for intracell modeling of metal/dielectric interfaces (Bushyager & Tentzeris, 2003), for static MEMS capacitor (Bushyager, McGarvey & Tentzeris, 2001).

#### **5.2.3 Finite Element Technique**

284 Micromachining Techniques for Fabrication of Micro and Nano Structures

There are a number of modeling techniques, with different facilities and capabilities, for modeling and simulation of MEMS at both component and system levels. However, we will concentrate on two mostly used techniques; the finite difference time domain (FDTD) and

FDTD technique is direct method of Solution of Maxwell's curt equation (Taflove, A. &

డ

డ

డ௧ (1)

డ௧ (2)

ߤ- =**Ē** Xߘ

߳6σ= Xߘ

where **E** and **H** are the vector electric and magnetic fields, ߤ, ߳ and ߪare the respectively,

FDTD method is a full-wave time stepping procedure that uses simple central—Difference approximation to evaluate the space and time derivatives. FDTD method has been successfully used for simulation of complex structure at macro scale, however one of the most constraints of FDTD is the requirement that cell size be equal to or smaller than the smallest feature of the device or system being simulated. This leads to large grid size when simulating structures consisting of several different elements or structure with large aspect ratios, such as MEMS circuits. The large grid size leads, in turn to large computing time. In order to overcome this difficulty, a number of modification and /or addition have been proposed, among which are the method of sub gridding, and the use of non-uniform grid (Shanmugavalli et. al., 2006; Bushyager et. al., 2001). The first, subgridding method subdivides sections of the FDTD grid, thus creating finer grids in areas of large field variation or interfaces between areas of different properties. This method can be used when small complex devices are embedded in a coarse grid, such as small passive compounds ( compactors' or inductors) connected by micro strip lines or when MEMS are packaged. Here a subroutine is needed to interface grids of different sizes. In the non uniform grid technique, each direction of the structure under investigation is treated separately, thus allowing each axis to be divided into a number of sections each having a different cell size.

MRTD employs multiresolution principles to discretize Maxwell's equations in wavelet expansions, which provide a set of functions with adaptive resolution. Higher resolution functions can be added and subtracted during simulation. Accordingly MRTD has built-in

1. MRTD can model changes in structure over time because it is a time-domain simulator

2. Variable grid capability and ability to arbitrarily place metals in structures allow for

3. Adaptive (variable) gridding allows modeling of MEMS in packages such as variable capacitor connected to large feed lines .In this case the surrounding structure can be modeled with a comparatively coarser grid, while the MEMS is modeled with a fine

modeling of a structure that changes configuration during simulation.

**5.2 MEMS modeling techniques** 

Hagness, S., 2001):

Finite Elements techniques and their variants.

**5.2.1 Finite – Difference – Time Domain (FDTD)** 

permeability, permittivity and conductivity of the medium.

**5.2.2 Multiresolution Time Domain (MRTD) technique** 

gridding capability in both time and time and space

grid (higher resolution ).

for example: variable parallel-plate capacitor.

The **finite element method (FEM)** is a numerical technique for finding approximate solutions of partial differential equations (PDE) and integral equations. The solution approach is based either on eliminating the differential equation completely (steady state problems), or rendering the PDE into an approximating system of ordinary differential equations, which are then numerically integrated using standard techniques such as Euler's method, Runge-Kutta, etc. Historically, finite element method originated from the need for solving complex material and structural problems by civil and mechanical engineers.

The first step in finite-difference analysis is to divide the structure under consideration into small homogeneous pieces or elements with corners called nodes. The nodes define the boundaries of each element and the entire collection of elements defines the analysis region or the geometry of the structure. These elements can be small where detailed geometric properties are required and much larger elsewhere. A simple variation of the field is assumed in each element and the main goal of FEA is to determine field quantities at each node. In general, FEA techniques solve for the unknown field quantities by minimizing an energy functional using variational techniques. The energy functional is an expression describing all the energy associated with the structure under consideration. For instance, this functional for 3-dimentional time-harmonic electromagnetic fields E and H is represented as:

$$F = \int\_{\upsilon} \frac{\mu |H|^2}{2} + \frac{\epsilon |E|^2}{2} + \frac{JE}{2j\omega} d\upsilon \tag{3}$$

Where the first two terms represent energy stored in the magnetic and electric fields and the third is the energy dissipated or supplied by the conduction currents.

