**2. Polycrystalline silicon thin film solar cells: 1 µm grains**

Polycrystalline silicon thin film solar cells in superstrate configuration have been fabricated industrially for some years by the company CSG in Thalheim, Germany. These are the only cells with grains above 1 µm ever fabricated industrially. The preparation steps are as follows (Green et al., 2004). On a borosilicate glass substrate spherical glass beads are deposited, which finally are responsible for light trapping. Then an about 70 nm thick SiN antireflection and barrier layer is deposited by PECVD. On top about 1.5 µm amorphous silicon (a-Si:H) is deposited again be PECVD including the final doping profile n+pp+. The silicon layer is crystallized in the solid state in an 18 h furnace annealing step at about 600°C during which grains of about 1 µm in size form. To activate the dopants a rapid (2 min)

crystallized in a furnace to result in grains about 1 µm in size. This process had been industrialized by the company CSG and is described in Sect. 2. In the lab CSG reached 10.4% efficiency on 90 cm² minimodules (Keevers et al., 2007). Alternatively pulsed excimer laser melting and solidification can been used, which is a standard process in flat panel display

Preparation of multicrystalline silicon thin film solar cells with grains exceeding 10 µm in size is under investigation. This topic is extensively dealt with in Sect. 3. Usually a two-step preparation scheme is used. In a first step a multicrystalline thin seed layer with the desired crystal structure is prepared (Sect. 3.3), which in a second step is epitaxially thickened (Sect. 3.4). For both, seed layer preparation and epitaxial thickening, different processes have been tested. There are, however, attempts to crystallize the complete layer stack of a thin film solar cell in one electron beam melting step (sect. 3.2). The idea for the multicrystalline cells is that in the large grains recombination is reduced, if the crystal quality is high enough, so that the efficiency should exceed that of cells with µm sized grains. Particularly, if the ratio of grain size to layer thickness is large (e.g. 50), such as in multicrystalline wafer cells, a similar efficiency potential is expected. This would require 100 µm large grains for 2 µm thick silicon layers. The preparation methods for large grained multicrystalline silicon layers divide in low and high temperature processes. The high temperature processes are rather straight forward for producing large grains (Beaucarne et al., 2004). However, temperature resistant substrates are required which are expensive. Much more demanding are preparation methods working at temperatures endured by low cost substrates such as glass. One such method is diode laser crystallization. The epitaxial thickening processes, as well, divide in high and low temperature processes with the same drawbacks and advantages. Several methods are presented in Sect. 3. The result of seed layer preparation as well as epitaxial thickening, via melt or in the solid state, depends on temperature history and is explained by the kinetics of phase transformation. The basic notions of this theory as far as they are important for silicon thin film solar cell preparation, are summarized in Sect. 5. Postcrystallization treatments such as rapid thermal annealing and hydrogen passivation are

Even single crystalline silicon thin film cells have been prepared by a transfer process starting from a silicon wafer from which a layer is detached and epitaxially thickened to several 10 µm thickness (Reuter et al., 2009; Brendel, 2001; Brendel et al., 2003; Werner et al., 2009) to reach an efficiency of 17%. This type of cells, which are much thicker than the polyand multicrystalline silicon thin film cells, and which cannot be prepared in the typical sizes

Polycrystalline silicon thin film solar cells in superstrate configuration have been fabricated industrially for some years by the company CSG in Thalheim, Germany. These are the only cells with grains above 1 µm ever fabricated industrially. The preparation steps are as follows (Green et al., 2004). On a borosilicate glass substrate spherical glass beads are deposited, which finally are responsible for light trapping. Then an about 70 nm thick SiN antireflection and barrier layer is deposited by PECVD. On top about 1.5 µm amorphous silicon (a-Si:H) is deposited again be PECVD including the final doping profile n+pp+. The silicon layer is crystallized in the solid state in an 18 h furnace annealing step at about 600°C during which grains of about 1 µm in size form. To activate the dopants a rapid (2 min)

of thin film technology such as > 1 m², is not the topic of this paper.

**2. Polycrystalline silicon thin film solar cells: 1 µm grains** 

production (Sect. 2).

explained in Sect. 4.

thermal annealing step at 900°C follows. The silicon layers are passivated by a hydrogen plasma treatment. Finally rather demanding structuring and contacting processes follow. In production, modules 1x1.4 m² in size reached about 7% efficiency. In the lab 10.4% efficiency were achieved on 92 cm² minimodules (Keevers et al., 2007). The production was stopped, probably because of the high cost PECVD deposition, which was used because the method was the only one available for silicon deposition in the m² range. In the lab, high rate electron beam evaporation was tested as an alternative which delivered minimodules with the efficiency of 6.7%, similar to that of the industrially produced modules (Egan et al., 2009; Sontheimer et al., 2009).

The grain size originating in the furnace anneal is dictated by the interplay of crystal nucleation within the amorphous matrix and growth of the nuclei (see Sect. 5). One can influence both processes by the temperature of the annealing step. Practically, however, there is not much choice. At lower temperature the annealing time required for complete crystallization would reach unrealistic high values so that this is not possible in production. Higher temperatures are not endured by the glass substrate for the time span needed for crystallization. Even at 600°C 18 h are required for crystallization and high temperature resistant borosilicate glass has to be used instead of a much cheaper soda lime glass.

As an alternative for the furnace crystallization pulsed excimer laser crystallization via the melt is a process industrially used in flat panel display production. For this application, however, rather thin films (<100 nm) are required and the resulting grain size is rather small, typically below 1 µm. In the context of solar cell preparation requiring films thicker than 1 µm this method has been mentioned only rarely (Kuo, 2009).

