Crystalline Silicon Thin Film Solar Cells

In the last few years the marked share of thin film solar cells increased appreciably to 16.8% (in 2009). The main part of that increase refers to CdTe modules (9.1%) followed by silicon thin film cells, that is amorphous silicon (a-Si) cells or tandem cells consisting of a-Si and nanocrystalline silicon (μc-Si). For a review on thin film solar cells in general see (Green, 2007) and on a-Si/μc-Si cells see (Beaucarne, 2007). The a-Si cells suffer from a low efficiency. In the lab the highest efficiency up to now is 10.1% on 1 cm2 (Green et al., 2011), whereas in the industrial production modules reach about 7%. In order to achieve the required electronic quality of hydrogenated amorphous silicon (a-Si:H), low deposition rate (max. 50 nm/min) PECVD (plasma enhanced chemical vapour deposition) is used for deposition which makes production more expensive as compared to CdTe modules. This is even worse for the layer system in a-Si/μc-Si tandem cells for which the more than 1 μm thick nanocrystalline μc-Si layer is deposited by PECVD, too, however with much lower deposition rates in the 10 nm/min range. Cells consisting just of μc-Si reached 10.1% efficiency (Green et al., 2011), just as a-Si-cells, whereas tandem cells arrived at 11.9%, both for lab cells, whereas in production the results are below 10%. The low deposition rate combined with the limited efficiency, make these cells not too competitive compared to CdTe cells, which, at lower cost, reach 11% in industrial production, or to CIGS (Copperindium-gallium-diselenide) cells with similar efficiencies. As an alternative, polycrystalline (grains in the μm range) or multicrystalline (grains >10 μm) silicon thin film solar cells receive growing interest (Beaucarne et al., 2006). The present paper reviews the status of these cells, and on the other hand gives details of laser based preparation methods, on which the authors have been working for many years. Both types, polyand multicrystalline silicon thin film cells, are prepared by depositing amorphous silicon followed by some crystallization process. One main advantage of the crystallization process is that the electronic quality of the virgin a-Si is not important. Therefore high rate deposition processes such as electron beam evaporation or sputtering can be used which are much less expensive as compared to low rate PECVD. In case of sputtering doped thin films can be deposited by using doped sputtering targets, whereas in electron beam evaporation the dopands are coevaporated from additional sources. So, in these deposition processes the use of toxic or hazardous gases such as silane, phosphine or diborane is avoided, reducing the abatement cost. Polycrystalline silicon layers for solar cells can be prepared in a single crystallization step. The layer system containing the doping profile is deposited in the amorphous state and is


Introduction
In the last few years the marked share of thin film solar cells increased appreciably to 16.8% (in 2009).The main part of that increase refers to CdTe modules (9.1%) followed by silicon thin film cells, that is amorphous silicon (a-Si) cells or tandem cells consisting of a-Si and nanocrystalline silicon (µc-Si).For a review on thin film solar cells in general see (Green, 2007) and on a-Si/µc-Si cells see (Beaucarne, 2007).The a-Si cells suffer from a low efficiency.In the lab the highest efficiency up to now is 10.1% on 1 cm² (Green et al., 2011), whereas in the industrial production modules reach about 7%.In order to achieve the required electronic quality of hydrogenated amorphous silicon (a-Si:H), low deposition rate (max.50 nm/min) PECVD (plasma enhanced chemical vapour deposition) is used for deposition which makes production more expensive as compared to CdTe modules.This is even worse for the layer system in a-Si/µc-Si tandem cells for which the more than 1 µm thick nanocrystalline µc-Si layer is deposited by PECVD, too, however with much lower deposition rates in the 10 nm/min range.Cells consisting just of µc-Si reached 10.1% efficiency (Green et al., 2011), just as a-Si-cells, whereas tandem cells arrived at 11.9%, both for lab cells, whereas in production the results are below 10%.The low deposition rate combined with the limited efficiency, make these cells not too competitive compared to CdTe cells, which, at lower cost, reach 11% in industrial production, or to CIGS (Copperindium-gallium-diselenide) cells with similar efficiencies.As an alternative, polycrystalline (grains in the µm range) or multicrystalline (grains >10 µm) silicon thin film solar cells receive growing interest (Beaucarne et al., 2006).The present paper reviews the status of these cells, and on the other hand gives details of laser based preparation methods, on which the authors have been working for many years.Both types, poly-and multicrystalline silicon thin film cells, are prepared by depositing amorphous silicon followed by some crystallization process.One main advantage of the crystallization process is that the electronic quality of the virgin a-Si is not important.Therefore high rate deposition processes such as electron beam evaporation or sputtering can be used which are much less expensive as compared to low rate PECVD.In case of sputtering doped thin films can be deposited by using doped sputtering targets, whereas in electron beam evaporation the dopands are coevaporated from additional sources.So, in these deposition processes the use of toxic or hazardous gases such as silane, phosphine or diborane is avoided, reducing the abatement cost.Polycrystalline silicon layers for solar cells can be prepared in a single crystallization step.The layer system containing the doping profile is deposited in the amorphous state and is 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 production (Sect.2).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 explained in Sect. 4.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 of thin film technology such as > 1 m², is not the topic of this paper.

