**1. Introduction**

Age related macular degeneration (AMD) and retinitis pigmentosa (RP) are two of the most common outer retinal degenerative diseases that have resulted in vision impairment and blindness of millions of people. Specifically, AMD has become the third leading cause of blindness on global scale, and is the primary cause of visual deficiency in industrialized countries (World Health Organization [WHO], 2011). In the United States, more than 500,000 people are suffering from RP and around 20,000 of them are legally blind (Artificial Retina Project, 2007). Whereas many treatment methods, including gene replacement therapy (Bennett et al., 1996), pharmaceutical therapy, nutritional therapy (Norton et al., 1993), photoreceptor and stem cell transplantations (MacLaren et al., 2006 & Tropepe et al., 2000), and dietary, have been explored to slow down the development of AMD and RP diseases in their early stages, vision impairment and blindness due to outer retinal degeneration currently remain incurable.

In the 1990's, researchers discovered that although the retinal photoreceptors are defective in RP patients, their optic nerves, bipolar and ganglion cells to which the photoreceptors synapse still function at a large rate (Humayun et al., 1999). Further studies showed similar results in AMD patients (Kim et al., 2002). These findings have created profound impact on the ophthalmology field, by providing the possibility of using artificial retinal prostheses to partially restore the lost vision function in AMD and RP patients. Two main retinal implant approaches are currently in development according to the layer of retina receiving the implanted device: subretinal (Chow et al., 2006; Rizzo, 2011; Zrenner et al., 1999) and epiretinal prostheses (Humayun et al., 1994; Weiland & Humayun 2008; Wong et al., 2009). Particularly, epiretinal implantation has received widespread attention over the last few years, for not only successful clinical trials demonstrating its efficacy in patients, but also its many advantages compared to others (Horch K.W. & Dhillon, G.S., 2004). First, the device implantation and follow-up examination only require standard ophthalmologic technologies, which can effectively reduce the risk of trauma during the surgery and also allow for the implant to be replaced easily. In addition, most of the implanted electronics are kept in the vitreous cavity so

Implantable Parylene MEMS RF Coil for Epiretinal Prostheses 5

Preliminary results suggest that the fold-and-bond technology is a very promising approach for making high Q MEMS coils in a simple, low-cost, and microfabrication compatible manner. Future direction for coil optimization will aim to increase the number of metal

Fig. 1. System schematic of an all-intraocular retinal prosthetic system, which contains two MEMS radio-frequency (RF) coils for power and data transmission, circuitry integrated on a flexible Parylene cable for converting the signal to simulation pulses, and a high-density

Microfabricated coils usually suffer from low self-inductances and inevitable parasitic effects, namely parasitic resistances and capacitances, due to their small physical dimensions and technical constrains of surface micromachining. Particularly, for intraocular retinal implants in human subjects, planar coils with a maximal outer diameter of ~ 10 mm and a minimal inner diameter of ~ 3 mm are desired, which is limited by the space availability in the anterior chamber of human eyes. To better understand the electrical properties and parasitic effects in such small devices, we studied several analytical models and have now applied them to a simplified circular-shape planar coil as illustrated in Fig. 2. In this section, the coil selfinductance, ERS, and parasitic capacitance will be discussed separately with respect to the

*do di d*

Fig. 2. Simplified model of a circular-shape planar MEMS coil.

Epiretinal

multielectrode array

Intraocular RF coil

Integrated circuit chips

*t*

*t*

Flexible cable

layers in order to enhance the Q factor and the power transfer efficiency.

MEMS electrode array for simulating the neural cells.

**2. Modeling of microcoils** 

geometric parameters of devices.

that potential thermal damage to the surrounding retinal tissues can be mitigated by taking the advantage of the heat dissipating properties of the vitreous. Finally, developments of modern microelectromechanical systems (MEMS) technologies enable complete intraocular retinal implants and high-density array stimulation.

Various epiretinal prosthetic technique options have been studied by a number of groups worldwide, by replacing the defective photoreceptors with a multielectrode array implanted on the surface of the inner retina between the vitreous and internal limiting membrane (Javaheri et al. 2006; Rizzo et al., 2004; Stieglitz et al. 2004). Although they have different system configurations, these systems generally utilize a pair of coils to transfer data and power wirelessly between an extraocular data acquisition system and intraocular electronics. In such implementations, the intraocular receiver coil, which resides inside the eye, has many constrains compared to the extraocular transmitter coil. First, the coil has to be mechanically durable to withstand the surgical procedure. It also needs to be flexible and small enough to facilitate the device implantation through small incisions. Finally, it should be chemically inert and biocompatible to prevent harmful interaction with surrounding tissue/cells. Currently, most systems are still using thick and stiff hand-wound coils as the receiver coils, which can cause notable degradation in the implant region. In addition, interconnections between the hand-wound coils and other components are usually formed by soldering, which will require additional hermetic package to ensure device biocompability. Planar microcoils that contain electroplated gold wires on flexible polyimide substrates have also been developed (Mokwa et al. 2008). While device functionality has been confirmed in animal models, several critical issues such as long-term reliability and biocompatibility need to be further addressed for human implantation.

