Growth of Multicrystalline Silicon for Solar Cells: The High-Performance Casting Method

  • C. W. LanEmail author
Living reference work entry

Latest version View entry history


The emergence of high-performance multicrystalline silicon (HP mc-Si) in 2011 has made a significant impact to photovoltaic (PV) industry. In addition to the much better ingot uniformity and production yield, HP mc-Si also has better material quality for solar cells. As a result, the average efficiency of solar cells made from HP mc-Si in production increased from 16.6% in 2011 to 18.5% or beyond in 2016. With an advanced cell structure, an average efficiency of more than 20% has also been reported. More importantly, the efficiency distribution became much narrower; the difference even from various wafer producers became smaller as well. Unlike the conventional way of having large grains and electrically inactive twin boundaries, the crystal growth of HP mc-Si by directional solidification is initiated from uniform small grains having a high fraction of random grain boundaries (GBs). The grains developed from such grain structures significantly relax thermal stress and suppress the massive generation and propagation of dislocation clusters. The gettering efficacy of HP mc-Si is also superior to the conventional one, which also increases solar cell efficiency. Nowadays, most of commercial mc-Si is grown by this approach, which could be implemented by either seeded with silicon particles or controlled nucleation, e.g., through nucleation agent coating. The future improvement of this technology is also discussed in this chapter.


A2: High-performance B2: Multicrystalline Si A1: Directional solidification A1: Casting A1: Dislocation cluster A1: Grain boundary A1: Lifetime A1: Gettering 


In 2015, the annual PV production was about 57 GW, and the solar cells made from mc-Si shared the production of 68% (Fraunhofer Institute for Solar Energy Systems 2016). The mc-Si has been grown by the directional solidification (DS) or casting since late 1970s due to its high throughput and low cost (Lan et al. 2015; Khattak and Schmid 1987). Although DS is a matured technology, as discussed in the previous chapters, due to the higher structure defects and impurities, the ingot quality of mc-Si has been much inferior to the dislocation-free single-crystalline Si (sc-Si) grown by the Cz method. To mimic sc-Si, over the past 30 years, tremendous effort has been focused on the growth of large grains with more electrically inactive GBs, especially the Σ3 twin boundaries. Among these efforts, the mono-like (Stoddard 2007; Stoddard et al. 2008) and dendritic casting techniques (Fujiwara et al. 2006; Li et al. 2011, 2012; Nakajima et al. 2010a, b; Wang et al. 2009; Yeh et al. 2010) are the most typical ones, and their schematics are shown in Fig. 1a, b, respectively. For the mono-like techniques, due to the limited size of the seeds from the Cz ingot, the splitting seeds are usually used as illustrated in Fig. 1a. The seed orientation is usually in (100) for the ease of alkaline texturing during solar cell production; the textures on the silicon wafer enhance light trapping, which is crucial for solar cell efficiency. On the other hand, instead of using the seeds, the dendritic casting technique, as illustrated in Fig. 1b, is to use the growth habit of silicon facetted dendrites to initiate a dendritic layer by high undercooling. Because the dendritic growth in the lateral direction at high undercooling (>10 K) is preferred in <110>, the grains in the ingot growth direction could then be controlled in <112> (Fujiwara et al. 2006; Nakajima et al. 2010a, b). Both approaches are effective in the growth of large grains having fewer GBs, and the GBs are mainly coherent Σ3 in the dendritic casting. The details could be found in other chapters. Since 2006, Lan’s group at NTU also has worked with Sino-American Silicon Products Inc. (SAS) to develop the dendrite casting method for mass production. For small ingots, they could reach a very high percentage of Σ3 GBs, up to 80%, with very high minority lifetime (Yeh et al. 2010). Even in industry-scale wafers, the twining area had rather low defects and high lifetime (Li et al. 2012; Lan 2011a). With this belief, they continued to develop the dendritic casting technique for several years until 2010; however, the progress was slow. During this period of time, most companies, including SAS, started to shift their effort to the mono-like technique, especially after the Mono2™ wafers of BP Solar appeared in the market in 2006 (Stoddard et al. 2008). Afterward, a few companies announced the success of mono-like production, such as the U-grade wafers from SAS, Virtus wafers from Renasolar, S2 wafers from GCL, and Maple wafers from JA Solar. SAS also branded its <110> mono-like wafers as E-wafers, which were found to have a better lifetime uniformity in the ingot production. Although the mono-like technology attracted much attention since 2006, it stayed in the market only for a short period of time (PV Magazine 2012). As the HP mc-Si emerged in late 2011 (Lan 2011a, b; Lan et al. 2012a, b, 2013, 2014; Stoddard et al. 2008; Wong et al. 2014; Yang et al. 2015), the mono-like wafers essentially disappeared from the market after 2012.
Fig. 1

Schematics of different DS technologies: (a) mono-like casting using splitting seeds; (b) dendritic casting; (c) HP mc-Si (Lan et al. (2016b) with permission of Elsevier)

The Emergence of HP mc-Si

Although high-quality ingots have been demonstrated for both mono-like and dendritic casting techniques, the multiplication and propagation of dislocation clusters due to thermal stress are still very difficult to control during crystal growth, especially for industrial-scale production. As a result, the solar cells fabricated from the wafers grown by both techniques have a very wide distribution in the conversion efficiency, and the low-efficiency tail in the distribution causes significant yield loss in the cell production. During the development of the dendritic casting technique, Lan’s group found that the control of undercooling was also not trivial. The thick quartz crucible wall (>30 mm) in production made the heat extraction from the crucible bottom much less effective. Moreover, the undercooling was sensitive to the silicon nitride coating as well. On the other hand, even large dendrites could be induced, massive dislocation clusters still appeared afterward due to thermal stress (Lan et al. 2012a, b; Yang et al. 2015). As the defect clusters appeared, they multiplied and propagated, so that the upper part of the ingot still had poor quality, i.e., low minority lifetime. Surprisingly, they occasionally induced small grains by controlling the undercooling, and the ingot grown from the small grains turned out to have a much better uniformity (Yang et al. 2015). The defect multiplication and propagation were significantly mitigated. For the wafers without massive dislocation clusters, they also noticed that the correlation between efficiency and grain size was small. This indicated that the GBs were not crucial to the wafer performance.

The Growth of HP mc-Si

To implement the small grain growth, at the beginning Lan’s group cut off the bottom part of the small-grain ingot and reused it as the seeds for the next run. However, the bottom red zone increased as the seed plates were reused; in addition, the seed preparation was also tedious. Accidentally, they found that using small silicon particles as the seeds for ingot growth turned out to have a much better quality for the grown ingot (Lan et al. 2014). Especially, during that period of time, low-cost granular silicon was available from the downgrade product of the fluidized-bed polysilicon. Different particle sizes were tested, but the size ranging from 2 to 5 mm was considered for production; the smaller particles could be easily contaminated due the larger specific surface area. They referred this seed layer for small grains as the incubation layer, as illustrated in Fig. 1c, and the growth of such small-grain ingots as the HP mc-Si technology. The use of nucleation agents was also found useful, and it will be discussed shortly. Using silicon particles was very robust in production in terms of ingot quality, but the melting stage required more care to keep a flat melting interface and a thin remaining seed layer. The red zone, the low-lifetime area due to the impurity diffusion from the crucible and the contaminated seeds, above the seed layer was found to be proportional to the remaining thickness of the seed layer. Surprisingly, in the HP mc-Si wafers, the percentage of noncoherent or random GBs was unprecedentedly high, more than 70%. However, the solar cell performance was significantly improved, especially the efficiency distribution was very narrow. The appearance of the conventional and HP mc-Si wafers are shown in Fig. 2a, b, respectively. As shown in Fig. 2a, the conventional one has large but nonuniform grains, and more importantly it contains many twins. On the other hand, the HP mc-Si wafer has uniform and small grains with many random GBs. Nevertheless, the grain structure is hardly visible from the appearance of the solar cell due to the isotropic acid texturing and antireflection coating, as shown in Fig. 2c. The adoption of HP mc-Si using silicon particles as the seeds was rather quick in industry. One of the major reasons was that the switch from the growth of mono-like to mc-Si ingots was straightforward; both required the control of seed melting.
Fig. 2

(a) A typical conventional multicrystalline silicon wafer (156 × 156 mm) having large grains and twin boundaries; (b) a typical high-performance multicrystalline silicon wafer having uniform small grains and many random GBs; (c) a typical multicrystalline silicon solar cell with four bus bars

The experimental results of HP mc-Si were first presented by Prof. Lan in the 5th International Workshop on Crystal Growth Technology in June of 2011 (Lan 2011a; Lan et al. 2013), and later in the 5th International Workshop on Crystalline Silicon Solar Cells held in Boston in October (Lan 2011b). The sample wafers were also sent to Solarworld for testing right after the Boston conference, and the feedback was excellent. The patent for such grain structures in the ingot and the wafer was filed in 2011 and first granted in 2014 (Lan et al. 2014); both silicon and nonsilicon particle seeds were used for ingot growth in the illustrated examples. After this finding, different approaches for getting such a grain structure have been explored and reported (Zhu et al. 2014; Wong et al. 2014a; Lan et al. 2016a, b; Zhang et al. 2016).

Figure 3a shows the comparison of HP mc-Si and conventional mc-Si. The lifetime mappings of both ingots are shown in Fig. 3a, where the HP mc-Si ingot was grown from small silicon particles (3 ~ 5 mm in size) (Lan et al. 2016b). The effect of particle size will be discussed shortly; however, the grains nucleated from the silicon particles depended on the microstructure of the seed materials, not the apparent size of the seeds. As shown, as compared with the conventional mc-Si ingot, the lifetime of HP mc-Si was very uniform; the low-lifetime areas due to the dislocation clusters were significantly reduced. Figure 3b shows the comparison of grain structures of the wafers at the top and bottom portions of the ingots. For HP mc-Si, the grains near the bottom part of the ingot were small and uniform. On the contrary, the grains were large but un-uniform in the conventional mc-Si. In general, during ingot growth the grain size increased with the increasing ingot height for both growth, but sometimes the grain size decreased in the conventional mc-Si due to the nucleation of new grains from grits, subgrains, or twining (Lan et al. 2013; Wong et al. 2014a). More importantly, as shown by the EPD mappings in Fig. 3c, in contrast to the conventional mc-Si, the high-EPD areas (EPD > 105/cm2) in HP mc-Si were much smaller and they were confined in the small grains. Because the columnar grain growth in HP mc-Si, the propagation of dislocation clusters from the bottom to the top was easily blocked by the random GBs. Furthermore, the defected grains tended to be softer, due to the plastic deformation assisted by dislocations, and they were easily overgrown or squeezed by others (Lan et al. 2012b; Yang et al. 2015). As will be discussed shortly, the large amount of random GBs seems to be quite effective in relaxing the thermal stress, presumably due to the sliding nature of the amorphous layer on the GBs (Lan et al. 2012b; Stokkan et al. 2014; Yang et al. 2015).
Fig. 3

Comparison of (a) ingot lifetime mapping; (b) grains and (c) EPD mapping at ingot bottom and top for conventional (top) and HP mc-Si; the ingot width in (a) is 780 mm and wafer size in (b) and (c) is 156 × 156 mm (Lan et al. (2016b) with permission of Elsevier)

The grain size indeed plays a key factor in the growth of the defects. Lan’s group also conducted a G5 experiment by putting several sc-Si chunks at different positions of the crucible along with the small silicon particles as the seeds, as shown in Fig. 4 (Lan et al. 2016b). The bricks were numbered by alphabets row by row as shown on the left of Fig. 4a, so that bricks C and F were next to the crucible wall and brick I was one brick away from the wall. The grain structures are shown in Fig. 4a. As shown in all cases, the big grains were gradually overgrown by the small grains from different directions; the areas of the chunk seeds are indicated by the red dashed-boxes. For silicon, it expands as it solidifies, so that the grains from the big chunks could be easily overgrown by the small grains. However, for brick F, some grains were grown from the crucible wall due to the slightly concave growth front near the wall. The resulted EPD mappings corresponding to the grain structure are shown in Fig. 4b. Interestingly, as shown, even though the grain size became uniform near the top of the ingot, the high EPD areas, inside the red dashed-boxes, seemed to correspond to the initial chunk seeds very well. The case for brick F was slightly affected by the grains grown from the wall. Because the new born grains from the wall had the lower EPD than the old grains, the high EPD area for brick F was slightly smaller than the original chunk size. The big grains from the chunk seeds being overgrown by the neighbor grains is often observed in mono-like growth. Again, it is presumed that during grain growth each grain expands as it solidifies; the more crowded columnar grains tend to grow outwards because they have more random orientations. Similarly, the weaker grains due to more dislocations for plastic deformation are overgrown by the stronger ones.
Fig. 4

(a) Grain growth from different seeds; (b) EPD mapping of the wafers from different ingot positions; the red dashed-boxes indicate the area of the chunk seeds at the crucible bottom (Lan et al. (2016b) with permission of Elsevier). The positions of bricks C, F, and I are indicated on the left of (a). The numbers on the top of the figures indicate the positions of the wafers in the bricks from the bottom to the top

The Performance of HP mc-Si

With the emergence of HP mc-Si in 2011, the solar cell efficiency has increased dramatically. As shown in Fig. 5, the biggest jump was from A+ to A3+; in fact, there was A2+ wafers in between that were the first generation of HP mc-Si using undercooling control and reused seed plates (Lan 2011a, b); the brand name for A+-series wafers was used by SAS since 2009. Interestingly, A2+ wafers appeared in the market only for a short period time due to the lower production yield. On the other hand, A3+ wafers were from the ingot grown from silicon particle seeds (Lan et al. 2012a, 2014), and this made the crystal growth much more robust both on the yield and quality. The progress to A4+ and A5+ required much better growth control including the hot zone, as well as the improvement of crucible/coating purity (Lan et al. 2016a, b; Yang et al. 2015). Nowadays, the average efficiency in production is about 18.3%, and the best could reach 18.5%, based on the back surface field (BSF) cell structure. If the passivated emitter and rear cell (PERC) structure is used, using alumina back surface passivation, an average efficiency up to 19.6% in production has been reached (Lan et al. 2016b). Some companies, such as REC and Trina, also reported an efficiency of more than 20% in their production line, while the champion cell recently reported by Jinko Solar was up to 21.63% using HP mc-Si wafers (156 × 156 mm). For these cells, reactive ion etching (RIE) was used for texturing, i.e., the so-called black silicon, instead of the traditional isotropic acidic etching. In addition, these wafers were also cut by diamond wire, so that the cutting damage was much less and the backside polishing was much easier for passivation.
Fig. 5

