Semiconductors and semimetals, volume 92

SERIES EDITORS
EICKE R. WEBER
Director
Fraunhofer-Institut
f€
ur Solare Energiesysteme ISE
Vorsitzender, Fraunhofer-Allianz Energie
Heidenhofstr. 2, 79110
Freiburg, Germany

CHENNUPATI JAGADISH
Australian Laureate Fellow
and Distinguished Professor
Department of Electronic
Materials Engineering
Research School of Physics
and Engineering
Australian National University
Canberra, ACT 0200
Australia


Academic Press is an imprint of Elsevier
225 Wyman Street, Waltham, MA 02451, USA
525 B Street, Suite 1800, San Diego, CA 92101-4495, USA
125 London Wall, London, EC2Y 5AS, UK
The Boulevard, Langford Lane, Kidlington, Oxford OX5 1GB, UK
First edition 2015
© 2015 Elsevier Inc. All rights reserved
No part of this publication may be reproduced or transmitted in any form or by any means,
electronic or mechanical, including photocopying, recording, or any information storage and

retrieval system, without permission in writing from the publisher. Details on how to seek
permission, further information about the Publisher’s permissions policies and our
arrangements with organizations such as the Copyright Clearance Center and the Copyright
Licensing Agency, can be found at our website: www.elsevier.com/permissions.
This book and the individual contributions contained in it are protected under copyright by
the Publisher (other than as may be noted herein).
Notices
Knowledge and best practice in this field are constantly changing. As new research and
experience broaden our understanding, changes in research methods, professional practices,
or medical treatment may become necessary.
Practitioners and researchers must always rely on their own experience and knowledge in
evaluating and using any information, methods, compounds, or experiments described
herein. In using such information or methods they should be mindful of their own safety and
the safety of others, including parties for whom they have a professional responsibility.
To the fullest extent of the law, neither the Publisher nor the authors, contributors, or editors,
assume any liability for any injury and/or damage to persons or property as a matter of
products liability, negligence or otherwise, or from any use or operation of any methods,
products, instructions, or ideas contained in the material herein.
ISBN: 978-0-12-801021-1
ISSN: 0080-8784
For information on all Academic Press publications
visit our website at store.elsevier.com


CONTRIBUTORS
Peter Dold
Fraunhofer CSP, Halle, Germany. (ch1)
Hans Joachim M€
oller
Fraunhofer Technology Center for Semiconductor Materials, Freiberg, Germany. (ch2)

Thomas Walter
Faculty of Mechatronics and Medical Engineering, University of Applied Sciences Ulm,
Ulm, Germany. (ch3)

vii


PREFACE
The rapid transformation of our energy supply system to the more efficient
use of increasingly renewable energies is one of the biggest challenges and
opportunities of the present century. Harvesting solar energy by photovoltaics is considered to be a cornerstone technology for this truly global transformation process, and it is well on its way. The speed of progress is
illustrated by looking at some figures of the cumulative installed PV peak
power capacity. In Part 1 of this series of “Advances of Photovoltaics,”
published in 2012, the introduction mentioned 70 GWp installed at the
end of 2011. As we write this preface of Part 4 in the spring of 2015, 1%
of the world electricity generation is now already supplied by PV, and in
the coming months the global PV installation figure will have tripled
compared with 2011! But this is just the beginning of the thousands of
GWp that are likely to be installed in the decades to come.
Key for this extraordinary development was the rapid decrease of PV
prices and thus the cost of solar electricity. This was fueled by a rapid
technology development with soaring efficiencies at reduced production
cost, coupled with an effective market introduction policy, especially the
well-designed German feed-in tariff. Today, we can harvest solar electricity
even in Germany—with insolation comparable to Alaska!—for about
10 $ct/kWh, and in sun-rich areas for half of this amount, far below the cost,
e.g., electricity obtained from Diesel generators.
As already mentioned above, this book presents the fourth volume in the
ongoing series “Advances in Photovoltaics” within Semiconductors and
Semimetals. This series has been designed to provide a thorough overview

of the underlying physics, the important materials aspects, the prevailing and
future solar cell design issues, production technologies, as well as energy system integration and characterization issues. The present volume deals with
three important issues, of crystallizing silicon, the dominating PV material,
the ways of how to transform it into wafers for solar cells, as well as the issue
of reliability of CIGS-based thin film solar cells and modules. Following the
tradition of this series, all chapters are written by world-leading experts in
their respective field.
As we write this text, the German PV market is likely to collapse from a
7.5 GWp/a market as recently as 2012 to a 1 GWp/a level in 2015, a market
size that we last had in 2007. Fortunately, other markets in China, Japan, and
ix


x

Preface

the USA are now taking over by currently developing into 10 GWp per year
and more markets.
The solar PV revolution has started irreversibly, it is now fueled by
economics in addition to the concern for reducing climate gas emissions,
and it takes rapid foothold beyond Europe in Asia and the Americas, the
other parts of our planet will follow in a few year’s time!
GERHARD P. WILLEKE AND EICKE R. WEBER
Fraunhofer ISE, Freiburg, Germany


CHAPTER ONE

Silicon Crystallization

Technologies
Peter Dold1
Fraunhofer CSP, Halle, Germany
1
Corresponding author: e-mail address:

Contents
1. Silicon Feedstock
1.1 Polysilicon: The Base Material for over 90% of All Solar Cells
1.2 The Chemical Path
1.3 Fluidized Bed Reactor
1.4 The Metallurgical Path: UMG-Si
1.5 Different Poly for Different Crystallization Techniques
2. Fundamental Parameters for Silicon Crystallization
2.1 Material Properties, Material Utilization, and Chemical Reactivity
2.2 Numerical Simulation
3. Crystallization Technologies
3.1 Pulling from the Melt: The Cz Technique
3.2 Directional Solidification: Growth of Multicrystalline Silicon
3.3 FZ Growth
4. Summary and Final Remarks
References

1
1
3
6
9
11
12

12
18
19
20
36
45
54
56

1. SILICON FEEDSTOCK
1.1 Polysilicon: The Base Material for over 90% of All
Solar Cells
The roller coaster ride of the polysilicon industry during the last 10 years was
quite extraordinary—even compared with the ups and downs of the semiconductor business over the last half century. The golden age of polysilicon
in the years 2007–2010, when companies could make billions of dollars if
they were able to deliver polysilicon at all, was followed by the severe crush
in the years 2011–2012, when most of the newcomers marched into bankruptcy and disappeared. And, even some of the old ones had to fight heavily
Semiconductors and Semimetals, Volume 92
ISSN 0080-8784
/>
#

2015 Elsevier Inc.
All rights reserved.

1


2

Peter Dold

to survive. During the golden years, spot market prices had reached highs of
200–300 or even 400 US$/kg polysilicon, simply because the market was
swept and the order books of the cell and module manufacturers were full.
The polysilicon industry was not prepared for such a fast ramp-up, investment is high,1 and equipment could not readily be ordered. The longestablished companies either have an exclusive partnership with a specific
equipment manufacturer, or they make the equipment in-house. Production capacity could not easily be ramped up, but once the train was running,
it also could not be stopped so easily and could not be adjusted to the then
changed market situation, partly because typical polysilicon projects take
several years from the financing phase all the way up to full production,
and partly because the players did not want to believe that the silicon
bonanza was over. The huge shortage was followed by a tremendous over
supply with spot market prices as low as 14–16 US$/kg in 2013—which was
below the actual production costs. Today, spot market prices leveled off
around 17–18 US$/kg and no significant changes are expected for the near
future.
As a consequence, all (or at least as good as all) of the new and innovative
approaches for polysilicon refinement, for upgrading metallurgical silicon
(an excellent review was given by Heuer, 2013), or for alternative production methods (compare Bernreuter and Haugwitz, 2010) could not find a
market share and disappeared again. The traditional Chemical Vapor Deposition (CVD)-based Siemens process (Fabry and Hesse, 2012), probably not
the most sophisticated technology for solar-grade-silicon production—but
for sure the most matured technique, was the match winner. A good overview of the market situation and an in-depth analysis of the trends are given
by Bernreuter every first or second year (Bernreuter, 2014).
Basically, two main routes might be distinguished for the refinement of
polysilicon: (I) the chemical path: bringing silicon into the gas phase and
purifying it by distillation, followed by thermal pyrolysis of the gaseous species; and (II) the metallurgical path, where impurities are removed from silicon by mixing it with another metal or with a slag, then let the impurities
segregate into the second phase, separate the different phases somehow
mechanically, and clean the surface of the silicon crystallites by chemical
etching.
1

Back in 2008, a polysilicon plant with a capacity of 10,000 t/a required an investment of at least 1 billion US$. Today, it might be something in the range of 400–600 M$, depending on the location.


Silicon Crystallization Technologies

3

1.2 The Chemical Path
The Siemens process (or modified Siemens process, as many manufacturers
like to call their variation) allows to produce ultrapure polysilicon, with
metallic bulk impurity levels as low as a few tens of ppt (parts per trillion)
or an equivalent of 10–11N. Electrically active elements (donors, acceptors)
are in the ppt range and only carbon and oxygen show up in higher concentrations, where lower single-digit parts per million levels are found. For
semiconductor applications, there is no alternative so far to the polysilicon
produced by the Siemens process.
The Siemens process itself goes back to a patent in the late 1950s filed by
the German electronics company Siemens (Reuschel, 1963; Schweickert
et al., 1961), which stepped out of the polysilicon business long ago. It
can be described by the following process steps:
I. Milling of the metallurgical silicon (purity: 98–99%) into millimeter/
submillimeter particles.
II. Reaction between the fine silicon particles and gaseous HCl at temperatures around 300–350 °C in a fluidized-bed reactor (FBR). The reactor might be heated from the outside, but the chemical reaction is also
strongly exothermic. Mainly copper is used as a catalyst. The main
product is TCS (trichlorosilane, SiHCl3).
III. Fractional distillation of the TCS and the by-products, like metal chlorides, boron, and phosphorus components, and so on. The result will
be ultrapure TCS.
IV. Pyrolytic decomposition of TCS in a bell-jar reactor (Fig. 1) at increased
pressure (normally 6 bar) and temperatures of 1000–1150 °C (Fig. 2).
High-purity polysilicon will be obtained (Fig. 3).

Steps I–III are relatively straightforward, although the installation of the
hardware reaches easily the size and complexity of a huge chemical plant
for typical production capacities of around 10,000 t/a. Step IV is more
difficult:
– The high temperature required for the silicon deposition is rather energy
intensive. The silicon rods on which the deposition takes place are
directly heated by an electrical current.
– Deposition rates on these U-shaped rods are on the order of 0.5–1 mm/h
(layer growth); beyond this rate, the rod morphology becomes unstable
and so-called “popcorn” or “broccoli” growth takes place.
– Only part of the TCS decomposes to silicon, and a significant part reacts
with the HCl formed during the deposition to STC (silicon tetrachloride,


4

Peter Dold

Figure 1 Schematic drawing of a Siemens bell-jar reactor for polysilicon deposition
from the gas phase. The U-shaped silicon rods are heated up to a temperature of
1000–1150 °C by direct current. The process gas enters and leaves the reactor chamber
through the base plate. By courtesy of Wacker Chemie AG.

Figure 2 Silicon deposition from TCS in a research reactor. Left: beginning of the deposition, right: after 30 h process time. In particular, in the elbow area, current and temperature distribution might be nonuniform.

SiCl4). Decomposition of STC is too low at the typical rod temperatures
in the bell-jar; therefore, it has to be removed from the reactor and has to
be back-converted into TCS.
In former times, back-conversion of STC to TCS was carried out mainly
in thermal STC converters (Paetzold et al., 2007; Sirtl et al., 1974), and the

process is also referred as “hydrogenation.” At high temperature in a hot
carbon rod reactor (>1200 °C), STC reacts with hydrogen back to TCS
(and other by-products), an another energy-intensive process step. Nowadays,


Silicon Crystallization Technologies

5

Figure 3 Polysilicon rods in an industrial multirod Siemens reactor. The rod length
might reach more than 3 m, at a maximum diameter of around 180 mm.

“hydrochlorination” is more and more used (see, e.g., />products-and-services-trichlorosilane-and-silane-production-packagesHydrochlorinationTCS-Plant.htm), especially by the newcomers. In this
process, hydrogen, metallurgical grade silicon, and STC are introduced into
an FBR. At high pressure (20 bar and more) and temperature T > 500 °C,
TCS is formed.
The Siemens process is a batch process. The U-shaped rods in the bell-jar
are heated with high current, starting with 6–8 mm starter rods (or slim
rods). Today, most of the slim rods are prepared in so-called slim rod pullers
by the pedestal method: The top area of a cylindrical silicon rod of some
4–600 in diameter is melted from above by an RF inductor with at least
one hole in center. Through this hole, the slim rod is pulled, comparable
to a crucible-free Czochralski (Cz) approach. In such a way, slim rods of
several meters are pulled, with pulling rates which might easily surpass half
a meter per hour.
At the beginning of the deposition process, just a few tens of amperes
are needed to keep the thin starter rods at deposition temperature.
A certain challenge is to bring the starter rods to temperatures where the
intrinsic carrier concentration of silicon becomes high enough that a decent
current can flow. To bridge the gap from room temperature to the required

300–400 °C, where the rods become electrically conductive, various
methods are in use: (I) preheating the starter rods with radiation lamps,
(II) use of medium- or high-voltage power supplies (see, e.g., http://
www.aegps.com/en/res/power-controllers/polysilicon-systems/), or (III)
use of slightly predoped starter rods (Aulich and Schulze, 2009). The latter


6

Peter Dold

is not an option for electronic grade material, but quite an option for solargrade polysilicon. At the end of the process cycle, when the rods have reached
their final size of 150–180 mm in diameter, several thousand amperes are
required to keep them at the specific deposition temperature. The whole
cycle takes about 100 h, depending on the deposition rates and the final size.
The maximum diameter is limited by the temperature gradient between the
rod surface (which has to stay around 1100 °C and which cools down by radiation and by convection) and the hotter core of the rod, where the current
flows preferentially. If the core or the elbow areas become too hot, there is a
risk that the silicon is melting, which results in a strong decrease of the electrical resistivity, and finally a local shortcut and a burned-through rod.
Some 10 years ago, with lower deposition rates, smaller reactors, and
less-optimized processes, power consumption to produce 1 kg of silicon
was in the range of 150–200 kWh/kg (including STC conversion). Today,
state-of-the-art reactors with some 48–72 rods (even 96 rod reactors are on
the market), and an annual capacity of some 400 t of silicon, high deposition
rates, integrated hydrochlorination, and proper debottlenecking, the power
consumption is as low as 50–70 kWh/kg. Some manufacturers are claiming
that they can even reach values below 50 kWh/kg.
As already mentioned in the beginning, the Siemens process is now very
matured, which also means that we cannot expect huge progress steps anymore, and further improvements will be rather incremental and less revolutionary. A significant cost reduction is promised by the FBR technology.

1.3 Fluidized Bed Reactor
In contrast to the batch-type Siemens process, the FBR operates in a continuous mode. Small seed particles (high-purity silicon with diameters of
some tens of micrometers) are fed into a heated reactor, a strong gas flow
(either TCS or silane, mixed with hydrogen) from the bottom part of the
reactor keeps the particles floating (Fig. 4). An excellent overview was given
in Ydstie and Du (2011). Reaction with TCS (or silane as used in the case of
the company REC) lets the silicon particles grow, until they reach a critical
mass and sink to the bottom area in the form of granules (or beads; Fig. 5),
where they can be harvested easily. The technology has a certain charm and
several advantages are obvious:
– Continuous operation—minimized downtime.
– High deposition rates due to a large silicon surface; different to the Siemens process where only toward the end of the process a large deposition


Silicon Crystallization Technologies

7

Figure 4 Sketch of an FBR reactor: seeds entering the chamber from the top are levitated by the strong gas stream and settle down once they have reached a certain
weight. At the bottom, the final granules are taken out of the process continuously.

Figure 5 Solar-grade silicon: poly chunks (left-hand side) and granular material (righthand side).


8

Peter Dold

surface is available, in the case of FBR it is provided right from the
beginning.2

– Significantly lower energy consumption, e.g., REC claims some 80–90%
less energy consumption for their silane-based FBR process compared to
TCS-based Siemens reactors ( />rec-silicons-fluidized-bed-reactor-process/).
– The spherical silicon beads are ready to be shipped (and filled into the
crystallization crucible right away), and no crushing or mechanical handling is required.
Of course, there are some obstacles to manage and one of the biggest is the
purity. The moving particles in the reactor might touch the reactor wall
where they might be contaminated, especially when steel-based/metalbased wall materials are in use. Today, granular silicon is about two to three
orders higher in metals than high-class Siemens silicon. Further, the swirling
and spinning in the reactor and the subsequent material handling produce
some fines in the form of a black dust, which should be removed or washed
off; otherwise, the acceptance of the material suffers. Finally, a major problem is the melting of granular silicon in the subsequent crystallization process: it has a tendency for popping and splashing, and small silicon droplets
might be found several centimeters away from the crucible. Most likely, this
is related to process gases (hydrogen and/or chlorines) stored in the granules
(Kajimoto et al., 1991) or it is related to stress at the interface seed shell. During crystal growth, evaporation of hydrogen might lead to a disturbed melt
surface during the Cz process. Release of chlorine is affecting the crystallization hardware, of course. The popping problem might be reduced by
proper charging of the crucible, blending the granular material with normal
polysilicon chunks, and avoiding that the granules are exposed to the free
crucible surface. In the case of recharge processes, the splashing problem
is more difficult to overcome.
Recently, quite some R&D activities are noticeable on FBR technology. For sure, it will not push the Siemens process out of the market, but
it might gain a certain share of the poly market. According to the 2014
ITRPV report, today, granular silicon has a market share of some 15%
( It has still a significant cost
savings potential, probably much more than the Siemens process. Combined
2

Just 1 kg of granules provides a reactive surface of about two-and-a-half square meters, assuming an
average diameter of 1 mm. On the other hand, a full-size Siemens U-rod of 150 mm in diameter
and a total length of 6.5 m possess a surface of about 3 m2 at a weight of 280 kg.

Silicon Crystallization Technologies

9

with broken poly chunks from Siemens reactors, an improved crucible fill
factor is achieved, an improvement of 29.3% was reported (REC Silicon
Inc., 2013), the small granules fill perfectly the space between the larger
chunks, and, the filling of crucibles with granules is fast.

1.4 The Metallurgical Path: UMG-Si
Over many years, photovoltaic industry (PV) used the leftovers from the
semiconductor industry, which was in most cases ultrapure poly-feedstock,
cutoffs from Cz ingots, and so on. The base material was in the range of 9–10
or even 11N purity. Using it for multicrystalline ingots, there is hardly any
difference noticeable whether 6N or 8N or 10N polysilicon is used. Therefore, the question seemed appropriate: Why not use silicon of purity just
clean enough for cell processes and simplify the purification process accordingly? The metallurgical path seemed highly promising: easy to scale,
low-energy consumption, low Capital Expenditures (CAPEX)—but still
delivering a fully usable product. Dozens and dozens of groups and companies tried it worldwide (Bernreuter and Haugwitz, 2010), and only about
two survived on a scale somewhere between pilot and full production:
Silicor Materials and Elkem (a subsidiary of China National Bluestar Group
Co. Ltd.). The U.S.-based company Silicor Materials (former Calisolar) had
purchased the UMG-process from the Canadian company 6N. The 6N process (Nichol, 2011) is based on the alloying of silicon with aluminum3: Metallurgical grade silicon of some 98–99% purity is mixed with aluminum, and
the hypoeutectic mixture becomes liquid in the range of 900–1000 °C,
depending on the silicon concentration. The eutectic temperature itself is
577 °C, with a silicon concentration of 12.2 at%. Cooling down the hypoeutectic mix, the excess silicon forms small crystallites or flakes, embedded in
the liquid Al–Si melt. In silicon, all metals show small segregation coefficients4 and, consequently, are enriched in the melt, or better, are accumulated in the solid–liquid boundary layer. The point with the accumulation
within the boundary layer is a bit problematic: a proper separation of the
silicon crystallites from the melt is essential and a chemical etching step is

required to dissolve the metals. To get a good cost structure, the residual
3

4

Instead of aluminum, tin would also be an option, but aluminum can be separated from silicon more
easily, either mechanically (e.g., centrifugation) or chemically etched off. Basically, all materials used in
former times for the liquid-phase epitaxy (LPE) of silicon could be used for alloying with silicon;
restrictions result mainly from practical considerations like availability in large quantities and price.
The lower temperature of the Al–Si melt compared to pure liquid silicon reduces the segregation coefficients even further (e.g., Morita and Yoshikawa, 2011).


10

Peter Dold

Al–Si melt—still slightly hypoeutectic—has to be sold, but there is a market
for this kind of alloys. The main trouble makers are, besides the proper
removal of the aluminum, which might be trapped in inclusions, the elimination or reduction of boron and phosphorus. Recently, plans for a 16,000 t
plant in Island had been released (Kaes et al., 2014).
Core features of the ELKEM process are chemical etching and slag treatment (Ceccaroli and Friestad, 2005; Heuer, 2013; Schei, 1998; Wang et al.,
2014). A calcium-based slag is used, and during the cooling-down phase,
most of the impurities are accumulated in the slag. After solidification,
the slag and the impurities are etched off and purified silicon is obtained.
The process works very well for the metallic impurities, but again, boron
and phosphorus are still present and the material is somewhat compensated.
Boron and phosphorus had been the greatest bottleneck for all the different UMG processes or better: their issue of failure. Boron shows a segregation coefficient of k0 ¼ 0.8 (somewhat lower at reduced temperatures) and
phosphorus 0.35. Removing boron and phosphorus simply by segregation is
not an option. All the methods developed so far are either costly or complicated (or both):
– oxidizing the boron out (the Becancour/Timminco process): huge loss of

silicon (Leblanc and Boisvert, 2008).
– removing it by slagging: expensive and risk of introducing other impurities (Ceccaroli and Friestad, 2005; Schei, 1998; Wang et al., 2014).
– gettering, forming a metal boride (e.g., TiB2): not efficient enough
(Yoshikawa et al., 2005).
– using low boron raw materials (SolSilc or SolSil process): helps significantly but requires a clean reduction process (Dosaj and Hunt, 1981;
Geerligs et al., 2002).
Phosphorus might be reduced by vacuum treatment of the melt or by plasma
(Alemany et al., 2002; Delannoy et al., 2002), but both approaches are cost
intensive. Work-around solutions had been suggested using compensated
feedstock (i.e., silicon-containing boron and phosphorus/adding boron or
phosphorus during the solidification; Dethloff and Friestad, 2007) or add
some gallium (Forster et al., 2011; Kirscht et al., 2010) in order to compensate the accumulated phosphor toward the end of the block, but the point is,
so far, all UMG products are not reaching the purity of CVD-based Siemens
or FBR material. Today, they are good with respect to metals, but boron in
particular is still an unsolved problem. And even if the user is adding boron
during crystallization, and maybe much more than the remaining boron
level in the UMG-feedstock had been, the product can be sold on the open
market only with a certain discount.


Silicon Crystallization Technologies

11

Today, UMG-Si suffers a hard time, but if the boron–phosphorus problem can be solved, it might be the path with the lowest cost structure, the
lowest CAPEX, and the easiest to scale up or down, according to the market
requirements.

1.5 Different Poly for Different Crystallization Techniques
1.5.1 Mono Growth, Single Batch Mode

In monocrystalline growth by the standard Cz method, the trend goes to
high-efficiency cells. Therefore, n-type cell structures will very likely gain
market shares. For these applications, high-quality wafers are essential and
polysilicon from CVD processes will be the standard. A certain mixing with
granular material is possible, but only if it is low in metals and low with
respect to trapped gases. During mono-crystal growth, the risk for structure
loss is always given and ingot producers try to avoid any potential source
which could jeopardize their yield. Since high-quality material is available
in sufficient quantities right now, consumers favor 9N or 10N poly material.
1.5.2 Feeding and Multipulling
Feeding and multipulling is used primarily for mono growth, although certain activities are visible in the multicrystalline sector (Mu¨ller et al., 2009),
too. Polysilicon for feeding processes has to show excellent transport properties, with a minimum risk for clogging and low material abrasion. For
mono-ingot growth, the introduction of particles has to be avoided and
accumulation of impurities in the residual melt has to be minimized. Melting should be smooth and fast. Theoretically, granular material would be
perfect for feeding, and the spherical shape and the rather small size give
them perfect transport properties. In practice, the high dust load, increased
metal concentrations, and trapped process gases (Kajimoto et al., 1991)
might cause problems. Problems, the poly manufacturers still have to work
on. An alternative to granular material are crushed chunks: they are available
from so-called “size 0” on (smaller than 10 mm, often rather chip-like) and
the maximum size for feeding should not exceed some 10–20 mm; otherwise, the impact and the splashing when the solid silicon hits the melt might
become serious.
1.5.3 Standard Multicrystalline Casting
The specifications for the polysilicon feedstock used for multicrystalline
growth are lower and mainly driven by cost reduction. A few particles or
a certain metal background are not affecting the quality of the ingot in
the same way as it would be in Cz growth. One reason is that in any case,


12

Peter Dold

the crucible and the crucible coating release a significant amount of impurities during the crystallization process anyway (Schubert et al., 2013).
Therefore, quite often a mix is used, composed of standard solar-grade polysilicon mixed with second-grade poly (8N and lower). Furthermore, most of
the side slabs of the ingot are recycled in order to minimize material losses.
Most of the granular material is used for multicrystalline growth, where it is
blended with poly chunks.
1.5.4 Float Zone
Float-Zone (FZ) growth requires specific feed rods: crack-free, smooth surface, minimum bending, high-purity, free of any oxide or nitride layers and
with uniform, microcrystalline morphology, to mention just the most
important characteristics. FZ feed rods are produced in CVD reactors dedicated to this purpose, and this requires special know-how with respect to
the control of the process gas composition and flow arrangements, as well
as a uniform temperature distribution and a specific cool-down procedure
(Freiheit et al., 2010). Only a very limited number of polysilicon producers
are able to deposit feed rods for FZ applications; thus, the availability is limited, production is low, and prices are high. Alternatives will be discussed in
Section 3.3.

2. FUNDAMENTAL PARAMETERS FOR SILICON
CRYSTALLIZATION
2.1 Material Properties, Material Utilization, and Chemical
Reactivity
Silicon shows some exceptional material properties which, on the one hand,
allow the growth of dislocation-free single crystals of several hundreds of
kilograms, but, on the other hand, require highly sophisticated crystallization strategies. The most relevant ones will be discussed in the following.
Essential material data for the analysis of silicon crystallization technologies
are summarized in Table 1 (after Zulehner et al., 2012).
The density of solid and liquid silicon differs by 10%. Silicon shows a
similar density anomaly like water: at the phase transition to the solid, it
expands. This property prevents the use of any kind of closed crucibles,

and the sufficient space for volume expansion is always critical. The density
change might be used for the measurement of the solidification rate during
directional solidification, as we will see later on, but it bears a significant risk


13

Silicon Crystallization Technologies

Table 1 Specific Material Parameters of Silicon
Melting Point
1410 °C
Density

Solid

2.3 g/cm3

Liquid

2.53 g/cm3

Heat capacity

20 °C

0.7 J/g K

1400 °C

1.0 J/g K (Rodriguez et al., 2011)

Latent heat of fusion

50.66 kJ/mol
3.3 kJ/cm3 (Rodriguez et al., 2011)

Thermal expansion coefficient 2.6 Â 10À6 /K
Electrical conductivity

Liquid

1.33 Â 106 ΩÀ-1 mÀ1 (Brandes and Brook, 1992)

for Cz and for vertical gradient freeze (VGF) growth. In case of a power failure, the melt freezes from top to bottom and will unavoidably crack the crucible and will spill liquid silicon into the furnace chamber.
The heat capacity for solid silicon is in the range of 0.7–1.0 J/g K and
might be described by a second-order polynomial fit (Gurvich et al.,
1990). Whether there is an anomaly around 560 K as described in Glazov
and Pashinkin (2001) or not does not really affect crystallization since it
was only described for slow heating rates, not relevant for our considerations. Quite significant is the high value for the latent heat of phase change.
Values given in literature vary somewhat in the range of 40–50 kJ/mol (or
3.3–4.2 kJ/cm3, see Table 1), but in any case, it is extremely high. Thus, a
large amount of energy is required for the melting process, which has to be
removed during crystallization. As a matter of fact, more energy is required
for the melting itself than for the heating from room temperature to the
melting point. Heating needs approximately 0.33 kWh/kg silicon (assuming
an average heat capacity of 0.85 J/g K) and melting requires 0.5 kWh/kg
silicon (assuming 50.6 kJ/mol for the latent heat of phase change, according
to Zulehner et al., 2012). Typical values for the crystallization by the Cz and
the VGF technique are summarized in Table 2.

In the case of Cz growth, the heat is released by radiation mainly, but in
case of VGF, it has to be extracted by heat conduction through the bottom of


14

Peter Dold

Table 2 Energy Balance for Heating, Melting, and Crystallization of Silicon
Heating and melting:
Czochralski

150 kg crucible load

(Heating: 49.5 kWh; melting: 75 kWh)
total: 124.5 kWh

Vertical gradient freeze

G4—250 kg load

(Heating: 82.5 kWh; melting: 125 kWh)
total: 207.5 kWh

G6—800 kg load

(Heating: 264 kWh; melting: 400 kWh)
total: 664.0 kWh

Crystallization:

Czochralski

Diameter: 900 ; growth rate: 1 mm/min 5.5 kg/h

À2.75 kWh

Vertical gradient freeze

G4—growth rate: 1 cm/h

10 kg/h

À5 kWh

G6—growth rate: 1 cm/h

22 kg/h

À11 kWh

the crucible, where the crucible made of sintered quartz ceramic acts as an
insulation barrier.
Discussing the crystallization of silicon for PV application, it is helpful to
have a look at the actual size and geometry of the ingots (Table 3). Today,
standard wafer size is 156 Â 156 mm2, either full square or pseudosquare in
case of certain mono ingots. Pseudosquare refers to the geometry with missing corners: to do without the four corners (the missing triangles have the
size of approximately 10 Â 10 Â 15 mm) reduces the active cell area by less
than 1% but allows to reduce the ingot diameter from 222 mm down to
206 mm (also referred to 900 vs. 800 , even this is not exactly correct). Only
a few cell manufacturers are still using the 125 Â 125 mm2 mono wafers.

It provides a higher utilization factor of the crystallized material but requires
more handling steps further down the manufacturing process chain in order
to get the same amount of active cell area. Quite likely, they will disappear
sooner or later from the market. In both cases, mono as well as multi, a significant amount of crystallized material cannot be used as wafers. In the case
of mono, it is due to the fact that the ingot is cylindrical but the wafer is
rectangular; in the case of multi, it is due to the impurity-rich areas near


15

Silicon Crystallization Technologies

Table 3 Geometry and Mass Balance for Czochralski (800 Pseudosquare) and VGF
(G6) Growth of Silicon
Czochralski

Initial charge

150 kg

Top–tail

7 kg

Residual melt

3 kg
00

Ingot cross-section area (8 )

333 cm2

Wafer area (pseudosquare)

241 cm2 (¼73%)

Side slabs

38.6 kg

Length of body

182 cm

Pseudosquare brick (weight)

101.4 kg (¼68%)

VGF

Initial charge (G6)

800 kg

Crucible

98.6 Â 98.6 cm2

Area for bricking

93.6 Â 93.6 cm2

Ingot height

35 cm

Brick height for wafering

31 cm

Bricks for wafering (weight)

625 kg (¼78%)

the walls, the bottom, and the top. For pseudosquare mono growth, the area
of the side slabs amounts to 27%, for full square even 37%. Adding some 7 kg
for the top and tail part and some 3 kg for the residual melt, a 150 kg crucible
charge results in 102 kg of bricks for wafering (pseudosquare) or 88 kg for
wafering full square, respectively. The material is not lost but will be
recycled, apart from the residual melt, which is difficult to separate from
the crucible. Nevertheless, it is affecting the energy balance. In the case
of VGF, the situation is slightly better, but still, about 2–2.5 cm from all
the edges have to be removed, which results in an optimistic scenario in
a material utilization of 73% (G4) and 77% (G6), respectively. Part of the
removed side slabs will be recycled, but they are somewhat contaminated
with iron, chromium, and copper.
The cutoff size of the edge areas of VGF blocks are average values and
might vary somewhat from manufacturer to manufacturer. In the case of
VGF, upscaling will improve the utilization factor somewhat, but the larger

16

Peter Dold

melt volumes and the longer process times also increase the width of the surface boundary layers with high metal contamination and low carrier lifetimes
(“electrically dead zone”). The rather large loss of material was always a
strong motivation for direct wafer casting technologies (until the final wafers
are ready for the cell process, an additional 40–50% of the silicon from the
ready-to-cut bricks will get lost in the wire saw). However, as long as the
direct wafer technologies do not reach the same thickness as the wafers from
the multiwire process, which is in moment between 150 and 180 μm, there
is not a real advantage from the viewpoint of material utilization. In any case,
the rather low material utilization factor for crystalline silicon wafer technology is a significant cost driver and it will be an important task for the future to
improve it.
An important material property of liquid silicon is its high chemical reactivity. In contrast to solid silicon, which is protected by an oxide passivation
layer and thus is very easy to handle, liquid silicon is a highly aggressive substance. So far, no material is known, which is fully inert against silicon. Even
in the oxidized state as Si4+ (e.g., as SiO2, SiC, or Si3N4), there is always an
interaction with the melt and a certain dissolution or formation of precipitates can be observed. In particular, in the case of SiO2, the reaction will not
stop since the oxygen vapor pressure of SiO is rather high and it will evaporate at the free melt surface. Thus, the equilibrium always favors the further
dissolution of the quartz crucible. The dissolution rate for fused quartz
glass in contact with liquid silicon was reported to be in the range of
1.15 Â 10À5 cm/min in the bulk of the melt and up to 8.4 Â 10À5 cm/min
at the triple point melt–crucible–gas (Chaney and Varker, 1976).
A correlation with melt stirring was reported by Hirata and Hoshikawa
(1980) and a certain correlation to the boron concentration was found by
Abe et al. (1998), but the reported values were all in the same range. To
get a better idea of the amount of quartz glass dissolved during the course
of the growth run, we might assume a process time of 50 h and an average

crucible surface in contact to the melt of 2300 cm2 (for a 2400 crucible; at the
beginning, it will be around 5600 cm2 but decreases continuously). The crucible wall would be reduced by about 0.35 mm on average, which correlates
to some 200–250 g of crucible material dissolved into the melt. The corrosion rate of the quartz glass crucible is a fundamental issue for multipulling or
for continuous Cz processes, and the development of high corrosionresistant crucible materials is essential. In the case of multicrystalline growth,
the crucible is protected by an Si3N4 coating, which cannot be used for Cz
growth, of course. Silicon nitride particles would result in structure loss.


17

Silicon Crystallization Technologies

Table 4 Classification of Binary Silicon Phase Diagrams with Respect to the Formation of
Solid Solutions, Silicides, or Eutectics

(A) Solid solutions

Ge

(B) Eutectics (low solubility in the solid)

Al, Ag, Au, Bi, Pb

(C) Intermetallic compounds/silicides

Cu, Ta, Fe, Mg, Mo,
Ni, Ti,

(D) Very limited solution in the solid, and complete
solubility in the liquid

Sn, In, Zn, Ga

With regard to metals, we might distinguish four classes (Table 4): silicon
might form (A) solid solutions, (B) eutectics, and/or (C) intermetallic components, or (D) shows a complete mixing in the liquid state, but as good as
no mixing in the solid. Quite often, eutectics and intermetallic components
are found in one phase diagram and sorting into the different classes is not
always a clear case. However, it helps to understand the interactions and
chemical reactions.
Some of the silicides have rather high melting points, e.g., MoSi2
(Tm ¼ 2020 °C) or TaSi2 (Tm ¼ 2040 °C). However, the tolerable levels
of these metals for solar applications are extremely low, and concentrations
in the ppt range affect the cell efficiency already heavily (Coletti et al., n.d.;
Davis et al., 1980). Metals from class (D) are used for LPE and class (B) or
class (D) elements are candidates for the use in silicon refinement.
Whereas the high reactivity in the liquid state makes it difficult to find the
right crucible material, the low solubility in the solid helps quite significantly
for purification. Despite a few exceptions, most elements show small segregation coefficients (the segregation coefficient k0 defines the ratio between
the concentration in the solid and the concentration in the liquid, under the
assumption of thermodynamic equilibrium) and will not be incorporated
into the crystal but will accumulate in the liquid boundary layer ahead of
the solid–liquid interface (Table 5). One exception is boron (k0 ¼ 0.8).
The large segregation coefficient of boron favors a uniform dopant distribution for p-type ingots—but it is quite troublesome for silicon purification.
A second exception is oxygen. With a segregation coefficient around 1, all
the oxygen near the solid–liquid interface will be incorporated into the crystal. To prevent this, the transport of oxygen toward the interface has to be
reduced, which is possible by proper melt flow configurations. The oxygenrich melt should be moved away from the growing interface and should be


18

Peter Dold

Table 5 Segregation Coefficient k0 for Various Elements in Silicon
Element
k0
Element
Element
k0

1 Â 10

À6

8 Â 10

À6

Al

2 Â 10

À3

Ga

8 Â 10

À3

As

0.3

Ge

Au

2.5 Â 10À3

B

0.8

Ag

À8

Fe

k0

O

1–1.25

P

0.35

0.33

Pb

2 Â 10À3

In

4 Â 10À4

S

10À5

Li

0.01

Sb

0.023

Se

<10À8

À6

Mg

8 Â 10

7 Â 10À4

Mn

10À5

Sn

0.016

C

7 Â 10À2

Mo

4 Â 10À8

Ta

10À7

Co

8 Â 10À6

N

10À7

Te

8 Â 10À6

Cr

<10À8

Na

2 Â 10À3

Cu

4 Â 10À4

Ni

8 Â 10À6

Ba

<10

Bi

transported toward the free surface, where the oxygen (in form of SiO) can
evaporate and subsequently be removed from the growth chamber.