Eliminating H using relation between E and H, and setting derivatives of this functional with respect to E to zero (for minimization), an equation of the form f(J,E)=0 is found. Appling Kth order approximation of F at each of the N nodes and enforcing boundary conditions results in a system of equations of the form:

$$
\begin{bmatrix} I\_1 \\ \vdots \\ I\_n \end{bmatrix} = \begin{bmatrix} \mathcal{V}\_{11} & \cdots & \mathcal{V}\_{1n} \\ \vdots & \ddots & \vdots \\ \mathcal{Y}\_{n1} & \cdots & \mathcal{Y}\_{nn} \end{bmatrix} \begin{bmatrix} E\_1 \\ \vdots \\ E\_n \end{bmatrix} \tag{4}
$$

Here Ji and yij ( i=1,2,..n and j=1,2,..,n) are known quantities and the electric fields E at each node are the unknowns to be determined. This needs inversion of the y matrix and the larger the number of nodes, the larger the size of the matrix and hence more computational requirements. For example a mesh of a 3-D cube with 1000 nodes (10 nodes per side) and 3 degree of freedom creates a 3000 by 3000 matrix. Doubling the number of nodes per side gives a system of 24000 by 24000 matrix. Therefore it is necessary to as fewer as possible number of nodes is preferable from the point of view of computer resources. On the other hand the more the number of nodes the more is the result accuracy. Usually a compromise is made between required accuracy and computational resources. For proper grasp of this useful technique, the reader is referred to any of the many text books on finite element analysis (see for instance Silvester & Ferrari (1990)).

#### **5.3 MEMS simulation software tools**

During the last decade, and specially in the period 2002-2010, there were a lot of research activities and developmental in the area of modeling and simulation tools dedicated to MEMS, and a number of such software simulation tools moved from research centers to the commercial world. These tools can be classified broadly as: *system level modeling and simulation tools* and *detailed device–level tools known as field solvers*. Technique used in type of tools, have been in use at macro level for some time and have gained maturity. the micro level , however . There tools are still going some modification due to the nature of MEMS, which comprises many physical domains such as; mechanical, thermal, electrical and optical domains. In the following sections, we give a brief overview of these techniques and tools.

#### **5.3.1 System level tools**

These tools predict the main behavior of the MEMS, using a set of ordinary differential equations and non linear functions at block diagram level in terms of input and output of each block. These tools are usually required by MEMS Designers for accurate prediction of the overall performance of these systems when connected or operated with other circuit components and for proper optimization (Lynn Khine & Moorthi P., Hine A. et. al. 2008). In addition they enable the designer predict the performance of the device under consideration for a set of defined input parameters, it allows him/her to study the effect of variation of one or a certain set of parameters relative to other parameters on the overall performance of the device, thus leading to optimized design.

#### **5.3.2 Field solvers**

The majority of these solvers are based on the well-known numerical techniques of finite elements (FE) or finite–difference time–domain, methods briefly out lined in the lost section, with only few tools based on analytical or asymptotic solution of the governing equation. These methods are virtual and can analyze complex geometric shapes.

#### **Finite-Elements–based tools**

The majority of MEMS software tools now available in the market are based on the well – established FE technique. These include ANSYS (COMSOL), ConvectorWare, Intellisuite, ABAQUS, and DELEN, to mention only some example. In the following we give a brief account of their mostly used tools in the field of MEMS analysis and design.

#### a. ConventorWare

This is a fully integrated MEMS Design environment, the latest version of which is convectorWare2010. It runs on PCS and work Stations based on sun Solaris and windows This package consists of four main parts; Designer, Architect, Analyzer and Integrator. Designer is concerned with design specification and modeling of the MEMS Structure including two-dimensional layout, process emulation and finite elements meshing. It includes a full- featured layout editor, foundry access kits , 2D and 3 D import/export capabilities, mesh generator and a library of 3D models of standard MEMS packages. It is tightly integrated with the other two modules ARCHITECT and ANALYZER, but it can also be used as stand- alone package to be integrated with other third-party FEA tools.


#### b. IntelliSuite:

286 Micromachining Techniques for Fabrication of Micro and Nano Structures

gives a system of 24000 by 24000 matrix. Therefore it is necessary to as fewer as possible number of nodes is preferable from the point of view of computer resources. On the other hand the more the number of nodes the more is the result accuracy. Usually a compromise is made between required accuracy and computational resources. For proper grasp of this useful technique, the reader is referred to any of the many text books on finite element

During the last decade, and specially in the period 2002-2010, there were a lot of research activities and developmental in the area of modeling and simulation tools dedicated to MEMS, and a number of such software simulation tools moved from research centers to the commercial world. These tools can be classified broadly as: *system level modeling and simulation tools* and *detailed device–level tools known as field solvers*. Technique used in type of tools, have been in use at macro level for some time and have gained maturity. the micro level , however . There tools are still going some modification due to the nature of MEMS, which comprises many physical domains such as; mechanical, thermal, electrical and optical domains. In the following sections, we give a brief overview of these techniques and tools.