To overcome these challenges, we have proposed a polymer-based MEMS technique for making microcoils. Devices typically consist of multi-layer conductive wires encapsulated by polymeric materials (Li et al., 2005; Li et al., 2006; Chen et al., 2008). In this design, Parylene C is selected as a main structural and packaging material because of its many unique properties, such as flexibility (Young's modulus ~ 4 GPa), chemical inertness, United States Pharmacopoeia (USP) Class VI biocompatibility (Rodger et al., 2006), and lower water permeability compared with other commonly used polymers (e.g. PDMS and polyimide) (Licari & Hughes, 1990). Microfabrication technology has several advantages over the handwinding including miniaturized structures, precise dimensional control, and feasibility for system integration. The proposed coils will finally be integrated with other system components, such as circuitry, discrete electrical components, a flexible cable, and a highdensity multielectrode array, to achieve a complete all-intraocular epiretinal prosthetic system (Rodger et al., 2008), as described in Fig. 1.

This chapter will concentrate on the design, microfabrication, and testing results of two types of Parylene-based intraocular MEMS coils for applications in epiretinal implantation. Specifically, Section 2 will introduce theoretical approaches for coil modeling and design. Coil's electrical properties, including self-inductance, effective series resistance (ESR), and parasitic capacitance, will be discussed theoretically with respect to their geometries. Section 3 describes a Parylene-metal-Parylene thin film technology for making the proposed microcoils. Two various types of microfabricated coil prototypes, including regular duallayered planar coils and novel fold-and-bond coils, are implemented and characterized in Section 4. Successful data transmissions through such devices have been demonstrated using inductive coupling tests. Experiment also confirms that the quality factor (Q factor) of microcoils can increase proportionally with the increase of the number of metal layers.

that potential thermal damage to the surrounding retinal tissues can be mitigated by taking the advantage of the heat dissipating properties of the vitreous. Finally, developments of modern microelectromechanical systems (MEMS) technologies enable complete intraocular retinal

Various epiretinal prosthetic technique options have been studied by a number of groups worldwide, by replacing the defective photoreceptors with a multielectrode array implanted on the surface of the inner retina between the vitreous and internal limiting membrane (Javaheri et al. 2006; Rizzo et al., 2004; Stieglitz et al. 2004). Although they have different system configurations, these systems generally utilize a pair of coils to transfer data and power wirelessly between an extraocular data acquisition system and intraocular electronics. In such implementations, the intraocular receiver coil, which resides inside the eye, has many constrains compared to the extraocular transmitter coil. First, the coil has to be mechanically durable to withstand the surgical procedure. It also needs to be flexible and small enough to facilitate the device implantation through small incisions. Finally, it should be chemically inert and biocompatible to prevent harmful interaction with surrounding tissue/cells. Currently, most systems are still using thick and stiff hand-wound coils as the receiver coils, which can cause notable degradation in the implant region. In addition, interconnections between the hand-wound coils and other components are usually formed by soldering, which will require additional hermetic package to ensure device biocompability. Planar microcoils that contain electroplated gold wires on flexible polyimide substrates have also been developed (Mokwa et al. 2008). While device functionality has been confirmed in animal models, several critical issues such as long-term reliability and biocompatibility need to be further addressed for human implantation. To overcome these challenges, we have proposed a polymer-based MEMS technique for making microcoils. Devices typically consist of multi-layer conductive wires encapsulated by polymeric materials (Li et al., 2005; Li et al., 2006; Chen et al., 2008). In this design, Parylene C is selected as a main structural and packaging material because of its many unique properties, such as flexibility (Young's modulus ~ 4 GPa), chemical inertness, United States Pharmacopoeia (USP) Class VI biocompatibility (Rodger et al., 2006), and lower water permeability compared with other commonly used polymers (e.g. PDMS and polyimide) (Licari & Hughes, 1990). Microfabrication technology has several advantages over the handwinding including miniaturized structures, precise dimensional control, and feasibility for system integration. The proposed coils will finally be integrated with other system components, such as circuitry, discrete electrical components, a flexible cable, and a highdensity multielectrode array, to achieve a complete all-intraocular epiretinal prosthetic

