Advanced Ceramic Materials Series Volume 13
edited by Aleksander Gurlo
David Karl | Segerkegel | 2017
ISSN (online) 2569-8338
David Uebel
Silicon grown from silicon-rich tin solution - from crystallization
of the seed layer to prototype solar cells
Sili
co
n gr
o
wn fr
om
sili
co
n
-
ri
c
h tin s
o
luti
o
n
- from crystallization of the seed layer
to prototype solar cells
vo
r
ge
l
eg
t
von
M.
Sc.
Dav
i
d
Uebe
l
ORC
I
D
:
0000-0003-4066-5691
an
de
r
Faku
lt
ä
t III
-
P
r
ozessw
i
ssenschaf
t
en
de
r
Techn
i
schen
Un
i
ve
r
s
it
ä
t
Be
rli
n
zu
r
E
rl
angung
des
akadem
i
schen
G
r
ades
Dok
t
o
r
de
r I
ngen
i
eu
r
w
i
ssenschaf
t
en
-
D
r
.-
I
ng
-
genehm
i
g
t
e
D
i
sse
rt
a
ti
on
P
r
omo
ti
onsausschuss
:
Vo
r
s
it
zende
:
P
r
of.
D
r
.
C
l
aud
i
a
F
l
eck
Gu
t
ach
t
e
r:
P
r
of.
D
r
.
A
l
eksande
r
Gu
rl
o
Gu
t
ach
t
e
ri
n
:
P
r
of.
D
r
.
As
tri
d
Ha
i
be
l
Gu
t
ach
t
e
r:
P
r
of.
D
r
.
Ma
tt
h
i
as
B
i
cke
r
mann
Tag
de
r
w
i
ssenschaf
tli
chen
Aussp
r
ache
:
20.
J
u
li
2022
Be
rli
n
2023
Zu
s
ammenfa
ss
ung
Im Rahmen dieser Arbeit werden zentrale Fortschritte bei der Abscheidungsmethode der Steady State
Flüssigphasenepitaxie (SSLPE) vorgestellt. SSLPE ermöglicht die kontinuierliche Abscheidung von
dünnen kristallinen Siliziumschichten aus einer Zinn-Lösung (Sn), die mit Silizium (Si) übersättigt ist. Als
Variante der Flüssigphasenepitaxie (LPE) wird eine hohe kristalline Qualität erreicht, mit dem
herausragenden Vorteil, dass das Wachstum auf Saatschichten aus amorphen Silizium möglich ist, die
auf Glas abgeschieden wurden. Mittels SSLPE können auch große Flächen von mehreren
Quadratmetern beschichtet werden. Bei der Verwendung als Absorber in Solarzellen könnte dies zu
erheblichen Einsparungen bei den Rohstoffen und beim Energieverbrauch während der Produktion
führen. Bei der Verarbeitung ist das Verhalten des amorphen Siliziums von großer Bedeutung, wobei
die selbstpassivierende Eigenschaft des amorphen Siliziums durch Bildung einer stabilen
Siliziumoxidschicht eines der Hauptprobleme des Wachstumsprozesses darstellt. Diese Arbeit
vergleicht verschiedene Maßnahmen zur Vermeidung des Siliziumoxids in Theorie und Experiment und
bietet erstmals eine umfassend anwendbare Lösung. Eine eingehende Untersuchung der Saatschicht
während des Wachstums zeigt, dass das amorphe Silizium gleichzeitig mit dem eigentlichen SSLPE-
Schichtwachstum kristallisiert. Folglich interagieren das kristallisierende Keimsubstrat und die
wachsende SSLPE-Schicht während des Wachstums und bestimmen so die Morphologie der
endgültigen Schicht. Darüber hinaus wurde das Wachstum auf monokristallinen Substraten
durchgeführt, um elektronische Messungen an den resultierenden monokristallinen SSLPE-Schichten
vorzunehmen. Die Analyse der Daten aus der Deep-Level-Transient-Spektroskopie (DLTS) zeigte, dass
die Deep-Level-Defekte von Kristallfehlern im SSLPE-Silizium herrühren. Diese Defekte entsprechen
Gitterfehlern, die seltene Kombinationen verschiedener Ladungszustände aufweisen, die sonst nur
nach einer Wärmebehandlung zu beobachten sind. Mit diesen neuen Ergebnissen konnten zum ersten
Mal funktionierende Solarzellen aus SSLPE-Material hergestellt werden.
Ab
st
rac
t
Within this work, central advances in the deposition method of steady state liquid phase epitaxy
(SSLPE) are presented. SSLPE allows for the continuous deposition of thin crystalline silicon (Si) films
from a tin (Sn) solution, which is supersaturated with Si. As a variant of liquid phase epitaxy (LPE) high
crystalline quality is reached, with the outstanding advantage that growth on seed layers of amorphous
silicon deposited on glass is possible. SSLPE layers can coat even large areas of square meters. When
used as absorbers in solar cells, this could result in considerable savings in raw materials and energy
consumption during production. During processing, the behavior of the amorphous silicon is of great
importance, with the self-passivating property of the amorphous silicon by forming a stable silicon
oxide layer posed one of the main problems of the growth process. This work compares various
measures to avoid the silicon oxide in theory and experiments and provides a comprehensively
applicable solution for the first time. An in depth investigation of the seed during growth reveals that
the amorphous silicon crystallizes simultaneously with the actual SSLPE layer growth. Consequentially
the crystallizing seed substrate and the growing SSLPE layer interact during growth, thus determining
the morphology of the final layer. Furthermore, growth on monocrystalline substrates was realized to
perform electronic measurements on the resulting monocrystalline SSLPE layers. Deep level transient
spectroscopy data analysis (DLTS) demonstrated that deep level defects originate from crystal
imperfections in SSLPE silicon. These defects correspond to lattice defects that exhibit rare
combinations of different charge states otherwise observed only after heat treatment. With these
novel results, functioning solar cells could be fabricated from SSLPE material for the first time.
Tab
l
e
of
con
t
en
ts
1
. Obj
ective
s
,
c
on
te
n
t
and s
tr
u
ct
u
re
of
t
h
e
w
o
rk
...............................................................................
5
2
. S
t
a
te
of
t
h
e
a
rt
................................................................................................................................
9
2
.
1
. L
i
qu
i
d
P
has
e
Ep
it
ax
y
(L
P
E).......................................................................................................
9
2
.
1
. S
te
ad
y
s
t
a
te
li
qu
i
d
e
p
it
ax
y
(SSL
P
E)........................................................................................
12
2
.
2
. Ox
i
d
e
c
on
tr
o
l
.........................................................................................................................
14
2
.
2
.
1
. Ox
i
d
e
c
on
tr
o
l
b
y
h
y
d
r
of
l
uo
ric
a
ci
d
etc
h
i
n
g
.......................................................................
15
2
.
2
.
2
. Ox
i
d
e
c
on
tr
o
l
b
y
l
as
er
tre
a
t
m
e
n
t
......................................................................................
17
2
.
3
. C
ry
s
t
a
lli
za
ti
on of amo
r
phous s
ilic
on .....................................................................................
25
2
.
4
.
I
n
tr
odu
cti
on E
lectric
a
l
p
r
op
ertie
s .........................................................................................
29
3
. Ma
teri
a
l
s and m
et
hods .................................................................................................................
33
3
.
1
.
I
nh
erite
d p
r
o
ce
ss and
ce
n
tr
a
l
p
r
o
ce
ss
i
mp
r
o
ve
m
e
n
t
s..........................................................
33
3
.
2
. Las
er
s
et
up ............................................................................................................................
38
3
.
3
. Subs
tr
a
te
s..............................................................................................................................
42
4
. Cha
r
a
cteri
za
ti
on m
et
hods.............................................................................................................
47
4
.
1
. S
c
ann
i
n
g
E
lectr
on M
icr
os
c
op
y
(SEM)....................................................................................
47
4
.
2
. T
r
ansm
i
ss
i
on E
lectr
on M
icr
os
c
op
y
(TEM).............................................................................
52
4
.
3
.
R
aman sp
ectr
os
c
op
y
.............................................................................................................
54
4
.
4
. E
lectric
a
l
p
r
op
ertie
s ..............................................................................................................
60
5
.
Re
su
lt
s and d
i
s
c
uss
i
on...................................................................................................................
63
5
.
1
.
I
n
ter
a
cti
on of s
ee
d
l
a
yer
and
li
qu
i
d
ti
n .................................................................................
63
5
.
2
. D
evel
opm
e
n
t
of
t
h
e
su
r
fa
ce
mo
r
pho
l
o
gy
du
ri
n
g
gr
o
wt
h .....................................................
67
5
.
3
.
I
nf
l
u
e
n
ce
of h
y
d
r
of
l
uo
ric
a
ci
d and
regr
o
w
n ox
i
d
e
................................................................
71
5
.
4
. C
ry
s
t
a
lli
n
e
qua
lity
..................................................................................................................
75
5
.
5
. S
ee
d
l
a
yer
ev
o
l
u
ti
on du
ri
n
g
gr
o
wt
h......................................................................................
78
5
.
6
. Las
er
tre
a
t
m
e
n
t
.....................................................................................................................
82
5
.
6
.
1
. Exp
eri
m
e
n
t
a
l
re
su
lt
s of
t
h
e
l
as
er
tre
a
t
m
e
n
t
.....................................................................
82
5
.
6
.
2
. S
i
mu
l
a
ti
on of
t
h
e
i
n
ter
a
cti
on b
etwee
n
l
as
er
and su
r
fa
ce
.................................................
85
5
.
7
. E
lectric
a
l
p
r
op
ertie
s ..............................................................................................................
89
6
. Summa
ry
and ou
tl
oo
k
.................................................................................................................
100
7
.
Ack
no
wle
d
ge
m
e
n
t
s.....................................................................................................................
105
8
. L
iter
a
t
u
re
.....................................................................................................................................
106
9
. L
i
s
t
of f
ig
u
re
s ...............................................................................................................................
116
10
. L
i
s
t
of
t
ab
le
s.............................................................................................................................
122
1. Objectives, content and structure of the work
1.
Objec
ti
ve
s,
con
t
en
t
and
st
ruc
t
ure
of
t
he
work
Energy supply is often considered a key concern of industrial policy, as cheap energy enables
energy-intensive heavy industry. But especially environmentally friendly energy conversion is a basic
prerequisite for the preservation and development of our civilization and enables better opportunities
for education, health and jobs [1] an in a digitalized world, access to the internet is a crucial
prerequisite for the everyday life of the vast majority of people [2]. As an everyday example, 83.89 %
of the global population own a smartphone and therefore are in need of regular access to a power
supply [3]. As a consequence of energy supply being this fundamental, supplying all households
worldwide with electricity until 2030 is a central goal of the United Nations [4]. Solar panels as a
decentralized source of electricity can supply even remote places with electricity. Recently,
photovoltaics have become the cheapest energy source in China and India [5]. A report of the
International Energy Agency (IEA) sees photovoltaics as the dominant source of electricity in all
discussed future scenarios [6]. Energy supply has a vast and complex influence on the life of individuals
in modern societies [7], with the most prominent challenge of climate change, caused by greenhouse
gases [8] set free by the human civilization and the industrialization since the 1800s. The last years
have been the hottest on record [9] with extreme weather phenomena already being a reality. In the
future, further warming of the planet might even cause a destabilization of the entire climate system,
itself resulting in an acceleration of climate change [10]. Renewable energy sources like wind, hydro
and solar power don’t produce greenhouse gases once their infrastructure has been set up and are of
central importance for preserving a livable planet.
Solar power has a special role in the broadening of renewable energy, since plants of solar modules
can be set up on very different scales, from tiny cells powering everyday devices’access to the internet
of things, up to few modules run by private persons on rooftops and even to large fields of modules
which substitute entire conventional power plants. With power grids designed to work with changing
power loads, this leads to a decentralization of power production. The unreliable nature of weather
creates voltage peaks when using renewable energy. It was long discussed that power grids would not
sustain these peaks. Recently it has become imminent that decentralized power grids with diversified
producers of renewable energy can be more resilient to power outages than conventional power
production [11].
Silicon (Si) is the chemical element with the atomic number 14. Its properties as elemental
semiconductor with a small band gap of E = 1.107 eV gives it a broad absorption spectrum which is
used in the conversion of sunlight to electric energy. Si solar cells account for the largest share of global
- 5 -
1. Objectives, content and structure of the work
demand [12], predominantly using bulk grown material, a top-down process to produce wafers.
Elemental Si semiconductors are the base of any home- or supercomputer, smartphone and of our
digital communication. Especially within photovoltaics its flexibility and adaptability prove to be
advantageous. Modern cells use a wide range of functionalization techniques to passivate and contain
electrons-hole pairs in their respective layers, to maximize the yield of the sun light and to harvest the
induced photovoltage [13], [14]. The knowledge of its growth allows to produce wafers of 300 mm in
diameter and larger, and the know-how of Si processing allows for handling within fab lines which are
constantly evolving, allowing for faster throughput and for reduced loss of chemicals. For the industrial
production of high performance cells, large rods of monocrystalline Si (mono-Si) are grown epitaxially at
temperatures above Si’s melting point of T
melt,Si
= 1414 °C, temperatures which have to be held for
hours. Then, the rods as intermediate products are cut down by elaborate sawing techniques. The
reduction of the cost of solar cells is pursued at every step of its value chain. A successful source of
cost reduction has been the reduction of raw polycrystalline Si (poly-Si) usage for some time [12].
Material is wasted due to two major areas. A typical commercial wafer has a final thickness of 180 µm,
but another 130 µm are lost during sawing. Of this material, only 50 µm are needed for the absorption of
sunlight (F
ig
u
re
1
a) [15]. The additional material is only needed for mechanical stability, or is wasted
during processing. The prospect of further reducing the gap between the actual active thickness and
the total material is limited. Reducing the wafer thickness decreases the bending moment, an objects
geometrical resistance to bending critically. If the wafer bends during processing, dislocations are
introduced which harm electrical properties or even lead to microcracks which render the material
unusable [12], [16], [17]. For n-type absorbers with diffusion lengths of L
N
= 200 µm, optimal wafer
thickness can be found around wafer thicknesses of 50 µm [18].
- 6 -
1. Objectives, content and structure of the work
Figure 1: Motivation for SSLPE-growth of Si on glass for PV. (a) Conventional absorbers only use a
fraction of the material invested, while most of the used material is electrically inactive due to sunlight
not reaching it or lost during sawing. The remaining material is only used for mechanical stability or is
lost during sewing (b) SSLPE offers a bottom up approach to grow high-quality crystalline Si on an
amorphous substrate. Thus, glass can be used for mechanical stability, which is much cheaper than
solar-grade Si (c) SSLPE resembles the float-glass process which promises a high scaling potential.
During the float-glass process a glass melt is poured onto a bed of molten metal, on which the glass
cools and solidifies, inheriting the flatness of the metal surface. After sufficient solidification, the glass
can be cut into large sheets.
For this reason, thin-film approaches are being investigated that dispense with the top-down approach
and instead grow the absorber layers to applicable size. The mechanical stability can be provided by a
cheap and easy-to-handle substrate like float-glass. This thesis presents the approach of steady-state
liquid phase epitaxy (SSLPE, F
ig
u
re
1
b) for growing on glass. Two thirds of energy needed for the solar
absorbers can be saved by growing Si directly on glass instead of processing a Si wafer. For Liquid Phase
Crystalized (LPC) Si this is demonstrated and provides an alternative technique for the realization of
thin film crystalline Si (c-Si) on glass. [19].
This work covers the entire process of SSLPE, an approach to grow c-Si bottom-up, developed at
Leibniz-Institut für Kristallzüchtung (IKZ) [20]–[27]. Liquid phase epitaxy (LPE) in general operates close
- 7 -
1. Objectives, content and structure of the work
to the thermodynamic equilibrium and thus produces crystals of high quality [28], [29]. SSLPE uses the
advantages of LPE with the additional benefit that it can be performed as a continuous process rather
than in single batches. With many similarities to the large-scale float-glass process (F
ig
u
re
1c
), there is a
perspective to scale SSLPE to several square meters. The entire process consists of three steps
described in chapter 2 in detail: seed layer preparation, oxide control and SSLPE growth itself.
The aim of this work has been to solve the remaining challenges with SSLPE, and lead towards the
production of prototype devices. As a consequence, the following chapters are targeted to investigate
specific problems or unknown behavior of SSLPE. Each chapter draws on one method of investigation to
look at a particular phenomenon of SSLPE growth. Since the substrate and the growing layer are
immersed in a tin (Sn) bath during the process, direct observation of the growth is not possible. The
overall process of growth has multiple stages interacting with each other, and the interplay of errors at
each stage imposes significant limitations on the interpretation and comparability of experiments. The
growth of polycrystalline material from an amorphous seed layer is virtually absent in literature or
practice. This raises the question of the role of the seed layer, and since its morphology changes during
growth, this change and its dynamic influence on growth must be mutually observed. Originally,
nanocrystalline regions on the surface of the seed layer were observed and consequentially evaluated
as the nuclei and origin of the superficially visible SSLPE crystallites. However, investigations within the
scope of this work did not find such nuclei in the a-Si substrates currently in use. Nevertheless, growth
still was successful on substrates without any nanocrystalline roots. In order to gain insight into their
development, a series of experiments with interrupted growth cycles was designed. This series of
experiments shows how the overall microstructure changes from the first contact of substrate and Sn
solution (1 s) to a pronounced growth morphology (1 h) take place. The experiment’s results are
investigated by scanning electron microscopy (SEM), and the results to lay out a novel SSLPE growth
model are described in chapter
5
.
2
. Chapter
4
.
2
uses an in-situ crystallization experiment performed in
a transmission electron microscope (TEM) and determines whether the crystallization of the seed is
induced by the present metal or merely induced thermally. Chapter
5
.
5
pursues the investigation of
the seed layer by Raman mapping of the different crystallization stages of the seed and discusses the
interplay between the morphologies of seed and SSLPE layer. Thereafter, the influence of the oxide
layer is discussed, both the influence and effects of hydrofluoric acid etching (chapter
5
.
3
) and by laser
treatment of the surface (chapter
5
.
6
). Due to promising electrical properties presented before [30],
prototype solar cells were manufactured. Chapter
5
.
7
compiles the electrical properties and presents
the properties of the prototype devices. As a conclusion, a novel growth model is presented in
chapter
6
.
- 8 -
2. State of the art
2.
S
t
a
t
e
of
t
he
ar
t
2.1.L
i
qu
i
d
Phase
Ep
i
taxy
(
LPE
)
LPE techniques in general use solution differences in binary or higher order material systems to deposit
materials, mostly semiconductors. An in-depth introduction of its possibilities and advantages can be
found in [29]. An example is the deposition of Gallium-Arsenide (GaAs). Arsenic (As) is first solved in
molten Gallium (Ga). Their phase diagram (F
ig
u
re
2
a) shows the high-melting point of GaAs and
naturally two liquidus lines, falling off towards the pure elements. One way of reading liquidus lines is a
change of solubility of the solid component in the solution. LPE uses the high As solubility at high
temperatures to saturate the melt. When cooling the solution, it gets oversaturated. First, the solution
will enter the Ostwald-Miers-area, where it is supersaturated but will remain metastable without any
precipitation. When temperatures decrease further or instabilities like mechanical vibrations or
chemical dopants are introduced, crystallization of GaAs will initiate. If a monocrystalline GaAs surface is
offered, atoms from the melt will attach to the surface and extend the lattice of the substrate. This
process is known as epitaxial growth, and specifically homoepitaxial since GaAs grows on GaAs. Thus,
the temperature process has to start underneath the melting point of GaAs to allow its precipitation
and will have no further effect underneath the solidification point of Ge. Therefore the cooling process is
constrained between these two temperatures and layers have to be grown in batch-like processes. In
the case of Si deposition from Sn melt (F
ig
u
re
2
b), pure Si precipitates.
SSLPE of Si from Sn (F
ig
u
re
3
) does not use the cooling of one batch for deposition, but applies a
temperature difference between a Si source and a substrate. Like the GaAs-LPE, SSLPE does exploit the
phase diagram of the involved elements and their changing solubility over time. In the typical growth
setting, Sn at 600 °C at the substrate can solve 3.10 × 10
-4
at-% silicon. Sn at the Si source is kept at 605
°C elevating Si solubility to 3.46 × 10
-4
at-% (F
ig
u
re
6
). Due to mixing of the solution, a steady atom flow
establishes and the hotter interface is undersaturated and dissolves, while the substrate interface is
oversaturated and crystallization and growth of nuclei is observed.
- 9 -
2. State of the art
Figure 2 Ga-As (a) and Sn-Si (b) are exemplary material systems for LPE. As is solved in Ga melt, or Si in
Sn melt and the mixture (1) is cooled (2) down so that GaAs or elementary Si precipitates.
Figure 3 SSLPE of Si is a three-step process of substrate preparation by PVD (a), oxide removal by
various methods (b) and the SSLPE growth-step itself (c). Unlike LPE of GaAs it is achieved by steadily
dissolving a Si source and precipitation of Si crystallites on the substrate
- 10 -
2. State of the art
In its currently deployed form, the SSLPE process consists of three separate process steps
i. Choice of substrate or substrate preparation,
ii. oxide removal or control,
iii. SSLPE growth,
of which the first two are not only arbitrary preparation steps, but have an extensive impact on the
eventual growth process as shown in the following chapters of this thesis. Thus all three will be
discussed in detail with special regard to their interaction concerning the complete process. For
substrates, the primary choice is amorphous Si (a-Si) on glass, which produces polycrystalline Si
(poly-Si) layers. The a-Si seed crystallizes in a polycrystalline manner and offers stochastically
orientated nuclei. An alternative are monocrystalline wafers (mono-Si), which provide an extended
crystalline surface for Si atoms to adsorb and bond to, resolving in monocrystalline, homoepitaxial
layers. Poly-Si layers and the use of glass as substrate have the prospective to just be limited to the
size of used glass sheets and thus several square meters. That size would be an immense advantage
over bulk silicon, which is limited to industrial float-zone rod diameters of 30 cm and even over thin
film techniques, which have area limitations due to uneven monomere flux. SSLPE deposition on
conventional wafers is important for electrical characterization, since poly-Si layers have a rough
surface which hamper measurement, are difficult to contact and electric currents flow uneven
depending on the varying thickness. Additionally, research into Epifoils® as substrates, ultra-thin Si
foils, has the potential for industrial application as already shown elsewhere, with no clear winner
concerning the processing yet [17]. Epifoils® minimize the amount of expensive monocrystalline
material when they are used as substrates instead of standard wafers. Applying them in SSLPE growth
could make homoepitaxial growth economically viable. In this work, they will be referred to as
reorganized porous silicon (RPS)
Si easily reacts with oxygen, a property often utilized for a self-passivating silicon-oxide (SiO
x
) layer.
Materials which are not self-passivating but form instable oxide layers corroding chemically easily
when not actively sealed with a diffusion barrier or held in inert atmosphere. When the diffusion
barrier is scratched or otherwise impaired, oxidation will start at the scratch and the reaction zone
growths from there. A self-passivating oxide will regrow instantly and prevent the extension of the
reaction zone. Unfortunately for SSLPE and other deposition techniques, SiO
x
hinders true contact
between precursor and Si and has to be removed ahead of any deposition. The contact is necessary
for incoming Si atoms to adsorb to the Si crystal and bond to it. The SSLPE cluster tool (described in
detail later) is spacious and sample handling takes time. Since the oxide regrows quickly and remains
- 11 -
2. State of the art
stable then, its control was a focal point of this research and will be discussed separately. Several
methods as fast handling, melt-back, laser treatment and hydrofluoric acid (HF) etching have been
applied and compared for this work. SSLPE growth itself is dependent on many technical factors which
had to be accounted for leading to a formerly proclaimed growth model being updated for some
growth phenomena which did not account for new results.
2.1.Steady
state
li
qu
i
d
ep
i
taxy
(
SSLPE
)
The central and defining step of the three-step process is the growth itself. It is achieved in a crucible
of graphite in a heat isolated inner chamber in hydrogen-rich atmosphere. The inner chamber is
surrounded by an outer chamber for further temperature isolation, for storage of samples and for gas
isolation.
The sample is inserted and stored in a handler arm in vacuum. When the chamber is closed it is flooded
with H
2
and heating will commence when overpressure is achieved. Heating is done with a linear
heating program and controlled by two Euroterm 2704 advanced multiloop temperature controllers.
They control three resistive graphite heaters located above, around and underneath the crucible. The
last two heat the crucible and thus the Sn melt indirectly. The first one is installed on top of the inner
chamber, and heats the melt directly via radiation. This is remarkable for the process - when a handler
arm passes through, it will shield the melt from the radiation and temperature drops immediately.
Simultaneously a sample being transported in will heat up quickly. This also indicates that a substrate
should not be stored within the inner chamber since it would be annealed.
The crucible contains a 4 mm thick Si wafer constrained on its bottom, and is filled with 1 cm of tin.
During the process, the temperature control ensures a vertical temperature difference in the range of
5 K in the melt (F
ig
u
re
7
). This means either a hotter surface and a colder source (sedimentation mode)
or, conversely, a hotter source and a colder surface or substrate (growth mode). The growth mode will
lead to a dissolution of the source and ideally to nucleation and growth on the substrate due to the
different solvability of Si in Sn solution (F
ig
u
re
6
). Additionally, parasitic nucleation and growth can
occur adjacent to it, leaving crystallites floating on the melt. These crystallites hinder wetting of the
sample in future runs, so they are dissolved in sedimentation mode. This is done after every process
run for an hour, or during lab downtime for several days. Naturally, growth appears differently on
amorphous and crystalline substrates. On the latter, SSLPE leads to epitaxial growth. Since (111)
surfaces are most stable in silicon, growth is also dependent on substrate orientation. On (111)
substrates full layers will be formed. On (100) substrates pyramids will form. Thus (111) wafers are
- 12 -
2. State of the art
used for homoepitaxial SSLPE layers. Growth from amorphous Si is very different from that. As revealed
in this research, a dynamic crystallization of the seed interacts with different growth stages and is
described in chapter
5
.
5
. Nevertheless the result is a polycrystalline layer.
Figure 4 SSLPE utilizes a temperature dependent solvability for the deposition from solution. If top and
bottom temperatures are identical, forward and backward reaction are identical at both surfaces.
Figure 5: Typical temperature control during processing in the inner chamber. Heat up of the melt (I),
sedimentation mode (II), growth mode (III), sedimentation mode (IV) and cool down (V). Typical is a
base temperature of 600 °C at the source with temperature differences for sedimentation and growth of
5 K.
- 13 -
2. State of the art
2.2.Ox
i
de
contro
l
Si is known to be self-passivating, which means that it always forms a stable, thin oxide layer [31],
which prevents it from oxidizing further. This property was famously used in the Avogadro-project to
produce a new primordial kilogram that does not react with its environment over time [32]–[34]. A
general prerequisite for self-passivation is a similar density of oxide and substrate (Tab
le
1
). If that is
not the case, like with iron (Fe) and its various oxides, the layer would form but since no geometrical fit
between Fe and its oxide is given, just flake off. Additionally, Si has a high chemical affinity to oxygen, and
the reaction of the two elements will occur instantaneous if they meet. The reaction resolves
differently depending on the temperature and condition of the Si, but at room temperature a SiO
x
layer
will form within minutes, and will reach a thickness of 2 nm. At 2 nm oxygen diffusion through SiO
x
is
too slow to feed the reaction at the interface significantly. The reaction thus stops and the Si is
passivated. Its self-passivation is used technologically, for example in traditional CMOS technology or
for the chemical stability of the novel primary kilogram made from isotope-pure
28
Si. Nevertheless, to
apply electrical contacts, for homoepitaxial growth or for deposition of contiguous layers of metal, the
oxide has to be removed.
Table 1: Density of Si and SiO
2
which originate self-passivating behavior of Si.
material
Si
SiO
2
density
2.336 g/cm
3
(20°C) [35]
2.19 – 2.66 g/cm
3
(20°C) [36]
Concerning SSLPE, the oxide was identified early to hinder growth. Therefore several methods for
oxide prevention have been tested and applied and will be presented in the following and are depicted
in F
ig
u
re
6
.
·Fast handling in vacuum after a-Si deposition: Since a-Si is produced in situ and the chambers
are connected by a vacuum cluster, fast handling leads to a thinner or less developed oxide.
Especially in combination with a melt-back step this leads to successful growth. Unfortunately
humidity in the lab in combination with handling delays make this approach unreliable. Also,
this method is naturally not applicable to pre-produced substrates like wafers, or to stored a-Si
substrates.
- 14 -
2. State of the art
·Melt-Back uses an undersaturated solution in the beginning of the process to etch-off the top
layer of the substrate (Tab
le
6
a). An oxide layer is removed in the process. In the presented
SSLPE process, melt-back is realized by an inverted temperature gradient, with a hotter
substrate and a colder source. As stated above, no reliable reproducibility could be reached by
melt-back.
·Laser Treatment: Due to former problems concerning the oxide, a laser treatment system was
implemented. It allows to remove the oxide directly in the growth chamber and thus directly
before growth (Tab
le
6
b). Lasing has two effects – it ionizes possible surface contaminations
which repels them from the surface. Simultaneously it heats up the surface, so the oxide is
brought to reaction with the chamber atmosphere. The manifold measures and realizations in
connection with lasing are described later in this chapter and in chapter
5
.
6
.
·Hydrochloric-Acid-Etching: HF-dipping is a popular method in Si technology (Tab
le
6c
).
Unfortunately, it was formerly not used to success for SSLPE with amorphous substrates. For
crystalline Si, HF-passivation is stable for hours. Amorphous Si has more surface states and
dangling bonds and is comparatively unstable. Its implementation was part of this research
and required process adaption and is described in detail later in chapter
5
.
3
.
Figure 6: The three different methods applied for oxide control. (a) Melt-back was the first successfully
implemented method. Here, insertion of the sample during sedimentation dissolves the upper layer of
the substrate and thus also the oxide. (b) HF-dipping etches the oxide selectively and passivates the
surface. (c) Scanning the surface with a UV-laser heats up the surface and brings the oxide to reaction
with the chamber atmosphere.
2.2.1.
Ox
i
de
contro
l
by
hydrof
l
uor
i
c
ac
i
d
etch
i
ng
On silicon surfaces grows a self-passivating, stable oxide. This is known for crystalline [37] as well as
for amorphous [38] material. In the research of silicon SSLPE growth this oxide was a detrimental factor
- 15 -
2. State of the art
from the beginning on. Unknown at first and revealed later, the oxide layer shields the Sn solution from
contact with the seed layer [39]. The resulting growth model already presumed that Sn and seed have to
be in direct contact so that solved silicon atoms would “see” the suspected nanocrystalline roots in the
seed. The interaction of Sn and seed have proven to be more complex, but the original theory
considering the contact of both has remained true and is proven in this chapter again. Five methods of
oxide removal or prevention have been discussed and / or used.
1. Thermal evaporation
2. Seed layer protection by a sacrificial layer of germanium [39]
3. Melt-back [22], [39]
4. Hydrogen passivation [40]
5. Laser cleaning [39], [41]
Since points 1. and 2. were discarded as a possibility in the early stage of research and they are not
discussed further here. Melt-back, the dissolution of the surface layers in the solution in the beginning
of growth, was applied successfully for some time but proofed not to be sufficiently reliable. When
SSLPE processing was changed, the substrates were prepared in batches and stored until they were
used in experiments. After this change the growth resulted in low-quality layers, ranging from no
growth at all to vastly inhomogeneous growth which only produced patches of crystallites. As a
consequence, a more reliable and stronger method of oxide control was needed, which was strong
enough to remove matured, thicker oxide layers. In order to enable growth on stored substrates,
approaches 4.) and 5.) were revisited and are presented in this and the following chapter.
HF-etching is a central method to all crystalline silicon based CMOS processing and a central step in
the famous RCA-cleaning. HF reacts with the SiO
X
and etches it off. Hydrogen ions are bonding with
dangling electrons on the surface of the now exposed silicon, passivating the surface. Despite its
popularity in science and industry, HF-dipping was not successfully applied to SSLPE before. The
absence of success was traced back to the higher density of surface states in a-Si – making the
passivation less stable. In crystalline Si, the passivation lasts for hours but only much shorter for a-Si.
Also the passivation decays at temperatures around 500°C [42], [43]. Regarding the long handling times
into the SSLPE chamber, and the long total time of storage during H
2
flooding, heat-up and during the
preparation of the melt by extended time for sedimentation, H-passivation was deemed impossible to
implement. Also preprocessing was plagued by errors during handling and flooding (see chapter
3
.
1
).
Nevertheless, the exclusion of sedimentation before growth and a more reliable handling and flooding
procedure shortened the wait time enough to make H-passivation last long enough. The following
- 16 -
2. State of the art
chapter reveals the influence of additional wait time on the growth layer, and thus also for the first
time the influence of oxide thickness or stability on SSLPE growth.
2.2.2.
Ox
i
de
contro
l
by
l
aser
treatment
The term “laser” describes coherent, monochromatic and intense light, as well as sources producing
laser light [44]. Pulsed lasers transfer highest amounts of energy in shortest amounts of time. For
physical experiments and material science, lasers give the opportunity for highly energetic interaction
between light and matter. Energy can be coupled into a surface precisely by just using one pulse, or a
precise number of them [45]. As HF-passivation, laser-cleaning of the substrate was adapted from
industrial applications for SSLPE. Laser-cleaning is known for a number of mechanisms of surface
alteration with positive impact on following processing. Laser treatment ionizes dirt-particles so that
they are statically repelled from surfaces. Depending on pulse lengths and energies, lasers heat-up
surfaces, again leading to cleaning effects and possibly to melting or even evaporation. In cases of high
intensities and short pulses, lasers exclusively heat-up just the electron subsystem, so fast that energy
cannot be emitted as phonons into the lattice. In that case, direct sublimation is observed. The
interaction of laser and surface proceeds in the following manner: The laser hits the surface and is
absorbed according to the Beer-Lambert-law [46]. The lattice is not affected directly, given that the
pulse energy is absorbed only by the electrons first by means of inverse Bremsstrahlung. Nevertheless,
the energy is initially only released into the interaction volume of laser beam and material [46]. Some
electrons with high kinetic energy will diffuse away from the interaction volume. Nevertheless the
majority of potential energy now stored in the electron subsystem is released into the lattice, creating
phonons and raising the temperature of the material. From there, heat also is conducted away from
the direct interaction volume. The high energy density in time and space of lasers leads to strong local
heat up and can lead to direct sublimation. Sublimation appears when the laser transfers energy faster
to the electron subsystem than it is transferred further into phonons. In that case electrons are
stimulated in a density which creates too many electron holes for the bulk to remain stable. For silicon,
pulse lengths of few picoseconds are the threshold for this effect. Thus, laser power and especially
pulse length must be considered in application and analysis [47]–[50].
This chapter presents two approaches to the understanding of the surface cleaning effect.
Experimental procedure and results show how the laser interacts with the surface and enables growth.
To accommodate the experimental growth results with precise evaluation of temperatures achieved,
heating by the laser was simulated with a thermodynamic approach [41], [45] and it was implemented
with the software package COMSOL Multiphysics [51]. The program allows for the combination of
- 17 -
0
0
2. State of the art
different physics packages. The simulation described in the following was achieved with a finite
element approach in which a cylinder is simulated, which represents a cut along the z-axis of the stack of
a-Si and SiO
x
. The laser is represented by a heat source modeled according to Beer-Lambert’s law, and
its heat is then transported into the stack but also absorbed by reactions which represent the
evaporation of SiO
x
or the melting of silicon.
A theoretical approach to the mechanisms leading to destabilization were already presented in the
author’s master thesis [52], of which the following considerations are extracts. As basis, we use the
knowledge about desorption of SiO
x
, a process which is well-known from its relevance in the field of
microelectronics or physical experiments like molecular beam epitaxy. For similar reasons as in this
work, oxide free silicon surfaces are a necessity for various processes. A common technique is the
heating of the sample to around 900 °C or higher under ultra-high vacuum conditions [53]–[56] and
also under gas deficiency [57]. Consistently it was stated that the silicon dioxide will most likely be
reduced by the pure silicon at their interface plane.
ᵄᵅᵄ
2
(s) + Si(s) ⇄ 2ᵄᵅO(ᵅ)( 1 )
Other reactions are also possible, but less likely, and will happen depending on the reached
temperatures and pressures. The most prominent two are:
2ᵄᵅᵄ
2
(s) ⇄ 2ᵄᵅO(ᵅ) + ᵄ
2
(ᵅ)( 2 )
ᵄᵅᵄ
2
(s) ⇄ ᵄᵅ (ᵆ) + ᵄ
2
(ᵅ)( 3 )
To predict which reaction happens at certain values of pressure and temperature, reaction enthalpies
for the possible reactions have to be calculated and compared. The chemical potential µ
i
of a substance
i interacting with other substances is described by the equation
µ
ᵅ
= µ
ᵅ
+ ᵄᵄᵅᵅᵄ
ᵅ
,( 4 )
with µ
ᵅ
as the p,T-depending term of the pure substance i and a term that contains the activity ᵄ
ᵅ
of
the substance in the mixture. R is the molar gas constant and T the thermodynamic temperature.
The molar free enthalpy ∆
ᵄ
ᵃ of the chemical reaction (molar Gibbs free energy of reaction) is formed by
summing the chemical potentials of the reacting partners thereby considering their
stoichiometry
ᵅ
- 18 -
0
0
ᵄ
0 0
ᵅ
ᵆ
ᵅ
0 0 0
ᵅ ᵅ
ᵄ
2. State of the art
∆
ᵄ
ᵃ = ∑
ᵅ
µ
ᵅ
.( 5 )
Putting ( 4 ) in ( 5 ) we get
∆
ᵄ
ᵃ = ∑
ᵅ
µ
ᵅ
+ ᵄᵄᵅᵅ ∏
(
ᵄ
ᵅ
ᵅ)
= ∆
ᵄ
ᵃ
0
+ ᵄᵄᵅᵅ ∏
(
ᵄ
ᵅ
ᵅ)
.( 6 )
Assuming the equilibrium state of the reaction ∆
ᵄ
ᵃ = 0 and having in mind that the quantities µ
ᵅ
are
constant at fixed total pressure p and temperature T we find
∆
ᵄ
ᵃ
0
(ᵄ) = −ᵄᵄᵅᵅ ∏
(
ᵄ
ᵅ
ᵅ)
= −ᵄᵄᵅᵅᵃ
ᵅ
ᵅ
−
∆
ᵃ
0
(
ᵄ
)
ᵃ
ᵅ
ᵅ
(ᵄ) = ᵅ
ᵄᵄ
,( 7 )
. ( 8 )
The equilibrium constant of a chemical reaction ᵃ
ᵅ
ᵅ
is therefore determined by the thermodynamic
potentials of the involved pure substances (at p and T). The chemical potential (or molar Gibbs free
energy) of such a pure substance i is calculated from its heat capacity at constant pressure ᵅ
ᵅ
= ᵅ
ᵅ
(ᵄ)
[58]. The calculation of µ
ᵅ
= ᵃ
ᵅ
bases on known values at standard conditions (p
⦵
= 1 bar,
T
⦵
= 298.15 K) and the integration to higher temperatures. Phase transitions (∆ᵃ
ᵆ
, ∆ᵄ
ᵅ
) have usually to
be considered but are not relevant here.
µ
ᵅ
= ᵃ
ᵅ
(ᵄ) = ᵃ
0
(ᵄ) − ᵄᵄ
ᵅ
(ᵄ), ( 9 )
= ᵃ
0
,⦵ + ∫
ᵄ
ᵅ
ᵅ
(ᵄ)ᵅᵄ − ᵄ (ᵄ⦵ + ∫
ᵄ
ᵅ
ᵅ
(ᵄ)ᵅᵄ), ( 10 )
ᵄ
⦵
ᵄ
⦵
Knowing ᵃ
ᵅ
ᵅ
at the regarded temperature in the range from room temperature to 873 K gives the
opportunity to find the process conditions where the potentials of synthesizing products from
reactants and reforming reactants from products are balanced. The formation of products is favored
when ᵃ < ᵃ
ᵅ
ᵅ
under actual process condition.
The dimensionless activities of the substances of Eq. 1 have to be suitable defined to obtain K. The
solid substances SiO
2
and Si are not mixed. Therefore, the activities are set 1 (pure substances, second
term of Eq. 1 vanishes). In a rough approximation not considering excess interactions with other gas
molecules the activity of gaseous SiO is derived from the ratio of its gas pressure ᵅ
ᵄ
ᵅ
ᵄ
to the gas
pressure at standard condition p
⦵
[59]. Then, the calculation of K reduces to
- 19 -
2
ᵄ
ᵄ
ᵄ ᵅ
ᵄ
ᵅ
ᵄ
−
ᵄ
1
0
ᵃ =
ᵅ
⦵
ᵅ
2
ᵃ
ᵄ
2. State of the art
ᵃ = ᵄ
ᵄ
ᵅ
ᵄ
2
ᵅ
ᵄ
ᵄ
ᵅ
= ( ᵅ⦵
)
2
.( 11 )
Using the equations ( 8 ) and ( 9 ) the equilibrium pressure ᵅ
ᵄ
ᵅ
ᵄ
can explicitly be written as
∆
ᵃ
0
(
ᵄ
)
ᵅ
ᵄ
ᵅ
ᵄ
,
1
(ᵄ) = ᵅ⦵ᵅ
2
ᵄᵄ
,( 12 )
with ∆
ᵄ
ᵃ
1
(ᵄ) referring to the Gibbs free energy of reaction of ( 1 ).
Also, removal of a SiO
X
-layer with a reducing H
2
-atmosphere is possible and is usually described with
the following reaction [53], [55], [60] and the corresponding reaction constant K.
ᵄᵅᵄ
2
(ᵆ) + ᵃ
2
(ᵅ) ⇄ ᵄᵅᵄ(ᵅ) + ᵃ
2
ᵄ(ᵅ)
ᵄ(ᵄᵅᵄ) ∙ ᵄ(ᵃ
2
ᵄ)
ᵄ(ᵄᵅᵄ
2
) ∙ ᵄ(ᵃ
2
)
( 13 )
( 14 )
To achieve a simplification similar to ( 11 ) several assumptions can be made:
·the activity a can be set to ᵄ = ᵅ / ᵅ⦵ for gaseous components and to ᵄ = 1 for solid ones
[59], as done before
·During heating, the partial pressure of H
2
is always set to ᵅ
ᵃ
2
= 1050 ᵅᵄᵄᵅ and we will
therefore approximate ᵄ(ᵃ
2
) = ᵅ
ᵃ
2
/ ᵅ⦵ ≈ 1.
ᵃ = ᵅ
ᵄ
ᵅ
ᵄ
∙ ᵅ⦵( 15 )
·The partial pressure of ᵃ
2
ᵄ as product is estimated to be equal to the vacuum pressure before
the H
2
flooding, since its content in the chamber atmosphere does not change. Even if the
partial pressure of the H
2
is higher, the partial pressures of this components still corresponds to
the vacuum pressure before the H
2
-flooding (ᵅ
ᵃ
2
ᵄ
= ᵅ
ᵆ
ᵄ
ᵅ
).
· ᵅ
ᵄ
ᵅ
ᵄ
represents the target value for this estimation and is supposed to clarify if the chamber
vacuum pressure is sufficient to promote the reaction. In this case of contemplation, the
- 20 -
ᵅ
⦵
−
ᵄ
2
0
2. State of the art
pressure of the ᵄᵅᵄ corresponds to the vacuum pressure (ᵅ
ᵄ
ᵅ
ᵄ
= ᵅ
ᵆ
ᵄ
ᵅ
) and we can therefore
conclude ᵅ
ᵄ
ᵅ
ᵄ
= ᵅ
ᵃ
2
ᵄ
as well.
ᵃ =
(
ᵅ
ᵄ
ᵅ
ᵄ
)
2
( 16 )
Using the equations ( 9 ) ( 17 ) and the equilibrium pressure ᵅ
ᵄ
ᵅ
ᵄ
can be written as
∆
ᵃ
0
(
ᵄ
)
ᵅ
ᵄ
ᵅ
ᵄ
,
2
(ᵄ) = ᵅ⦵ᵅ
2
ᵄᵄ
,( 17 )
with ∆
ᵄ
ᵃ
2
(ᵄ) referring to the Gibbs free energy of reaction of ( 13 ). The calculations ( 12 ) and ( 17 )
differing only slightly in the values of ∆
ᵄ
ᵃ
0
by the varying stoichiometry of the constituents result in
similar values of critical pressure ᵅ
ᵄ
ᵅ
ᵄ
and imply the thermodynamic probability within the
experimental conditions. Since both, especially ( 17 )include several estimates, the reactions ( 1 )and
( 13 ) have been analyzed with FactSage© as well.
The calculation routines in FactSage© apply the principles described in this chapter for critical values
of pressure and temperature (F
ig
u
re
7
) analogously to the calculations. They are refined by taking into
account excess interactions of the involved materials if yet determined. Here, reaction kinetics is
obviously ignored and the values have to be understood in terms of thermodynamic possibility. The
plots show the equilibrium values of chamber pressure and temperature. Lower pressures or higher
temperatures will promote the reaction.
The plots were produced with the “reaction” function of FactSage©, and for different pressure values
the critical temperatures ᵄ(ᵅ,∆
ᵄ
ᵃ = 0) were taken. For ( 13 ), the pressure of ᵃ
2
ᵄ was set to ᵅ
ᵃ
2
=
1050 ᵅᵄᵄᵅ for the reasons explained above. F
ig
u
re
8
shows a solution of ( 26 ), with the peak
temperature over time (a) and the temperature profile in the direction after different durations.
- 21 -
2. State of the art
Figure 7 Critical temperature subjected to ambient pressure in reference to the conditions in the
growth chamber. Although the chamber is flushed with H
2
at 1050 mbar, the partial pressure of any
other component is expected to correlate to the vacuum pressure before the flooding around 10
-7
mbar exemplarily. Temperatures above the critical values are expected to favor the products of the
reaction.
As F
ig
u
re
7
shows for the most relevant pressure of 10
-6
mbar that the forward reactions are both
favored above 750 °C, with only a difference of around 30 K. This being a completely theoretical
discussion, several comments have to be made, since these calculations ignore any kinetical
circumstances of the reactions
·The reactions most possibly require a relevant additional activation energy, which would
increase the critical temperature. This is especially relevant for laser-heated systems, since the
amount of energy available for surpassing critical points is limited by singular pulses’ energies.
·For reaction ( 1 ) experimental values around 900 °C at 10
-9
mbar have been reported [53]–
[56], differing highly from the calculated value of around 600 °C, and a similar addition can be
expected for the reaction with H
2
.
·The reaction ( 13 ) with H
2
will happen at the interface of SiO
X
and H
2
, and possibly also within
the SiO
X
layer, since gas is able to infiltrate the bulk. Also, the gaseous reactant H
2
is steadily
transported to the interface by convection. In contrary, reaction ( 1 ) will happen at the solid-
solid interface of the Si-bulk and the SiO
X
, obviously relying on diffusion in the solids alone and
- 22 -
ᵅᵃ
ᵅᵃ
ᵄ
=
2. State of the art
without for example the faster convection for both supply and exhaust of the reactants and
products respectively.
Concluding, we expect both reactions to happen from a similar critical temperature, considerably
above 750 °C, while the reaction with H
2
should then take place faster due to the mass transport
advantage.
As mentioned before, the energy dependence of the temperature is a dynamic heating problem and
requires an appropriate solution. An analytical approach was done in [46], and will be presented in the
following. An elevation of temperature above its ambience represents a change in potential and can be
understood as a rise in the potential of the concerning volume, and is usually denoted as enthalpy,
which can be written as
∆ᵃ = ᵅᵅ
ᵅ
∆ᵄ ( 18 )
Or for constant pressure with the relevant specific heat constant or equivalently for constant volume
∆ᵃ
ᵄ
= ᵰᵅ
ᵄ
∆ᵄ ( 19 )
The change of heat through heat flow from warm to cold can be written as
∇ ⋅ ᵄ = − ᵅᵆ
ᵄ
, ( 20 )
where a positive value represents a loss of heat in the observed element. To elevate the temperature,
a negative outcome is required.
If a heat source S like the laser irradiation is considered, it expands the equation to
S(z) − ∇ ⋅ ᵄ = ᵅᵆ ,( 21 )
To which a one dimensional solution is (with ( 19 ) ( 20 ) and ( 21 ):
ᵅᵄ ᵅ ᵯ
2
ᵄ( 22 )
ᵅᵆ ᵰᵅ
ᵅ
ᵯᵆ
2
,
with the yielded energy of the beam being depending on the depth z in the bulk, since in smaller depths
more energy will be absorbed due to the remaining energy in the beam. As mentioned above, the
- 23 -
ᵅ
1
ᵅ
1
ᵆ
2. State of the art
converted energy is also dependent on the optical absorption coefficient α. The remaining intensity in
a depth z can be written as
ᵃ(ᵆ) = ᵃ
0
exp (−ᵯᵆ)( 23 )
and the source term would be
ᵄ(ᵆ) = ᵯ ᵃ(ᵆ) = ᵯ ᵃ
0
exp (−ᵯᵆ). ( 24 )
Combined, solutions are possible for the temperature development during an ongoing heat transfer:
ᵄ(ᵆ,ᵆ < ᵰ) = 2 ᵯ ᵃ
0
(1 − ᵄ)(ᵃ ᵆ)
2
ᵅᵅᵅᵅᵅ
[
2(ᵃ
ᵆ
ᵆ)
2
1
]
( 25 )
and for a combination of irradiation and the following cooling:
ᵄ(ᵆ,ᵆ < ᵰ) = 2 ᵯ ᵃ
0
(1 − ᵄ){(ᵃ ᵆ)
2
ᵅᵅᵅᵅᵅ
[
2(ᵃ
ᵆ
ᵆ)
1
2
]
( 26 )
− (ᵃ ᵆ − ᵰ)
1
2
ᵅᵅᵅᵅᵅ
[
2(ᵃᵆ − ᵰ)
1
2
]
}
Figure 8 Shows plots of equation ( 26 ) solved with MathLab for the surface temperature over time (a)
and for the temperature in the depth of the specimen, at the end of a pulse (1.5 ns) and at later time
points (3.0 ns and 9.0 ns). Naturally, the distribution will even out with time, cooling the hotter and
warming the cooler sections. Calculations were done with a laser intensity of I
0
= 0.4 W / m² and the
absorption coefficient = 1.47 × 10
-9
1 / m.
- 24 -
2. State of the art
2.3.Crysta
lli
zat
i
on
of
amorphous
s
ili
con
The formation of microcrystalline Si layers from amorphous seed layers will be dealt with in this
section. In previous experimental research and growth models of the SSLPE process as a whole,
crystallites nucleated and matured to nano-size in the amorphous seed layer by annealing as an
inherent part of the a-Si deposition. As stated above, these crystallites were not found any more after
changes in the process were implemented. Novel substrates which did not exhibit nanocrystalline
roots in the seed layer still lead to successful growth. In the first instance the seed layer appeared as a
black box, with the only recognizable initial amorphous state before growth and its polycrystalline state
afterwards. In literature the crystallization of amorphous Si is well documented, since it attained
technological relevance concerning thin film photovoltaics and flat panel displays.
Amorphous Si is metastable and annealing leads to crystallization. Temperatures can be as low as 0.6
times of the melting temperature of Si [61], [62]. Crystallization does not commence immediately but
is preceded by an incubation time. It must be noted that both incubation time and crystallization speed
itself vary vastly depending on the nature of deposition method, in example if it was achieved by
physical evaporation of Si or chemical reaction of Si-containing gases. Both incubation time and
crystallization are faster if annealing is done at higher temperatures, so that annealing at 600 °C leads
to incubation time of 2 h and crystallization time of additional 2 h. At 638 °C, both are reduced to 1 h
[63]. Also, layer thickness and dopants or impurities in the amorphous material have a notable
influence on the crystallization (compare Tab
le
2
).
- 25 -
2. State of the art
Table 2 Examples from literature and this work showing the high range of results in the crystallization of
a-Si on glass depending on various process parameters with d: layer thickness; T: operation
temperature; t
inc
incubation time; and t
cryst
crystallization time.
fab
r
ication
c
hara
c
teri
z
ation d
µm
treat
m
ent
T
tin
c
°
C
min
t
c
ryst sour
c
e
min
1
CV
D 0
.
06 nm
/s
S
i2H6 450 °
C
2 P
VD
0
.
3
n
m
/s
o
ptical tran
s
mi
ss
i
o
n
electrical c
o
nductivity
0.3 thermal
0.2 thermal
570 1800
600 240
638 60
550 240
600 30
780 [63]
120
30
780 [64]
120
3 P
ECV
D
TE
M
250 °
C
0.2 thermal 700 [61]
MI
C
(
Pd
)
500
4 P
ECV
D
250 °
C
0
.
31 A
/s
TE
M, x
-
ray, electrical 0.6 thermal 700 4 [62]
c
o
nductivity
5
n
ot re
po
rted n
o
t rep
o
rted thermal 800 [65]
MI
C
600
6 P
VD
400 °
C
1
.
5
n
m
/s
7 P
VD
400 °
C
1
.
5
n
m
/s
in
s
itu
TE
M
Raman
2 MI
C
2 MI
C
(
A
)
MI
C
(
B
)
600 5 10 this work
600 0 5 this work
600
>
1 h
- 26 -
2. State of the art
The use of a metal as catalyst accelerates the phase transformation remarkably compared with simple
annealing. This procedure is called Metal Induced Crystallization (MIC). It is long known that
amorphous semiconductors crystallize at lower temperatures and faster when in contact with a metal
layer. This includes a reduced incubation time. The exact mechanism was not revealed for a
considerable time of MIC research. For the popular model-system Si-Sn it was long thought that a
eutectic mixture would form and its reduced melting point would lead to a liquid solution. Material
transport in liquid would be promoted by convective effects and as a result would lead to a fast
material transport. In situ observed experiments eventually disproved this theory [66] and revealed
the true mechanisms. The formation energy for Si crystals is reduced if another crystalline material like
solid Sn is in contact with amorphous Si and the lattice of the Sn acts as a structural template. Si atoms
adhere to Sn grain surfaces or interfaces and are directed to their respective lattice positions, an effect
which can be interpreted as reduced nucleation energy. Diffusion is promoted by Sn as transport
medium, another important factor in MIC. This concerns not the bulk of the tin, but the grain
boundaries which contain an increased density of defects. Si atom`s diffusion is enabled by those
impurities and they jump from vacancy to vacancy [67]. The enhanced diffusion in combination with
the reduced nucleation energy leads to crystallization with shorter incubation time and at lower
temperatures. It is even reported that MIC can lead to a layer inversion of seed and catalyst metal. In
that case Si atoms diffuse from the seed through the metal layer and form a layer on the opponent
side of it. When the new layer grows, the original layer is depleted and pushes forward the metal layer.
Catalyzation can not only appear with full layers of metal but also with co-deposition during deposition
of amorphous silicon, leading to doping with precipitation. An outstanding example is the use of
aluminum as catalyst for both examples, layers as well as doping. Aluminum leads to crystallization
temperatures of under 300 °C while other metals usually lead to crystallization temperatures around
600°C [65].
Tab
le
2
shows the wide range of crystallization phenomena in crystallization of amorphous Si as
reported in literature as well as the variations how researchers focus on different aspects. [63] and
[64] report similar crystallization times, but incubation times differ widely. The reason of varying
incubation time might be found in the very slow deposition rate of Si in [63] supposedly leading to a
truly amorphous layer. [61] mentions no specific times but reports a reduction of critical temperature
by application of a super-thin palladium layer. The metal catalyst reduces crystallization temperature
from 700 °C for annealing without palladium to 500 °C when the layer is applied, demonstrating the
effect of MIC. The palladium layer is not stable but forms droplets. At the critical temperature of 500 °C,
crystallization is only observed at palladium-Si interfaces but not on Si surfaces which are not covered
- 27 -
2. State of the art
by palladium. [62] demonstrated ultra-fast crystallization at 700 °C rapid thermal annealing in just 4
min outperforming the other examples given here. [65] gives as rule of thumb a critical temperature of
800 °C for thermal crystallization of amorphous Si and 600 °C for metal-induced one.
In the case of SSLPE, Sn acts as a liquid solvent and is not solid like in the examples discussed above.
Since liquid Sn is used, contrary to crystalline Sn, its role has to be interpreted anew, and was
investigated earlier but not fully understood [68]. Primarily it is assured that Sn cannot provide a
heteroepitaxial template which would support crystal formation further. Also there are no grain
boundaries which act as diffusion channel as in MIC. Nevertheless liquid Sn excels the solid phase in
solvability and diffusivity, enabling rapid atom transport through the liquid´s bulk. The material
transport should also be enhanced by convection. Since the volume which is available for diffusion is
much larger in liquid state, it is very likely that more Si diffuses in liquid Sn than in solid one. Since the
chemical potential of amorphous Si is higher than that of the crystalline phase, the liquid will solve a-Si
and precipitate c-Si even at steady temperatures without a gradient present, an effect labeled
amorphous-liquid-crystalline (ALC) process [68]. This also leads to etching of the amorphous layer, as
used technically in this work for melt-back of the oxide layer. Since etching creates a larger surface
area and crystallization is promoted at surfaces, it could be argued that etching of the seed supports
crystallization additionally.
Summarizing it can be said that the exact mechanisms of liquid Sn interacting as possible catalyst with
the amorphous Si seed is not described in literature. The different possibilities of mechanisms which
could be enhancing or restrictive to the crystallization demand for an in-depth investigation of the
crystallization. The final experiment should be able to observe the crystallization quasi in situ so that
incubation and crystallization periods can be recorded. Also one should be able to clarify wether the
presence of Sn propagates or changes crystallization.
It was decided to perform an in situ transmission electron microscopy (TEM) experiment. As described
in the following, TEM in general is a method which combines the strength of scattering experiments
with actual microscopic imaging of the microstructure. Thus TEM is able to observe phase changes and
extract crystalline information. The in situ experiment designed simulates the reaction between Sn and
Si and by this the crystal formation can be observed directly.
- 28 -
2. State of the art
2.4.
I
ntroduct
i
on
e
l
ectr
i
ca
l
propert
i
es
SSLPE was designed and invented as a method to cover large surfaces of inexpensive glass with
crystalline silicon. A simple, exemplary silicon solar cell can be produced with only two active layers
(F
ig
u
re
9
). With the p-type silicon which is produced by the presented SSLPE setup, the thicker
absorber layer could be realized. As emitter, a p-doped layer has to be produced, for example by
further doping of the surface, or by applying additional doped material, for example doped a-Si by
chemical vapor deposition. Contacts have to be applied in a way that sunlight reaches the active
material. An additional anti-reflection coating on the sunny side can be used to reduce reflection and
electrically passivate the silicon, both enhancing efficiency.
The manufacture of solar cells was attempted on two different routes. The first approach included
SSLPE layers grown on RPS substrates, which were detached from their substrate wafer.
Functionalization was done at Institut National de l'Énergie Solaire (INES), France and metallization at
IMEC. Electrical measurements pursued at Helmholtz-Zentrum Berlin für Materialien und Energie (HZB)
revealed that the devices were not working. As most likely reason, heavy shunting was identified which
prohibits build-up of an open-circuit voltage V
OC
. This was likely since the super-thin foils of few tens of
micrometers inhibited internal stress and did bend after detachment. Also, due to their low
thicknesses well below 40 µm, mechanical damage during transport, processing and measurement lost
some substrates and prototypes. It is very likely that other foils which were not lost still have suffered
microcracks or mechanical stress, which is reported to reduce efficiencies of cells as well [12], [16],
[17]. In a second attend to manufacture prototypes in cooperation with HZB, different substrates were
used to circumvent reliability on one substrate type being improper for processing or prone to
mechanical failure. To succeed with the manufacture and to have an easy-to-handle system,
homoepitaxial SSLPE silicon grown on a silicon wafer was used. Additionally, RPS substrates were used,
but detachment from the parent wafer was foregone to minimize mechanical stress. Alongside both,
polycrystalline SSLPE silicon on glass was also processed. A functionalization method for crystallized
silicon on glass [13], [92]–[95] was adapted for SSLPE material. The method originally is used on LPC
substrates, but was adapted to SSLPE material. For polycrystalline SSLPE silicon the adaption proved to
be futile, since the surface with exposed crystallites of several tens of micrometers was too rough for
the different materials to adequately wet or bond on the surface and form connected layers. The
monocrystalline materials where polished and could be coated. In both cases, a p-doped substrate
provides backside passivation to the n-type SSLPE silicon and contacts as well as an emitter layer can be
applied from only the sunny side of the device.
- 29 -
2. State of the art
Figure 9 Sketch and band model of a simple solar cell with a n-type absorber (light teal), a p-type
emitter (dark teal), metal contacts (grey) and functional coating (blue). The energy difference between
the differently doped semiconductors causes bending of the energy bands. When light enters the
absorber and interacts with electrons, they are excited and electron-hole pairs are created. The
potential between n- and p-type pushes positive electron holes in the valence band to the front
contacts and negative electrons in the conduction band towards the backside contact, creating
electrical charge.
Thus the charge separation is not accomplished vertically from frontside to backside of the active layer.
Instead, charge is separated horizontally in a volume which is covered by emitter and a transparent
conductive oxide (TCO), and positive holes are collected at that emitter, while negative electrons drift
to the metal contacts which surround the emitter (F
ig
u
re
10
). This mechanism of separation underlines
the need for the typical material parameters which indicate the efficiency of a solar cell: high charge
carrier density, long life-time of minority carriers and consequentially long diffusion lengths of charge
carriers. A high density of charge carriers defines the maximum charge which could be created in an
ideal cell. If the doping density is too high, charge carriers could affect each other negatively, but for n-
type silicon a density in the range of N
d
= 10
16
cm
-1
has been proven to achieve the best outcomes and
balance V
OC
and J
SC
[16], [95], [96]. High lifetimes before recombination allow for long diffusion
lengths, and are needed so the charge carriers created can reach the contacts. In the case of a regular
state-of-the-art, crystalline solar cell this amounts to the thickness of the active material around
150 µm to 300 µm.
Deep level defects obscure the electrical properties in multiple ways. They can be identified as charge
carries, making charge carrier density appear higher than its effective value. Additionally, deep level
- 30 -
2. State of the art
defects can act as recombination centers and will trap drifting charge carriers before they are collected.
With solar cell production as final goal of the research, the electrical properties of the SSLPE silicon
were always in focus of research. Previously conducted measurements which are already published in
[30] are compiled in the experimental and result section of this chapter again to demonstrate the
theoretical suitability of the SSLPE material as solar cell absorber material. The goal of this chapter is
the manufacture and assessment of a working device. High values in electron mobility and lifetime
suggest efficiencies between 12 % and 18 %, compared to publications concerning crystalline silicon
on glass and by exploratory simulations with the software AFORS-HET [97]. Data gathered before
shows that mobility measured is especially high for the measured carrier concentration [30]. As this
chapter shows, this in disagreement to devices manufactured from SSLPE layers which exhibit only
minor efficiencies. Especially the formerly obtained data from Deep Level Transient Spectroscopy
(DLTS) is presented again and analyzed with a special focus on the effect of deep level defects on cell
efficiency. High carrier concentrations could stem from deep level defects [98]. While generally a high
carrier concentration is desirable for an absorber material, deep level defects can be detrimental. DLTS is
used to distinguish between minority and majority carrier traps and can be used to assess activation
energy and density of possible deep level defects. Capacitive DLTS saturates defects with an electric
bias and uses an electric pulse to activate them [99]. The transient is then recorded at different
temperatures.
- 31 -
2. State of the art
Figure 10 sketch of the produced prototype devices and band models along one lateral and two
horizontal cuts. Due to the configuration of the active layer (n-type) being grown on a silicon wafer,
functionalization has to be done just from the sunny side of the cell. Electrical contacts are realized
with metal representing the back side and TCO on the p-type emitter representing the front side of
the cell. The p-type substrate passivates the backside of the absorber, pushing back excited electrons,
letting them drift laterally. Then they are collected by the metal contacts. Holes are collected by the
emitter layer and consequently the TCO layer.
- 32 -
3. Materials and methods
3.
Ma
t
er
i
a
ls
and
me
t
hod
s
3.1.
I
nher
i
ted
process
and
centra
l
process
i
mprovements
The SSLPE cluster tool contains the SSLPE growth reactor and the PVD chamber (Fehler! Verweisquelle
konnte nicht gefunden werden.). The handling systems were designed to transport samples from and
into the loadlocks and into the two different growth chambers. It also provides switching of the sample
holder for PVD deposition and inserting of the standard substrate holder into the swivel arms in the
growth chamber. Originally it was designed for automated handling and processing of substrates.
Understanding its layout is important, because its possibilities and restrictions define experiments.
Central to its design are sample carriers made from graphite. Substrates, independent from their type,
are fixated in them. Transport is based on grabbing the carrier (as with handler arm 2), or dropping it
into designated cavities (as with the swivel arms).
As depicted in F
ig
u
re
11
the cluster tool consists of
·SSLPE chamber (f): consisting of the inner and the outer growth chamber. Two swivel arms (f1
and f2) are used for holding and transporting the carrier, which is inserted by handler 2 (d1)
and dropped into the rectangular opening of the arm, with the substrate hanging upside down.
The inner chamber (f3) contains the crucible (f4) surrounded by the heating system heater as
descripted above. When moved from their holding position into the liquid tin, the swivel arms
move in a programmed combination of lateral and pivoting movements, so that only the
carrier`s edge is inserted at first. With a lateral movement, residual material like floating Si on
the Sn surface is moved aside before full insertion of the carrier and thus the substrate. When
the insertion is completed, the design of the arm allows substrate and carrier to float on the
Sn and the arm just fixates its lateral position. For retraction, the described movement is
applied backwards. The secondary arm is used with the same movement to clean the Sn
surface once before growth. In the outer chamber different waiting positions for the swivel
arms and thus substrates are possible. They can be used to keep the substrate for reduced
heat exposure during heating. Also laser treatment is applied in the outer chamber.
·Handler chamber 2 (d): Connects loadlocks and the growth chamber and is therefore used
extensively and during every growth run. A robotic arm (d1) with a grabbing mechanism is
used to transport the graphite holder in and out of the attached chambers. Since it is
automated and the operator could not directly intervene if problems occur, several initiation
- 33 -
3. Materials and methods
steps are included in its program for safety. This elongates handling times and samples have
to be kept in this chamber for several minutes before they can be inserted.
·Loadlock 1 (e): equipped for the handling of the graphite holder
·Loadlock 2 (c): equipped to transfer the graphite holder into another holder designed for the
PVD chamber and back. The process for doing so is slow and produces errors often. If an error
occurs, the operator has to open the chamber and do the transfer manually.
· Handler chamber 1 (b): transfers the PVD holder between the loadlock and the PVD chamber
· PVD chamber (a): consists of a spinning sample holder on the top and an electron evaporator
on the bottom. Four heater lamps are installed around the chamber and heat the substrate.
Technical detail are noted in the process section
- 34 -
3. Materials and methods
Figure 11: Sketch of the cluster tool, harboring most importantly the PVD chamber and the SSLPE
growth chamber. They are connected by an automated handling system, that allows for transport
through vacuum by two robotic arms and two loadlocks. A detailed description is given in the text.
- 35 -
3. Materials and methods
Additionally to the chambers itself, the substrate carriers are important. The mainly used graphite
holder can carry substrates up to a thickness of 3 mm and a maximum area of 4 x 4 cm
2
. It can be
handled in the reactor chamber, handler chamber 1 and in loadlock 1. For PVD, an aluminum adapter is
designed to take in the graphite carrier in loadlock 2 and handled by handler 1.
The process the author inherited was the following: a singular glass substrate would be mounted in
the graphite carrier and both into the aluminum carrier. All three would be mounted in the PVD
chamber where a-Si deposition would take place. Since PVD is controlled manually and challenged by
incidental errors, the exact deposition takes place differently every run. Afterwards the assembly of
carriers would be handled into loadlock 2, were a lift mechanism would separate the two carriers and
handler 2 would pick up the graphite carrier and handle it further to the reactor chamber. The handling
process takes 30 min at minimum, but can take up to 90 min easily if errors occur. The unreliability
lead to a modification of the process eventually. Instead of trying to handle the sample in vacuum with
unpredictable success, the operator would extract the sample from the PVD chamber directly and
switch handlers for SSLPE manually. The short but calculated exposition to air was in total superior to
the unpredictable handling in vacuum. After insertion into the SSLPE chamber, H
2
flooding would
proceed and after 7 min, when overpressure is achieved, a temperature program would heat the inner
chamber to working temperature (600 °C most of the time). When reaching that temperature after 25
min, it would be held in sedimentation mode for 30 to 60 min. Afterwards insertion into Sn would be
executed. During that period after PVD substrates were in high vacuum of 10
-5
to 10
-7
mbar, but still
residue O
2
would lead to Si oxide grown on their surface. It has to be noted that even highly controlled
processing and highly focused lab work did not seem to produce adequately predictable outcomes.
Regularly fringe growth results were produced despite seemingly identical lab routines. Since the list of
preprocessing steps was long these results could not be tracked down to singular steps.
Oxide control was recognized early as the major challenge and as an opportunity to make the process
more reliable and the grown layers more often satisfactory (see above). Nevertheless the outlook for
improvements of the handling process was small. Another major problem recognized by the author
was the sensitivity of the PVD process and subsequently of the characteristics of the a-Si layer.
Research shows that small changes in the parameters of the deposition of a-Si lead to vastly different
annealing behavior (see Tab
le
2
). Also, the described nanocrystalline roots were not observed
anymore, possibly due to faster deposition implemented. In summary it can be said that results of
SSLPE growth were hardly to predict and results would vary even under strict process routine and good
lab conditions. A high number of preprocessing steps made it difficult to pin down changes of growth
- 36 -
3. Materials and methods
results to changes in processing. The volatility and sensitivity of PVD and handling, as well as
completely uncontrollable circumstances as air moisture in the lab complicated improvements and
development of a holistic growth model. Eventually, four goals for a predictable process were defined
i. Elimination of PVD sensitivity as a factor in growth
ii. Elimination of Si oxide in growth or reduction of its volatile influence or a reliable assessment
of its influence and boundaries
iii. Asses the role of the amorphous silicon, its crystallization behavior and the role of
nanocrystalline seeds.
iv. Elimination of highly varying durations of the preprocessing steps
After evaluation of the possibilities the following measures where applied
a) The use of a holder for multiple substrates during PVD was designed, manufactured and
implemented. It requires PVD`s chambers loading through its door and makes it impossible to
directly process the produced substrate, since carriers are now incompatible. Nevertheless
using this novel holder eliminates PVD sensitivity of the entire process (i.) but requires absolute
oxide control (ii.). If implemented successfully, it allows for an assessment of the seed layer`s
role (iii.), since comparative processes will be possible with the virtually identical substrates
from one PVD process run.
b) Originally deemed improper to SSLPE with a-Si substrates, HF dipping eventually was made
possible. As shown in the respective chapter, it requires a reduction of the preprocessing time
to a minimum. Former experiments implicated that H-passivation is too unstable on a-Si. After
H-passivation was implemented and the process was adapted adequately, full oxide control
(ii.) was achieved. Additionally, all handling steps which were heavily prone to errors (iv.) are
circumvented.
c) Instead of processing freshly prepared substrates in-line and in vacuum, samples are extracted
and stored between days and months now. Before an experiment, a single substrate is cleaned
and HF-dipped. Since it is inserted directly into loadlock 1, it can be put fast into the growth
chamber and thus into passivating atmosphere. As an additional benefit, characterization and
quality control can be done for on one of the substrates for the entire batch.
d) The combination of measures a, b and c have enabled nearly perfect reproducibility between
the 8 substrates which can be prepared in a single PVD run. HF-dipping is very reliable and
allows to just use the reliable and fast handling route from loadlock 1 to the reactor. It has to be
noted that the three measures presented had to be implemented in combination and would
- 37 -
3. Materials and methods
not have been helpful as singular alterations. These improvements lay the bases for a number
of experiments presented in this thesis and are the key to the holistic growth model which is
now available.
3.2.Laser
setup
To allow for a laser treatment directly before growth and within the growth chamber a laser system
was installed underneath the SSLPE cluster tool and is shown F
ig
u
re
12
. Its layout shapes the beam
and directs it into the chamber. A scan head moves the beam across the substrate surface. The
particularities of the elements are:
a) Laser source: as laser source a diode-pumped passively Q-switched solid state laser (FQSS-266-
200, CryLas) with a wavelength of 266 nm and a pulse width of 1.4 ns (FWHM) is used.
b) Mirror 1: this mirror is necessary to use the space on the breadboard effectively
c) Iris: the manually operated iris diaphragm has two purposes. First it can be used to change the
size of the beam and thus the spot. Additionally it is possible to move the opening in x- and y-
direction within the beam. This is useful to select a mostly flat part of the original beam and to
produce a gauss-like energy profile without vast intensity differences (compar
e
F
ig
u
re
30
a and b
showing an unadjusted and adjusted beam profile)
d) Lens 1: this lens focusses (focus length f1 = 250 mm) the beam and can be used to adjust the
intensity of the spot on the substrate. It is mounted on a rail and can be moved in z-direction.
Since the beam is parallel in front of the element, movement in the z direction moves the
focusing point (close to f) by the same amount. Nevertheless it is important not to move the
focus point into mirror 2 to prevent damage.
e) Attenuator: a manually attenuator wheel with a gradual, radial shading to dim the beam. It is
adjusted manually and enables the operator to work freely with beam transmissions from 0 %
to 100 % gradually.
f) Mirror 2: this mirror directs the beam upwards towards the chamber and the scan head
g) Lens 2: this lens is fixed directly underneath the scan head and focusses the divergent beam
directly in front of the substrate (focus length f2 = 400 mm).
h) Scan head: The scan head moves 2 mirrors with electromotors in the x-y plane of the beam
(SCANcube, SCANLAB). It is programmed with vector lists written in Hewlett Packard Graphics
Language (HP-GL), a language originally created to program pen-plotters [69]. Thus the
mirrors’ movement is not directly linked with the laser’s pulsing and writing dots is not part of
- 38 -
3. Materials and methods
the language, since its original purpose are vectors plotted by lateral movement. HPGL
contains commands to rest pen movements after jumps originally intended to prevent
wobbling of the pens. To plot precisely dot-by-dot anyway, these wait times after vector jumps
are programmed to concur to laser pulse length. If calibrated correctly only one laser pulse will
occur while mirrors rest on one spot.
i) Mirror 3: this mirror is used to direct the beam into the chamber window
j) Substrate: mounted close to the focus plane, the substrate receives high intensities of the
focused beam. Nevertheless it is mounted slightly behind focus so that the beam gained some
width and its Gaussian shape can be used. Precise energy measurements (PEM100, LTB
Lasertechnik) and beam profiling (CinCam CMOS-1202, Cinogy) are also done in substrate plane
ex situ. In situ the chamber is closed and the intensity measurements are performed outside
of the growth chamber.
Lasing was performed in situ to destabilize the oxide layer. When the oxide is removed successfully,
growth is enabled and the existence of the growth layer is proof of the destabilization reaction. Due to
the in situ nature of the experiment direct observation of it is impossible. Even without the enclosed
system of the SSLPE chamber direct observation of the change in oxide would require dedicated x-ray
equipment and the laser would need to be removed and reinstalled at the x-ray measurement facility.
An alternative could have been a heat camera in order to directly observe heat up but again the space
around the chamber is very limited.
- 39 -
3. Materials and methods
Figure 12 Laser beam (purple) in a technical drawing of the optical system and in principle. (a) laser
source, (b) mirror 1, (c) iris diaphragm, (d) lens 1 (focus length f1 = 250 mm), (e) attenuator, (f) mirror 2,
(g) lens 2 (focus length f2 = 400 mm), (h) scan head, (i) mirror 3, (j) specimen, (k) electronics.
- 40 -
3. Materials and methods
For a better understanding of the interaction of laser beam and silicon a simulation was designed with
COMSOL Multiphysics® software package. COMSOL is a finite-element (FEM) based solver which uses
different physics-model packages which can be combined to solve specific problems. FEM in general is a
numerical approach to solve differential equations. For this work, the heat transfer module and
MUMPS [51] solver was used to model the heat up of the interaction volume and the release of the
heat into the stack. The stack consists of 200 nm of a-Si with a 2 nm SiO
x
layer covering it. Several
educated simplifications can be made to make the model and the simulation less demanding for the
hardware and processing time. The simulations were created with Stefan Kayser and Christian Ehlers
and published with extensive description [41], [45]. The presented case consists of a small simulated
element with a radius of 2 nm, and for the purpose of the simulation is heated up by one discrete pulse
energy within the energy range of the experiment and in with the energy is not distributed laterally (r
direction), but only vertically (z-direction). The assumption are:
·The element has to have a small radius compared to the Gaussian width
·Heat is induced vertically and also lead away only vertically. No heat diffusion occurs laterally
·Heat is not induced in the SiO
x
volume, since it is virtually transparent to the laser light
·The temperature of the lower interface of the a-Si cylinder is stable at ambient temperature
With the described assumptions, the case becomes quasi-1-dimensional, and the simulation can be
simplified to sweep through energies depending on the point in the radius of the Gaussian energy
profile. Two reactions have been implemented in the model, both with an onset range of 10 K.
1. The melting and solidification of a-Si at T = 1147 °C
2. The evaporation of SiO
x
at 750 °C
Over the course of this work very different beam shapes and energy densities have been applied, in
combinations with differently grown and matured SiO
x
layers. Therefore simulations will be presented
in a more qualitative manner to understand how parameters influence heating, evaporation and
entually growth. In regard to the practical results achieved, spot sizes of 30 µm to 200 µm and pulse
energies of 0.9 µJ to 11.0 µJ are presented.
- 41 -
3. Materials and methods
Table 3 material parameters used in the simulation
Property
amorphous silicon
Density
g/cm
3
2.2 [70]
Melting
temperature
°C
1147 [71]
Thermal
conductivity
W/(m K)
2 [72]
Specific heat
J/(g K)
0.865 [70]
Latent heat
J/g
1320 [73]
liquid silicon 2.52 [70] 44.6 [74] 1.1 [70]
3.3.Substrates
Amorphous Si on glass is the mayor substrate for SSLPE as applied here. Glass in general is known as
excellent high temperature material, and for chemical, corrosion and erosion resistance. It serves a
number of functions during processing and in the final layer stack. Due to its high performance in the
relevant categories, Corning Eagle XG® glass was chosen (see Tab
le
4
).
Table 4: Properties of Corning Eagle XG glass® as used in for this work. The high-performance float
material inhibits great chemical and thermal resistance. Additionally, its thermal expansion comes
close to that of Si and thus minimizes thermal stress generation between layers
property
Mechanical temperature
resistance
Chemical resistance
Thermal expansion
Roughness
Thickness
parameter
Softening Point (10
7.6
poise)
Annealing Point (10
13
poise)
Strain Point (10
14.5
poise)
HCI – 5% (24 h; 95 °C)
HNO
3
– 1M (24 h; 95 °C)
HF – 10% (20 min; 20 °C)
25 °C – 675 °C
R
q
d
value
971 °C
722 °C
669 °C
0.79 mg/cm²
0.49 mg/cm²
5.18 mg/cm²
3.55 x 10
-6
/K
41,3 nm
0,7 nm
source
[75]
[75]
[75]
Bruker
Dimension Icon
[76]
- 42 -
3. Materials and methods
In regard to process parameters described later, the specifications were identified as follows: The
substrate is heated and held at 600 °C for hours while held in a rigid graphite holder. Due to the
expansion this leads to thermally induced stress. If the strain point of the material would be exceeded,
the internal stress would release at high temperatures by material creep and plastic deformation, but
then lead to stress and deformation when cooling down. Its thermal expansion α conforms to the one of
Si which exhibits a thermal expansion α between 2.55 × 10
-6
/K (at 25 °C) and 4.26 × 10
-6
/K (at 727 °C) [77].
This is important since the substrate stack undergoes various temperature changes of different speeds
and vastly differing coefficients of the involved materials would lead to delamination. The glass comes
in contact with HF during oxide removal and with Aqua Regia (HCl + HNO
3
) during post processing.
A generally low chemical resistance would lead to mechanical instabilities, but could also lead to
chemical impurities being released from the glass and integrated into the silicon.
The seed layer preparation takes places in a custom physical vapor deposition (PVD) chamber at
vacuum pressures of 5 x 10
-5
mbar to 5 x 10
-7
mbar constructed build at IKZ. As target, 6N Si chips in a
molybdenum crucible are used and evaporated by an electron beam (EVM-5 from AP+T) at a constant
voltage of 8 kV. The beam current is manually controlled between 150 mA and 215 mA to ensure
comparatively constant growth rates within one deposition run around 1.6 nm/s. Growth rates are
measured by a quartz crystal sensor and an Inficon SQM-160 thin film deposition monitor. Manual
operation is necessary since final thicknesses of 1 to 3 µm require operation of target and crucible at
critical temperatures over long process times (around 20 min), making attentive and educated
observation necessary. To ensure that the melt pool and especially the electron beam does not interact
with the molybdenum, the electron beam has to be operated carefully and adapted to changes in the
melt or target volume. Another reason for manual operation is sensitivity of the deposition system to
circumstances in the laboratory which are hard to control and lead to composition and structure of the Si
target. The crucible is used excessively to achieve thick layers, which exhausts the limitations of the
system. As a result, the volume of the residual melt varies, and with this the geometry of the residue as
well. SiO
X
on the residual Si from a deposition run leads to discontinuous melting in the next run. Used
Si is restocked between process runs by feeding Si chips into the cavity. As a consequence of the
combination of factors, melting the Si in the crucible varies between runs and the body of the Si which is
heated for evaporation is different every run. These problems are compounded by the fact that the
humidity in the laboratory and the working pressure in the chamber have an interfering effect, too.
Overall, this leads to the fact that a given deposition rate can rarely be reproduced by re-setting the
same combination of voltage and current. Humidity and chamber pressure also affect which and how
many impurities are integrated. The specifics of the final amorphous layer are therefore not the same
- 43 -
3. Materials and methods
between processes. As shown in chapter 5, the specifics of its deposition can have vast influence on its
annealing behavior (compare Tab
le
5
). As this work reveals, dynamic crystallization of amorphous Si
has a major effect in SSLPE and could explain and solve several problems of reproducibility.
Table 5: Processing parameters during PVD growth of the seed layer. Parameters may vary dependent
on the functionality of the chamber and lab environment (pressure, heater lamps, evaporation rate)
while others are set from the operator to adjust or to reach certain experimental goals (electron beam
current and voltage, evaporation rate, final thickness of the layer)
property
Pressure
Electron beam current
Electron beam voltage
Evaporation rate
Heater Lamps
Final thickness of the layer
min
5 x 10
-7
mbar
150 mA
8 kV
0.5 nm / s
3 x 34 V
1 µm
max
5 x 10
-5
mbar
215 mA
8 kV
2.5 nm / s
4 x 34 V
3 µm
The strength and unique feature of SSLPE is the deposition of thick, crystalline layers on glass.
Nevertheless, SSLPE was performed on different kinds of substrates, including two monocrystalline
ones. Commercial wafers allow for the ideal deposition of monocrystalline, homoepitaxial layers. The
grown Si can then be used for electrical characterization. With the primarily pursued SSLPE on glass,
rough layers with highly varying thickness and stochastic crystal orientation are produced. This is
problematic for electrical measurements. Electrical contacts are hard to apply and a reliable
characterization requires defined and constant thickness. To characterize just the grown material,
commercial wafers are used as substrates. Since the SSLPE Si produced here is n-type, p-type wafers
have been used, so that a depletion zone between substrate and layer electrically isolates forms
between the two.
A monocrystalline substrate alternative which inhibit a commercial perspective are ultrathin Si foils,
reorganized porous Si (RPS). F
ig
u
re
13
shows the manufacturing steps of RPS substrates and their
application for SSLPE. The original method of preparation was demonstrated independently in 1998 in
Japan [78] and Germany [79]. Central to the process is etching of the upper layer of a conventional Si
wafer to make it porous. In an annealing step the mostly freestanding, monocrystalline foil forms and
- 44 -
3. Materials and methods
the underlying pores connect during annealing. Surface energies would be minimized when the surface
is fully closed and the pores are fully connected to one void underneath the foil. Substrates used in
this work were produced and provided by Interuniversity Microelectronics Centre (IMEC), Belgium.
Their process structures commercial wafers with a deep-UV photolithography step. A photomask of
open circles defines the profile after SF
6
/O
2
plasma etching, consisting of periodically arranged tubular
voids surrounded by walls. Subsequent annealing produces foils of 1 µm thickness, with some few
remaining pillars [80]–[82].
Due to their thickness of just 1 µm, epi-Si is not suitable as an absorber on its own. Sunlight requires
more material to be absorbed and their thickness does not provide adequate stability. Nevertheless
RPS substrates are a potent candidate for epitaxy. When a foil is reduced from its parent wafer that
wafer can be polished and reused for epi-Si processing. The advantage in regard to bulk processing is
the minimal loss of material per substrate or produced layer [17]. RPS substrates are commonly used in
chemical vapor deposition (CVD) epitaxy [17]. For our experiments, foils attached to their parent
wafer were used to provide mechanical stability during growth.
Thus, conventional wafers as substrate are pursued for their academic use, especially for electrical
measurement and the manufacture of prototype solar cells. Epi-RPS on the contrary are debated as
commercial substrates [17] and SSLPE could be a fast epitaxy method for their enlargement.
The presented solution growth process allows for a flexible epitaxy of crystalline silicon. Data
presented here is used for the assessment of two different substrates, Si(111) and RPS. Si(111)
produces a layer most comparable to conventional wafers and growth is well controllable. Since the
substrate is monocrystalline over a large area, epitaxy can take place unobstructed. The production by
float-zone melting also produces large bulks with low density of crystallographic defects. Thus during
epitaxy, few defects will be promoted into the SSLPE epilayer. These layers then can be used for the
assessment of the potential of the material.
- 45 -
3. Materials and methods
Figure 13: Preparation and usage of RPS substrates as presented in this work. A conventional Si(111)
wafer (a) is prepared by mask lithography and an array of holes is applied (b). By annealing, a
monocrystalline layer forms (c) to minimize surface energy and pores coalesce (d) for the same reason.
Eventually a freestanding foil of 1 µm is formed (e) The foil is then used as substrate in the SSLPE
process, while still attached to the parent wafer (f) and cut of (g). After separation, the foil can be used
for solar cell preparation and the substrate wafer is polished to be reused f or RPS preparation (h).
- 46 -
4. Characterization methods
4.
Charac
t
er
i
za
ti
on
me
t
hod
s
4.1.Scann
i
ng
E
l
ectron
M
i
croscopy
(
SEM
)
SEM is a versatile method to characterize surfaces. It uses a focused electron beam which scans a
rectangular matrix on the sample. Primary electrons enter the sample surface and interact with
electrons located there. These interactions have different physical origins and produce different
material contrasts. An in-depth introduction of SEM and related methods can be found in [83]. The
primary beam functions identically for the different methods, but different detectors allow to collect
different data:
·Secondary electrons (SE mode): primary electrons interact with electrons in the conduction
band and ionize them by impulse transmission. A detector located sideways in the chamber
collects them.
·Back scattered electrons (BSE mode): primary electrons enter the material and do not lose
their momentum but are scattered electrostatically by the positively charged atom centers.
Only electrons scattered roughly directed towards the incident beam and thus the sample
surface can leave the bulk again. Therefore the detector for back scattered electrons is located
close to or around the incidence beam.
·Characteristic x-ray: when primary electrons interact with lower energy bands by impulse
transfer and when the electrons become ionized, the resulting electron hole is filled by an
electron of higher energy. The energy difference is emitted as characteristic x-ray radiation.
Purposefully placed detectors in the chamber collect these secondary electrons and generates images
by scanning the surface with the primary beam and building up images dot by dot. The primary beam is
generated at a cathode and electromagnets are used for beam-shaping, utilizing Lorentz-force to
push electrons which are charged and moving rectangular to their flight path. Lorentz-force is used in
the electromagnetic lenses for focusing and in electromagnetic scanners for lateral movement. This
build up enables very flexible imaging with some very useful contrast and detection properties,
especially relevant to this work.
Contrast in SE mode is driven by the ability of the electrons to enter and to travel through the material, to
interact with the bulk. For the induced signal to be collected, being it secondary electrons, back
scattered electrons or x-ray-radiation. This leads to significantly higher signals from edges (F
ig
u
re
14
)
and to a medium contrast from different elements, and to a soft and thus hard to interpret contrast for
crystal orientation (F
ig
u
re
14c
) or contrast between crystalline and amorphous material of the
- 47 -
4. Characterization methods
same element. The specific interactions between electrons and material is described in the following,
with focus on how to interpret images, what imaging artifacts involved and what the limitations
concerning this research are:
·Electrical conductivity: Contrast depended on conductivity can be high, especially if electrical
isolating material is involved. The latter leads to electrostatic charging of the material and
consequently electrostatic reflection of the incoming beam. In this research, well-contacted Si
is sufficiently conductive and does not charge. Nevertheless substrate glass does charge and
hence is not observable in SEM. In some growth experiments the amorphous layer is resolved
partially and the growth layer is not contiguous. In these cases it is barely possible to produce
SEM-images of the surface (F
ig
u
re
14
d). Highly conducting material transports electrons away
from the surface fast and appears darker.
·Edge contrast: Since the signal is strongly depended on the electron’s path length through the
investigated material before emission, edges, vertices and especially fins produce intense
signal in images and appear bright (F
ig
u
re
14
b and c). At these elevated and exposed sights
electrons face a heavily increased surface to bulk ratio. While traveling chaotically through the
bulk, their chance to interact with the surface and to be ejected back into the chamber is much
higher. For the rough surfaces in this work, overrepresentation is to be considered during
microscopy and analysis of images, but does not lead to major problems.
·Element contrast: the most important elemental contrast in this work is the one between Sn
and silicon. Due to the high conductivity of the metal Sn is dark in images and
well-distinguishable from silicon.
·Orientation / Structure: hard to interpret in the given material system of amorphous and
crystalline silicon, but noticeable in a number of circumstances. The laser-induced
crystallization and manipulation of the amorphous seed leads to visible contrast in the laser-
spots. Remarkably, that contrast seems to be also depended on the angle of the electron beam
(figure 10a). Another effect of interest is a contrast between different crystal orientations
making (111) facets darker than other, possibly due to better conduction away from the
surface (F
ig
u
re
14
d)
·Surface angle in relation to the detector: The electron detector can be installed in different
positions in a SEM chamber. As in the employed microscope it is often positioned to the side
of the chamber. When electrons leave the surface, electrons leaving facets facing the detector
will be collected more likely and those planes appear bright. When facing sideways or even
away they will appear darker. To the observers eye this effect is easily readable, since it
- 48 -
4. Characterization methods
correlates optically to a surface illuminated from the detectors position and observed from the
primary beam’s direction (F
ig
u
re
14
b). Also the incident angle of the beam can have an effect
(F
ig
u
re
14
a)
A specialty of SEM is the high depth of focus especially compared to light microscopy. The long focus
distance makes the beam narrow over a large distance. Since the sharpness of the image is directly
dependent on the width of the interaction spot, even rough surface can be characterized and imaged
properly. In the case of polycrystalline Si on glass, a surface could not be captured with light
microscopy. In light microscopy the focal depth is narrow and without image processing like focus
stacking the morphology is not to understand for the observer’s eye. Additionally the great number of
facets of the crystallites projects refracted light and disturb the image.
- 49 -
4. Characterization methods
Figure 14 Different imaging artifacts in SEM as they occurred in this research. a) laser spots on a Si(111)
surface. The incidence angle of the beam leads to dark-to-light gradient of the untreated sample, and
an inverse one in the laser treated areas. b) (100)-oriented Si crystallite. The edges offer more
possibilities for the electrons to escape the bulk, making the edges appear brighter. c) plate-like and
stochastic morphology. The (111) surfaces (top right) appear much darker than surfaces orientated
differently. d) seed layer after melt-back with exposed substrate glass. Since glass is not conducting
electrons away from their incidence point it becomes charged. Further incoming electrons are
reflected, leading to bright artifacts with an optical effect appearing like light reflection on bend
surfaces.
- 50 -
4. Characterization methods
SEM images of polycrystalline SSLPE layers are difficult to use for automated image analysis.
Crystallites exhibit varying numbers of facets and depending on their particular angle and the shading of
the facets vary. An optical counting method was adapted from [84]–[86] (F
ig
u
re
15
). To allow for
objective counting and to minimize human error, a square grid is overlaid on the SEM image in
question. To obtain the number of crystallites which are included in its area on average, it is neither
adequate to just count fully included crystallites, nor just the number of crystallites which touch the
square. Instead, lines of inclusion and exclusion are designed purposefully (se
e
F
ig
u
re
15
a). Counted is
every particle which is fully included in the square or which just crosses the green line of inclusion.
Every particle which is not in the square or does cross the red line of exclusion is excluded. The method
ensures for every particle to be just counted despite analyzing larger areas in total by counting particles
in a number of counting tiles. In addition to the original method, small, parasitic crystallites of less than 5
µm have been indicated as well (see purple dots in F
ig
u
re
15
a). They exclusively appear between
large crystallites on exposed seed layer. By counting every particle over a grid of squares (25 squares of
25 µm x 25 µm), this method also shows how much crystallite density varies between samples and
serves as measure for homogeneity.
Figure 15 (a) number of visible crystallites within a square of 25 µm after oxide growth. Substrates
where HF-etched and passivated. Afterwards, time for oxide growth was given. Particle counting was
done manually according to [84]–[86]. (b) green bars represent fully countable particles, red bars
represent excluded ones, and purple bars represent crystallites which were identified as parasitic
particles with a size underneath 5 µm. Lines in (a) represent the inclusion (green) and exclusion (red)
lines. Indication and counting was performed manually over a 5x5 grid and the ranges of data found
are shown in b). a) shows a detail of F
ig
u
re
25c
, growth after 15 min additional oxide growth
- 51 -
4. Characterization methods
4.2.Transm
i
ss
i
on
E
l
ectron
M
i
croscopy
(
TEM
)
TEM in general uses a high-energetic electron beam. The samples of well underneath 1 µm thickness
are mostly transparent to the beam and the largest fraction of the beam just passes through it.
Nevertheless some part of the beam is scattered by periodic objects, in our case atoms. As in a
transmitted light microscopy the operator can chose to observe the beam not scattered (bright field) or
the scattered beam (dark field). TEM uses electromagnets as lenses since they are able to bend the
electron beam via the Lorentz force. This makes beam shaping much more flexible than with optical
lenses in light microscopes and the complete beam path can be changed by adjusting the flux of the
electromagnets. This to use is one of TEM´s greatest strengths. The imaging beam system, so the part of
the beam path which collects the beam behind the sample and focusses it onto the detector, can be
switched to either display the virtual image of the sample (the image plane) or to display the virtual
image of the diffraction pattern (the diffraction plane). No parts of the setup of the microscope like
sample or lens position has to be changed for this operation and it can be performed arbitrary during
analysis. By selecting an area in the image plane with an aperture, the operator can extract the
scattering information of just this area (Selective Area Diffraction). Vice versa one or several reflexes in
the diffraction plane can be selected. When switched to imaging mode, areas which originate these
reflexes will light up and can be analyzed. This also enables imaging of dislocations. The high
acceleration voltage of the used FEI Titan 80-300 up to 300 kV creates electrons with a low wavelength
and thus a high resolution and can resolve rows of atoms. It also has an aberration correction to
eliminate this artifact, an operation which is much more complicated with electromagnetic lenses than it
would be with optical ones. An important note are the typical diffraction patterns of amorphous,
polycrystalline and monocrystalline material. The latter shows a distinct, singular pattern of reflexes
depending on the original crystal´s lattice planes. Amorphous material has no lattice planes and the
pattern is only depended on the distance of the atoms, which varies, and the resulting pattern consists of
rings without sharp intensity changes. Additionally amorphous material does not scatter electrons as
much as crystalline does. Polycrystalline patterns consist of rotated monocrystalline diffraction
patterns and thus of sharp rings, or if few crystals are involved, distinctive reflexes along several circles.
One problem with TEM is the tedious sample preparation. Samples are first cut and glued into blocks
of few millimeters diameter. Then they are polished down to few micrometers thickness and again
glued onto a sample holder. Preparation of samples can take up to one week and even experienced
technicians can damage or even lose a sample during preparation making TEM less suited for broad
screening of samples. Nevertheless in situ observation of the interaction of catalyst metal and a
- 52 -
4. Characterization methods
semiconductor was presented to great success with rows of gallium-arsenide precipitating in gold [87].
The experiment presented in the following uses a heater chip to observe crystallization in situ with the
phase transition being revealed by diffraction effects in imagining mode as well as with diffraction
patterns.
Figure 16: Setup of the in situ TEM experiment depicted as a sketch (a) and with light microscopy (b).
After annealing, the former a-Si is fully crystallized as visible in bright field TEM (c).
To achieve this several growth runs have been conducted with identical conditions, just the time of
extraction has been altered to freeze the growth of the SSLPE layer in intermediate stages. The seed
layer of 2 µm amorphous Si was deposited at a rate of 1.6 nm/s, in vacuum of 2.9∙10
-6
mbar. The
electron beam was operated at 8 kV and 190 mA. The stored substrates were treated with 4 %
hydrofluoric acid solution and rinsed in de-ionized water afterwards. SSLPE was pursued at 600°C /
605°C and substrates were inserted into Sn shortly after the temperatures were set to growth mode.
The processes were conducted regularly, but samples were extracted after 1 s, 30 s, 1 min, 5
min, 10 min and 1 h. Sn residue was removed by aqua regia etching. Images were obtained by SEM and
morphology features were colorized manually for better readability.
For an in situ observation of the interaction of amorphous silicon and liquid tin, 2 µm of a-Si was
deposited via electron beam PVD on glass, just like described in the growth method chapter. After HF-
dipping Sn was also deposited onto the amorphous silicon, but evaporated by resistance heating.
Resistance heating has low deposition rates and it is most likely that hydrogen-passivation was not
stable for the entirety of the deposition and oxide did grow back. Sn formed differently sized droplets,
with maximum thicknesses of 1 µm but some areas are not covered. The stack was then glued face-to-
face and cut down to perform interface preparation. For resin hardening, temperatures were kept as
- 53 -
4. Characterization methods
low as 150 °C so that the amorphous material would not be damaged or pre-annealed. Initial thinning by
plan parallel mechanical polishing was performed down to 10 µm thickness. Further thinning was done
in a PIPS Gatan Inc. system by argon-ion milling with liquid nitrogen cooling. In a last step of focused
ion beam manipulation performed with a FEI Nova 600 NanoLab DualBeam™ SEM/FIB, a section of
30 by 40 µm including all the layers was extracted and placed on a Fusion Select heating cell of
Protochips, Inc. With the experiment it is now possible to observe the interaction of the materials
when heated. To ensure a maximum of image information extracted despite the relatively thick sample
and highly varying material contrast, imaging was performed in bright field mode with a minimum of
aperture applied to the beam.
4.3.Raman
spectroscopy
Silicon is found in two different configurations in SSLPE: amorphous in the seed layer and crystalline in
both seed and growth layer. As reported before with TEM imaging, the seed crystallizes during SSLPE
growth [22]. TEM is an excellent method to reveal single states of a morphology in detail, up to the
atomic level and with vast insight into its crystalline structure. Nevertheless TEM preparation and
imaging is tedious and consumes the sample, making it inadequate to investigate a sequence of
experiments.
As discussed before, the seed layer is central in the growth model of the entire process but some
findings or assumptions are not in agreement:
a) nanoscopic seed crystallites with an areal density of 1 µm
-2
within the seed are suspected to
initiate growth of macroscopic crystallites in the Sn solution [30] p.43. Unfortunately these
crystallites were not always found in the seed layers
b) crystallite densities in the seed layer do not correlate to the density in the growth layer
c) extensive pre-crystallization of the layer leads to worse growth results, a phenomenon
observed with annealing of the seed in example during high energy laser treatment
d) some morphology phenomena where not explained by the established growth model, but
could be explained by dynamic crystallization of the seed still taking place during growth
In summary the growth of crystals cannot necessarily originate from preexisting crystallites (a), and
even if they would there must an interaction between seed and growth layer which prevents every
root leading to a SSLPE crystallite (b). Also, there was no explanation for the negative effect of pre-
crystallization (c). For a holistic growth model the dynamic of the seed`s crystallization must be
investigated and described precisely. A special requirement for the method is a large-scale, lateral
- 54 -
4. Characterization methods
acquisition of the layer (several 100 µm
2
) while still offering resolution or at least sensitivity in the nm
scale.
Candidate methods for observation could be
·TEM could differentiate amorphous and crystalline phase. As explained sample preparation
and observation takes approximately one week in total. Complication during preparation could
lead to even further extension of characterization. Samples and especially the areas of interest
are lost. Preparation of a large lateral sample area would be at least complicated if not even
impossible.
·Light-microscopy with transmitted, polarized light could reveal crystallized areas and would
be applicable as large-scale observation tool. Unfortunately the seed is embedded between
glass and SSLPE layer, and the polycrystalline Si would obscure the seed. Removing the growth
layer would be possible but would require difficile polishing and too much material could be
removed easily and destroy the seed layer. Even if possible, preparation would be again time-
consuming and samples would be lost for additional characterization. Thus, this method would
likely be more adequate than TEM, but still not suitable for the observation of a high number of
successive morphology states
·Raman spectroscopy with the extensive optimizations presented here allows for the
measurement of the embedded seed layer without any impairment of the sample. The data
processing allows for measurement times of 1 s, which makes large-area mapping possible.
Measurement points are actuated with the motorized sample holder, so mapping allows to
capture large areas of 1x1 mm or larger. Minimal resolution can be assumed to be roughly the
wavelength of the used light of 633 nm.
Raman spectroscopy uses the asymmetric oscillation of the electron cloud of molecules, the vibrations
of the whole lattice in crystals or of bonds in amorphous solids. The small resulting energy difference
due to deteriorations of the perfect solid is detected in the scattered laser light as a so-called Raman
shift. The special behavior of the underlying effect can now be used for analyses. Examples are the
chemical composition of crystals of silicon and germanium, phase analysis or the detection of
electrically active defects in semiconductors. Overall Raman spectroscopy can be used to estimate
crystalline quality. Nearly perfect lattices, i.e. extended to infinity and with minimal disturbances,
produce sharp peaks. If the lattice is disturbed, less sharp signals appear. Small crystallites show a blue
shift because the lattice function is not fully developed and lattice expansion occurs in these
crystallites. In silicon, this effect can be observed below 10 nm [88], [89].
- 55 -
4. Characterization methods
In this work Raman spectroscopy is used for the differentiation of two silicon phases; amorphous and
crystalline. The two spectra are very different. a-Si is characterized by the point group of the silicon
and thus primarily by the atomic spacing. Since the atomic distance varies in an amorphous phase,
oscillations between the atoms vary and a broad band forms with a maximum at 480 cm
-1
. The
crystalline signal on the other hand is characterized by the space group. In distinct crystals, this results in
a sharp signal at 520 cm
-1
. Due to the clearly distinguishable positions and expressions, the two
phases are easily distinguishable.
In order to comparatively detect the crystallization in larger areas of the seed layer a targeted mapping
method is developed and presented in the following. It allows to record Raman spectra of the seed
layer through the substrate glass as well. Some difficulties arise in this process. Depending on the
Raman signal strength, measurement times of several minutes or hours are usually used to record
precise signals. For mapping, this is not an option. With the used measurement matrix of 100 x 100
measurement points it is necessary to minimize the measurement time for single points. A long total
measurement time is not only unfavorable for the daily laboratory routine, but also bears the risk that
the device produces less comparable results, for example due to temperature fluctuations. This is
aggravated by signal degradation due to the passed substrate glass. For spectrography, a He-Ne laser
with a wavelength λ = 633 nm is used. A hundredfold objective for the near-infrared range with a focus
length of 4.7 mm allows focusing through the glass. Due to the confocal design of the spectrometer
the high magnification thereby provides a greatly improved signal yield as the lens system both
provides high source intensity and selectively returns signal from the focal point to the detector. The
laser light used is absorbed in silicon within 1 µm.
Despite of these measures, the signal-to-noise ratio is still unfavorable (F
ig
u
re
17
). At the targeted
measurement time of 1 s, the amorphous band cannot be distinguished from a vacuum measurement.
Even if the crystalline signal turns out to be significantly higher overall and thus easier to detect,
medium strength signals are often masked by the background. Thus a simple approach to quantify or
represent its intensity is not possible, for example by just using signal height. Also stronger peaks can
be misrepresented when a blue-shift deviates the highest point of the peak from the observed
wavenumber when just comparing a single position at λ = 520 cm
-2
. This is the case when nanoscopic
crystals aggregate and a large volume is crystalline but the lattice function of the crystallites cannot
develop fully. An alternative is to fit the spectrum using mathematical distribution curves. Usually two or
more Gaussian distributions are used [88]–[91]. This allows to take different Raman modes of the
same phase into account which produce different, but too closely positioned peaks to be identified.
- 56 -
2ᵰ
4. Characterization methods
Without fitting they might appear as one peak and would deteriorate the analysis of the data. Fitting
can also be used to separate the amorphous signal from the crystalline or vice versa, if both appear in a
single measurement. When fitting the weak signal of 1 s dwell time, the main focus is to measure the
peak position and strength of the sharp crystalline signal. In addition, since the amorphous signal is
too weak and thus omitted only the following Gaussian distribution is used for the region around
520 cm
-1
(F
ig
u
re
18
).
ᵃ(ᵆ) = ᵃ
0
+ ᵃ · exp (−1
(
ᵅ
ᵅ
ᵅ
− ᵅ
0
)
2
), ( 27 )
with the signal height A, the characteristic signal of crystalline silicon k
cr
at 520 cm
-1
, the deviation of
the actual peak x
0
and the standard deviation σ. The fitting is able to find values even in very noisy
spectrograms and weak signals. The algorithm for point-by-point fitting of the peaks was created in
close collaboration with Stefan Kayser. Optimization of the measurement itself were made possible by
the expertise and extensive help of Andreas Fiedler.
- 57 -
4. Characterization methods
Figure 17: Raman spectra of amorphous and crystallized silicon. The darker lines show the spectra after
dwell times of 8 min, the lighter lines of 5 s and the lightest after just 1 s. For dwell times of 1 s, blue and
red are not distinguishable.
Figure 18: Exemplary raw Raman data (dwell time 1 s) and the Gaussian fit used for mapping.
- 58 -
4. Characterization methods
The Gaussian distribution offers different possibilities to measure and visualize "crystallinity" (F
ig
u
re
19A
lready with the use of height A, the spreading of crystallites in the amorphous matrix can be
noticed (F
ig
u
re
19
a). For example it can already be seen that nucleation centers are not distributed in a
finely dispersed manner, but that dendrite-like structures are present as drivers of crystallization.
However a disadvantage of proceeding this way is the z-dependence of A; crystallites in the focal plane
provide strong signals. However if crystallites are no longer in focus their signal’s strength decreases
rapidly. Thus the signal is primarily dependent on the z-position, and thus easily loses general
information about crystals present. An alternative is to use the width of the Gaussian curve
represented by the standard deviation σ (F
ig
u
re
19
b). By the nature of the distribution, it is already
normalized to the area under the curve. This allows to read information that unlike height A does not
depend on the overall strength of the signal but depends very directly on the crystal properties.
Extremely narrow signal peaks with σ < 0.6 cm
-1
can be neglected, since they correspond to twice the
pitch width of the optical grating and thus to the fit of a single measurement point recording noise.
Signals just above this value correspond to crystals with high perfection and size. If crystallinity or size
decrease the signal widens. This is accompanied by a blue shift of the characteristic peak at 520 cm
-1
.
The plot of the standard deviation σ reveals a much larger spread of dendrites than observable with
peak height A. This means that the fit also takes into account very flat signals. For an analysis this
method observes both
·crystals outside the focal plane, which would have a strong native signal, but are only weakly
detected by the measurement, and
·early stages of crystallization, before a clear sharp peak appears in the graph but instead a
recognizably broad peak can be analyzed
Thus, the standard deviation σ gives a good indication of the position of crystal structures but does not
provide an intuitively comprehensible differentiation within the affected area. As described, signals
are mapped similarly, independent of origin strength or measurement quality.
To take advantage of both criteria, a quality factor A/σ is introduced (F
ig
u
re
18c
). This means for the
different cases:
·High crystalline quality and crystallites are in focus: characterized by high A and small σ,
subsequently A/σ is large.
·High crystalline quality and crystallites are not in focus: characterized by small A and small σ,
subsequently A/σ takes medium values
- 59 -
4. Characterization methods
·Low crystalline quality and crystallites are in focus: characterized by small A and medium σ,
subsequently A/σ takes small values
·Low crystalline quality and crystallites are not in focus: not detectable.
In the case of incomplete crystallization of the amorphous layer, intermediate stages are thus
displayed particularly well. At the same time, areas with advanced crystallization are also displayed
particularly clearly. The quality factor A/σ is therefore used for other maps shown in this work. The
setup of the growth experiment and the surface morphology is described in chapter 3.3 and the line of
experiments and the produced sample are the ones presented in this chapter. The mapping
algorithm was developed in close collaboration with Stephan Kayser.
Figure 19: Different approaches for plotting of the Gaussian fits. Just plotting the peak height
overemphasizes strong signals and thus z-position of the crystals (a). Using the width of the fit
represented by the standard deviation σ reveals more crystalline areas, but no sufficient contrast
between them is achieved (b). Best morphology contrast is given by a quotient of the two, combining
general crystal location with their z-position.
4.4.E
l
ectr
i
ca
l
propert
i
es
Essential for solar absorber material is its majority carrier type, and the concentration and mobility of
charge carriers. All of which can be calculated from the areal resistivity, which is measured in the van-
der-Pauw setup [100]. Ohmic contacts were applied with indium-gallium paste and measurements
were performed with an in-house build system. For the measurement of the Hall Effect, a temperature
range of 20 K to 350 K was established with a nitrogen cryostat. The thickness of the tested growth
layers was 150 µm on the Si(111) substrate and 40 µm on the RPS(111) substrate respectively. For the
confirmation of conductivity type, electro-paramagnetic resonance (EPR) was used. With the analysis
of spectra, specific dopants are isolated and are used further for the simulation of an ideal crystal.
- 60 -
4. Characterization methods
As a method to measure minority carrier lifetimes, microwave detected photoconductivity (MDP) is
used. Charge carriers are injected at low rates by laser illumination and remnant voltage is measured
over time. The transient reveals minority carrier lifetime and defect emission. Voltage is measured
with a microwave resonator. To identify minority carrier traps, they are saturated with applied voltage in
comparative measurements, so decreased signal can be traced back to saturated traps. MDP`s
penetration depth can be adjusted so it is possible to selectively characterize homoepitaxial layers
without measuring their substrates in the same time. Mainly bulk lifetime is measured, while surface
states have a low influence or can be extracted from the transients [101]. Additionally, surfaces were
passivated with iodine/alcohol solution (1 g/100 ml). The fast method allows for mapping and thus
provides information of a material`s homogeneity.
With MDP, deep level defects are also identified as charge carriers, since they also store charge under
illumination, release it afterwards and contribute to life-time signals [102]. Nevertheless, they act as
traps for the charge carriers and therefore must be avoided. In solar cells, they cause vast loss of
efficiency [103]. To identify and eventually understand how deep level effects are introduced into the
material, DLTS was performed with a PhysTech FT1030 DLTS to characterize these defects.
Measurements were done at temperatures from 50 K to 350 K and signals were obtained at intervals of
10 ms, 20 ms, 200 ms and 2000 ms. DLTS has a detection limit for this setup of 10
12
cm
-3
impurities
[104]. DLTS measures the capacitance generated by crystal defects and dopants underneath the
material’s Fermi-level. A Schottky contact induces a zone of saturated defects. Traps are then saturated
with an electrical pulse and voltage transients are observed. Since the defects show varying activity at
different temperatures, their activation energy can be extracted from the transients. For this method,
temperature transients are Fourier transformed so that overlapping peaks can be resolved and
additional data for Arrhenius plotting is obtained [105]. Schottky contacts were applied with platinum.
Electrical measurements were conducted by and with Julian Stöver and Christian Ehlers and are already
published in [30] and compiled here for analysis.
Solar cells were produced by functionalizing one side of the substrate as shown in F
ig
u
re
20
. Starting
with a commercial Si wafer (a), a SSLPE layer is grown and RCA cleaned (b). A heterojunction is created in
the form of amorphous p-doped a-Si (Si:H(i,p
+
)) by plasma enhanced chemical vapor deposition
(PECVD) under low pressure. TCO is deposited by sputtering. A polymer spacer made from polyimide
heat resistant up to 350 °C (e) is applied and the Si:H(i,p
+
)-TCO stack wet chemically removed from
areas not covered by the spacer (f). Afterwards, a titanium-aluminum stack is evaporated and forms
the metallization (g). As final step, the spacer is removed together with the metal still covering the
- 61 -
4. Characterization methods
TCO (h). A number of 28 rectangular solar cells were produced with sizes varying from 0.270 cm² to
0.506 cm². Initial assessment of their quality was done with the Suns-V
OC
characterization technique
[106], which uses a light-pulse from a xenon lamp and correlates the subsequent open-circuit voltage
V
OC
to the declining light intensity. As a result, a pseudo-efficiency and a pseudo-fill factor are
calculated with the assumption of a short-circuit current J
SC
according to electrical values. Work
functions of the most promising cells were recorded with a sun simulator (Wacom WXS-156S-L2) with
AAA characteristics under AM1.5 spectrum and 1000 Wm
−2
and V
OC
, J
SC
, and FF values are extracted. A
set-up build in-house at HZB measures internal and external quantum efficiency. The absorption
spectrum is measured with a PerkinElmer LAMBDA 1050 by reflection data.
Figure 20 processing of prototype solar cells. Onto a silicon substrate (a) SSLPE silicon is grown (b). The
emitter (c) is applied by CVD and the TCO contact (d) by sputtering. A spacer foil (e) prevents TCO and
emitter from etching (f). Metal contacts are applied by PVD (g) and surplus metal is removed together
with the polymer spacer (h)
- 62 -
5. Results and discussion
5.
Re
s
u
lts
and
d
is
cu
ssi
on
5.1.
I
nteract
i
on
of
seed
l
ayer
and
li
qu
i
d
t
i
n
The initial state before annealing is seen in figure 15a. In general, a number of residual droplets of the
resin used for preparation are visible which can obstruct precise observation. Nevertheless, the
morphology can be observed and also the diffraction patterns are not affected. The seed layer is
completely amorphous and does not show any of the suspected nanocrystalline areas. Also no
crystalline phase is detected by diffraction. The interface of a-Si and Sn is sharp and no initial damage to
the seed is visible. Nevertheless, the silicon surface appears not perfectly flat due to polishing which
removed some material. The Sn does not form a contiguous layer but partially connected droplets
(F
ig
u
re
21
).
Temperature experiments around the melting point of Sn T
M,Sn
= 232 °C confirm the sensitivity and
precision of the heater chip. Also, no relevant hysteresis was observed. Melting and solidification
occurred repeatedly at T
M,Sn
and thus no influence of the small size to the melting temperature of the
droplets is assumed. The fast phase switching also confirms quick total heating of the sample and the
steep temperature adjustment of the heater chip. As an important result it can be stated that the
droplets were keeping their shape and no movement appeared, either laterally or onto the sample fin.
As the only contact between silicon and Sn the initial interface remains. Diffraction patterns of Sn
during temperature hysteresis reveal that Sn crystallizes always in the same configuration, but in
different orientations.
During annealing at 600 °C loss of Sn was observed, most likely due to evaporation into the TEM
chamber vacuum. Nevertheless, silicon and resin remained stable. Initial crystallization of silicon was
visible after an incubation time of 5 min, with first crystallites being observed at the Si-Sn interface.
From the same interface dendritic growth is initiated into the seed. Full crystallization is reached after
additional 4 min. For the whole duration, no changes were observed, neither for the interface Si-tin,
nor for the Si-vacuum surface.
- 63 -
5. Results and discussion
Figure 21: TEM bright field images of the crystallization during the in situ experiment. Already
crystallized seed is colored red for visibility. Crystallization initializes at the interface tin-silicon after an
incubation time of 5 min.
The in situ TEM experiment confirms the catalytic effect of liquid Sn on the crystallization reaction of
a-Si. This statement is qualified first by the start of the crystallization at the interface of tin-silicon. This
is remarkable since crystallization onset could be just be associated with the surface of the silicon fin,
but if it would be, crystallization could also start in the former bulk of the seed. Crystallization then
continues into the bulk. No clear crystal orientation is visible from diffraction and mostly weak
crystalline diffraction patterns are recorded which are arbitrarily rotated (F
ig
u
re
21
). Notably there is
one stronger crystalline signal visible, which means that the crystallized volume has a preferred
orientation. The preferred growth orientation should be [100], since it has the highest surface energy.
Since the bulk is now prepared as a fin, all visible area is close to a surface, but no crystallization starts
within. Comparing these results to data from literature (compar
e
Tab
le
2
), it is obvious that incubation
time and crystallization speed are reduced both. This underlines the catalytic effect of the Sn towards
the crystallization reaction, and possible effects will be discussed in the following.
Notably there is no material shrinkage of the Si observed, or that the Sn advanves into the Si. Thus, an
etching effect as described in [68] or [110] is not taking place. During ALC, a-Si is solved in a indium
droplet while the droplet moves forward to the front where the solution reaction happens. The
dissolved atoms are precipitated as crystalline material at the trailing edge of the droplet on its
pathway. The material system Si-indium should behave similar to the system here, Si-tin. Nevertheless,
- 64 -
5. Results and discussion
an effect like ALC droplet movement is not observed here. The difference to the experiments
presented in [68] might arise from the following:
·In [68] a temperature of 320 °C was applied compared to the 600 °C here. The lower
temperature leads to less crystallization pressure and crystallization onset is spread to fewer
sites over the samples. At 600 °C the amorphous state is less stable and crystallization starts
at more points simultaneously. Thus more or all material crystallize immediately and no
advancement of the droplet towards residual amorphous material can be made.
·Since no advancement of one droplet is possible, no steady state of absorption and
precipitation is established. This also means that the droplet can just absorb silicon atoms until
its saturation. The low solvability of just 0.0885 at-% at a minimal solution volume leads to an
unnoticeable removal of silicon from the seed layer.
·According to [68], saturation mixture of silicon in indium is higher regarding an amorphous
source than a crystalline one, promoting crystalline precipitation. Since no crystalline seeds
are available, this state will remain until crystallization starts. After incubation the silicon at
the interface crystallize quickly and further influence of the droplet on remaining crystalline
material is negligible.
As described above, MIC is observed and described with solid metal catalysts. In contrary, diffraction
in the experiment presented here confirms that the Sn is fully molten during incubation and for the
full duration of crystallization. Without an epitaxial template, other mechanism which support
crystallization must be active. Since no chemical reaction is suspected to be involved, atomistic and
physical effects must be debated. As origin of the catalytic effect several microscopic processes are
possible:
·Mass transport: for a crystallite to be formed from atoms, a critical density of these atoms has to
be given, at least spontaneously. These atoms must coalesce in a point where nucleation is
possible, a probability that rises with higher mass transport. In MIC diffusion along grain
boundaries with their high density of defects improves diffusivity. With liquid tin, the diffusion
coefficient is higher than in bulk material. Additionally, convection in the Sn adds mass flow of
silicon. Also the volume of diffusive medium includes the whole droplet, a distinctly greater
volume than any grain boundary, which represents a two-dimensional object. In total, mass
transport should be much higher in liquid media than in solid ones.
· Adsorption effects: not only mass transport but also agglomeration of singular atoms can lead
to atom coalescence. The surface of the seed layer attracts atoms electrostatically. The newly
- 65 -
5. Results and discussion
formed layer is likely to show crystalline properties. If an initial crystallite is formed, it
influences atoms within the seed and can act as a blueprint for them.
·It must be noted that solving Si in Sn up to the saturation limit does not apply to many atoms.
Nevertheless, once solved in Sn, atoms are mobile and are much more likely to participate in
a nucleation reaction. If an unstable nucleus forms, atoms can complete its growth to a critical
size. In the moment a reaction starts, atoms are depleted in its vicinity and the improved mass
transport supports the reaction.
·Micro-etching on the atomic level can open up atomic bonds. Stable structures of silicon atoms
are more likely to prevail. This would especially include tetrahedrons excavated by micro-
etching. These could then act as nuclei for epitaxial growth. The melt-back process used to
destabilize the oxide etches the surface and is well established and it is very likely that
described micro-etching happens as a side effect.
·Fast heat transport: nucleation is an endothermic formation process. The energy needed
exclusively comes from heat. In the reaction chamber, heat is introduced to the seed layer by
radiation and convection from the top heater and thus the hydrogen-rich atmosphere (heat
diffusion through gas should be negligible), and from the heated Sn and thus through
convection and heat diffusion. Altogether, liquid Sn is the superior heat conductor compared
to hydrogen. Also, heat conduction in interfaces is superior for liquid-solid compared to
gaseous-solid. In the TEM chamber the atmosphere is vacuum, reducing heat transport even
more.
·In all cases, Sn is a better heat absorber then a-Si, and also a better heat conductor (Tab
le
6
).
Since heat is consequently collected and transmitted easily it serves as a heat exchange
system. In both the TEM and reactor chamber, Sn should act as a heat collector regarding the
heat sources and as a heat donator regarding the seed.
Concluding the in situ crystallization experiment proves the catalytic effect of the tin, with effects of
micro-etching, atom transport and heat collection being the most likely effects. Incubation and
reaction periods of the amorphous to crystalline reaction are reduced compared to literature. Due to
the low absolute solvability of the Sn droplets no notable silicon precipitation was observed.
- 66 -
5. Results and discussion
Table 6: heat conductivity coefficients influencing the crystallization speed in SSLPE
Material
Sn liquid
Sn solid
Si crystalline
Si amorphous
Heat conductivity coefficient
W/(mK)
35
67
150
3
Applicable temperature source
K
750 [107]
300 [108]
300 [108]
300 [109]
5.2.Deve
l
opment
of
the
surface
morpho
l
ogy
dur
i
ng
growth
Two morphologies have been observed over the time of this research. Most prominent and already
reported before is a morphology of randomly orientated and differently sized crystals [111], further
referred to as stochastic morphology or respectively stochastic crystallites (figure 11a). Clearly
distinguishable from that is a morphology consisting of plates larger than stochastic crystallites. These
plates form round patches and their [111] crystal facets are parallel to the substrate surface, resulting in
a generally more flat appearance (F
ig
u
re
22
b). Despite not representing one identical crystal,
vertices meet in the center and are parallel for several crystallites. The appearance described will be
referred to as plate-like morphology. Additionally, both morphologies have been observed in
combination in one sample. There fully developed circles of plate-like areas coexist, often fully
surrounded by stochastic growth layers (F
ig
u
re
22c
).
Figure 22: SEM images of different characteristic morphologies which develop during SSLPE growth of
Si. The stochastic orientation is most often observed (a). Additionally a plate-like morphology can be
observed (b), producing larger crystallites which form round spots with shared lateral orientation. The
latter can be observed best during bimodal growth (c), where the two major morphologies coexist
(colorized, magenta: plate-like, teal: stochastic).
- 67 -
5. Results and discussion
In order to understand the development of SSLPE layer and the whole growth system a series of
experiments was conducted. The novel multi-substrate holder allows for using virtually identical
substrates for growth and in combination with HF-dipping eliminates most of inconsistencies between
runs. A single substrate is taken from storage and etched before each run following the same
temperature program and general procedure. Growth then is interrupted after different processing
times, eventually revealing a number of consecutive states. F
ig
u
re
23
shows the series of SEM images
after different extraction times, which have been colorized to accentuate stochastic surface crystallites
(teal) and plate-like morphology (magenta). In the final morphology after 1 h (F
ig
u
re
23
a-6/b-6), a
multimodal SSLPE layer shows plate-like and stochastic surfaces simultaneously.
Immediately after growth starts, or respectively after first contact of Sn and seed, a high density of
small crystallites populates the surface densely (F
ig
u
re
23
b-1). Crystallites show no observable
preferred orientation. Within the first minute (F
ig
u
re
23
a-1/b-1), the number of crystallites decreases,
while their size increases.
From 1 min on, a secondary type of crystallite is visible, which is broader and flatter. Also, less empty
space is observable between them compared to the primary morphology. From 5 min on, the
secondary morphology is clearly recognizable as the plate-like morphology described before. At this
timestamp, it already exhibits the characteristic round shape with a seemingly shared
[111]-orientation. Nevertheless at 1 min, the round shape is not yet refined.
Figure 23: Colorized SEM images of the growth layer during SSLPE. The first row (a) shows an overview
and the second row (b) shows corresponding details. The columns (1 to 6) represent different
timestamps from 1s to 1 h as indicated. Three phases can be observed: surface nucleation and partial
dissolving (1 and 2), formation of plates and refining of stochastic areas (2 and 3), growth and
stabilization of the crystals formed before.
- 68 -
5. Results and discussion
The immersion of the substrate into the Sn solution results in a high number of crystallites on the
substrate’s surface. Not only are no nanocrystalline areas (roots) found in the seed layer, the density of
surface crystallites even supersedes the formerly reported density for roots in the seed of 1 µm
-1
[22],
contradicting the original theory that surface crystallites are solely originating from these
crystalline seeds (compare chapter
2
). Without an epitaxial nucleus of the surface crystallites, their
origin must be understood.
An investigation of the first 5 min shows a very high initial density, with crystallite numbers per area
decreasing quickly. This effect can only be explained by crystallites being re-dissolved into the tin.
Subsequently not all crystallites are stable in that phase of growth, but an effect of instability must
introduce the high number of crystallites in the beginning. Either way, mechanisms of supersaturation
have to be considered. At that point, the Sn melt has been held at working temperature for some
minutes and was heating over 30 min before, so saturation of Si in Sn is achieved. If the crystallites are
not originating in the seed, supersaturation leading to increased nucleation pressure is a central
candidate for the formation of crystallites. During dip-in the tin’s temperature profile is already set to
growth mode, with the Sn surface being colder and Si’s solvability being lower. Nevertheless directly
before growth and during heat-up, the profile was inverted and the surface was hotter with a higher
solvability. When inverting the direction of the temperature gradient, the inertia of the system keeps
the share of the solved Si atoms locally stable, while temperature changes fast. Additionally, former
studies have shown that temperature is not evenly pronounced in the Sn and especially laterally on
the tin’s surface. Instead simulations found mass flux vertices induced by the temperature difference.
The vertices naturally also lead to heat flux and an uneven temperature profile. Both effects combined
could very well lead to a supersaturation at the surface during insertion.
As one process improvement, substrates are kept at the coldest zones in the outer chamber, to
improve stability of their passivation during reactor heat-up. When dip-in approaches, the substrates
are moved into the inner chamber and are warmed up to reactor temperature. Nevertheless, a short
temperature drop of the Sn bath is visible during dip-in. It is well-possible that this temperature drop
leads to spontaneous nucleation in the beginning of growth.
Summarizing, it is likely that thermal fluctuations do exist due to different reasons, and that they
consequently cause supersaturation locally. Especially the beginning of growth leads to fluctuations
increasing these effects when the temperature gradient is inverted and the substrate transport
interfering with otherwise steady conditions.
- 69 -
5. Results and discussion
From 5 min on no further changes in the areal distribution of the two morphologies is observed, but
still crystallites increase in size. Further growth of crystallites is only natural since the supersaturation at
the surface established earlier does not change. Instead the crystallites have stabilized and will not
dissolve again, implying that they have reached a critical size and process conditions have stabilized. As
a consequence, the ongoing supersaturation leads to the growth of the crystallite which is
energetically favored over nucleation.
The plate-like morphology only establishes itself between 1 min and 5 min. They share a
(111)-orientation, and also their symmetry is 6-fold, with influence of some mosaicity. This strongly
indicates that one patch shares one crystalline origin. Two effects can be imagined to be the origin of
these plates (F
ig
u
re
24
). One hypothesis is that the origin lies in the seed (b-1 to b-3), which would
mean the seed crystallizing first from that spot and forming a circle already representing the later
surface patch. Afterwards this crystal embedded in the seed would grow out, resulting in the observed
morphology. As an alternative hypothesis, one surface nuclei would originate the surface patch, most
likely from its center. Afterwards the plates would grow over the seed laterally. Of the three low Miller
indexed facets (100), (110) and (111) the latter is the most stable one and promotes the enlargement of
that facet (
Tab
le
7
). Simultaneously (100) is the least stable facet and thus the preferred growth direction. This
perspective would indicate that flat surface patches are preferred due to their (111)-surface being the
energetically most stable. The flat surface crystallites would act as a seed for the underlying amorphous
material, promote its crystallization and the seed crystallite would inherit its orientation.
The question which of the two theories holds true is central to the growth process as a whole. Until
now, the evolution of the seed layer was a black box, with only the initial state (mostly or fully
amorphous) and the final state (fully poly-, or specifically micro-crystalline) being known, and the
theory was established that the seed acts just as a seed, which would be close to the theory as depicted
in F
ig
u
re
24
b. A more complex interaction where a dynamic interaction between SSLPE layer and seed
takes place over the whole growth would change that perspective. Therefore the evolution of the seed
layer will be investigated in the following chapters. Especially comparing surface and seed
morphologies offers an opportunity to understand that dynamic.
- 70 -
5. Results and discussion
Figure 24 two theories of the origin of plate-like crystallite patches. In either case, the patch must share a
uniting root. It only is of question a crystal nucleates out of the seed first (a-1) or if a root develops
laterally in the seed first (b-1) .In the first case, the crystallite on the surface would originate the patch
(a-2) and only afterwards crystallize the seed underneath. In the secondary case, a larger crystal in the
seed (b-2) would grow out and become pronounced as a surface patch later. Both cases result in similar
morphologies (a/b-3)
Table 7 Surface energies of the three lowest Miller indexed facets of Si [112]
facet energy
mJ / cm
2
(100) 21.3
(110) 15.1
(111) 12.3
5.3.
I
nf
l
uence
of
hydrof
l
uor
i
c
ac
i
d
and
regrown
ox
i
de
Identical substrates were prepared by simultaneous PVD-deposition on glass as described. Substrates
where stored and HF-dipped directly before growth. HF dipping was achieved with 4 % HF solution in
de-ionized water. The processing steps and waiting segments between HF-treatment were minimized,
as well as the heat budget by storing the samples in the outer chamber isolated from direct heat. This
means samples after dipping were first exposed to air for 5 min, then inserted into loadlock 1 which
- 71 -
5. Results and discussion
quickly applies a vacuum of 5 x 10-5 mbar within 5 min. Afterwards the samples are handled and are
thus stored in vacuum at 5 x 10-8 mbar. Afterwards H2-flooding commences and heating is initiated
when pressure supersedes 1 bar. Heating of the chamber takes 30 min. Afterwards the four samples
are inserted into the melt after varying times, ranging from 1 to 45 min. During waiting the samples
have a temperature of 70 °C.
From a first optical inspection of the SEM images (F
ig
u
re
25
) no clear trend concerning particle size,
density or other morphological features depending on the wait time is observable. The morphologies
are described in detail in the following:
a) 1 min: the crystallites appear dense, contiguous and homogenous. The density seems to be
the highest among the samples and crystallite size the overall smallest. Crystallites show
stochastic morphology.
b) 5 min: the crystallites appear dense, contiguous and homogenous. Crystallites are overall
larger than in a) and partially exhibit accentuated {111} facets. The latter gives the morphology
tendencies of the plate-like structures described, but overall the morphology can also be
described as stochastic.
c) 15 min: the crystallites do not form a contiguous layer and thus the seed layer is partially
exposed. On the seed, small, parasitic crystallites with diameters of less than 5 µm are located.
Some crystallites exhibit large {111} facets, consequentially giving the surface a plate-like
appearance. Overall the morphology can be described as a mixture of plate-like and stochastic.
d) 45 min: crystallites are not dense, contiguous nor homogeneous. The seed surface is exposed
and exhibits parasitic crystals. No enlarged {111} facets are visible and the existing crystallites
can be described as stochastic
Figure 25: SEM images of HF-treated samples. Substrates where HF-etched and passivated.
Afterwards, time for oxide growth was given.
- 72 -
5. Results and discussion
Summarizing, a) (1 min) and b) (2 min) show enclosed, homogenous growth. For c) (15 min) and d
(45 min), crystallite size varies more and larger areas exhibit bare seed surface which do not contain
large crystals. These bare areas do indeed contain small, parasitic crystallites of less than 5 µm.
These observations are confirmed in the data extracted from the SEM images (F
ig
u
re
15
b). Data for a)
and b) is similar, with the mean crystallite density being 7/625 µm
2
for both a) and b). The density of c is
the lowest with 4/625 µm
2
and intermediate with 5/625 µm
2
for d). Notably the density of parasitic
crystals is lowest for b, a number increasing heavily for c) and d), for which also the variation of density
per square is increasingly large.
As stated in the results section no clear trend is observed, neither with qualitative analysis of the SEM
images nor when using the data extracted from them. For example the crystalline density has its
minimum at 15 min, while shorter or longer oxide growth is leading to higher density. In engineering
science, problems in which the variation of one parameter causes two conflicting effects are well
known and they demand not maximizing the varied parameter, but the evaluation of its optimum
outcome. In the case presented here, the optimum is rather obvious as discussed later. What is not
obvious are the two contradicting effects. As central explanation the following model is proposed:
The H-passivation starts to decay immediately after dipping, but the decay commences slowly. The
sample is inserted into the vacuum and soon after into passivating hydrogen atmosphere. In the
waiting position, the sample has an elevated temperature of 70 °C which supports desorption of
hydrogen and growth of oxide. Thus samples which are inserted quicker have thinner and less stable
oxide. When they come in contact with the Sn solution, the contact itself is better and more sites on
the surface are suited for nucleation. This leads to a higher number of crystallites in the concerning
area, of which a high amount grows successfully. Samples which are inserted with a longer delay grow a
thicker and more stable oxide before insertion. When inserted, fewer nucleation sites are offered and
less crystallites form. When they grow out, they can span larger areas before they touch other
growing crystals and their lateral growth is stopped. For very long wait times the oxide itself hinders
the lateral growth additionally. Resulting open seed areas do not offer {111} facets which are the
energetically best surface for epitaxy. During growth incoming silicon atoms do not find sites for
epitaxy and eventually parasitic nuclei will form. Since this is a question of chance, they are formed
late during growth and stay small. As a result, also {111}-facets are pronounced less.
Former experiments have shown that these conditions are not suited to keep the substrates oxide
free. In these experiments, wait times of 30 to 60 min were used. Oxide prevention was usually
achieved by fast handling directly after PVD and in some experiments also by HF-preparation. With no
- 73 -
5. Results and discussion
other measures applied, these experiments lead to unsuccessful growth [41]. This is in agreement with
wait times over 15 min here, which also lead to unsatisfactory growth.
A possible reason for different growth results could be annealing of the substrate with elevated
temperature possibly starting or promoting a crystallization reaction. The substrate is held at 70 °C for
45 min in the longest case. No effect of annealing is reported at these settings (compare Tab
le
2
).
So what are the implications of the collected data? Firstly, potential and limitations of the method
must be evaluated, and what information can be extracted from the data. Desired properties of the
layer are: large crystals in combination with a homogenous and contiguous layer. Large crystals are
naturally signaled by a low crystallite density (low number of included crystallites, green bars). A low
number of crystallites per area could either stem from large crystallites, but also from a low areal
coverage with smaller crystallites. To differentiate those two cases, the data can be checked for a high
exclusion (red bars) to inclusion ratio. Mostly large crystallites will cross the counting grid and are
identified in that way. Also, a high number of parasitic crystals (purple bars) in combination with low
inclusion numbers indicate the absence of large crystals. Parasitic crystals are also a sufficient proxy
for a bare seed layer surface and should be avoided in general. Lastly, a homogenous layer is very
important for back end processing. This is indicated at first by low variation of any one of the three
parameters. Additionally it could be checked if meridian and mean diverge from each other but this is
not the case.
As a conclusion, it can be said that the 5 min sample reached the best results. It shows the lowest
amounts of parasitic crystals and it is very homogenous. Also it offers larger crystals than the 1 min
sample. The 1 min sample is another close candidate for a potential application, but it shows more
parasitic crystals and is less homogenous. Also, it has more small crystals. The large crystals of the 15
min sample are also of interest, but the large open seed areas are undesired.
Asides from the application at hand the author wants to emphasize the uniqueness of a process for
refinement or crystallite size which is carried out by oxide manipulation of the substrate. Especially in
the production of engineering materials like steel or aluminum alloys grain refinement plays an
important role for the material properties and is well researched. Usually specific grain sizes are
obtained by rapid or respectively slow cooling from melt, or by annealing of the intermediate product.
Another common measure is inclusion of additives which act as nuclei and promote crystallite
formation in this manner. Nevertheless, manipulation of the oxide thickness leading to grain
refinement is yet unreported and has led to a patent [40]. Exemplarily in the presented case an
enlargement of the crystallites is desirable. Larger crystallites lead to fewer electron-hole
- 74 -
5. Results and discussion
recombination at grain boundaries and thus produce high-power solar cells. Better controllable
homogeneity could make layers better to contact in further post processing. Contacting SSLPE layers is
difficult as shown in later chapters because of the large height-differences of the crystallites and thus of
the SSLPE layer’s macro-roughness. Smaller crystallites as provided with thinnest oxide layer could
produce smaller roughness in further prototype production.
As a contribution to the new growth model, this chapter reveals that thicker SiO
X
layers lead to fewer
stable nucleation sites. At intermediate SiO
X
layer thickness, this also leads to pronunciation of {111}
facets, most likely by overgrowth of area covered by SiO
X
. With thinner SiO
X
layer thickness, more
stable nucleation sites are available, leading to a higher crystal density in the SSLPE layer and to a
stochastic morphology. As known before, thick SiO
X
layer hinder growth and only lead to sporadic
crystal coverage.
5.4.Crysta
lli
ne
qua
li
ty
The analysis of crystalline quality has to be differentiated between SSLPE and seed layer (F
ig
u
re
26
).
Raman analysis of the growth layer shows a high, sharp peak at 520 cm
-1
. Its brilliance is generally
comparable to the signal of the commercial wafer used for calibration of the Raman system and does
not show a specific shift or additional asymmetry. The amorphous seed shows the typical broad signal of
a-Si with the center at 480 cm
-1
. Relevant to this work is also the signal of glass, since Raman
spectrographs for mapping were recorded through glass. Compared to the signal of crystalline silicon as
well as a-Si on glass shows no recognizable signal at the used wavelength. The same is true for a-Si
measured through glass, making it indistinguishable from glass. Nevertheless, crystallized seed
material shows clear peaks around 520 cm
-1
.
Since nanocrystals in the seed have been a central finding in earlier publications regarding this
research, and crystallization of the seed is a central question of this work, crystallites in the seed have
been investigated further. No crystallites are observed in the as-manufactured seed layer (F
ig
u
re
28
a-0/b-0). By Gaussian fitting of all measurement points of seeds after 1 s and 1 h after contact with Sn
at 600 °C, central wave number (shown here as deviation from 520 cm
-1
) and width (given by the
standard deviation of the Gaussian curve) have been determined and the pairs have been plotted to
investigate their correlation. This was done as described for the mapping procedure. In both cases of
different annealing time, a preferential value is observed (F
ig
u
re
27
).
After long annealing, central wave numbers vary between 517.5 and 520 cm
-1
, and widths of the
peaks σ vary between 1 and 2 cm
-1
, with virtually no outliers. In the case of quick annealing, a similar
- 75 -
5. Results and discussion
accumulation for peak widths is observed, but σ varies between 517 and 519 cm
-1
und thus showing a
blue shift. Additionally, a number of outliers exist for both values. Highly shifted peaks are correlated
with extensive peak broadening. Summarizing the annealing of the seed leads to a more homogeneous
signal, and shifted as well as broad peaks disappear.
- 76 -
5. Results and discussion
Figure 26: Raman spectra relevant to this work (dwell time 8 min). Notably, some spectra are not
distinguishable from another. This concerns SSLPE Si and a reference wafer and a-Si seeds which is
promising for their crystalline properties. Unpractically for the characterization is the identical
appearance of glass and the amorphous seed observed through glass.
Figure 27: Correlation of standard deviation and central wave number directly at the beginning of
crystallization and after 1 h. Notably standard deviation decreases and central wavenumber increases
on average. Both values get more homogenous with crystallization. Unfit data points are discarded
regarding the rules used for the mapping described in the text
- 77 -
5. Results and discussion
5.5.Seed
l
ayer
evo
l
ut
i
on
dur
i
ng
growth
F
ig
u
re
28
shows the evolution of a seed layer crystallizing during SSLPE growth. The resulting SSLPE
layer is shown in F
ig
u
re
23
and exhibits bimodal morphology described before.
As central observation, two phases are visible during the whole process. Naturally a crystalline phase
evolves. Additionally initial a-Si at least partially remains until the end of the 1 h of processing time.
Before growth, no crystalline signal is observed at all. Immediately after dip-in of the sample, partial
crystallization initiates. A dendritical network forms in the amorphous matrix but does not span the
whole area immediately at 1 s. Nevertheless after 30 s, the whole area is spanned by the network.
With further contact with Sn and increasing temperature, dendrites grow thicker (1 to 5 min). Before,
a-Si represented the matrix and crystalline material was embedded in it. Now, crystalline phase forms a
connected matrix and amorphous, residual islands remain within it. After 5 min, morphology has
reached a status which does not change further and is remained in the final stage after 1 h.
Figure 28: Development of crystallites during growth. A crystalline network develops during 1 h of
growth and eventually leads to matrix inversion between amorphous and crystalline material.
- 78 -
5. Results and discussion
It must be noted that crystallization starts immediately when a-Si and Sn come in contact. This is in
disagreement with observations from the TEM in situ experiment described in this work, where an
incubation time was observed. Several differences in the setups of the two experiments could be the
reason.
·Oxide: In the TEM experiment, oxide control was not possible due to long deposition time of
Sn and an oxide layer is expected between a-Si and Sn. In the line of experiment presented
here, the oxide thickness should be reduced or the oxide should be non-existent. Since Sn has
a catalytic effect on crystallization, Sn would need time to pass through the oxide first before
interacting with a-Si.
·Smaller Sn volume: In the in situ TEM experiment just small droplets act as catalysts. All
discussed catalytic effects depend on the size of the Sn volume: dissolving / etching of seed-
material, heat transport, agglomeration / adhesion of solved atoms and mass transport in
general.
·(No) supersaturation: As discussed in the chapter chapter
2
.
1,
Sn was saturated or
supersaturated with silicon when coming into contact with the seed layer, in the TEM in situ
experiment no silicon was solved before contact. The initial surface crystallites reported there
might play a role for the onset of crystallization, acting as initial nucleus. In this case surface
crystallites would form and a-Si underneath would crystalize from these nuclei.
·Temperature / heat effect: During the actual SSLPE process a large volume of Sn is already at
operation temperature. Thus, the high amount of heat stored in the Sn is injected fast into the
substrate while the large volume in total will only cool down negligibly. In contrary in the TEM
experiment, Sn is not pre-heated and will store only a small amount of heat according to its
volume.
Crystallization of the seed can be separated into three stages displayed in Tab
le
8
. A special focus
should be put into the analysis of the amorphous islands and specifically why they persist while their
surrounding volume is crystallized. Both parts receive equal thermal treatment so the difference has to
be originated in microscopic dynamic effects during the process. Comparing to chapter 3 a
coincidence between amorphous islands and plate-like surface patches is imminent, both show the
same size, shape and areal distribution. In the same manner the crystallized seed is connected to
stochastic morphology. In the following it is to be discussed whether the surface morphology originates
from the one of the seed or vice versa.
- 79 -
5. Results and discussion
Table 8 Stages of morphology development of the seed
stage time
1 1 s to
30 s
2 1 min
to 5
min
3 10 min
to 1 h
short description
Immediate crystallization
on-set and early
development of dendrites
Ripening and thickening
of dendrites eventually
leading to a matrix
inversion
Stabilization of
morphology after matrix
inversion
detailed description
Crystallization starts from a fully amorphous seed, in
agreement with TEM results. In contrary to earlier
research no initial nuclei must be present in the seed, but
the seed crystallizes quickly after substrate and Sn come
in contact. At the 1 s mark, the crystalline network is not
developed over the whole area but expands from a not
exactly observed starting point SEM images reveal a
large amount of initial surface crystallites with a high
density. Their number or distribution is in no correlation
to the morphology observed in the seed.
A crystalline network is developed after stage 1 and it is
embedded in an amorphous matrix. In stage 2 dendrites
of the network grow thicker and dendrites interconnect
as a consequence. This leads to a matrix inversion so that
the majority of the material is crystalline and only
residue a-Si remains. It is not interconnected anymore
but surrounded by crystalline material.
From 10 min on the composition of amorphous and
crystalline material remains stable especially including
stage 2`s final morphology of a-Si islands embedded in
former dendrites of crystalline silicon. That stabilization
is of special interest since annealing continues for the
whole duration. It must be questioned which effect
changes its behavior between islands and matrix,
effectively protecting amorphous remainders.
- 80 -
5. Results and discussion
By comparing crystallization and incubation speeds in Tab
le
2,
the observations during TEM in this
chapter, a number of observations can be made:
·Crystallization time is well underneath data reported for thermal annealing at 600 °C, and
especially showing no incubation time, uncharacteristic for thermal annealing without metal
catalyst. [65] indicates 600 °C as rule-of-thumb temperature for metal induced crystallization
of a-Si.
·Amorphous residue does not crystallize despite an annealing time of 60 min, pointing at a just
simple thermal annealing without metal influence (compare sources [63], [64] which present
total crystallization times of 150 min or even 360 min).
This means that the amorphous islands must be shielded from contact with metal while crystalline
volume is catalyzed by metal in some form.
In comparison with F
ig
u
re
24
this points toward theory b, where a surface crystallite overgrows an
area first. The plate-like crystal would protect the seed from interaction with metal atoms quickly.
Atom diffusion through bulk material is driven by the availability of three-dimensional inter-atomic
sites where the diffusing atoms rest during diffusion [67] and large two-dimensional areas spanned by
atoms through which atoms can pass. Both are much more available in amorphous bulks making a
crystal an efficient confinement for the metal catalyst F
ig
u
re
21
shows that the crystallization of a-Si
does propagate individually for the different dendrites and only singular dendrites grow into the seed`s
bulk. Other areas just crystallize close to the surface and no further expansion is observed. When the
volume close to the surface crystallizes it would also act as a diffusion barrier. This would support the
hypothesis depicted in F
ig
u
re
24
a. Dendrites in the seed show a preferred growth orientation. The
initial direction nevertheless should be arbitrary, which could lead dendrites growing from an initial
point in the seed just crystallizing the surface of the layer.
In conclusion bulk a-Si crystallizes due to interaction with tin. Crystallized Si forms a diffusion barrier
between Sn and Si which leads to areas remaining amorphous throughout the growth. It cannot be
clearly determined whether the barrier comes into existence as part of the growing SSLPE layer or as
partial crystallization of the seed. Nevertheless at growth time of 1 s, the spread of SSLPE crystallites on
the surface is in no correlation to the crystallized expansion in the seed. Concluding, initial SSPLE
crystallites form without connection to the crystallized Si in the seed. This observation is underlined by
the partial dissolution of early SSLPE crystallites. If connected to crystallized seed material would be
more stable. With the theory of surface crystallites forming first, also flat, shielding SSLPE crystallites
should develop before the crystallization of underlying seed a-Si.
- 81 -
5. Results and discussion
5.6.Laser
treatment
5.6.1.
Exper
i
menta
l
resu
l
ts
of
the
l
aser
treatment
With the presented method of laser cleaning, successful growth was demonstrated, as shown with the
best result obtained in F
ig
u
re
29
. Despite the untreated area only exhibiting few crystallites and mostly
exposed seed layer (upper half), the laser treated area (lower half) exhibits a dense homogenous SSLPE
layer. This result was made possible by laser-scanning with a beam with 16.8 µJ per pulse, with spot
diameters of d = 90 µm, resulting in an energy density of E/A = 0.264 J/mm². For this result, a quadratic
lasing pattern of 68 x 68 µm² was used and the seed layer was produced directly before growth without
following hydrogen-passivation, leading to a thin, naturally grown oxide which cannot be characterized
further. Successful growth was indeed reproducible, but only with unsatisfactory reliability.
The intensity distribution of the uncorrected laser beam (F
ig
u
re
30F
ehler! Verweisquelle konnte nicht
gefunden werden.a) exhibits two maxima as well as pronounced shoulders. The laser optic allows for
the isolation of one of the maxima (F
ig
u
re
30
b), nevertheless the shoulders in the intensity distribution
remain. In comparison with a Gaussian distribution (light green graph in the projection), the real energy
density show steeper increments towards the center, and flatter segments towards outer radii.
The interaction of laser and seed layer is depicted in F
ig
u
re
31
. SEM (a) shows a homogenous crater
area in the center, surrounded by a wall of piled up material. Outside of that high impact interaction
area different shadings are visible, without any morphology changes. An analysis of the impact area
with the Raman mapping algorithm presented in chapter
4
.
3
reveals cracks in the crater area, which
are not susceptible with SEM. Looking at the Raman spectra, a number of observations can be made.
The crater area is fully crystallized (green area and yellow cracks), with the crystalline peak being higher
outside of the cracks but more refined and thus narrower within the cracks. Both show the same
deviation from k
cr
= 520 cm
-1
, indicating comparable crystallite size. The direct surrounding of the
crater of the crater is also crystallized, but exhibits further deviation, and a broader and less intense
peak.
Over the course of the research into laser-induced oxide removal or weakening, growth was
successfully induced by lasing. Nevertheless, no reliable treatment routine was found. A critical factor
hindering growth was identified in the melting regime and subsequent crystallization of the seed.
Underneath the melting regime, growth is enabled and supported, independent from the substrates. In
exploratory runs on monocrystalline substrates, growth was possible with a higher range of energies and
especially at much higher energy dosages.
- 82 -
5. Results and discussion
Figure 29 contiguous and homogenous growth achieved by laser treatment (bottom half). In the
untreated area (upper half) only scarce and inhomogenous crystallites are found
- 83 -
5. Results and discussion
Figure 30 intensity map of the unadjusted beam (a) and the adjusted beam (b). A Gaussian
approximation of both (green curve in the projections) shows notable deviations – measured values in
the center are higher and both spots have pronounced shoulders unlike the Gaussians used.
Nevertheless Gaussians will be used for a generalized description of the beams. (a) Average standard
deviation σ = 96.0 µm, ellipticity (σ
x
-σ
y
)/σ
x
= 0.72 (b) Average standard deviation σ = 46.4 µm,
ellipticity (σ
x
-σ
y
)/σ
x
= 0.91
Figure 31 a-Si surface treated with an adjusted, high energy laser pulse. a) the SEM image shows a
crater with a wall of accumulated material surrounding it. Outside of the krater area, different shadings
are visible on the surface. b) the Raman map was processed with the algorithm presented before.
Different exemplary spectra are shown in c). the Krater area is crystallized, as well as the direct
surroundings of the wall area. Cracks within the silicon are visible throughout the crystallized area. c)
shows exemplary Raman spectra obtained at the indicated locations 1 to 4.
- 84 -
5. Results and discussion
5.6.2.
S
i
mu
l
at
i
on
of
the
i
nteract
i
on
between
l
aser
and
surface
The simulation calculates which temperature is reached by a definite energy density in a finite element.
The correlation between both is presented in F
ig
u
re
32
. Heating in regimes without any reaction
happening lead to faster temperature increment per energy unit than the regimes with a reaction
present, dampening heat-up. The comproportionation reaction of SiO
x
starts at E/A = 4.5 J/cm² and
melting of a-Si from E/A = 35.0 J/cm². If a Gaussian energy distribution is applied, then the critical
energy densities are reached at different radii. As result a temperature profile over the spot radius is
obtained with the steps in the heating profile also being visible. This leads to an abrupt increase in the
center of the spot when the dampening of the comproportionation reaction is lost and the energy
distribution is steep as well. In F
ig
u
re
33
Gaussian energy distributions (c and d) are applied to the
simulated correlation between energy density and reached temperature. In the resulting temperature
distribution (a and b), the steps visible in F
ig
u
re
32
are found again.
Figure 32 temperature achieved in an infinitesimal element depending on laser intensity in that
segment. The chemical comproportionation reaction of silicon dioxide takes places starting at T =
750 °C (green highlighting), and the melting of a-Si at T = 1147 °C (red highlighting). Both reactions are
endothermic and require considerably more energy than just heating the volume which results in a
slower temperature increment over the energy range in which heating and a reaction takes place.
- 85 -
5. Results and discussion
Figure 33 simulated temperatures resulting from laser treatment with an ideally gauss-shaped laser
beam. The chemical comproportionation reaction starts at T = 750 °C (green highlighting) and melting at
T = 1147 °C (red highlighting), both dampening heating of the substrate. a) Different total spot
energies with a constant beam width (σ = 100 µm). b) Different beam widths σ with a constant total
pulse energy (E = 3.0 µJ). c) and d) depict the energy distributions leading to the respective
temperature profiles in a) and b).