These tools predict the main behavior of the MEMS, using a set of ordinary differential equations and non linear functions at block diagram level in terms of input and output of each block. These tools are usually required by MEMS Designers for accurate prediction of the overall performance of these systems when connected or operated with other circuit components and for proper optimization (Lynn Khine & Moorthi P., Hine A. et. al. 2008). In addition they enable the designer predict the performance of the device under consideration for a set of defined input parameters, it allows him/her to study the effect of variation of one or a certain set of parameters relative to other parameters on the overall performance of

The majority of these solvers are based on the well-known numerical techniques of finite elements (FE) or finite–difference time–domain, methods briefly out lined in the lost section, with only few tools based on analytical or asymptotic solution of the governing equation.

The majority of MEMS software tools now available in the market are based on the well – established FE technique. These include ANSYS (COMSOL), ConvectorWare, Intellisuite, ABAQUS, and DELEN, to mention only some example. In the following we give a brief

This is a fully integrated MEMS Design environment, the latest version of which is convectorWare2010. It runs on PCS and work Stations based on sun Solaris and windows This package consists of four main parts; Designer, Architect, Analyzer and Integrator. Designer is concerned with design specification and modeling of the MEMS Structure including two-dimensional layout, process emulation and finite elements meshing. It

These methods are virtual and can analyze complex geometric shapes.

account of their mostly used tools in the field of MEMS analysis and design.

analysis (see for instance Silvester & Ferrari (1990)).

**5.3 MEMS simulation software tools** 

**5.3.1 System level tools** 

**5.3.2 Field solvers** 

a. ConventorWare

**Finite-Elements–based tools** 

the device, thus leading to optimized design.

This is an integrated design environment specifically developed to link the entire MEMS organization together. It enables designers to manage their MEMS product throughout it life cycle. Intellisuite consists of a number of advanced tools that work together, each covering a certain MEMS development stage. The main components of Intellisuite are:


COMSOL Multiphysics is a simulation software specifically designed to tackle multi physics problem, usually encountered all disciplines in Engineering Design starting from the definition of geometry, through meshing and material property definition and ending with solving and visualization of results. It is based on solution of the partial differential equations that given different physical phenomenon be it electromagnetic, heat transfer, structural mechanics, magnetostatics, simultaneously or sequentially solved using finite element analysis.

By its very nature of structure it is particularly appropriate module for MEMS, modeling and simulation, and in addition , it has a specialized module for MEMS, called MEMS Module that simulates standard MEMS problems, such as piezoelectric and thin film damping problems.

#### **Finite difference time domain tools**

a. XFDTD Package:

This is based on finite difference time domain technique (FDTD), for the solution of vector Maxwell's equations developed by REMCOM Inc. It is an easy to use package, but it does not have the capability of dealing with multidomain (multiphysics) problems.

#### **5.3.3 Other tools**

In addition to the above mentioned main MEMS Simulation software tools, there are other tools most important of which is the hybrid approach in which both system-level and field solvers in circuit simulation environment. They rely on a new class of parameterized behavioral method that can automatically access relational databases that contain field software solution results. These models have been successfully used for analyzing packaging effects of inertial sensors and design of PZR pressure sensors (Lorenz G., Greiner, K., & Breit S. 2006).

#### **6. Selected MEMS-based Modelling Examples**

Modeling and simulation can not only be acquired by reading and studying different methods and techniques, but through actual involvement in building models, simulating and practicing. This is similar to driving a car. Car driving can be learned through reading instructors notes, books and articles about car driving, but can only mastered by actually driving a car.

In this section, we illustrate the process of modeling and simulation described in section 2.2, by going through an actual case carried out by the author and his coworkers, and highlight some few successful cases done by other research workers.

#### **6.1 Detailed illustrative example on modeling and simulation process**

#### **Micromachined DC and high frequency power sensors**

In order to show how the modeling and simultion process described in section 3.3 is implemented in practice, modeling and simulation of micromachined power sensors at both the dc and high frequencies (microwave and millimeter waves) is presented. The sensor fabrication techniques is based on bulk and surface micromachining of AlGaAs/GaAs heterostructures, and the sensor concept is based on transduction principle that converts the RF into thermal power, which is then measured by thermoelectric means (Mutamba et al., 2001; Ali, I. Kabula M. & Hartnagel H., 2010)).

Modeling and simulation is based on the solution of the coupled electromagnetic field and heat transfer equations, using hybrid finite-element/finite-difference techniques. The model can be used as an effective tool for adjustment of design parameters, such as geometric configuration of the sensor resistive element, and arrangement of thermocouples around the resistor for optimum sensitivity and noise figure.

Fig. 3. The structure of the power sensor. left-hand : top view, right-hand-side sectional view along A-A'.

Step 1: Description of the Physical Problem:

A sectional view of the configuration of the sensor to be modeled is depicted in Fig. 3. The sensor is composed of a thermally isolated thin (2 μm) AsGaAs/GaAs membrane region with a terminating resistor, in which heat generated by DC or RF power is dissipated and converted into heat. A high-thermally resistive membrane region is obtained by selective etching of the GaAs against AlGaAs. This helps to increase temperature gradient between the resistor region and the rest of the chip, thus leading to high sensitivity. Temperature increase in the resistor region is sensed by a set of Gs/Au-Cr thermoelements, whose dc output is proportional to the input RF power. Detailed description of technical realization of this sensor is given by Mutamba et al., (2001).

Steps 2,3: The Governing Equations and Approximations:

Here we consider both the thermal and the electric models. The mathematical model is required to be simple enough to be handled by known methods which demand reasonable time and cost, and at the same time give an adequate description of the physical problem under investigation.

#### **Thermal model**

288 Micromachining Techniques for Fabrication of Micro and Nano Structures

By its very nature of structure it is particularly appropriate module for MEMS, modeling and simulation, and in addition , it has a specialized module for MEMS, called MEMS Module that simulates standard MEMS problems, such as piezoelectric and thin film

This is based on finite difference time domain technique (FDTD), for the solution of vector Maxwell's equations developed by REMCOM Inc. It is an easy to use package, but it does

In addition to the above mentioned main MEMS Simulation software tools, there are other tools most important of which is the hybrid approach in which both system-level and field solvers in circuit simulation environment. They rely on a new class of parameterized behavioral method that can automatically access relational databases that contain field software solution results. These models have been successfully used for analyzing packaging effects of inertial sensors and design of PZR pressure sensors (Lorenz G., Greiner,

Modeling and simulation can not only be acquired by reading and studying different methods and techniques, but through actual involvement in building models, simulating and practicing. This is similar to driving a car. Car driving can be learned through reading instructors notes, books and articles about car driving, but can only mastered by actually

In this section, we illustrate the process of modeling and simulation described in section 2.2, by going through an actual case carried out by the author and his coworkers, and highlight

In order to show how the modeling and simultion process described in section 3.3 is implemented in practice, modeling and simulation of micromachined power sensors at both the dc and high frequencies (microwave and millimeter waves) is presented. The sensor fabrication techniques is based on bulk and surface micromachining of AlGaAs/GaAs heterostructures, and the sensor concept is based on transduction principle that converts the RF into thermal power, which is then measured by thermoelectric means (Mutamba et al.,

Modeling and simulation is based on the solution of the coupled electromagnetic field and heat transfer equations, using hybrid finite-element/finite-difference techniques. The model can be used as an effective tool for adjustment of design parameters, such as geometric configuration of the sensor resistive element, and arrangement of thermocouples around the

not have the capability of dealing with multidomain (multiphysics) problems.

damping problems.

a. XFDTD Package:

**5.3.3 Other tools** 

K., & Breit S. 2006).

driving a car.

**Finite difference time domain tools** 

**6. Selected MEMS-based Modelling Examples** 

some few successful cases done by other research workers.

**Micromachined DC and high frequency power sensors** 

2001; Ali, I. Kabula M. & Hartnagel H., 2010)).

resistor for optimum sensitivity and noise figure.

**6.1 Detailed illustrative example on modeling and simulation process**

A simplified pictorial view of the sensor is shown in Fig. 3 for the purpose of thermal modeling. Here the thermocouple wires are not shown, and heat generated by the NiCr resistor, is assumed to be distributed throughout the sensor structure mainly by conduction. Due to axial symmetry of the sensor structure about it longitudinal axis, we only consider half of its geometry as shown in Fig.4. We select Cartesian coordinates (x,y,z) with its origin at the sensor input and the direction of power transmission along the positive z-direction.

In order to obtain a simple and manageable model, the following simplifying assumptions, are made:

i. The presence of the thermocouples is ignored as the metallic part (gold) is very small compared to other dimensions.

Fig. 4. Simplified half-section used for modeling.


The equation governing heat flow as a result of microwave or dc heating is the well-known heat conduction equation, also known as Fourier equation:

$$\frac{\partial T}{\partial t} = \alpha \nabla^2 T + Q \;/\; \rho \mathbf{C} \tag{5}$$

where T is the temperature, α is thermal diffusivity , Q heat generation term, ρ is the density and C is the specific heat of the medium .

#### **Thermal boundary conditions**

These conditions can be obtained from the prevailing thermal conditions at the boundaries and the interface between different material layers of the sensor.


iv. Assume initially that the sensor is at a constant temperature To.

The heat source term, Q, in equation (5), is determined by solving either of two different sets of equations, depending on the operation mode of the sensor (dc or high frequency) as shown below.

#### **Electric model**

Here we consider the case of AC and DC case separately, because of their different governing equations.