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) 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).

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.
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 SiO 2 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.

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 (10 17 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 j sc = 12.4 mA/cm², V oc = 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.

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.

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.
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 10 19 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.

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 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.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, 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.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). 100 µm 300 µm

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.

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 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.

Solid phase epitaxy in furnace
Technically the most simple way to achieve epitaxial growth is to deposit first an amorphous layer on top of the cleaned seed layer, and then to epitaxially crystallize the layer by furnace annealing in the solid state.The layer to be crystallized can already contain the desired doping profile which remains during the annealing step.The main critical point with this simple procedure is that not only an epitaxial crystallization front moves into a-Si, but also spontaneous nucleation will occur within a-Si followed by growth of crystallites.So there exists a competing process to the desired epitaxy.The question arises, which of the two succeeds.The speed of the epitaxial front of course depends on temperature (described by Jackson-Chalmers equation, see Sect.5.1) and so does nucleation, described by classical nucleation theory (Sect.5.2), and growth of nuclei, the latter phenomena described together by ).An important point, which makes SPE possible, is that, if no nuclei pre-exist in the amorphous matrix, nucleation does not start immediately.Instead it needs some time, called time lag of nucleation, until a stationary population of nuclei evolves (Sect.5.3).Only after that time lag the stationary nucleation rate applies at fixed temperature, described by classical nucleation theory, and crystal nuclei appear.So any successful epitaxy relies on the time lag of nucleation.The thickness of an epitaxially crystallized layer is just given by the time lag of nucleation times the speed of the epitaxial crystallization front.After the time lag, in the virgin amorphous silicon crystalline nuclei of random orientation appear resulting in fine grained material, such as is generated by direct furnace crystallization (see Sect. 2) without seed.For successful epitaxy one has to make sure that within the amorphous phase there are no nuclei present which could form during deposition already.In the last few years we developed the technique of SPE on diode laser crystallized seed layers on borosilicate glass substrates (Andrä et al., 2008a;Schneider et al., 2010).The virgin a-Si layers including a doping profile were deposited at high rate (typically 300 nm/min) by electron beam evaporation at a substrate temperature in the 300°C range.At that temperature no nuclei form within a-Si.The layer system was then annealed in a furnace under ambient air.To control the progress of crystallization, an in situ measurement technique was installed.For this purpose, the beam of a low power test laser was sent through the sample.The transmitted intensity was monitored by a photocell.Since a-Si has a different optical absorption from c-Si, the progress of crystallization can be monitored easily.In particular, the crystallization process is complete when the transmission does not change any more.Fig. 5 shows a transmission electron micrograph of a cross section of an epitaxially thickened silicon film.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, this did not reflect in the achieved solar cell efficiencies.

Layered laser crystallization
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 deposition, as we do it in our lab scale equipment.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 method were 4.8% at an open circuit voltage of 517 mV (Andrä et al., 2005b;Andrä et al., 2007).These values were measured on cells without any light trapping.