This chapter will concentrate on the design, microfabrication, and testing results of two types of Parylene-based intraocular MEMS coils for applications in epiretinal implantation. Specifically, Section 2 will introduce theoretical approaches for coil modeling and design. Coil's electrical properties, including self-inductance, effective series resistance (ESR), and parasitic capacitance, will be discussed theoretically with respect to their geometries. Section 3 describes a Parylene-metal-Parylene thin film technology for making the proposed microcoils. Two various types of microfabricated coil prototypes, including regular duallayered planar coils and novel fold-and-bond coils, are implemented and characterized in Section 4. Successful data transmissions through such devices have been demonstrated using inductive coupling tests. Experiment also confirms that the quality factor (Q factor) of microcoils can increase proportionally with the increase of the number of metal layers.

implants and high-density array stimulation.

system (Rodger et al., 2008), as described in Fig. 1.

Preliminary results suggest that the fold-and-bond technology is a very promising approach for making high Q MEMS coils in a simple, low-cost, and microfabrication compatible manner. Future direction for coil optimization will aim to increase the number of metal layers in order to enhance the Q factor and the power transfer efficiency.

Fig. 1. System schematic of an all-intraocular retinal prosthetic system, which contains two MEMS radio-frequency (RF) coils for power and data transmission, circuitry integrated on a flexible Parylene cable for converting the signal to simulation pulses, and a high-density MEMS electrode array for simulating the neural cells.

### **2. Modeling of microcoils**

Microfabricated coils usually suffer from low self-inductances and inevitable parasitic effects, namely parasitic resistances and capacitances, due to their small physical dimensions and technical constrains of surface micromachining. Particularly, for intraocular retinal implants in human subjects, planar coils with a maximal outer diameter of ~ 10 mm and a minimal inner diameter of ~ 3 mm are desired, which is limited by the space availability in the anterior chamber of human eyes. To better understand the electrical properties and parasitic effects in such small devices, we studied several analytical models and have now applied them to a simplified circular-shape planar coil as illustrated in Fig. 2. In this section, the coil selfinductance, ERS, and parasitic capacitance will be discussed separately with respect to the geometric parameters of devices.

Fig. 2. Simplified model of a circular-shape planar MEMS coil.

Implantable Parylene MEMS RF Coil for Epiretinal Prostheses 7

voltage profile can then be obtained by averaging the beginning and ending potential across the coil structure. With known voltage variations between the correlated sections of adjacent turns and layers, the total capacitive energy stored in the coil structure can be calculated using the distributed capacitance of each segment. The equivalent capacitance can then be approximated from the distributed capacitances, using the ideal double plate capacitor

<sup>1</sup> <sup>C</sup> C l [d(k 1) d(k 1)] ,

<sup>1</sup> <sup>A</sup> <sup>C</sup> (C C ...) [4 2d(k 1) 2d(k)] <sup>4</sup> <sup>m</sup>

Equations (4), (5), and (6) show the analytical formulas for calculating parasitic capacitance, where Cii (in F) denotes the capacitance per unit length between adjacent metal turns, Cm, m-1 (in F) is the capacitance per unit area between the m-th and (m-1)-th metal layer, Ak (in m2) is the trace occupied area of the k-th turn on each layer, and d(k)=h1+ h2+…+ hk, in which hk is defined as the ratio of the wire length of the k-th turn (lk ) to the total wire length (ltot). This simplified model neglects the second order parasitic capacitances between non-

Q factor is an important metric for evaluating the efficiency of a coil, which is theoretically defined as the ratio of total stored energy to dissipated energy per cycle in a resonating system. With known Ls, Rs, and Cs, the Q factor of the coil can be derived from a 3-element circuit model (Fig. 3) (Wu, 2003). In order to obtain the coil Q factor mathematically, the total equivalent impedance (Zs) is first studied, which can be written as the sum of a real

<sup>1</sup> <sup>A</sup> (C C ...) [2d(k 1) 2d(k)] , <sup>4</sup> <sup>m</sup>

adjacent turns and layers, which are much less than the first order capacitances.

2

k 2

(5)

(4)

k 2

C C C, eq total eq turn eq la yer (6)

2 22

 

(7)

j .

s s ss ss

s 22 2 22 2 ss ss ss ss

<sup>R</sup> (L R C L C ) <sup>Z</sup> (1 L C ) ( C R ) (1 L C ) ( C R )

 

n 1

4

eq turn ii k k 1

eq layer m,m 1 m 2,m 3 2

m 1,m 2 m 3,m 4 2

N

k 1

resistance and an imaginary reactance (equation [7]).