A central goal for optimization and control of the lasing process was an understanding of the heating
induced by the beam and understanding of the root cause of the inhomogeneous markings F
ig
u
re
31
a
and F
ig
u
re
34
a left after lasing on substrates and how they are interconnected with SSLPE growth
morphology. Intensity measurements of the beam, simulations of the establishing temperature profile,
Raman measurements, and micrographs of the seed as well as the SSLPE layer are in good agreement.
The laser beam was used in variations either of the full uncorrected beam with two local maxima or a
corrected beam which was shaped close to a Gaussian profile with only a single maximum. In either
- 86 -
5. Results and discussion
case the laser did not show any discontinuities or vast jumps in intensities. SEM of laser treated seed
material at higher energies in contrary shows different interaction regimes of the seed surface. There
are two origins of this behavior
·Melting and chemical reaction are reached at certain temperature steps. If they are reached by
a continuous beam profile, discontinuities are created in the temperature profile. For
melting this leads to an obvious observation – to crystallized material over the whole area and
to piled up molten and crystallized material in its vicinity.
·The SiO
x
serves as a temperature buffer during heat up. When all of it is evaporated,
temperature increment per energy is increased drastically. Even if oxide forms again after
cooling down, temperature steps created by the endothermic reaction could still lead to
shading of the treated area.
As optimization for the process it was eventually decided to use the larger but uncorrected beam over
the smaller but corrected spot. Removing the secondary intensity maximum did promise better areal
coverage due to its rotation-symmetric shape making spots easier to arrange. Nevertheless the
uncorrected beam covers a larger area. Also, energy densities do rise less steep in critical parts of the
beam. F
ig
u
re
33
b demonstrates that increasing width of the beam from σ = 40 µm to 200 µm will
enlarge the reacting radius at constant energy, but at no point the melting regime is reached.
Experimental results show clearly that growth can be supported or created by laser treatment (F
ig
u
re
34
). Nevertheless, melting of the seed prevents growth in the affected area. The derivative
optimization problem explains the wide range of results after laser treatment. A critical, homogenous
temperature is needed in a possibly large area, but the critical energy of melting must not be breached.
This is supported by results on monocrystalline substrates. Here, much higher energy dosages still lead
to successful growth, a fact easily explained by the high melting temperature of crystalline silicon of
1420 °C compared to the one of a-Si at 1147 °C.
- 87 -
5. Results and discussion
Figure 34 interaction of a small, corrected beam with the seed surface and on the subsequent SSLPE
growth. Interaction area depicted as a) SEM image of the seed, b) Raman map of the seed, and c) SEM
image of the resulting SSLPE morphology. Complementary Raman spectrums d) and lateral plot of the
crystalline Raman peak e) are provided. The molten and recrystallized center show a satisfaction
intensity of the crystalline Raman peak and leads to prevented growth during SSLPE. Partial
crystallization without melting does not hinder growth and the lasing supports SSLPE.
The observation that on seed layer crystallized before insertion into Sn, no SSLPE layer is formed in
agreement with former experiments with fully crystallized seed layers, also leading to prevention of
SSLPE growth. This is a clear indication that a dynamic interaction between seed and SSLPE layer is
necessary to stabilize early crystallites (F
ig
u
re
24,
theory a)
The simulation reveals when a critical reaction temperature is reached at a certain pulse energy, and
which proportion of volume is affected. In reality a minor reaction at the interface of Si and SiO
x
could
be sufficient to destabilize and ablate the oxide layer. Also imaginable is the necessity to bring the full
- 88 -
5. Results and discussion
volume to reaction. The high variety in successful growth parameters obtained are explained by the
interplay of different factors. As a general observation it must be differentiated between small spots
with low energy pulses and large spots with high energy. Small spots reach high temperatures even at
low pulse energies. Their steep intensity profile leads to a noteworthy interaction only in their center
and allows for precise manipulation in that center. Unfortunately, the slope of the curve is not used.
Broad spots use the slowly flattening flank and are less dependent on the beam center (F
ig
u
re
33
d). As
a complicating factor, thickness of the oxide layer is detrimental, but during processing highly
dependent on the daily environment in the laboratory like humidity and processing times. As a result,
its thickness and its configuration is unknown and the laser energy cannot be tailored for a singular
run. The reaction in SiO
X
consume a high amount of energy, and functions as a temperature buffer.
This effect is dependent on the volume. Thus a thick oxide can lead to better moderated temperatures
and can prevent melting. At the same time a thicker oxide is harder to bring to reaction. In total the
interaction is very volatile, depending on various factors and hard to predict.
5.7.E
l
ectr
i
ca
l
propert
i
es
Hall measurements with van-der-Pauw setup show n-type conductivity in both epilayers. EPR confirms
that by revealing As and P as electron donor dopants in the material. A fitting of the Hall-measured
data of the Si(111) growth layer shows best accordance with 1.79 × 10
16
cm
-3
As and 6.6 × 10
15
cm
-3
,
dampened by 2 × 10
14
cm
-3
unspecified acceptors. Epi-Si(111) layers show carrier concentration of
2.45 × 10
16
cm
-3
and mobility of 890 cm
2
/(Vs). Epi-RPS layers are in the same magnitude with 2.2
× 10
16
cm
-3
and 825 cm
2
/ (Vs). MDP shows only a minor reduction in saturation voltage for both
materials when minority carrier traps are saturated with a bias, confirming low density of minority
traps. Saturation voltages induced during MDP show high pairs of voltages and minority carrier bulk
lifetimes of 785 mV with 158 ms for epi-Si(111) layers and lower values of 60 mV with 55 ms for epi-
RPS(111). Additional testing of a layer grown on an RPS and subsequently separated from the parent
wafer shows vast change in properties. Saturation voltage is decreased and drops even further down to
15 mV.
- 89 -
cm
5. Results and discussion
Table 9 electrical data as already published in [30]. The high measured lifetimes make the material a
candidate for solar cell build-up.
substrate Si(111) RPS(111) a-Si / glass
DLTS
van-der-
Pauw
majority-
carrier traps
conductivity
type
carrier
-3
concentration
Dislocations, point
defects
n-type
2.45 × 10
16
Dislocations, point
defects
n-type
2.2 × 10
16
inconclusive
n-type
2.0 × 10
16
mobility cm
2
/(Vs) 890 825 330
As Donor cm
-3
P Donor cm
-3
Aceptors cm
-3
lifetime
MDP overall µs
(no bias)
1.79 × 10
16
6.6 × 10
15
2.0 × 10
14
41.3 14.9
lifetime
overall (bias) µs 40.9 14.6
lifetime
ᵰbulk
bulk µs 158 55
diffusion
length µm 608.8 419
EPR Donors As / P
- 90 -
5. Results and discussion
Figure 35 temperature dependent b1 coefficients (a-1 and b-1) and their Fourier transformation (a-2
and b-2) revealing energy and effective cross section of the defects
DLTS shows broad temperature transients for both SSLPE layers grown on Si(111) and on RPS(111)
layers, indicating extended lattice defects [113]. The defects will be indicated as epi-Si1 and 2, as well
as epi-RPS1 to 3 to clarify that they were found in epilayers grown on the noted substrates. The specific
cells discussed later are noted with Si(111)a and b, as well as RPS(111). The transients itself appear
different, in example showing different number of peaks and differently broad, and they reveal
different trap activation energies. DLTS of Epi-Si(111) exhibits two peaks with energies of 0.318 eV
(3.92 × 10
-17
cm
2
) and 0.428 eV (9.8 × 10
-17
cm
2
). Epi-RPS has three defects with energies of 0.149 eV
(2.38 × 10
-15
cm
2
), 0.147 eV (1.24 × 10
-18
cm
2
) and 0.459 eV (3.52 × 10
-15
cm
2
).
Adaption of the solar cell manufacturing process proved to be complicated for epitaxial substrates
(SSLPE layers on wafers and RPS) and impossible for polycrystalline SSLPE silicon on glass. In both cases,
- 91 -
5. Results and discussion
the macroscopic roughness hindered bonding of the added layers and the polymer spacer. The original
process was designed for LPC substrates. For the LPC process, a stable SiO
x
cap would be added before
laser crystallization which stabilizes the flat surface of the a-Si seed layer. After first functionalization
efforts were unsuccessful, runs on SSLPE polysilicon were discarded. Further modification of the
sensitive surface seemed unpromising without comprehensive changes in the manufacturing method.
In a following attempt, SSLPE epilayers nevertheless were polished and processing was possible
afterwards.
Results from Suns-V
OC
measurements exhibited best values up to V
OC,max
= 265 mV, an average of
V
OC,av
= 221 mV and a standard deviation of V
OC,σ
= 47 mV. The most promising cells were characterized
further with a sun simulator, producing J-V curves, reflectivity and quantum efficiencies depicted in
F
ig
u
re
36
.
Table 10 characteristics of the best prototype solar cells for each substrate type
short-circuit current density
open-circuit voltage
[mA / cm²]
[mV]
Si(111)a
16.88
238
Si(111)b
10.35
259
RPS(111)
16.39
266
fill factor [%] 29.5 34.8 23.3
efficiency
U
MPP
J
MPP
MPP
series resistance
shunt resistance
[%] 1.18
[mV]
120 [mA
/ cm²] 9.87
[mW / cm²] 1.18
[Ω · cm²] 14.2
[Ω · cm²] 16.5
0.93 1.02
150 120
6.22 8.47
0.93 1.02
13.9 25.6
39.1 i n c onclusive
- 92 -
5. Results and discussion
Figure 36 Cell build-up and data of cells build from SSLPE-epilayers, grown either on Si(111) (a, b and c-
1) or on an RPS(111) (a, b and c-2). (a) photographs of the cell arrays with the quasi-backside contact
covering all inactive front area. The active area is covered with ITO and is used as front-side contact. (b)
J-V curves of one cell of the respective wafer and (c) reflectivity and external quantum efficiency with
the resulting internal quantum efficiency
- 93 -
5. Results and discussion
The J-V curve of the cells with the highest V
OC
and J
SC
for Wafer and RPS substrates both exhibit less
pronounced exponential behavior (F
ig
u
re
36,
Si(111)a and RPS(111)), with fill factors of
FF
Si(111)a
= 29.5 % respectively FF
RPS(111)
= 23.3 %). The cell with the most pronounced solar cell work
curve reaches FF
Si(111)b
= 34.8 %, but only reaches lower J
SC
and thus has no improvement of efficiency. All
three cells have similar V
OC
between V
OC
= 238 mV and V
OC
= 266 mV. Reflectivity is similar for both
functionalized substrates, but internal and external quantum efficiency differs vastly between the two
substrates, exemplarily represented by maximum external quantum efficiency of EQE
Si(111),max
= 33 %
and EQE
RPS(111),max
= 24 %.
As mentioned in the introduction of this chapter, the good electrical properties of the material
measured and analyzed before are not in agreement with the disappointing performance of the
prototype devices manufactured. This discussion analyzes electrical properties of the material and
characteristics of the prototype devices. Special regard is put on reviewing the DLTS data, as well as
the manufacture of the cells. DLTS can explain shortcomings in material quality which poses problems
amplified by cell architecture. The adapted manufacturing process poses problems especially
concerning interface and recombination there.
To determine if the defects have an incisive effect on the cells’ properties and are in connection to the
presented shortcomings, possible origins are discussed in the following. While narrow DLTS-transients
are associated with point defects as dopants, extended or broadened transients point towards low
dimensional traps like dislocations. Along a dislocation, several charge carriers can be trapped, but the
coulomb field of an already trapped one will change the relaxation behavior of others within the same
dislocation, resulting in different transients [113], [114]. Also, presented energies are within the range
of reported data for extended lattice defects, which range from 0.07 eV [114] to 0.54 eV [113], both
reported for dislocations. It also has been reported that energy of lattice defects vary depending on
their exact configuration, allowing for variation of their activation energy instead of discrete
values [115]. For metallic impurities discrete energies have been reported for the respective
donors [116], [117]. Contrary to extended defects, less diversion from reported trap energies
introduced by metal dopants is be acceptable for interpretation. Both investigated epilayers are
showing transients relating to crystal defects. It is likely that lattice defects are introduced by the
growth method but pronounced differently on the different substrates. If there were metal impurities
within one epilayer, they would have been introduced during growth, which again should produce
identical signals from both layers. Nonetheless crystal defects could very likely stem from a substrate
- 94 -
5. Results and discussion
and be promoted into the epilayer. It also is imaginable that the growth reaction differs slightly
between the respective substrate and the solvent, due to offcut or internal stress in the surface.
Table 11: Assumed origins of the defects found in monocrystalline SSLPE silicon. All recorded defects
can be explained by differently pronounced vacancy complexes, respectively dislocations. If the growth
method originates the defect, similarities are likely. E3 and PHR-V3 would represent annealed states of
other defects, and the increased capture cross section of the dislocations could stem from annealing as
well. V2 is known to appear in coexistence with E3.
defect
label
epi-Si1
epi-Si2
epi-
RPS1
epi-
RPS2
epi-
RPS3
Energy levels (eV)
and capture cross
sections (cm
2
)
Ec - 0.318 eV
(3.92 × 10
-17
cm
2
)
Ec - 0.428 eV
(9.80 × 10
-17
cm
2
)
Ec - 0.149 eV
(2.38 × 10
-15
cm
2
)
Ec - 0.147 eV
(1.24 × 10
-18
cm
2
)
Ec - 0.459 eV
(3.52 × 10
-15
cm
2
)
Assignment
and
particularities
E3
V2 -/0 (single)
dislocations
at 2.2 %
deformation
PHR-V3 -/0
(single)
comment
cross section not in alignment with
literature. Possible hidden peak
cross section not in alignment with
literature. Possible hidden peak
various defects reported within the
range of the measured defects.
Annealing is reported to increase
capture cross section
increases LC
reported as /
similar defect
Ec - 0.32 eV
(3.0 × 10
-15
cm
2
)
Ec - 0.42 eV
(1.7 × 10
-5
cm
2
)
Ec - 0.07
(1.0 × 10
-15
cm
2
)
Ec - 0.11 eV
(6.0 × 10
-18
cm
2
)
Ec - 0.458 eV
(2.4 × 10
-15
cm
2
)
source
[118]–
[120]
[118]
[114]
[114]
[121]
- 95 -
5. Results and discussion
Concerning their energies, the 0.32 eV traps could be originated in platinum (Pt) or palladium (Pd)
impurities [116]. The 0.43 eV trap could stem from aluminum (Al) [117]. As stated before, metal
impurities are expected to cause narrower transients than the transients observed, making metal
impurities unlikely. Additionally, SIMS measurements conducted before did not show Pt or Pd, and low
concentrations (< 2 × 10
13
cm
-3
) of Al. Al defect concentrations have been reported to be 3 orders of
magnitude smaller than the actual doping concentration and to have a capture cross section of 3.2
× 10
-13
cm
2
[122].
The lattice complex V2 / E3 as reported by [118] would explain the epi-Si2 trap as dual vacancy (V2)
and epi-Si1 trap as the defect E3 which is reported to stem from annealed dual-vacancy (V2) effects. V2
is a well reported defect often produced for further research by bombardment with various
particles [118], [121], [123], [124]. V2 has two significant charge states, but depending on particle
bombardment, Si production method and annealing, it shows very different DLTS spectra [124].
Additionally to the 0.42 eV single-charged state mentioned above it often shows its double charged
state at 0.23 eV. The latter was reported to be not existent depending on the origin of the Si
experimented on [124]. E3 is yet not appointed to a specific lattice phenomenon, but was traced back to
arise during annealing of V2 defects. During heat treatments of 30 min at 250 °C or 300 °C, V2 signal is
reduced drastically while the E3 signal rises. Hereby, the double charged state can be annihilated
completely. Also, E3 appears to be a dissociation product of VO [118]. It is likely that V2 lattice defects
are produced during growth, but that a majority of them reorganize during the growth due to
annealing and form E3 defects.
Concluding, the epi-Si(111) layer represents highly annealed Si, with heavy influence of the E3 defect.
All of the defects in question appear in as-grown bulk Si [118], [124], so they could stem from the
substrate and be promoted into the epilayer. Only the high ratio of E3 would then be in question. The E3
defect’s crystallography is more complex but thermodynamically also more stable [118]. Since SSLPE
growth takes place close to thermodynamic equilibrium, formation of said defect could be promoted.
Also, the SSLPE process temperature of 700 °C will anneal any already formed layer over the two hours of
process time. The degenerate ratio of high E3 to low V2 is yet unreported in literature.
For epi-RPS1, 2 and 3 no commonly reported metal impurities show the energy levels in question. Only
indium (In) is close with 0.16 eV [125] which for a metal impurity is too much of a deviation. For lattice
defects other than point defects, a broader energy peaks are more likely.
A study of [114] investigated strain induced dislocations in Si and found several energy levels from
0.05 eV to 0.14 eV, with capture cross sections ranging from 4 × 10
-18
cm
2
to 1 × 10
-13
cm
2
. While no
- 96 -
5. Results and discussion
specific defect fitting the presented data was reported, the report shows how different dislocations
can appear within this energy range. Additionally, the group reports a connection between annealing
and higher values of the capture cross section with unchanged energy.
Annealing of strained crystals leads to cross slip of dislocations, leading to their accumulation and
eventual formation of highly ordered dislocation networks. It is likely that this effect of higher order
between defects causes a change in their cross section. Since dislocations show a multitude of energy
levels and capture cross sections depending of the maturity (annealing state) of the silicon, we think it
is likely that the energy states of epi-RPS1 and epi-RPS2 are caused by a specific dislocation network
not reported yet.
Nevertheless, the defect epi-RPS3 0.459 eV (3.52 × 10
-15
cm
2
) has a nearly perfect resemblance in a tri-
vacancy (V3) reported by [121]. V3 was observed in silicon after heavy ion-bombardment, simulating
conditions in the detectors of the large hadron collider at CERN. It exhibits two configurations, with a
total of three energy levels. One of which, the PHR configuration, exhibits a single charged state of
0.458 eV (2.4 × 10
-15
cm
2
), notably close to the 0.459 eV (3.52 × 10
-15
cm
2
) measured in epi-RPS layers.
Unfortunately it is reported as responsible for increased leakage current. The PHR dual charged state is
reported as 0.359 eV (2.15 × 10
-15
cm
2
) and to diminish during annealing of 1850 min at 80 °C. Thus, it
would be feasible that the dual charged state diminishes due to annealing during the growth period,
while the single charge state remains.We think the observed epi-RPS3 defect is very likely the reported
single charged PHR-V3 defect. The V3 data reported coincides closely to epi-RPS3. As theorized for epi-
Si(111), our SSLPE material would represent a heavily annealed state of the material. V3 is a rare defect
observed after unique ion treatment of Si, but we think that these defects might possibly arise during
SSLPE growth. As stated above, SSLPE produces very mature crystals and might produce and preserve
rare lattice defects, otherwise only observed after ion bombardment. Also, the reported high leakage
current of V3 resembles the low saturation voltage achieved in our MDP measurements. Also, the
inconclusive measurement of shunting resistance in the cell measurement could point to a negligible
shunting resistance of the material. The J-V-curves exhibit a mostly flat curvature. Both shunting and
series resistance can lead to this behavior.
·A low shunting resistance R
SH
enables current to flow back through the device. As a result,
charge cannot build up and is reduced. This leads to a flattening of the J-V-curve first, but with
very low R
SH
also V
OC
decreases.
·High series resistance R
SE
hinders current to flow, thus also flattening the J-V-curve. With
higher R
SE
, it influences current flowing even if short circuited, reducing I
SC
.
- 97 -
5. Results and discussion
Regarding the recorded J-V-curves and especially the data of low I
SC
and V
OC
, both low R
SH
and high R
SE
must be effecting the efficiency of the cells. In combination, they could originate both the J-V curves
with weak exponential beahavior and would explain overall low values. Additionally, both internal and
external quantum efficiency are low, with acceptable reflectivity. Low shunting resistance could
originate simply from metal residue from SSLPE forming a contact. Or, due to the not optimized cell
manufacturing process, from the application of metal or TCO being imperfect and contacting the two.
Looking at the diffusion path which charge carriers take through the device (compare F
ig
u
re
10
), the
backside interface plays an important role in the properties of the solar cell. The used manufacturing
technique for solar cells was adapted from cells build from laser-crystallized silicon on glass. In that
original development of the cells, backside passivation has proven to be detrimental for the cells
properties [19]. Electrons as majority charge carriers have to pass by that interface for several
millimeters during which they could recombine there, thus this interface plays an important role. In an
optimized cell produced with the technique used here as well, finger widths of d = 1080 µm were used
[95], compared to several millimeters electrons have to travel in the here presented prototypes. As a
first observation, the passivation effect of the p-n-junction might just be too shallow for sufficient
passivation. Apart from electrical factors of material and cell, the manufacturing method and
adaptability to the SSLPE material must be discussed. Band structures theorized for the stack could
well be obstructed by problems during material application, and as well already during epitaxy. As
backside material, p-type silicon was chosen to create a potential barrier for electrons as the majority
charge carriers. A number of factors make it likely that this specific interface is imperfect:
·Etching at the beginning moments of epitaxy possibly creates a macroscopically rough
interface. This might include simply a curved interface which should not influence the electrical
properties further. If dendrites of one material reach into the other, the electrical potentials
might create depletion zones. As a result, the effective area which is electrically passivated be
reduced. As a consequence, less charge carriers are reflected. Additionally, macro-vacancies
might be created which create surfaces which are not passivated and lead to loss of charge
carriers.
·Long processing times of the epitaxy represent annealing of several hours for that interface.
During that time, diffusion of dopants is likely which prevents a sharp interface. Also, p- and
n-dopants coexisting close to each other might have unpredictable outcomes, for example it
could create more complex defects or additional vacancies. Diffusion-barrier layers like Si-
- 98 -
5. Results and discussion
nitride (SiN
x
) have been reported to hinder diffusion of impurities between substrates and
improve interface quality [19], [92], [126].
The long travel distance for the charge carriers is detrimental as well. It extends problems with the
back-side interface, as charge carriers travel along them longer. Also, bulk properties play an important
role, as errors in the bulk will create more chances for recombination with longer charge carrier
diffusion paths.
In summary, performance of the cells seems to be dependent on electrical properties of the material,
from insufficiencies in the back-side interface and from the newly adapted cell processing method. As a
result, all three areas should receive further investigation towards a reworked approach in which all of
them are optimized to work together.
An extended TEM investigation could clarify a number of different points in this discussion.
Dislocations and vacancies assumed in the bulk from DLTS data could be confirmed, as well as other
crystalline defects in bulk, interface and substrate. A TEM investigation could as well be extended to
interfaces and layers created during cell build up. This should be precluded by an investigation of the
stack via optical microscopy to exclude microscopical shortcomings of the included layers.
This chapter shows how detrimental the components and their functionality and optimization is for
the final functionality of the cell. And might require more than an adapted manufacturing method.
This starts at the interface between substrate and SSLPE layer, which might be improved drastically by
a layer stack preventing diffusion during epitaxy and which isolate electrically within the final cell. A
diffusion barrier could improve bulk properties, but the SSLPE process would need to be adapted to
the new substrate surface in a way the intermediate layer is protected. The SSLPE process itself might
profit from slower growth, so that higher crystalline perfection is reached. Finally, the manufacturing
itself might need further improvement, and might even use different possibilities if not being applied to
an LPC substrate. For the currently presented prototypes, these optimizations were not given,
instead processing steps which not conceived to work with each other. Thus it has to be emphasized
that this chapter has demonstrated the possibility to manufacture solar cells from SSLPE silicon for the
first time.
- 99 -
6. Summary and outlook
6.
Summary
and
ou
tl
ook
This presents a novel comprehensive model of SSLPE growth. The growth itself takes place in an
enclosed chamber while the growing layer is embedded in liquid tin and is not directly observable. To
overcome this limitation, the author approached open questions with deliberately designed
experiments and methods, presented in the respective chapters of this thesis.
Originally, the surrounding preparation-steps of substrate preparation and oxide control were
time-consuming and prone to errors. Subsequently, many factors influenced the outcome of growth
and morphology phenomena could not be traced back to processing steps. Another open question was
the applicability of SSLPE silicon as absorber in a solar cell. Despite promising electrical properties and
efforts to produce prototypes, no functioning solar cell was demonstrated before. As a consequence,
the three following goals were formulated and achieved:
1. Technical optimization: the overall process is complex and the three individual sub-steps have
not been fully exploited before in terms of their technical capabilities. By revisiting each of
them it was possible to make the experiments very controllable.
2. Understanding of the processes involved in growth: With improved reliability, it was possible to
design sensitive growth experiments which were not anymore disturbed by multiple and
uncontrollable interferences and led to an improved understanding of the growth process as a
whole
3. Prototype device manufacture: with an improved process and understanding thereof, SSLPE
was validated by using SSLPE layers for the manufacture of prototype solar cells.
Central to the further in-depth research of SSLPE growth was the technical exclusion of disturbances
during the process chain. A major weakness of the original process was its unreliability. With the
implementation of a novel method to facilitate multiple substrates or respectively seed layers
simultaneously, it was ensured that multiple SSLPE growth runs could be pursued with merely identical
seed morphology. Central for this approach was successful oxide control. For this work, laser treatment
of the oxide and etching with hydrofluoric acid was looked at in depth. Both led to successful results
when applied directly before growth. Nevertheless only hydrofluoric acid etching proved to be
applicable undisturbed from changing lab conditions, unlike laser treatment. With etching, it is possible
to clean oxide layers fully and independently from other preparation parameters. In this work it is
shown that laser cleaning is highly dependent from SiO
X
thickness, since the SiO
X
serves as a heat
buffer. If thinner than aspired, the seed will have to withstand hotter annealing temperatures, and if
the SiO
X
layer is too thick, more energy is necessary to reach critical temperature. This work also
- 100 -
6. Summary and outlook
demonstrates that different SiO
X
thicknesses can be used to adjust crystallite density in the final SSLPE
layer, by changing the density of nucleation sites. Additionally to the academic value of a novel method
to change nucleation density by changing the seed’s oxide layer [127], this method has also resulted in
a patent application [40].
The evolution of the seed layer acted as a blackbox throughout the research of SSLPE, but with the
implementation of identical substrate preparation (see chapter 2), it is now possible to investigate and
reveal its different states during growth (chapter 5). Before, growth results between runs were hard
to compare, since the PVD seed layer preparation was influenced heavily between different runs, and
consequentially changed the seeds behavior during growth. This works compiles research results to
demonstrate how much variance the crystallization progression of a-Si can be dependent on changing
preparation techniques. Chapter 4 focused on another uncertainty: whether the interaction of seed
and metal solvent is relevant to the process or crystallization is rather caused by annealing or the
vicinity of surface crystallites. To reveal the metal’s influence, an in situ TEM experiment was designed
and revealed that crystallization starts at the interface Si-Sn and that incubation time is reduced
drastically by the layer’s contact. The different states of crystallites on the surface were investigated
with unprecedented depth, again enabled by the implementation of identical substrates.
As a result, a novel growth model can be presented (F
ig
u
re
37
), which explains the formation of
different types of surface morphologies existing simultaneously, which recognizes the dynamic
crystallization of the seed, as well as its interaction with the continuously developing SSLPE layer. The
seed layer is fully amorphous before insertion into Sn solution. The Sn solution is supersaturated with
silicon atoms. When a-Si and Sn surface come in contact, crystallites immediately nucleate on the
substrate, forming a dense layer of small crystallites which do not coalesce. At the same time,
crystallization commences within the seed. Crystalline dendrites start growing from the interface Si-Sn
in different angles into the seed. Some of these dendrites reach towards the interface silicon-glass fast
and are detectable by Raman measurements conducted from the back side. On the corresponding
areas of the seed, stochastically orientated crystallites are visible. Other dendrites grow laterally,
effectively shielding the seed from further crystallization and forming islands of remaining a-Si. On top
of these laterally expanded dendrites, plate-like crystallites are appearing. From there on,
crystallization of the seed continues by the dendrites enlarging, growing as well longer and thicker.
Stochastic surface crystallites undergo Oswald ripening – their number decreases while their individual
size increases. Plate like crystallites enlarge and form round patches on top of amorphous islands. With
- 101 -
6. Summary and outlook
prolonged growth, this morphology stabilizes, and epitaxial growth of existing surface crystallites leads
to their enlargement, without any further influence of Oswald ripening appearing.
Figure 37 summary of results concerning morphology development during the growth of a
polycrystalline SSLPE layer and dynamic seed layer crystallization. (a) before insertion into tin, the seed
layer is fully amorphous and located on a substrate glass (b) in the moment of insertion, nucleation of
surface SSLPE crystallites in high density takes place. The surface of the seed layer crystallizes
immediately. Some crystallized dendrites reach the interface of glass and a-Si (c) crystallization of the
seed commences, with dendrites growing longer and broader within the a-Si. In these areas,
stochastically orientated crystallites persist. Their number is reduced by Oswald ripening, while the
remaining crystallites enlarge. Some volume of the seed in surface area does crystallize laterally, and
that lateral crystallization is represented in larger, flat SSLPE crystallites on the surface. (d) Oswald
Ripening and growth on the surface continues, while crystallization state of the seed stabilizes as seen in
(e).
- 102 -
6. Summary and outlook
Further insight into the crystal quality can be gained by interpretation of the DLTS. A total of five
different charge states were identified on the two different substrates. All charge states are explicable
by lattice defects and share the property that they represent highly annealed states of lattice defects.
The otherwise unreported ratios between detected energy levels of these defects indicates annealed
and matured conditions of known and reported lattice defects. This demonstrates that SSLPE silicon
represents thermodynamically stable or respectively highly annealed material.
Due to the promising electrical properties of SSLPE material reported before, manufacturing of
prototype devices seemed feasible. Modern solar cells are complex and require specialized tools for
the deposition of various functionalization layers, which in themselves require optimized substrates.
This work successfully demonstrates solar cells made from SSLPE material for the first time. Since
preparation of SSLPE layers was the focus this work, adaption of the raw material to a cell
manufacturing process could only be realized after the necessities of reliable growth were met. As a
result, monocrystalline silicon substrates were used and a cell manufacturing process, originally
developed for LCP material and adapted for the SSLPE material stack [13], [92]–[95]. This approach
came with a number of shortcomings. For the original LCP-cell concept, functionalization layers for
passivation of the backside are applied on the substrate glass before the future absorber, achieving
back side passivation. In the adaption for the monocrystalline SSLPE layer here, backside passivation
was realized by using a p-type substrate wafer with n-type SSLPE. Low shunting resistance and high
series resistance burden the presented cells. A reason might be that the backside interface is
compromised by melt-back in the beginning of growth, and by annealing during growth. Thus problems
with the adaption of the manufacturing process were identified as a reason to the low quality of the
prototypes. Additional insight in electrical properties was gained by revisiting DLTS data acquired
before. Lattice defects form singular vacancies and vacancy constellations like dislocations or clusters.
Neither are reported to have a detrimental influence of the electrical properties of silicon, but are
reported to lower efficiencies of solar cells, and lead to unusable layers in extreme cases [12], [16],
[17]. In order to qualify produced cells further, several points of improvement can be stated. Foremost,
existing cells must undergo investigation concerning their layering and especially concerning their
backside interface. Basis to this would be the preparation of the interfaces and layers with the
metallurgy toolset. Optical microscopy, SEM and energy-dispersive X-ray spectroscopy (EDX) could be
used to find imperfections of the layers and inclusions, for example of tin at the interface Si-Si.
Methods able to reveal doping of the silicon could then be used to detect electrical properties of the
different heterojunctions, for example by MDP or by conductive AFM, revealing conductivity of the
different materials. In general, for cells to be improved sustainably, a specified cell concept for SSLPE
- 103 -
6. Summary and outlook
stacks must be designed which respects the boundaries of the growth method, but also uses the
possibilities of low growth temperatures to apply backside functionalization layers and possibly a
metallization layer. A possibility would be silicon-carbide due to its high conductivity and high thermal
resistance. Most importantly, cell build-up must be included in the growth and optimization cycle, with
the prototyping tools available easily and with the option to build up cells occasionally, so that
promising layers can be manufactured to cells repeatedly. If these prerequisites are given, the focus of
the SSLPE growth can transition away from the growth mechanisms which are now fully understood
and are presented in this work.
As a new goal, optimization of the process towards a material stack and to interfaces suitable for cell
build-up can be set. Simultaneously, the electrical quality of the bulk material must be improved. Until
now, questions of growth-onset and general reproducibility were the center of the research. With the
high reproducibility reached, the process can now be adapted carefully to grow layers of higher
crystalline quality. This can be controlled by DLTS and Van-der-Pauw measurements. As a possible
perspective, slower growth rates can be targeted, so that the solution and dissolution reaction on the
substrate surface happen closer to thermodynamic equilibrium, giving the lattice more time to evolve
and thus achieve better epitaxy and higher crystalline perfection. The technical solution is to reduce
the temperature difference between silicon source and substrate during growth. This might lead to
unsteady or interrupted flux of silicon atoms in the melt in the current setup [27]. An alternative would
be to solve lower concentrations of silicon mobility in the Sn solution, by reducing temperature during
growth, unfortunately leading to similar problems. Both would require a larger crucible with less
temperature deviations induced by the heating system. A larger crucible, both deeper and wider would
allow for the temperature gradient to confine itself more evenly and thus for a more evenly spread
material flux.
Numerous process improvements have enabled deep research into the nature of SSLPE. This work
presents research into the growth phenomena during SSLPE growth and develops a new understanding
of the dynamic interaction between the seed and the SSLPE growth layer. This knowledge has been
applied to produce functional solar cells from SSLPE material for the first time.
- 104 -
7. Acknowledgements
7.
Acknow
l
edgemen
ts
I thank Torsten Boeck for his leadership and the freedom he gave me during my time working in his
group. I have learned much more from him than physics and material science, especially how
important it is to focus on diplomacy and advertising to third parties next to science itself in a career.
His group SiGe-Nanostructures has helped me tremendously during my PhD-time. Thomas Teubner
always gave meticulous professional consultation and kept the SSLPE-cluster tool running together
with Hans-Peter Schramm, for which I am grateful to both. With Owen Ernst connects me a love for
odd peritectic systems, a start-up project and a more productive relationship than I ever had. Katharina
Eylers and Roman Bansen have become friends and I enthusiastically follow their careers to see which of
their paths would work or not work for me. I thank Stefan Kayser who aides his colleagues
relentlessly with his sharp scientific mind and who made parts of this thesis possible. I also thank
Christian Ehlers, Yujia Liu, Julian Stöver, Lucinda Matiwe and Andreas Fielder for their contributions to
this work and friendly discussions during coffee breaks. Sabine Bergmann and Raimund Grünberg were a
great support with various measurements. Klaus Böttcher and Setareh Zahedi-Azad contributed by
correctional reading of this thesis. An ongoing conversation and collaboration with Helmholtz-Zentrum
Berlin (HZB) made the presented prototype solar cells possible, and I thank Cham Thi Trinh, Daniel
Amkreutz and Darja Erfurt for their support. Despite being critical, Thomas Schröder supports and
trusts me in producing isotope-pure silane within the IKZ and I could not be more humbled.
A big thank you goes to my reviewers Aleksander Gurlo, Astrid Haibel and Matthias Bickermann for
their interest in my work and their willingness to take the effort of reviewing this thesis, and to Claudia
Fleck for presiding over the commission.
I want to thank my friends and family for all the support I got during my life, my PhD-time and especially
during the last year when I needed them more than ever. Especially I thank Insa Tusch for sticking with
me. Bianka Koschke and parents gave me a roof over my head without further questioning, thank you
to you all. Despite not being in my life anymore, I thank Anne Lindner and Laura Grimm for sharing
paths for the longest time, I am sorry for the pain I caused and grateful for all the support they gave
me. My grandmother passed two years ago, but I want to thank her for always offering a warm and
save haven and for being the backbone of my family.
- 105 -
8. Literature
8.
L
it
era
t
ure
[1] “A Global Scorecard for Policy Makers,” Washington, 2017.
[2] J. Poushter and R. Stewart, “Smartphone Ownership and Internet Usage Continues to Climb in
Emerging Economies,” Washington; D.C., 2016.
[3] Ash Turner, “How Many Smartphones Are In The World?,” Bankmycell. [Online]. Available:
https://www.bankmycell.com/blog/how-many-phones-are-in-the-world. [Accessed: 15-Feb-
2022].
[4] “Global Tracking Framework 2017: Progress Towards Sustainable Energy,” 2017.
[5] S. Evans, “Solar is now ‘cheapest electricity in history’, confirms IEA,” Carbon Brief, 2020.
[Online]. Available: https://www.carbonbrief.org/solar-is-now-cheapest-electricity-in-history-
confirms-iea. [Accessed: 21-Feb-2022].
[6] “IEA (2020),”World Energy Outlook 2020, 2020. [Online]. Available:
https://www.iea.org/reports/world-energy-outlook-2020.
[7] F. Mahdavian, S. Platt, M. Wiens, M. Klein, and F. Schultmann, “Communication blackouts in
power outages: Findings from scenario exercises in Germany and France,” Int. J. Disaster Risk
Reduct., vol. 46, no. April, p. 101628, 2020.
[8] IEA - International Energy Agency, “Redrawing the Energy-Climate Map: World Energy Outlook
Special Report,” 2013.
[9] M. Jaganmohan, “Global climate change - statistics & facts,” Global climate change, 2022.
[Online]. Available: https://www.statista.com/topics/1148/global-climate-
change/#dossierKeyfigures. [Accessed: 15-Feb-2022].
[10] W. Steffen et al., “Trajectories of the Earth System in the Anthropocene,” Proc. Natl. Acad. Sci.
U. S. A., vol. 115, no. 33, pp. 8252–8259, 2018.
[11] F. Wulff, “Versorgungsunterbrechungen Strom 2020,” Bonn, 2021.
[12] VDMA Photovoltaic Equipment, “International Technology Roadmap for Photovoltaic (ITRPV) -
2018 Results,” 2019.
[13] H. Akbari et al., “Pilot Line Results of NType IBC Cell Process in Mass Production Environment,”
Sol. Energy Mater. Sol. Cells, vol. 9, no. 1, pp. 153–158, 2012.
[14] M. H. Shubbak, “Advances in solar photovoltaics: Technology review and patent trends,”
- 106 -
8. Literature
Renew. Sustain. Energy Rev., vol. 115, no. July, p. 109383, 2019.
[15] M. Imaizumi, T. Ito, M. Yamaguchi, and K. Kaneko, “Effect of grain size and dislocation density
on the performance of thin film polycrystalline silicon solar cells,” J. Appl. Phys., vol. 81, no. 11,
pp. 7635–7640, 1997.
[16] J. Kroon, “Cheetah - Project Final Report,” 2017.
[17] H. Sivaramakrishnan Radhakrishnan et al., “Freestanding and supported processing of sub-70
μm kerfless epitaxial Si and thinned Cz/FZ Si foils into solar cells: An overview of recent progress
and challenges,” Sol. Energy Mater. Sol. Cells, vol. 203, no. July, p. 110108, 2019.
[18] A. G. Aberle, “Crystalline Silicon Solar Cells : A Review,” Prog Photovolt Photovolt, vol. 8, no.
February, pp. 473–487, 2000.
[19] J. Haschke, D. Amkreutz, and B. Rech, “Liquid phase crystallized silicon on glass: Technology,
material quality and back contacted heterojunction solar cells,” Jpn. J. Appl. Phys., vol. 55, no.
4, 2016.
[20] R. Bansen et al., “Crystalline silicon on glass by steady-state solution growth using indium as
solvent,” Appl. Phys. A, vol. 119, no. 4, pp. 1577–1586, Jun. 2015.
[21] T. Boeck, T. Teubner, K. Schmidt, and P. M. Wilde, “A method to grow silicon crystallites on
glass,” J. Cryst. Growth, vol. 198–199, no. PART I, pp. 420–424, 1999.
[22] R. Bansen, C. Ehlers, T. Teubner, T. Markurt, J. Schmidtbauer, and T. Boeck, “Continuous
polycrystalline silicon layers on glass grown from tin solutions,” CrystEngComm, vol. 18, pp.
1911–1917, 2016.
[23] T. Boeck et al., “Kristallisationsverfahren zur Erzeugung kristalliner Halbleiterschichten,” DE 10
2010 044 014 A1, 2010.
[24] T. Teubner, T. Boeck, and K. Schmidt, “Modelling of solution growth of silicon from small indium
droplets homogeneous nucleation,” J. Cryst. Growth, vol. 198–199, no. PART I, pp. 425–429,
1999.
[25] R. Heimburger, “Solution Growth of Microcrystalline Silicon on Amorphous Substrates,”
Humboldt-Universität zu Berlin, 2010.
[26] T. Teubner, R. Heimburger, K. Böttcher, T. Boeck, and R. Fornari, “Equipment for low
temperature steady-state growth of silicon from metallic solutions,” Cryst. Growth Des., vol. 8,
- 107 -
8. Literature
no. 7, pp. 2484–2488, 2008.
[27] R. Bansen, “Solution Growth of Polycrystalline Silicon on Glass using Tin and Indium as
Solvents,” Humboldt-Universität zu Berlin, 2016.
[28] B. J. Baliga, “Silicon Liquid Phase Epitaxy: A Review,” J. Electrochem. Soc., vol. 133, no. 1, pp. 5C-
14C, 1986.
[29] M. G. Mauk, “Liquid-Phase Epitaxy,” in Handbook of Crystal Growth, Elsevier, 2015, pp. 225–
316.
[30] B. C. Ehlers, “Growth and Characterization of Silicon on Glass and Reorganized Porous Silicon
Substrates for Photovoltaic Applications,” Technische Universität Berlin, 2020.
[31] A. Stapf, C. Gondek, E. Kroke, and G. Roewer, “Wafer Cleaning, Etching, and Texturization,” in
Handbook of Photovoltaic Silicon, D. Yang, Ed. Berlin, Heidelberg: Springer Berlin Heidelberg,
2019, pp. 311–358.
[32] Y. Azuma et al., “Improved measurement results for the Avogadro constant using a 28 Si-
enriched crystal,” Metrologia, vol. 52, p. 360, 2015.
[33] C. A. Sanchez, B. M. Wood, R. G. Green, J. O. Liard, and D. Inglis, “A determination of Planck ’ s
constant using the NRC watt balance,” Metrologia, vol. 51, no. 2, 2014.
[34] A. Pramann and O. Rienitz, “Mass Spectrometric Investigation of Silicon Extremely Enriched in
28 Si: From 28 SiF 4 (Gas Phase IRMS) to 28 Si Crystals (MC-ICP-MS),” Anal. Chem., vol. 4, 2016.
[35] N. N. Greenwood and A. Earnshaw, Chemie der Elemente, 1. Aufl.,. Weinheim: VCH, 1990.
[36] J. Hartmann-Schreier and H. F. Krug, “Siliciumdioxid,” Thieme Römpp, 2016. [Online]. Available:
https://roempp.thieme.de/lexicon/RD-19-02447. [Accessed: 22-Mar-2022].
[37] M. Morita, T. Ohmi, E. Hasegawa, M. Kawakami, and M. Ohwada, “Growth of native oxide on a
silicon surface,” J. Appl. Phys., vol. 68, no. 3, pp. 1272–1281, 1990.
[38] H. Ohsaki, K. Miura, and Y. Tatsumi, “Oxidation mechanism of amorphous silicon in air,” J. Non.
Cryst. Solids, vol. 93, no. 2–3, pp. 395–406, Sep. 1987.
[39] R. Bansen et al., “Crystalline silicon on glass by steady-state solution growth using indium as
solvent,” Appl. Phys. A Mater. Sci. Process., vol. 119, no. 4, pp. 1577–1586, 2015.
[40] D. Uebel, O. Ernst, R. Bansen, T. Teubner, H.-P. Schramm, and T. Boeck, “Silizium-basierte Wafer
und Verfahren zur Herstellung von Silizium-basierten Wafern,” DE 10 2020 132 900.8, 2020.
- 108 -
8. Literature
[41] C. Ehlers et al., “In situ removal of a native oxide layer from an amorphous silicon surface with
a UV laser for subsequent layer growth,” CrystEngComm, vol. 20, no. 44, pp. 7170–7177, 2018.
[42] K. Kawase, J. Tanimura, H. Kurokawa, and K. Wakao, “XPS Study of H-Terminated Silicon Surface
under Inert Gas and UHV Annealing,” pp. 163–167, 2005.
[43] S. R. Kasi et al., “Hydrocarbon reaction with HFcleaned Si (100) and effects on metaloxide
semiconductor device quality Hydrocarbon reaction with HF-cleaned Si (lOQ) and effects on
metal-oxide-semiconductor device quality,” vol. 59, pp. 107–110, 1991.
[44] G. R. Gould, “The LASER, Light Amplification by Stimulated Emission of RadiationNo Title,” in
The Ann Arbor Conference on Optical Pumping, the University of Michigan, 15 June through 18
June 1959., 1959, p. 128.
[45] O. C. Ernst, D. Uebel, S. Kayser, F. Lange, T. Teubner, and T. Boeck, “Revealing all states of
dewetting of a thin gold layer on a silicon surface by nanosecond laser conditioning,” Appl. Surf.
Sci. Adv., vol. 3, no. August 2020, 2021.
[46] D. Sands, “Pulsed Laser Heating and Melting,” in Heat Transfer - Engineering Applications, V.
Vikhrenko, Ed. InTech, 2011, pp. 47–53.
[47] H.-G. Rubahn, Laser applications in surface science and technology. Chichester: Wiley, 1999.
[48] R. Larciprete, E. Borsella, and P. Cinti, “KrF-excimer-laser-induced native oxide removal from Si
(100) surfaces studied by Auger electron spectroscopy,” Appl. Phys. A Mater. Sci. Process., vol.
62, no. 2, pp. 103–114, Feb. 1996.
[49] B. N. Chichkov, C. Momma, S. Nolte, F. von Alvensleben, and A. Tünnermann, “Femtosecond,
picosecond and nanosecond laser ablation of solids,” Appl. Phys. A Mater. Sci. Process., vol. 63,
no. 2, pp. 109–115, 1996.
[50] D. M. Zehner, C. W. White, G. W. Ownby, D. M. Zehner, C. W. White, and G. W. Ownby,
“Preparation of atomically clean silicon surfaces by pulsed laser irradiation,” Appl. Phys. Lett.,
vol. 56, no. 1, p. 56, 1980.
[51] “COMSOL Multiphysics Reference Manual.” COMSOL, 2019.
[52] D. Uebel, “Laser-induced Removal of SiOx from Si Seed Structures for Epitaxial Growth of
Absorber Layers for Low Cost Solar Cells,” Technische Universität Berlin, 2017.
[53] H. Habuka, H. Tsunoda, M. Mayusumi, N. Tate, and M. Katayama, “Roughness of Silicon Surface
- 109 -
8. Literature
Heated in Hydrogen Ambient,” J. Electrochem. Soc., vol. 142, no. 9, pp. 3092–3098, 1995.
[54] A. Ishitani, Y. Ohshita, K. Tanigaki, K. Takada, and S. Itoh, “Prebaking and silicon epitaxial growth
enhanced by UV radiation,” J. Appl. Phys., vol. 61, no. 6, pp. 2224–2229, 1987.
[55] J. Cressler, SiGe and Si Strained-Layer Epitaxy for Silicon Heterostructure Devices. CRC Press,
2007.
[56] N. Miyata, M. Shigeno, Y. Arimoto, and T. Ito, “Thermal decomposition of native oxide on Si
(100),” J. Appl. Phys., vol. 74, p. 5275, 1993.
[57] H. E. Sasse and U. König, “SiO diffusion during thermal decomposition of SiO2,” J. Appl. Phys.,
vol. 67, no. 10, pp. 6194–6196, 1990.
[58] I. Barin and O. Knacke, Thermochemical Properties of Inorganic Substances. Berlin, Düsseldorf:
Springer Verlag, Berlin und Verlag Stahleisen mbH, Düsseldorf, 1974.
[59] P. W. Atkins and A. Höpfner, Physikalische Chemie, 1. Aufl.,. Weinheim: VCH, 1990.
[60] B. M. Gallois, T. M. Besmann, and M. W. Stott, “Chemical Etching of Silicon (100) by Hydrogen,”
J. Am. Ceram. Soc., vol. 77, no. 11, pp. 2949–2952, 1994.
[61] G. Liu and S. J. Fonash, “Selective area crystallization of amorphous silicon films by low-
temperature rapid thermal annealing,” Appl. Phys. Lett., vol. 660, no. June 1989, pp. 1–4, 1989.
[62] R. Kakkad, J. Smith, W. S. Lau, S. J. Fonash, and R. Kerns, “Crystallized Si films by low-
temperature rapid thermal annealing of amorphous silicon,” J. Appl. Phys., vol. 65, no. August,
p. 2069, 1989.
[63] R. B. Bergmann, G. Oswald, M. Albrecht, and J. Werner, “Polycrystalline Silicon on Glass
Substrates for Thin Film Solar Cells,” Solid State Phenom., vol. 51–52, pp. 515–520, 1996.
[64] K. Zellama, P. Germain, S. Squelard, J. C. Bourgoin, and P. A. Thomas, “Crystallization in
amorphous silicon,” J. Appl. Phys., vol. 50, no. July, p. 6995, 1979.
[65] A. R. Zanatta and F. A. Ferri, “Metal-Induced Crystallization by Homogeneous Insertion of
Metallic Species in Amorphous Semiconductors,”in Metal-Induced Crystallization
Fundamentals and Applications, 1st ed., Z. Wang, L. P. H. Jeurgens, and E. J. Mittemeijer, Eds.
Pan Stanford Publishing, 2015, pp. 113–132.
[66] Z. Wang, L. P. H. Jeurgens, and E. J. Mittemeijer, “Thermodynamics and Atomic Mechanisms of
Metal-induced Crystallization of Amorphous Semiconductors at Low Temperatures,” in Metal-
- 110 -
8. Literature
Induced-Crystallization-Fundamentals-and-Applications, 1st ed., Z. Wang, L. P. H. Jeurgens, and
E. J. Mittemeijer, Eds. Pan Stanford Publishing, 2015, pp. 25–55.
[67] W. R. Runyan, Silicon Semiconductor Technology. New York: McGraw-Hill Book Company, 1965.
[68] R. Heimburger, N. Deßmann, T. Teubner, H. P. Schramm, T. Boeck, and R. Fornari,
“Polycrystalline silicon films on glass grown by amorphous-liquid- crystalline transition at
temperatures below 330 °C,” Thin Solid Films, vol. 520, no. 6, pp. 1784–1788, 2012.
[69] “Hewlett-Packard Graphics Language Manual Revision C.” Hewlett Packard, 1999.
[70] L. Calcagnile, “Laser Induced Phase Transitions in Si and Ge Implanted Si Substrates,” Phys.
status solidi, vol. 23, pp. 23–32, 1995.
[71] E. P. Donovan, F. Spaepen, D. Turnbull, J. M. Poate, and D. C. Jacobson, “Heat of crystallization
and melting point of amorphous silicon,” Appl. Phys. Lett., vol. 42, no. 8, p. 698, 1983.
[72] M. C. Wingert, J. