Documente Academic
Documente Profesional
Documente Cultură
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
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
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 photovol-
taics is considered to be a cornerstone technology for this truly global trans-
formation 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 sys-
tem 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!
Silicon Crystallization
Technologies
Peter Dold1
Fraunhofer CSP, Halle, Germany
1
Corresponding author: e-mail address: peter.dold@ise.fraunhofer.de
Contents
1. Silicon Feedstock 1
1.1 Polysilicon: The Base Material for over 90% of All Solar Cells 1
1.2 The Chemical Path 3
1.3 Fluidized Bed Reactor 6
1.4 The Metallurgical Path: UMG-Si 9
1.5 Different Poly for Different Crystallization Techniques 11
2. Fundamental Parameters for Silicon Crystallization 12
2.1 Material Properties, Material Utilization, and Chemical Reactivity 12
2.2 Numerical Simulation 18
3. Crystallization Technologies 19
3.1 Pulling from the Melt: The Cz Technique 20
3.2 Directional Solidification: Growth of Multicrystalline Silicon 36
3.3 FZ Growth 45
4. Summary and Final Remarks 54
References 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 semi-
conductor 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 bank-
ruptcy and disappeared. And, even some of the old ones had to fight heavily
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, invest-
ment is high,1 and equipment could not readily be ordered. The long-
established companies either have an exclusive partnership with a specific
equipment manufacturer, or they make the equipment in-house. Produc-
tion 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 produc-
tion methods (compare Bernreuter and Haugwitz, 2010) could not find a
market share and disappeared again. The traditional Chemical Vapor Depo-
sition (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 over-
view 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 spe-
cies; and (II) the metallurgical path, where impurities are removed from sil-
icon 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 bil-
lion US$. Today, it might be something in the range of 400–600 M$, depending on the location.
Silicon Crystallization Technologies 3
Figure 2 Silicon deposition from TCS in a research reactor. Left: beginning of the depo-
sition, right: after 30 h process time. In particular, in the elbow area, current and tem-
perature distribution might be nonuniform.
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.
is not an option for electronic grade material, but quite an option for solar-
grade 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 radi-
ation 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 elec-
trical 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 any-
more, and further improvements will be rather incremental and less revolu-
tionary. A significant cost reduction is promised by the FBR technology.
Figure 4 Sketch of an FBR reactor: seeds entering the chamber from the top are levi-
tated 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 (right-
hand side).
8 Peter Dold
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.
the crucible and the crucible coating release a significant amount of impu-
rities during the crystallization process anyway (Schubert et al., 2013).
Therefore, quite often a mix is used, composed of standard solar-grade poly-
silicon 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.
for Cz and for vertical gradient freeze (VGF) growth. In case of a power fail-
ure, the melt freezes from top to bottom and will unavoidably crack the cru-
cible 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 consider-
ations. 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
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 156 156 mm2, either full square or pseudosquare in
case of certain mono ingots. Pseudosquare refers to the geometry with miss-
ing corners: to do without the four corners (the missing triangles have the
size of approximately 10 10 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 125 125 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 sig-
nificant 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
Silicon Crystallization Technologies 15
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
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 sur-
face 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 technol-
ogy 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 reac-
tivity. 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 sub-
stance. 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 precip-
itates 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 evap-
orate 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
1.15 105 cm/min in the bulk of the melt and up to 8.4 105 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 cru-
cible 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 corro-
sion rate of the quartz glass crucible is a fundamental issue for multipulling or
for continuous Cz processes, and the development of high corrosion-
resistant 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.
Silicon Crystallization Technologies 17
Table 4 Classification of Binary Silicon Phase Diagrams with Respect to the Formation of
Solid Solutions, Silicides, or Eutectics
With regard to metals, we might distinguish four classes (Table 4): silicon
might form (A) solid solutions, (B) eutectics, and/or (C) intermetallic com-
ponents, 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 segre-
gation 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 distribu-
tion 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 crys-
tal. To prevent this, the transport of oxygen toward the interface has to be
reduced, which is possible by proper melt flow configurations. The oxygen-
rich melt should be moved away from the growing interface and should be
18 Peter Dold
transported toward the free surface, where the oxygen (in form of SiO) can
evaporate and subsequently be removed from the growth chamber.
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 sim-
ulate 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 temper-
ature. 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 sur-
face 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 forma-
tion, 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 sci-
entist 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 crystalliza-
tion 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 multi-
crystalline 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 sig-
nificant 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 activ-
ities 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 pre-
dominant 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).
between FZ and Cz. The crucible-free growth and the lack of any graphite
or insulation material seemed to be in favor of the FZ method, but after the
development of proper quartz glass crucibles and hot zones based on purified
graphite, the easy scale-up option for Cz and the easier handling and oper-
ation of the crystallization process shifted the pendulum clearly toward Cz.
The point that for most semiconductor devices a certain oxygen concentra-
tion is beneficial for device manufacturing gave Cz additional credit and
soon the Cz technique had a market share for semiconductor ingots of more
than 90%. Exceptions are the low-in-oxygen wafers for power electronics,
which are still the domain of FZ silicon. A schematic drawing of a Cz puller
is given in Fig. 6, and state-of-art machines for the growth of up to 1200
ingots are shown in Fig. 7.
The basic features of the Cz technique might be summarized as follows: the
feedstock material (it might be broken chunks, or chips, or etched cutoffs, or
even granules) is loaded into a quartz glass crucible, which is sitting in a graphite
susceptor (Fig. 8). The material is heated by a graphite-based resistance heater;
in most cases, it is a single, fence-shaped heater, sitting in a fixed position
and surrounding the crucible. The crucible can be moved up and down, sim-
ilar to the seed crystal. The monocrystalline seeds of specific crystallographic
orientation, which is machined from a dedicated Cz crystal, are clamped in
a seed holder and connected to a stainless steel rod or a wire. Both the crucible
and the seed/growing crystal are rotating, normally in counter directions.
The whole assembly is sitting inside a vacuum chamber. During the pro-
cess, a continuous argon flow is purging the puller, and the argon flow is in
Ingot
Heat shield
Heater
Quartz
crucible Si-melt
Figure 6 Sketch of the Czochralski method: The ingot is pulled upward and the crucible
is lifted according to the amount of solidified silicon in order to keep the melt level at a
fixed position. Crystal and crucible rotate in counter-direction. For the increase of pro-
ductivity, the radiation shield became an essential part in modern Cz puller.
22 Peter Dold
Figure 7 Industrial-size Czochralski puller for the growth of 800 and 900 mono crystals,
with an ingot length of up to 2 m. All the subsystems like vacuum unit, power supply,
and dust filter are located at a lower level not visible in the image. The total height of a
puller might easily reach some 6–8 m. (PVA TePla puller EKZ-3500 at Fraunhofer CSP.)
Figure 8 Loaded Cz crucible. The quartz glass crucible is sitting in a graphite support
unit, surrounded by a fence-type heater. For the picture, the outer insulation, as well as
the water-cooled jacket, has been removed.
Silicon Crystallization Technologies 23
Figure 9 Fully mounted hot zone, the radiation shield covers most of the crucible
surface.
24 Peter Dold
depends on the system used for growth control and for controlling the crystal
diameter. Since PV-crystal pullers are rather low-budget machines, quite
often, no fancy load cells or redundant measurements of weight and diam-
eter are used, but simple optical control of the diameter using CCD cameras.
Since the observation angle is rather steep with respect to the crystal axis
(there is no option for any optical access to the hot zone from the side or
any optical access through the crucible), accuracy is limited; e.g., an uncer-
tainty of half a millimeter on the crystal radius (for an 800 ingot) accumulates
over a crystal length of 200 cm to an incorrect calculation of the crystal
weight of approximately 1.5 kg and consequently a misinterpretation of
the position of the melt level.
Table 6 Summary of the Basic Parameters for Typical PV-Related Czochralski Ingot
Growth
Figure 11 The Czochralski process: top from left to right: the loaded crucible—melting
of the material—necking. Bottom from left to right: shoulder growth—shortly after the
transition to the body—body growth.
Figure 12 Pot scrap: 2–5 kg of residual melt remains in the crucible and sticks to the
quartz glass. The crucible cracks during the cooling-down process and has to be dis-
posed. Recycling of the pot scrap is difficult; the separation of the silicon from the quartz
requires mechanical and chemical process steps. Further, impurities are accumulated in
the pot scrap.
glass crucible, and due to the different thermal expansion coefficients, the
quartz glass brakes. In addition, a phase change of the quartz glass into
cristobalite starts during the crystallization process, and these two substances
have different thermal expansion coefficients. Since the modern machines
are rather well insulated, the cooling down from over 1500 °C to temper-
atures low enough for cleaning and crucible replacement takes several hours,
a downtime of the puller which is unproductive of course.
Hot zone: The graphite parts are consumables, too, and their life span is
normally in the range of 10 to maybe 50 cycles, depending on their work and
heat load. Since a full hot zone might easily cost some 20,000–40,000 US$
depending on the graphite purity and manufacturing costs, the hot zone
contributes quite significantly to the operating cost. Multipulling might help
to increase the lifetime of graphite parts, because they will see less heating
cycles, but at the end, there is not really a lot which can be done to lower
the cost without affecting the purity and the quality of the final product.
Coated graphite parts (e.g., SiC-coated) or CFC-enforced graphite felt is
beneficial for the purity and the life span of the corresponding part, but it
is significantly more expensive than the standard materials.
Argon: During the whole growth cycle, a continuous argon flow is essen-
tial in order to remove the SiO. Quartz glass is permanently dissolved by
liquid silicon. Fortunately, more than 99% of the oxygen evaporates as
SiO, which condensates at cooler surfaces. The argon flow helps to transport
most of the SiO to the outside of the growth chamber, where it might be
oxidized in a controlled manner after the growth.5 Further, the continuous
removal of SiO from the melt surface keeps the equilibrium on the silicon-
rich side and supports the evaporation of the SiO. Argon recovery systems
are available and they might be an option to reduce the Operational Expen-
ditures (OPEX).
Process time/productivity: A scale-up as we see it for directional solidifica-
tion, going from G4 to G5 to G6 and so on, is not possible for Cz growth.
Increasing the diameter in such a way that four bricks are cut out of the
grown ingot instead of one brick would require a crystal diameter of
445 mm, which is not cost effective. In addition to the higher OPEX
and CAPEX, the growth rate for such large ingots is reduced compared
to the one of an 800 standard ingot, since the removal of the latent heat is
more difficult. Thus, an increase of the productivity has to be coupled with
an increase of the growth rate and/or a reduction of the downtime.
Yield/structure loss: Structure loss is an important (or maybe the most
important) parameter for the calculation of the real cost of ownership and
the final dollar per wafer price. One of the great features of the Cz method
is the fact that a loss of single crystallinity is detected immediately or it is even
anticipated by careful observation of the structure and pronunciation of the
four growth lines. Figure 13 shows the growth facets during the crystalliza-
tion of the shoulder. If the facets are not all equally well pronounced or if one
5
As a fine powder, SiO is a pyrophoric substance, which has to be handled with great care.
30 Peter Dold
Figure 13 Growth facets (or growth lines), indicated by the blue (gray in the print ver-
sion) arrows: During the shoulder growth, the pronunciation of the four facets reveals
relevant information about the defect-free character of the crystal. (In the given image,
the fourth facet is partially shadowed by the neck.)
of them looks different compared to the other three, a structure loss is very
likely. If a structure loss is unavoidable, a decision can be drawn whether it is
cheaper to melt back the already grown part or to take it out and grow
another ingot from the remaining melt.6 A problem with structure losses
is that quite often it is not clear why it happened. Reasons might be:
– Particles, dust, etc., have been brought in during the loading process or
even earlier, during the harvesting and crushing of the polysilicon. In this
case, backmelting would not help, of course. The particle-contaminated
ingot has to be removed.
– Necking was not successful or the seed crystal was recycled too often.
– Growth conditions are not appropriate (too fast, temperature fluctuations
too high)
– Polysilicon was not pure enough, metallic impurities accumulated and
destabilized the growth interface (generation of morphological
instabilities).
– Particles are transported or introduced by the argon flow either from the
argon itself or picked up by the argon.
6
In the case of FZ growth, it will be seen immediately, too, but due to the growth setup, melting back is
impossible.
Silicon Crystallization Technologies 31
– Crucible corrosion results in a flaking off of quartz glass which might stick
to the growth interface.
For sure, there are many other reasons not listed here, but it becomes clear
that the growth of single crystals by the Cz method does not allow many
compromises and shortcuts. And, it requires skilled operators. Figure 14
shows an 800 ingot for the production of 156 156 mm2 pseudosquare
p-type wafers. In Fig. 15, the ingot is seen after cutting top and tail, remov-
ing the side slabs, and preparing wafers out of the brick.
Figure 15 After the growth process, the ingot is squared and wafered. The side slabs,
the top, and the tail are recycled.
spiral (http://www.pvatepla.com/en/products/crystal-growing-systems/
pva/cz—equipment/active-crystal-cooling). The axial position is some
20–50 cm above the solid–liquid interface. Since the optical access to the
meniscus area is mandatory for diameter control, there is a limited degree
of freedom with respect to positioning of the water-cooling system. Accord-
ing to the manufacturer, the active cooling might increase the pulling veloc-
ity by 20–40%. Obviously, inserting a water-cooled system into the hot zone
and in particular within the proximity of liquid silicon bears a certain risk.
Since it has to be machined from metal, any contact between the water-
cooled device and the growing crystal has to be avoided. Anyway, it is an
interesting approach and it will increase the productivity and thus will fur-
ther reduce cost. Furthermore, it shows that even for the rather matured Cz
pullers, there is still room for improvements.
Figure 16 Feeder system for batch feeding of small-size polysilicon. The feeder is
designed for topping-up the crucible after the initial melting of the first charge or
for recharging the crucible after the first crystal has been grown and moved into the
gate chamber for cooling down.
feeding. The feeder can be opened and recharged during ingot pulling, and it
is equipped with an independent gas and vacuum system.
(I) Batch feeding: Granular silicon or small-size chips are fed into the cru-
cible before or in between growth periods, i.e., either before the
growth starts, the crucible is topped up with additional silicon, or
the crucible is refilled after an ingot has been pulled, in which case,
the time required to cool down and to remove the crystal out of the
gate chamber is used to melt additional silicon. The latter is illustrated
in Fig. 17. Using batch feeding, the pulling process itself is not affected.
The equipment manufacturer PVA TePla has batch feeding systems in
their portfolio, Kayex had offered one too, but in 2013, Kayex stepped
out of the Cz business.
Silicon Crystallization Technologies 35
Figure 17 Feeding process. Feeding of small-size polysilicon chips into molten silicon
for a multipulling process.
Figure 18 Multicrystallizer for the production of G4 ingots. Maximum load: 250 kg. The
furnace is loaded from the bottom and is equipped with three heaters (top–side–
bottom).
Silicon Crystallization Technologies 37
The loading of a crucible and a resulting G4 block, together with a side slab
and a finished brick for wafering, are seen in Fig. 20. Typical cycle times are
in the range of 50–70 h, depending on the crucible load and the ingot
height. Since modern crystallizers are loaded with some 450 (in the case
of G5)–800 kg (G6)—and an end of the scaling process is still not
reached—the melting step requires quite sometime. In case of an 800 kg
load, the energy needed to melt the material is roughly half a MWh, which,
of course, has to be removed during the solidification process. The solidi-
fication itself takes place with typical rates of 1–1.5 cm/h. A certain difficulty
is the extraction of the latent heat from the system, which is released during
the solidification. In order to keep the interface flat, the heat has to be
removed through the bottom, as uniformly as possible.
In contrast to manufacturers of Cz pullers, with just a few companies
active, a large number of equipment providers tried to get a foot into the
door of the multicrystallizer furnace market during the booming period a
few years ago. Most of them sold just a few crystallizers, big business was
concentrated at GT-Solar (now GT-Advanced Technology), which
equipped at that time more than 80% of the market. Somewhat surprisingly,
they stepped out of the multicrystallizer furnace business by the end of 2012
(although they modified the statement somewhat in the meantime), opening
a new situation for the case the furnace market should pick up again in the
near future. On the other hand, today the Chinese market is primarily sup-
plied by local hardware manufacturers.
38 Peter Dold
Figure 20 Top left: A quartz crucible (sinter quartz) is sitting inside a graphite susceptor.
The quartz crucible is coated with a Si3N4 layer (or more precisely, a silicon oxinitride
layer) in order to prevent sticking of the silicon melt to the crucible. Crucibles are con-
sumables designed for single use only. Top right: Crucible loaded with 250 kg of poly-
silicon. Loading is done manually; any damage of the coating has to be avoided
carefully. Bottom: Multicrystalline ingot (G4) together with a brick and a cutoff side slab.
case of induction heating, but specific graphite parts are still required, like
the susceptor or the insulation material. In any case, induction heating bears
a certain risk of water spilling since the copper tubes have to be water-
cooled. In order to provide a high degree of flexibility, three heaters are
in use in most cases: bottom, top, and a wall heater. The bottom heater is
primarily used during the melting phase, and the top heater is required to
create flat isotherms and to prevent the formation of solid islands on the melt
surface. Sophisticated variations are in use for the bottom area. During the
melting part, as much heat as possible has to be introduced into the system,
and during growth, the latent heat has to be removed through the bottom
plate. Gas cooling might be used for this purpose (e.g., Li et al., 2012a), or
more often, the base plate is water cooled.
An advanced heater design had been developed by Lange et al. (2008)
and Rudolph et al. (2011): separating the wall heater into sections and
powering them by phase-shifted AC of a well-defined frequency, the heater
acts as magnetic field device and generates a controlled melt flow. In such a
way, the amount of inclusions and precipitates might be reduced and the
axial heat flux might be enhanced.
Crucial is the internal gas flow. The heaters, the susceptor for the quartz
crucible, and all the insulation parts are machined out of graphite. Due to the
interaction with SiO, with residual moisture, or with oxygen, rather high
concentrations of CO are generated. The CO will be absorbed at the free
melt surface and consequently forms SiC precipitates in the silicon block
as soon as a certain supersaturation is reached. To minimize the CO trans-
port into the melt, the system is continuously purged with argon, and the
inlet nozzle faces the melt surface directly. There are different philosophies,
whether a closed system (i.e., crucible and susceptor are covered by a graph-
ite plate or a SiC-coated lid) or an open system results in lower carbon con-
centrations. In any case, a well-designed argon flow is essential to avoid SiC
precipitates (Kimbel et al., 2012). During the process, the argon flow is in the
range of 10–40 l/min and the pressure inside the furnace chamber is in the
range of 500–800 mbar. The argon inlet contributes to the melt mixing
quite significantly (Li et al., 2011, 2012b). The argon generates a cold spot
in the center of the melt surface and triggers a strong surface tension-driven
flow (Marangoni flow), which improves melt mixing substantially. By
nature, the VGF configuration shows stable density stratification and con-
vective melt flow is driven by radial temperature gradients only. Well-
defined convective flows help to remove accumulated impurities in front
40 Peter Dold
of the growing interface and they enhance the heat transport. In conclusion,
a well-designed argon flow stabilizes the growth process.
Figure 21 During solidification, silicon expands by approximately 10%. The white line
marks the initial solid–liquid interface, the red (dark gray in the print version) arrow indi-
cates the initial position of the melt surface, and the blue (light gray in the print version)
one the final one.
all the major ingot manufacturers came up with mono-like ingots, and con-
sequently, modules based on mono-like wafers. In 2012, the ITRPV report
(www.itrpv.net/Reports/Downloads/2012) forecasted that by 2020,
mono-like will be the predominant growth technology with a total market
share of all crystalline silicon of 50%. The benefits seemed too tempting:
– Existing equipment can be used and only minor modifications might be
required in order to realize a flat solid–liquid interface.
– Simply covering the bottom area of the crucible with approximately 2 cm
strong seed plates is sufficient to obtain a mono-like structure of the ingot,
comparable to Cz ingots.
– Process time and cost of consumables are as good as not affected in com-
parison to standard directional solidification (despite the seed plates, of
course).