#### DC Operation Mode:

In the case of the dc operation mode, the electrostatic potential equation is used:

$$-\nabla(\cdot \sigma \nabla V) = q\_i \tag{7}$$

where V is the electrical potential, σ is the electrical conductivity, and qi is the current source. In the case of constant electrical conductivity, equation (7) reduces to the simple Poisson's equation:

$$
\nabla^2 V = q\_{\rm i} \,/\,\sigma \tag{8}
$$

Boundary conditions assumed for solution of equation (8) were that a constant voltage +V applied at one arm and –V at the other arm of the CPS With the symmetry condition along the longitudinal axis of the sensor, +V was assumed at one arm at input of the sensor and zero voltage at the end of the resistor (see Fig. 3). Having found V from equation (8), the heat generation term, Q, is obtained as:

$$Q = \sigma \left| E^2 \right| \tag{9}$$

where E is the electric field intensity, given by:

$$
\bar{E} = \nabla V \tag{10}
$$

High Frequency Mode:

290 Micromachining Techniques for Fabrication of Micro and Nano Structures

ii. As a first approximation, heat distribution is assumed to be two-dimensional (in Y-Z plane). This assumption can be justified due to the presence of the thermally isolated

iv. Radiation losses are ignored as the sensor were to work in a very confined region and

The equation governing heat flow as a result of microwave or dc heating is the well-known

<sup>2</sup> / *<sup>T</sup> TQ C <sup>t</sup>* 

where T is the temperature, α is thermal diffusivity , Q heat generation term, ρ is the density

These conditions can be obtained from the prevailing thermal conditions at the boundaries

iii. At the interface between different sensor material layers: assuming perfect thermal contact leads to the continuity of heat flux and temperature at these interfaces between

The heat source term, Q, in equation (5), is determined by solving either of two different sets of equations, depending on the operation mode of the sensor (dc or high frequency) as

Here we consider the case of AC and DC case separately, because of their different

In the case of the dc operation mode, the electrostatic potential equation is used:

(5)

thin membrane whose area dominates the horizontal dimensions of the sensor. iii. Constant thermal properties. Although thermal properties, especially thermal conductivity of GaAs vary with temperature, these properties were assumed to be

Fig. 4. Simplified half-section used for modeling.

the expected temperature rise is limited.

and C is the specific heat of the medium .

**Thermal boundary conditions** 

layers.

shown below. **Electric model** 

governing equations. DC Operation Mode:

constant due to the expected moderate temperature rise.

and the interface between different material layers of the sensor. i. Convective heat transfer at boundaries y=0 and y=a. ii. Specified temperature at boundaries z=0 and z=L.

iv. Assume initially that the sensor is at a constant temperature To.

heat conduction equation, also known as Fourier equation:

In the case of high frequency operation mode, the heat generation term, Q, is obtained from the solution of Vector Maxwell's equations:

$$
\bar{\nabla} \mathbf{X} \bar{\mathbf{E}} = -\mu \frac{\partial \bar{\mathbf{H}}}{\partial t} \tag{11}
$$

$$
\bar{\nabla} \mathbf{X} \cdot \bar{\mathbf{H}} = \sigma \bar{\mathbf{E}} + \varepsilon \frac{\partial \bar{\mathbf{E}}}{\partial t} \tag{12}
$$

in which *E* and *H* are electric and the magnetic field vectors, and , , are respectively the permittivity, the permeability and the conductivity of the material (medium) through which electromagnetic wave propagation takes place.

Steps 4 & 5: Solution of Governing Equations:

If the simplifying assumption (i) – (iv) are used, equation (1) with boundary conditions ii), iii) and iv) can be solved analytically using the method of separation of variables. However, in the more general case of three-dimensional form of equation (1) with boundary conditions (i) to (iv), the versatile numerical methods of finite difference in time-domain or finite elements are more appropriate, since they can deal with any sensor structure. Investigation of available software packages revealed that XFDTD package, based on finite difference time domain technique, is most appropriate for the solution of vector Maxwell's equations and FEMLAB, a package based on FEM for solution of the heat equation. As the two packages (XFDTD and FEMLAB) are based on different solution techniques using different simulation tools, it was necessary to have an interface that links the two packages. This was achieved by a special script file written using MATLAB built-in functions. Material properties used in the simulation are shown in table 1.


Table 1. Electrical and thermal properties used in the simulation.

Step 6: Verification of Solution:

Due to the small dimensions of the sensor structure, it was difficult to determine temperature distribution by direct temperature measurements. Therefore, a technique based on thermal imaging was used. The top surface of the test structures were coated with a thin film of liquid crystal (R35CW 0.7 from Hallcrest Inc, UK), that changes color with the changes in the sensor surface temperature. This change in temperature was monitored by a CCD camera connected to a microscope and a personal computer was used to store the recorded shots for later analysis. For dc operation mode, a stable current source was used, and both the input voltage and current were monitored for accurate determination of input power. For RF mode of operation, the current source was replaced by an RF probe that connected directly to the input pads of the CPS (see Fig.3).