Fig. 3. Equivalent RLC circuit of a planar MEMS coil.

N

k 1

**2.4 Quality factor** 

formula.

#### **2.1 Self-inductance**

The self-inductance (Ls) of a multi-layer circular coil with rectangular cross-section can be calculated using the following equation (Dwight , 1945):

$$\mathcal{L}\_s = 2\pi \text{d(nN)}^2 \times 10^{-9} \left[ \left( \ln \frac{4\text{d}}{\text{t}} \right) \left( 1 + \frac{\text{t}^2}{24\text{d}^2} \cdots \right) - \frac{1}{2} + \frac{43\text{t}^2}{288\text{d}^2} \cdots \right] \quad \text{(in } \text{Henry)}, \tag{1}$$

where d (in cm) is the mean diameter of the coil, t (in cm) is the coil width, n is the number of turns on each layer, and N is the total number of layers. This expression is valid only when the coil is operated at a low-frequency, i.e., no skin effect. The skin effect can be evaluated using a frequency-dependent factor, which is known as skin depth δ and can be calculated as

$$
\delta = \sqrt{\frac{2\rho}{o\mu}} \quad \text{(in }\text{ meter)}.\tag{2}
$$

where ρ is the electrical resistivity of metal (in Ω·m), ω is the angular frequency (in rad/s), and µ is the permeability of metal (in H/m). In our proposed system, the data signal is modulated on a ~ 22 MHz carrier, and the power transfer is taken place at a frequency within 1-2 MHz. Therefore, the estimated skin depths at these low frequencies are much bigger than the thickness of metal thin films produced from physical vapor deposition (PVD). In this case, the skin effect can be negligible with an assumption of uniform current distribution in conductive wires.

#### **2.2 Effective series resistance**

ERS (Rs) is commonly used to estimate coil losses, which plays an important role in designing a power efficient inductive link. The ESR can be divided into two parts: DC resistance and frequency dependent resistance. Assuming the width of metal traces is much larger than the separation distance between adjacent turns that can be ignored, the DC resistance of the proposed coil can be calculated with the Ohm's law as given in (3),

$$\mathbf{R}\_s = \rho \frac{\mathbf{n}^2 \mathbf{N} \pi \mathbf{d}}{\mathbf{t} \times \mathbf{h}} \quad \text{(\Omega)}, \tag{3}$$

where ρ is the metal resistivity (in Ω·m), and h is the metal thickness (in m). As a result of the skin-effect, the frequency dependent part can be neglected at the low operating frequencies as mentioned earlier. Therefore, the equivalent ESR can be simply written as a DC resistance.

#### **2.3 Parasitic capacitance**

Parasitic capacitance (Cs) places a limit to the self-resonant frequency of the coil, above which the coil will not behave as an inductor any more. In a first-order approximation, the parasitic capacitance of a planar MEMS coil usually consists of two main components: the capacitance between turns and the capacitance between layers. A distributed model has been developed to estimate the equivalent parasitic capacitance, as discussed elsewhere (Wu et al. 2003 & Zolfaghari et al. 2001). In this method, a planar coil can be decomposed into equal sections by assuming consistent thickness and width of metal traces everywhere. The voltage profile can then be obtained by averaging the beginning and ending potential across the coil structure. With known voltage variations between the correlated sections of adjacent turns and layers, the total capacitive energy stored in the coil structure can be calculated using the distributed capacitance of each segment. The equivalent capacitance can then be approximated from the distributed capacitances, using the ideal double plate capacitor formula.

$$\mathbf{C}\_{\text{eq\\_turn}} = \sum\_{\mathbf{k}=1}^{n-1} \frac{1}{4} \mathbf{C}\_{\text{ii}} \mathbf{l}\_{\text{k}} [\mathbf{d}(\mathbf{k}+1) - \mathbf{d}(\mathbf{k}-1)]^2 \tag{4}$$

$$\begin{aligned} \mathbf{C\_{eq-layer}} &= \frac{1}{4} \sum\_{\mathbf{k}=1}^{\mathrm{N}} (\mathbf{C\_{m,m-1}} + \mathbf{C\_{m-2,m-3}} + \ldots) \frac{\mathbf{A\_{k}}}{\mathrm{m}^{2}} [4 - 2\mathrm{d}(\mathbf{k} - 1) - 2\mathrm{d}(\mathbf{k})]^{2} \\ &+ \frac{1}{4} \sum\_{\mathbf{k}=1}^{\mathrm{N}} (\mathbf{C\_{m-1,m-2}} + \mathbf{C\_{m-3,m-4}} + \ldots) \frac{\mathbf{A\_{k}}}{\mathrm{m}^{2}} [2\mathrm{d}(\mathbf{k} - 1) + 2\mathrm{d}(\mathbf{k})]^{2}, \end{aligned} \tag{5}$$

$$\mathbf{C}\_{\text{eq\\_total}} = \mathbf{C}\_{\text{eq\\_turn}} + \mathbf{C}\_{\text{eq\\_layer}} \tag{6}$$