– Alkaline texturing of the wafers is possible which increases cell efficiency
quite substantially.
– The best-in-class cells based on mono-like wafers were just some 0.2%
below standard Cz cells.
– The mono-like wafers showed significantly lower oxygen levels than Cz
wafers.
A good overview of the state of the art in 2012 was provided by Gu et al.
(2012). The benefits had been pretty clear—and the difficulties had been
underestimated:
– Costs for the seed plates are high. Whether they can be recycled or not
and if so, how many times, is still unclear. Different data and statements
have been released over the years.
– The connection points and the interfaces between the seed plates are cru-
cial. They might generate dislocations which grow into dislocation clus-
ters or dislocation cascades, which will affect the carrier lifetime
significantly.
– The crucible walls act as nucleation centers for new grains, and very
likely, they will grow inward and into the outer ring of bricks, i.e., for
a G5 block; this could mean that only the inner 9 bricks have really a
mono-like structure but the outer 16 bricks show a mix of mono-like
orientation and multicrystalline structure, which is particularly unfavor-
able for alkaline texturing.
– The variation of the cell efficiency within one block was wider than it was
the case for standard multi. This requires intensive sorting.
For some of the problems, solutions had been presented, like the best way
how to pattern the seed plates (Birkmann and Kropfgans, 2013; Oriwol
44 Peter Dold
et al., 2013), or how to sort and classify the wafers into specific grades. But at
the end, the process turned out to be more complicated than expected and
many manufacturers lost their enthusiasm and went back to the standard
process. Or they switched to a new method: high-performance multi.
3.3 FZ Growth
3.3.1 State of the Art
Maybe the most fascinating feature of FZ is that neither the growing crystal
nor the melt is in contact with any other material. With the FZ technique,
silicon crystals with extremely low-impurity concentrations can be grown,
purer than it is possible with any other method (Ciszek and Wang, 2000,
2002). The inside of an FZ puller looks rather minimalistic: stainless steel
walls and, in the center, a water-cooled radiofrequency inductor made of
copper. Growth direction is opposite to Cz growth. The growing crystal
moves downward and the feed rod is lowered from the top. For silicon,
the needle-eye technique is in use, where the feed rod and the growing crys-
tal are connected by a liquid bridge only, with a diameter much smaller than
the crystal or the feed rod (Fig. 22). The inductor is sitting between the feed
and the crystal (would the bridge freeze, the inductor would be captured
between feed and crystal; Fig. 23). The inner diameter of the inductor hole
is in the range of 100 –1.500 , depending on the anticipated crystal diameter.
The outer diameter has to be slightly larger than the crystal to be grown
and the feed to be molten, respectively. Operational frequencies are in
the range of 500 kHz up to 3 MHz, preferably in the range of 2–3 MHz.
If the frequency of teh RF inductor is too low, then there is a risk that vibra-
tions are induced into the melt (Bohm et al., 1994). The FZ process provides
a perfect optical access (Fig. 24) and process control is either performed man-
ually by the operator according to direct observation of the growth process
or it is based on image processing.
Heat transfer by radiofrequency is extremely fast. In contrast to resistance
heating used for Cz, where the heat is transferred from the heater to the cru-
cible by radiation absorbed at the surface only, induction heating generates
heat in a boundary layer of several tens or some hundreds of micrometers,
depending on the frequency and the material. The localized heat induction
(from the bottom in the case of the feed rod and from the top for the growing
crystal) results in very steep axial temperature gradients. At the solid–liquid
interface, several hundred K/cm can be achieved (Dold, 2004), a feature to
be discussed later in more detail. One problem is the starting of the process.
At room temperature, the resistivity of normal feed rods is high—too high
for any heat transfer by radiofrequency. Therefore, the feed rod has to be
preheated. One option is to move a small graphite ring between the inductor
and the feed rod: the electrically conductive graphite couples well with the
RF, generates heat, and therefore warms the feed rod by radiation to the
46 Peter Dold
Feedstock
Liquid silicon
Inductor
FZ crystal
Seed
Neck
Figure 22 Float Zone growth of silicon by the needle-eye technique. Different to the
Czochralski growth, the crystal growth is downward and only a small portion of the
material is liquid at a time.
Figure 23 The floating zone. The dark ring marks the RF inductor. The silicon melt
appears darker than the grown crystal due to the lower emissivity of the liquid. In
the center, the liquid bridge is visible. The liquid silicon forms a kind of melt lake,
and the solid–liquid interface is bent downward.
Silicon Crystallization Technologies 47
Figure 24 Process control is performed optically either by the operator or by the help of
CCD cameras and image processing. Compared to the Cz process, FZ growth is consid-
erably less automated and skilled operators are mandatory.
Table 7 Float Zone Versus Czochralski Growth: Comparison of the Main Features
Float Zone Czochralski
Power consumption (Vedde et al., 2008) 6.5 kWh/kg Si 15.0 kWh/kg Si
Crucible None 5–10 US$/kg Si
Hot-zone, graphite consumables None 3–4 US$/kg Si
00
Time per kg crystallized Si (based on 8 ) 0.15 h 0.3 h
00
Growth rate (8 ) 2 mm/min 1.0–1.3 mm/min
Heat-up phase 1 h 5–8 h
Cool-down phase 1 h 5–6 h
Time needed for cleaning, charging, etc. 2 h 3 h
point where sufficient intrinsic carriers are generated in the silicon. Table 7
compares the main features of the FZ and the Cz method.
Despite the fact that the inside of an FZ chamber looks rather simple,
requirements for the mechanical rigidity and stiffness, for the vacuum capa-
bility, and for the excellence of the RF-generator are very stringent.
Industrial-size FZ puller for the growth of 800 ingots might easily reach a total
height of 12 m and a weight of 12 tons (Fig. 25). Worldwide, only two man-
ufacturers of commercial FZ machines for large-scale silicon growth are
active on the open market: PVA TePla (after purchasing the Danish
48 Peter Dold
Figure 25 A Float-Zone machine for the growth of 800 ingots. Only the upper part is vis-
ible, and the total height of the machine is 12 m. Courtesy of PVA TePla.
company Haldor Topsoe some years ago) and Steremat, both from Germany.
Other manufacturers work exclusively for certain ingot producers. Some
industrial producers of FZ ingots often have their own puller design,
constructed in-house in order to keep the know-how secured. A bit unclear
is the situation in China, whether there are manufacturers producing FZ
hardware or not.
The most exciting features of the FZ process are:
– Large energy transfer: only a small part of the material is heated at a time.
– Large thermal gradients result in a fast removal of the latent heat: growth
rates are faster by a factor of 2 compared to Cz.
– In situ doping of the material by the gas phase (diborane—B2H6 for p-type
or phosphine—PH3 for n-type) is easily carried out which results in very
uniform axial resistivity profiles (Goorissen and van Run, 1960).
– Oxygen concentration in the ingots is more than two orders of magni-
tude lower than it is the case in Cz crystals (Zulehner et al., 2012).
– Carbon contamination is low and metallic impurities are low: FZ ingots
are the prime material on the silicon market (Ciszek and Wang, 2002;
Richter et al., 2012).
Silicon Crystallization Technologies 49
Figure 26 The feed rods for the FZ process have to fulfill specific requirements with
respect to the geometry and the density of the material. Normally, the rods are cut
to a specific length, and they are grinded and etched before they can be used for
the FZ process. The picture on the left-hand side was provided by courtesy of Silicon Prod-
ucts Bitterfeld GmbH & Co. KG.
50 Peter Dold
Figure 27 Feed rod. The feed rod is covered by a thin layer of silicon melt only, which is
driven by surface tension toward the liquid bridge and then floats through the needle
eye into the melt lake. Spikes (or so-called “noses” or “needles”), sticking out of the feed
rod as unmolten silicon, are dangerous for the process: if they touch the inductor, there
will be a shortcut and the inductor will be damaged.
For the discussion and the evaluation of the FZ technique, the situation
at the growth interface has to be analyzed in more detail. The steep axial and
radial temperature gradients lead to a strongly concave interface shape,
which results in large thermal stresses. As long as the crystal is dislocation
free, the stress is still below the critical resolved shear stress and the crystal
simply relaxes during the cooling process and will be stress free once it
has reached a uniform temperature. On the other hand, if there are any dis-
locations, the thermal stress results immediately in multiplication of disloca-
tions and a loss of structure. In other words, an FZ crystal has to be
dislocation free. Once the structure gets lost, the ingot becomes useless. Bac-
kmelting is impossible and wafering of a multicrystalline FZ ingot is not fea-
sible because of the high stress level. Furthermore, the carrier lifetimes are
worse than the ones from any multicrystalline sample (Rost et al., 2012),
most likely due to the small grain size and the high defect structure.
Silicon Crystallization Technologies 51
Figure 28 A 400 FZ crystal, p-doped with a resistivity of 1 Ω cm. Diameter of the neck is
2.8 mm.
but with the consequence that the heat loss is reduced and the growth rate
has to be lowered.
3.3.1.2 Feedstock
As mentioned in the beginning already, proper feed rods are absolutely
essential for successful FZ growth. In most cases, the feed rods are produced
by the Siemens technology, but the process differs from the standard
method. In order to get a crack-free rod with a good morphology and a
smooth surface, deposition rates are lower. Lower deposition rates mean
longer process times and thus higher energy costs. Still, only part of these
special feed rods can be sold for FZ growth, i.e., the straight part without
the bridge area and with a certain distance to the elbow and to the carbon
electrodes. Due to these reasons (and the fact that there are only a limited
number of producers able to handle the process properly), the price for
FZ feed rods is several times higher than it is the case for solar-grade poly-
silicon. Alternatives are under investigation.
One option reducing costs for FZ feed rods is the use of granular poly-
silicon from an FBR and feed it directly into an FZ furnace (von Ammon
and Altmannshofer, 2011; von Ammon et al., 2011a,b). The furnace looks a
bit different from a conventional one: a second inductor is implemented for
melting of the granules. Of course, special care is necessary that no granules
come close to the solid–liquid interface, or the structure would be lost
immediately. Further, the granular material has to be of high quality with
a low dust load and low metallic impurity levels.
Another option is the prepulling of the feed rods (so-called pp-FZ—
prepulled Float Zone): standard solar-grade polysilicon is melted in a cruci-
ble and, similar to the Cz process, a feed rod for the FZ process is pulled
(Dold et al., 2014b). Since the structure of the feed rod does not play a role,
the feed rods might be pulled with high velocity. Combining the prepulling
process with multipulling and other cost-reducing procedures, the ready-to-
use feed rod obtained from the prepulling process is quite cost competitive to
the Siemens process. The oxygen picked up during the prepulling process
will evaporate during the FZ process.
It is still an open question whether FZ will ever play a significant role in PV or
not (of course besides the use of FZ wafers in, e.g., Insulated-Gate Bipolar Tran-
sistor (IGBTs) for PV inverters). The development of commercial 800 pullers with
a high degree of automation would be one step. Under these boundary condi-
tions, FZ would be a highly attractive alternative to Cz growth, considering the
lower operational costs, the high growth rates, and the excellent wafer quality.
54 Peter Dold
Czochralski
Pros Cons
Matured technique, highly automated Cost for consumables is high
Good optical access, good diameter control Oxygen levels are in the range of
(5–10) 1017 at/cm3
Necking allows the growth of dislocation-free Structure loss in some 5–10% of
ingots the cases
Remelting option is positive for productivity Cylindrical shape, but quadratic
bricks are needed
Ingots show high minority carrier lifetime 2–4% of material loss by pot scrap
Defined crystallographic orientation allows Axial dopant profile difficult to
texturing (option for multipulling) control
Silicon Crystallization Technologies 55
REFERENCES
Abe, K., Terashima, K., Matsumoto, T., Maeda, S., Nakanishi, H., 1998. Fused quartz dis-
solution rate in silicon melts: influence of boron addition. J. Cryst. Growth
186, 557–564.
Alemany, C., Trassy, C., Pateyron, B., Li, K.-I., Delannoy, Y., 2002. Refining metallurgical-
grade silicon by inductive plasma. Sol. Energy Mater. Cells 72, 41–48.
Altekrüger, B., 1994. Apparatus for continuously feeding material to a melting, crucible.
5,360,480.
Aulich, H., Schulze, F.W., 2009. Verfahren zur Herstellung von polykristallinen
Siliziumstäben und polykristalliner Siliziumstab. DE 10 2007 041 803 A1.
Bender, D.L., 2013. Systems and processes for continuous growing of ingots. US
2013/0133567 A1, 30.05.2013.
Bender, D.L., Smith, D.E.A., 2012. Weir design providing optimal purge gas flow, melt con-
trol, and temperature stabilization for improved single crystal growth in a continuous
Czochralski process. US 8,262,797 B1.
Bernreuter, J., 2014. The 2014 Who’s Who of Solar Silicon Production: Bernreuter
Research Polysilicon Market Reports, p. 124.
Bernreuter, J., Haugwitz, F., 2010. The Who’s Who of Solar Silicon Production: Bernreuter
Research Polysilicon Market Reports, p. 174.
Birkmann, B., Kropfgans, F., 2013. Verfahren zur Herstellung eines gerichtet erstarrten
Materialk€ orpers, insbesondere eines Metall- oder Halbmetallk€ orpers, sowie
Verwendungen hiervon. DE 10 2012 102 597 A1, 26.09.2013.
Bohm, J., Lüdge, A., Schr€ oder, W., 1994. Crystal growth by floating zone melting.
In: Hurle, D.T.J. (Ed.), Handbook of Crystal Growth vol. 2. Elsevier Science
Publishers, North-Holland, Amsterdam, London, New York, Tokyo, pp. 213–257.
Brandes, E.A., Brook, G.B., 1992. Smithells Metals Reference Book, seventh ed.
Butterworth-Heinemann Ltd., London.
Ceccaroli, B., Friestad, K., 2005. Refining of metallurgical grade silicon. US 6,861,040 B1.
Chaney, R.E., Varker, C.J., 1976. The dissolution rate of fused silica in molten silicon.
J. Cryst. Growth 33, 188–190.
Ciszek, T.F., Wang, T.H., 2000. Silicon float-zone crystal growth as a tool for the study of
defects and impurities. High Purity Silicon Vi 4218, 105–117.
Ciszek, T.F., Wang, T.H., 2002. Silicon defect and impurity studies using float-zone crystal
growth as a tool. J. Cryst. Growth 237, 1685–1691.
Coletti, G., Geerligs, L.J., Manshanden, P., Swanson, C., Riepe, S., Warta, W.,
Arumughan, J., Kopecek, R., Oct. 2007. Impact of Iron and Molybdenum in
Mono and Multicrystalline Float-Zone Silicon Solar Cells. Solid State Phenomena
131–133, 15–20.
Collet, Y., Magotte, O., VanDenBogaert, N., Rolinsky, R., Loix, F., Jacot, M., Regnier, V.,
Marchal, J.-M., Dupret, F., 2012. Effective simulation of the effect of a transverse mag-
netic field (TMF) in Czochralski silicon growth. J. Cryst. Growth 360, 18–24.
Czochralski, J., 1918. Ein neues Verfahren zur Messung der Kristallisationsgeschwindigkeit
der Metalle. Z. Phys. Chem. 92, 219–221.
Dash, W.C., 1958. Silicon crystals free of dislocations. J. Appl. Phys. 29, 736–737.
Dash, W.C., 1959. Growth of silicon crystals free from dislocations. J. Appl. Phys. 30 (4),
459–474.
Davis Jr., J.R., Rohatgi, A., Hopkins, R.H., Blais, P.D., Rai-Choudhury, P.,
McCormick, J.R., Mollenkopf, H.C., 1980. Impurities in silicon solar cells. IEEE Trans.
Electron Devices 27 (4), 677–687.
Delannoy, Y., Alemany, C., Li, K.-I., Proulx, P., Trassy, C., 2002. Plasma-refining process
to provide solar-grade silicon. Sol. Energy Mater. Cells 72, 69–75.
Silicon Crystallization Technologies 57
DeLuca, J.P., Delk, F.S., Johnson, B.K., Luter, W.L., Middendorf, N.D., Williams, D.S.,
Ostrom, N.P., Highfill, J.N., 2012. Silicon single crystal doped with gallium, indium,
or aluminum US2012/0056135 A1.
Dethloff, C., Friestad, K., 2007. Method for producing directionally solidified silicon ingots.
WO 2007/001184 A1, 04.04.2007.
Dold, P., 2003. Czochralski growth of doped germanium with an applied rotating magnetic
field. Cryst. Res. Technol. 38 (7–8), 659–668.
Dold, P., 2004. Analysis of microsegregation in RF-heated float zone growth of silicon—
comparison to the radiation-heated process. J. Cryst. Growth 261, 1–10.
Dold, P., Benz, K.W., 1995. Convective temperature fluctuations in liquid gallium in
dependence on static and rotating magnetic fields. Cryst. Res. Technol. 30 (8),
1135–1145.
Dold, P., Ernst, M., Kunert, R., 2014a. In-situ measurement of the crystallization rate during
block casting. In: 29th EU PVSEC Conference, Amsterdam, The Netherlands.
Dold, P., Zobel, F., Kunert, R., Lauer, K., Michl, B., 2014b. Low oxygen, high life-time
wafers produced by a modified Float Zone process (pp-FZ). In: 29th EU PVSEC Con-
ference, Amsterdam, The Netherlands.
Dosaj, V.D., Hunt, L.P., 1981. Method for producing solar-cell-grade silicon. 4,247,528,
27.07.1981.
Fabry, L., Hesse, K., 2012. Crystalline silicon feedstock preparation and analysis.
In: Willeke, G.P., Weber, E.R. (Eds.), Semiconductors and Semimetals, vol. 87.
Elsevier Inc, pp. 185–261.
Forster, M., Fourmond, E., Einhaus, R., Lauvray, H., Kraiem, J., Lemiti, M., 2011. Ga
co-doping in Cz-grown silicon ingots to overcome limitations of B and
P compensated silicon feedstock for PV applications. Phys. Status Solidi C 8 (3),
678–681.
Freiheit, H.C.D., Kraus, H., Sofin, M.D., 2010. Polykristalliner Siliciumstab für das Zon-
enschmelzverfahren und ein Verfahren zu dessen Herstellung. EP 1 992 593 B1,
13.10.2010.
Galindo, V., Gerbeth, G., von Ammon, W., Tomzig, E., Virbulis, J., 2002. Crystal growth
melt flow control by means of magnetic fields. Energy Convers. Manag. 43, 309–316.
Geerligs, L.J., Wyers, G.P., Jensen, R., Raaness, O., Waernes, A.N., Santén, S., Reinink, A.,
Wiersma, B., 2002. Solar-grade silicon by a direct route based on carbothermic reduction
of silica: requirements and production technology. In: Sopori, B.L. (Ed.), 12th Work-
shop on Crystalline Silicon Solar Cell Materials and Processes, Breckenridge, CO.
NREL National Renewable Energy Laboratory, Breckenridge, CO, pp. 215–218.
Glazov, V.M., Pashinkin, A.S., 2001. The thermophysical properties (heat capacity and ther-
mal expansion) of single-crystalline silicon. High Temp. 39 (3), 413–419.
Goorissen, J., van Run, A.M.J.G., 1960. Gas-phase doping of silicon. In: The Institution of
Electrical Engineers, pp. 858–860, Paper No. 3022 E.
Gu, X., Yu, X., Guo, K., Chen, L., Wang, D., Yang, D., 2012. Seed-assisted cast quasi-single
crystalline silicon for photovoltaic application: towards high efficiency and low cost sil-
icon solar cells. Sol. Energy Mater. Sol. Cells 101, 95–101.
Gunjai, P., Ramchandran, P., 2009. Tailoring oxygen distribution in 300 mm Czochralski
crystal of pure silicon using cusp magnetic field. In: Seventh International Conference on
CFD in the Mineral and Process Industries CSIRO, 9–11 December 2009, Melbourne,
Australia.