In order to show how closely the simulated results resemble the actually expected temperature distribution, we compare simulated current density distribution in Fig.5 with the an experimental shots of the sensor while it was burning shown in Figs 6,7; one with thermocouple (Fig. 6) and the other without thermocouples (Fig. 7). The experimental results were obtained by increasing the input power level at small increments until the resistor was destroyed at an input power level of about 80 mW. The accumulation of current density around the inner corner of the resistor in the simulated result (Fig. 5), explains the destruction of the resistor at one of its sharp inner corners. furthermore experimental results show that the degree of destruction is more severe ( as illustrated by the size of the elliptical shape surrounding the resister) when the thermocouples are removed (Fig. 7). This can be attributed to spreading of heat away from the resistive termination by thermocouples when they present, thus decreasing level of destruction.

Fig. 5. Current density (J) at the inner corner of the resistive element

Fig. 6. Measured temperature distribution on the top surface of the sensor structure including thermocouple s

Fig. 7. Measured temperature distribution on the top surface of the sensor structure without thermocouples

Step 7: Using the Model

#### **2-D simulation**

292 Micromachining Techniques for Fabrication of Micro and Nano Structures

*GaAs 44 334 5360 12.8 5x10-4 NiCr 22 450 8300 - 9.1x105 Au 315 130 1928 - 4.5x107 Al/GaAs 23.7 445 3968 - 5x10-4* 

Due to the small dimensions of the sensor structure, it was difficult to determine temperature distribution by direct temperature measurements. Therefore, a technique based on thermal imaging was used. The top surface of the test structures were coated with a thin film of liquid crystal (R35CW 0.7 from Hallcrest Inc, UK), that changes color with the changes in the sensor surface temperature. This change in temperature was monitored by a CCD camera connected to a microscope and a personal computer was used to store the recorded shots for later analysis. For dc operation mode, a stable current source was used, and both the input voltage and current were monitored for accurate determination of input power. For RF mode of operation, the current source was replaced by an RF probe that

In order to show how closely the simulated results resemble the actually expected temperature distribution, we compare simulated current density distribution in Fig.5 with the an experimental shots of the sensor while it was burning shown in Figs 6,7; one with thermocouple (Fig. 6) and the other without thermocouples (Fig. 7). The experimental results were obtained by increasing the input power level at small increments until the resistor was destroyed at an input power level of about 80 mW. The accumulation of current density around the inner corner of the resistor in the simulated result (Fig. 5), explains the destruction of the resistor at one of its sharp inner corners. furthermore experimental results show that the degree of destruction is more severe ( as illustrated by the size of the elliptical shape surrounding the resister) when the thermocouples are removed (Fig. 7). This can be attributed to spreading of heat away from the resistive termination by thermocouples when

*Density (kg/m3)* 

*Relative permittivity* 

*Electrical conductivity (mho)* 

*Specific heat (J/kg K)* 

Table 1. Electrical and thermal properties used in the simulation.

connected directly to the input pads of the CPS (see Fig.3).

they present, thus decreasing level of destruction.

Fig. 5. Current density (J) at the inner corner of the resistive element

*Material* 

*Thermal conductivity (W/m K)* 

Step 6: Verification of Solution:

Results obtained from 2-D simulation are shown in figure 8 through 11. Fig. 8 shows a 3D color plot of the normalized temperature distribution on the top surface of the sensor. It is shown here how temperature is concentrated in the Ni.Cr. resistor region with sharp decrease with distances from the resistor edges. In order to have a more quantitative picture, an enlarged view of the temperature contour around the resistor region is shown in Fig. 9. It can be seen that temperature decreases to about 67% of the peak value (at the resistor) at a distance of about 20 μm from the side arm of the resistor (y-direction). Further away from the resistor, temperature level reaches about only 33% of its peak value at a distance of about 180 μm. Along the x-axis away from the short arm of the resistor, the drop in temperature level is even more sharp reaching 67% at 10 μm, and 33% at 105 μm.

Fig. 8. Color plot of relative temperature distribution around the NiCr. Resistor.

Fig. 9. Enlarged view of temperature distribution around the corner of the resistive termination

#### **3-D simulation**

The three-dimensional form of equations (5-8) were solved with the assumption of negligibly small resistor thickness. Fig. 10 shows a color plot of the temperature distribution on the top surface of a thin resistor on bulk substrate 150 μm thick. This figure compares

Fig. 8. Color plot of relative temperature distribution around the NiCr. Resistor.

Fig. 9. Enlarged view of temperature distribution around the corner of the resistive

The three-dimensional form of equations (5-8) were solved with the assumption of negligibly small resistor thickness. Fig. 10 shows a color plot of the temperature distribution on the top surface of a thin resistor on bulk substrate 150 μm thick. This figure compares

termination

**3-D simulation** 

very well with the experimentally obtained result of Fig. 9. 6,7, and clearly illustrates the elliptic form of surface temperature distribution. To see the effect of the third, z-dimension, temperature distribution on a plane cut along the line y-y on Fig. 10 is plotted in Fig.11. This figure illustrates the diffusion of heat through the GaAs substrate with peak values of temperature directly under the two long arms of the resistor. Thus it shows the effect of the bulk substrate that leads to the spreading of temperature away from the resistor region and down into substrate. This reinforces the idea behind using thin membrane technology, in which case the bulk substrate is removed, in the construction of thermoelectric power sensors.