Liquid or solid phase epitaxy by diode laser irradiation
The layered laser crystallization method described in the last section has the drawback that up-scaling into the industrial scale is not so easy.This is due to the fact, that the laser beam has to be fed into the deposition chamber and several pulses have to be applied at each position.That was the motivation for us to look for a method in which the complete absorber thickness is deposited in the amorphous state on top of the seed and to apply a single laser treatment to epitaxially crystallize the whole system in one run after deposition outside the deposition chamber.The most obvious way to achieve epitaxy is via the liquid phase similar to layered laser crystallization.The main difference is that the whole amorphous absorber precursor layer is melted in one step down to the seed, so that epitaxial solidification is to occur after irradiation.It is a challenge to melt about 1 µm of a-Si without completely melt the about 200 nm thin c-Si seed beneath which would hamper any epitaxy.To crystallize a layer system more than 1 µm thick, a short pulse laser is useless.To get the required energy into the system the pulse fluence would have to be so large that ablation would occur at the surface.Moreover, the cooling rate of the melt after a short laser pulse would be so high, that nucleation is expected to occur in a surface near region before the epitaxial solidification front reaches the surface.Therefore we decided to use a scanned cw diode laser for this purpose with irradiation times in the ms range.In this case the cooling rate is low enough so that the melt stays long enough in a slightly undercooled state with low nucleation rate until the epitaxial solidification front reaches the surface.We succeeded in epitaxially crystallizing 500 nm in one run.However, forming of cracks is an issue.Moreover, due to the strong diffusion in the melt which intermixes any pre-existing doping profile, absorber and emitter cannot be crystallized in one step.An alternative is solid phase epitaxy in which the amorphous layer is heated by the laser to a temperature of about 1100°C, below the melting point of a-Si.At such high temperature the solid phase epitaxial speed was determined to several 100 nm/s high so that epitaxy of 1 µm should be complete within several seconds.

Emitter preparation
The emitter of the final solar cell can be prepared in different ways.One is to include emitter doping into the deposition sequence of the layer system so that no additional emitter preparation step is needed.This way has been chosen in the CSG process and in layered laser crystallization.It cannot be applied in case of liquid phase epitaxy of the whole layer stack (Sect.3.4.4)since during melting for several ms, diffusion in the liquid state would intermix any dopand profile introduced during deposition.In this case, phosphorus doping of a boron doped absorber as in conventional wafer cells can be performed.The only difference is that the doping profile has to be much shallower.Another variant is to use amorphous heteroemitters.IMEC has found that this is the best emitter for their thin film solar cells prepared by the high temperature route (Gordon et al., 2007).

RTA and hydrogen passivation
To improve the solar cell performance some post-crystallization treatment is required.One point is dopand activation, the other defect passivation.In order that dopand atoms like boron or phosphorus really lead to a free carrier concentration higher than the intrinsic one, it is necessary that the dopand atoms are included substitutionally in the lattice, i.e. that they rest on regular lattice positions replacing a silicon atom.If they are included interstitially, resting not on regular lattice positions, they are useless.If the silicon lattice forms from the melt the mobility of atoms is high enough so that the dopand atoms can occupy lattice positions.In this case no additional means are needed to make them active.This is not so in case of solid state crystallization.There most of the dopand atoms are included interstitially so that they are inactive.To let them replace silicon atoms substitutionally an additional heat treatment is needed, which is realized by a rapid thermal annealing (RTA) step.In the CSG process, for example, the whole system is heated to about 900°C for 2 min to achieve dopand activation (Keevers et al., 2007).It has been a lot of speculation if this RTA step also improves the grain structure by reducing the number of extended defects.This seems not to be the case (Brazil & Green, 2010).In any case a hydrogen passivation step has to follow, in which different types of defects e.g.dislocations and grain boundaries, are passivated.Usually, a remote hydrogen plasma is applied to the layer system for 10 to 30 min at about 500°C.Crucial is that during cooling down at the end of the process the plasma has to be applied for some time.A lot of optimization work has been devoted to this passivation step (Rau et al., 2006), which easily can improve the open circuit voltage of the cell by 200 mV.

Kinetics of phase transformation
In Sect. 5 the basics of phase transformation relevant for silicon thin film crystallization, both from the melt and in the solid state are summarized (Falk & Andrä, 2006).The Section divides in the propagation of already present phase boundaries and in nucleation, including non-stationary nucleation.Kinetics of aluminum induced crystallization has already been reviewed (Pihan et al., 2007) and is not treated in the following.The facts presented in this section are the background for any successful crystallization of amorphous silicon, in the furnace or by laser irradiation.Quantitative values following from the equations depend on the material parameters of the system involved.These are rather well known for crystalline and for liquid silicon, mostly in the whole range of temperature involved in the processes.This is not the case for amorphous silicon, the properties of which strongly depend on the preparation conditions.They may appreciably differ for hydrogenated a-Si prepared by PECVD and hydrogen free a-Si deposited by electron beam evaporation.Therefore, quantitative predictions have to be taken with some care.