Equations (4), (5), and (6) show the analytical formulas for calculating parasitic capacitance, where Cii (in F) denotes the capacitance per unit length between adjacent metal turns, Cm, m-1 (in F) is the capacitance per unit area between the m-th and (m-1)-th metal layer, Ak (in m2) is the trace occupied area of the k-th turn on each layer, and d(k)=h1+ h2+…+ hk, in which hk is defined as the ratio of the wire length of the k-th turn (lk ) to the total wire length (ltot). This simplified model neglects the second order parasitic capacitances between nonadjacent turns and layers, which are much less than the first order capacitances.

#### **2.4 Quality factor**

6 Microelectromechanical Systems and Devices

The self-inductance (Ls) of a multi-layer circular coil with rectangular cross-section can be

4d t 1 43t L 2 d(nN) 10 ln 1 (in Henry), t 2 24d 288d

where d (in cm) is the mean diameter of the coil, t (in cm) is the coil width, n is the number of turns on each layer, and N is the total number of layers. This expression is valid only when the coil is operated at a low-frequency, i.e., no skin effect. The skin effect can be evaluated using a frequency-dependent factor, which is known as skin depth δ and can be

<sup>2</sup> (in meter),

where ρ is the electrical resistivity of metal (in Ω·m), ω is the angular frequency (in rad/s), and µ is the permeability of metal (in H/m). In our proposed system, the data signal is modulated on a ~ 22 MHz carrier, and the power transfer is taken place at a frequency within 1-2 MHz. Therefore, the estimated skin depths at these low frequencies are much bigger than the thickness of metal thin films produced from physical vapor deposition (PVD). In this case, the skin effect can be negligible with an assumption of uniform current

ERS (Rs) is commonly used to estimate coil losses, which plays an important role in designing a power efficient inductive link. The ESR can be divided into two parts: DC resistance and frequency dependent resistance. Assuming the width of metal traces is much larger than the separation distance between adjacent turns that can be ignored, the DC

resistance of the proposed coil can be calculated with the Ohm's law as given in (3),

s

2

 

nN d <sup>R</sup> ( ), t h 

where ρ is the metal resistivity (in Ω·m), and h is the metal thickness (in m). As a result of the skin-effect, the frequency dependent part can be neglected at the low operating frequencies as mentioned earlier. Therefore, the equivalent ESR can be simply written as a

Parasitic capacitance (Cs) places a limit to the self-resonant frequency of the coil, above which the coil will not behave as an inductor any more. In a first-order approximation, the parasitic capacitance of a planar MEMS coil usually consists of two main components: the capacitance between turns and the capacitance between layers. A distributed model has been developed to estimate the equivalent parasitic capacitance, as discussed elsewhere (Wu et al. 2003 & Zolfaghari et al. 2001). In this method, a planar coil can be decomposed into equal sections by assuming consistent thickness and width of metal traces everywhere. The

s 2 2

2 2

(2)

(3)

(1)

**2.1 Self-inductance** 

calculated as

DC resistance.

**2.3 Parasitic capacitance** 

distribution in conductive wires.

**2.2 Effective series resistance** 

calculated using the following equation (Dwight , 1945):

2 9

Q factor is an important metric for evaluating the efficiency of a coil, which is theoretically defined as the ratio of total stored energy to dissipated energy per cycle in a resonating system. With known Ls, Rs, and Cs, the Q factor of the coil can be derived from a 3-element circuit model (Fig. 3) (Wu, 2003). In order to obtain the coil Q factor mathematically, the total equivalent impedance (Zs) is first studied, which can be written as the sum of a real resistance and an imaginary reactance (equation [7]).

Fig. 3. Equivalent RLC circuit of a planar MEMS coil.

$$Z\_{s} = \frac{\text{R}\_{\text{s}}}{\left(1 - \rho^{2}\text{L}\_{\text{s}}\text{C}\_{\text{s}}\right)^{2} + \left(\rho\text{C}\_{\text{s}}\text{R}\_{\text{s}}\right)^{2}} + \frac{\rho\text{\left(L\_{\text{s}} - \text{R}\_{\text{s}}\text{ $^{2}$ C}\_{\text{s}} - \rho^{2}\text{L}\_{\text{s}}\text{ $^{2}$ C}\_{\text{s}}\right)}{\left(1 - \rho^{2}\text{L}\_{\text{s}}\text{C}\_{\text{s}}\right)^{2} + \left(\rho\text{C}\_{\text{s}}\text{R}\_{\text{s}}\right)^{2}}.\tag{7}$$

Implantable Parylene MEMS RF Coil for Epiretinal Prostheses 9

A multi-layer Parylene-metal thin film technology for making the proposed microcoils has been developed (Li et al., 2005). In this approach, thin-film metal conductive wires are sandwiched between multiple layers of Parylene C and interconnections between two adjacent layers are implemented using through holes in the Parylene insulation layer. Fig. 5 depicts a typical process flow for making a dual-metal-layer structure. Briefly, a layer of sacrificial photoresist is optionally spun on a standard silicon wafer, followed by Parylene C deposition (PDS 2120 system, Special Coating Systems, Indianapolis, IN, USA) (*Step 1*). A layer of metal is then deposited on top of the Parylene using an electron beam (e-beam) evaporator (SE600 RAP, CHA Industries, Fremont, CA, USA), and patterned using a wet etching process (*Step 2*). After that, a thin layer of Parylene C is deposited as an insulation layer between two metal layers, and the interconnection vias are selectively opened with oxygen plasma in a reactive ion etch system (RIE) (Semi Group Inc. T1000 TP/CC) using a photoresist mask(*Step 3*). After removing the photoresist mask, the second metal layer is evaporated and patterned, followed by another Parylene C deposition to conformally cover the exposed metal wire (*Step 4*). A photoresist mask is then patterned to expose the contact pads, as well as to define the coil shape (*Step 5*). Finally, oxygen plasma etch is performed to remove unwanted Parylene C, and the entire flexible device is released from the silicon

The reliability of the interconnections between nearby metal layers highly depends on the step coverage of the Parylene sidewall during metal evaporation, which can be improved by a slightly isotropic O2 plasma etch (Meng et al., 2008). A special design of rotating wafer holder inside the e-beam evaporator also helps adjust the angle of attack of metal evaporant for best coverage. Microcoils comprising more than two layers of metal can be fabricated with similar procedure by alternating the Parylene C deposition, interconnection via fabrication, and metal evaporation process steps. Although it is specifically developed for microcoil fabrication, this technology can also be applied to the fabrication of other flexible, implantable devices with multi-layer Parylene-metal structures, such as dual-metal-layer

Fig. 4. Simulated self-inductance and ESR of the sample coil.

**3. Parylene-metal-Parylene thin film technology** 

substrate by dissolving the sacrificial photoresist (*Step 6*).

electrode arrays (Rodger et al., 2008).

Then the self-resonant frequency ωs can be expressed as:

$$
\rho \alpha\_{\rm s} = \sqrt{\frac{1}{\mathcal{L}\_{\rm s} \mathcal{C}\_{\rm s}} - \frac{\mathcal{R}\_{\rm s}^2}{\mathcal{L}\_{\rm s}^2}} \approx \sqrt{\frac{1}{\mathcal{L}\_{\rm s} \mathcal{C}\_{\rm s}}}, \text{ when } \mathcal{R}\_{\rm s} \ll \sqrt{\frac{\mathcal{L}\_{\rm s}}{\mathcal{C}\_{\rm s}}}. \tag{8}
$$

For a retinal implant system, when both the external and internal units of the inductive link are tuned to a same resonant frequency ωr, the maximal coupling energy can be delivered to the implanted system. In this case, the coil Q factor can be expressed with the following equation:

$$\mathbf{Q}\_{\mathbf{r}} = \frac{\mathrm{Im}(\mathbf{Z}\_{\mathrm{s}})}{\mathrm{Re}(\mathbf{Z}\_{\mathrm{s}})} \approx \frac{\alpha\_{\mathrm{r}} \mathrm{L}\_{\mathrm{s}}}{\mathrm{R}\_{\mathrm{s}}} \tag{9}$$

Combining with equations (1) and (3), equation (9) can be rewritten as

$$\mathbf{Q}\_{\mathbf{r}} = \frac{2\alpha\_{\mathbf{r}}\mathbf{N}\mathbf{t}\mathbf{t}}{\rho} \left[ \left( \ln \frac{\mathbf{4d}}{\mathbf{t}} \right) \left( \mathbf{1} + \frac{\mathbf{t}^2}{2\mathbf{4d}^2} \cdots \right) - \frac{1}{2} + \frac{\mathbf{43t}^2}{288\mathbf{d}^2} \cdots \right] \times \mathbf{10}^{-9} \tag{10}$$

Ideally, the Q factor of a coil should be as high as possible in order to minimize the power loss in the device as well as to maximize power transfer efficiency of the system. It can be seen from equation (10) that Qr can be enhanced by increasing the number of coil layers (N), the coil width (t), and/or the thickness of the conductive layer (h). For an intraocular retinal implant, however, there is not much zoom to improve the coil width due to the small coil dimensions (inner diameter, outer diameter, etc.) confined by the eyeball size. Therefore, a more applicable way to increase a coil's Q factor is to increase the number of stacking layers as well as the thickness of conductive wires.