Fig. 10. Pictorial color plots of temperature distribution on the surface of the sensor (initial temperature 293K)

Fig. 11. Simulated temperature distribution on the top surface of the sensor structure along x-x

#### **6.2 Other examples**

Table 2a,b show respectively selected double and triple physics illustrative examples showing the type of MEMS, physical phenomena on which their operations are based, type of equations involved in their model and the technique and/or software tools used for their simulation. Information given in these tables are only highlights on the main principles and the interested reader is advised to consult cited references for more details.


Table *2.*a. Double-Physics Problems


Table 2.b. Triple-Physics Problems

296 Micromachining Techniques for Fabrication of Micro and Nano Structures

Table 2a,b show respectively selected double and triple physics illustrative examples showing the type of MEMS, physical phenomena on which their operations are based, type of equations involved in their model and the technique and/or software tools used for their simulation. Information given in these tables are only highlights on the main principles and

*Equations* 

*Electric current flow + heat equation* 

*Maxwell's + heat equation* 

*Maxwell's + heat equation* 

*Maxwell's + Transport equation* 

*Maxwell's + equation of motion* 

*Maxwell's + equation of motion* 

*Stress tensor + magnetic field equation* 

*Stress tensor + magnetic field equation* 

*Simulation Software tool/ technique* 

> *XFDTD+ FEMLAB ( FDTD+ FEM method)*

FDTD method

COMSOL Multiphysics (F.E.M)

Shell–Element Method

*IntelliSuite TM Ijaz et. al.* 

ן െܨ݈ݓ Ji-Song, et. al.,

*FDTD method* Bushyager et.

*FDTD method* Bushyager et.

*Reference* 

*(2005)* 

*Ali, I.et. al. (2010)* 

(1999)

al. (2002)

al. (2001)

Yamagata, Michiko Kuroda & Manos M. Tentzeris (2005)

Bohdana Marvalova (2008)

Heung-Shik Lee et. al. (2008)

> *Riccardo & Roberto (2008)*

the interested reader is advised to consult cited references for more details.

*Phenomenon MEMS type Type of* 

*Thermal convertors for gas sensors* 

*Microwave power sensors* 

*Fingerprint sensors*

*Parallel-plate capacitors* 

*Stub tuners* 

*Antennas with moving parts* 

*Magneto- sensitive Elastometers* 

*Magnetostrictive thin-film actuators*

*Ring laser and fiber optic gyroscopes* 

**6.2 Other examples** 

*Physical* 

*Electromagnetism and Thermal* 

*Electromagnetism and Mechanics* 

*Magnetostaticts And Mechanics* 

> *Optics And Mechanics*

Table *2.*a. Double-Physics Problems

#### **7. Concluding remarks and future outlook**

In this chapter we carefully looked at all MEMS features and nature which make their modeling and simulation a challenging task have been identified and summarized in the following:


Different types of simulation techniques, as well as software tools based on these techniques, have been considered with advantages and limitations of each type. A detailed case study that illustrate proper modeling and simulation steps was made and some other successful modeling and simulation examples have been highlighted with proper reverence to their sources for interested reader.

The field of MEMS is very promising and much work is needed in the following areas:


#### **8. References**

Ali, I. Kabula M. & Hartnagel H. L.(2010). Electro-thermal Simulation of Micromachined Power Sensors at DC and Microwave Frequencies. Presented at The Seventh International Symposium on Machatronics and its Applications SIAM'10. American University of Sharjah, Sharjah, UAE, April 20-22-2010.

Available from: http://academic.research.microsoft.com/publication/4577492

Bohdana Marvalova (2008). Modeling of Magnetoresistive Elastometers. Modelling and Simulation, Giuseppe Petrone and Giuliano Cammarata (Ed.), ISBN: 978-3-902613- 25-7, InTech, Available from:


http://academic.research.microsoft.com/publication/4571811

Bushyager, N. Manos M.Tentzeris, Larry Gatewood, Jeff DeNatale, (2003). A Novel Adaptive Approach to Modeling MENS Tunable Capacitors Using MRTD and FDTD Techniques. Available from:

http://academic.research.microsoft.com/publication/4569517


http://www.intechopen.com/articles/show/title/multiphysics\_modelling\_and\_si mulation\_in\_engineering

Available from:

298 Micromachining Techniques for Fabrication of Micro and Nano Structures

Ali, I. Kabula M. & Hartnagel H. L.(2010). Electro-thermal Simulation of Micromachined

