Vacuum 128 (2016) 66e72

Contents lists available at ScienceDirect

Vacuum
journal homepage: www.elsevier.com/locate/vacuum

Influence of different vacuum conditions on the distribution of

insoluble inclusions in a multi-crystalline silicon ingot
Shi Qiu a, Shiqiang Ren b, Liuqing Huang a, Tianyu Tang a, Ming Fang a, b, Xuetao Luo a, *
a
Fujian Provincial Key Laboratory of Advanced Materials, College of Materials, Xiamen University, Xiamen 361005, China
b
School of Material Science and Engineering, Dalian University of Technology, Dalian 116023, China

a r t i c l e i n f o a b s t r a c t

Article history: The distribution of insoluble inclusions in multi-crystalline (m-c) silicon was investigated by directional
Received 20 December 2015 solidification under different vacuum conditions using recycled scraps. We found an abnormal peak
Received in revised form (weight ratio of 1.19%) of SiC and Si3N4 with an average size of 100 mm located at the bottom of the silicon
11 March 2016
ingot obtained under low-vacuum conditions. At the top of the ingot, the impurity phase featured a
Accepted 12 March 2016
Available online 14 March 2016
weight ratio exceeding 10.62% and a size of 20 mm. Results showed that large particles sank to the bottom
of the ingots whereas small particles floated upward. Further investigation suggested that the dominant
factor influencing inclusion distribution is related to the presence of metallic impurities (e.g., Ca, Al, and
Keywords:
Directional solidification
Na) and their corresponding microstructure. A mechanism of migration was proposed to reveal that
Vacuum bonding of metallic impurities and insoluble inclusions contribute to the abnormal distribution of im-
Multi-crystalline silicon purities. The results provide guidance for controlling the distribution of insoluble inclusions in recycled
Nitrides m-c silicon.
Carbides © 2016 Elsevier Ltd. All rights reserved.

1. Introduction with large amounts of insoluble inclusions have yet to be reported.

Multi-crystalline (m-c) silicon, one of the main materials used to
manufacture solar cells, must show acceptable contamination
levels in solar-grade silicon products [1e3]. Metallurgical methods
are environment-friendly and inexpensive m-c production opera-
tions that aim to purify silicon from metallurgical-grade (99%) to
solar-grade (99.9999%) [4]. Extensive efforts have been exerted to
remove impurities, such as Al, Ca, Fe, B, and P, from metallurgical
silicon effectively [5e7]. However, insoluble inclusions (including
SiC and Si3N4) in silicon have yet to be sufficiently investigated to
fulfill the requirements of the photovoltaic industry. During block
casting, insoluble inclusions frequently appear in certain portions
of silicon ingots [8,9]. The presence of SiC leads to severe ohmic
shunt current flow in large-angle grain boundaries, causing
degradation in solar cell efficiency [10,11]. Broken wires caused by
insoluble inclusions result in wafer defects [12]. Both SiC and Si3N4
can induce impurity accumulation, which causes larger local de-
fects in the final wafers [13,14]. Methods to recycle silicon scraps
* Corresponding author. Fujian Key Laboratory of Advanced Materials, College of
Materials, Xiamen University, No. 422, Siming South Road, Xiamen, Fujian 361005,
China.
E-mail address: xuetao@xmu.edu.cn (X. Luo).
Insoluble inclusions are expected in ingots, which feature
various contamination sources, including silicon scraps recycling,
Si3N4 coating, dust particles on the long-term insulation plate, and
reactions of the silica crucible and the graphite supporter
[12,14e16]. Therefore, the behavior of insoluble inclusions in a sil-
icon ingot must be understood to improve the solar cell efficiency of
wafers. Researchers have investigated a number of insoluble in-
clusion characteristics, including their existing forms, sizing, dis-
tribution, and segregation behavior, near the solideliquid interface
during directional solidification [17e22]. However, previous
studies have only concentrated on individual particles of insoluble
inclusions during normal directional solidification (DS) (non-vac-
uum conditions). Other impurities, such as volatile impurities (e.g.,
Ca, Na) and non-volatile impurities (e.g., Fe, B), may also be found in
m-c silicon. Removal of volatile impurities has been investigated by
considering the segregation behavior and evaporation mechanism
of DS under low-vacuum conditions [23e27]. Considering that the
interactions of impurities with each other influence the corre-
sponding redistribution, studying the behavior of insoluble in-
clusions while accounting for the existence of other impurities are a
significant endeavor.
In this paper, the distributions of insoluble inclusions and other
impurities along the height of an ingot were measured. Interactions