Gurvich, L.V., Vetyts, I.V., Alcock, C.B., 1990. Thermodynamic Properties of Inorganic
Substances, vol. 2. Hemisphere, New York.
Heuer, M., 2013. Metallurgical grade and metallurgically refined silicon for photovoltaics.
In: Willeke, G.P., Weber, E.R. (Eds.), Semiconductors and Semimetals, vol. 89.
Elsevier Inc, pp. 77–134.
58 Peter Dold
Hirata, H., Hoshikawa, K., 1980. The dissolution rate of silica in molten silicon. Jpn. J. Appl.
Phys. 19 (8), 1573–1574.
Hoshi, K., Isawa, N., Suzuki, T., Ohkubo, Y., 1985. Czochralski silicon crystals grown in a
transverse magnetic field. J. Electrochem. Soc. 132 (3), 693–700.
Huang, X., Huang, M., Ding, C., Ming, I., 2013. Seed-assisted growth for high-quality
multi-crystalline silicon ingots. In: CSSC-7, Fukuoka, Japan, pp. 91–96.
Hurle, D.T.J., Cockayne, B., 1994. Czochralski growth. In: Hurle, D.T.J. (Ed.), Handbook
of Crystal Growth, vol. 2a. Elsevier Science Publishers, North-Holland, Amsterdam,
London, New York, Tokyo, pp. 99–211.
Johnson, B.K., DeLuca, J.P., Luter, W.L. Silicon ingot having uniform multiple dopants and
method and apparatus for producing same. US 2012/0301386 A1, 2012.
Kaes, M., Heuer, M., Kirscht, F., Turenne, A., Hofbauer, C., Jester, T., 2014. Metallurgi-
cally purified silicon for photovoltaics 2AV.1.1. In: European PVSEC, Amsterdam.
Kajimoto, T., Horie, D., Sakurada, S., 1991. Method for growing silicon single crystal.
5,037,503.
Kakimoto, K., 2002. Effects of rotating magnetic fields on temperature oxygen distribution
in silicon melt. J. Cryst. Growth 237–239, 1785–1790.
Kakimoto, K., Shyo, T., Eguchi, M., 1995. Correlation between temperature and impurity
concentration fluctuations in silicon crystals grown by the Czochralski method. J. Cryst.
Growth 151, 187–191.
Kakimoto, K., Yi, K.-W., Eguchi, M., 1996. Oxygen transfer during single silicon crystal
growth in Czochralski system with vertical magnetic fields. J. Cryst. Growth
163, 238–242.
Kanda, T., Hourai, M., Miki, S., Shigematsu, T., Tomokage, H., Miyano, T., Morita, H.,
Shintani, A., 1996. Influence of melt-temperature fluctuations on striation formation in
large-scale Czochralski Si growth systems. J. Cryst. Growth 166, 663–668.
Kaneko, K., 2010. Method for casting polycrystalline silicon. US 7,682,472 B2.
Kaneko, K., Nakagawa, K., Onizuka, T., Sasatani, K., 2006. Recent status on electromag-
netic casting. NREL National Renewable Energy Laboratory, (Ed.), 16th Workshop on
Crystalline Silicon Solar Cells and Modules: Materials and Progresses, Denver, Colorado,
pp. 2–7.
Kim, K.M., Smetana, P., 1985. Striations in CZ silicon crystals grown under various axial
magnetic field strengths. J. Appl. Phys. 58 (7), 2731–2735.
Kimbel, S.L., Chen, J., Schrenker, R., Ferry, L., 2012. Directional solidification furnace for
reducing melt contamination and reducing wafer contamination. US8,317,920 B2.
Kirscht, F., Walerysiak, M., Heuer, M., Jouini, A., Ounadjela, K., 2010. Method and system
for controlling resistivity in ingots made of compensated feedstock silicon. US 2010/
0258768 A1.
Kobayashi, N., 1978. Computational simulation of the melt flow during Czochralski growth.
J. Cryst. Growth 43, 357–363.
Kurz, M., Pusztai, A., Müller, G., 1999. Development of a new powerful computer code
CrysVUN++ especially designed for fast simulation of bulk crystal growth processes.
J. Cryst. Growth 198/199, 101–106.
Lange, R.P., Jockel, D., Ziem, M., Rudolph, P., Kießling, F., Frank-Rotsch, C., Czupalla,
M., Nacke, B., Kasjanow, H., 2008. Vorrichtung zur Herstellung von Kristallen aus
elektrisch leitenden Schmelzen. DE 10 2007 028 547 A1.
Leblanc, D., Boisvert, R., 2008. Process and apparatus for purifying low-grade silicon
material.
Li, Z.Y.Y., Liu, L.J., Ma, W.C., Kakimoto, K., 2011. Effects of argon flow on heat transfer in
a directional solidification process for silicon solar cells. J. Cryst. Growth 318 (1),
298–303.
Silicon Crystallization Technologies 59
Li, T.F., Huang, H.C., Tsai, H.W., Lan, A., Chuck, C., Lan, C.W., 2012a. An enhanced
cooling design in directional solidification for high quality multi-crystalline solar silicon.
J. Cryst. Growth 340, 202–208.
Li, Z.Y., Liu, L.J., Liu, X., Zhang, Y.F., Xiong, J.F., 2012b. Effects of argon flow on melt
convection and interface shape in a directional solidification process for an industrial-size
solar silicon ingot. J. Cryst. Growth 360, 87–91.
Lorenzini, R.E., Iwata, A., Lorenz, K., 1977. Continuous crystal growing furnace.
4,036,595.
Lu, Z., 2007. Controlling melt-solid interface shape of a growing silicon crystal using variable
magnetic field. US 7,223,304 B2, May 29.
Ma, W.C., Zhong, G.X., Sun, L., Yu, Q.H., Huang, X.M., Liu, L.J., 2012. Influence of an
insulation partition on a seeded directional solidification process for quasi-single crystalline
silicon ingot for high-efficiency solar cells. Sol. Energy Mater. Sol. Cells 100, 231–238.
Mackintosha, B., Seidl, A., Ouellettea, M., Batheya, B., Yatesa, D., Kalejs, J., 2006. Large
silicon crystal hollow-tube growth by the edge-defined film-fed growth (EFG) method.
J. Cryst. Growth 287, 428–432.
Meléndez-Martı́nez, J.J., Domı́nguez-Rodrı́guez, A., 2004. Creep of silicon nitride. Prog.
Mater. Sci. 49, 19–107.
Morita, K., Yoshikawa, T., 2011. Thermodynamic evaluation of new metallurgical refining
processes for SOG-silicon production. Trans. Nonferrous Metals Soc. China 21, 685–690.
Müller, A., Schwichtenberg, G., Leistner, J., Seidel, J., 2004. Vorrichtung sowie Verfahren
zur Überwachung der Kristallisation eines Mediums, insbesondere Silizium. DE 102
34 250 A1, 5.2.2004.
Müller, M., Sahr, U., Lentes, F., Büllesfeld, F., 2009. Verfahren und Vorrichtung zum
Kristallisieren eines Halbleitermaterials, insbesondere Silizium. DE 102008022882 A1.
Nichol, S., 2011. Method for purifying silicon. US 7,883,680.
Oriwol, D., Mark, H., Wahl, B., 2013. Verfahren zur Herstellung von Silizium-Ingots. DE
10 2012 203 706 A1, 08.08.2013.
Paetzold, U., Reisbeck, A., Surner, M., 2007. Process and apparatus for the hydrogenation of
chlorosilanes. US 2007/0073075 A1.
Raming, G., Altmannshofer, L., Ratnieks, G., Landrichinger, J., Lobmeyer, J., Holzinger,
A., 2013. Method and apparatus for producing single crystals.
Rancoule, G., 2012. Crucible for the crystallization of silicon and process for making the
same.
REC Silicon Inc., 2013. White paper: benefit of FBR granular polysilicon to increase
efficiency of PV ingot manufacturing by increasing crucible polysilicon charge weight.
In: FBR White Paper—V4 10/13, REC Webpage, pp. 1–8.
Reuschel, K., 1963. Vorrichtung zur Gewinnung reinsten kristallinen Halbleitermaterials
fuer elektrotechnische Zwecke. DE1155759 B, 17.10.1963.
Richter, A., Glunz, S.W., Werner, F., Schmidt, J., Cuevas, A., 2012. Improved quantitative
description of Auger recombination in crystalline silicon. Phys. Rev. B 86 (16), 14.
Rodriguez, H., Guerrero, I., Koch, W., Endr€ os, A., Franke, D., Häsler, C., Kalejs, J.,
M€ oller, H.J., 2011. Bulk crystal growth and wafering for PV. In: Luque, A.,
Hegedus, S. (Eds.), Handbook of Photovoltaics: Science and Engineering, second ed.
John Wiley & Sons, Ltd, pp. 218–264.
Rost, H.-J., Luedge, A., Riemann, H., Kirscht, F., Schulze, F.-W., 2012. Float zone (FZ)
silicon: a potential material for advanced commercial solar cells? Cryst. Res. Technol.
47 (3), 273–278.
Rudolph, P., Frank-Rotsch, C., Kiessling, F.M., Kudla, C., 2011. Current and next steps of
bulk crystal growth to meet the challenges of photovoltaics. J. Ceram. Process. Res.
12, S159–S166.
60 Peter Dold
Watanabe, M., Eguchi, M., Hibiya, T., 1998. Flow and temperature field in molten silicon
during Czochralski crystal growth in a cusp magnetic field. J. Cryst. Growth
193, 402–412.
Wetzel, T., Muiznieks, A., Mühlbauer, A., Gelfgat, Y., Gorbunov, L., Virbulis, J.,
Tomzig, E., Ammon, W.v., 2001. Numerical model of turbulent CZ melt flow in
the presence of AC and CUSP magnetic fields and its verification in a laboratory facility.
J. Cryst. Growth 230, 81–91.
Yang, Y.M., Yu, A., Hsu, B., Hsu, W.C., Yang, A., Lan, C.W., 2015. Development of high-
performance multicrystalline silicon for photovoltaic industry. Prog. Photovolt. Res.
Appl. 23, 340–351.
Ydstie, B.E., Du, J., 2011. Producing poly-silicon from silane in a fluidized bed reactor.
In: Kosyachenko, L.A. (Ed.), Solar Cells—Silicon Wafer-Based Technologies,
Available from: http://www.intechopen.com/books/solar-cells-silicon-wafer-based-
technologies/producing-polysilicon-from-silane-in-a-fluidized-bed-reactor, ISBN:
978-953-307-747-5. www.intechopen.com.
Yoshikawa, T., Arimura, K., Morita, K., 2005. Boron removal by titanium addition in solid-
ification refining of silicon with Si-Al melt. Metall. Mater. Trans. B 36 (6), 837–842.
Zulehner, W., 1982. Device and process for pulling high-purity semiconductor rods from a
melt. 4,330,362.
Zulehner, W., 1999. Historical overview of silicon crystal pulling development. In: E-MRS
Spring Meeting, Strasbourg, France, Symposium E—Advances in Silicon Substrates,
Conference proceedings.
Zulehner, W., Huber, D., 1982. Czochralski-grown silicon. In: Grabmeier, J. (Ed.), Crystals
Growth, Properties, and Applications, vol. 8. Springer, Berlin, Heidelberg, New York,
pp. 3–143.
Zulehner, W., Neuer, B., Rau, G., 2012. Silicon. In: Ullmann´s Encyclopedia of Industrial
Chemistry, vol. 32. Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim, pp. 573–601.
CHAPTER TWO
Wafering of Silicon
€ ller1
Hans Joachim Mo
Fraunhofer Technology Center for Semiconductor Materials, Freiberg, Germany
1
Corresponding author: e-mail address: hans.joachim.moeller@ise.fraunhofer.de
Contents
1. Introduction 63
2. Multiwire Sawing Technology 65
2.1 Slurry-Based Sawing 66
2.2 Fixed Abrasive Sawing 69
2.3 Experimental Sawing Results 71
2.4 Determination of Wafer Properties 80
2.5 Electronic Grade Silicon 89
3. Basic Sawing Mechanisms 90
3.1 Slurry-Based Sawing 90
3.2 Damage of the Wafer Surface 99
4. Alternative Wafering Technologies 102
4.1 Cleavage Technologies 102
4.2 Layer Transfer Technologies 103
References 105
1. INTRODUCTION
Multiwire sawing has been developed as the main wafer-slicing tech-
nique for large multi- and monocrystalline silicon crystals in the photovol-
taic industry. The advantages are a high yield in production and a good wafer
quality even for thin wafers. Since 80% of today’s solar cell production is
based on crystalline silicon material, one can expect that multiwire sawing
will remain an important processing technology (Fig. 1). In the past decade,
the technology has also been introduced into the microelectronics industry
because it also has advantages there.
The cost pressure on the wafer fabrication is, however, high because the
production of wafers for solar cells has a cost share of about 15% of the final
module cost (International Technology Roadmap for Photovoltaic, 2012).
One of the reasons is that wire sawing needs expensive supply materials and
Figure 1 Annual development of the worldwide market shares of solar cell technolo-
gies (Solar PV Market Forecast, 2012).
Figure 4 Principle of different wire guide arrangements of multiwire saws (Assi, 2013).
The crystals (ingots or bricks) which have to be cut into wafers are pushed
through the wire web until the cut is finished. The motion can be either
up- or downward into the wire web, depending on the machine type. This
so-called slurry or loose abrasive sawing technique is worldwide applied in
the photovoltaic industry (Bidiville et al., 2010a; Endr€ os et al., 2002; M€oller,
2006; Nasch & Schneeberger, 2005).
SiC is the most commonly used abrasive today, but diamond or other hard
material can be used as well. The abrasive is suspended as a powder with a
mean grain size between 5 and 15 μm. Table 1 shows some typical parameters
of today’s SiC powders. The volume fraction of solid SiC particles can vary
between 18% and 25%. Most of the commercial slurries are based on polyeth-
ylene glycol (PEG)-based fluids. By changing the chain length of the polymer,
the viscosity can be varied. Others fluids such as monoethylene glycol,
diethylene glycol, dipropylene glycol, or oils are also used. In fact, oil has been
the main fluid in previous years and is still partly used in some Asian countries
today. Both fluid and abrasive particles deteriorate after some cuts. It is, how-
ever, possible to recycle the fluid and the SiC and mix a new slurry with addi-
tions of virgin SiC. Since recycling of the slurry and cleaning of the wafers is
cheaper for PEG than for oil, this slurry is mainly used in industry today.
The main purpose of the slurry is to transport the abrasive particles into
the sawing channel. Material is continuously removed through the interac-
tion of the SiC particles under the moving wire with the silicon surface. The
abrasive action of the SiC depends on many factors such as the wire speed,
Table 1 Size Specifications of Commercial SiC Powders and Viscosities of Carrier Fluids
at 20 °C
Grit D50 (μm) D97 (μm) D3 (μm) Fluid Viscosity (Pa s)
F2000 1.2 0.3 3.5 PEG 200 0.07
F1500 2.0 0.4 5.0 PEG 300 0.09
F1200 3.0 0.5 7.0 Oil 0.06–0.2
F1000 4.5 0.8 10 Water 0.001
F800 6.5 1.0 14
F600 9.3 1.0 19
F500 12.8 1.0 25
F400 17.3 1.0 32
50% of the SiC particles are smaller than D50, 3% are smaller than D97 and larger than D3.
68 Hans Joachim M€
oller
the force between wire and crystal, the solid fraction of SiC in the suspen-
sion, the viscosity of the suspension, the size distribution, and the shape of
SiC particles. The viscosity of the slurry depends on the temperature and the
solid fraction of particles and changes because of the continuous abrasion of
silicon, iron, and brass from the wire. The width of the sawing channel or
kerf is determined by the diameter of the wire, the size distribution of the
SiC particles, and the vibrations of the wire. A kerf loss around 120–160 μm
per wafer of 180 μm thickness is typical today.
In today’s industrial machines, the wires move in one direction with a
speed between 10 and 20 m/s. Therefore in a standard cutting process,
between 300 and 500 km of wire is used. The wires are used only once;
the slurry can be used two to four times and is recycled then. The cutting
speeds are around 0.3–0.5 mm/min, which yields a total cutting time of
about 8–13 h for a standard ingot size.
The ingots, which are sliced into wafers, have currently a size of
156 156 mm2. Modern machines allow to cut several ingots simulta-
neously and yield between 2000 and 5000 wafers in one run. Currently,
wafers with a thickness of about 180 μm are cut (Fig. 2). The thickness var-
ies, however, along the direction of the wire motion and is higher at the wire
outlet. This is due to the unidirectional motion of the wire and changes of
the sawing action along the wire channel. The change of the wafer thickness
should be, however, less than 25 μm (specified by the total thickness vari-
ation parameter (TTV)) for high-quality wafers.
The objective of efficient sawing is to slice with a high throughput, a
minimum loss of slurry and silicon, and a high quality of the resulting wafers.
The control of the process is very complex, since many parameters can be
varied and have an impact on the wafer quality and yield. The optimization
of the sawing is therefore a difficult task. Important are the control of wire
speed, wire tension, cutting speed, and other machine parameters, but also
parameters concerning the composition and properties of the abrasive SiC
powders and the carrier fluids.
Currently, two improvements of the sawing technology are tested: the
use of structured steel wires and the use of wires, which are coated with dia-
mond particles (fixed abrasive wires). In the first case, the wires have small
kinks at certain distances which result in a better transport of slurry into and
through the sawing channel. In the second case, the cutting is achieved by
the fixed abrasive diamond particles on the wire. The advantage in both cases
is that the cutting speed can be increased. The slightly higher kerf loss can be
compensated by a thinner core wire.
Wafering of Silicon 69
Figure 5 Steel wire coated with diamond particles (electroplating) (Buchwald et al.,
2013a).
70 Hans Joachim M€
oller
the expense of the cost. One can expect, however, that a mass production of
such wires will reduce the wire cost substantially.
In fixed abrasive sawing, the machines are operated in the so-called pil-
grim mode, where the wire is running back and forth. Typically, the wire is
running for several hundred meters in one direction before it is stopped and
the motion direction is reversed. In each cycle, a few meters of new wire are
added in one direction.
Since higher cutting speeds can be achieved (>1 mm/min), the cutting
times are reduced to 2–3 h. The wire consumption is about 5 km then and
thus very much reduced. Many of the current industrial saws allow high wire
and cutting speeds. The sawing machines therefore require only slight mod-
ifications to operate it with the diamond-coated wires. These modifications
mainly require a change of the fluid supply system. Since the fluid in a dia-
mond wire machine has the function to cool the ingot and clean the wire,
one only has to modify the nozzle system in the saw slightly.
The competiveness of the fixed abrasive technique depends on the cost
per wafer. In recent years, the cost of the fixed abrasive wires could already
be reduced, but other factors are also important. During sawing, the wafer
surface is damaged, as will be discussed in Section 2.3 in detail. This reduces
the mechanical stability of the wafers, which is important for the following
processing steps of the wafers such as cleaning, solar cell processing, and
module manufacturing. With increasing cutting speeds, the wafers become
more unstable and the wafer yield is reduced.
The experiences so far indicate that fixed abrasive sawing is able to cut
monocrystalline ingots into wafers of standard thickness of 180 μm. The
wafer surface quality is good but may require an adjustment of the following
etching processes for texturization and further solar cell processes. The
results also show that sawing multicrystalline wafers works less well
(Buchwald et al., 2013b, 2014). The wafer surface quality is lower and wafer
breakage is more frequent. The reasons for these differences between mono-
and multicrystalline silicon will be discussed in Section 2.3. Such a difference
between the two materials is not observed for slurry-based sawing.
Considering the long-term industrial development, wafers of less than
100 μm may be required in the future. At present, such thin wafers cannot
be handled in the downstream process of solar cell and module fabrication.
Thinner wafers require adapted solar cell designs because they absorb less
light. Such processes have not been implemented into the standard produc-
tion chain. In addition, the following industrial processing steps, which
Wafering of Silicon 71
require mechanical handling of the wafers, are even more prone to breakage
because of the lower fracture toughness.