#### **2.5 Finite element simulation**

To validate the effectiveness of the theoretical models, finite element simulations (FES) are performed using a built-in package in CoventorWare (Coventor Inc., Cary, NC). As a demonstration, a coil with two layers of metal is designed, and its electrical characteristics are evaluated using both analytical models and FES, as summarized in Fig. 4 and Table 1.

During the simulation, an octagonal coil is used to approximate a circular shape due to memory constraint in CoventorWare. The coil self-inductance and the ESR are simulated over a frequency range from 10 kHz to 1 GHz. It can be seen that the self-inductance at 1 MHz shows only 2.2% deviation, and the ESR deviates by less than 6%, suggesting good agreement with the analytical models. The slight deviations might be introduced by the approximation of coil shape. Note that Ls and Rs both remain stable at frequencies below 10 MHz, indicating that skin effect or proximity effect is negligible at target frequencies of 1 or 2 MHz.


Table 1. Coil characteristics estimated using both analytical models and FES.

when <sup>s</sup> <sup>s</sup>

For a retinal implant system, when both the external and internal units of the inductive link are tuned to a same resonant frequency ωr, the maximal coupling energy can be delivered to the implanted system. In this case, the coil Q factor can be expressed with the following

> s rs <sup>r</sup> s s

r 9

 

Ideally, the Q factor of a coil should be as high as possible in order to minimize the power loss in the device as well as to maximize power transfer efficiency of the system. It can be seen from equation (10) that Qr can be enhanced by increasing the number of coil layers (N), the coil width (t), and/or the thickness of the conductive layer (h). For an intraocular retinal implant, however, there is not much zoom to improve the coil width due to the small coil dimensions (inner diameter, outer diameter, etc.) confined by the eyeball size. Therefore, a more applicable way to increase a coil's Q factor is to increase the number of stacking layers

To validate the effectiveness of the theoretical models, finite element simulations (FES) are performed using a built-in package in CoventorWare (Coventor Inc., Cary, NC). As a demonstration, a coil with two layers of metal is designed, and its electrical characteristics are evaluated using both analytical models and FES, as summarized in Fig. 4 and Table 1. During the simulation, an octagonal coil is used to approximate a circular shape due to memory constraint in CoventorWare. The coil self-inductance and the ESR are simulated over a frequency range from 10 kHz to 1 GHz. It can be seen that the self-inductance at 1 MHz shows only 2.2% deviation, and the ESR deviates by less than 6%, suggesting good agreement with the analytical models. The slight deviations might be introduced by the approximation of coil shape. Note that Ls and Rs both remain stable at frequencies below 10 MHz, indicating that skin effect or proximity effect is negligible at target frequencies of 1 or

> Trace cross section (µm × µm)

Table 1. Coil characteristics estimated using both analytical models and FES.

Calculations 10 3 220 × 2 28 / Layer 5.0 28.9 65.5 1.1 FES 10 3 220 × 2 28 / Layer 4.9 27.4 -- 1.12

Number of turns

Ls (µH)

Rs (Ω)

Cs (nF)

Q at 1MHz

t 2 24d 288d

2 2

Im(Z ) L <sup>Q</sup> Re(Z ) R

r 2 2 2 Nth 4d t 1 43t <sup>Q</sup> ln 1 <sup>10</sup>

s

(9)

(8)

(10)

<sup>L</sup> R . C

2 <sup>s</sup> <sup>s</sup> <sup>2</sup> s s <sup>s</sup> s s 1 1 <sup>R</sup> , LC LC <sup>L</sup>

Combining with equations (1) and (3), equation (9) can be rewritten as

Then the self-resonant frequency ωs can be expressed as:

as well as the thickness of conductive wires.

**2.5 Finite element simulation** 

OD (mm)

ID (mm)

2 MHz.

equation:

Fig. 4. Simulated self-inductance and ESR of the sample coil.