http://dx.doi.org/10.1016/j.vacuum.2016.03.010
0042-207X/© 2016 Elsevier Ltd. All rights reserved.
 S. Qiu et al. / Vacuum 128 (2016) 66e72
between insoluble inclusions and volatile impurities were studied
during DS under low-vacuum and non-vacuum conditions in m-c
silicon.
2. Experiment
A pilot-scale electromagnetic induction furnace (DPS-650) was
used to produce a silicon ingot, which is shown in Fig. 1. A graphite
heater wound with copper coils absorbed the majority of the
magnetic flux and generated heat by electromagnetic induction. A
protective layer of Si3N4 was painted on the inner wall of a cylin-
drical silica crucible (F810 mm
620 mm) to act as a barrier for
impurities and relieve thermal strain from silicon. An insulation
plate on top of the silicon was designed to reduce carbon
contamination from graphite components. Heater was used to
control temperature field in the furnace chamber. During the DS
process, silicon, as well as the crucible, moved downwards gradu-
ally from the hot zone to the cold zone. The increasing of crucible
pulling down heights led to larger heat dissipating area at the
bottom of silicon. Maintaining the net heat flux along axial direc-
tion was in favor of generating a vertical, homogeneous, columnar
grain structure. Two thermocouples were adopted to monitor
temperature fields in the furnace. Thermocouple 1 (TC1) was
installed on the surface of the graphite heater to record the heater
temperature, while thermocouple 2 (TC2) was installed at the
bottom of the crucible to confirm melting completion.
The DS process involved three stages: melting stage, crystalli-
zation stage and cooling stage. In melting stage, the insulator was
closed. Heater temperature increased gradually to 1850 K, and then
maintained the temperature until melting completion. In crystal-
lization stage, heater temperature decreased to 1800 K, and the
crucible started pulling downwards. Heat was transferred to the
bottom of crucible wall. Temperature gradient from bottom to top
of silicon melt was formed. Crystal growth occurred once the
temperature decreased to melting point of silicon. The height of
crystal growth was measured by a quartz rod immersing into the
molten silicon every hour. The solidified stage was considered
Fig. 1. Schematic of the pilot scale apparatus.
67
completed with the same measured value for three times, then
entered into the cooling stage. Crucible didn't pulling downwards
continually, and heater temperature decreased gradually to room
temperature.
Silicon feedstock originated from the edges, head, and tail of the
ingots, which included large amounts of foreign inclusions. Fig. 2
shows the process parameters of vacuum degree and solidified
height during the DS process. In the melting stage, a protective
atmosphere of argon was flowed and the pressure increased
gradually to 6
104 Pa. At the start of the crystallization, argon also
flowed under a pressure of 6
104 Pa until the solidified height
reached 50 mm. The furnace chamber was then evacuated to 0.3 Pa
in 30 min, and the vacuum degree was maintained until the end of
crystallization. In cooling stage, the pressure of argon increased to
6
104 Pa again. It should be note that the crystal growth rate, or
the ratio of solidified height to solidification time, was controlled to
a constant value to ignore the influence of crystal growth rates on
the segregation behavior of impurities.
After cooling down, the grown ingot (410 kg,
F810 mm
350 mm) was cleaned by abrasive blasting and
sectioned by a mechanical saw. Two axial rods close to the center
region were selected for subsequent tests. For on silicon rod, nine
pieces axial direction were chosen to be completely etched off with
an acid mixture of HNO3 and HF (1:1), followed by filtration and
weighing. For further identification, a scanning electron micro-
scope (SEM; Hitachi SU-70, Japan) with an energy dispersive
spectrometer (EDS; INCAE250, Oxford, England) was employed to
investigate impurity components of silicon rods and the acid-
etched residues. The other silicon rod was equally divided into 34
blocks, and eight pieces (10
10
10 mm3) along the axial di-
rection were selected to determine impurity concentrations by
inductively couple plasma mass spectrometry (ICP-MS; DRC II,
Perkin Elmer, US).
3. Results
The weights of initial samples with different solidified heights
and corresponding insoluble inclusions after acid etching are
shown in Table 1. Under the non-vacuum condition (solidified
height of 0e30 mm), no residues are collected. Under the low-
vacuum condition, three regions are found along the DS direc-
tion: insoluble precipitation at the start of solidification (50 and
60 mm), empty space (80e300 mm), and a large amount of
Fig. 2. Process parameters of vacuum degree and solidified height during the DS
process.
68 S. Qiu et al. / Vacuum 128 (2016) 66e72