Although the fixed abrasive sawing technique is certainly not optimized,
estimations of the cost per wafer have been made for the Japanese market
(Solar Battery Technologies, 2012). Based on current parameters for mono-
crystalline silicon the cost of consumables are 10 Yen/W, which is still
higher compared to 7 Yen/W for a slurry sawn wafer. An introduction
of the technology in mass production is not expected before the year
2020. Whether multicrystalline silicon can be sawn on an industrial scale
with this technique is still uncertain.
Figure 6 Temperature distribution on the side face of an ingot during sawing (mea-
sured by infrared thermography).
Figure 7 Average force per wire in cutting direction (ingot feed motion) as a function of
the ratio of feed velocity/(wire velocity)n for different grit sizes (n ¼ 1.125; Meißner et al.,
2012, 2013). The force per wire is determined from the total measured force divided by
the number of wires in the experiments.
for finer grit sizes. It has also been observed that there is a slight dependence
on the shape of the size distribution.
The friction is responsible for the temperature increase along the sawing
channel. Since the viscosity is exponentially dependent on the temperature
(see Section 2.3.1.3), there is a substantial change of the viscosity along the
sawing channel. All factors, which affect the friction force, thus also have an
impact on the viscosity.
Sawing experience shows that small changes of the slurry temperature
can have a significant influence on the sawing result. Therefore, a good con-
trol and stabilization of the slurry temperature is very important in the saw-
ing process. The ramifications of the slurry temperature on the wire sawing
process will be discussed in Section 3.2.2.
Figure 9 Dependence of friction and feed force per wire on the slurry viscosity.
Wafering of Silicon 75
Figure 10 Relative viscosity as a function of the volume fraction of SiC in a PEG 300
carrier fluid for different grit size distributions. The solid lines are Krieger–Dougherty fits
with appropriate parameters.
76 Hans Joachim M€
oller
During sawing, the solid volume fraction in the fluid changes in the saw-
ing channel because of the removed silicon and the wear of the wire. The
debris is finer than the average SiC grain size. Therefore, one can expect an
increase in the viscosity due to the increased volume and the contribution of
finer particles. It has also been reported that the shape of the particles has an
effect on the slurry viscosity, but systematic data are not available so far.
Because of the temperature changes in the sawing channel, one also has
to consider the temperature dependence of the viscosity. Generally, the vis-
cosity decreases with increasing temperature T according to an Arrhenius
law ηo ¼ ηoo eQ=RT , where Q is an activation energy and R the universal
gas constant. Carrier fluids from different suppliers can have quite different
activation energies because the basic molecules and the composition of main
carrier fluids, PEG, and oil can be modified quite easily (Doolittle, 1951).
Considering all factors, the viscosity at the entry and exit of the sawing
channel differs and varies along the channel. Figure 11 shows the calculated
variation of the slurry viscosity along the sawing channel due to temperature
and solid fraction changes for a standard slurry with PEG 200, 23 vol% SiC,
and a typical wire velocity of 15 m/s. The temperature changes from 20 to
80 °C in this case, which is quite significant. The addition of solid material
due to debris increases the viscosity, but this effect is overcompensated by
the temperature increase induced viscosity decrease.
So far, experimental results today are compared using the original slurry
viscosity measured at room temperature. No procedures exist how to take
Figure 11 Calculated change of the viscosity of a PEG 200/F800 slurry due to the tem-
perature and solid fraction increase along the wire channel.
Wafering of Silicon 77
into account the various factors that cause viscosity changes along the sawing
channel and how to determine a relevant viscosity value correctly. Since the
influence of the slurry on the sawing performance is a key factor today, fur-
ther research is required to better understand the role of the slurry in the
sawing process.
Figure 12 Preston coefficient as a function of the wire diameter for different grit sizes
(Meißner et al., 2012).
the wire will break. Commercial steel wires have fracture stresses around
4 GPa and can bear forces around 40–50 N. Calculations of the wire stresses
as a function of the total applied forces, which have been obtained from
experimental results for different slurries, wire velocities, tensions, diame-
ters, and grit sizes, show that for thinner wires and finer grit sizes, one
approaches already the fracture stress (Wagner & M€ oller, 2008).
Because of the limits of fracture stresses of commercial wires, one cannot
increase the cutting velocity too much. At present, it appears that F600 or
F800 is already the best choice, so that it is not clear how much the wafer
thickness and the kerf will actually be reduced in the future. At present, one
cannot saw 100 μm wafers with wires of 100 μm or less under production
conditions where one needs a high yield.
Figure 13 Part of an as-sawn multicrystalline silicon wafer surface showing saw marks
ending at the right edge of the wafer (wire outlet side).
industrial process where thousands of wafers are cut, but systematical data
have not been published. There are indications that a high sawing velocity
(vs), a low SiC load, and high slurry temperatures enhance the probability for
the occurrence of saw marks (M€ oller et al., 2013; Park & Dibiase, 2012;
Retsch et al., 2014).
In the industrial process, sawing conditions have to be chosen where saw
marks can be avoided. The drawback is that in particular higher sawing
speeds cannot be reached and this reduces the productivity. Recently, an
improvement of the slurry-based sawing could be achieved by the use of
structured wires. These are wires which contain periodic kinks that can
improve the slurry transport in the sawing channel, so that higher sawing
speeds (by about 20%) become possible, however, at the expense of a slightly
higher kerf loss.
Considering the observed dependencies, one can assume that a homo-
geneous slurry transport along the entire sawing channel is important for
a stable process and a good wafer quality. A possible explanation for this
assumption, which has recently been proposed, will be given in
Section 3.2.1.
Figure 14 Surface height topograms of a slurry sawn wafer (A) and of a fixed abrasive
sawn wafer (B) as measured by confocal optical microscopy.
Figure 15 Roughness profiles (Rz) from the wire inlet to the wire outlet region for wafers
sawn with different grit sizes. The Rz value gives the difference between minimum and
maximum surface height.
Figure 16 (A) SEM image of microcracks at a cross section of a slurry sawn wafer.
(B) Etched traces of microcracks at a beveled (slope angle 1°), polished surface of a slurry
sawn wafer.
Figure 17 Optical microscope image of an etched, beveled wafer surface after fixed
abrasive sawing. The microcracks are aligned along the wire motion direction.
Roughness and crack depth are therefore both dependent on the posi-
tion on the wafer surface. They change in the same way because both param-
eters originate from the same indentation process during wire sawing.
Microcracks also occur on wafers sawn with fixed abrasives, but their dis-
tribution and sizes differ (W€urzner et al., 2015). The cracks are aligned along
the direction of the sawing grooves (see Fig. 14) and show a distinct pattern
which is repeated periodically (Fig. 17).
A further difference is that the surface can locally become amorphous.
This has been observed by Raman spectroscopy. The extension of the amor-
phous region depends on the sawing conditions (M€ oller, 2008). No system-
atic investigations exist so far about the extension of these regions. It is also
still unknown how and under which conditions the crystalline to amorphous
transformation of silicon occurs. Single scratching experiments as described
in Section 2.4.2 may give some indications about these transformation
processes.
The amorphization of the surface can have an impact on the etching
behavior. Since this is an important processing step before solar cell fabrica-
tion, it may be difficult then to produce a uniform, textured surface.
the material removal process, and the particles at the side of the sawing
channel determine the damage on the wafer surface.
The individual particle interaction processes have been studied by inden-
tation experiments with a microhardness Vickers tester. A geometrically
defined sapphire or diamond tip is indented under load into a silicon surface.
The resulting damage pattern and the extension of the cracks are determined
then. From the large number of results that have been published for silicon
(Evans & Marshall, 1981; Feltham & Banerjee, 1992; Lawn, 1993; Li &
Bradt, 1996; Malkin & Ritter, 1989), the following sequence of events
can be deduced (Fig. 18).
At low loads, a plastic zone forms first. The diameter a depends on the
applied load Fp and can be described by the following equation:
!1=2
α
a¼ Fp (3)
4H tan ½φ2
H is the hardness of the material, φ the angle of the indentation tip, and α a
geometry factor. If the load is increased, the material breaks and different
types of crack systems can occur depending on the crystal orientation
and the tip geometry. So-called median and/or radial cracks directly under
the load tip extend vertically into the surface. The force dependence of the
median crack length is given by
Figure 18 Schematic diagram of the indentation of a single particle into the surface (A).
Under the action of the normal forces, a plastic zone and cracks are formed. The exten-
sion of the lateral cracks and the depth of the plastic zone determine the chipped
volume. (B) Optical micrograph of a Vickers indentation showing median cracks of
length c and the plastic zone of diameter a.
Wafering of Silicon 85
2=3
β
c¼ Fp (4)
Kc
Kc is the fracture toughness of the material and β a geometry factor. For the
Vickers indenter, the parameter α ¼ 2, for a Knoop indenter α ¼ 4/π, while
the best fit value β ¼ 1/7 varies considerably in the published data. Measure-
ments for monocrystalline silicon at room temperature yield H ¼ 10.6 GPa
(Hamblin & Stachowiak, 1995; Poon & Bushan, 1995) and
Kc ¼ 0.75 MPa m1/2 in large-grained polycrystalline silicon (Anstis et al.,
1981; Chen & Leipold, 1980).
When the load is removed, the stress difference between the plastic zone
and the crystalline silicon leads to the formation of lateral cracks. When they
reach the surface, a piece of material is removed (chipping). This process is
responsible for the material removal in the sawing process. The volume of
the chipped material has been determined experimentally and depends on
the applied load according to
Vo ¼ γFp2:2 (5)
where γ is a geometry factor (Funke & M€ oller, 2003). Since the formation of
the median cracks occurs before the chipping, these cracks partially remain
in the crystal surface at the side of the sawing channel. They form the saw
damage of the as-sawn wafers (Chauhan et al., 1993; Cook, 1990; Larsen-
Basse, 1993; Siekman, 1987; Verspui et al., 1995).
Figure 19 Single scratch test showing periodic crack patterns. The upper image shows
the damage at the surface. In the lower image, the surface is inclined by about 1° from
left to right. The right-hand side of the image is about 10 μm below the surface.
motion direction, the resulting force faces obliquely into the bulk. The stress
which builds up has not been calculated so far, but one can assume that it
consists of tensile stresses behind the tip and compressive stress components
in front of the tip. The stresses depend on the tip shape and the orientation of
the crystal. This could be confirmed by the experimental results. The stresses
eventually will lead to cracking. Usually, one can expect a crack directly in
front of the moving direction and several cracks to the sides. Because of the
oblique force, some cracks will chip material away mainly in front of the tip.
Once the material breaks, the tip can move a certain distance before it is
blocked again. Then, the process of stress build-up and breakage repeats
itself.
The lengths of the cracks depend on the applied forces both in wire
direction and in the vertical (indentation) direction. Quantitative measure-
ments yielded a crack length dependence on the applied force as given in
Fig. 20. A slight dependence on the crystal orientation was also observed
(W€ urzner et al., 2014b).
Recent measurements of the crack length distribution in multicrystalline
silicon also showed a correlation with the grain orientation (W€ urzner et al.,
2015). This result may give an explanation for the observed difference of the
weaker stability of multicrystalline wafers compared with monocrystalline
silicon after fixed abrasive sawing. If one assumes that there are always unfa-
vorable grain orientations, which yield deeper cracks, these “weaker” cracks
determine the overall fracture stability of the wafer (see Section 2.4.3).
Wafering of Silicon 87
The single scratch tests also show that in some areas, a transformation of
the crystalline silicon into an amorphous phase can occur (Fig. 21). The con-
ditions under which this occurs are, however, not clear yet. Single Vickers
indentations experiments have shown that in the regions of very high stresses
in the plastic zone, phase transformations of the silicon can occur. Several
phase changes have been observed by Raman spectroscopy, in particular
a metallic high-pressure phase (Ballif et al., 2005; Domnich & Gogotsi,
2002; Gogots et al., 1999; Jang et al., 2005; Weppelmann et al., 1993;
W€ urzner et al., 2014a). Under loading at 11.8 GPa, an endothermic trans-
formation to metallic silicon (Si II) occurs, which partly transforms back to
another high-pressure phase (Si III) upon unloading. In the metallic state,
the silicon can plastically deform and material can be removed by processes
known for ductile metals. These regions are probably very thin and are
removed when the material is further stressed and breaks (compare Fig. 18).
In the scratching mode, highly strained regions occur in front (com-
pressed region) and behind (tensile region) the indenting tip. Since chipping
occurs mainly in front of the tip, any phase changes that occur here will be
removed. But phase changes behind the moving tip could remain. This may
explain that one can observe for instance amorphous regions on the surface
after sawing. Since the details of these processes are not known in much
detail, it is too early to control the scratching and sawing process. But
one can expect that further scratching tests will also deepen the understand-
ing of the sawing process with fixed abrasive wires.
88 Hans Joachim M€
oller
Figure 21 Raman spectroscopy map along a single scratch. It shows the shift of the
Raman line due to the local stresses. The shapes of the corresponding Raman lines indi-
cate the presence of an amorphous phase.
Figure 22 Weibull plots (failure probability) for as-sawn and etched wafers. The thick-
nesses of the removed damage layers are indicated.
90 Hans Joachim M€
oller
planarity in the nanometer range over the entire wafer surface is very impor-
tant. Although the following processing steps, grinding, lapping, and
polishing, produce the final surface quality, also the sawing step has to be
very precise. Thickness variations that occur here require higher efforts in
the following steps to remove them and are therefore expensive.
The wafers are cut in the oscillating mode which avoids the wedge shape
of the slurry sawn wafer. The adjustment of the wire speed, tension, and
other parameters enable one already to produce high-quality wafer surfaces.
A problem, however, that remains is when the wires first cut into the crystal.
It has been observed that the wafers often show a shallow bump of a few
micrometers at the rim of the surface which has to be removed afterward.
It is assumed that the wires move sideways before they cut into the material
and straighten out afterward during cutting. The reason for the sideward
motion is not quite clear but may have to do with some irregularities on
the crystal surface. In general, when cutting monocrystalline wafers, the pre-
cise control of the wire motion through the crystal is the most demanding
requirement.
Figure 23 Schematic diagram of the sawing channel. It shows the main forces that are
assumed in the model.
interaction of rolling SiC particles that are randomly indented into the crystal
surface until small silicon pieces are chipped away. This “rolling–indenting
grain” model forms the physical basis of the wire sawing process. The same
mechanism also occurs in lapping brittle material surfaces with loose abrasive
particles (Buijs & Korpel-van Houten, 1993a,b; Su et al., 1996).
It is assumed that only the larger particles which are in direct contact par-
ticipate in the material removal process. During sawing, the crystal is pushed
against the wires with a constant forward (feed) velocity. With increasing
velocity, the total force Ftot on the wire increases and more and more par-
ticles are indented. Since the wire is elastic, it will bow under the force like a
string. The resulting wire curvature adjusts to achieve an overall constant
feed velocity.
One has to consider here that the forces vary below the wire, depending
on the local conditions in the sawing channel. Considering the cross-
sectional view, the resulting force perpendicular to the wire surface which
pushes the particles into the surface also varies along the contact area. The
forces are maximal directly below the wire and decrease toward the side faces
(Fig. 24). The cutting process at the wafer side and the applied force there is
important because it determines the final surface quality of the sliced wafers.
Furthermore, the fast-moving slurry builds up a hydrodynamic pressure,
which may exert additional force on the wire. The hydrodynamic pressure
also decreases toward the sides of the channel because it is zero at the free
slurry surface. Particles in the channel will be pushed sideways because of
a pressure difference and eventually be removed from the sawing channel
when they reach the free slurry surface. Since the pressure difference
depends on the particle size, one can expect a rearrangement of the particle
size distribution by the removal of larger particles from the sawing channel
below the wire. In fact, several results have been reported, which show that
92 Hans Joachim M€
oller
Figure 24 Schematic diagram of the cross section of the sawing channel and the var-
iation of the force perpendicular to the wire surface as a function of the angle α.
the size distribution varies along the sawing channel. This aspect will be
discussed in more detail in Section 3.2.1.
All processes together determine the local forces on the indenting par-
ticles. The individual process of the interaction of a single particle with
the surface has been studied by microindentation experiments as discussed
in Section 2.4.2. The main process for material removal during sawing is
the formation of lateral cracks and the chipping of material. The median
and radial cracks partly remain. This crack system is part of the sawing dam-
age which has to be removed for further processing of the wafers. Combin-
ing the rolling–indenting process of free abrasive grains with the fracture
mechanics of brittle materials, a quantitative description of the material
removal process could be derived. For a detailed review of the basic ideas,
see references M€ oller (2004, 2008).
mVo
vs ¼ : (7)
Atot Δt
The material volume Vo that can be removed in a single event is deter-
mined by the extension of the lateral cracks below the indentation of a single
grain (Fig. 18). It depends on the applied normal force Fp on the particle
according to Eq. (5).
The velocity profile of the laminar flow in the slurry leads to a rotation of
the particles. If a rolling grain makes one indentation per cycle, the average
time interval for a single indentation event is given by △t ¼ 2dp/v, where v is
the wire velocity and dp the particle diameter which is determined by the
largest particles in the size distribution.
The experimental results show that wire distance, particle size distribu-
tion, the temperature, and other parameters vary along the sawing channel.
Therefore, it is necessary to consider the local situation. Each wire segment
exerts a force Fn on the particles, which depends on the local curvature and
the wire tension. For a rigid wire segment, the resulting average force on a
single particle Fp is given by Fp ¼ Fn =m, where m is the number of all the
particles in contact under the segment. For a short segment of length dx,
one can write
@Fn =@x
Fp ¼ : (8)
@m=@x
Combining the equations yields for the local sawing rate, the fundamen-
tal relationship which forms the basis for the following theoretical
description:
@m=@x @Fn =@x n
vs ¼ v∗so v , (9)
dp rw @m=@x
where the prefactor vso ∗ summarizes material and geometry parameters. The
contact area below the wire has been expressed by Atot ¼ πrwdx here, with
the wire radius rw. The remaining problem that has to be solved is to deter-
mine the number m of indenting grains in the slurry and the average force Fn
per wire segment dx.
The number m of particles in contact and the average force Fn are cal-
culated from the grain size distribution. Commercial virgin SiC particle sizes
can be described rather well by a Gaussian distribution. In general, the size
distribution will differ from this form, particularly when slurry is transported
along the sawing channel and particles are lost or break. It is useful therefore
94 Hans Joachim M€
oller
Figure 25 Schematic particle size distribution g(l) for a typical total number of particles
in contact no ¼ 70. Grains with a diameter l > ho are in contact both with wire and crystal
surface. The total number m of particles in contact is proportional to the shaded area.
where hmax is the maximum distance where just one particle is in contact in
the contact area.
If a force Fn is applied, the distance h decreases and more grains come into
contact with the surface. Neglecting the indentation in the wire, the force Fn
is equal to the force on all particles Fn ¼ Fnp, which are actually indented into
the crystal surface
ð hmax
Fnp ¼ no Find ðl hÞgo ðl Þdl, (11)
h
where Find(x) is the force law that applies when a single particle is indented
by a distance x into a surface. It depends on the elastic, plastic, and fracture
behavior of the material. For sharp Vickers indentations and a brittle material
Find as a function of the indentation depth, x ¼ l h could be given by
Wafering of Silicon 95
Eq. (3), but other force laws have been derived as well, depending on the
shape of the indented grains.
Combining the Eqs. (1–5), one can calculate the average force Fp ¼ Fnp/m
on a single particle for a given size distribution go(l). The result is shown in
Fig. 26 for a Gaussian distribution as a function of the wire distance h and dif-
ferent tip angles φ of the indenting particle tip.