Progress of solar cell efficiency using HP-mc Si wafers for BSF and PERC cells; A+ is not HP mc-Si

Furthermore, from the recent investigation by Sio and Macdonald (2016), as shown in Fig. 6, the recombination activity of the GBs in HP mc-Si wafers turned out to be very low (the recombination velocity was about 200 cm/s) as compared with that in the traditional mc-Si wafers (~1000 cm/s). The reason for the low recombination activity at the GBs for HP mc-Si remains unknown and needs further investigation. Moreover, the gettering efficacy of the HP mc-Si wafers was much higher than the convectional mc-Si wafers as well (Castellanos et al. 2016). In general, the lifetime could be easily enhanced by gettering for HP mc-Si wafers, but the lifetime often deteriorated for the conventional ones. The main reason was believed to be the lower dislocation clusters to trap the metals in the HP mc-Si, so that the gettering could be more efficient. The gettering efficacy is very important to enhance the lifetime of the wafer in the emitter formation during solar cell production. The higher wafer lifetime gives the higher solar cell efficiency. Therefore, HP mc-Si has not only the better uniformity and production yield, but also the better solar cell performance owing to its better gettering efficacy and lower GB activities. Recently, Lan et al. (2016b) also grew G6 n-type HP mc-Si, and the lifetime of the wafers before and after phosphorous gettering from the center brick was shown in Fig. 7. As shown, the best lifetime could be increased to 1 ms after gettering. In fact, after hydrogenation, the best lifetime was further increased to above 2 ms (Phang et al. 2016). Although boron-gettering was found not effective, the additional phosphorous gettering and hydrogen passivation were found effective in gaining the lifetime. This indicates that N-type HP mc-Si is also a potential material for high-efficiency solar cells based on the lifetime results (Phang et al. 2016). Recently, Schubert et al. (2016) further reported that the simulated solar cell efficiency was up to 22% for this material.
Fig. 6

Variation of surface recombination activity of p-type mc-Si, p-type HP mc-Si, and n-type mc-Si wafers after gettering, hydrogenation, and both treatments (Sio and Macdonald (2016) with permission of Elsevier)

Fig. 7

Maximum minority lifetime distribution of n-type HP mc-Si wafers from a G6 ingot (P/Ga co-doped) before and after P-gettering (Lan et al. (2016b) with permission of Elsevier)

Growth of HP mc-Si Without Seeding

In addition to the use of silicon particles for seeding, using nucleation agent for getting small grains could be useful as well (Lan et al. 2016a, b; Wong et al. 2014b; Zhang et al. 2016). A few nucleation coatings have been considered including silica, silicon nitride, and their mixtures with silicon particles (Lan et al. 2016a), and some have been adopted in production. Figure 8a shows an example of using a Si3N4/Si nucleation agent; the agent was paint on the left part of a Si3N4-coated quartz crucible. The grain structures grown from the coating are shown in Fig. 8b. As shown, uniform small grains could be early nucleated from the nucleation agent. Because the thickness of the nucleation agent coating was only 500 μ, the thermal condition in the crucible bottom for different coatings should be similar. The roughness due to the voids occupied by silicon might be key factor for the nucleation of small grains. Recently, Kupka et al. (2016) used SiO2 and SiC particles with different particle sizes for the nucleation agents. They also had a similar observation that the grains nucleated from the coating decreased with the increasing roughness. Therefore, the small and uniform grains could be induced by the nucleation agent; however, the portion of the random GBs usually was not as high as that from the silicon particle seeding. Figure 9 shows the comparison of the columnar growth and the lifetime mapping of a brick in a G5 growth obtained by from two different seeding approaches; silica (0.1 mm in size) was used as the nucleation agent for Fig. 9b. As shown, the columnar grains from both seeding methods were very similar. Both methods had good uniformity in the lifetime mappings. The major difference was in the thickness of the bottom red zone. Using the nucleation agent significantly reduced the red zone thickness. As a result, the growth yield was increased. Nevertheless, as compared with the silicon seeding, the initial percentage of non-Σ or random GBs (~50%) of the ingot was found lower for using the silica nucleation agent. Kupka et al. (2016) also observed similar results, especially for SiC particles. They observed that although small grains could be initiated from SiC particles, the fraction of random GBs was not high; still many Σ3 GBs appeared during nucleation. On the other hand, the silica nucleation layer gave the grain structures much closer to HP mc-Si, i.e., small grains with a high fraction of random GBs. In addition, the silica particles used previously were presumed to be in the β-cristobalite phase during nucleation.
Fig. 8

An example of nucleation agent for grain control: (a) the additional Si3N4/Si coating on the left bottom of the Si3N4-coated quartz crucible; (b) the grain structures; the dashed-line indicates the boundary of the nucleation agent coating

Fig. 9

Grains and lifetime time of the ingots from (a) seeded growth; (b) the silica nucleation agent; the thinner bottom red zone in (b) could be seen (Lan et al. (2016b) with permission of Elsevier)

Moreover, increasing temperature gradients was found easier to get smaller grains from the nucleation; however, the fraction of Σ3 GBs depended on the coatings. Magnetic stirring could also be used, e.g., by REC Inc. in Singapore. Nowadays, the quality and thus the solar cell performance of the ingot grown from the nucleation agent were still found slightly inferior to that from the silicon particles, even through several companies claimed that they could grow high quality HP mc-Si without using silicon particle seeds. Therefore, this indicated that the amount of “random GBs” would be more crucial for stress relaxation and thus defect reduction, instead of just the small grains. This observation was further confirmed by Reimann et al. in their recent experiments (2016). They conducted small ingot growth using different seeding materials. As shown in Fig. 10, the longer random GB length fraction from the seeds led to the smaller area of high EPD clusters in the grown ingot. More importantly, the seed materials played a crucial role. For example, if the seeds were from Siemen’s feedstock or fluidized-bed-reactor, which was mc-Si with very small grains, the seed size had little effect on the grown grain size and the length fraction of random GBs (Reimann et al. 2016). In other words, to initial the grain structures for HP mc-Si, the grain structures were more important than the apparent size for the seed materials.
Fig. 10

Area fraction of EPD > 105 cm−2 versus the random grain boundary length fraction for all performed experiments for the 25 mm cut; seeds were from single crystalline crushed (SCC), fluidized-bed-reactor (FBR) and Siemens (SIE) feedstock; the conventional mc-Si was included for comparison (Reimann et al. (2016) with permission of Elsevier)

The Properties of HP mc-Si

Figure 11 summaries the typical grain structures (a), crystallographic orientations and GB types (c) of the wafers taken from different positions of a HP mc-Si ingot seeded by silicon particles (Lan et al. 2016c). The statistics in Figs. 11b and 6c was obtained from Fig. 11a. As shown in Fig. 11b, the grain orientations were mainly located in the low-energy planes, such as (111) and (112). Other orientations such as and (115) and (313) had nontrivial portions and this might be generated due to twining (Wong et al. 2014a). More importantly, as shown in Fig. 11c, the percentage of the non-Σ or random GBs at the lower part of the ingot was greater than 70%. Even in the top portion of the ingot, the percentage of the random GBs was also more than 60%.
Fig. 11

Development of (a) grain structures; (b) crystallographic orientations; and (c) GB types along the height of an HP mc-Si ingot. The orientations and GB types were obtained from the wafers at different heights of the central brick of the ingot (Lan et al. (2016b) with permission of Elsevier)

Wong et al. (2014a) did a detailed analysis of the grain structures developed from small silicon beads (0.9 mm in diameter), and they also found that the lowest-energy orientation, i.e., (111) tended to dominate during grain competition. However, the twining from the tri-junctions generated new grains with different orientations (Wong et al. 2014a); this might be the reason for more (112) orientation at the top ingot. Indeed, the (111)- or (112)-dominated orientations and the high fraction of random GBs are the typical characteristics nowadays in commercial HP mc-Si wafers. The high-percentage of the random GBs shown in Fig. 11c upsets the previous understanding for high quality mc-Si wafers; however, they played a crucial role in the reduction of dislocation clusters, especially for industrial production. Before Lan’s group reported this finding (Lan et al. 2012a, b, 2014; Yang et al. 2015), most people believed that twins or Σ3 GBs were needed for better lifetime (Fujiwara et al. 2006; Li et al. 2011, 2012; Nakajima et al. 2010a, b; Wang et al. 2009; Yeh et al. 2010), which was the core concept of the dendritic casting approach. In fact, if the wafers from the dendritic casting are carefully examined, one could find that the twin areas often have very few defects (Li et al. 2012; Ryningen et al. 2011); however, this is based on the condition without massive dislocation clusters. During the growth of large ingot, the relaxation of thermal stress for reducing the multiplication of dislocation clusters is crucial, and the large amount of random GBs in HP mc-Si plays a critical role.

Outlook for HP mc-Si

Although the growth of HP mc-Si having low defects and good uniformity is rather robust, especially seeded with silicon particles, the random GBs decrease and defects increase during crystal growth. The defect growth is strongly affected by the evolution of grain structures, and the grain growth, which leads to the reduction of random GBs, is very important. Lehmann et al. (2016) analyzed HP and conventional mc-Si wafers from various companies, as shown in Fig. 12. They found that indeed the GB types were very different as shown in Fig. 12a. The HP mc-Si in general had more than 60% of random GBs, while the conventional one had high percentage of Σ3 GBs. However, as shown in Fig. 12b, their difference became smaller near the top of the ingots due to grain growth and the decrease of random GBs, as well as the increase of Σ3 GBs. As a result, if the ingot grows taller, there are less random GBs for stress relaxation, so that the dislocation clusters even in HP mc-Si will become more problematic.
Fig. 12

Comparison of GB types in HP and conventional mc-Si at the bottom (a) and the top (b) of the ingots from different suppliers (Lehmann et al. (2016) with permission of Elsevier)

In fact, Wong et al. (2014a) was the first to perform detailed experiments to investigate the structure evolution of HP mc-Si in a lab-scale furnace for different growth speeds ranging from 10 to 200 mm/h. They analyzed the grain structures of the wafers cut from the ingot by using electron backscattered diffraction (EBSD). For grain growth, some of their results are shown in Fig. 13a for illustration. In general, the low-energy (111) grain became dominant during grain coarsening due to the reduction of overall interfacial energy. The random GBs decreased as grains coarsened. On the other hand, as shown in Fig. 13b, they also found that Σ3 GBs increased due to twining (Wong et al. 2014a; Duffar and Nadri 2010) and decreased due to the blocking by the random GBs. As shown, the new grain nucleated from the tri-junction, and this twining process could repeat based on the twining probability (Duffar and Nadri 2010). Lin and Lan (2017) recently found that the GB type at the faceted groove, i.e., tri-junction, between grains played a crucial role in reducing the nucleation barrier of twining. With a random GB, the twining probability could be greater than 10−6 at an undercooling of 0.5 K, which gave a twin spacing about 300 μm. The reduction of random GBs and the generation of twin boundaries eventually tended to be close. As a result, near the top of the small ingots, the fraction of both GBs became comparable. Nevertheless, because twining became more frequent with the increasing speed or undercooling, the generation of Σ3 GBs and the reduction of random GBs became faster as the growth speed increased. Therefore, the control of grain growth and twining is crucial to HP mc-Si for taller ingots. Nowadays, even for the normal ingot height, e.g., 35 cm in G6, the high defect density near the top of the ingot for HP mc-Si is still problematic due to the reduction of random GBs. In fact, if one carefully examines the amount of random GBs near the top ingots in (Lehmann et al. 2016), it is clear that some ingots could still keep a high percentage of random GBs. This indicates that the growth conditions could be optimized for HP mc-Si in practice. Recently, Trempa et al. (2017) compared the grain structures and defects of the conventional and HP mc-Si ingots from eight successive growths to simulate the extraordinary ingot height of 710 mm. They found that the grain structures and the defects of both ingots became similar after 350 mm. Especially, as shown in Fig. 14 for the comparison of defects and recombination area fraction for convectional and HP mc-Si in their study (Trempa et al. 2017). In other words, the advantage of HP mc-Si is limited to the first grown 350 mm of the ingot.
Fig. 13

(a) Grain coarsening (left figures are EBSD and right figure are GBs); (b) nucleation of twins and their movement; twin boundary was blocked by the random GBs (from wafers 4 to 6). The twin nucleation from the faceted groove at tri-junction is illustrated on the right (Wong et al. (2014a) with permission of Elsevier)

Fig. 14

Recombination active area fraction versus total ingot height for the 710 mm G1 conventional ingot (red circles) and the 710 mm G1 HP mc-Si (HPM) ingot (green squares). Additionally, the values for a 300 mm industrial HPM ingot (violet triangles) are shown. On the sides, PL images of both 710 mm G1 ingots are shown for 80 mm and 710 mm total ingot height (Trempa et al. (2017) with permission of Elsevier)

In addition to the dislocation clusters, the control of impurities is also important to ingot quality. The back diffusion of the metals from the silicon seeds increases the red zone and deteriorates the quality of bottom ingot, even though the EPD is the lowest there. Therefore, to further improve HP mc-Si, the reduction of seed layer thickness and the improvement of crucible/coating purity would be important. Recently, the crucibles coated with high purity silica are available in the market; using a diffusion barrier could also be feasible (Hsieh et al. 2014). Eventually, if the seeding layer is replaced by simple coating, while keeping small grains and high fraction of random GBs after nucleation, both quality and yield of HP mc-Si could be improved.


Nowadays, nearly 70% of solar cells are made from mc-Si wafers, and most of them are produced from the HP casting method. The method is very robust, so that it has been widely adopted by industry. In addition to the excellent ingot quality, the yield of the HP casting is also very high due to the much less massive propagation of dislocation clusters. The excellent quality of the HP mc-Si has been reflected on the significant progress of the solar cell efficiency in the recent years. Besides the p-type champion cell with an efficiency higher than 21.23% reported by Trina Solar Inc., the very recent world record (21.9%) on the n-type HP mc-Si solar cell has been made by Fraunhofer ISE in early 2007. Nevertheless, to compete with the monocrystalline silicon solar cells, further improvement of the ingot quality by HP casting can be expected.




CWL is grateful for the generous support by the Ministry of Science and Technology of Taiwan, National Taiwan University, and SAS.


  1. S. Castellanos, K.E. Ekstrøm, A. Autruffe, M.A. Jensen, A.E. Morishige, J. Hofstetter, P. Yen, B. Lai, G. Stokkan, C. del Canizo, T. Buonassisi, IEEE J. Photovolt. 6, 632 (2016)CrossRefGoogle Scholar
  2. T. Duffar, A. Nadri, Scr. Mater. 62, 955 (2010)CrossRefGoogle Scholar
  3. Fraunhofer Institute for Solar Energy Systems, ISE, Freiburg, Photovoltaics Report. Accessed 17 Nov 2016
  4. K. Fujiwara, W. Pan, K. Sawada, M. Tokairin, N. Usami, Y. Nose, A. Nomura, T. Shishido, K. Nakajima, J. Cryst. Growth 292, 282 (2006)CrossRefGoogle Scholar
  5. C.C. Hsieh, A. Lan, C. Hsu, C.W. Lan, J. Cryst. Growth 401, 727 (2014)CrossRefGoogle Scholar
  6. C.P. Khattak, F. Schmid, in Silicon Processing for Photovoltaics II, ed. by C. P. Khattak, K. V. Ravi (Elsevier Science Publishers, North Holland, 1987), p. 153Google Scholar
  7. I. Kupka, C. Reimann, T. Lehmann, D. Oriwol, F. Kropfgans, J. Friedrich, in Abstracts of Technical Digest of the 18th International Conference on Crystal Growth and Epitaxy, Nogoya, 7–12 Aug 2016Google Scholar
  8. C.W. Lan, in Abstracts of the 5th International Workshop on Crystal Growth Technology, Berlin, 26–30 June 2011aGoogle Scholar
  9. C. W. Lan, in Abstracts of the 5th International Workshop on Crystalline Silicon Solar Cells, Boston, 1–3 Nov 2011bGoogle Scholar
  10. C.W. Lan, W.C. Lan, T.F. Li, A. Yu, Y.M. Yang, C. Hsu, B. Hsu, A. Yang, From A+ to A+++, in Abstracts of the 22nd Workshop on Crystalline Silicon Solar Cells & Modules: Materials and Processes, Vail, 22–25 July 2012aGoogle Scholar
  11. C.W. Lan, Y.M. Yang, A. Yu, B. Hsu, A. Yang, in Abstracts of the 27th European Photovoltaic Solar Energy Conference (27th EU PVSEC), Frankfurt, 24–28 Sept 2012bGoogle Scholar
  12. C.W. Lan, W.C. Lan, T.F. Li, A. Yu, Y.M. Yang, C. Hsu, B. Hsu, A. Yang, J. Cryst. Growth 360, 68 (2013)CrossRefGoogle Scholar
  13. C.W. Lan, W.H. Yu, Y.M. Yang, H.S. Chou, C.L. Hsu, W.C. Hsu, TW Patent I452185(B), 11 Sept 2014Google Scholar
  14. C.W. Lan, C. Chuck, K. Nakajima, in Handbook of Crystal Growth 2A: Bulk Crystal Growth: Basic Techniques, 2nd edn., ed. by P. Rudolph (Elsevier Science Publishers, Amsterdam, 2015), p. 373Google Scholar
  15. C.W. Lan, Y.M. Yang, A. Yu, Y.C. Wu, B. Hsu, W.C. Hsu, A. Yang, Solid State Phenom. 242, 21 (2016a)CrossRefGoogle Scholar
  16. C.W. Lan, A. Lan, C.F. Yang, H.P. Hsu, M. Yang, A. Yu, B. Hsu, W.C. Hsu, A. Yang, J. Cryst. Growth, in press (2016b), Accessed 28 Oct 2016
  17. C.W. Lan, C.F. Yang, A. Lan, M. Yang, A. Yu, H.P. Hsu, B. Hsu, C. Hsu, Cryst. Eng. Comm. 18, 1474 (2016c)CrossRefGoogle Scholar
  18. T. Lehmann, C. Reimann, E. Meissner, J. Friedrich, Acta Mater. 106, 98 (2016)CrossRefGoogle Scholar
  19. T.F. Li, K.M. Yeh, W.C. Hsu, C.W. Lan, J. Cryst. Growth 318, 219 (2011)CrossRefGoogle Scholar
  20. T.F. Li, H.C. Huang, H.W. Tsai, A. Lan, C. Hsu, C.W. Lan, J. Cryst. Growth 340, 202 (2012)CrossRefGoogle Scholar
  21. H.K. Lin, C.W. Lan, Acta Mater. 131, 1 (2017)Google Scholar
  22. P.V. Magazine, Sep 2012, pp. 94–99.
  23. K. Nakajima, K. Kutsukake, K. Fujiwara, N. Usami, S. Ono, I. Yamasaki, in Abstracts of the 35th IEEE Photovoltaic Specialists Conference, Honolulu, 20–25 June 2010aGoogle Scholar
  24. K. Nakajima, K. Kutsukake, K. Fujiwara, N. Usami, S. Ono, I. Yamasaki, in Abstracts of the 25th European Photovoltaic Solar Energy Conference and Exhibition (25th EU PVSEC), The 5th World Conference on Photovoltaic Energy Conversion (WCPEC-5), Spain, 6–9 Sep 2010bGoogle Scholar
  25. S. P. Phang, H. C. Sio, C.F. Yang, C.W. Lan, Y.M. Yang, A. Yu, B. Hsu, C. Hsu, D. Macdonald, in Abstracts of the 26th Photovoltaics Science and Engineering Conference (PVSEC-26), Singapore, 24–28 Oct 2016Google Scholar
  26. C. Reimann, M. Trempa, T. Lehmann, K. Rosshirt, J. Stenzenberger, J. Friedrich, K. Hesse, E. Dornberger, J. Cryst. Growth 434, 88 (2016)CrossRefGoogle Scholar
  27. B. Ryningen, G. Stokkan, M. Kivambe, T. Ervik, O. Lohne, Acta Mater. 59, 7703 (2011)CrossRefGoogle Scholar
  28. M. Schubert, F. Schindler, J. Schön, W. Kwapil, B. Michl, J. Benick, in Abstracts of the 9th International Workshop on Crystalline Silicon Solar Cells, Tempe, Arizona, 10–12 Oct 2016Google Scholar
  29. H.C. Sio, D. Macdonald, Sol. Energ. Mat. Sol. Cells 144, 339 (2016)CrossRefGoogle Scholar
  30. N. Stoddard, BP Corp., WO Patent 2007084936, A2 July 2007Google Scholar
  31. N. Stoddard, B. Wu, I. Witting, M. Wagener, Y. Park, G. Rozgonyi, R. Clark, Solid State Phenom. 1, 131 (2008)Google Scholar
  32. G. Stokkan, Y. Hu, O. Mjos, M. Juel, Sol. Energ. Mat. Sol. Cells 130, 679 (2014)CrossRefGoogle Scholar
  33. M. Trempa, I. Kupka, C. Kranert, T. Lehmann, C. Reimann, J. Friedrich, J. Cryst. Growth 459, 67–75 (2017)CrossRefGoogle Scholar
  34. T.Y. Wang, S.L. Hsu, C.C. Fei, K.M. Yei, W.C. Hsu, C.W. Lan, J. Cryst. Growth 311, 263 (2009)CrossRefGoogle Scholar
  35. Y.T. Wong, C. Hsu, C.W. Lan, J. Cryst. Growth 387, 59 (2014a)CrossRefGoogle Scholar
  36. Y.T. Wong, C.T. Hsieh, A. Lan, C. Hsu, C.W. Lan, J. Cryst. Growth 404, 59 (2014b)CrossRefGoogle Scholar
  37. Y.M. Yang, A. Yu, B. Hsu, W.C. Hsu, A. Yang, C.W. Lan, Prog. Photovolt. Res. Appl. 23, 340 (2015)CrossRefGoogle Scholar
  38. K.M. Yeh, C.K. Hseih, W.C. Hsu, C.W. Lan, Prog. Photovolt. Res. Appl. 18, 265 (2010)Google Scholar
  39. H. Zhang, D. You, C. Huang, Y. Wu, Y. Xu, P. Wu, J. Cryst. Growth 435, 91 (2016)CrossRefGoogle Scholar
  40. D. Zhu, L. Ming, M. Huang, Z. Zhang, X. Huang, J. Cryst. Growth 386, 52 (2014)CrossRefGoogle Scholar

Copyright information

© Springer-Verlag GmbH Germany, part of Springer Nature 2019

Authors and Affiliations

  1. 1.Department of Chemical EngineeringNational Taiwan University (NTU)TaipeiTaiwan

Section editors and affiliations

  • Kazuo Nakajima
    • 1
  1. 1.Tohoku University, Institute for Materials ResearchTohoku University; FUTURE-PV InnovationFukushimaJapan

Personalised recommendations