Table 1
Weight percentages of insoluble inclusions in the nine 1 cm3 pieces from the bottom to the top of the ingot under the mixture of non- and low-vacuum condition.

Sample no. Solidified height (mm) Initial weight M1 (g) Weight after etching M2 (g) Weight ratio of inclusions (%)a

1 10 3.5042 –b e
2 30 3.3663 e e
3 50 3.4724 0.0414 1.1922
4 60 3.6945 0.0168 0.4547
5 80 3.4298 e e
6 150 3.4585 e e
7 240 3.5846 e e
8 300 3.7642 e e
9 340 4.4131 0.4689 10.6252
a
Weight ratio of inclusions equals to M2 =M1 , revealing the distribution of insoluble inclusions.
b
“–” indicates that the amount of inclusions is either too low for collection or completely absent in the sample.

precipitation on top of the ingot (340 mm). The main impurities of Table 2
the residues are confirmed as Si3N4 rods and SiC particles, and the Concentrations of impurities of recycled m-c silicon by ICP-MS.
relevant morphologies are shown in Fig. 3. Element Concentration(ppm wt) Element Concentration(ppm wt)
It is interesting that a peak of insoluble inclusions occurred at
B 0.13 Na 1.97
the start of low-vacuum condition, and the vacuum condition may P <1 Fe 5.68
be an important factor. However, we also had reason to doubt that Al 3.06 Cu 0.07
the other impurities in raw material caused the redistribution of Ca 5.15 Mn 0.06
Ni 0.01 Ti 0.02
Si3N4 and SiC or the risk of air leakage into the chamber at the start
of vacuum pump working caused the changing of the oxygen/car-
bon equilibrium.
To further investigate the cause of the peak of insoluble in-
clusions, it was important that a series of test experiments were
added to clarify the nature of raw materials. Concentrations of
impurities by ICP-MS were shown in Table 2. The metallic impu-
rities, such as Fe, Ca, Al, Na, were slightly higher than normal m-c
silicon (>6 N silicon). Afterwards the raw material was also chosen
to be completely etched off, and the residues are shown in Fig. 4.
The size of Si3N4 particle was over hundreds of microns, which
indicated that recycled Si already contained large insoluble in-
clusions. A possible explanation of the peak of insoluble inclusions
in our experiments was that large insoluble inclusions in raw ma-
terials deposited at the bottom region.
Then, a different vacuum condition was adopted that the vac-
uum degree in the furnace chamber was set as 0.3 Pa at the start of
the crystallization. After cooling down to room temperature, the
grown ingot also experienced acid pickling and weighing. Weight
ratios of insoluble inclusions, as a function of solidification fraction
from the bottom to the top of an ingot, are under the low-vacuum Fig. 4. Concentrations of impurities along the solidified height from the bottom to the
top of the ingot: (a) mixture of non- and low-vacuum condition during crystallization;
(b) low-vacuum condition during crystallization.