An important result is that Fp remains almost constant for sharp tips
(φ < 90°), when the wire distance decreases. Only the number m of indented
particles increases. One can simplify Eq. (3) and obtain the following expres-
sion for a wire segment of length lo
∗Fpn1
vso
vs ¼ vso vFn vso ¼ : (12)
dp rw lo
T @2h T @Fn
p ¼ po + ph ph ¼ po ¼ , (15)
πreff @x 2 πreff @x
where the surface curvature has been expressed by a constant pressure po and
a constant normal force Fno. Experimental investigations of the shape of the
surface during cutting have shown that it is slightly curved and deviates only
very little from a circular shape at least over longer segments of the cutting
length. In equilibrium between wire and slurry, the pressures in Eqs. (13)
and (15) have to be equal. Since po is assumed constant over the segment
length, the derivative is zero and one obtains by inserting
@ 3 h 6πvηreff h ho
+ ¼ 0: (16)
@x3 T h3
This equation is the basis for the investigation of the hydrodynamic slurry
behavior. It has no analytical solution but previous numerical solutions show
that the wire distance h differs only slightly from ho, so that one can approx-
imate the equation quite well by (M€ oller et al., 2013)
@ 3 h 6πvηreff
+ ðh ho Þ ¼ 0: (17)
@x3 Th3o
This equation has the following analytical solution:
hpffiffiffi i hpffiffiffi i
h ¼ ho + c1 ea x + c2 ea x=2 cos 3a1=3 x=2 + c3 ea x=2 sin 3a1=3 x=2 ,
1=3 1=3 1=3
(18)
where
6πvηreff
a¼ : (19)
Th3o
The integration constants c1, c2, and c3 are determined from boundary
conditions. Appropriate boundary conditions are the pressure at wire inlet
p(0) ¼ pa and outlet p(lo) ¼ 0, and the wire distance at the outlet h(lo) ¼ ho.
The total pressure can be derived from Eq. (15) if po is known. If ph is
small compared to po, the pressure p and po are approximately equal (p po).
In this case, the pressure is determined by the local force Fno resulting from
the ingot feed and wire bow. It is equal to the force Fnp, which has been
calculated from the total force on all indented particles by Eq. (8).
A significant hydrodynamic pressure will push the wire away from the
ingot surface. This reduces the sawing rate since the force on a single particle
98 Hans Joachim M€
oller
is reduced and fewer particles are in contact (see Eq. 6). Numerical results
show that this is particularly the case in the exit region of the sawing channel,
when ph becomes large. For ph > 0, one obtains in this case for the force per
wire segment length
@Fnp
h
@Fnp
¼ ph (20)
@x @x
where ph is given in Eq. (15). When ph is negative, the hydrodynamic pres-
sure p has to be neglected because p < po then. This means it has no effect on
the local distance of the wire, which is entirely determined by the particles in
contact then.
Figure 27 Calculated hydrodynamic pressure p in the slurry film along the wire length
for two different feed velocities.
vs ¼ 0.35 mm/min in Fig. 27). This means that the basic assumption of a
continuous slurry flow is not possible here. It has only recently been recog-
nized that the hydrodynamic pressure situation in this part of the channel can
lead to an instability of the wire motion (M€ oller et al., 2013). This will be
discussed in Section 3.2.
Figure 28 Feed velocity versus wire velocity, showing stability regions of wire motion
for different SiC and silicon volume concentrations.
Wafering of Silicon 101
REFERENCES
Anspach, O., Schulze, F.-W., 2007. Glycol to water-based slurry, investigation on the wire
sawing process. In: Proceedings of the 22nd EU PVSEC (WIP, M€ unchen, 2007), p. 952.
Anstis, G.R., Chantikul, P., Lawn, B.R., Marshall, D.B., 1981. A critical evaluation of
indentation techniques for measuring fracture toughness. J. Am. Ceram. Soc. 64, 533.
Aoyama, T., Kondo, Y., Asai, H., Takato, H., Sakata, I., 2009. Fabrication of single-
crystalline silicon solar cells using wafers sliced by a diamond wire saw.
In: Proceedings of the 25th European PVSEC, p. 2429.
Assi, F., 2013. Industrial diamond wire sawing. In: Freiberger Sawing Symposium (Freiberg,
2013).
Baer, S.C., 2009. Patent No. US 2009/0056513 A1.
Ballif, C., Gasser, P., Buerki, G., Michler, J., 2005. In situ micro-Raman compression. Phys.
Status Solidi A 202, 2858.
Behm, T., F€ utterer, W., Funke, C., Kaminski, S., M€ oller, H.J., Rietzschel, R., Wagner, T.,
2011. Challenges of the wire saw wafering process. Photovolt. Int. 11, 38.
Bergmann, R.B., Berge, C., Rinke, T.J., Schmidt, J., Werner, J.H., 2002. Advances in
monocrystalline Si thin film solar cells by layer transfer. Sol. Energy Mater. Sol. Cells
74, 213.
Bhagavat, M., Prasad, V., Kao, I., 2000. Elastic hydrodynamic interaction in the free abrasive
wafer slicing using a wire saw. J. Tribol. 122, 394.
Bidiville, A., Wasmer, K., Kraft, R., Ballif, C., 2009. Diamond wire-sawn silicon wafers—
from the lab to the cell production. In: Proceedings of the 24th European PVSEC,
p. 1400.
Bidiville, A., Wasmer, K., Michler, J., Nasch, P.M., Van der Meer, M., Ballif, C., 2010a.
Mechanism of sawing and impact on wafer properties. Prog. Photovolt. Res. Appl.
18, 563.
Bidiville, A., Heiber, J., Wasmer, K., Habegger, S., Assi, F., 2010b. Diamond wire wafering:
wafer morphology in comparison to slurry sawn wafers. In: Proceedings of the 25th
European PVSEC (WIP, M€ unchen, 2010), p. 1673.
Brailove, A., Kang, S., Fujisaka, A., Henley, F., 2010. First demonstration of high volume
manufacturing of kerf-free polymax wafers. In: 25th European PVSEC (WIP, M€ unchen,
2010), p. 1613.
Buchwald, R., Fr€ ohlich, K., W€ urzner, S., Lehmann, T., Sunder, K., M€ oller, H.J., 2013a.
Sub-surface damage of silicon wafers sawn with diamond-coated wire.
In: Proceedings of the 28th EU PVSEC (Paris, 2013), p. 1502.
Buchwald, R., Fr€ ohlich, K., W€ urzner, S., Lehmann, T., Sunder, K., M€ oller, H.J., 2013b.
Analysis of the sub-surface damage of mc- and cz-Si wafers sawn with diamond-plated
wire. Energy Procedia 38, 901.
Buchwald, R., W€ urzner, S., Fr€
ohlich, K., Fuchs, M., Retsch, S., Lehmann, T., M€ oller, H.J.,
2014. Analysis of the topography and the sub-surface damage of Cz- and mc-silicon
wafers sawn with diamond wire. In: Proceedings of the 40th IEEE Conference
(Denver 2014), p. 654.
Buijs, M., Korpel-van Houten, K., 1993a. A model for lapping of glass. J. Mater. Sci.
28, 3014–3020 (Chapman & Hill).
Buijs, M., Korpel-van Houten, K., 1993b. Three-body abrasion of brittle materials as studied
by lapping. Wear 166, 237.
Buijs, M., Korpel-van Houten, K., 1993c. Micromechanics of material removal mechanisms
from brittle surfaces. J. Mater. Sci. 28, 3014.
Bye, J.-I., Jensen, S.A., Aalen, F., Rohr, C., Nielsen, Ø., Gaumann, B., Hodsden, J.,
Lindemann, K., 2009. Silicon slicing with diamond wire for commercial production
of PV wafers. In: Proceedings of the 24th European PVSEC, p. 1269.
106 Hans Joachim M€
oller
Chauhan, R., Ahn, Y., Chandrasekar, S., Farris, T., 1993. Role of indentation fracture in free
abrasive machining of ceramics. Wear 166, 246.
Chen, C.P., Leipold, M., 1980. Fracture toughness of silicon. J. Am. Ceram. Soc. 59, 342.
Ciszek, T.F., 1981. Crystals 5. Springer, Berlin, p. 183.
Cook, L.M., 1990. Chemical processes in glass polishing. J. Non-Cryst. Solids 120, 152.
Depauw, V., Gordon, I., Beaucarne, G., Poortmans, J., Mertens, R., Celis, J.-P., 2009. Large
area monocrystalline silicon thin films by annealing of macroporous arrays. J. Appl. Phys.
106, 033516.
Depauw, V., Qiu, Y., Van Nieuwenhuysen, K., Gordon, I., Poortmans, J., 2010. Epitaxy-
free monocrystalline silicon thin film: first steps beyond proof-of-concept solar cells.
In: 25th European PVSEC (WIP, M€ unchen, 2010), p. 2843.
Depauw, V., Qiu, Y., Van Nieuwenhuysen, K., Gordon, I., Poortmans, J., 2011. Epitaxy-
free monocrystalline silicon thin film: first steps beyond proof-of-concept solar cells.
Prog. Photovolt. Res. Appl. 19, 844.
Domnich, V., Gogotsi, Y., 2002. Phase transformations in silicon under contact loading.
Rev. Adv. Mater. Sci. 3, 1.
Doolittle, A.K., 1951. Studies in Newtonian flow. The dependence of the viscosity of liquids
on free space. J. Appl. Phys. 22, 1471.
Dross, F., Robbelein, J., Vandevelde, B., Kerschaver, E., Gordon, I., 2007. Stress-induced
large area lift off of crystalline Si films. Appl. Phys. A 89, 149.
Endr€os, A., Franke, D., Häßler, C., Kaleis, J., Koch, W., M€ oller, H.J., 2002. Bulk crystal
growth and wafering for photovoltaics. In: Luque, A., Hegedus, S. (Eds.), Handbook
of Photovoltaic Engineering. Wiley-VCH, Weinheim, Germany, p. 1254.
Evans, A.G., Marshall, D.B., 1981. Fundamentals of Friction and Wear. ASM, Metals Park,
OH, p. 441.
Feltham, P., Banerjee, R., 1992. Theory and application of microindentation in studies of
glide and cracking in single crystals of elemental and compound semiconductors.
J. Mater. Sci. 27, 1626.
Ferris, T., Chandrasekar, S., 1994. Wear mechanisms in ceramic machining. Appl. Mech.
Rev. 47, 204.
Fujisaka, A., Kang, S., Tian, L., Chow, Y., Belyaev, A., 2011. Implant-cleave processes with-
out kerf loss. Photovolt. World, p. 21–24.
Funke, C., M€ oller, H.J., 2003. Microscopic mechanisms of multi-wire sawing.
In: M€ oller, H.J., Roewer, G. (Eds.), Freiberger Forschungshefte B 327. TU
Bergakademie Freiberg, Freiberg, p. 206.
Funke, C., Kullig, E., Kuna, M., M€ oller, H.J., 2004a. Biaxial fracture test of silicon wafers.
Adv. Eng. Mater. 6, 594.
Funke, C., Sciurova, O., M€ oller, H.J., Stephan, M., Fr€ ohlich, K.J., Seifert, C.,
Bachmann, A., M€ uller, A., 2004b. Surface damage from multi wire sawing and mechan-
ical properties of silicon wafers. In: Proceedings of the 19th European PVSEC (WIP,
M€ unchen, 2004), p. 1226.
Funke, C., Kiriyenko, O., Sciurova, O., M€ oller, H.J., 2005. Surface damage from multi wire
sawing and mechanical properties of silicon wafers. In: Proceedings of the 20th European
PVSEC (WIP, M€ unchen, 2005), p. 1128.
Funke, C., Wolf, S., Stoyan, D., 2009. Modeling the tensile strength and crack length of wire
sawn silicon wafers. J. Sol. Energy Eng. 131, 011012-1.
Ghannam, M.Y., Alomar, A., Poortmans, J., Mertens, R., 2011. Role of hydrogen in pro-
moting the splitting of silicon film during high temperature annealing of porous silicon.
In: 26th European PVSEC (WIP, M€ unchen, 2011), p. 2821.
Global Market Outlook, 2012. Report of european photovoltaic industry association.
Gogots, Y., Baek, C., Kirscht, F., 1999. Raman microspectroscopy study of process-induced
phase transformations and residual stresses in silicon. Semicond. Sci. Technol. 14, 936.
Wafering of Silicon 107
Hamblin, M., Stachowiak, G., 1995. A multi-scale measure of particle abrasivity, and its rela-
tion to two-body abrasive wear. Wear 190, 237.
Henley, F., 2010. Kerf-free wafering: technology overview and challenges for thin PV
manufacturing. In: 35th IEEE PVSC, p. 387.
Henley, F., Brailove, A., Lamm, A., Heerwagen, T., Sauar, E., Nese, M., Steeman, R.,
Hammel, B., 2008a. Direct film transfer (DFT) technology for kerf-free silicon wafering.
In: Proc. 23rd European PVSEC (WIP, M€ unchen, 2008), p. 2017.
Henley, F., Lamm, A., Kang, S., Liu, Z., Tian, L., 2008b. Direct film transfer (DFT)
technology for kerf-free silicon wafering. In: Proc. 23rd European PVSEC (WIP,
M€ unchen, 2008), p. 1090.
Henley, F., Kang, S., Liu, Z., Tian, L., Lamm, A., 2009a. Beam induced wafering technology
for kerf-free thin PV manufacturing. In: 34th IEEE PVSC, p. 56.
Henley, F., Lamm, A., Kang, S., Liu, Z., Tian, L., Wang, J., Chow, Y., 2009b. Kerf-free
20–150 μm c-Si wafering for thin PV manufacturing. In: 24th European PVSEC
(WIP, M€ unchen, 2009), p. 886.
International Technology Roadmap for Photovoltaic, 2012. SEMI Europe Photovoltaics
Group.
Jang, J., Lance, M.J., Wen, S., Tui, T.Y., Pharr, G.M., 2005. Indentation induced phase
transformations in silicon. Acta Mater. 53, 1759.
Johnsen, L., Gastinger, K., Bjerkan, L., Rietzschel, R., M€ oller, H.J., 2009. Absolute temper-
ature measurements in wire sawing process. In: Proceedings of the 24th European
PVSEC (WIP, M€ unchen, 2009), p. 1248.
Kaminski, S., Rietzschel, R., Wagner, T., Funke, C., M€ oller, H.J., 2009. Multi-wire sawing
down to 80 μm. In: Proceedings of the 24th European PVSEC (WIP, M€ unchen, 2009),
p. 1299.
Kaminski, S., Wagner, T., Rietzschel, R., F€ utterer, W., Funke, C., M€ oller, H.J., 2010. The
role of the slurry transport in multi-wire sawing of solar silicon wafers. In: Proceedings of
the 25th European PVSEC (WIP, M€ unchen, 2010), p. 1315.
Kasai, T., Doy, T.K., 1992. Tribology from a view point of ultraprecision. Jpn. J. Tribol.
37 (5), 539.
Kondo, Y., Watanabe, N., Ide, D., Matsuki, T., Takato, H., Sakata, I., 2008. Character-
ization of multicrystalline silicon wafers for solar cell applications sliced with a fixed
abrasive wire. In: Proceedings of the 23rd European PVSEC (WIP, M€ unchen,
2008), p. 1297.
Lange, H., Schwirtlich, I., 1990. Process of manufacture: an improved columnar grained
polycrystalline silicon solar cell. J. Cryst. Growth 104, 108.
Larsen-Basse, J., 1993. Effect of atmospheric humidity on the dynamic fracture strength of
SiC abrasives. Wear 166, 93.
Lawn, B., 1993. Fracture of Brittle Solids. Cambridge University Press, New York.
Li, H., Bradt, R., 1996. The effect of indentation-induced cracking on the apparent micro-
hardness. J. Mater. Sci. 31, 1065.
Li, J., Kao, I., Prasad, V., 1998. Modelling stresses of contacts in wire saw slicing of polycrys-
talline and crystalline ingots: application to silicon wafer production. J. Electron. Packag.
120, 123.
Malkin, S., Ritter, J.E., 1989. ASME J. Eng. Industry 111, 167.
Meißner, D., Hurka, B., Rietzschel, R., M€ oller, H.J., Anspach, O., 2012. Decrease of wire
tension in the wire web: towards stable cutting conditions for all wafers. In: 27th EU
PVSEC (WIP, M€ unchen, 2012), p. 1076.
Meißner, D., Sch€ onfelder, S., Hurka, B., Zeh, J., Sunder, K., K€ opge, R., Wagner, T.,
Gr€ un, A., Hagel, H., M€ oller, H.J., Schwabe, H., Anspach, O., 2013. Loss of wire tension
in the wire web during the slurry based multi wire sawing process. Energy Procedia
38, 83.
108 Hans Joachim M€
oller
Mizushima, I., Sato, T., Taniguchi, S., Tsunashima, Y., 2000. Empty-space-in-silicon tech-
nique for fabrication of a silicon-on-nothing structure. Appl. Phys. Lett. 77, 3290.
M€oller, H.J., 2004. Technology and basic mechanism of multi-wire sawing. Adv. Eng.
Mater. 6, 501.
M€oller, H.J., 2006. Wafering of silicon crystals. Phys. Status Solidi A 203 (4), 657.
M€oller, H.J., 2008. Crystal sawing technology. In: Scheel, H.J., Capper, P. (Eds.), Crystal
Growth Technology. Wiley-VCH, Weinheim, p. 457.
Möller, H.-J., 2015. Wafer processing. In: Rudolph, P. (Ed.), Handbook of Crystal Growth,
second ed. Elsevier, Boston, pp. 715–755 (Chapter 18).
M€oller, H.J., Buchwald, R., Winstanley, S., 2012. Basic mechanism of diamond wire sawing.
In: Proceedings of the 6th International Conference on Crystalline Silicon for Solar
Cells.
M€oller, H.J., Retsch, S., Rietzschel, R., 2013. On the origin of wafer saw marks in slurry
based multi—wire sawing. In: Proceedings of the 28th EU PVSEC (WIP, M€ unchen,
2013), p. 927.
Mueller, A., Cherradi, N., Nasch, P.M., 2003. Proceedings of the 3rd World Conference on
Photovoltaic, Energy Conversion.
Nasch, P.M., Schneeberger, S., 2005. A theoretical modelling of consumables as a tool for
cost reduction in silicon wafering using multi-wire slurry saw. In: Proceedings of the
20th European PVSEC, p. 1139.
Nasch, P.M., Cherradi, N., Mueller, A., 2004. In: Proceedings of the 19th European
PVSEC, p. 357.
Orellana Pérez, T., Schmid, C., Riepe, S., M€ oller, H.J., Reber, S., 2009. Mechanical char-
acterization of wafer equivalent substrate materials from alternative silicon feedstock.
In: Proceedings of the 24th EU PVSEC (WIP, M€ unchen, 2009), p. 1228.
Park, J.E., Dibiase, N., 2012. Correlation of wafer surface defects with wire sawing param-
eters. In: Proceedings of the 27th European PVSEC (WIP, M€ unchen, 2012), p. 957.
Petermann, J.H., Zielke, D., Schmidt, J., 2012. 19% efficient and 43 μm thick crystalline Si
solar cell from layer transfer using porous silicon. Prog. Photovolt. Res. Appl. 20, 1.
Poon, C., Bushan, B., 1995. Comparison of surface roughness measurements by stylus pro-
filer, AFM and non-contact optical profiler. Wear 190, 76.
Proventa, E., Brize, V., Dreveta, B., Coustier, F., Flahaut, E., 2012. Diamond wire slicing:
towards silicon kerf recycling. In: Proceedings of the 26th European PVSEC, p. 1965.
Retsch, S., Jentsch, S., M€oller, H.J., 2012. Influence of shape and size distribution of abrasives
on the cutting performance in lapping related to multi wire sawing. In: Proceedings of
the 27th EU PVSEC (WIP, M€ unchen, 2012), p. 947.
Retsch, S., Fuchs, M., M€ oller, H.J., 2014. Investigation of the impact of ingot temperature
and silicon debris on the generation of saw marks in multi-wire sawing. In: Proceedings
of the 29th EU PVSEC (WIP, M€ unchen, 2014), p. 738.
Reuter, M., Brendle, W., Tobail, O., Werner, J., 2008. 50 μm thin solar cells with 17.0%
efficiency. Sol. Energy Mater. Sol. Cells 93, 704.
