- Email: [email protected]
Evolution of grain structure and recombination active dislocations in extraordinary tall conventional and high performance multi-crystalline silicon ingots
Journal of Crystal Growth 459 (2017) 67–75
Contents lists available at ScienceDirect
Journal of Crystal Growth journal homepage: www.elsevier.com/locate/jcrysgro
Evolution of grain structure and recombination active dislocations in extraordinary tall conventional and high performance multi-crystalline silicon ingots
crossmark
⁎
M. Trempaa, , I. Kupkab, C. Kranertb, T. Lehmannb,1, C. Reimanna,b, J. Friedricha,b a b
Fraunhofer IISB, Schottkystr. 10, 91058 Erlangen, Germany Fraunhofer THM, Am St.-Niclas-Schacht 13, 09599 Freiberg, Germany
A R T I C L E I N F O
A BS T RAC T
Communicated by P. Rudolph
In this work one high performance multi-crystalline silicon ingot and one conventional multi-crystalline silicon ingot, each with an extraordinary ingot height of 710 mm, were replicated by the successive growth of eight G1 ingots to evaluate the potential advantage of extraordinary tall HPM ingots in industrial production. By analyzing different grain structure parameters like mean grain size, grain orientation and grain boundary type distribution as well as the recombination active dislocation area over the complete ingot height, it was observed that the material properties strongly differ in the initial state of growth for the two material types. However, at ingot heights above 350 mm, the difference has vanished and the grain structure properties for both materials appear similar. It is shown that the evolution of the grain structure in both material types can be explained by the same grain selection and grain boundary generation/annihilation mechanisms whereas the current grain structure determines which mechanisms are the most dominant at a specific ingot height. Since the grain structure directly influences the dislocation content in the silicon material, also the recombination active dislocation area becomes equal in high performance and conventional multi-crystalline silicon material at ingot heights above 350 mm. From these results it is concluded that the advantage of high performance silicon material is limited to the first grown 350 mm of the ingot.
Keywords: A1.Directional solidification A1.High performance multi-crystalline silicon A1.Grain structure A1.Grain boundaries A1.Recombination active defects
1. Introduction Nowadays, directionally solidified multi-crystalline (mc) silicon is one of the most used materials for the production of silicon solar cells. For a few years, the conventional mc-silicon with the silicon melt solidifying directly at the Si3N4 releasing coating at the crucible bottom, is being replaced more and more by the so called “high performance mc-silicon” or HPM-silicon which was firstly reported by Sino-American Silicon Productions Inc. (SAS) in 2011 [1,2]. The HPM material, which is produced either by nucleation on a feedstock particle layer [2–6] or more recently on a structured crucible bottom/functional coating [7–9], is characterized by a very fine initial grain structure (mean grain size < 4 mm2), a homogeneous orientation distribution (small coefficient of variation CVGO < 1.5, explanation see Section 2) and a high length fraction of random grain boundaries > 60% (see e.g. [10]). These grain structure properties lead to a significantly smaller amount of recombination active dislocations and therefore result in ~0.5% absolute higher solar cell efficiencies [2,10] in
⁎
1
comparison to the conventional mc-silicon which is characterized by larger grains containing many dendrites and twins. Recently, it was observed by Lehmann et al. [10], who investigated the grain structure properties of several industrially grown HPM and conventional mc-Si ingots, that the difference in grain structure properties and also the amount of recombination active defects between the two material types was not as significant at the top region as it initially occurs at the bottom region of the ingots. From this observation, it seems that the advantage of the HPM silicon mainly occurs in the lower parts of the ingot where the difference of the HPM grain structure to the conventional mc grain structure is largest. Currently, silicon ingots produced in industrial scale exhibit an ingot height of about 250–400 mm leading to the above described benefit of HPM silicon. In order to clarify if there is some advantage of the growth of industrial HPM ingots with an ingot height > 400 mm within the present work, the evolution of grain structure properties and the recombination active dislocation area in extraordinary tall silicon ingots was investigated. For that purpose one HPM silicon ingot and
Correspondence to: Fraunhofer Institute IISB, Schottkystr. 10, 91058 Erlangen, Germany E-mail address: [email protected] (M. Trempa). New address: Dr. Toni Lehmann, Siltronic AG, Berthelsdorfer Str. 113, 09599 Freiberg, Germany.
http://dx.doi.org/10.1016/j.jcrysgro.2016.11.030 Received 13 September 2016; Received in revised form 3 November 2016; Accepted 4 November 2016 Available online 11 November 2016 0022-0248/ © 2016 Elsevier B.V. All rights reserved.
Journal of Crystal Growth 459 (2017) 67–75
M. Trempa et al.
pattern. Analysis of the Laue pattern is done automatically by the system. As a result, the grain orientation perpendicular to the wafer surface, which nearly equals the growth direction, is obtained. Furthermore, the grain boundary type is determined if the grain orientation of two neighboring grains was measured. The measured grain orientations were simplified to hkl values up to 20. In the following, the results are depicted as circles drawn into inverse pole figures (IPF). The center of a circle represents the grain orientation itself and its diameter is proportional to the area fraction of that grain orientation. The grain orientation distribution is also described by using the coefficient of variation CVgrain orientation (CVGO). In this case, the CVGO is defined as the standard deviation of the area fractions of the grain orientations divided by the mean value. After measuring the grain orientation, a polishing etch containing HF:HNO3 was applied on the same wafers and subsequently band to band photoluminescence imaging (PLI) was performed to investigate the area fraction of recombination active dislocations. The PLI measurements were done with the OPTECTION imaging tool [13] using a 175 W laser (wavelength λ 790 nm) and a 40 s exposure time. The area fraction of recombination active dislocations was determined by post-processing and analysis of the resulting image.
Fig. 1. Sketch of the procedure of growing extraordinary high conventional and HPM silicon ingots with a total ingot height of 710 mm by successively growing of eight G1 ingots. In the case of the 1st grown conventional G1 silicon ingot, there is no initial seeding layer.
one conventional mc-silicon ingot, each with an extraordinary ingot height of 710 mm, were replicated by the successive growth of eight G1 ingots. 2. Experimental setup and characterization A crystallization furnace, which allows to grow G1 silicon ingots with dimensions of 220×220×130 mm3 and a weight of 15 kg, was used for the experiments [11]. For all experiments, standard fused silica crucibles with a standard Si3N4 release coating and polysilicon feedstock from Wacker were applied. To obtain silicon ingots with a “total” ingot height of 710 mm, successive growth of eight G1 silicon ingots was carried out. The relating procedure is sketched in Fig. 1. For the first HPM as well as for the first conventional G1 ingot the crucible was filled completely with the silicon feedstock. In the case of the HPM ingot a 20 mm seeding layer containing only small silicon chips was placed on the crucible bottom. During the crystallization process 10 mm of this seeding layer was kept unmelted in order to achieve a very fine HPM grain structure by solidifying the silicon melt directly on the feedstock particles. In the case of the conventional ingot, the feedstock was completely melted resulting in a more coarse and dendritic grain structure by solidifying on the standard Si3N4 coating at the crucible bottom. Besides the small difference during the melting phase, the crystallization processes for both the HPM and the conventional silicon ingot were quite similar with a growth rate of about 1 cm/ h and an almost flat solid-liquid interface shape during the whole growth process. After growing the first ingots, 20 mm thick slices were cut from 80 mm to 100 mm grown ingot length and used as seed plates for the second crystallization experiments. This step was repeated six times until eight silicon ingots were grown corresponding to a “total” growth length of 710 mm designated as “total ingot height” in the following text. In terms of characterization, three horizontal wafers with an area of 156×156 mm2 and a thickness of 2 mm were prepared from the center of each ingot at 10 mm, 45 mm and 80 mm grown ingot length. After mechanical preparation of the wafer surface, the grains of the wafers were automatically detected in full wafer scale by an optical system (GEMINI tool from Intego [12]) which is based on reflectivity. Twenty images of the wafer surface are taken by illuminating the wafer by LED modules placed at different angles to the wafer surface. Subsequently, the position and the size of each grain are determined by an automatic image processing step. The minimum detectable grain size for the present investigations was around 0.075 mm2. A description of the grain size distribution was done by using the so-called coefficient of variation CVgrain size (CVGS) which is defined by the ratio of the standard deviation of the mean grain size and the mean grain size itself. This value, often used in statistics, reflects the homogeneity of the grain sizes means the smaller the CV value, the more homogeneous is the grain size distribution. The crystallographic orientation for grains larger than 3 mm2 was determined in a second step by using a Laue scanner [12]. This system irradiates each grain with white x-rays and detects the resulting Laue
3. Results First, the results gained from the grain structure analysis are presented and discussed. Fig. 2 shows the mean grain size as well as the coefficient of variation CVGS over the total ingot height for the 710 mm G1 HPM ingot and the 710 mm G1 conventional ingot, respectively. As expected, the mean grain size at the ingot bottom is significantly larger for the conventional ingot than for the HPM ingot. With increasing ingot height the mean grain size increases in both ingots. Additionally, the difference between the two curves becomes smaller and vanishes within the margin of error at about 150 mm. Over the further ingot height up to 710 mm, the mean grain size remains constant for both ingot types. This can also be seen from the grain structure images obtained by the GEMINI tool, which are shown in Fig. 3. While there is a large difference in the initial grain structure of both ingot types at 10 mm grown ingot length, the grain structures becomes more and more similar with increasing ingot height and no significant difference can be observed by eye at 710 mm. In the case of the coefficient of variation CVGS a similar trend is observed, but even more clearly. Again, the CVGS values are distinctly different at the ingot bottom for both material types. While the conventional ingot exhibits a high initial CVGS=6.6 which decreases with increasing ingot height, the HPM ingot has a small initial CVGS=1.5 slightly increasing with increasing ingot height. After about 330 mm ingot height, both level at CVGS≈3 and exhibit an almost constant trend up to the ingot top at 710 mm. Again, this is comprehensible if we look on the grain structure images in Fig. 3. The initial grain size distribution is quite inhomogeneous in the case of the conventional ingot due to several dendrites and twins. Oppositely, the initial HPM structure exclusively consists of very small and nearly isometric grains without any twins or dendrites. With increasing ingot height the grain size distribution for both ingot types looks more and more similar. For comparison, the correspondent values for an industrially grown HPM ingot with an ingot height of about 300 mm were added to the curves in Fig. 2. The curves for the 300 mm industrial HPM ingot correlate well to the values of the first 300 mm of the successively grown 710 mm ingot. Only in the second ingot half, the mean grain size is slightly higher for the industrial ingot. This good agreement validates that the used method of the successive growing of several 130 mm tall ingots yields highly comparable results with respect to growth within one process as it is done in industry. This allows us the transfer of the results from our experiments to extraordinary high industrial ingots. 68
Journal of Crystal Growth 459 (2017) 67–75
M. Trempa et al.
G1 HPM
8
8
7
7
6
6
5
5
CVGS
mean grain size [mm²]
G1 conventional
4
industrial HPM
4
3
3
2
2
1
1 0
0 0
100
200
300
400
500
600
0
700
100
200
300
400
500
600
700
total ingot height [mm]
total ingot height [mm]
Fig. 2. Mean grain size (left) and coefficient of variation CVGS (right) vs. total ingot height for the 710 mm G1 conventional ingot (red circles) and the 710 mm G1 HPM ingot (green squares). Additionally the values for a 300 mm industrial HPM ingot (violet triangles) are shown. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
were added, too. Again this grain structure parameter matches the values of the successively grown HPM ingot in our laboratory underlining the validity of our experiments. To have a closer look on the initial grain orientation distribution and its evolution over the ingot height, all measured grains per wafer were assigned to nine defined grain orientation sectors as shown in Fig. 5. For that purpose all exactly measured grain orientations hkl were assigned to this orientation which exhibits the lowest deviation angle compared to the exact hkl value. The selection of the nine orientations was done by considering the most popular orientations ((001), (111), (101) and (112)) and simultaneously keeping the section areas as similar as possible. The area fractions of these nine orientations at 10 mm ingot height are shown in Fig. 6, while their evolution over the complete ingot height is shown in Fig. 7. For comparison the ideal orientation distribution obtained from Monte Carlo integration over all possible orientations on the Ewald sphere was added in Fig. 6 as shaded columns. Thereby, ideal distribution means that the area fraction of each of the nine selected orientations on the wafer corresponds to their surface fraction on the Ewald sphere. It has to be noted that these fractions are not equally distributed and are not exactly identical to the
Beside the grain size and its distribution, the grain orientation was determined by Laue method on full wafer scale as described in Section 2. The results over the ingot height are plotted in Fig. 4 in terms of the coefficient of variation CVGO. Again, the formerly observed trends occur. At the ingot bottom, the conventional ingot exhibits a high CVGO value of about 4.5. This correlates to a very inhomogeneous orientation distribution which is also illustrated by the inverse pole figure IPF where the large circles at the (112) and (110) regions indicate a large area fraction of these predominant orientations. In contrast the HPM ingot reveals a very homogeneous orientation distribution (CVGO=0.8) which means that there is no dominant orientation existent corresponding to the absence of large circles in the IPF. With increasing ingot height, the same trend as observed for the CVGS values in Fig. 2 appears. While the CVGO value drastically decreases for the conventional ingot, the CVGO value for the HPM ingot becomes slightly larger with increasing ingot height. Both curves align to each other with CVGO=1.3 at about 270–350 mm ingot height. For taller ingot heights, the CVGO values remain constant for both ingot types. Consequently, the IPF figures at 710 mm ingot height on the right of Fig. 4 appear similar. In Fig. 4 the corresponding values for the industrial HPM ingot
10mm
80mm
170mm
710mm
Conv. multi
HPM
Fig. 3. Grain structure images obtained by the GEMINI tool for the G1 conventional ingot (upper row) and the G1 HPM ingot (lower row) at different ingot heights.
69
Journal of Crystal Growth 459 (2017) 67–75
M. Trempa et al.
Fig. 4. Coefficient of variation CVGO vs. total ingot height for the 710 mm G1 conventional ingot (red circles) and the 710 mm G1 HPM ingot (green squares) as well as for a 300 mm industrial HPM ingot (violet triangles). Additionally, the IPFs for both G1 ingots at 10 mm and 710 mm total ingot height are shown. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
(111)
((112), (111) and (115)), while others decrease ((315), (313) and (105)) over ingot height. Additionally, some orientations remain nearly constant ((001), (101) and (102)). Further, it is remarkable that the surface fractions of all grain orientations become similar in the two ingot types at an ingot height of about 350 mm and do not change significantly until 710 mm ingot height. This indicates that the grain orientation distribution reaches a stable state after a particular ingot height. The grain boundary type distribution over the complete wafer area is another important grain structure parameter to investigate. For this purpose the grain boundaries were determined from the orientation matrices of neighboring grains obtained from the orientation measurements. In Fig. 8 the grain boundary length fraction is plotted versus the grown ingot height. Again, there is a large difference in the grain boundary distribution at the bottom of the two ingots. The HPM ingot (left plot in Fig. 8) exhibits a low Σ3 grain boundary length fraction of 15% and a very high random grain boundary length fraction of 70%, whereas the conventional ingot (right image in Fig. 8) reveals an opposite trend meaning a high Σ3 grain boundary length fraction of 55% and a very low random grain boundary length fraction of 15%. The higher value decreases and the lower value increases with increasing ingot height for both ingots as indicated by the arrows in Fig. 8. Above a particular total ingot height, the grain boundary length fraction values for the Σ3 and the random grain boundaries remain constant at
(112)
(313)
(115) (315) (001)
(101)
(102)
(105)
Fig. 5. Orientation triangle with the assignment of the exact hkl orientation to nine orientation sectors.
area fractions in the IPF shown in Fig. 5. The diagrams in Fig. 6 confirm the statement which was given from the initial CVGO values in Fig. 4. The distribution at the bottom of the conventional ingot is much more inhomogeneous than in the HPM ingot, mainly due to the large surface fraction of the (112) orientation. On the other side the distribution of the HPM ingot approximates the ideal distribution. By following the trends over ingot height in Fig. 7 it is obvious that the individual orientation fractions evolve differently along the ingot height, but this evolution does not depend on the ingot type. In particular, the fraction of some grain orientations increases
29%
25
23%
20 15%
15 12%
5,6%
11%
10
2,5% 1,5%
25
20%
20 16% 16%
15
8,8% 13% 10% 9,5%
10 4,3%
5
5
1,9%
0,35%
0
HPM
30
Conventional area fraction [%]
area fraction [%]
30
0
(001) (115) (112) (111) (313) (101) (102) (105) (315)
(001) (115) (112) (111) (313) (101) (102) (105) (315)
simplified grain orientation
simplified grain orientation
Fig. 6. Area fraction of distinct grain orientations in the G1 conventional ingot (left) and the G1 HPM ingot (right) at 10 mm total ingot height. Additionally the ideal homogeneous distribution is shown as dashed columns.
70
Journal of Crystal Growth 459 (2017) 67–75
M. Trempa et al.
HPM
30
30
25
25
area fraction [%]
area fraction [%]
Conventional
20 15 10
20 15 10
5
5
0
0
0
100
200
300
400
500
600
(001) (115) (112) (111) (313) (101) (102) (105) (315)
700
0
100
200
300
400
500
600
700
total ingot height [mm]
total ingot height [mm]
Fig. 7. Evolution of area fractions of distinct grain orientations for the 710 mm G1 conventional ingot (left) and the 710 mm G1 HPM ingot.
increases rapidly up to 27% at 170 mm ingot height, while the area fraction in the HPM ingot increases only moderately up to 13% at 440 mm. This is also apparent from the PL images taken from 80 mm ingot height on the left hand side in Fig. 9. In the case of the conventional ingot, a higher amount of large dark areas indicating low PL-intensity is visible. At taller ingot heights, the area fraction decreases to about 13% within the conventional ingot, while it approaches this value from below for the HPM ingot. This reveals that also the recombination active area fraction in both ingot types align to each other at a specific ingot height which is about 350–440 mm in our experiments. Further, it was repeatedly observed that the values for the industrial HPM ingot match quite well those of the 710 mm HPM ingot grown in our laboratory. In reflection of all experimental results shown so far, it is striking that all grain structure parameters of both ingot types are distinctly different at the ingot bottom. During the further growth, these parameters (mean grain size, CVGS, CVGO) change in a way that they approximate each other and become more or less equal at specific ingot heights. This ingot height is different for the different parameters reaching from 150 mm for the mean grain size up to 330 mm for the CVGS and 270–350 mm for the CVGO values. Additionally, the grain
approximately 30% and 50%, respectively. It is noted that the ingot height at which the values become constant is different for the HPM ingot (80–170 mm) and the conventional ingot (170−260 mm). The length fraction of the other grain boundary types, namely Σ9+Σ27 and Σothers, is low ( < 10%) within the HPM ingot. In the case of the conventional ingot, the Σ9+Σ27 grain boundaries exhibit a similar trend as the Σ3 grain boundaries meaning a decrease over the first 230 mm ingot height followed by constant values up to 710 mm. However, the values are at a significant lower level than for the Σ3 grain boundaries. Additionally, the grain boundary fractions of the industrial HPM ingot were added to Fig. 8. As for the other values discussed above, the grain boundary length fraction values of the industrially grown HPM ingot match well those of the successively grown 710 mm HPM ingot grown in our laboratory. Besides the grain structure properties, the area fraction of recombination active dislocations was determined by means of PL imaging. The results are shown in Fig. 9. Close to the ingot bottom, the conventional ingot exhibits only a slightly higher recombination active area fraction (3%) than the HPM ingot (1%). During further growth, the recombination active area fraction for the conventional ingot
grain boundary length fraction [%]
grain boundary length fraction [%]
80
Conventional
70 60 50 40 30 20 10 0 0
100
200
300
400
500
600
Σ9 + Σ27
Σ3
Random (R)
700
Other Σ
80
HPM
70 60 50 40 30 20 10 0 0
100
200
300
400
500
600
700
total ingot height [mm]
total ingot height [mm]
Fig. 8. Grain boundary length fraction vs. total ingot height for the 710 mm G1 conventional ingot (left) and the 710 mm G1 HPM ingot (right). Additionally the values for a 300 mm industrial HPM ingot (unfilled symbols) are shown in the right image.
71
Journal of Crystal Growth 459 (2017) 67–75
M. Trempa et al.
G1 HPM
recombination active area fraction [%]
156 mm
G1 conventional
industrial HPM
35 30 25 20 15 10 5 0 0
100
200
300
400
500
600
700
total ingot height [mm]
Fig. 9. Recombination active area fraction vs. total ingot height for the 710 mm G1 conventional ingot (red circles) and the 710 mm G1 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. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
silicon under moderate growth rates of ~ 10 mm/h, is the grain selection reducing the surface energy at the growth interface [14]. This means that the area fraction of grains with higher surface energy decreases and the area fraction of grains with lower surface energy increases. This mechanism, which was already investigated by other groups in small scaled HPM material [4,15], occurs in our experiments, too. The trends for both ingot types in Fig. 7 show that the surface fraction of the high energetic (101) orientation decreases from 4% to 0.2% in the HPM ingot and remains constant at a low level of about 1% in the conventional ingot. Also the high energetic (100) orientation is on a constant low level of smaller than 2% in both ingots. A more extensive decrease from 20–23% to 11–12% could be observed in both ingots for the (315) orientation which is assumed to exhibit a relatively high surface energy, too. Oppositely, the surface fraction of the low energetic (111) orientation increases from 11–13% to 20–23% in both ingots. The dominance of the (112) orientation occurring in both ingots was similarly observed in former works in conventional mc-silicon [16] as well as in HPM silicon [4,15] on laboratory scale. Also in industrial HPM ingots a predominance of the (112) orientation was reported in [2] where the authors proposed that the smallest interfacial energy of (112) next to (111) could be the reason. Additionally, the 29% fraction of the (112) orientation at the bottom of the conventional ingot can be correlated to large dendrites which typically occur with this grain orientation [17]. During further growth, first the (112) fraction slightly decreases due to the overgrowth by other grains (compare grain structure image in 80 mm ingot height in Fig. 3), followed by an increase due to the energy-driven grain selection. Beside the grain selection due to the energy minimization of the surface energy, a further change in the orientation distribution can be induced by a twin formation at {111} growth facets frequently occurring during growth of mc-silicon [18,19]. After generation of every new twin boundary, new grain orientations appear in dependence of the original grain orientation(s). This can explain the increase of the (115) surface fraction from 12% to 20% in the conventional ingot because the (115) orientation is the Σ3-twin orientation of a (111) grain [20] which is available in a high amount of about 20% area fraction in the upper ingot half. Even if the observed trends of some orientations can be explained, the trends for the other measured grain orientations, namely (102), (105) and (313) in both ingots could not be directly correlated to special growth mechanisms as described above. Probably these orientations are just the crystallographic result from twin generation starting from non-(111) orientations, from other grain interactions or from
boundary type distributions of both ingot types change over the ingot height in a way that a constant type distribution (roughly 50% random, 30% Σ3, 15% Σ9+Σ27) is reached exceeding a specific ingot height, whereas the height value differs from the HPM ingot (80–170 mm) to the conventional ingot (170–260 mm). Also the recombination active areas within both ingots - after an initial spread along the first 170 mm - become more and more equal at longer ingot heights and finally match each other at 350–440 mm. 4. Discussion The observed trends of all investigated grain structure parameters point out that the initially formed grain structures tend to a stable state during the solidification process which is reached at a specific ingot height. Further, the initial grain structure appears to be irrelevant for the properties of the stable state, but relevant for the height at which this state is reached. In the following, the observed trends for the grain orientation, the grain boundary type distribution and the grain size for both ingot types are discussed by taking into account grain selection and different grain boundary annihilation/formation mechanisms. 4.1. Grain orientation Concerning the grain orientation it has to be pointed out that the observed trends in the 710 mm HPM ingot up to 250 mm ingot height (Fig. 7) are in good agreement with the trends of a 250 mm industrial HPM ingot published by Yang et al. [2] considering the different choice of plotted orientations. This is a further validation that the method of successive growth of several G1 ingots can be used as replication of tall industrially grown silicon ingots. By observation of Fig. 7 it is remarkable that the area fraction of all grain orientations becomes similar in both ingot types at an ingot height of about 350 mm which indicates that the grain orientation distribution reaches a stable state after a particular ingot height. In dependence on the initial value, the fraction of some grain orientations increases ((112), (111) and (115)), while others decrease ((315), (313) and (105)) over ingot height. If the starting value is close to the final value in the stable state the fraction remains nearly constant ((001), (101) and (102)), whereas this constant area fraction results from a simultaneous generation and annihilation of particular grain orientations due to different growth mechanisms which are discussed in the following. One mechanism, which takes place during the solidification of 72
Journal of Crystal Growth 459 (2017) 67–75
M. Trempa et al.
annihilation processes, respectively (see also Table 1). In the bottom part of the 710 mm HPM ingot an extremely high amount of R grain boundaries of about 70% exists (compare Fig. 8 on the left). Therefore, the generation mechanisms (1), (5) and the annihilation mechanism (6) should be the most dominant ones which is affirmed by the fraction values of 50% for (1), 20% for (5) and 40% for (6). Also the annihilation mechanism (7) is very prominent (40%) despite the Σ3 length fraction is only 20%. Due to the significantly lower length fraction of Σ3n grain boundaries, the generation mechanisms (3)–(5) and the annihilation mechanisms (8)–(10) are very rare at this stage of growth. Summing up the fractions for the generation and annihilation of R and Σ3 grain boundaries results in 20% R generation (5), 40% R annihilation (6), 55% Σ3-generation (1)–(4) and 45% Σ3annihilation (7)–(10). Considering the overall ratio of 45 to 55% of generation to annihilation, this indicates a strong decrease of R grain boundaries over ingot height (generation 20%*45% « annihilation 40%*55%) which typically appears in HPM ingots and also in the 710 mm HPM ingot in this work. In the case of the Σ3 grain boundaries the analogue calculation reveals a nearly constant trend over ingot height (generation 55%*45% = annihilation 45%*55%), which obviously does not match with the increasing trend typically observed in HPM material (Fig. 8 and e.g. [2,4]). But the grain boundary length fraction typically used for such diagrams is not the most suitable parameter to investigate frequencies of grain boundary formation/annihilation processes. Therefore, the absolute number of determined grain boundaries, corrected by the length ratio between determined and total grain boundaries, was plotted over ingot height as shown in Fig. 10. It is obvious from the right hand side diagram that now the expected constant trend of Σ3 grain boundaries over the ingot height arises, while the R boundary trend shows still a strong decrease at the initial state of growth as shown in Fig. 8. Due to the ongoing change in the grain boundary type distribution during the first 150 mm of HPM growth, the fractions of the different generation and annihilation mechanism change, too. This was also presented by Prakash et al. [22] who observed significant changes already after 30 mm growth of their small HPM ingot (see values in Table 1). According to the decreasing number of R grain boundaries, the fractions of the mechanisms where R grain boundaries are involved have drastically decreased from 50% to 20% (1), from 20% to 5% (5) and from 40% to 0% (6). Simultaneously, the mechanisms related to Σ3n grain boundaries occur with higher fractions, whereas the number of Σ3n grain boundaries remains nearly constant. In detail the generation mechanisms (2)–(4) increase from 5% to 50% and the annihilation mechanisms (8)(10) from 5% to 35%. Summing up the generation and annihilation of R and Σ3 grain boundaries yields 5% R generation (5), < 5% R annihilation (6), 70% Σ3-generation (1)–(4) and 65% Σ3-annihilation (7)–(10). Considering again the overall generation to annihilation ratio, this predicts a nearly constant trend for R grain boundaries (generation 5%*45% ≈ annihilation < 5%*55%) and Σ3 boundaries (generation 70%*45% ≈ annihilation 65%*55%), too. This is in good agreement with the obtained data from the 710 mm HPM ingot at heights above 170 mm shown in Fig. 8 and Fig. 10. In the following, the model introduced above is adapted to conventional mc silicon. For that purpose the fractions for each grain boundary mechanism (1)–(10) were estimated according to the values experimentally determined by Prakash et al. for the HPM material given in Table 1. Due to the almost identical trends for the R and Σ3 grain boundaries observed in the upper ingot regions (stable state) of the 710 mm ingots, the same fraction values as in the HPM ingot were assumed for the conventional ingot. For the lower part of the conventional ingot it was assumed that the fractions of mechanisms (1), (5), (6) and (7) should be relatively smaller, hence the amount of R grain boundaries is smaller in comparison to the stable state region. In
Table 1 Overview of experimentally observed grain boundary generation and annihilation mechanisms after [15,22], the fraction of theses mechanisms in HPM silicon (after [22]) and in conventional mc-silicon (estimated in this work). HPM
Conventional
No
Generation
Fraction (initial) after [22]
Fraction (stable state) after [22]
Fraction (Initial) estimated
Fraction (stable state) estimated
(1)
R→Σ3+R (R=random) Σ3→Σ3+Σ9 Σ9→Σ3+Σ3/ Σ27/R* Σ27→Σ3+Σ9/R R→R+R
50%
20%
< 20%
20%
5%
50%
> 50%
50%
20%
5%
< 5%
5%
40% 40% 5%
< 5% 30% 35%
0% < 30% > 35%
< 5% 30% 35%
(2) (3) (4) (5)
(6) (7) (8) (9) (10)
Annihilation R+R→R Σ3+R→R Σ3+Σ3→Σ9 Σ3+Σ9→Σ3/ Σ27/R* Σ3+Σ27→Σ9
R* are in fact large tilted Σ3 grain boundaries, but practically determined as R grain boundary according to the Brandon criterion [23].
grain boundary generation/annihilation processes (see next passage) which permanently appear during growth. 4.2. Grain boundary type distribution The second phenomenon to be discussed is the evolution of the grain boundary (GB) type distribution with increasing ingot height which is determined by the mechanisms of grain boundary generation and annihilation. The probably most prominent generation mechanism in mc-silicon is the already mentioned energetically driven twin formation at undercooled {111} growth facets, e.g. in existent grain boundary grooves [18,19]. According to this mechanism, new low energetic Σ3 twin boundaries were created during growth by the generation rule GB1→GB2+Σ3. The two other involved grain boundaries GB1 and GB2 can in principle be of various types [21]. Recently, this grain boundary generation process was investigated by different groups in small lab-scale HPM silicon ingots [4,22] as well as in ribbon growth mc-silicon samples [15]. They could show that several grain boundary generation mechanisms including twin formation occur in this type of material (see mechanisms (1)-(4) in Table 1). Additionally, the splitting of R grain boundaries into two R grain boundaries was experimentally observed in [22] (mechanism (5) in Table 1). On the other side, grain boundaries can annihilate by merging with other grain boundaries [4,15,22]. Also in this case, several mechanisms with different types of grain boundaries exist (see mechanisms (6)– (10) in Table 1). In principle, all mechanisms (1)–(10) will happen in HPM silicon as well as in conventional mc silicon. However, the frequency of each mechanism differs in dependence on the amount of the existent grain boundary types in the silicon material. Further, in the case of the annihilation mechanisms (6)–(10), the distance and the angle between the grain boundaries is an indicator for the probability of the merging process. Recently, Prakash et al. [22] have analyzed the frequency of the different grain boundary generation and annihilation mechanisms in small HPM silicon ingots by detailed investigation of more than 2000 grain boundaries. They could show that the overall fraction ratio of generation to annihilation processes is roughly 45% to 55% which means that the total amount of grain boundaries decreases over ingot height. Further, they determined the fraction of each observed mechanism related to the total amount of observed generation and 73
Journal of Crystal Growth 459 (2017) 67–75
M. Trempa et al.
Σ3
Σ9 + Σ27 No. of grain boundaries [arb. units]
No. of grain boundaries [arb. units]
Random (R)
Conventional
0
100
200
300
400
500
600
700
all
Other Σ
HPM
0
100
200
300
400
500
600
700
Total ingot height [mm]
Total ingot height [mm]
Fig. 10. Number of measured grain boundaries vs. total ingot height for the 710 mm G1 conventional ingot (left) and the 710 mm G1 HPM ingot (right).
the mean grain size increases more clearly. The distribution of the grain size becomes more homogeneous (CVGS value decreases), mainly due to the overgrowth of the large dendrites. During further growth the mean grain size in both material types remains constant which can be explained by the more and more balance between annihilation and formation of grain boundaries and grains, respectively.
consequence the fractions of the mechanisms (2)–(4) and (8)–(10) should become relatively larger even if the absolute number of Σ3n grain boundaries remains constant. After sum up the total fractions for R generation (5), R annihilation (6), Σ3-generation (1)–(4) and Σ3annihilation (7)–(10) are < 5%, 0%, 50–75% and 35–65%. This shows that the R generation is higher than the annihilation resulting in an increasing amount of R grain boundaries over ingot height which can be observed in the 710 mm ingot (Fig. 8 and Fig. 10). Hence the fraction of 5% for R generation is not as big, maybe some more R grain boundaries were gained from the mechanisms (3) and (9) exhibiting quite high fractions. The fractions for Σ3 generation are slightly higher than for annihilation; however they equal each other if the overall ratio of generation and annihilation of 55–45% is considered. As the result also in the conventional 710 mm ingot, the absolute amount of Σ3 grain boundaries remains constant in the stable state region above 300 mm as shown in Fig. 8 and Fig. 10. In summary it can be stated that the by Prakash et al. proposed “grain boundary model” is valid for both of our HPM and conventional mc silicon ingots with extraordinary height up to 710 mm.
4.4. Recombination active dislocations As shown in Fig. 9, the area fraction of recombination active dislocations at the bottom region of both ingots is comparatively low. In the HPM material, the dislocation movement is firstly kept on a low level due to the small grain size and the high amount of R grain boundaries which prohibit the movement of the dislocations into neighboring grains [24]. However, during further growth, the number of R grain boundaries decreases while the grain size becomes larger. Additionally, the length fraction of Σ3 grain boundaries increases according to the mechanisms described above. That allows the dislocations to propagate within the larger grains, cross over grain boundaries into neighboring grains and in consequence to spread in the ingot volume resulting in the observed increasing trend of the recombination active area. However, a wide spreading of dislocation clusters in the ingot volume is limited by the still relatively high R grain boundary length fraction > 50%. Further, dislocated grains can be overgrown by non-dislocated grains during due to the apparent grain selection. Therefore, the increase of recombination active dislocation area becomes smaller over ingot height until the final state is reached at about 350 mm where the generation and propagation of new dislocation clusters is balanced by the described diminishing processes. In the case of the conventional ingot, the initial increase of the dislocated area up to 170 mm ingot height is much more extensive in comparison to the HPM material due to the larger grain size (which also increases) and the high amount of Σ3 grain boundaries enabling the propagation in the ingot volume. Simultaneously, more and more R grain boundaries were formed as described in Section 4.2. Consequently, the dislocation propagation should be more and more inhibited with increasing ingot height. Additionally, existent dislocation clusters diminish by ending up at R grain boundaries or they are overgrown by non-dislocated grains. Both should lead to a decrease of the recombination active area as can be observed in the ingot from
4.3. Grain size and grain size distribution The observed trends for the other investigated grain structure parameters, the mean grain size as well as the CVGS (Fig. 2), are a direct consequence of the grain orientation and grain boundary evolution described in the sections above. The initial dominant annihilation processes of R grain boundaries in the HPM ingot lead to a large decrease of the total number of grain boundaries (and also grains) over the first 170 mm ingot height (Fig. 10) implicating the observed increase of mean grain size in Fig. 2. Further, in combination with the grain orientation selection this leads to a more inhomogeneous grain size distribution indicated by the increasing CVGS also shown in Fig. 2. In the case of the conventional ingot the initial mean grain size is higher due to large apparent dendrites paired with some smaller grains. While in the regions of the smaller grains a grain coarsening due to the grain boundary mechanisms and the grain selection takes place (including a slight decrease of the grain boundary number), the large dendrite areas are overgrown by newly formed grains (including a strong increase of grain boundary number). In summary the total amount of grain boundaries just slightly increases (see Fig. 10) while 74
Journal of Crystal Growth 459 (2017) 67–75
M. Trempa et al.
From the results it is concluded that the advantage concerning the grain structure parameters and the recombination active area of HPM silicon in comparison to conventional mc silicon is limited to the first 350 mm of the ingot. Therefore, it has to be evaluated by the industrial producers if the growth of extraordinary high HPM ingots higher than 350 mm is profitable.
170 mm to 350 mm (Fig. 9). The sharp transition from the increase to the decrease of the recombination active area could maybe explained by the fact that there is no further increase of the grain size at about 150 mm (see Fig. 2) which would promote the dislocation propagation while the increase of R grain boundaries is still ongoing until ~250 mm (see Fig. 8). At longer ingot heights above 350 mm, the recombination active area reaches a stable state characterized by an equality of permanent dislocation formation and diminishing, analogously to the HPM material.
Acknowledgement These research and development activities are carried out within the ENOWA-II project (0325805E) funded by the German Federal Ministry for Economic Affairs and Energy.
4.5. General remarks In summary, it could be shown that in both silicon materials, conventional and HPM, the same growth mechanisms principally occur. Which mechanism or interaction is dominant at specific ingot heights depends on the current grain structure and therefore which kind of grain orientations and grain boundaries are available. After a particular growth length – which can be larger than those actually applied in industry – a stable configuration of the grain structure concerning the mean grain size, the grain orientation and the grain boundary types is reached. It is assumed that the observed stable configuration in this work is just a local energy minimum which probably depends on the actual growth conditions like growth rate or temperature gradients. This assumption is supported by the results in [10] which show slightly different values for orientation and grain boundary type distribution at the top of several industrially grown ingots in comparison to our work.
References [1] C.W. Lan, W.C. Lan, T.F. Lee, A. Yu, Y.M. Yang, W.C. Hsu, B. Hsu, A. Yang, J. Crys. Growth 360 (2012) 68–75. [2] Y.M. Yang, A. Yu, B. Hsu, W.C. Hsu, A. Yang, C.W. Lan, Prog. Photovolt. Res. Appl. 23 (3) (2015) 340–351. [3] C. Reimann, M. Trempa, T. Lehmann, K. Rosshirt, J. Stenzenberger, J. Friedrich, K. Hesse, E. Dornberger, J. Crys. Growth 434 (2016) 88–95. [4] Y.T. Wong, C. Hsu, C.W. Lan, J. Crys. Growth 387 (2014) 10–15. [5] D. Zhu, L. Ming, M. Huang, Z. Zhang, X. Huang, J. Crys. Growth 386 (2014) 52–56. [6] R.R. Prakash, T. Sekiguchi, K. Jiptner, Y. Miyamura, J. Chen, H. Harada, K. Kakimoto, J. Crys. Growth 401 (2014) 717–719. [7] G.A. Babu, I. Takahashi, S. Matsushima, N. Usami, J. Cryst. Growth 441 (2016) 124–130. [8] Y.T. Wong, C.T. Hsieh, A. Lan, C. Hsu, C.W. Lan, J. Crys. Growth 404 (2014) 59–64. [9] H. Zhang, D. You, C. Huang, Y. Wu, Y. Xu, P. Wu, J. Crys. Growth 435 (2016) 91–97. [10] T. Lehmann, C. Reimann, E. Meissner, J. Friedrich, Acta Mater. 106 (2016) 98–105. [11] M. Beier, M. Trempa, J. Seebeck, C. Reimann, P. Wellmann, B. Gründig-Wendrock, J. Friedrich, K. Dadzis, L. Sylla, T. Richter. In: Proceedings of the 29th European Photovoltaic Solar Energy Conference: Amsterdam, 2014, p. 717–721. [12] T. Lehmann, M. Trempa, E. Meissner, M. Zschorsch, C. Reimann, J. Friedrich, Acta Mater. 69 (2014) 1–8. [13] T. Lehmann, Einfluss der Korngefüge industriell hergestellter mc-Siliziumblöcke auf die rekombinationsaktivenKristalldefekte und die Solarzelleneffizienz (Dissertation), Technische Universität Bergakademie Freiberg, Freiberg, 2016. [14] K. Fujiwara, Y. Obinata, T. Ujihara, N. Usami, G. Sazaki, K. Nakakjima, J. Crys. Growth 266 (2004) 441–448. [15] H.K. Lin, M.C. Wu, C.C. Chen, C.W. Lan, J. Cryst. Growth 439 (2016) 40–46. [16] I. Brynjulfsen, K. Fujiwara, N. Usami, L. Arnberg, J. Crys. Growth 356 (2012) 17–21. [17] K. Nagashio, K. Kuribayashi, Acta Mater. 53 (10) (2005) 3021–3029. [18] T. Duffar, A. Nadri, Scr. Mater. 62 (12) (2010) 955–960. [19] A. Nadri, Y. Duterrail-Couvat, T. Duffar, J. Crys. Growth 385 (2014) 16–21. [20] F. Wilhelm, J. Appl. Cryst. 4 (1971) 521–523. [21] V. Randle, Acta Mater. 52 (14) (2004) 4067–4081. [22] R.R. Prakash, K. Jiptner, J. Chen, Y. Miyamura, H. Harada, T. Sekiguchi, Appl. Phys. Express 8 (3) (2015) 35502. [23] D.G. Brandon, Acta Metall. 14 (11) (1966) 1479–1484. [24] G. Stokkan, Y. Hu, Ø. Mjøs, M. Juel, Sol. Energ. Mat. Sol. C. 130 (2014) 679–685.
5. Conclusion In this work, an extraordinary high conventional mc silicon ingot as well as HPM silicon ingot was replicated by the successive growth of eight G1 ingots. The results show that all studied grain structure parameters like mean grain size, grain orientation and grain boundary type distribution are distinctly different in the initial state of growth for both material types. However, with increasing ingot height the differences becomes smaller and the grain structures for both materials become very similar at long ingot heights above 350 mm. This behavior was found to result from grain selection and grain boundary generation/annihilation mechanisms which are driven by energy reduction or simply by crystallographic relationships and happen in both material types in a similar manner. Further, it was shown that the grain structure evolution over ingot height directly influences the dislocation propagation within the silicon ingots leading also to a comparable recombination active dislocation area in the conventional and the HPM silicon material at long ingot heights.
75
Contents lists available at ScienceDirect
Journal of Crystal Growth journal homepage: www.elsevier.com/locate/jcrysgro
Evolution of grain structure and recombination active dislocations in extraordinary tall conventional and high performance multi-crystalline silicon ingots
crossmark
⁎
M. Trempaa, , I. Kupkab, C. Kranertb, T. Lehmannb,1, C. Reimanna,b, J. Friedricha,b a b
Fraunhofer IISB, Schottkystr. 10, 91058 Erlangen, Germany Fraunhofer THM, Am St.-Niclas-Schacht 13, 09599 Freiberg, Germany
A R T I C L E I N F O
A BS T RAC T
Communicated by P. Rudolph
In this work one high performance multi-crystalline silicon ingot and one conventional multi-crystalline silicon ingot, each with an extraordinary ingot height of 710 mm, were replicated by the successive growth of eight G1 ingots to evaluate the potential advantage of extraordinary tall HPM ingots in industrial production. By analyzing different grain structure parameters like mean grain size, grain orientation and grain boundary type distribution as well as the recombination active dislocation area over the complete ingot height, it was observed that the material properties strongly differ in the initial state of growth for the two material types. However, at ingot heights above 350 mm, the difference has vanished and the grain structure properties for both materials appear similar. It is shown that the evolution of the grain structure in both material types can be explained by the same grain selection and grain boundary generation/annihilation mechanisms whereas the current grain structure determines which mechanisms are the most dominant at a specific ingot height. Since the grain structure directly influences the dislocation content in the silicon material, also the recombination active dislocation area becomes equal in high performance and conventional multi-crystalline silicon material at ingot heights above 350 mm. From these results it is concluded that the advantage of high performance silicon material is limited to the first grown 350 mm of the ingot.
Keywords: A1.Directional solidification A1.High performance multi-crystalline silicon A1.Grain structure A1.Grain boundaries A1.Recombination active defects
1. Introduction Nowadays, directionally solidified multi-crystalline (mc) silicon is one of the most used materials for the production of silicon solar cells. For a few years, the conventional mc-silicon with the silicon melt solidifying directly at the Si3N4 releasing coating at the crucible bottom, is being replaced more and more by the so called “high performance mc-silicon” or HPM-silicon which was firstly reported by Sino-American Silicon Productions Inc. (SAS) in 2011 [1,2]. The HPM material, which is produced either by nucleation on a feedstock particle layer [2–6] or more recently on a structured crucible bottom/functional coating [7–9], is characterized by a very fine initial grain structure (mean grain size < 4 mm2), a homogeneous orientation distribution (small coefficient of variation CVGO < 1.5, explanation see Section 2) and a high length fraction of random grain boundaries > 60% (see e.g. [10]). These grain structure properties lead to a significantly smaller amount of recombination active dislocations and therefore result in ~0.5% absolute higher solar cell efficiencies [2,10] in
⁎
1
comparison to the conventional mc-silicon which is characterized by larger grains containing many dendrites and twins. Recently, it was observed by Lehmann et al. [10], who investigated the grain structure properties of several industrially grown HPM and conventional mc-Si ingots, that the difference in grain structure properties and also the amount of recombination active defects between the two material types was not as significant at the top region as it initially occurs at the bottom region of the ingots. From this observation, it seems that the advantage of the HPM silicon mainly occurs in the lower parts of the ingot where the difference of the HPM grain structure to the conventional mc grain structure is largest. Currently, silicon ingots produced in industrial scale exhibit an ingot height of about 250–400 mm leading to the above described benefit of HPM silicon. In order to clarify if there is some advantage of the growth of industrial HPM ingots with an ingot height > 400 mm within the present work, the evolution of grain structure properties and the recombination active dislocation area in extraordinary tall silicon ingots was investigated. For that purpose one HPM silicon ingot and
Correspondence to: Fraunhofer Institute IISB, Schottkystr. 10, 91058 Erlangen, Germany E-mail address: [email protected] (M. Trempa). New address: Dr. Toni Lehmann, Siltronic AG, Berthelsdorfer Str. 113, 09599 Freiberg, Germany.
http://dx.doi.org/10.1016/j.jcrysgro.2016.11.030 Received 13 September 2016; Received in revised form 3 November 2016; Accepted 4 November 2016 Available online 11 November 2016 0022-0248/ © 2016 Elsevier B.V. All rights reserved.
Journal of Crystal Growth 459 (2017) 67–75
M. Trempa et al.
pattern. Analysis of the Laue pattern is done automatically by the system. As a result, the grain orientation perpendicular to the wafer surface, which nearly equals the growth direction, is obtained. Furthermore, the grain boundary type is determined if the grain orientation of two neighboring grains was measured. The measured grain orientations were simplified to hkl values up to 20. In the following, the results are depicted as circles drawn into inverse pole figures (IPF). The center of a circle represents the grain orientation itself and its diameter is proportional to the area fraction of that grain orientation. The grain orientation distribution is also described by using the coefficient of variation CVgrain orientation (CVGO). In this case, the CVGO is defined as the standard deviation of the area fractions of the grain orientations divided by the mean value. After measuring the grain orientation, a polishing etch containing HF:HNO3 was applied on the same wafers and subsequently band to band photoluminescence imaging (PLI) was performed to investigate the area fraction of recombination active dislocations. The PLI measurements were done with the OPTECTION imaging tool [13] using a 175 W laser (wavelength λ 790 nm) and a 40 s exposure time. The area fraction of recombination active dislocations was determined by post-processing and analysis of the resulting image.
Fig. 1. Sketch of the procedure of growing extraordinary high conventional and HPM silicon ingots with a total ingot height of 710 mm by successively growing of eight G1 ingots. In the case of the 1st grown conventional G1 silicon ingot, there is no initial seeding layer.
one conventional mc-silicon ingot, each with an extraordinary ingot height of 710 mm, were replicated by the successive growth of eight G1 ingots. 2. Experimental setup and characterization A crystallization furnace, which allows to grow G1 silicon ingots with dimensions of 220×220×130 mm3 and a weight of 15 kg, was used for the experiments [11]. For all experiments, standard fused silica crucibles with a standard Si3N4 release coating and polysilicon feedstock from Wacker were applied. To obtain silicon ingots with a “total” ingot height of 710 mm, successive growth of eight G1 silicon ingots was carried out. The relating procedure is sketched in Fig. 1. For the first HPM as well as for the first conventional G1 ingot the crucible was filled completely with the silicon feedstock. In the case of the HPM ingot a 20 mm seeding layer containing only small silicon chips was placed on the crucible bottom. During the crystallization process 10 mm of this seeding layer was kept unmelted in order to achieve a very fine HPM grain structure by solidifying the silicon melt directly on the feedstock particles. In the case of the conventional ingot, the feedstock was completely melted resulting in a more coarse and dendritic grain structure by solidifying on the standard Si3N4 coating at the crucible bottom. Besides the small difference during the melting phase, the crystallization processes for both the HPM and the conventional silicon ingot were quite similar with a growth rate of about 1 cm/ h and an almost flat solid-liquid interface shape during the whole growth process. After growing the first ingots, 20 mm thick slices were cut from 80 mm to 100 mm grown ingot length and used as seed plates for the second crystallization experiments. This step was repeated six times until eight silicon ingots were grown corresponding to a “total” growth length of 710 mm designated as “total ingot height” in the following text. In terms of characterization, three horizontal wafers with an area of 156×156 mm2 and a thickness of 2 mm were prepared from the center of each ingot at 10 mm, 45 mm and 80 mm grown ingot length. After mechanical preparation of the wafer surface, the grains of the wafers were automatically detected in full wafer scale by an optical system (GEMINI tool from Intego [12]) which is based on reflectivity. Twenty images of the wafer surface are taken by illuminating the wafer by LED modules placed at different angles to the wafer surface. Subsequently, the position and the size of each grain are determined by an automatic image processing step. The minimum detectable grain size for the present investigations was around 0.075 mm2. A description of the grain size distribution was done by using the so-called coefficient of variation CVgrain size (CVGS) which is defined by the ratio of the standard deviation of the mean grain size and the mean grain size itself. This value, often used in statistics, reflects the homogeneity of the grain sizes means the smaller the CV value, the more homogeneous is the grain size distribution. The crystallographic orientation for grains larger than 3 mm2 was determined in a second step by using a Laue scanner [12]. This system irradiates each grain with white x-rays and detects the resulting Laue
3. Results First, the results gained from the grain structure analysis are presented and discussed. Fig. 2 shows the mean grain size as well as the coefficient of variation CVGS over the total ingot height for the 710 mm G1 HPM ingot and the 710 mm G1 conventional ingot, respectively. As expected, the mean grain size at the ingot bottom is significantly larger for the conventional ingot than for the HPM ingot. With increasing ingot height the mean grain size increases in both ingots. Additionally, the difference between the two curves becomes smaller and vanishes within the margin of error at about 150 mm. Over the further ingot height up to 710 mm, the mean grain size remains constant for both ingot types. This can also be seen from the grain structure images obtained by the GEMINI tool, which are shown in Fig. 3. While there is a large difference in the initial grain structure of both ingot types at 10 mm grown ingot length, the grain structures becomes more and more similar with increasing ingot height and no significant difference can be observed by eye at 710 mm. In the case of the coefficient of variation CVGS a similar trend is observed, but even more clearly. Again, the CVGS values are distinctly different at the ingot bottom for both material types. While the conventional ingot exhibits a high initial CVGS=6.6 which decreases with increasing ingot height, the HPM ingot has a small initial CVGS=1.5 slightly increasing with increasing ingot height. After about 330 mm ingot height, both level at CVGS≈3 and exhibit an almost constant trend up to the ingot top at 710 mm. Again, this is comprehensible if we look on the grain structure images in Fig. 3. The initial grain size distribution is quite inhomogeneous in the case of the conventional ingot due to several dendrites and twins. Oppositely, the initial HPM structure exclusively consists of very small and nearly isometric grains without any twins or dendrites. With increasing ingot height the grain size distribution for both ingot types looks more and more similar. For comparison, the correspondent values for an industrially grown HPM ingot with an ingot height of about 300 mm were added to the curves in Fig. 2. The curves for the 300 mm industrial HPM ingot correlate well to the values of the first 300 mm of the successively grown 710 mm ingot. Only in the second ingot half, the mean grain size is slightly higher for the industrial ingot. This good agreement validates that the used method of the successive growing of several 130 mm tall ingots yields highly comparable results with respect to growth within one process as it is done in industry. This allows us the transfer of the results from our experiments to extraordinary high industrial ingots. 68
Journal of Crystal Growth 459 (2017) 67–75
M. Trempa et al.
G1 HPM
8
8
7
7
6
6
5
5
CVGS
mean grain size [mm²]
G1 conventional
4
industrial HPM
4
3
3
2
2
1
1 0
0 0
100
200
300
400
500
600
0
700
100
200
300
400
500
600
700
total ingot height [mm]
total ingot height [mm]
Fig. 2. Mean grain size (left) and coefficient of variation CVGS (right) vs. total ingot height for the 710 mm G1 conventional ingot (red circles) and the 710 mm G1 HPM ingot (green squares). Additionally the values for a 300 mm industrial HPM ingot (violet triangles) are shown. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
were added, too. Again this grain structure parameter matches the values of the successively grown HPM ingot in our laboratory underlining the validity of our experiments. To have a closer look on the initial grain orientation distribution and its evolution over the ingot height, all measured grains per wafer were assigned to nine defined grain orientation sectors as shown in Fig. 5. For that purpose all exactly measured grain orientations hkl were assigned to this orientation which exhibits the lowest deviation angle compared to the exact hkl value. The selection of the nine orientations was done by considering the most popular orientations ((001), (111), (101) and (112)) and simultaneously keeping the section areas as similar as possible. The area fractions of these nine orientations at 10 mm ingot height are shown in Fig. 6, while their evolution over the complete ingot height is shown in Fig. 7. For comparison the ideal orientation distribution obtained from Monte Carlo integration over all possible orientations on the Ewald sphere was added in Fig. 6 as shaded columns. Thereby, ideal distribution means that the area fraction of each of the nine selected orientations on the wafer corresponds to their surface fraction on the Ewald sphere. It has to be noted that these fractions are not equally distributed and are not exactly identical to the
Beside the grain size and its distribution, the grain orientation was determined by Laue method on full wafer scale as described in Section 2. The results over the ingot height are plotted in Fig. 4 in terms of the coefficient of variation CVGO. Again, the formerly observed trends occur. At the ingot bottom, the conventional ingot exhibits a high CVGO value of about 4.5. This correlates to a very inhomogeneous orientation distribution which is also illustrated by the inverse pole figure IPF where the large circles at the (112) and (110) regions indicate a large area fraction of these predominant orientations. In contrast the HPM ingot reveals a very homogeneous orientation distribution (CVGO=0.8) which means that there is no dominant orientation existent corresponding to the absence of large circles in the IPF. With increasing ingot height, the same trend as observed for the CVGS values in Fig. 2 appears. While the CVGO value drastically decreases for the conventional ingot, the CVGO value for the HPM ingot becomes slightly larger with increasing ingot height. Both curves align to each other with CVGO=1.3 at about 270–350 mm ingot height. For taller ingot heights, the CVGO values remain constant for both ingot types. Consequently, the IPF figures at 710 mm ingot height on the right of Fig. 4 appear similar. In Fig. 4 the corresponding values for the industrial HPM ingot
10mm
80mm
170mm
710mm
Conv. multi
HPM
Fig. 3. Grain structure images obtained by the GEMINI tool for the G1 conventional ingot (upper row) and the G1 HPM ingot (lower row) at different ingot heights.
69
Journal of Crystal Growth 459 (2017) 67–75
M. Trempa et al.
Fig. 4. Coefficient of variation CVGO vs. total ingot height for the 710 mm G1 conventional ingot (red circles) and the 710 mm G1 HPM ingot (green squares) as well as for a 300 mm industrial HPM ingot (violet triangles). Additionally, the IPFs for both G1 ingots at 10 mm and 710 mm total ingot height are shown. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
(111)
((112), (111) and (115)), while others decrease ((315), (313) and (105)) over ingot height. Additionally, some orientations remain nearly constant ((001), (101) and (102)). Further, it is remarkable that the surface fractions of all grain orientations become similar in the two ingot types at an ingot height of about 350 mm and do not change significantly until 710 mm ingot height. This indicates that the grain orientation distribution reaches a stable state after a particular ingot height. The grain boundary type distribution over the complete wafer area is another important grain structure parameter to investigate. For this purpose the grain boundaries were determined from the orientation matrices of neighboring grains obtained from the orientation measurements. In Fig. 8 the grain boundary length fraction is plotted versus the grown ingot height. Again, there is a large difference in the grain boundary distribution at the bottom of the two ingots. The HPM ingot (left plot in Fig. 8) exhibits a low Σ3 grain boundary length fraction of 15% and a very high random grain boundary length fraction of 70%, whereas the conventional ingot (right image in Fig. 8) reveals an opposite trend meaning a high Σ3 grain boundary length fraction of 55% and a very low random grain boundary length fraction of 15%. The higher value decreases and the lower value increases with increasing ingot height for both ingots as indicated by the arrows in Fig. 8. Above a particular total ingot height, the grain boundary length fraction values for the Σ3 and the random grain boundaries remain constant at
(112)
(313)
(115) (315) (001)
(101)
(102)
(105)
Fig. 5. Orientation triangle with the assignment of the exact hkl orientation to nine orientation sectors.
area fractions in the IPF shown in Fig. 5. The diagrams in Fig. 6 confirm the statement which was given from the initial CVGO values in Fig. 4. The distribution at the bottom of the conventional ingot is much more inhomogeneous than in the HPM ingot, mainly due to the large surface fraction of the (112) orientation. On the other side the distribution of the HPM ingot approximates the ideal distribution. By following the trends over ingot height in Fig. 7 it is obvious that the individual orientation fractions evolve differently along the ingot height, but this evolution does not depend on the ingot type. In particular, the fraction of some grain orientations increases
29%
25
23%
20 15%
15 12%
5,6%
11%
10
2,5% 1,5%
25
20%
20 16% 16%
15
8,8% 13% 10% 9,5%
10 4,3%
5
5
1,9%
0,35%
0
HPM
30
Conventional area fraction [%]
area fraction [%]
30
0
(001) (115) (112) (111) (313) (101) (102) (105) (315)
(001) (115) (112) (111) (313) (101) (102) (105) (315)
simplified grain orientation
simplified grain orientation
Fig. 6. Area fraction of distinct grain orientations in the G1 conventional ingot (left) and the G1 HPM ingot (right) at 10 mm total ingot height. Additionally the ideal homogeneous distribution is shown as dashed columns.
70
Journal of Crystal Growth 459 (2017) 67–75
M. Trempa et al.
HPM
30
30
25
25
area fraction [%]
area fraction [%]
Conventional
20 15 10
20 15 10
5
5
0
0
0
100
200
300
400
500
600
(001) (115) (112) (111) (313) (101) (102) (105) (315)
700
0
100
200
300
400
500
600
700
total ingot height [mm]
total ingot height [mm]
Fig. 7. Evolution of area fractions of distinct grain orientations for the 710 mm G1 conventional ingot (left) and the 710 mm G1 HPM ingot.
increases rapidly up to 27% at 170 mm ingot height, while the area fraction in the HPM ingot increases only moderately up to 13% at 440 mm. This is also apparent from the PL images taken from 80 mm ingot height on the left hand side in Fig. 9. In the case of the conventional ingot, a higher amount of large dark areas indicating low PL-intensity is visible. At taller ingot heights, the area fraction decreases to about 13% within the conventional ingot, while it approaches this value from below for the HPM ingot. This reveals that also the recombination active area fraction in both ingot types align to each other at a specific ingot height which is about 350–440 mm in our experiments. Further, it was repeatedly observed that the values for the industrial HPM ingot match quite well those of the 710 mm HPM ingot grown in our laboratory. In reflection of all experimental results shown so far, it is striking that all grain structure parameters of both ingot types are distinctly different at the ingot bottom. During the further growth, these parameters (mean grain size, CVGS, CVGO) change in a way that they approximate each other and become more or less equal at specific ingot heights. This ingot height is different for the different parameters reaching from 150 mm for the mean grain size up to 330 mm for the CVGS and 270–350 mm for the CVGO values. Additionally, the grain
approximately 30% and 50%, respectively. It is noted that the ingot height at which the values become constant is different for the HPM ingot (80–170 mm) and the conventional ingot (170−260 mm). The length fraction of the other grain boundary types, namely Σ9+Σ27 and Σothers, is low ( < 10%) within the HPM ingot. In the case of the conventional ingot, the Σ9+Σ27 grain boundaries exhibit a similar trend as the Σ3 grain boundaries meaning a decrease over the first 230 mm ingot height followed by constant values up to 710 mm. However, the values are at a significant lower level than for the Σ3 grain boundaries. Additionally, the grain boundary fractions of the industrial HPM ingot were added to Fig. 8. As for the other values discussed above, the grain boundary length fraction values of the industrially grown HPM ingot match well those of the successively grown 710 mm HPM ingot grown in our laboratory. Besides the grain structure properties, the area fraction of recombination active dislocations was determined by means of PL imaging. The results are shown in Fig. 9. Close to the ingot bottom, the conventional ingot exhibits only a slightly higher recombination active area fraction (3%) than the HPM ingot (1%). During further growth, the recombination active area fraction for the conventional ingot
grain boundary length fraction [%]
grain boundary length fraction [%]
80
Conventional
70 60 50 40 30 20 10 0 0
100
200
300
400
500
600
Σ9 + Σ27
Σ3
Random (R)
700
Other Σ
80
HPM
70 60 50 40 30 20 10 0 0
100
200
300
400
500
600
700
total ingot height [mm]
total ingot height [mm]
Fig. 8. Grain boundary length fraction vs. total ingot height for the 710 mm G1 conventional ingot (left) and the 710 mm G1 HPM ingot (right). Additionally the values for a 300 mm industrial HPM ingot (unfilled symbols) are shown in the right image.
71
Journal of Crystal Growth 459 (2017) 67–75
M. Trempa et al.
G1 HPM
recombination active area fraction [%]
156 mm
G1 conventional
industrial HPM
35 30 25 20 15 10 5 0 0
100
200
300
400
500
600
700
total ingot height [mm]
Fig. 9. Recombination active area fraction vs. total ingot height for the 710 mm G1 conventional ingot (red circles) and the 710 mm G1 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. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
silicon under moderate growth rates of ~ 10 mm/h, is the grain selection reducing the surface energy at the growth interface [14]. This means that the area fraction of grains with higher surface energy decreases and the area fraction of grains with lower surface energy increases. This mechanism, which was already investigated by other groups in small scaled HPM material [4,15], occurs in our experiments, too. The trends for both ingot types in Fig. 7 show that the surface fraction of the high energetic (101) orientation decreases from 4% to 0.2% in the HPM ingot and remains constant at a low level of about 1% in the conventional ingot. Also the high energetic (100) orientation is on a constant low level of smaller than 2% in both ingots. A more extensive decrease from 20–23% to 11–12% could be observed in both ingots for the (315) orientation which is assumed to exhibit a relatively high surface energy, too. Oppositely, the surface fraction of the low energetic (111) orientation increases from 11–13% to 20–23% in both ingots. The dominance of the (112) orientation occurring in both ingots was similarly observed in former works in conventional mc-silicon [16] as well as in HPM silicon [4,15] on laboratory scale. Also in industrial HPM ingots a predominance of the (112) orientation was reported in [2] where the authors proposed that the smallest interfacial energy of (112) next to (111) could be the reason. Additionally, the 29% fraction of the (112) orientation at the bottom of the conventional ingot can be correlated to large dendrites which typically occur with this grain orientation [17]. During further growth, first the (112) fraction slightly decreases due to the overgrowth by other grains (compare grain structure image in 80 mm ingot height in Fig. 3), followed by an increase due to the energy-driven grain selection. Beside the grain selection due to the energy minimization of the surface energy, a further change in the orientation distribution can be induced by a twin formation at {111} growth facets frequently occurring during growth of mc-silicon [18,19]. After generation of every new twin boundary, new grain orientations appear in dependence of the original grain orientation(s). This can explain the increase of the (115) surface fraction from 12% to 20% in the conventional ingot because the (115) orientation is the Σ3-twin orientation of a (111) grain [20] which is available in a high amount of about 20% area fraction in the upper ingot half. Even if the observed trends of some orientations can be explained, the trends for the other measured grain orientations, namely (102), (105) and (313) in both ingots could not be directly correlated to special growth mechanisms as described above. Probably these orientations are just the crystallographic result from twin generation starting from non-(111) orientations, from other grain interactions or from
boundary type distributions of both ingot types change over the ingot height in a way that a constant type distribution (roughly 50% random, 30% Σ3, 15% Σ9+Σ27) is reached exceeding a specific ingot height, whereas the height value differs from the HPM ingot (80–170 mm) to the conventional ingot (170–260 mm). Also the recombination active areas within both ingots - after an initial spread along the first 170 mm - become more and more equal at longer ingot heights and finally match each other at 350–440 mm. 4. Discussion The observed trends of all investigated grain structure parameters point out that the initially formed grain structures tend to a stable state during the solidification process which is reached at a specific ingot height. Further, the initial grain structure appears to be irrelevant for the properties of the stable state, but relevant for the height at which this state is reached. In the following, the observed trends for the grain orientation, the grain boundary type distribution and the grain size for both ingot types are discussed by taking into account grain selection and different grain boundary annihilation/formation mechanisms. 4.1. Grain orientation Concerning the grain orientation it has to be pointed out that the observed trends in the 710 mm HPM ingot up to 250 mm ingot height (Fig. 7) are in good agreement with the trends of a 250 mm industrial HPM ingot published by Yang et al. [2] considering the different choice of plotted orientations. This is a further validation that the method of successive growth of several G1 ingots can be used as replication of tall industrially grown silicon ingots. By observation of Fig. 7 it is remarkable that the area fraction of all grain orientations becomes similar in both ingot types at an ingot height of about 350 mm which indicates that the grain orientation distribution reaches a stable state after a particular ingot height. In dependence on the initial value, the fraction of some grain orientations increases ((112), (111) and (115)), while others decrease ((315), (313) and (105)) over ingot height. If the starting value is close to the final value in the stable state the fraction remains nearly constant ((001), (101) and (102)), whereas this constant area fraction results from a simultaneous generation and annihilation of particular grain orientations due to different growth mechanisms which are discussed in the following. One mechanism, which takes place during the solidification of 72
Journal of Crystal Growth 459 (2017) 67–75
M. Trempa et al.
annihilation processes, respectively (see also Table 1). In the bottom part of the 710 mm HPM ingot an extremely high amount of R grain boundaries of about 70% exists (compare Fig. 8 on the left). Therefore, the generation mechanisms (1), (5) and the annihilation mechanism (6) should be the most dominant ones which is affirmed by the fraction values of 50% for (1), 20% for (5) and 40% for (6). Also the annihilation mechanism (7) is very prominent (40%) despite the Σ3 length fraction is only 20%. Due to the significantly lower length fraction of Σ3n grain boundaries, the generation mechanisms (3)–(5) and the annihilation mechanisms (8)–(10) are very rare at this stage of growth. Summing up the fractions for the generation and annihilation of R and Σ3 grain boundaries results in 20% R generation (5), 40% R annihilation (6), 55% Σ3-generation (1)–(4) and 45% Σ3annihilation (7)–(10). Considering the overall ratio of 45 to 55% of generation to annihilation, this indicates a strong decrease of R grain boundaries over ingot height (generation 20%*45% « annihilation 40%*55%) which typically appears in HPM ingots and also in the 710 mm HPM ingot in this work. In the case of the Σ3 grain boundaries the analogue calculation reveals a nearly constant trend over ingot height (generation 55%*45% = annihilation 45%*55%), which obviously does not match with the increasing trend typically observed in HPM material (Fig. 8 and e.g. [2,4]). But the grain boundary length fraction typically used for such diagrams is not the most suitable parameter to investigate frequencies of grain boundary formation/annihilation processes. Therefore, the absolute number of determined grain boundaries, corrected by the length ratio between determined and total grain boundaries, was plotted over ingot height as shown in Fig. 10. It is obvious from the right hand side diagram that now the expected constant trend of Σ3 grain boundaries over the ingot height arises, while the R boundary trend shows still a strong decrease at the initial state of growth as shown in Fig. 8. Due to the ongoing change in the grain boundary type distribution during the first 150 mm of HPM growth, the fractions of the different generation and annihilation mechanism change, too. This was also presented by Prakash et al. [22] who observed significant changes already after 30 mm growth of their small HPM ingot (see values in Table 1). According to the decreasing number of R grain boundaries, the fractions of the mechanisms where R grain boundaries are involved have drastically decreased from 50% to 20% (1), from 20% to 5% (5) and from 40% to 0% (6). Simultaneously, the mechanisms related to Σ3n grain boundaries occur with higher fractions, whereas the number of Σ3n grain boundaries remains nearly constant. In detail the generation mechanisms (2)–(4) increase from 5% to 50% and the annihilation mechanisms (8)(10) from 5% to 35%. Summing up the generation and annihilation of R and Σ3 grain boundaries yields 5% R generation (5), < 5% R annihilation (6), 70% Σ3-generation (1)–(4) and 65% Σ3-annihilation (7)–(10). Considering again the overall generation to annihilation ratio, this predicts a nearly constant trend for R grain boundaries (generation 5%*45% ≈ annihilation < 5%*55%) and Σ3 boundaries (generation 70%*45% ≈ annihilation 65%*55%), too. This is in good agreement with the obtained data from the 710 mm HPM ingot at heights above 170 mm shown in Fig. 8 and Fig. 10. In the following, the model introduced above is adapted to conventional mc silicon. For that purpose the fractions for each grain boundary mechanism (1)–(10) were estimated according to the values experimentally determined by Prakash et al. for the HPM material given in Table 1. Due to the almost identical trends for the R and Σ3 grain boundaries observed in the upper ingot regions (stable state) of the 710 mm ingots, the same fraction values as in the HPM ingot were assumed for the conventional ingot. For the lower part of the conventional ingot it was assumed that the fractions of mechanisms (1), (5), (6) and (7) should be relatively smaller, hence the amount of R grain boundaries is smaller in comparison to the stable state region. In
Table 1 Overview of experimentally observed grain boundary generation and annihilation mechanisms after [15,22], the fraction of theses mechanisms in HPM silicon (after [22]) and in conventional mc-silicon (estimated in this work). HPM
Conventional
No
Generation
Fraction (initial) after [22]
Fraction (stable state) after [22]
Fraction (Initial) estimated
Fraction (stable state) estimated
(1)
R→Σ3+R (R=random) Σ3→Σ3+Σ9 Σ9→Σ3+Σ3/ Σ27/R* Σ27→Σ3+Σ9/R R→R+R
50%
20%
< 20%
20%
5%
50%
> 50%
50%
20%
5%
< 5%
5%
40% 40% 5%
< 5% 30% 35%
0% < 30% > 35%
< 5% 30% 35%
(2) (3) (4) (5)
(6) (7) (8) (9) (10)
Annihilation R+R→R Σ3+R→R Σ3+Σ3→Σ9 Σ3+Σ9→Σ3/ Σ27/R* Σ3+Σ27→Σ9
R* are in fact large tilted Σ3 grain boundaries, but practically determined as R grain boundary according to the Brandon criterion [23].
grain boundary generation/annihilation processes (see next passage) which permanently appear during growth. 4.2. Grain boundary type distribution The second phenomenon to be discussed is the evolution of the grain boundary (GB) type distribution with increasing ingot height which is determined by the mechanisms of grain boundary generation and annihilation. The probably most prominent generation mechanism in mc-silicon is the already mentioned energetically driven twin formation at undercooled {111} growth facets, e.g. in existent grain boundary grooves [18,19]. According to this mechanism, new low energetic Σ3 twin boundaries were created during growth by the generation rule GB1→GB2+Σ3. The two other involved grain boundaries GB1 and GB2 can in principle be of various types [21]. Recently, this grain boundary generation process was investigated by different groups in small lab-scale HPM silicon ingots [4,22] as well as in ribbon growth mc-silicon samples [15]. They could show that several grain boundary generation mechanisms including twin formation occur in this type of material (see mechanisms (1)-(4) in Table 1). Additionally, the splitting of R grain boundaries into two R grain boundaries was experimentally observed in [22] (mechanism (5) in Table 1). On the other side, grain boundaries can annihilate by merging with other grain boundaries [4,15,22]. Also in this case, several mechanisms with different types of grain boundaries exist (see mechanisms (6)– (10) in Table 1). In principle, all mechanisms (1)–(10) will happen in HPM silicon as well as in conventional mc silicon. However, the frequency of each mechanism differs in dependence on the amount of the existent grain boundary types in the silicon material. Further, in the case of the annihilation mechanisms (6)–(10), the distance and the angle between the grain boundaries is an indicator for the probability of the merging process. Recently, Prakash et al. [22] have analyzed the frequency of the different grain boundary generation and annihilation mechanisms in small HPM silicon ingots by detailed investigation of more than 2000 grain boundaries. They could show that the overall fraction ratio of generation to annihilation processes is roughly 45% to 55% which means that the total amount of grain boundaries decreases over ingot height. Further, they determined the fraction of each observed mechanism related to the total amount of observed generation and 73
Journal of Crystal Growth 459 (2017) 67–75
M. Trempa et al.
Σ3
Σ9 + Σ27 No. of grain boundaries [arb. units]
No. of grain boundaries [arb. units]
Random (R)
Conventional
0
100
200
300
400
500
600
700
all
Other Σ
HPM
0
100
200
300
400
500
600
700
Total ingot height [mm]
Total ingot height [mm]
Fig. 10. Number of measured grain boundaries vs. total ingot height for the 710 mm G1 conventional ingot (left) and the 710 mm G1 HPM ingot (right).
the mean grain size increases more clearly. The distribution of the grain size becomes more homogeneous (CVGS value decreases), mainly due to the overgrowth of the large dendrites. During further growth the mean grain size in both material types remains constant which can be explained by the more and more balance between annihilation and formation of grain boundaries and grains, respectively.
consequence the fractions of the mechanisms (2)–(4) and (8)–(10) should become relatively larger even if the absolute number of Σ3n grain boundaries remains constant. After sum up the total fractions for R generation (5), R annihilation (6), Σ3-generation (1)–(4) and Σ3annihilation (7)–(10) are < 5%, 0%, 50–75% and 35–65%. This shows that the R generation is higher than the annihilation resulting in an increasing amount of R grain boundaries over ingot height which can be observed in the 710 mm ingot (Fig. 8 and Fig. 10). Hence the fraction of 5% for R generation is not as big, maybe some more R grain boundaries were gained from the mechanisms (3) and (9) exhibiting quite high fractions. The fractions for Σ3 generation are slightly higher than for annihilation; however they equal each other if the overall ratio of generation and annihilation of 55–45% is considered. As the result also in the conventional 710 mm ingot, the absolute amount of Σ3 grain boundaries remains constant in the stable state region above 300 mm as shown in Fig. 8 and Fig. 10. In summary it can be stated that the by Prakash et al. proposed “grain boundary model” is valid for both of our HPM and conventional mc silicon ingots with extraordinary height up to 710 mm.
4.4. Recombination active dislocations As shown in Fig. 9, the area fraction of recombination active dislocations at the bottom region of both ingots is comparatively low. In the HPM material, the dislocation movement is firstly kept on a low level due to the small grain size and the high amount of R grain boundaries which prohibit the movement of the dislocations into neighboring grains [24]. However, during further growth, the number of R grain boundaries decreases while the grain size becomes larger. Additionally, the length fraction of Σ3 grain boundaries increases according to the mechanisms described above. That allows the dislocations to propagate within the larger grains, cross over grain boundaries into neighboring grains and in consequence to spread in the ingot volume resulting in the observed increasing trend of the recombination active area. However, a wide spreading of dislocation clusters in the ingot volume is limited by the still relatively high R grain boundary length fraction > 50%. Further, dislocated grains can be overgrown by non-dislocated grains during due to the apparent grain selection. Therefore, the increase of recombination active dislocation area becomes smaller over ingot height until the final state is reached at about 350 mm where the generation and propagation of new dislocation clusters is balanced by the described diminishing processes. In the case of the conventional ingot, the initial increase of the dislocated area up to 170 mm ingot height is much more extensive in comparison to the HPM material due to the larger grain size (which also increases) and the high amount of Σ3 grain boundaries enabling the propagation in the ingot volume. Simultaneously, more and more R grain boundaries were formed as described in Section 4.2. Consequently, the dislocation propagation should be more and more inhibited with increasing ingot height. Additionally, existent dislocation clusters diminish by ending up at R grain boundaries or they are overgrown by non-dislocated grains. Both should lead to a decrease of the recombination active area as can be observed in the ingot from
4.3. Grain size and grain size distribution The observed trends for the other investigated grain structure parameters, the mean grain size as well as the CVGS (Fig. 2), are a direct consequence of the grain orientation and grain boundary evolution described in the sections above. The initial dominant annihilation processes of R grain boundaries in the HPM ingot lead to a large decrease of the total number of grain boundaries (and also grains) over the first 170 mm ingot height (Fig. 10) implicating the observed increase of mean grain size in Fig. 2. Further, in combination with the grain orientation selection this leads to a more inhomogeneous grain size distribution indicated by the increasing CVGS also shown in Fig. 2. In the case of the conventional ingot the initial mean grain size is higher due to large apparent dendrites paired with some smaller grains. While in the regions of the smaller grains a grain coarsening due to the grain boundary mechanisms and the grain selection takes place (including a slight decrease of the grain boundary number), the large dendrite areas are overgrown by newly formed grains (including a strong increase of grain boundary number). In summary the total amount of grain boundaries just slightly increases (see Fig. 10) while 74
Journal of Crystal Growth 459 (2017) 67–75
M. Trempa et al.
From the results it is concluded that the advantage concerning the grain structure parameters and the recombination active area of HPM silicon in comparison to conventional mc silicon is limited to the first 350 mm of the ingot. Therefore, it has to be evaluated by the industrial producers if the growth of extraordinary high HPM ingots higher than 350 mm is profitable.
170 mm to 350 mm (Fig. 9). The sharp transition from the increase to the decrease of the recombination active area could maybe explained by the fact that there is no further increase of the grain size at about 150 mm (see Fig. 2) which would promote the dislocation propagation while the increase of R grain boundaries is still ongoing until ~250 mm (see Fig. 8). At longer ingot heights above 350 mm, the recombination active area reaches a stable state characterized by an equality of permanent dislocation formation and diminishing, analogously to the HPM material.
Acknowledgement These research and development activities are carried out within the ENOWA-II project (0325805E) funded by the German Federal Ministry for Economic Affairs and Energy.
4.5. General remarks In summary, it could be shown that in both silicon materials, conventional and HPM, the same growth mechanisms principally occur. Which mechanism or interaction is dominant at specific ingot heights depends on the current grain structure and therefore which kind of grain orientations and grain boundaries are available. After a particular growth length – which can be larger than those actually applied in industry – a stable configuration of the grain structure concerning the mean grain size, the grain orientation and the grain boundary types is reached. It is assumed that the observed stable configuration in this work is just a local energy minimum which probably depends on the actual growth conditions like growth rate or temperature gradients. This assumption is supported by the results in [10] which show slightly different values for orientation and grain boundary type distribution at the top of several industrially grown ingots in comparison to our work.
References [1] C.W. Lan, W.C. Lan, T.F. Lee, A. Yu, Y.M. Yang, W.C. Hsu, B. Hsu, A. Yang, J. Crys. Growth 360 (2012) 68–75. [2] Y.M. Yang, A. Yu, B. Hsu, W.C. Hsu, A. Yang, C.W. Lan, Prog. Photovolt. Res. Appl. 23 (3) (2015) 340–351. [3] C. Reimann, M. Trempa, T. Lehmann, K. Rosshirt, J. Stenzenberger, J. Friedrich, K. Hesse, E. Dornberger, J. Crys. Growth 434 (2016) 88–95. [4] Y.T. Wong, C. Hsu, C.W. Lan, J. Crys. Growth 387 (2014) 10–15. [5] D. Zhu, L. Ming, M. Huang, Z. Zhang, X. Huang, J. Crys. Growth 386 (2014) 52–56. [6] R.R. Prakash, T. Sekiguchi, K. Jiptner, Y. Miyamura, J. Chen, H. Harada, K. Kakimoto, J. Crys. Growth 401 (2014) 717–719. [7] G.A. Babu, I. Takahashi, S. Matsushima, N. Usami, J. Cryst. Growth 441 (2016) 124–130. [8] Y.T. Wong, C.T. Hsieh, A. Lan, C. Hsu, C.W. Lan, J. Crys. Growth 404 (2014) 59–64. [9] H. Zhang, D. You, C. Huang, Y. Wu, Y. Xu, P. Wu, J. Crys. Growth 435 (2016) 91–97. [10] T. Lehmann, C. Reimann, E. Meissner, J. Friedrich, Acta Mater. 106 (2016) 98–105. [11] M. Beier, M. Trempa, J. Seebeck, C. Reimann, P. Wellmann, B. Gründig-Wendrock, J. Friedrich, K. Dadzis, L. Sylla, T. Richter. In: Proceedings of the 29th European Photovoltaic Solar Energy Conference: Amsterdam, 2014, p. 717–721. [12] T. Lehmann, M. Trempa, E. Meissner, M. Zschorsch, C. Reimann, J. Friedrich, Acta Mater. 69 (2014) 1–8. [13] T. Lehmann, Einfluss der Korngefüge industriell hergestellter mc-Siliziumblöcke auf die rekombinationsaktivenKristalldefekte und die Solarzelleneffizienz (Dissertation), Technische Universität Bergakademie Freiberg, Freiberg, 2016. [14] K. Fujiwara, Y. Obinata, T. Ujihara, N. Usami, G. Sazaki, K. Nakakjima, J. Crys. Growth 266 (2004) 441–448. [15] H.K. Lin, M.C. Wu, C.C. Chen, C.W. Lan, J. Cryst. Growth 439 (2016) 40–46. [16] I. Brynjulfsen, K. Fujiwara, N. Usami, L. Arnberg, J. Crys. Growth 356 (2012) 17–21. [17] K. Nagashio, K. Kuribayashi, Acta Mater. 53 (10) (2005) 3021–3029. [18] T. Duffar, A. Nadri, Scr. Mater. 62 (12) (2010) 955–960. [19] A. Nadri, Y. Duterrail-Couvat, T. Duffar, J. Crys. Growth 385 (2014) 16–21. [20] F. Wilhelm, J. Appl. Cryst. 4 (1971) 521–523. [21] V. Randle, Acta Mater. 52 (14) (2004) 4067–4081. [22] R.R. Prakash, K. Jiptner, J. Chen, Y. Miyamura, H. Harada, T. Sekiguchi, Appl. Phys. Express 8 (3) (2015) 35502. [23] D.G. Brandon, Acta Metall. 14 (11) (1966) 1479–1484. [24] G. Stokkan, Y. Hu, Ø. Mjøs, M. Juel, Sol. Energ. Mat. Sol. C. 130 (2014) 679–685.
5. Conclusion In this work, an extraordinary high conventional mc silicon ingot as well as HPM silicon ingot was replicated by the successive growth of eight G1 ingots. The results show that all studied grain structure parameters like mean grain size, grain orientation and grain boundary type distribution are distinctly different in the initial state of growth for both material types. However, with increasing ingot height the differences becomes smaller and the grain structures for both materials become very similar at long ingot heights above 350 mm. This behavior was found to result from grain selection and grain boundary generation/annihilation mechanisms which are driven by energy reduction or simply by crystallographic relationships and happen in both material types in a similar manner. Further, it was shown that the grain structure evolution over ingot height directly influences the dislocation propagation within the silicon ingots leading also to a comparable recombination active dislocation area in the conventional and the HPM silicon material at long ingot heights.
75