condition and the mixture of non- and low-vacuum condition, as is

Fig. 3. Secondary electron SEM micrograph of silicon samples obtained after etching
(in the solidified height of 50 mm) shows that insoluble inclusions are rod-like Si3N4
and small SiC particles.
shown in Fig. 5. The data of the red dash dot line come from Table 1,
whereas the blue dash line is under low-vacuum condition. Large
amounts of residues appeared on top of the ingots both under the
low-vacuum condition and the mixture vacuum condition. The
abnormal peak (solidified height of 50 mm on the red dash dot line)
indicated insoluble inclusion aggregation under low-vacuum con-
dition, but this relationship did not apply to the peak of the blue
solid line (non- and low-vacuum condition). The main difference of
the process parameters was vacuum condition. Although it is
possible that the distribution of insoluble inclusions is related to
the changes vacuum conditions, insoluble inclusions of Si3N4 and
SiC cannot be volatilized to their gas phase under vacuum condi-
tions. Moreover, if air leakage at the start of vacuum pump working
caused the abnormal deposition, the peak values of insoluble in-
clusions under the mixture of non- and low-vacuum condition and
the low-vacuum condition should be identical. However, a differ-
ence shown in Fig. 5 indicates that the peak value at the start of
 S. Qiu et al. / Vacuum 128 (2016) 66e72 69

Fig. 5. Weight ratios of insoluble inclusions as a function of solidification fraction from

the bottom to the top of an ingot are under low-vacuum, the mixture of non- and low-
vacuum condition. The data of the red dash dot line come from Table 1, whereas the
blue dash line indicates the distribution under low-vacuum condition. (For interpre-
tation of the references to colour in this figure legend, the reader is referred to the web
version of this article.)