Rietzschel, R., Senf, A., Seelmann-Eggebert, H.P., Kaminski, S., Wagner, T., Funke, C.,
M€ oller, H.J., 2010. Investigation of polyglycol based slurries for multi wire sawing with
lower forces and less hygroscopicity. In: Proceedings of the 25th European PVSEC
(WIP, M€ unchen, 2010), p. 1596.
Siekman, G., 1987. Polishing of brittle amorphous alloy. Wear 117, 359.
2012. Solar Battery Technologies—Present Market Situation and Future Perspective. vol. 2,
Fuji Keizai Co., Japan.
Solar PV Market Forecast, 2012. Report Navigant Consulting.
Struble, L.J., Sun, G.K., 1993. Flow and microstructure of dense suspensions. In: Struble, Z.,
Maitland, G.C. (Eds.), Materials Research Society (MRS), Symposia Proceedings, Pitts-
burgh MRS, vol. 289, p. 173.
Wafering of Silicon 109
Su, Y.-T., Horng, C.-C., Hwang, Y.-D., Guo, W.-K., 1996. Effects of tool surface irreg-
ularities on machine rate of a hydrodynamic polishing process. Wear 199, 89.
Van Nieuwenhuysen, K., Depauw, V., Martini, R., Govaerts, J., Debucquoy, M.,
Radhakrishnan, H.S., Gordon, I., Bearda, T., Baert, K., Poortmans, J., 2012. High qual-
ity epitaxial foils, obtained by a porous silicon based layer transfer process, for integration
in back contacted solar cells. In: 27th European PVSEC (WIP, M€ unchen, 2012), p. 2471.
Verspui, M., de With, G., van der Varst, P., Buijs, M., 1995. Bed thickness and particle size
distribution in three-body abrasion. Wear 188, 102.
Wagner, T., M€ oller, H.J., 2008. A 3D wire saw model of the slurry flow to predict forces
exerted upon silicon ingots during cutting. In: Proceedings of the 23rd European PVSEC
(WIP, M€ unchen, 2008), p. 1315.
Wagner, T., Behm, T., Rietzschel, R., Kaminski, S., Funke, C., M€ oller, H.J., 2010. Spatial
distribution of wafer subsurface damage induced by wire sawing. In: Proceedings of the
25th European PVSEC (WIP, M€ unchen, 2010), p. 1267.
Wald, F.V., 1981. Crystals 5. Springer, Berlin, p. 147.
Wei, S., Kao, I., 2000. Vibration analysis of wire and frequency response in the modern wire
saw manufacturing process. Trans. Int. J. Vib. Sound 231, 1383.
Weppelmann, E., Field, J., Swain, M., 1993. Mechanical properties of silicon. J. Mater. Res.
8, 830.
W€urzner, S., Buchwald, R., M€ oller, H.J., 2014a. Surface damage and mechanical stability of
silicon wafers. Phys. Stat. Sol. 156.
W€urzner, S., Buchwald, R., Retsch, S., M€ oller, H.J., 2014b. Investigation of the develop-
ment of the microcrack structure in scratch tests with single diamond particles on mono-
crystalline silicon wafers. In: Proceedings of the 29th EU PVSEC (WIP, M€ unchen,
2014), p. 742.
W€urzner, S., Falke, A., Buchwald, R., M€ oller, H.J., 2015. Determination of the surface dam-
age of diamond wire sawn silicon wafers depending on wire velocity. Energy Procedia 40.
Yang, F., Kao, J., 1999. Interior stress for axisymmetric abrasive indentation in the free abra-
sive machining process. J. Electron. Packag. 121, 191.
Yang, F., Kao, I., 2000. Free abrasive machining in slicing brittle materials with wire saw.
J. Electron. Packag. 123, 254.
CHAPTER THREE
Contents
1. Reliability 111
2. Metastabilities 115
2.1 Conclusions on Metastabilities 117
3. Partial Shading and Hotspots 117
3.1 Conclusions on Partial Shading 132
4. Potential-Induced Degradation 132
4.1 Conclusions on PID 138
5. Back Contact 138
Conclusions on Back Contact 147
References 148
1. RELIABILITY
For the success of a PV technology, several conditions have to be ful-
filled. Efficiency and the associated costs certainly exhibit the dominating
criteria. However, the reliability of a technology or the long-term stability
of a PV system cannot be neglected at all (Noufi and Zweibel, 2006), which
can be made plausible from a simple calculation example. Let us assume that
a module manufacturer guarantees 80% of the initial output power after
20 years assuming a linear degradation with time. That means that on aver-
age (over 20 years), only 90% of the power rating is to be expected. Con-
sequently, the return on investment should be calculated based on this
reduced averaged power rating. From this (perhaps too) simple example,
it is evident that the long-term stability plays a nonnegligible role for the
economic success of a PV technology (Wendlandt et al., 2012).
These considerations impose several challenges for the development of a
PV technology:
teol
a¼ (1)
tD
with teol being the end-of-life time when the specifications of a module are
violated and tD being the equivalent time under stress conditions in order to
accelerate the degradation. Parameter drifts of solar cells can be induced (and
therefore accelerated) by several factors. The operation temperature accel-
erates or slows down parameter drifts with a characteristic activation energy
EA describing the temperature dependence:
EA TD TS
1 1
aT ¼ e k (2)
n
EAT
ΔP ¼ AT *e kT + AI *
ID EAI
*e kT *Δt (4)
IS
(AT, AI: prefactors; EAT and EAI: activation energies in the dark and under
illumination, respectively).
That means that a certain solar cell parameter P drifts within a small
period of time Δt by a value ΔP as defined by the equation above under
the assumption of constant conditions within this small period of time.
For a given temperature and illumination profile under field conditions,
these parameter drifts have to be added up or integrated in order to obtain
and predict parameter drifts under field conditions. When looking again at
Eq. (4), it is evident that under constant conditions a linear change of the
parameters is predicted. This has to be considered more carefully as satura-
tion effects may occur (e.g., Voc only rises or drops to a certain saturated
value). In order to describe such processes, more complex mathematical
descriptions and models have to be applied. It is nowadays certainly not
the problem to solve complex differential equations, however, to determine
and develop the underlying model with the associated parameters certainly
imposes a severe challenge which is far from being completely understood
and solved. Equation (4) is suitable to describe Cu(In,Ga)Se2 (CIGS)-based
solar cells as a dark anneal reduces the open-circuit voltage Voc and the fill
factor FF, whereas keeping the devices even under a pretty low level of illu-
mination enhances these solar cell parameters.
In the following, the determination of acceleration factors and especially
the extrapolation to field conditions will be discussed based on experimental
results published in Mack et al. (2009). Figure 1 shows parameter drifts (nor-
malized) as a consequence of a dark anneal at different temperatures. It is
obvious that this solar cell parameter decreases faster with increasing temper-
ature. For these experiments, the end-of-lifetime was defined in such a way
that teol is reached when the FF falls below 90% of its initial value (t0.9). These
t0.9 values were extracted at the different temperatures and plotted in an
Arrhenius diagram as shown in Fig. 2 in order to determine the activation
energy and extrapolate t0.9 to field conditions.
As shown in Fig. 2, these results yield a straight line such that t0.9 can be
easily extrapolated graphically (or mathematically) to other temperatures. As
indicated in Fig. 2, t0.9 can be estimated to 45 years at 45 °C. 45 °C is a typ-
ical value for NOCT (normal operation cell temperature). Therefore, this
device could be operated around 45 years (in the dark) at NOCT without
violating the degradation criterion introduced above. Furthermore, it
114 Thomas Walter
1.0
0.9
Fill factor (normalized)
0.8
0.7
85 °C
125 °C
0.6 165 °C
205 °C
0.5
0.4
0 100 200 300 400 500 600
Time (h)
Figure 1 FF parameter drift of a CIGS-based solar cell for dark anneals at various
temperatures (Mack et al., 2009). (Reprinted, with permission from EUPVSEC).
Temperature (°C)
227 182 144 112 84 60 40
106 102
45 a
105 5a 101
104 100
t0.9 (a)
t0.9 (h)
101 10−3
65 °C 45 °C
100 10−4
2.0 2.2 2.4 2.6 2.8 3.0 3.2
1000 / T (1/K)
Figure 2 Arrhenius diagram of t0.9 (Mack et al., 2009). (Reprinted, with permission from
EUPVSEC).
Reliability Issues of CIGS-Based Thin Film Solar Cells 115
2. METASTABILITIES
One of the most puzzling properties also of highly efficient CIGS-
based devices is the appearance of so-called metastabilities which have
accompanied the development of CIGS-based devices since its early days.
These metastabilities are in most cases reversible parameter drifts of primarily
Voc and FF with extremely long time constants (hours and days). These
parameter drifts are mainly initiated by a high-temperature dark storage
of the devices (dark anneal) leading to a decrease of the output power or
by keeping the modules under illumination (light soak) which improves
the performance. Regarding the reliability and also certification of a
CIGS-based technology, these parameter drifts impose a severe problem
for measurements of modules and devices. In order to verify that modules
withstand a certain stress test, the measurements prior and after the stress
have to be undertaken with the module being in the same metastable state.
116 Thomas Walter
Therefore, usually the modules are light-soaked for a certain time prior to
the measurements as this appears to be the “natural state” of a solar cell.
However, the duration and the conditions of such treatments are the content
of an ongoing discussion.
What is now the physics behind these metastable parameter drifts? It is
widely accepted that a change of the charge distribution in the device is
responsible for the observed parameter drifts. One model postulates ampho-
teric defect complexes which can change their charge state depending on the
position of the quasi-Fermi levels and/or optical transitions (Lany and
Zunger, 2006, 2008). The extremely long time constants associated with
these metastabilities are explained in terms of a configuration change as a con-
sequence of the modification of interatomic spacings. Taking the associated
capture cross sections for electrons and holes into account, a significant por-
tion of the observed metastabilities can be explained. A second model
resulting also in a change of the charge distribution is based on the existence
of mobile ions in these chalcopyrite semiconductors. Not only Cu ions but
also the extremely mobile Na are candidates for this model. It should be
noted that the existence of Na in these materials is not an unwanted and det-
rimental impurity but is one of the most important prerequisites for the suc-
cess of CIGS (Bl€ osch et al., 2013; Salome et al., 2014). Recently, it was
demonstrated that a postdeposition treatment involving KF might even lead
to devices with higher efficiencies (Chirila and Reinhard, 2013).
From a first glance, these two models—i.e., electronic defects and mobile
ions—should be easily distinguishable. However, this discussion is still
ongoing with a lot of arguments from both sides. For a discussion on these
metastabilities, the reader should refer to the numerous literature including
some excellent review articles (Igalson et al., 2011, 2013; Kempa et al., 2013;
K€otschau et al., 2001; Moore et al., 2014).
In order to explain a certain impact of partial shading on Voc parameter
drifts, the influence of a bias across the heterojunction will be considered
in the following (Mack et al., 2009). Figure 3 shows Voc parameter drifts
of a CIGS-based solar cell under different conditions (5 h, 165 °C), from left
to right:
• In the dark unbiased
• Open-circuit conditions: 0.01 sun
• Open-circuit conditions: 0.31 sun
• Open-circuit conditions: 1 sun
• Short-circuit conditions (KS): 1 sun
• In the dark: Applied bias of +400 mV (roughly corresponding to MPP)
Reliability Issues of CIGS-Based Thin Film Solar Cells 117
50
Delta Voc (mV)
0
0% 1% 31% 100% 100% & 400 mV
−50 KS
−100
−150
−200
Figure 3 Voc parameter drifts under various stress conditions (Mack et al., 2009). The
stress conditions are explained in the text. (Reprinted, with permission from EUPVSEC).
glance, this seems to be not critical as the power loss due to the shading of
one cell in a module, which consists of up to 100 interconnected devices,
appears to be negligible. However, such a configuration can lead under
certain circumstances to the damage of the module which might show a
significant power loss. This phenomenon will be explained and discussed
in the following. Consider a module consisting of n interconnected cells
with one shaded cell and the module being externally short circuited as illus-
trated in Fig. 4.
Looking at the equivalent circuit of this configuration (Fig. 5), it
becomes evident that the photocurrent produced by the illuminated devices
cannot flow through the shaded cell since a current flow through a perfectly
rectifying diode assumed for the moment is not possible. Thus, the illumi-
nated devices operate at Voc condition. Looking now at the externally short-
circuited module and taking into account Kirchhoff’s laws, it is clear that a
reverse bias amounting to (n 1) * Voc (with Voc being the open-circuit
voltage of a single cell) drops across the shaded device. That means that
for a module with a module Voc of 60 V, a negative bias of about 59 V drops
across the shaded device. Such a high reverse bias can lead to an irreversible
junction breakdown damaging the device and also the module severely.
A countermeasure which can be practically only taken on a module level
is the addition of bypass diodes which allow the generated photocurrent
to bypass the shaded device, therefore limiting the arising reverse bias.
To protect each cell by an external bypass diode in a thin-film module is
practically (and especially economically) not feasible. However, there seems
to be a mechanism inherent to CIGS-based devices which exhibit a certain
self-protection mechanism. This is illustrated in Figs. 6 and 7 which show
the measured dark IV characteristics of a CIGS-based solar cell produced
by the coevaporation method.
URev
URev
n interconnected devices
1E+2
1E+1
Current density (mA/cm2)
1E+0
−8 −7 −6 −5 −4 −3 −2 −1 0 1
1E−1
1E−2
1E−3
1E−4
Voltage (V)
Figure 6 Dark IV characteristics of a CIGS-based solar cell produced by coevaporation.
The red (gray in the print version) line indicates an exponential current increase for a
bias below approximately 5.5 V.
110
90
70
Current density (mA/cm2)
50
30
10
−8 −7 −6 −5 −4 −3 −2 −1 −10 0 1 2
−30
−50
Voltage (V)
Figure 7 Reverse IV characteristics of a CIGS-based solar cell in the dark. The horizontal
red (gray in the print version) line indicates the value of the photocurrent density
resulting in a reverse bias of approximately 7.5 V across the shaded cell.
dissipated in a possibly small area shunt resistance that might lead to a vast
increase of the local temperature most likely accompanied by a local
melting of the layers or even by the development of cracks in the module
glass. It is obvious that this heat dissipation of 15 W represents an upper
limit as series resistances limit the current flowing through the resistance
to much smaller values. Therefore, it is evident that for a complete
understanding of the spatial distribution of the currents, an electrical
2D modeling of a module is necessary. Approaches for such an advanced
simulation will be presented and discussed below.
• Negative bias metastabilities: As pointed out earlier, CIGS-based devices
“like” a positive bias dropping across the junction. On the other hand,
they “hate” a negative bias as a result of a shaded cell in a module leading
to a significant drop of Voc as shown below. However, these parameter
drifts appear to be reversible. Especially after a subsequent light soak
(which is a realistic case for field conditions after shading), Voc recovers
to its initial value with a time constant typical for these metastabilities.
Thus, this negative bias metastability can be considered as reversible
leading to no permanent damage of the module. However, it neverthe-
less leads to certain power losses since the time constants of these meta-
stabilities are rather in the minutes than in the seconds range.
In the following, this current breakthrough of CIGS-based devices that
limits the arising reverse bias of a shaded cell will be considered and discussed
in more detail. Figure 9 compares the reverse characteristics in the dark
and under illumination using spectral-edge filters with various cutoff
wavelengths.
Using these spectral-edge filters, it is possible to illuminate the devices
with a spectrum beginning (at its shorter wavelength end) at the cutoff
wavelength of these optical filters. From these measurements, it can be con-
cluded that absorption and photogeneration in the CdS buffer have a distinct
impact on the reverse breakthrough voltage. Without absorption in the CdS
buffer layer (cutoff wavelength of the edge filter >500 nm), the superposi-
tion principle is fulfilled. Therefore, the reverse breakdown voltage is similar
for the dark and illuminated (without photogeneration in the CdS) device,
whereas for sufficient photogeneration in the CdS the reverse breakdown
voltage is reduced significantly to values around 2 V (Mack et al., 2009;
Szaniawski et al., 2013). That means that white light illumination can reduce
the reverse bias which is necessary to carry a certain reverse bias. At a first
glance, this appears to be without practical use regarding the critical break-
down voltage in the shadowed cell. However, such a shading is under real
Reliability Issues of CIGS-Based Thin Film Solar Cells 123
0.00
−0.01
Current (A)
Dark
−0.02 700 nm
600 nm
500 nm
450 nm
400 nm
−0.03
−7 −6 −5 −4 −3 −2 −1 0 1
Voltage (V)
Figure 9 Dark and illuminated reverse characteristics using different spectral filters
(Mack et al., 2009). (Reprinted, with permission from EUPVSEC).
−2
Bias at which reverse current
−3 equals photocurrent
Reverse bias (V)
−4
−5
−6
−7
−8
1E−3 0.01 0.1 1 10 100
Illumination intensity (mW/cm2)
Figure 10 Breakdown voltage for varying illumination intensity (Mack et al., 2008).
(Reprinted, with permission from EUPVSEC).
Ev Ev
−2 −2
CIGS
−4 −4
−6 −6
CdS CdS
−8 −8
0 0.5 1
0 0.5 1
x (µm)
x (µm)
Figure 11 Simulated band diagrams under reverse bias with and without interface
states (Mack et al., 2008). x ¼ 0 μm denotes the position of the back contact, x ¼ 1 μm
the position of the heterointerface with CdS and ZnO. (Reprinted, with permission from
EUPVSEC).
Reliability Issues of CIGS-Based Thin Film Solar Cells 125
et al., 2011; Tran et al., 2011). Figure 13 illustrates the principle of this 2D
network simulation.
The network consists of a multitude of small solar cells which are inter-
connected by lumped resistances representing the corresponding ZnO and
Mo resistances. Consequently, a 2D network is formed which can be solved
by network simulators such as PSPICE or ADS. It is therefore possible to
determine the electric potential and currents at each solar cell or lumped
resistance. Furthermore, defects can be easily inserted into the network
and simulated. The device characteristics discussed above have to be
modeled as an input for the network simulator. Figure 14 exhibits a possible
underlying model for a single solar cell in the network.
This model consists of a principal diode with an associated photocurrent
source. Additionally, a voltage-dependent current source has been added. In
order to account for the reverse characteristics, a reverse diode was
implemented. The solar cell model was completed by a series and a parallel
resistance. The parameters of these lumped elements can be determined
from measurements of de-embedded solar cells. In addition, the used diode
models can usually account for complex device characteristics including the
temperature dependence of the device parameters. Therefore, experimen-
tally determined device parameters can be modeled quite accurately
y
z
NonlinVCCS
CSRC1
Coeff = list(0.005)
I_DC
SRC1
Idc = inte*iph*lae*bre/1000*1e3
R
R2
R = rs Ohm
Diode
DIODE2
Model = DIODEM2
Area =
Temp =
Mode = nonlinear
Figure 14 Equivalent circuit of a solar cell used for network simulation (Mack et al.,
2010). (Reprinted, with permission from EUPVSEC).
Figure 15 Measured and simulated EL and LIT images (Tran et al., 2011). (Reprinted, with
permission from EUPVSEC).
Figure 16 EL image for low current injection (Mack et al., 2010). (Reprinted, with permis-
sion from EUPVSEC).
Reliability Issues of CIGS-Based Thin Film Solar Cells 129
Figure 17 EL image for high current injection (Mack et al., 2010). (Reprinted, with
permission from EUPVSEC).
(in forward direction) taken at a low and high current—exhibit some inter-
esting features. At a low current density, the luminescence of the complete
cell is bleached due to the fact that such a rather low current is completely
carried by the shunt defect. Thus, the cell areas (of the defected cell) outside
the shunt defect do not have to carry any current with the consequence that
no luminescence arises out of this cell due to the missing diode current.
When looking at the neighboring cells, the phenomenon of the current
being directed to the shunt defect is clearly visible. A cone-like area with
increased luminescence points at the position of the shunt defect indicating
a current flow to this point. Such a current flow leads to a voltage drop espe-
cially in the ZnO window layer due to the finite conductivity of the TCO.
As a consequence of this voltage drop, the bias across the pn junction which
determines the luminescence intensity due to the associated splitting of the
quasi-Fermi levels changes and therefore also the luminescence intensity.
Looking now at the condition of a high enforced current density, the EL
image exhibits a completely different pattern. On the defected cell, only
the immediate surroundings of the defect is bleached whereas the remaining
130 Thomas Walter
0.70
Open-circut voltage (V)
0.65
0.60
0.55
0.50
0 50 100 150 200
Time (h)
Figure 18 Voc parameter drifts as a consequence of an applied bias (Ott et al., 2011).
(Reprinted, with permission from EUPVSEC).
Reliability Issues of CIGS-Based Thin Film Solar Cells 131
parameter should be taken into account when designing modules which are
immune against partial shading.
As discussed above, the application of a negative bias (for a certain time)
to a CIGS-based device leads to distinct metastabilities with associated
parameter drifts especially of Voc and FF. In Fig. 18, the parameter drifts
of Voc for endurance tests at 165 °C under different bias conditions are
shown (Ott et al., 2011). For a positive bias of 0.4 V (and under an illumi-
nation of 0.2 sun), the light-soaking effect described earlier is clearly shown
leading to a remarkable increase of Voc with respect to the initial values.
However, unbiased conditions or a small negative bias of 0.1 V lead to
a significant initial drop of Voc followed by a saturation. Thus, negative
biases which arise as a consequence of partial shading of modules can lead
to significant parameter drifts affecting the module performance signifi-
cantly. However, as also indicated above such parameter drifts appear to
be metastable in the sense that these parameter drifts relax (especially after
the removal of the origin of partial shading) to an eventual recovery of
the original (or at least very close to the original) performance.
In Fig. 19, the result of a partial shading (one cell was shaded in a mini-
module) and the subsequent relaxation is illustrated. The module exhibits an
initial performance drop of about 6% as a consequence of 1 h shadowing
followed by an almost complete relaxation after a light soak of together
approximately 45 min. However, it should be noted that these metastable
changes as a consequence of partial shading definitively depend on the mod-
ule size and the associated number of cells and also on the number of shaded
1.05
Module power (norm.)
1.00
0.95
0.90
0.85
0.80
Initial
shadowing
After 5 min
After 15 min
After 25 min
light soak
After 1 h
light soak
light soak
4. POTENTIAL-INDUCED DEGRADATION
PID is a mechanism which can lead to a dramatic damage of affected
PV modules. Hereby, a potential drop across the front or substrate glass leads
to a leakage current associated with mobile Na ions in the glass. PID was first
discovered in highly efficient monocrystalline Si modules (Pingel et al.,
2010). Nevertheless, it was pointed out that these parameter drifts were
reversible leading to a complete recovery of the initial module performance.
For thin-film Si modules, irreversible damage was reported as a result of
TCO corrosion associated with the out-diffusion of Na from the cover glass
into the TCO where a reaction with in-diffused water occurred. However,
PID is not only restricted to Si-based technologies but can be observed for
almost all PV technologies. For CIGS, only a few publications exist regard-
ing the impact of PID on the module performance (Fjallstrom et al., 2013;
Lechner et al., 2013). The latter publication is based on research undertaken
at the Angstrom Center in Uppsala, Sweden. It was clearly shown in this
publication (Fjallstrom et al., 2013) that the efficiency of CIGS-based
devices deposited by simultaneous evaporation could be reduced to
“zero” by applying a positive bias onto the back side of the glass substrate
with respect to the Mo back contact. Hereby, the high voltage was applied
between the Mo back contact and a metallic tape attached to the backside of
Reliability Issues of CIGS-Based Thin Film Solar Cells 133
1.0
0.9
0.8
Normalized initial efficiency
0.7 1—SLG
2—High Na
0.6 3—Na only
0.5 4—Na-free
5—Low Na
0.4
0.3
0.2
0.1
0.0
0 10 20 30 40 50
Duration of stress (h)
Figure 20 PID parameter drifts for various glass substrates (SLG: soda lime glass).
Copyright IEEE 2013. Reprinted, with permission, from Fjallstrom et al. (2013).
the substrate glass. This is shown in Fig. 20 which also clearly points out that
this degradation is related to the amount of Na in the glass substrate (as can be
deduced from the different substrate glasses). In other words, the use of
Na-free or Na-poor substrate glasses results in an immunity against PID,
whereas substrates with a higher Na content are affected by this severe deg-
radation mechanism.
A closer look at the degraded IV characteristics as illustrated in Fig. 21
reveals that PID primarily affects Voc (and FF) in the initial phase (compare,
for example, the 24-h case) followed by an additional decrease of the pho-
tocurrent in the final stadium which is characterized by an almost negligible
Voc. This decrease of Voc could be at a first sight attributed to a significantly
reduced minority carrier lifetime. However, this should also have a severe
impact on the diffusion length and therefore on the photocurrent collection
which is not really the case when looking at Fig. 21 in the initial degradation
phase. Alternatively, a significant reduction of the net acceptor density could
also explain this severe parameter drift of Voc as with a decreasing net accep-
tor density in the absorber the barrier in the junction decreases which then
reduces Voc. At least in the initial stage of the degradation, this mechanism
134 Thomas Walter
50
40 SLG—sample 1
0h
30 1h
Current density (mA/cm2)
7h
20
24 h
50 h
10
–10
–20
–30
–40
–0.4 –0.2 0.0 0.2 0.4 0.6 0.8
Voltage (V)
Figure 21 Evolution of IV characteristics during PID stress test for a soda lime glass
substrate (SLG). Copyright IEEE 2013. Reprinted, with permission, from Fjallstrom et al.
(2013).
should not affect the photocurrent collection severely as the electric field
even stretches deeper into the p-CIGS. In the final stadium, the electric field
apparently gets too low in order to assure a rather complete current collec-
tion as demonstrated by the reduced photocurrent (the 50-h case). There-
fore, changes of the charge distribution in the heterojunction which also
underlie the metastabilities described above might be the physical cause
for the observed parameter drift due to PID damage.
In the course of this work, additional chemical analysis was carried out in
order to understand the associated changes in the material as a consequence
of a PID damage. As can be deduced from this work (Fig. 22), the applica-
tion of PID stress to the device leads to a significant agglomeration of Na in
the CdS buffer layer and presumably also in the absorber region close to the
interface. These GDOES results (glow discharge optical emission spectros-
copy) led to the conclusion that Na out-diffuses from the front glass and
agglomerates in the buffer layer. This out-diffusion seems to be supported
by the high electric field in the glass substrate, whereas the transport into
the CIGS should occur via diffusion. One would therefore expect a concen-
tration gradient from the Mo back contact to the buffer layer which is, how-
ever, not the case. Furthermore, relaxation of damaged devices which will
be discussed later would be based on out-diffusion of Na from the CdS
Reliability Issues of CIGS-Based Thin Film Solar Cells 135
CdS CIGS Mo
Counts (a.u.)
Se
Mo
Cd
Na control
Na PID
buffer layer. However, what should be the driving force for such a process?
In principle, diffusion is from a thermodynamic point of view irreversible. It
is therefore hard to explain that PID-damaged CIGS can relax. In an addi-
tional work of the group in Uppsala, GDOES studies were performed after
an almost complete relaxation of a severely damaged device showing a very
similar Na distribution as measured in the damaged state. These consider-
ations make clear that the microscopic origin of the PID damage is not
yet understood. Similar to existing models for PID damage of crystalline
Si solar cells, models involving the out-diffusion of Na or models based
on pure charging mechanisms are discussed. However, so far the published
results cannot exclude or verify either of these models.
Table 1 clearly shows that the overwhelming part of the damage could be
restored after 6 months under lab conditions without additional measures
136 Thomas Walter
such as an opposite polarity of the PID stress test. Quasi all solar cell param-
eters could be restored with the photocurrent being even slightly higher
compared to the initial state. This could be an indication that due to a larger
space charge width as a consequence of a still reduced net acceptor density,
the photocurrent collection gets improved. However, more chemical and
electrical characterizations are necessary in order to support or rule out
the underlying mechanisms which are still in a somewhat speculative state.
In a second publication (Lechner et al., 2013), several PV technologies
were compared with respect to their PID immunity. As can be deduced
from Fig. 23, the different CIGS-based modules in this study are affected
in an extremely different way. Some modules or technologies (no additional
information about the differences between these CIGS-based modules is
given) seem to be affected not at all, whereas other modules degrade to
below 40% of their initial performance.
In this publication also the development of the module IV characteristics
under illumination is given as illustrated in Fig. 24. Similar to the parameter
drifts observed on a cell level, Voc (and FF) are affected primarily, whereas Isc
seems to be only affected slightly even in the final state supporting the model
of changes of the charge distribution as a consequence of the PID stress.
Another extremely interesting aspect published in this work concerns the
overall charge induced by PID (integrated leakage current). This parameter
seems to be a fundamental parameter to describe the degree of PID damage.
As illustrated in Fig. 25, PID stress performed at two different temperatures
1.1
Si-TF 1
1 Si-TF 2
0.9 Si-TF 3
0.8 Si-TF 4
Relative power
Si-TF 6
0.7
CIGS 2
0.6 CIGS 3
0.5 CIGS 4
0.4 CIGS 5
c-Si 1
0.3
c-Si 2
0.2 c-Si 3
0.1 c-Si 4
c-Si 5
0
0 200 400 600 800 1000
PID exposure time (h)
Figure 23 PID parameter drifts of various PV technologies (Lechner et al., 2013).
(Reprinted, with permission from EUPVSEC).
Reliability Issues of CIGS-Based Thin Film Solar Cells 137
1.5
Current (A)
Initial
1 50 h PID
100 h PID
0.5
0
0 20 40 60 80 100
Voltage (V)
Figure 24 PID parameter drifts on the module level (Lechner et al., 2013). (Reprinted,
with permission from EUPVSEC).
1.1
1
0.9 CIGS 60/85
0.8 CIGS 85/85
CIGS 85/85
Relative power
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0 5 10 15 20
Charge (c/m2)
Figure 25 PID parameter drift versus integrated leakage current (charge) (two modules:
85 °C, 85% relative humidity (RH); one module: 60 °C, 85% RH) (Lechner et al., 2013).
(Reprinted, with permission from EUPVSEC).
5. BACK CONTACT
Another issue regarding the long-term stability of thin-film solar cells
in general is the stability of the contacts. Especially, the back contact stability
issue of CdTe-based devices has been discussed in literature to a large extent
(Sites, 2010). Regarding CIGS-based solar cells, a Mo back contact is used
quite frequently. However, it should be noted that the actual interface
between the semiconductor and the metal is formed by a MoSex or
MoSx,Sey layer, respectively (Lee et al., 2013). In addition, it was published
that the formation of this intermediate layer is beneficial with respect to elec-
trical contact properties (Wada et al., 2001). In the following, the impact of a
Schottky back contact on the device performance and on the reliability of
solar cells will be described and discussed. First, some well-known charac-
teristics of CIGS-based solar cells that deviate significantly from an ideal
behavior will be considered with respect to the impact of a back
contact diode.
In Fig. 26, low-temperature IV characteristics of a highly efficient CIGS-
based solar cell as a function of illumination intensity (0–100%) are
reproduced (Ott et al., 2013a). These characteristics exhibit a behavior
which deviates from a “normal” solar cell with respect to several properties:
• A crossover of the dark and illuminated characteristics is clearly visible.
• Voc is independent of the illumination intensity.
• The dark current is blocked in forward direction (at least in the regime
shown in Fig. 26).
Reliability Issues of CIGS-Based Thin Film Solar Cells 139
1200
Open-circuit voltage (mV)
1100
1000
900
100%
800 55%
40%
700 25%
600 10%
0 50 100 150 200 250 300
Temperature (K)
Figure 27 Voc(T) for the various relative illumination intensities indicated (Ott et al.,
2013a). Reprinted with permission from Eisenbarth et al. (2010). Copyright 2010, AIP
Publishing LLC.
0.5
0
Energy (eV)
-0.5
-1
-1.5
-2
-2.5
0 0.5 1 1.5
x (µm)
Figure 28 Band diagram with back contact barrier. The back contact is located at
x ¼ 0 μm, the heterojunction at x ¼ 1.5 μm.
Reliability Issues of CIGS-Based Thin Film Solar Cells 141
junction where the associated electric field collects these injected electrons
leading to the shown part of the collector current (αIE), where α is the com-
mon base gain of the transistor. A minor part of the injected electrons
recombine with holes in the base, resulting in the base current. However,
in the considered case, this recombination current consists of
photogenerated holes. Therefore, the photocurrent of the solar cell acts as
the base current of the phototransistor. In this model, photogenerated holes
have two options. A part of these photogenerated holes recombine with
injected electrons. The remaining holes can overcome the Schottky barrier
at the back contact, adding to the collector current. Under certain assump-
tions (which will not be discussed in this context), Voc can be calculated
using the fact that the external currents have to cancel each other:
kT IPhe αIPhe
Voc ¼ ln + 1 ln +1 (5)
q IE0 IC0
with
ΦB
IC0 ¼ IC00 exp (6)
kT
and
Eg
IE0 ¼ IE00 exp (7)
kT
142 Thomas Walter
being the diode currents for the back contact and the principal diode, respec-
tively. In the high-temperature regime, photogenerated holes overcome
the Schottky barrier. Thus, the device acts there as a solar cell whose Voc
dependence can be derived as follows:
kT IPhe
Voc ¼ ln +1 (8)
q IE0
In this regime, Voc extrapolates to the bandgap energy at zero Kelvin. In
the low-temperature regime, Eq. (5) simplifies to (see Ott et al., 2013a):
kT IC0
Voc ¼ ln (9)
q αIE0
It should be noted that in this regime (which is the phototransistor
regime), Voc does not depend on the illumination intensity any longer as
required by the experimental observation shown in Fig. 26. Using Eqs. (6)
and (7), the following temperature dependence of Voc can be derived:
Eg ΦB kT αIE00
Voc ¼ ln (10)
q q IC00
Thus in the low-temperature regime, Voc extrapolates to the bandgap
energy reduced by the Schottky barrier height. Therefore, looking at the
low-temperature characteristics allows one to determine the back contact
barrier of the solar cell device. SCAPS1D simulations, which do not include
the assumptions for this analytical derivation, can be used in order to make
this phototransistor model for the low-temperature behavior more plausible.
In Fig. 30, the IV characteristics for different illumination intensities in
two temperature regimes have been simulated. It is quite obvious from this
figure that lowering the temperature coincides with a transition from a solar
cell to a phototransistor regime as seen in experimental data. Furthermore,
Fig. 31 shows that the experimentally observed Voc over temperature char-
acteristics can also be explained by the existence of such a back contact bar-
rier. It should be noted that the back barrier height ΦB inserted as a
simulation parameter also came out of the simulations in Fig. 31 as the dif-
ference between the extrapolated value from the high-temperature part
(solar cell regime) and the low-temperature part (phototransistor regime).
At this point, it has to be mentioned that the interpretation of character-
istics of CIGS-based devices especially at low temperatures with significant
deviations from a “normal” solar cell behavior has been the subject of
Reliability Issues of CIGS-Based Thin Film Solar Cells 143
50
1000 W/m2 300 K
500 W/m2 300 K
Current density (mA/cm2) Dark 300 K
1000 W/m2 150 K
500 W/m2 150 K
Dark 150 K
0
-50
0 0.2 0.4 0.6 0.8 1
Voltage (V)
Figure 30 Simulated IV characteristics at high and low temperatures. Reprinted from Ott
et al. (2014). Copyright 2014, with permission from Elsevier.
1.4
1.2
fB
1
0.8
Voc (V)
0.6
1000 W/m2
0.4
500 W/m2
barrier, in order to explain the relatively complex change of the device char-
acteristics from the high-temperature to the low-temperature regime, quite
a number of arguments can be found in the recent literature supporting the
importance of the back contact for the performance of CIGS-based solar
cells and modules. Summarizing these arguments, the following features
of a CIGS-based solar cell are indicative for a back contact barrier:
• A blocking of the dark forward current when lowering the temperature.
• The occurrence of a crossover between the dark and illuminated IV
characteristics.
• Voc(T) exhibits two regimes with different extrapolated values at 0 K.
• A transition from a solar cell to a phototransistor regime with Voc being
independent of the illumination intensity in the phototransistor regime.
As shown in Fig. 31, such a back barrier only affects the device performance
below about 200 K. However, the transition temperature from the solar cell
to the phototransistor regime depends on the barrier height. For a barrier
height exceeding about 350 meV, an impact on Voc can be seen already
at room temperature. The question which will be discussed next concerns
the stability of the back contact with respect to the long-term behavior.
Figure 32 (Ott et al., 2013b) shows the parameter drift of Voc for a device
which was held at a forward bias of +400 mV during the endurance testing at
two temperatures (105 and 145 °C) for more than 300 h. This forward bias
held the device close to the MPP emulating therefore “field” conditions
during daytime as discussed in the chapter about metastabilities associated
with CIGS. Looking at Fig. 32, an initial increase of Voc can be observed
0.78
+400 mV @ 105 °C
0.64
t100 t100
0.62 145 °C
105 °C
0.60
0 3000 6000 9000
Time (h)
Figure 32 Voc parameter drifts under an applied positive bias. Copyright IEEE 2013.
Reprinted, with permission, from Ott et al. (2013b).
Reliability Issues of CIGS-Based Thin Film Solar Cells 145
which is due to the commonly observed light (or current) soaking. How-
ever, especially as a result of the endurance test at the higher temperature,
Voc drops again reaching its initial value after about 1700 h at 145 °C. At
105 °C, this initial value is not reached within the total endurance test time
of 3600 h. An extrapolation indicates that the initial state would be reached
after more than 8000 h. From these decays of Voc at the two temperatures,
an activation energy could be extracted according to Eq. (11):
EA
t100 ¼ t0 *ekT (11)
with t100 being the time before Voc reaches its initial value after the initial
increase due to the light soaking. Knowing EA and t0 from measured data,
t100 can be extrapolated to field conditions. An analysis of this data yielded a
lifetime of more than 30 years for this parameter indicating that this degra-
dation mechanism seems not to be relevant for field conditions. However, in
the following this parameter drift will be analyzed in more detail and related
to the existence of a back contact barrier.
Figure 33 shows the development of the illuminated IV characteristics
from the initial state to the final state after an endurance test at 145 °C for
about 3600 h. With increasing time, a kind of blocking behavior of the for-
ward current develops. Regarding Voc, the initial increase as mentioned and
discussed is clearly seen from Fig. 33 as well as the decrease with increasing
time. At the final stage, Voc falls below the initial value as also shown in Fig. 33.
A more detailed analysis of the degraded device (not shown in this context)
0.06 Initial
state
20 h
0.04
470 h
I (A)
0.02
870 h
1700 h
0.00
3600 h
−0.02
0.0 0.2 0.4 0.6 0.8 1.0
Voltage (V)
Figure 33 Development of illuminated IV characteristics. Copyright IEEE 2013. Reprinted,
with permission, from Ott et al. (2013b).
146 Thomas Walter
Figure 34 Voc(T) prior and after stress test. Copyright IEEE 2013. Reprinted, with permis-
sion, from Ott et al. (2013b).
Figure 35 Ultrathin CIGS-based solar cell with point contacts. Copyright IEEE 2014.
Reprinted, with permission, from Vermang et al. (2014).
30
Unpassivated reference cell
20 J0 = 8.8 × 10–8 mA/cm2; n = 1.6;
Rs = 0.65 Ωcm2; Gsh = 2.6 mS/cm2
10
J (mA/cm2)
−20
−30
0 100 200 300 400 500 600 700 800
V (mV)
Figure 36 Illuminated IV characteristics showing the impact of back contact passiv-
ation. Copyright IEEE 2014. Reprinted, with permission, from Vermang et al. (2014).
absorber thicknesses below 500 nm were used which is about one quarter of
the usual thickness. Figure 36 compares the passivated with the unpassivated
back contact regarding the device performance. From that diagram, it is
quite obvious that a passivated back contact both improves Voc and jsc
due to reduced recombination. The impact of such a back contact passiv-
ation on the Schottky barrier has not been investigated so far. This very
interesting work is still in a research state. Possible implementations in a
production line have not been considered so far.