Bohdana Marvalova (2008). Modeling of Magnetoresistive Elastometers. Modelling and

http://www.intechopen.com/articles/show/title/multiphysics\_modelling\_and\_si

Bushyager, N. Intracell Modeling of Metal/Dielectric Interfaces for EBG/MEMS RF

Bushyager, N. Krista Lange, Manos Tentzeris, John Papapolymerou. (2002). Modeling and

Bushyager, N. Manos M.Tentzeris, Larry Gatewood, Jeff DeNatale, (2003). A Novel

Bushyager, N. Mc Garvey, B. & Tentzeris, M.M. (2001). Adaptive Numerical Modeling of RF

Bushyager, N., Dalton, E., Papapolymerou, J. Tentzeris, M. (2002).Modeling of Large Scale

Bushyager, N.A. & Tentzeris, M.M. (2001). Modeling and Design of RF MEMS Structures

Fengyuan Li & Jason Vaughn Clark, (2009). An Online Tool for Simulating Electro-Thermo-

Transducers Journal, Vol.7, Special Issue, pp, 101-115, October 2009. Heung-Shik Lee, Sung-Hoon Cho, Jeong-Bong Lee & Chongdu Cho, (2008).Numerical

Ijaz H. Jafri, Frank DiMeo Jr, Jeffrey, Neuner, Sue DiMascio, James Marchettl, (2005).

used for monitoring process chambers in semiconductor manufacturing.

http://academic.research.microsoft.com/publication/4560023.

American University of Sharjah, Sharjah, UAE, April 20-22-2010. Available from: http://academic.research.microsoft.com/publication/4577492

http://academic.research.microsoft.com/publication/4575444.

25-7, InTech, Available from:

Time-Domain Techniques. Available from:

FDTD Techniques. Available from:

Student Paper Competition.

and Technology Conference.

2008.

http://academic.research.microsoft.com/publication/4571811

http://academic.research.microsoft.com/publication/4569517

mulation\_in\_engineering

Power Sensors at DC and Microwave Frequencies. Presented at The Seventh International Symposium on Machatronics and its Applications SIAM'10.

Simulation, Giuseppe Petrone and Giuliano Cammarata (Ed.), ISBN: 978-3-902613-

Structures Using the Multiresolution Time-Domain Method, (2003). Available from:

Optimization of RF-MEMS Reconfigurable Tuners with Computationally Efficient

Adaptive Approach to Modeling MENS Tunable Capacitors Using MRTD and

Structures Requiring the Coupling of Maxwell's, Mechanical and Solid –State Equations. Submitted to The 17th Annual Review of Progress in Applied Computational Electromagnetics Session: "Wavelets in Electromagnetics" –

RF-MEMS Circuits Using Efficient Time-Domairr Techniques. Available from:

Using Computationally Efficient Numerical Techniques. Electronic Components

Mechanical Flexure Using Distributed and Lumped Analyses. Sensors &

Modelling and Characterization of Micromachined Flexible Magnetostrictive Thin Film Actuator. IEEE Transactions on Magnetics, Vol.44, No. 11, November

Experimental investigation, modeling, and simulations for MEMS based gas sensor

**8. References** 

http://academic.research.microsoft.com/publication/11448778


consumers-and-industrial-applications


http://academic.research.microsoft.com/publication/4577492

Yong Zhu & Horacio Espinosa, D. (2003). Electromechanical Modeling and Simulation of RF MEMS Switches. Proceedings of the 4th International Symposium on MEMS and Nanotechnology, the 2003 SEM Annual Conference and Exposition on Experimental and Applied Mechanics, June 2-4, Charlotte, North Carolina, Session 03, Paper 190, pp.8-11, 2003.

03, Paper 190, pp.8-11, 2003.

Nanotechnology, the 2003 SEM Annual Conference and Exposition on Experimental and Applied Mechanics, June 2-4, Charlotte, North Carolina, Session

### *Edited by Mojtaba Kahrizi*

Micromachining is used to fabricate three-dimensional microstructures and it is the foundation of a technology called Micro-Electro-Mechanical-Systems (MEMS). Bulk micromachining and surface micromachining are two major categories (among others) in this field. This book presents advances in micromachining technology. For this, we have gathered review articles related to various techniques and methods of micro/ nano fabrications, like focused ion beams, laser ablation, and several other specialized techniques, from esteemed researchers and scientists around the world. Each chapter gives a complete description of a specific micromachining method, design, associate analytical works, experimental set-up, and the final fabricated devices, followed by many references related to this field of research available in other literature. Due to the multidisciplinary nature of this technology, the collection of articles presented here can be used by scientists and researchers in the disciplines of engineering, materials sciences, physics, and chemistry.

Photo by lnzyx / iStock

Micromachining Techniques for Fabrication of Micro and Nano Structures

Micromachining Techniques

for Fabrication of Micro and

Nano Structures

*Edited by Mojtaba Kahrizi*