low-vacuum condition is much larger than that under the mixed

vacuum conditions. Hence, the vacuum conditions cannot directly
influences the distribution of Si3N4 and SiC, and air leakage cannot
explain the abnormal peak values of insoluble inclusions as well.
The concentrations of impurities from the bottom to the top of
the m-c silicon ingot during crystallization are shown in Fig. 6: (a)
under the mixture of non- and low-vacuum condition; (b) under
the low-vacuum condition. The abnormal peak values of Ca, Al
appear at the start of low-vacuum condition in both of Fig. 5(a) and
(b), while no changes of B, Fe occur under different vacuum con-
ditions. In general, the impurities are divided into two types. For
one type (e.g., B and Fe), the distribution of impurities are hardly
influenced by vacuum conditions, and mainly determined by the
solubility in solid and liquid phase of silicon [2]. For the other type
(e.g., Ca, Al, Na), the distribution of impurities is mainly determined
by the segregation behavior caused by the diffusibility and evapo-
rability of impurity atoms [25,26,28]. High concentrations of im-
purities in the final grown silicon are due to accumulation of
impurities on the solideliquid interface. Considering low segrega-
tion coefficient, the concentrations of Al, Ca, and Na at the start of
solidification should be minimal. In our experiment under mixture
of non- and low-vacuum condition shown in Fig. 6 (a), however,
peaks of impurities (e.g., Al, Ca, and Na) are located at the bottom
(50 mm) of an ingot. The initial concentrations of Ca, Al, and Na are
C0(Ca) ¼ 5.15 ppm wt., C0(Al) ¼ 3.06 ppm wt., and C0(Na) ¼ 1.97 ppm
wt, respectively, and their peak values are 4.22, 2.83, and 1.72 ppm
wt. The transition tendency of Ca, Al, and Na is similar to that of
insoluble inclusions shown in Fig. 5. We can expect that aggrega-
tions of SiC and Si3N4 at the start of processing in low-vacuum
conditions may be attributed to the presence of Al, Ca, and Na.
The morphology of the high impurity phase in the silicon ingot
under the mixture of non- and low-vacuum condition is shown in
Fig. 7. At a solidified height of 50 mm, the average particle size of
Phase І is over 100 mm, as also observed in Fig. 3. Considering the
20 mm size of Phase III at the top of an ingot, the result shows that
larger particles precipitate at the bottom of the ingots, whereas
smaller particles float upward. According to Table 1, the weight
ratios of Phases І and II exceed 1.19%, whereas that of Phase III is
beyond 10.62%. Comparison of the impurity phase indicates that a
Fig. 6. Concentrations of impurities along the solidified height from the bottom to the
top of the ingot: (a) mixture of non- and low-vacuum condition during crystallization;
(b) low-vacuum condition during crystallization.
large portion of the impurities in grown silicon ingots belongs to
small particles. The corresponding chemical compositions are
confirmed in Table 3. Phase І reveals straight hexagon-shaped rods
of Si3N4, whereas Phases II and III show the coexistence of carbides
and metallic impurities (Ca, Al, and Na). After Phases I and II
completely etching, Fig. 3 indicates that carbides in Phases II are
small SiC particles. With Si3N4 acting as nucleation sites for im-
purities [8], various kinds of impurities got together to form im-
purity clusters (Phase II) around insoluble inclusions (Phase I). This
observation confirms that the presence of impurity clusters influ-
enced the distribution of insoluble inclusions under vacuum con-
ditions. An explanation is that, as the density of Ca and Na were
lower than melt silicon, impurity clusters covering small SiC par-
ticles carried large insoluble inclusions suspending in the melt
silicon (solidified height of 0e50 mm under non-vacuum condi-
tion), while deposited if Ca and Na were not enough (solidified
height of 50 mm at the start of low-vacuum condition).
It is possible that the occurrence of abnormal peak at the change
of vacuum condition is due to the existence of volatile metallic
impurities (Ca, Na, Al) in recycled silicon, especially for Ca and Na.
In addition to the metallic impurities are around insoluble in-
clusions, the densities of Ca and Na are all lower than that of melt
70 S. Qiu et al. / Vacuum 128 (2016) 66e72
silicon and the concentrations are easily influenced by vacuum
conditions. If there are a number of Ca and Na around inclusions
(non-vacuum stage in the mixture of non- and low-vacuum con-
dition), some small SiC particles covered by impurities clusters and
large inclusions were pushed from solid-liquid interface to the melt
silicon. Once impurities of Ca, Na are separated from inclusions
(low-vacuum stage in the low-vacuum condition or the mixture of
non- and low-vacuum condition), large inclusions were engulfed
by the solideliquid interface, whereas small insoluble inclusions