REFERENCES
Bart Vermang, V.F., Gao, X., Edoff, M., 2014. Improved rear surface passivation of Cu(In,
Ga)Se2 solar cells: a combination of an Al2O3 rear surface passivation layer and nanosized
local rear point contacts. IEEE J. Photovolt. 4 (1), 486–492.
Bl€
osch, P., Nishiwaki, S., Chirila, A., Kranz, L., Fella, C., Pianezzi, F., Adelhelm, C.,
Franzke, E., Buecheler, S., Tiwari, A., 2013. Sodium-doped molybdenum back contacts
for flexible Cu(In, Ga)Se2 solar cells. Thin Solid Films 535, 214–219.
Boone, T., Soltz, D., Kinkhabwala, A., Zapalac, G., Pinarbasi, M., 2012. Optoelectronic
methods for characterizing uniformity in CIGS thin film solar cells. In: Proc. 38th IEEE
PVSC, Seattle, pp. 1732–1735.
Borowski, P., Karg, F., 2014. PID-free CIGS thin film modules. In: Proc. 29th EU PVSEC,
Amsterdam, pp. 2356–2362.
Chirila, A., Reinhard, P., 2013. Potassium-induced surface modification of Cu(In, Ga)Se2
thin films for high-efficiency solar cells. Nat. Mater. 12, 1107–1111.
Dongaonkar, S., Sheets, E., Agrawal, R., Alam, M., 2012. Reverse stress metastability of
shunt current in CIGS solar cells. In: Proc. 38th IEEE PVSC, pp. 868–872.
Dongaonkar, S., Deline, C., Alam, M., 2013. Performance and reliability implications of
two-dimensional shading in monolithic thin-film photovoltaic modules. IEEE J. Photo-
volt. 3 (4), 1367–1375.
Eisenbarth, T., Caballero, R., Nichterwitz, M., Kaufmann, C., Schock, H.W., Unold, T.,
2010. Interpretation of admittance, capacitance-voltage, and current–voltage signatures
in Cu(In,Ga)Se2 thin film solar cells. J. Appl. Phys. 107, 034509.
Fjallstrom, V., Salome, P., Hultqvist, A., Edoff, M., Jarmar, T., Aitken, B., Zhang, K.,
Fuller, K., Williams, C., 2013. Potential-induced degradation of CuIn1-xGaxSe2 thin
film solar cells. IEEE J. Photovolt. 3 (3), 1090–1094.
Hoffmann, S., Koehl, M., 2012. Modeling the leakage current for potential induced degra-
dation. In: Proc. 28th EUPVSEC, Frankfurt, pp. 3336–3339.
Igalson, M., Urbaniak, A., Macielak, K., Tomassini, M., Barreau, N., Spiering, S., 2011. Bar-
riers for current transport in CIGS structures. In: Proc. 37th IEEE PVSC, Seattle,
pp. 2727–2731.
Igalson, M., Urbaniak, A., Zabierowski, P., Maksoud, H., Buffiere, M., Barreau, N.,
Spiering, S., 2013. Red-blue effect in Cu(In, Ga)Se2-based devices revisited. Thin Solid
Films 535, 302–306.
Janssen, G., Sloof, L., Bende, E., 2012. 2D-finite element model of a CIGS module. In: Proc.
27th EU PVSEC, Frankfurt, pp. 2368–2371.
Kempa, H., Rissom, T., Hlawatsch, U., Gaudig, M., Obereigner, F., Kaufmann, C.,
Scheer, R., 2013. Metastability of solar cells based on evaporated chalcopyrite absorber
layers prepared with varying selenium flux. Thin Solid Films 535, 340–342.
K€otschau, I.M., Turcu, M., Rau, U., Schock, H.W., 2001. Structural and electronic
properties of polycrystalline Cu(In,Ga)(S,Se)2 alloys. In: Mat. Res. Soc. Symp.
Proc. 668.
Reliability Issues of CIGS-Based Thin Film Solar Cells 149
Lany, S., Zunger, A., 2006. Light- and bias induced metastabilities in Cu(In, Ga)Se2 based
solar cells caused by the (VSe-VCu) vacancy complex. J. Appl. Phys. 100, 113725.
Lany, S., Zunger, A., 2008. Intrinsic DX centers in ternary chalcopyrite semiconductors.
Phys. Rev. Lett. 100, 016401.
Lechner, P., Hummel, S., Geyer, D., Mohring, H.-D., 2013. PID-behaviour of thin-film-
and c-Si PV-modules. In: Proc. 27th EU PVSEC, Paris, pp. 2810–2815.
Lee, S., Koo, J., Kim, S., Cheon, T., Oh, J., Kim, S., Kim, W., 2013. Characteristics of
MoSe2 formation during rapid thermal processing of Mo-coated glass. Thin Solid Films
535, 206–213.
Mack, P., Walter, T., Kniese, R., Hariskos, D., Schäffler, R., 2008. Reverse bias and reverse
currents in CIGS thin film solar cells and modules. In: Proc. 23rd EU PVSEC, Valencia.
Mack, P., Walter, T., Hariskos, D., Schäffler, R., Dimmler, B., 2009. Endurance testing and
accelerated ageing of CIGS thin film solar cells. In: Proc. 24th EU PVSEC, Hamburg,
pp. 2439–2442.
Mack, P., Ott, T., Schwäble, F., Runai, F., Walter, T., 2010. 2D-network simulation and
modelling of CIGS modules. In: Proc. 25th EU PVSEC, Valencia.
Moore, J., Dogaonkar, S., Vamsi, R., Chavali, K., Alam, M., Lundstrom, M., 2014. Cor-
relation of built-in potential and I–V crossover in thin-film solar cells. IEEE J. Photovolt.
4 (4), 1138–1148.
Noufi, R., Zweibel, K., 2006. High-efficiency CdTe and CIGS thin-film solar cells: high-
lights and challenges. In: Proc. 4th World Conf. Photovolt. Energy Convers. vol. 1,
pp. 317–320.
Ott, T., Mack, P., Trudel, F., Schulz, Y., Walter, T., 2011. Accelerated ageing
and luminescence of CIGS solar cells. In: Proc. 26th EU PVSEC, Hamburg,
pp. 2421–2424.
Ott, T., Runai, F., Schwäble, F., Walter, T., 2012. 2D network simulation and luminescence
characterization of CIGS thin film modules. Prog. Photovolt. Res. Appl. 20 (5),
600–605.
Ott, T., Walter, T., Unold, T., 2013a. Phototransistor effects in Cu(In, Ga)Se2 solar cells.
Thin Solid Films 535, 275–278.
Ott, T., Walter, T., Hariskos, D., Kiowski, O., Schäffler, R., 2013b. Accelerated aging and
contact degradation of CIGS solar cells. IEEE J. Photovolt. 3 (1), 514–519.
Ott, T., Sch€ onberger, F., Walter, T., Hariskos, D., Kiowski, O., Salomon, O., Schäffler, R.,
2014. Verification of phototransistor model for Cu(In,Ga)Se2 solar cells. Thin Solid
Films 582, 392–396.
Pingel, S., Frank, O., Winkler, M., Daryan, S., Geipel, T., Hoehne, H., Berghold, J., 2010.
Potential induced degradation of solar cells and panels. In: Proc. 35th IEEE PVSC,
pp. 2817–2822.
Runai, F., Schwäble, F., Walter, T., Fidler, A., Gorse, S., Hahn, T., K€ otschau, I., 2011.
Imaging and performance of CIGS thin film modules. In: Proc. 37th IEEE PVSC,
Seattle, pp. 3399–3403.
Salome, P., Fjällstr€
om, V., Hultqvist, A., Edoff, M., 2014. Incorporation of Na in Cu(In, Ga)
Se2 thin-film solar cells: a statistical comparison between Na from soda-lime glass and
from a precursor layer of NaF. IEEE J. Photovolt. 4 (6), 1659–1664.
Sites, J., 2010. Thin-film photovoltaics: what are the reliability issues and where do they
occur? In: Proc. Reliab. Phys. Symp., Anaheim, pp. 494–498.
Szaniawski, P., Lindahl, J., T€
orndahl, T., Zimmermann, U., Edoff, M., 2013. Light-enhanced
reverse breakdown in Cu(In, Ga)Se2 solar cells. Thin Solid Films 535, 326–330.
Sze, S.M., 1981. Physics of Semiconductor Devices. Wiley & Sons Inc., New York,
Chichester, Brisbane, Toronto, Singapore.
Tran, T.M.H., Pieters, B.E., Siegloch, M., Gerber, A., Ulbrich, C., Kirchartz, T.,
Schäffler, R., Rau, U., 2011. Characterization of shunts in Cu(In, Ga)Se2 solar modules
150 Thomas Walter
via combined electroluminescence and dark lock-in thermography analysis. In: Proc.
26th EU PVSEC, Hamburg, pp. 2981–2985.
Vermang, B., Wätjen, J.T., Frisk, C., Fjällstr€
om, V., Rostvall, F., Edoff, M., Salome, P.,
Borme, J., Nicoara, N., Sadewasser, S., 2014. Introduction of Si PERC rear contacting
design to boost efficiency of Cu(In, Ga)Se2 solar cells. IEEE J. Photovolt. 4 (6),
1644–1649.
Wada, T., Kohara, N., Nishiwaki, N., Negami, T., 2001. Characterization of the Cu(In, Ga)
Se2/Mo interface in CIGS solar cells. Thin Solid Films 387, 118–122.
Wendlandt, S., Giese, A., Drobisch, A., Tornow, D., Hanusch, M., Berghold, J., Krauter, S.,
Grunow, P., 2012. The temperature as the real hot spot risk factor at PV-modules.
In: Proc. 27th EU PVSEC, Frankfurt, pp. 3553–3557.
Westin, P.-O., Zimmermann, U., Stolt, L., Edoff, M., 2009. Reverse bias damage in CIGS
modules. In: Proc. 24th EU PVSEC, Hamburg, pp. 2967–2970.
INDEX
151
152 Index
155
156 Contents of Volumes in this Series
Part B
T. Misawa, IMPATT Diodes
H. C. Okean, Tunnel Diodes
R. B. Campbell and Hung-Chi Chang, Silicon Junction Carbide Devices
R. E. Enstrom, H. Kressel, and L. Krassner, High-Temperature Power Rectifiers of GaAs1 xPx
A. Leitoila and J. F. Gibbons, Applications of CW Beam Processing to Ion Implanted Crystalline Silicon
N. M. Johnson, Electronic Defects in CW Transient Thermal Processed Silicon
K. F. Lee, T. J. Stultz, and J. F. Gibbons, Beam Recrystallized Polycrystalline Silicon: Properties,
Applications, and Techniques
T. Shibata, A. Wakita, T. W. Sigmon and J. F. Gibbons, Metal-Silicon Reactions and Silicide
Y. I. Nissim and J. F. Gibbons, CW Beam Processing of Gallium Arsenide
Part B
J. I. Pankove, Introduction
G. D. Cody, The Optical Absorption Edge of a-Si: H
N. M. Amer and W. B. Jackson, Optical Properties of Defect States in a-Si: H
P. J. Zanzucchi, The Vibrational Spectra of a-Si: H
Y. Hamakawa, Electroreflectance and Electroabsorption
J. S. Lannin, Raman Scattering of Amorphous Si, Ge, and Their Alloys
R. A. Street, Luminescence in a-Si: H
R. S. Crandall, Photoconductivity
J. Tauc, Time-Resolved Spectroscopy of Electronic Relaxation Processes
P. E. Vanier, IR-Induced Quenching and Enhancement of Photoconductivity and Photoluminescence
H. Schade, Irradiation-Induced Metastable Effects
L. Ley, Photoelectron Emission Studies
Part C
J. I. Pankove, Introduction
J. D. Cohen, Density of States from Junction Measurements in Hydrogenated Amorphous Silicon
P. C. Taylor, Magnetic Resonance Measurements in a-Si: H
K. Morigaki, Optically Detected Magnetic Resonance
J. Dresner, Carrier Mobility in a-Si: H
T. Tiedje, Information About Band-Tail States from Time-of-Flight Experiments
A. R. Moore, Diffusion Length in Undoped a-S: H
W. Beyer and J. Overhof, Doping Effects in a-Si: H
H. Fritzche, Electronic Properties of Surfaces in a-Si: H
C. R. Wronski, The Staebler-Wronski Effect
R. J. Nemanich, Schottky Barriers on a-Si: H
B. Abeles and T. Tiedje, Amorphous Semiconductor Superlattices
Part D
J. I. Pankove, Introduction
D. E. Carlson, Solar Cells
G. A. Swartz, Closed-Form Solution of I–V Characteristic for a s-Si: H Solar Cells
I. Shimizu, Electrophotography
S. Ishioka, Image Pickup Tubes
P. G. Lecomber and W. E. Spear, The Development of the a-Si: H Field-Effect Transistor and its Possible
Applications
Contents of Volumes in this Series 161
Part B
J. P. van der Ziel, Mode Locking of Semiconductor Lasers
K. Y. Lau and A. Yariv, High-Frequency Current Modulation of Semiconductor Injection Lasers
C. H. Henry, Special Properties of Semi Conductor Lasers
Y. Suematsu, K. Kishino, S. Arai, and F. Koyama, Dynamic Single-Mode Semiconductor Lasers with a
Distributed Reflector
W. T. Tsang, The Cleaved-Coupled-Cavity (C3) Laser
Part C
R. J. Nelson and N. K. Dutta, Review of InGaAsP InP Laser Structures and Comparison of Their
Performance
N. Chinone and M. Nakamura, Mode-Stabilized Semiconductor Lasers for 0.7–0.8- and 1.1–1.6-μm
Regions
Y. Horikoshi, Semiconductor Lasers with Wavelengths Exceeding 2 μm
B. A. Dean and M. Dixon, The Functional Reliability of Semiconductor Lasers as Optical Transmitters
R. H. Saul, T. P. Lee, and C. A. Burus, Light-Emitting Device Design
C. L. Zipfel, Light-Emitting Diode-Reliability
T. P. Lee and T. Li, LED-Based Multimode Lightwave Systems
K. Ogawa, Semiconductor Noise-Mode Partition Noise
Part D
F. Capasso, The Physics of Avalanche Photodiodes
T. P. Pearsall and M. A. Pollack, Compound Semiconductor Photodiodes
162 Contents of Volumes in this Series
Part E
S. Wang, Principles and Characteristics of Integrable Active and Passive Optical Devices
S. Margalit and A. Yariv, Integrated Electronic and Photonic Devices
T. Mukai, A. Yamamoto, and T. Kimura, Optical Amplification by Semiconductor Lasers
W. M. Becker, Band Structure and Optical Properties of Wide-Gap AII1 x Mnx BIV Alloys at Zero Magnetic
Field
S. Oseroff and P. H. Keesom, Magnetic Properties: Macroscopic Studies
T. Giebultowicz and T. M. Holden, Neutron Scattering Studies of the Magnetic Structure and Dynamics of
Diluted Magnetic Semiconductors
J. Kossut, Band Structure and Quantum Transport Phenomena in Narrow-Gap Diluted Magnetic
Semiconductors
C. Riquaux, Magnetooptical Properties of Large-Gap Diluted Magnetic Semiconductors
J. A. Gaj, Magnetooptical Properties of Large-Gap Diluted Magnetic Semiconductors
J. Mycielski, Shallow Acceptors in Diluted Magnetic Semiconductors: Splitting, Boil-off, Giant Negative
Magnetoresistance
A. K. Ramadas and R. Rodriquez, Raman Scattering in Diluted Magnetic Semiconductors
P. A. Wolff, Theory of Bound Magnetic Polarons in Semimagnetic Semiconductors
T. G. Brown and D. G. Hall, Radiative Isoelectronic Impurities in Silicon and Silicon-Germanium Alloys
and Superlattices
J. Michel, L. V. C. Assali, M. T. Morse, and L. C. Kimerling, Erbium in Silicon
Y. Kanemitsu, Silicon and Germanium Nanoparticles
P. M. Fauchet, Porous Silicon: Photoluminescence and Electroluminescent Devices
C. Delerue, G. Allan, and M. Lannoo, Theory of Radiative and Nonradiative Processes in Silicon
Nanocrystallites
L. Brus, Silicon Polymers and Nanocrystals
F. H. Pollak, Effects of External Uniaxial Stress on the Optical Properties of Semiconductors and
Semiconductor Microstructures
A. R. Adams, M. Silver, and J. Allam, Semiconductor Optoelectronic Devices
S. Porowski and I. Grzegory, The Application of High Nitrogen Pressure in the Physics and Technology of
III–N Compounds
M. Yousuf, Diamond Anvil Cells in High Pressure Studies of Semiconductors
Volume 64 Electroluminescence I
M. G. Craford, S. A. Stockman, M. J. Peansky, and F. A. Kish, Visible Light-Emitting Diodes
H. Chui, N. F. Gardner, P. N. Grillot, J. W. Huang, M. R. Krames, and S. A. Maranowski, High-Efficiency
AIGaInP Light-Emitting Diodes
R. S. Kern, W. Gōtz, C. H. Chen, H. Liu, R. M. Fletcher, and C. P. Kuo, High-Brightness Nitride-Based
Visible-Light-Emitting Diodes
Yoshiharu Sato, Organic LED System Considerations
V. Bulovic´, P. E. Burrows, and S. R. Forrest, Molecular Organic Light-Emitting Devices
Volume 65 Electroluminescence II
V. Bulovic´ and S. R. Forrest, Polymeric and Molecular Organic Light Emitting Devices: A Comparison
Regina Mueller-Mach and Gerd O. Mueller, Thin Film Electroluminescence
Markku Leskelā, Wei-Min Li, and Mikko Ritala, Materials in Thin Film Electroluminescent Devices
Kristiaan Neyts, Microcavities for Electroluminescent Devices
Christoph Lienau and Thomas Elsaesser, Spatially and Temporally Resolved Near-Field Scanning Optical
Microscopy Studies of Semiconductor Quantum Wires
K. T. Tsen, Ultrafast Dynamics in Wide Bandgap Wurtzite GaN
J. Paul Callan, Albert M.-T. Kim, Christopher A. D. Roeser, and Eriz Mazur, Ultrafast Dynamics and Phase
Changes in Highly Excited GaAs
Hartmut Hang, Quantum Kinetics for Femtosecond Spectroscopy in Semiconductors
T. Meier and S. W. Koch, Coulomb Correlation Signatures in the Excitonic Optical Nonlinearities of
Semiconductors
Roland E. Allen, Traian Dumitrică, and Ben Torralva, Electronic and Structural Response of Materials to
Fast, Intense Laser Pulses
E. Gornik and R. Kersting, Coherent THz Emission in Semiconductors
Franz Freibert, Timothy W. Darling, Albert Miglori, and Stuart A. Trugman, Thermomagnetic Effects and
Measurements
M. Bartkowiak and G. D. Mahan, Heat and Electricity Transport Through Interfaces
J. Cibert, L. Besombes, D. Ferrand, and H. Mariette, Quantum Structures of II–VI Diluted Magnetic
Semiconductors
Agnieszka Wolos and Maria Kaminska, Magnetic Impurities in Wide Band-gap III–V Semiconductors
Tomasz Dietl, Exchange Interactions and Nanoscale Phase Separations in Magnetically Doped
Semiconductors
Hiroshi Katayama-Yoshida, Kazunori Sato, Tetsuya Fukushima, Masayuki Toyoda, Hidetoshi Kizaki, and An
van Dinh, Computational Nano-Materials Design for the Wide Band-Gap and High-TC
Semiconductor Spintronics
Masaaki Tanaka, Masafumi Yokoyama, Pham Nam Hai, and Shinobu Ohya, Properties and Functionalities of
MnAs/III–V Hybrid and Composite Structures