#### **3. Multicrystalline silicon thin film solar cells: grains > 10 µm**

#### **3.1 Basic considerations**

As mentioned in the last paragraph, grains larger than about 1 µm cannot be prepared by direct deposition of crystalline silicon, nor by solid phase crystallization of a-Si nor via melting a-Si by short laser pulses. Large grains can be produced from the melt only if the melt is cooled below the equilibrium melting point slowly so that the melt stays long enough in a region of low nucleation rate and there is time enough for the few nucleating crystallites to grow to large size. Low cooling rate means low heat flow into the substrate following from a low temperature gradient in the substrate. This can be achieved if the melting time of the silicon layer is larger than in excimer laser crystallization, i.e. much larger than 100 ns. To reach longer melting times the energy for melting has to be delivered on a longer time scale. For energy delivery scanned electron beams or scanned laser beams have been used. However, the longer melting time has the consequence, that dopand profiles, introduced into the virgin a-Si for emitter, absorber, and back surface field, get intermixed due to diffusion. Typical diffusion constants in liquid silicon are in the 10-4 to 10-3 cm²/s range (Kodera, 1963) so that dopands will intermix over a distance of 1 µm within 10 to 100 µs. Nevertheless a one-step crystallization procedure for a solar cell layer system has been done by electron beam melting, discussed in Sect. 3.2. Alternatively a two-step procedure has been used. In a first step a thin seed layer is crystallized to large grains from a-Si by laser irradiation. In a second step the seed is thickened epitaxially. Seed and epitaxial layer can be differently doped so that the seed can act as the emitter and the epitaxial layer as the absorber of the solar cell. Alternatively, the seed may act as a highly doped back surface field layer with the epitaxial layer acting as a moderately doped absorber. The emitter is generated on top in a third preparation step.

Crystalline Silicon Thin Film Solar Cells 141

When the Al/a-Si layer system is heated (350°C…550°C below the eutectic temperature of the Al-Si system at 577°C) a layer exchange process takes place combined with silicon crystallization, which is completed, at 500°C, in about 30 min. (Pihan et al., 2007). Finally, a crystalline silicon layer rests on the glass and is covered by an aluminium layer, which may contain silicon islands. The silicon layer is highly p-doped typically by 1019 cm-3 Al (Antesberger et al., 2007). It has been reported that the details of the process and the properties of the final silicon layer depend on the thickness of an aluminum oxide layer which was present between Al and a-Si before the tempering step. Typical resulting silicon grain sizes are in the range of 10 µm. The preferred grain orientation is (100) but other orientations occur as well (Schneider et al., 2006a). Typical layer thicknesses are 300 nm for Al and 375 nm for Si (Fuhs et al., 2004), which is a bit high for seed layers. However, even silicon films thinner than 100 nm have been crystallized by AIC (Antesberger, 2007). Some work has been done to understand the thermodynamics and the kinetics of the process (Wang et al., 2008; Sarikov et al., 2006; Schneider et al., 2006b). It seems that silicon diffuses through the thin alumina layer into the aluminum where it preferably further diffuses towards the glass along the aluminum grain boundaries. When aluminum gets supersaturated by silicon, nucleation of silicon crystallites starts preferably at the interface to the glass substrate. The driving force for the process is the free energy difference between metastable amorphous and absolutely stable crystalline silicon. Finally, the a-Si completely has diffused through the aluminum which then rests on top. Before the crystalline silicon layer can be used as a seed, the aluminum layer has to be removed, e.g. by wet chemical etching using HCl. Challenging is the removal of the silicon islands included in the aluminum layer and of the aluminum oxide film. The removal of both is crucial for good epitaxy (Rau et al., 2004). The inverse process with the starting sequence glass/a-Si/Al and the final sequence glass/Al/c-Si works as well (Gall et al., 2006). It has some advantages for cells in substrate configuration, e.g. that a Al back contact is formed automatically. However, the Al/Si contact has the consequence that any further processing steps, e.g. epitaxy, cannot be performed above the eutectic temperature of the Al-Si system of 577°C. For this reason the inverse

There has been done a lot of work on silicon crystallization by other metals, e.g. Au, Ni, but

To get large silicon crystals by laser crystallization the beam of a cw laser is scanned so that the irradiation time at each position is in the ms range, much larger than during pulse laser irradiation mentioned in Sect. 2. Under these conditions the temperature gradient and therefore the heat flow in the substrate is low so that the melt undercools only slowly, nucleation rates are low, and nucleated crystals have time enough to grow to large sizes (see Sect. 5). First results on this method date back to the late 1970ies (Gat et al., 1978; Colinge et al., 1982). At these times laser crystallization was performed for applications in microelectronics. Therefore amorphous silicon on wafers covered by oxide was used as starting material. The only available well suited lasers were argon ion lasers emitting green light at 514 nm wavelength with a total power of up to 15 W. Typically a circular Gaussian beam with diameter in the 40 µm range was scanned across the sample. At a scanning rate of 12.5 cm/s already in 1978 grains 2x25 µm in size were produced (Gat et al., 1978). Due to the high thermal conductivity of the wafer substrate a rather high power density is needed for melting and crystallization in this case. Only later glass was discovered as a useful substrate

these methods did not find application in solar cell preparation.

**3.3.2 Laser crystallization for seed preparation** 

process was abandoned.

An important issue in any of the mentioned preparation steps is the choice of the substrate. This choice depends on the thermal load the substrate experiences during the silicon crystallization process. Plastic substrates are not useful for any of the processes described in Sect. 3 since the substrate temperature well exceeds 200°C. One usually divides the crystallization methods into low temperature processes for which glass can be used as a substrate and high temperature processes for which glass is not sufficient. Instead, ceramics (e.g. alumina) or graphite has been used. These substrate materials, however, are much more expensive than glass so that the economic consequences for the high temperature routes are not so pleasant. Typically, in high as in low temperature processes some barrier layer is used to prevent the diffusion of foreign atoms from the substrate material into the silicon layer during the processing steps. The barrier layer has to fulfil different requirements except of its main purpose. First of all it has to withstand liquid silicon, i.e. it should not decompose or react with the silicon melt. Moreover, it should not release gases which would blow off the silicon layer. Then it should be well wetted by liquid silicon. Otherwise the silicon film during

melting could dewet to form droplets. This latter requirement is the reason that SiO2 is not useful as a barrier layer. Silicon nitride or silicon carbide are better suited. However, if deposited by PECVD the layers contain too much hydrogen which is released during silicon melting so that the silicon films are destroyed. According to our experience sputtered silicon nitride is well suited if prepared correctly.

#### **3.2 Single step layer preparation - electron beam crystallization**

As mentioned in Sect. 3.1 silicon solar cell absorbers in substrate configuration have been prepared by electron beam crystallization in a one step process (Gromball et al., 2004; Amkreuz et al., 2009). On a glass substrate with a barrier layer (e.g. SiC) 7 to 15 µm of pdoped (1017 cm-3 B) nanocrystalline silicon was deposited by high rate (up to 300 nm/min) PECVD from trichlorosilane. This layer was crystallized by scanning a line shaped electron beam (15 cm x 1 mm). At a scanning rate of 1 cm/s a beam energy density of 500mJ/cm² has been used so that any position is treated for about 0.1 s. The resulting grain size is in the mm range. To get a solar cell a 30 nm thick n-doped a-Si heteroemitter was deposited onto the crystalline absorber by PECVD. The maximum solar cell parameters achieved so far were jsc = 12.4 mA/cm², Voc = 487 mV, and an efficiency of 3.5% (Amkreuz et al., 2009). Obviously the absorber doping is too high and a back surface field is missing. Work is ongoing to improve these cells.

#### **3.3 Two-step process - seed preparation**

In the two-step preparation method first a thin seed layer with the desired crystal structure is prepared which can be used as a back surface field layer or as emitter in the final solar cell. The absorber is then prepared by epitaxial thickening of the seed. In case of a cell in superstrate configuration (illumination through the glass), the seed layer should be rather thin. This is to reduce light absorption in the seed which is highly doped (as emitter or as back surface field layer) and shows only low photovoltaic activity. Two seed preparation methods have been investigated: aluminium induced crystallization (Fuhs et al., 2004) as well as laser crystallization.

#### **3.3.1 Aluminum induced crystallization for seed preparation**

Aluminum induced crystallization (AIC) works as follows: On to the substrate an aluminum layer is deposited by sputtering or evaporation. On top follows an amorphous silicon layer.

An important issue in any of the mentioned preparation steps is the choice of the substrate. This choice depends on the thermal load the substrate experiences during the silicon crystallization process. Plastic substrates are not useful for any of the processes described in Sect. 3 since the substrate temperature well exceeds 200°C. One usually divides the crystallization methods into low temperature processes for which glass can be used as a substrate and high temperature processes for which glass is not sufficient. Instead, ceramics (e.g. alumina) or graphite has been used. These substrate materials, however, are much more expensive than glass so that the economic consequences for the high temperature routes are not so pleasant. Typically, in high as in low temperature processes some barrier layer is used to prevent the diffusion of foreign atoms from the substrate material into the silicon layer during the processing steps. The barrier layer has to fulfil different requirements except of its main purpose. First of all it has to withstand liquid silicon, i.e. it should not decompose or react with the silicon melt. Moreover, it should not release gases which would blow off the silicon layer. Then it should be well wetted by liquid silicon. Otherwise the silicon film during melting could dewet to form droplets. This latter requirement is the reason that SiO2 is not useful as a barrier layer. Silicon nitride or silicon carbide are better suited. However, if deposited by PECVD the layers contain too much hydrogen which is released during silicon melting so that the silicon films are destroyed. According to our experience sputtered silicon

As mentioned in Sect. 3.1 silicon solar cell absorbers in substrate configuration have been prepared by electron beam crystallization in a one step process (Gromball et al., 2004; Amkreuz et al., 2009). On a glass substrate with a barrier layer (e.g. SiC) 7 to 15 µm of pdoped (1017 cm-3 B) nanocrystalline silicon was deposited by high rate (up to 300 nm/min) PECVD from trichlorosilane. This layer was crystallized by scanning a line shaped electron beam (15 cm x 1 mm). At a scanning rate of 1 cm/s a beam energy density of 500mJ/cm² has been used so that any position is treated for about 0.1 s. The resulting grain size is in the mm range. To get a solar cell a 30 nm thick n-doped a-Si heteroemitter was deposited onto the crystalline absorber by PECVD. The maximum solar cell parameters achieved so far were jsc = 12.4 mA/cm², Voc = 487 mV, and an efficiency of 3.5% (Amkreuz et al., 2009). Obviously the absorber doping is too high and a back surface field is missing. Work is ongoing to

In the two-step preparation method first a thin seed layer with the desired crystal structure is prepared which can be used as a back surface field layer or as emitter in the final solar cell. The absorber is then prepared by epitaxial thickening of the seed. In case of a cell in superstrate configuration (illumination through the glass), the seed layer should be rather thin. This is to reduce light absorption in the seed which is highly doped (as emitter or as back surface field layer) and shows only low photovoltaic activity. Two seed preparation methods have been investigated: aluminium induced crystallization (Fuhs et al., 2004) as

Aluminum induced crystallization (AIC) works as follows: On to the substrate an aluminum layer is deposited by sputtering or evaporation. On top follows an amorphous silicon layer.

nitride is well suited if prepared correctly.