particles remained suspended by remained metallic clusters and
finally accumulated at the top of m-c silicon ingot.
4. Discussion
4.1. Transport characteristics of Ca, Al, and Na
To understand the migration of insoluble inclusions further, the
transport characteristics of impurities must be investigated. The
mass transfer coefficient of impurities reflects the mass of impurity
atoms transferred from melted silicon to the gas phase per unit of
time. The mass transfer coefficient k depends on the solidification
rate v, the density of solid silicon rs, the density of liquid silicon rl,
the height of the liquid silicon ingot H, the initial concentration C0,
an arbitrary time t during solidification, and the corresponding
concentration of evaporation Ct, which can be expressed as [26]:
8
> 0 < t  2:4
104
>0
>
>
<
k¼  (1)
>
>
>
> vrl Ct      
: ln ln H  v t  2:4
104 rS =rl  lnðHÞ
2:4
10 < t < 12
10
r S C0
Given that v ¼ 3.54
106 (m/s), rs ¼ 2.33
103 (kg/m3),
rl ¼ 2.45
103 (kg/m3), H ¼ 0.35 (m), C0(Ca) ¼ 5.15 ppm wt.,
C0(Al) ¼ 3.06 ppm wt., C0(Na) ¼ 1.97 ppm wt., solidification time
t ¼ 12
104 s, and concentrations of Ct(Ca) ¼ 1.50 ppm wt.,
Ct(Al) ¼ 1.70 ppm wt., and Ct(Na) ¼ 0.85 ppm wt. may be obtained
after evaporation.
Fig. 8 shows that mass transfer coefficients of impurities in
molten silicon under the mixture of non- and low-vacuum condi-
tion as a function of directional solidification time. The mass
transfer coefficients of Ca, Al, and Na show the same tendency of
time changes. At the start of processing in low-vacuum conditions,
the mass transfer coefficients of Ca, Al, and Na decrease sharply and
then plateau at a low value with increasing solidification time, as
shown in Fig. 8. This trend indicates that the majority of the im-
purity atoms rapidly evaporate from the melt to the gas phase at the
beginning of processing, subsequently reaching equilibrium with
the extraction of several atoms by the vacuum system.
4.2. Migration of insoluble inclusions
Due to the fact that melting points of SiC and Si3N4 are
considerably higher than that of silicon, their existing forms are
solid particles. Force analysis in molten silicon should was con-
ducted to determine whether the particles are engulfed or pushed
by the plane front [20]. Forces acting on particles for a planar sol-
ideliquid interface include the repulsive force (Fr), the weight force
(FW), and the buoyancy force (FB). Fr is determined by the surface
area of inclusions and the surface energy difference (Ds) between
liquid and impurity particles [19].
pRDs
Fr ¼ (2)
2
The buoyancy force is associated with the density difference
between molten silicon and impurities (Dr), particle size (R), and
gravitational acceleration (g), shown as [29].
4 3
FB ¼ pR gDr: (3)
3
FW is related to the weight of insoluble inclusions
FW ¼ mg: (4)
Particles can remain stably floating in molten silicon only if Fr
and FB are equal to FW. If this balance is broken, the particles are
engulfed/pushed by the solidification front.
A mechanism of insoluble inclusion migration is used to reveal
the abnormal peak of insoluble inclusions, which is shown in Fig. 9.
The state of insoluble particles in molten silicon under non-vacuum
conditions is shown in Fig. 9 (a). The transfer behavior of particles
under low-vacuum conditions is shown in Fig. 9(b)e(d). Although
the densities of nitrides and carbides are higher than that of silicon,
4
4
bands of particles are found ahead of the solideliquid interface
during crystal growth [19]. Among these bands, solid insoluble
particles act as nucleation sites to adsorb volatile metallic
impurities.
Toward the end of processing in non-vacuum stage under the
mixture vacuum condition in Fig. 9 (a), abundant volatile metallic
atoms aggregate on the surface of insoluble particles. FB is suffi-
ciently large to guarantee that
FB þ Fr ¼ FW (5)
Thus, regardless of size, insoluble particles are suspended above
the solideliquid interface.
As seen in Fig. 9 (b), at the start of processing in low-vacuum
stage, large amounts of impurity atoms are extracted from the
bands of particles as a result of high mass transfer coefficients. Parts
of the impurities separated from insoluble particles lead to de-
creases in FB. For large particles, FW is larger than the sum of the Fr
and FB. Large particles are engulfed by the solideliquid interface,
whereas small particles remain suspended.
During crystallization, the mass transfer coefficients sharply
decrease to low values, and few metallic atoms of Ca, Al, Na move
away from the insoluble particles. Hence, insoluble inclusions
gradually move upward, as shown in Fig. 9 (c). In the final stage of
crystallization shown in Fig. 9 (d), the suspended particles are
aggregated at the top the solideliquid interface, leading to a high
weight ratio at the top of the silicon ingot.
 S. Qiu et al. / Vacuum 128 (2016) 66e72 71

Fig. 8. Mass transfer coefficients of Ca, Al, and Na in molten silicon under the mixture
of non- and low-vacuum condition as a function of directional solidification time.

Fig. 7. Secondary electron SEM micrographs of the silicon samples: (a) Solidified
height of 50 mm, (b) solidified height of 340 mm, and (c) detailed view of the impurity
phase in (b).

Table 3
Chemical composition (wt.%) of impurity phases in a silicon ingot based on Fig. 7 as
measured by EDS testing.
Fig. 9. Schematic of insoluble inclusion migration in m-c silicon ingots under the
Phase Si Al Ca Na C N O
mixture of non- and low-vacuum condition along the solidification direction. Large
І 69.95 e e e 0.53 28.5 1.01 and small blue particles indicate different sizes of insoluble inclusions, whereas green
II 62.23 2.58 0.39 0.40 2.66 3.35 28.38 particles show volatile metallic atoms. (For interpretation of the references to colour in
III 37.27 2.74 0.30 e 1.51 44.90 13.27 this figure legend, the reader is referred to the web version of this article.)

5. Conclusions insoluble particles are proposed to be the dominant driving force of

DS under different vacuum conditions is adopted to investigate
the migration behavior of insoluble inclusions in m-c silicon. An
abnormal peak (weight ratio of 1.19%) of SiC and Si3N4 with average
sizes of 100 mm is located at the bottom of the silicon ingot obtained
under the mixture of non- and low-vacuum condition. At the top of
the ingot, the weight ratio of the impurity phase exceeds 10.62%
and its size is 20 mm. This large size of impurity particles is due to
agglomeration of insoluble inclusions and impurities of Ca, Al, Na at
the bottom of the ingots. The interactions of Ca, Al, Na, and
impurity transition from the bottom to the top surface of the
samples. Extraction of atoms of Ca, Al, Na from insoluble particles,
which is caused by high mass transfer coefficients at the beginning
of processing at low-vacuum conditions, is the main cause of
abnormal peaks. The results provide guidance for the removal of
insoluble inclusions from recycled silicon.
Acknowledgments
The authors grateful acknowledge the financial support from
72 S. Qiu et al. / Vacuum 128 (2016) 66e72
Qingdao Longsun Silicon Technology company Ltd, and support of
the National Natural Science Foundation of China (51334004) as
well as the Scientific Technological Innovation Platform of Fujian
Province (2006L2003). Especially, we thank very much Mr. Zheng
Gu and Mr. Zhenhai Hou for the helpful discussions.
References
[1] P.Y. Gan, Z. Li, Quantitative study on long term global solar photovoltaic
market, Renew. Sustain. Energy Rev. 46 (2015) 88e99.

vre, C. del Can
[2] J. Hofstetter, J.F. Lelie ~ izo, A. Luque, Acceptable contamination
levels in solar grade silicon: From feedstock to solar cell, Mater. Sci. Eng. B
159e160 (2009) 299e304.
[3] S. Pizzini, Towards solar grade silicon: Challenges and benefits for low cost
photovoltaics, Sol. Energy Mater. Sol. Cells 94 (2010) 1528e1533.
[4] T. Buonassisi, A.A. Istratov, M.A. Marcus, B. Lai, Z. Cai, S.M. Heald, E.R. Weber,
Engineering metal-impurity nanodefects for low-cost solar cells, Nat. Mater. 4
(2005) 676e679.
[5] M. Fang, C. Lu, L. Huang, H. Lai, J. Chen, X. Yang, J. Li, W. Ma, P. Xing, X. Luo,
Multiple Slag Operation on Boron Removal from Metallurgical-Grade Silicon
Using Na2O-SiO2Slags, Ind. Eng. Chem. Res. 53 (2014) 12054e12062.
[6] S. Zheng, T.A. Engh, M. Tangstad, X.-T. Luo, Numerical Simulation of Phos-
phorus Removal from Silicon by Induction Vacuum Refining, Metall. Mater.
Trans. A 42 (2011) 2214e2225.
[7] S. Pizzini, M. Acciarri, S. Binetti, From electronic grade to solar grade silicon:
chances and challenges in photovoltaics, Phys. Status Solidi (A) 202 (2005)
2928e2942.
[8] N. Chen, B. Liu, S. Qiu, G. Liu, G. Du, Study of SiC and Si3N4 inclusions in in-
dustrial multicrystalline silicon ingots grown by directional solidification
method, Mater. Sci. Semicond. Process. 13 (2010) 231e238.
[9] L. Zhang, A. Ciftja, Recycling of solar cell silicon scraps through filtration, Part
I: Experimental investigation, Sol. Energy Mater. Sol. Cells 92 (2008)
1450e1461.
[10] J. Bauer, O. Breitenstein, J.-P. Rakotoniaina, Electronic activity of SiC pre-
cipitates in multicrystalline solar silicon, Phys. Status Solidi (A) 204 (2007)
2190e2195.
[11] O. Breitenstein, J. Bauer, J.P. Rakotoniaina, Material-induced shunts in multi-
crystalline silicon solar cells, Semiconductors 41 (2007) 440e443.
[12] G. Du, L. Zhou, P. Rossetto, Y. Wan, Hard inclusions and their detrimental
effects on the wire sawing process of multicrystalline silicon, Sol. Energy
Mater. Sol. Cells 91 (2007) 1743e1748.
[13] S.M. Myers, M. Seibt, W. Schro €ter, Mechanisms of transition-metal gettering in
silicon, J. Appl. Phys. 88 (2000) 3795.
[14] A.K. Søiland, E.J. Øvrelid, T.A. Engh, O. Lohne, J.K. Tuset, Ø. Gjerstad, SiC and
Si3N4 inclusions in multicrystalline silicon ingots, Mater. Sci. Semicond.
Process. 7 (2004) 39e43.
[15] A. Lotnyk, J. Bauer, O. Breitenstein, H. Blumtritt, A TEM study of SiC particles
and filaments precipitated in multicrystalline Si for solar cells, Sol. Energy
Mater. Sol. Cells 92 (2008) 1236e1240.
[16] W. Ma, G. Zhong, L. Sun, Q. Yu, X. Huang, L. Liu, Influence of an insulation
partition on a seeded directional solidification process for quasi-single crys-
talline silicon ingot for high-efficiency solar cells, Sol. Energy Mater. Sol. Cells
100 (2012) 231e238.
[17] C.P.K.F. Schmid, T.G. Digges Jr., Larry Kaufman, Origin of SiC impurities in
silicon crystals grown from the melt in vacuum, Electrochem. Soc. 126 (1979)
935e938.
[18] O.B.J.P. Rakotoniaina, M. Werner, M. Hejjo Al Rifai, T. Buonassisi, M.D. Pickett,
M. Gosh, A. Müller, N. Le Quang, Distribution and formation of silicon carbide
and silicon nitride precipitates in block-cast multicrystalline silicon, in: Pro-
ceedings of the 20th European Photovoltaic Solar Energy Conference, Barce-
lona, 2005, pp. 773e776.
[19] A.M. Doru, M. Stefanescu, A. Sedat Kacar, Brij K. Dhindaw, The influence of
buoyant forces and volume fraction of particles on the particle pushing
entrapment transition during directional solidification of Al SiC and Al
graphite compisites, Metall. Trans. A 21A (1990) 9.
[20] J.K.K.a.P.K. Rohatgi, The effect of the diffusion of solute between the particle
and the interface on the particle pushing phenomena, Acta Mater. 46 (1998)
1115e1123.
[21] M.G.W.A.W. Rempel, The interaction between a particle and an advancing
solidification front, J. Cryst. Growth 205 (1999) 427e440.
[22] E.M. Agaliotis, C.E. Schvezov, M.R. Rosenberger, A.E. Ares, A numerical model
study of the effect of interface shape on particle pushing, J. Cryst. Growth 354
(2012) 49e56.
[23] T. Luo, G. Lv, W. Ma, K. Wei, X. Yang, S. Li, Numerical and experimental study
of vacuum directional solidification purification process for SoG-Si in metal-
lurgical route, J. Cryst. Growth 384 (2013) 122e128.
[24] X. Yang, W. Ma, G. Lv, K. Wei, T. Luo, D. Chen, A modified vacuum directional
solidification system of multicrystalline silicon based on optimizing for heat
transfer, J. Cryst. Growth 400 (2014) 7e14.
[25] Y. Tan, S. Ren, S. Shi, S. Wen, D. Jiang, W. Dong, M. Ji, S. Sun, Removal of
aluminum and calcium in multicrystalline silicon by vacuum induction
melting and directional solidification, Vacuum 99 (2014) 272e276.
[26] S. Ren, P. Li, D. Jiang, S. Shi, J. Li, S. Wen, Y. Tan, Removal of Cu, Mn and Na in
multicrystalline silicon by directional solidification under low vacuum con-
dition, Vacuum 115 (2015) 108e112.
[27] K. Liu, J. Wu, K. Wei, W. Ma, K. Xie, S. Li, B. Yang, Y. Dai, Application of mo-
lecular interaction volume model on removing impurity aluminum from
metallurgical grade silicon by vacuum volatilization, Vacuum 114 (2015)
6e12.
[28] J.S.a.M. Tangstad, Vacuum refining of molten silicon, Metall. Trans. B 43B
(2012) 1427e1445.
[29] S.S. Andris, V. Bune, Sundeep Mukherjee, Adrian Catalina, Doru M. Stefanescu,
Effect of melt convection at various gravity levels and orientations on the
forces acting on a large spherical particle in the vicinity of a solidification
interface, J. Cryst. Growth 211 (2000) 446e451.