Propagation of phase boundaries
The propagation speed of already present phase boundaries into a metastable phase, i.e. the growth of a crystal into the undercooled melt or into amorphous silicon, can quantitatively be described by the The prefactor v 0 = a 0 ν depends on the atomic vibration frequency (Debye frequency) ν, the jump distance a 0 of the order of the lattice parameter of silicon and on a geometry factor of the order of 1. Δµ>0 is the difference in chemical potential of the phases involved.For the transition from liquid to crystalline Δµ may be approximated by (1 ) c mc where Δh c is the latent heat per mole for melting and T mc is the equilibrium melting temperature of 1685 K.For the crystallization of amorphous silicon Δµ is given in the literature (Donovan et al., 1983).g* is an activation energy for the jump of an atom from the parent to the final phase and is related to the self-diffusion coefficient D according to Results for crystallization from the melt and in the solid state are given in Figs. 6 and 7.In the melt the crystallization speed vanishes at the equilibrium melting point T mc to increase to a maximum of about 16 m/s at 200 K undercooling.At even lower temperature the solidification front gets slower due to the increasing influence of the activation energy.At temperatures above the melting point the phase front runs into the crystal, i.e. the crystal melts and the speed changes sign.In Fig. 6. also the melting speed of amorphous silicon is shown (with opposite sign as compared to c-Si).Melting of a-Si starts at T ma , which, depending on the deposition conditions of a-Si, is 200 to 300 K lower than the melting point of c-Si.The crystallization speed in amorphous silicon shown in Fig. 7 increases with temperature, and reaches about 1 mm/s near the melting point of a-Si.At 600°C the speed is only about 0.2 nm/s which well correlates with the results obtained in furnace solid phase epitaxy (Sect.3.4.2).

Stationary nucleation rate
Classical nucleation theory gives the nucleation rate J, i.e. the number of nuclei appearing in a metastable phase per volume and time interval at given temperature.The value applies after some induction time (Sect.5.3) and as long as not too much of the parent phase is consumed.
In this formula V m is the atomic volume and j c and ΔG c are the number of atoms in and the free energy of a critical nucleus of the new phase in the matrix of the parent phase, respectively.These are given by is the interface energy between both the phases, which, however, is hard to determine independently of nucleation phenomena, and, in addition, may depend on temperature.Moreover, strongly influences the nucleation rate since via Eqs.5&6 it enters Eq. 4 in the third power within the exponential.For crystallization in an undercooled silicon melt the stationary nucleation rate is plotted in Fig. 8 for a temperature dependent interfacial energy according to = (43,4+0.249T/K) mJ/m 2 (Ujihara et al., 2001).Down to about 300 K below the equilibrium melting point the nucleation rate is very low to change within 100 K of further cooling by 35 orders of magnitude.Below 1200 K the nucleation rate gets rather flat at a value of 10 35 m -3 s -1 = 0.1 nm -3 ns -1 .The stationary nucleation rate of crystallization in amorphous silicon is plotted in Fig. 9.There the values increase by 16 orders of magnitude when temperature is increased from 600 K to 1200 K.The nucleation rate then flattens off at 10 17 m -3 s -1 = 0.1 µm -3 s -1 up to the melting point of a-Si of 1400 K.

Non-stationary nucleation
When the temperature of a system is changed abruptly from a value where the parent phase is absolutely stable and there are no nuclei present to another temperature where it gets metastable, then a population of nuclei evolves.Finally, a stationary distribution of nuclei emerges which leads to the stationary nucleation rate of Eq. 4. The master equation for the population of nuclei can be solved numerically.By some approximations a closed form for non-stationary nucleation rate has been derived (Kashchiev, 1969), which leads to the stationary value after some time lag of nucleation, which is given by Here g is an accommodation coefficient of the order of 1.The result for nucleation of c-Si from the melt is shown in Fig. 10.The time lag diverges at the equilibrium melting point and has a minimum of 30 ps around 1350 K.At all relevant temperatures the time lag is so small that it does not play any role in laser crystallization with pulses longer than 1 ns.This is different for solid phase crystallization of amorphous silicon as shown in Fig. 11.The time lag goes down from 10 13 s (or 300.000 years) at 600 K to 0.01 s at the melting point of a-Si (1400 K).That means that below 300°C crystallization never occurs whereas in the CSG process of furnace crystallization at 600°C the time lag is in the range of 2 h which does not play a major role when complete crystallization takes 18 h.However, it gives an upper limit for epitaxial growth by furnace annealing as described in Sect.3.4.2.

Complete kinetics of transformation
Stationary nucleation together with the growth of supercritical nuclei according to the Jackson-Chalmers equation leads to a continuous increase of the amount of the new phase on account of the parent phase.When one takes account that during the progress of phase transformation more and more parent phase is consumed and less volume is available for actual transformation, one ends up with the Avrami-Mehl equation (Avrami, 1940) for the volumetric amount of the new phase  J is the stationary nucleation rate of Eq. 4 and v is the speed of propagation of a phase front according to Jackson-Chalmers Eq. 1.In deriving Eq. 9 the time lag of nucleation has been neglected.To include this effect, one simply replaces t by (t-) in Eq. 9 for t> .The resulting average grain size when the parent phase has been consumed completely is given by 4 So the grains are the larger the higher the Jackson-Chalmers speed and the lower the nucleation rate is, which sounds reasonable.To get large grains from an undercooled melt one should keep the temperature in a range of not too high undercooling, where nucleation rate is low and growth rate is high (Figs.6 and 8).Fig. 12 shows the expected final grain size in solid phase crystallization of amorphous silicon.It shows that in the CSG process at about 600°C (see Sect. 3.) grains of several µm are to be expected, which is in accordance with experiments.By increasing the crystallization temperature one cannot change the grain size appreciably.Lowering the temperature would lead to a rather high time needed for crystallization due to higher time lag of nucleation (Fig. 11), lower nucleation rate (Fig. 9), and lower growth rate (Fig. 7).

Conclusion
Multi-and polycrystalline silicon thin film solar cells receive growing interest worldwide.Presently, the maximum efficiency reached by these types of cells is 10.4%.Different cell concepts and preparation methods are under investigation and no clear favourite way is identified up to now.The concepts differ in the resulting grain structure, i.e. size and quality, but also in the preparation technologies used and the processing time needed.Today it is not clear which of the methods will succeed in industrial production.In all the methods, pin holes in the films are an issue since they lead to shunting of the final cells.
Another issue is dopand deployment, particularly along grain boundaries.This also may lead to shunting, which today limits the open circuit voltage to slightly above 500 mV.A further point is that TCO cannot easily be used as a front contact in superstrate cells since it hardly withstands the temperatures needed for crystallization.Usually a highly doped silicon layer is used instead, which, however, has somewhat low transparency.Very important for thin film crystalline solar cell is a perfect light management so that about 2 µm of silicon is enough to absorb the solar spectrum.This can be achieved either by structured substrates or by texturing the surface.In the first case, the irregular substrate surface should not influence the crystallization behaviour.In the second case, the rough surface should not increase surface recombination.Generally, passivation of defects and of the surface is a crucial preparation step.
Concerning the theoretical description of the processes involved in crystallization, the basic equations are well understood.However, there are some issues with the material parameters involved, which, particularly for amorphous silicon, strongly depend on deposition conditions and therefore need to be determined individually.But even if numerical predictions may not completely coincide with experiments due to inadequate numerical values of the materials parameters, general trends can reliably be predicted.All the mentioned issues need further investigation.Careful study of these topics is expected to lead to full exploitation the potential of the material.Multicrystalline thin film cells with a ratio of grain size over film thickness similar to multicrystalline wafer cells should deliver, if prepared correctly, comparable efficiencies.Therefore we expect the polyand multicrystalline silicon thin film solar cells to gain increasing significance and may replace microcrystalline silicon cells.Multicrystalline silicon also can act as one partner in tandem cells which would further increase the efficiency.

Acknowledgment
This work partly was funded by the European Commission under contract 213303 (HIGH-EF), and by the German state of Thuringia via Thüringer Aufbaubank under contract 2008 FE 9160 (SolLUX).We would like to thank J. Lábár and G. Sáfrán (MFA Budapest) for TEM investigations.

Fig. 1 .
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,

Fig. 2 .
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).

Fig. 3 .
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.

Fig. 5 .
Fig. 5. Transmission electron microscopic cross section image of a film epitaxially thickened by furnace annealing.

Fig. 6 .Fig. 7 .
Fig.6.Speed of the phase boundaries liquid-crystalline (lc) and amorphous liquid (al) for crystalline solidification form the melt and melting of a-Si, respectively.

Fig. 10 .
Fig. 10.Time lag of nucleation for crystallization from the melt for a fixed value of interfacial energy of 400 mJ/m²

Fig. 12 .
Fig. 12.Average grain size after solid phase crystallization of amorphous silicon as depending on temperature