#### **3. Parylene-metal-Parylene thin film technology**

A multi-layer Parylene-metal thin film technology for making the proposed microcoils has been developed (Li et al., 2005). In this approach, thin-film metal conductive wires are sandwiched between multiple layers of Parylene C and interconnections between two adjacent layers are implemented using through holes in the Parylene insulation layer. Fig. 5 depicts a typical process flow for making a dual-metal-layer structure. Briefly, a layer of sacrificial photoresist is optionally spun on a standard silicon wafer, followed by Parylene C deposition (PDS 2120 system, Special Coating Systems, Indianapolis, IN, USA) (*Step 1*). A layer of metal is then deposited on top of the Parylene using an electron beam (e-beam) evaporator (SE600 RAP, CHA Industries, Fremont, CA, USA), and patterned using a wet etching process (*Step 2*). After that, a thin layer of Parylene C is deposited as an insulation layer between two metal layers, and the interconnection vias are selectively opened with oxygen plasma in a reactive ion etch system (RIE) (Semi Group Inc. T1000 TP/CC) using a photoresist mask(*Step 3*). After removing the photoresist mask, the second metal layer is evaporated and patterned, followed by another Parylene C deposition to conformally cover the exposed metal wire (*Step 4*). A photoresist mask is then patterned to expose the contact pads, as well as to define the coil shape (*Step 5*). Finally, oxygen plasma etch is performed to remove unwanted Parylene C, and the entire flexible device is released from the silicon substrate by dissolving the sacrificial photoresist (*Step 6*).

The reliability of the interconnections between nearby metal layers highly depends on the step coverage of the Parylene sidewall during metal evaporation, which can be improved by a slightly isotropic O2 plasma etch (Meng et al., 2008). A special design of rotating wafer holder inside the e-beam evaporator also helps adjust the angle of attack of metal evaporant for best coverage. Microcoils comprising more than two layers of metal can be fabricated with similar procedure by alternating the Parylene C deposition, interconnection via fabrication, and metal evaporation process steps. Although it is specifically developed for microcoil fabrication, this technology can also be applied to the fabrication of other flexible, implantable devices with multi-layer Parylene-metal structures, such as dual-metal-layer electrode arrays (Rodger et al., 2008).

Implantable Parylene MEMS RF Coil for Epiretinal Prostheses 11

 (a) (b) Fig. 6. (a) A fabricated dual-metal-layer coil sitting on a penny. (b) The microscope image

shows the interconnection via between two metal layers. (Li et al., @ 2005 IEEE)

Fig. 7. Demonstration of the coil's flexibility and foldability. (Li et al., @ 2005 IEEE)

imaginary part of the impedance is zero (Wu, 2003).

The electrical properties of the fabricated coil are characterized experimentally. Recall equation (7) in Section 2, by setting the derivation of the real part to zero and equating the imaginary part to zero, the self-inductance (Ls) and the parasitic capacitance (Cs) can be extracted using equations (11) and (12), where ω0 is defined as the frequency at which the real part of the impedance is maximum, and ωz is the zero-reactance frequency at which the

<sup>s</sup> <sup>S</sup> 2 2

<sup>R</sup> L , 2( ) 

0 z

2 2 0 z S 2 2 s0 z 2( ) C . R (2 ) 

(12)

 

For the coil in Fig. 6, the ESR (Rs) is measured to be around 72 Ω and the resistivity of ebeam deposited gold is calculated to be around 2.25×10-6 Ω·cm. This number agrees with the resistivity of bulk gold (2.2×10-6 Ω·cm), implying that the E-beam evaporated metal is voidfree. The coil impedance is swept with an HP 4192A LF impedance analyzer over a frequency range from 5 Hz to 13 MHz. From the impedance versus frequency curves (Fig. 8), f0 and fz can be read with values of 7.5 MHz and 3.3 MHz respectively. Knowing Rs, ω0, and ωz, the coil self-inductance and capacitance are therefore calculated as Ls = 1.19 µH and

(11)

Fig. 5. Fabrication process flow of a dual-metal-layer Parylene-based MEMS structure.

For implantable devices, our Parylene-metal thin film technology has several unique advantages compared with conventional semiconductor-based microfabrication technologies. Using biocompatible Parylene directly as the actual substrate greatly simplifies the device integration and packaging procedures. Devices fabricated in this way are very flexible and foldable so that they can be implanted through small surgical incisions, allowing wounds to heal quickly. Moreover, the metal lines are completely padded by the Parylene material, and can therefore withstand repeated bending during surgical handling. Finally, a post-fabrication heat-molding process has been developed to modify the skins into various shapes that match the curvatures of the target implant areas (Tai et al., 2006).

### **4. Coil designs and fabrication results**