2.2 Numerical Simulation
Today, numerical simulation is a standard tool for industrial crystallization.
In most cases, it is an integral part for any hardware or hot-zone development. It helps to understand the heat fluxes (and losses), the material transport, and reveals which areas are crucial for the optimization of the energy
consumption. The first attempts for computational simulation of crystallization processes go back to the 1970s (e.g., Kobayashi, 1978). At that time, it
was still restricted to 2D axisymmetrical calculations based on finite differences and nonstructured grids. Now, modern software packages are running
on PC systems and are able to handle transient processes, 3D flows, and some
of them even chemical reactions. In particular, for the crystallization of silicon, commercial codes are now tailored to specific growth technologies.
Examples for software packages dedicated to silicon crystal growth are,
e.g., CGSim ( Smirnov and
Kalaev, 2009), FEMAGSoft ( Collet et al.,
2012), or CrysVUn (Kurz et al., 1999; />de/abteilungen/kristallzuechtung/crysmas.html) to mention just some of
them, or of course ANSYS ( as a more general


Silicon Crystallization Technologies

19

software code for any kind of fluid dynamic problems. Quite often, the user
is enabled to add and integrate user-based subroutines, e.g., in order to simulate external magnetic fields. Therefore, numerical simulations became a
reliable and indispensable tool for any crystal grower. Nevertheless, certain
points have to be kept in mind when analyzing the results of numerical
simulations:
– In the simulation, the heat transfer is always idealized. In reality, it will be
reduced due to small gaps, surface layers, cracks, etc., or it might be
increased by altered material properties, enhanced emissivities, etc.
– Today, the material data are known much better than some 20 years ago.
Still, they are often idealized or not available as a function of the temperature. Furthermore, they might change over time.
– Materials exposed to high temperatures and aggressive media will change
their structure and their surface. In particular, surface corrosion and surface coatings have a huge impact on the temperature. Changes in the

emissivity affect the radiative heat transfer, which has a T4 impact on
the heat flux.
– The different length scales are difficult to handle. We have to deal with
macroscopic features in the meter range, but at the same time, chemical
reactions and surface-related phase changes have to be resolved in the
micrometer or even submicrometer range.
– Certain features have a 3D or a time-dependent characteristic. VGF is
nonaxisymmetric by definition. The large melt volumes result in large
Grashof and Reynolds numbers, indicating time-dependent 3D flow
structures.
– For certain aspects like defect formation, structure loss, or grain formation, the physics behind is not fully understood yet and the physical
models are not always adequate.
As long as these limitations are kept in mind, numerical simulations are an
extremely helpful tool. Most software programs became rather user-friendly
and the profile of a typical operator is shifting from a highly specialized scientist toward an engineer with experimental background. But in any case,
the proper validation of numerical results by experimental data is absolutely
crucial.

3. CRYSTALLIZATION TECHNOLOGIES
In the following chapter, the main technologies for silicon crystallization are described in detail: the Cz technique used for the majority of all


20

Peter Dold

mono ingots, the directional solidification or VGF method used for multicrystalline ingot production, and finally the FZ technique, a method well
established for the crystallization of electronic grade ingots, whenever low
oxygen material is required, but not yet adapted to the PV market. Also,
FZ would provide many benefits, and there are certain bottlenecks which

prevented the cost-competitive introduction of FZ wafers for solar cell
manufacturing until now. One serious problem is the availability of suitable
feedstock.
Other crystallization techniques for silicon could not gain a significant
market share so far. For example, the electromagnetic casting had made significant progress; e.g., the Japanese company Sumco had shown impressive
pictures of 7 m tall ingots (“taller than a giraffe”—as they claimed in their
portfolio and their webpage; Kaneko, 2010; Kaneko et al., 2006;
SUMCO Annual Report, 2008), but the technique was considered not cost
competitive and production was stopped. There had also been many activities with respect to sheet growth (EFG - Edge-defined Film Fed Growth by
Schott (Mackintosha et al., 2006), String Ribbon (van Glabbeek et al., 2008)
by Evergreen/Sovello, to mention just the most prominent ones), but so far,
none of them had really been able to reach the cost structure and/or the
quality of Cz and VGF. We will therefore focus in the following on the predominant and most promising PV silicon bulk crystallization technologies.
A detailed discussion of the different ribbon and foil techniques is provided
by Rodriguez et al. (2011).

3.1 Pulling from the Melt: The Cz Technique
Initially, pulling a monocrystalline material from a melt goes back to
Czochralski (1918). Although the initial intention was not the growth of
large single crystals but the measurement of solidification velocities and
latent heat, it was soon realized that this method was perfectly suited for
the pulling of monocrystalline ingots. There is no direct interaction of
the growing crystal with the crucible material, and in situ observation of
the success (or failure) of the growth process is easily carried out. An excellent overview about the historical development of silicon pulling from the
melt was given by Zulehner (1999); unfortunately, to the knowledge of the
author, it is only available as a conference proceeding paper.
3.1.1 Standard Cz Growth
Since more than half a century, the Cz technique is the workhorse for the
semiconductor industry. At the very beginning, there was a competition