**3.3 Two-step process - seed preparation** 

improve these cells.

well as laser crystallization.

**3.2 Single step layer preparation - electron beam crystallization** 

**3.3.1 Aluminum induced crystallization for seed preparation** 

When the Al/a-Si layer system is heated (350°C…550°C below the eutectic temperature of the Al-Si system at 577°C) a layer exchange process takes place combined with silicon crystallization, which is completed, at 500°C, in about 30 min. (Pihan et al., 2007). Finally, a crystalline silicon layer rests on the glass and is covered by an aluminium layer, which may contain silicon islands. The silicon layer is highly p-doped typically by 1019 cm-3 Al (Antesberger et al., 2007). It has been reported that the details of the process and the properties of the final silicon layer depend on the thickness of an aluminum oxide layer which was present between Al and a-Si before the tempering step. Typical resulting silicon grain sizes are in the range of 10 µm. The preferred grain orientation is (100) but other orientations occur as well (Schneider et al., 2006a). Typical layer thicknesses are 300 nm for Al and 375 nm for Si (Fuhs et al., 2004), which is a bit high for seed layers. However, even silicon films thinner than 100 nm have been crystallized by AIC (Antesberger, 2007). Some work has been done to understand the thermodynamics and the kinetics of the process (Wang et al., 2008; Sarikov et al., 2006; Schneider et al., 2006b). It seems that silicon diffuses through the thin alumina layer into the aluminum where it preferably further diffuses towards the glass along the aluminum grain boundaries. When aluminum gets supersaturated by silicon, nucleation of silicon crystallites starts preferably at the interface to the glass substrate. The driving force for the process is the free energy difference between metastable amorphous and absolutely stable crystalline silicon. Finally, the a-Si completely has diffused through the aluminum which then rests on top. Before the crystalline silicon layer can be used as a seed, the aluminum layer has to be removed, e.g. by wet chemical etching using HCl. Challenging is the removal of the silicon islands included in the aluminum layer and of the aluminum oxide film. The removal of both is crucial for good epitaxy (Rau et al., 2004). The inverse process with the starting sequence glass/a-Si/Al and the final sequence glass/Al/c-Si works as well (Gall et al., 2006). It has some advantages for cells in substrate configuration, e.g. that a Al back contact is formed automatically. However, the Al/Si contact has the consequence that any further processing steps, e.g. epitaxy, cannot be performed above the eutectic temperature of the Al-Si system of 577°C. For this reason the inverse process was abandoned.

There has been done a lot of work on silicon crystallization by other metals, e.g. Au, Ni, but these methods did not find application in solar cell preparation.

#### **3.3.2 Laser crystallization for seed preparation**

To get large silicon crystals by laser crystallization the beam of a cw laser is scanned so that the irradiation time at each position is in the ms range, much larger than during pulse laser irradiation mentioned in Sect. 2. Under these conditions the temperature gradient and therefore the heat flow in the substrate is low so that the melt undercools only slowly, nucleation rates are low, and nucleated crystals have time enough to grow to large sizes (see Sect. 5). First results on this method date back to the late 1970ies (Gat et al., 1978; Colinge et al., 1982). At these times laser crystallization was performed for applications in microelectronics. Therefore amorphous silicon on wafers covered by oxide was used as starting material. The only available well suited lasers were argon ion lasers emitting green light at 514 nm wavelength with a total power of up to 15 W. Typically a circular Gaussian beam with diameter in the 40 µm range was scanned across the sample. At a scanning rate of 12.5 cm/s already in 1978 grains 2x25 µm in size were produced (Gat et al., 1978). Due to the high thermal conductivity of the wafer substrate a rather high power density is needed for melting and crystallization in this case. Only later glass was discovered as a useful substrate

Crystalline Silicon Thin Film Solar Cells 143

have rather limited power so that it is impossible to crystallize seed layers for large area solar cells in an industrial environment. For example, a 1 m² module would require many hours laser treatment. Therefore, when looking for high power lasers we ended up with diode lasers, emitting in the near infrared. However, the absorption coefficient of a-Si for 806 nm radiation, the shortest wavelength available for high power diode lasers, at room temperature is only about 0.3 µm-1, as compared to 25 µm-1 for green light. Fig. 2 shows the absorption of 806 nm radiation in amorphous silicon (electron beam deposited, hydrogen free) as calculated from optical properties (*n* and *k*) measured from room temperature up to 600°C and extrapolated up

Obviously there exists a problem for thin films, particularly at room temperature. In thin films, only a small amount of the incoming radiation is absorbed at room temperature. Therefore, to heat the silicon film, a rather high power density is needed. When heating started successfully then the absorption increases and a run-off sets in which is only limited after melting, when the reflectivity jumps up. So the process has some inherent instability, which can be handled only when one preheats the substrate to about 600°C so that laser heating starts at a higher absorption already. The substrate heating has another positive effect, namely to reduce the cracking tendency of the glass substrate, for which we use a borosilicate glass (Schott boro 33) with a thermal expansion coefficient very near to that of

to 1000°C. The maxima and minima are due to interference effects in the silicon layer.

silicon. Work using diode lasers for crystallization started 2006 (Andrä et al., 2006).

layer on the glass substrate.

0 ,0

0 ,1

0 ,2

0 ,3

Absorption

0 ,4

0 ,5

0 ,6

0 ,7

For our seed layer crystallization we use LIMO line focus lasers (806 nm wavelength, 13 mm x 0.1 mm focus and 30 mm x 0.1 mm focus) with maximum power density of up to 25 kW/cm² (Andrä et al., 2006), allowing for scanning speeds up to several cm/s. Fig. 3 shows an EBSD map of a crystallized region demonstrating large grains in the 100 µm range in 450 nm thick films. With the diode laser we can go down to 100 nm thin films. In these the grains size is in the 30 µm range. A further problem with thin films is dewetting. This means that holes form when the silicon film is liquid. It even happens that the holes grow to large sizes and only a part of the substrate is covered by silicon. Dewetting can be reduced if the wetting angle of liquid silicon on the substrate is low. This can be influenced by the barrier

0 100 200 300 400 500 600 700 800 900 1000

Film Thickness [nm] Fig. 2. Absorption of 806 nm diode laser radiation in an amorphous silicon thin film on glass as depending on film thickness. Film temperature 20°C (blue), 600°C (red), and 1000°C (black).

for thin film transistor applications (Michaud et al., 2006) as well as for solar cells (Andrä et al., 1998; Andrä et al. 2000). On glass with low thermal conductivity power densities of about 20 kW/cm² are needed at scanning speeds of several cm/s. Due to the limited laser power the spot diameter was limited to about 100 µm.

Fig. 1. Optical micrograph of a silicon layer crystallized by scanning the circular beam of an argon ion laser. Left: single scan; right: overlapping scans

Fig. 1 (left) shows an optical micrograph of a single trace produced by scanning a circular Ar ion laser beam. At the rim very fine crystals were produced. There the laser power just was able to generate a temperature a bit above the melting point of a-Si, which is well below the melting point of crystalline silicon (see Sect. 5.1). In the rim region a strongly undercooled melt is generated which immediately (that is must faster than the irradiation time) crystallizes to fine grained (about 100 nm) silicon. Towards the center of the trace the power density increases so that the temperature gets higher, the undercooling gets lower, and a bit larger grains solidify. In the central part the laser power is high enough to produce a silicon melt above the equilibrium melting point of crystalline silicon (1412°C). There solidification occurs only when the laser beam already has passed. The slowly undercooling liquid silicon is in contact with the small crystallites of the rim region which crystallized earlier. From these, lateral epitaxial growth takes place. The crystallization direction coincides with the temperature profile following the scanned laser beam. Those of the many nuclei are successful in epitaxy for which the fastest crystallographic growth direction coincides with the temperature gradient. Therefore, a selection mechanism is active and only few of the potential nuclei grow. As a consequence large grains form several 10 µm wide and over 100 µm long. To get not just one crystalline trace but a completely crystallized area, one just has to scan the laser beam in overlapping rows (Fig. 1, right). In the second row the laser beam remelts part of the previous row with the consequence that now the melt is in contact with the large grains produced in the previous row. Therefore large crystals are already present for lateral epitaxy to occur. In this way large areas covered by large grains can be produced. Defect population in films generated in this way has been investigated (Christiansen et al., 2000). The dislocation density was rather low. Grain boundaries are mostly Σ3 and Σ9 twin boundaries which are expected to be not active electrically. The grain orientation is at random with no preferential texture.

Later on, for crystallization the argon ion laser was replaced by a solid state cw Nd:YAG laser, emitting green light of 532 nm wavelength after frequency doubling. Similar results were obtained with this laser type (Andrä et al., 2005a). Both, argon ion as well as Nd:YAG lasers,

for thin film transistor applications (Michaud et al., 2006) as well as for solar cells (Andrä et al., 1998; Andrä et al. 2000). On glass with low thermal conductivity power densities of about 20 kW/cm² are needed at scanning speeds of several cm/s. Due to the limited laser

 Fig. 1. Optical micrograph of a silicon layer crystallized by scanning the circular beam of an

Fig. 1 (left) shows an optical micrograph of a single trace produced by scanning a circular Ar ion laser beam. At the rim very fine crystals were produced. There the laser power just was able to generate a temperature a bit above the melting point of a-Si, which is well below the melting point of crystalline silicon (see Sect. 5.1). In the rim region a strongly undercooled melt is generated which immediately (that is must faster than the irradiation time) crystallizes to fine grained (about 100 nm) silicon. Towards the center of the trace the power density increases so that the temperature gets higher, the undercooling gets lower, and a bit larger grains solidify. In the central part the laser power is high enough to produce a silicon melt above the equilibrium melting point of crystalline silicon (1412°C). There solidification occurs only when the laser beam already has passed. The slowly undercooling liquid silicon is in contact with the small crystallites of the rim region which crystallized earlier. From these, lateral epitaxial growth takes place. The crystallization direction coincides with the temperature profile following the scanned laser beam. Those of the many nuclei are successful in epitaxy for which the fastest crystallographic growth direction coincides with the temperature gradient. Therefore, a selection mechanism is active and only few of the potential nuclei grow. As a consequence large grains form several 10 µm wide and over 100 µm long. To get not just one crystalline trace but a completely crystallized area, one just has to scan the laser beam in overlapping rows (Fig. 1, right). In the second row the laser beam remelts part of the previous row with the consequence that now the melt is in contact with the large grains produced in the previous row. Therefore large crystals are already present for lateral epitaxy to occur. In this way large areas covered by large grains can be produced. Defect population in films generated in this way has been investigated (Christiansen et al., 2000). The dislocation density was rather low. Grain boundaries are mostly Σ3 and Σ9 twin boundaries which are expected to be not active electrically. The grain orientation is at

Later on, for crystallization the argon ion laser was replaced by a solid state cw Nd:YAG laser, emitting green light of 532 nm wavelength after frequency doubling. Similar results were obtained with this laser type (Andrä et al., 2005a). Both, argon ion as well as Nd:YAG lasers,

power the spot diameter was limited to about 100 µm.

argon ion laser. Left: single scan; right: overlapping scans

random with no preferential texture.

have rather limited power so that it is impossible to crystallize seed layers for large area solar cells in an industrial environment. For example, a 1 m² module would require many hours laser treatment. Therefore, when looking for high power lasers we ended up with diode lasers, emitting in the near infrared. However, the absorption coefficient of a-Si for 806 nm radiation, the shortest wavelength available for high power diode lasers, at room temperature is only about 0.3 µm-1, as compared to 25 µm-1 for green light. Fig. 2 shows the absorption of 806 nm radiation in amorphous silicon (electron beam deposited, hydrogen free) as calculated from optical properties (*n* and *k*) measured from room temperature up to 600°C and extrapolated up to 1000°C. The maxima and minima are due to interference effects in the silicon layer.

Obviously there exists a problem for thin films, particularly at room temperature. In thin films, only a small amount of the incoming radiation is absorbed at room temperature. Therefore, to heat the silicon film, a rather high power density is needed. When heating started successfully then the absorption increases and a run-off sets in which is only limited after melting, when the reflectivity jumps up. So the process has some inherent instability, which can be handled only when one preheats the substrate to about 600°C so that laser heating starts at a higher absorption already. The substrate heating has another positive effect, namely to reduce the cracking tendency of the glass substrate, for which we use a borosilicate glass (Schott boro 33) with a thermal expansion coefficient very near to that of silicon. Work using diode lasers for crystallization started 2006 (Andrä et al., 2006).

For our seed layer crystallization we use LIMO line focus lasers (806 nm wavelength, 13 mm x 0.1 mm focus and 30 mm x 0.1 mm focus) with maximum power density of up to 25 kW/cm² (Andrä et al., 2006), allowing for scanning speeds up to several cm/s. Fig. 3 shows an EBSD map of a crystallized region demonstrating large grains in the 100 µm range in 450 nm thick films. With the diode laser we can go down to 100 nm thin films. In these the grains size is in the 30 µm range. A further problem with thin films is dewetting. This means that holes form when the silicon film is liquid. It even happens that the holes grow to large sizes and only a part of the substrate is covered by silicon. Dewetting can be reduced if the wetting angle of liquid silicon on the substrate is low. This can be influenced by the barrier layer on the glass substrate.

Fig. 2. Absorption of 806 nm diode laser radiation in an amorphous silicon thin film on glass as depending on film thickness. Film temperature 20°C (blue), 600°C (red), and 1000°C (black).

Crystalline Silicon Thin Film Solar Cells 145

In the two step preparation method on top of the multicrystalline seed layer the absorber of the solar cell is prepared by epitaxial growth. Several methods have been used which can be classified into direct epitaxial deposition and deposition as amorphous silicon followed by epitaxial crystallization, either in the solid state by furnace or by laser annealing or via laser melting. Particularly in the cases without melting the cleanliness of the interface between crystalline seed and amorphous silicon to be epitaxially crystallized is an issue. Any contaminants present, even small amounts of a monolayer, will jeopardize epitaxial crystallization or at least increase the amount of extended defects in the epitaxial layer appreciably. First of all, any silicon oxide has to be removed from the seed surface. This can be achieved by HF. A 2% to 5 % solution in water is most useful. Success can be observed by the naked eye. When HF has removed the oxide the silicon surface gets hydrogenated which makes the surface hydrophobic and the etching solution dewets, i.e. forms droplets. Then the HF solution can be blown off by nitrogen. The hydrogenated surface state remains stable in ambient air at room temperature for about 1 h so that there is time enough to introduce the sample into a deposition chamber for a-Si deposition. However, other possible contaminants are not so easily removed. It turned out as useful to start with an RCA cleaning

step before HF treatment. The RCA step removes e.g. organic contaminants.

The simplest epitaxial thickening procedure is direct epitaxial deposition of silicon on top of the seed layer. Several processes have been investigated in the past, high temperature CVD and, at intermediate temperature, electron beam evaporation, ECRCVD, and hot wire CVD. The high temperature route has been reviewed recently (Beaucarne et al., 2004). The highest efficiency reached so far with this method is 8% (Gordon et al., 2007). On an alumina substrate seed layers were prepared by aluminium induced crystallization. Epitaxial thickening for the p-doped absorber with rates up to 1.4 µm/min was done by thermal CVD at 1130°C from trichlorosilane. The final emitter was prepared by phosphorus diffusion, or an a-Si heteroemitter was deposited by PECVD. Corresponding to the seed layer the grain size in the absorber is several 10 µm. It is expected that the efficiency is not so much limited by the grain size but by intragrain defects, which have been thoroughly investigated (van Gestel et al.,

Even higher efficiencies of 11.1% were reached on seed layers crystallized by lamp heater zone melting on graphite and high temperature epitaxy for absorber growth (Kunz et al., 2008). The high temperature process has the advantage that it works on any grain orientation of the seed. However, high temperature resistant substrates such as alumina, silica, glass ceramics, or

At intermediate temperature both, electron beam evaporation, partly modified by ion assisted deposition, or ECR-CVD (electron cyclotron resonance CVD) has been tested for epitaxy on AIC seed layers. ECR-CVD was successfully applied at 585°C substrate temperature (Rau et al., 2004). However, epitaxy worked well only on (100)-oriented grains, which is the most common orientation following from AIC, but not the only one. At 670°C epitaxy by hot wire CVD worked on any grain orientation with a rate of 100 nm/min. Ion assisted deposition, that is electron beam evaporation plus some ionization of the silicon atoms, was tested for epitaxy as well. For the deposition a temperature ramp was carefully optimized with maximum temperature below 700°C. The deposition rate was 300 nm/min. The highest achieved open circuit voltage of solar cells was 453 mV (Straub et al., 2005). Direct epitaxy during electron

graphite are needed, which are not very feasible for large scale production.

**3.4 Two step process - epitaxial thickening** 

**3.4.1 Direct epitaxial deposition** 

2009).

Fig. 3. EBSD map (inverse pole figure) of diode laser crystallized seed layers 450 nm (left) and 110 nm (middle) thick. Color code for grain orientation is shown on the right.

Concerning the throughput, laser companies are just developing line focus diode lasers with long lines (Lichtenstein 2010) which would allow crystallization of a 1 m² module within minutes.

If seed layers thinner than 100 nm are to be crystallized diode lasers cannot be used due to too low absorption even when preheated. We tested a pulsed green laser (JenLas ASAMA) emitting 515 nm wavelength radiation (Andrä et al., 2010). This laser has a line focus up to 100 mm long and 5 to 10 µm wide and it delivers 600 ns pulses at a repetition rate of up to 80 kHz. At a fluence of about 1.2 J/cm² the sample was shifted 1.5 µm between subsequent pulses. In this way 60 nm thin seed layers were crystallized without any preheating with resulting grains several µm wide and several 10 µm long (Fig. 4). Obviously, the melt generated during each laser pulse solidifies by lateral epitaxy so the grains generated by the previous pulse grow stepwise. Finally long grains form, which continue over many pulses. Since the width of the melt is 5 µm in our case and the melt exists for a time interval in the several µs range, the solidification speed is in the m/s range. This value is near the maximum following from solidification kinetics (see Sect. 5).

Fig. 4. EBSD map (inverse pole figure) of pulse laser crystallized seed layers 60 nm thick.

300 µm 100 µm

maximum following from solidification kinetics (see Sect. 5).

minutes.

Fig. 3. EBSD map (inverse pole figure) of diode laser crystallized seed layers 450 nm (left)

Concerning the throughput, laser companies are just developing line focus diode lasers with long lines (Lichtenstein 2010) which would allow crystallization of a 1 m² module within

If seed layers thinner than 100 nm are to be crystallized diode lasers cannot be used due to too low absorption even when preheated. We tested a pulsed green laser (JenLas ASAMA) emitting 515 nm wavelength radiation (Andrä et al., 2010). This laser has a line focus up to 100 mm long and 5 to 10 µm wide and it delivers 600 ns pulses at a repetition rate of up to 80 kHz. At a fluence of about 1.2 J/cm² the sample was shifted 1.5 µm between subsequent pulses. In this way 60 nm thin seed layers were crystallized without any preheating with resulting grains several µm wide and several 10 µm long (Fig. 4). Obviously, the melt generated during each laser pulse solidifies by lateral epitaxy so the grains generated by the previous pulse grow stepwise. Finally long grains form, which continue over many pulses. Since the width of the melt is 5 µm in our case and the melt exists for a time interval in the several µs range, the solidification speed is in the m/s range. This value is near the

Fig. 4. EBSD map (inverse pole figure) of pulse laser crystallized seed layers 60 nm thick.

and 110 nm (middle) thick. Color code for grain orientation is shown on the right.

#### **3.4 Two step process - epitaxial thickening**

In the two step preparation method on top of the multicrystalline seed layer the absorber of the solar cell is prepared by epitaxial growth. Several methods have been used which can be classified into direct epitaxial deposition and deposition as amorphous silicon followed by epitaxial crystallization, either in the solid state by furnace or by laser annealing or via laser melting. Particularly in the cases without melting the cleanliness of the interface between crystalline seed and amorphous silicon to be epitaxially crystallized is an issue. Any contaminants present, even small amounts of a monolayer, will jeopardize epitaxial crystallization or at least increase the amount of extended defects in the epitaxial layer appreciably. First of all, any silicon oxide has to be removed from the seed surface. This can be achieved by HF. A 2% to 5 % solution in water is most useful. Success can be observed by the naked eye. When HF has removed the oxide the silicon surface gets hydrogenated which makes the surface hydrophobic and the etching solution dewets, i.e. forms droplets. Then the HF solution can be blown off by nitrogen. The hydrogenated surface state remains stable in ambient air at room temperature for about 1 h so that there is time enough to introduce the sample into a deposition chamber for a-Si deposition. However, other possible contaminants are not so easily removed. It turned out as useful to start with an RCA cleaning step before HF treatment. The RCA step removes e.g. organic contaminants.

#### **3.4.1 Direct epitaxial deposition**

The simplest epitaxial thickening procedure is direct epitaxial deposition of silicon on top of the seed layer. Several processes have been investigated in the past, high temperature CVD and, at intermediate temperature, electron beam evaporation, ECRCVD, and hot wire CVD. The high temperature route has been reviewed recently (Beaucarne et al., 2004). The highest efficiency reached so far with this method is 8% (Gordon et al., 2007). On an alumina substrate seed layers were prepared by aluminium induced crystallization. Epitaxial thickening for the p-doped absorber with rates up to 1.4 µm/min was done by thermal CVD at 1130°C from trichlorosilane. The final emitter was prepared by phosphorus diffusion, or an a-Si heteroemitter was deposited by PECVD. Corresponding to the seed layer the grain size in the absorber is several 10 µm. It is expected that the efficiency is not so much limited by the grain size but by intragrain defects, which have been thoroughly investigated (van Gestel et al., 2009).

Even higher efficiencies of 11.1% were reached on seed layers crystallized by lamp heater zone melting on graphite and high temperature epitaxy for absorber growth (Kunz et al., 2008). The high temperature process has the advantage that it works on any grain orientation of the seed. However, high temperature resistant substrates such as alumina, silica, glass ceramics, or graphite are needed, which are not very feasible for large scale production.

At intermediate temperature both, electron beam evaporation, partly modified by ion assisted deposition, or ECR-CVD (electron cyclotron resonance CVD) has been tested for epitaxy on AIC seed layers. ECR-CVD was successfully applied at 585°C substrate temperature (Rau et al., 2004). However, epitaxy worked well only on (100)-oriented grains, which is the most common orientation following from AIC, but not the only one. At 670°C epitaxy by hot wire CVD worked on any grain orientation with a rate of 100 nm/min. Ion assisted deposition, that is electron beam evaporation plus some ionization of the silicon atoms, was tested for epitaxy as well. For the deposition a temperature ramp was carefully optimized with maximum temperature below 700°C. The deposition rate was 300 nm/min. The highest achieved open circuit voltage of solar cells was 453 mV (Straub et al., 2005). Direct epitaxy during electron

Crystalline Silicon Thin Film Solar Cells 147

In summary we could epitaxially crystallize up to 1.6 µm of a-Si at a temperature of 630°C within 3 h. The epitaxial quality as determined by EBIC was best in (100) oriented grains and worst in (111) grains. Moreover, the epitaxial crystallization speed depends on orientation and on the doping level. Higher doped layers crystallize faster. Solar cells prepared on these layers reached an efficiency of 4.9% after hydrogen passivation (Schneider et al., 2010). By TEM cross section investigations it was shown that the seed layers contain only very few extended defects such as dislocations, whereas the epitaxial layer contains much more. It seems that the cleaning procedure of the seed surface prior to a-Si deposition is crucial for good epitaxial quality. At least the dislocation density in the epitaxial layer could be reduced by an additional RCA cleaning step before removal of oxide by HF. However,

The epitaxy method of layered laser crystallization has been developed in our group years ago (Andrä et al. 2005b, Andrä et al., 2008a). The principle is simple. During deposition of a-Si on top of the seed layer excimer laser pulses are applied repeatedly, which melt the newly deposited a-Si and a bit of the crystalline silicon beneath so that after each pulse epitaxial solidification occurs. Again, the layer thickness to be crystallized by one laser shot is limited by a competing nucleation process in the undercooling melt after the laser pulse. According to our experience about 200 nm of a-Si can be epitaxially crystallized by one laser pulse. The typical laser fluence needed is 550 mJ/cm². However, when during the whole thickening process the thickness of the crystalline layer beneath the newly deposited a-Si increases from the initial seed layer (say 200 nm) to the final absorber thickness (say 2 µm) the laser parameters or the thickness of the newly deposited a-Si have to be adjusted so that the laser pulse just melts the a-Si and bit of c-Si beneath. This adjustment is necessary because the thermal properties of glass, c-Si, and a-Si differ so that the temperature profiles change during the process if the laser energy would be kept constant. In the layered laser crystallization process epitaxy works independently of the grain orientation, which is an advantage since crystal orientation in the seed is at random. For the process, the laser pulse has to be fed through a window in the deposition chamber onto the growing layer. In this way the pulses can be applied without stopping deposition. For a-Si deposition we use electron beam evaporation which has first the advantage of high deposition rate, at least an order of magnitude higher than for PECVD, and secondly the advantage that deposition is directed so that no deposition occurs at the laser window. Doping is achieved by codeposition of boron or phosphorus. In our device we can deposit and laser irradiate substrates of up to 10x10 cm². The single laser spot has a size of 6x6 mm² with top hat profile. To cover the whole substrate area the laser spot is scanned over the substrate by a scanning mirror placed outside the deposition chamber. In order to avoid cracks in the glass substrate heating to about 600°C helps. Upscaling the system to m² surely is a challenge but not outside the technical possibilities. If properly optimized, about 10 laser pulses are needed at each position during absorber deposition to prepare a 2 µm thick epitaxial film. This makes sense only if the laser is fed into the deposition chamber and is applied without braking

In the epitaxial layer prepared by layered laser crystallization the number of extended defects like dislocations is much lower as compared to solid state epitaxy. This is because the mobility of crystallizing atoms is much higher in the melt than in a-Si so that correct placement is easier. The highest efficiencies achieved in solar cells prepared using the

this did not reflect in the achieved solar cell efficiencies.

deposition, as we do it in our lab scale equipment.

**3.4.3 Layered laser crystallization** 

beam evaporation at 550°C substrate temperature has successfully been demonstrated (Dogan et al., 2008). Solar cells prepared with this process reached 346 mV open circuit voltage and 2.3% efficiency, which is a bit low as compared to the values achieved by other methods.
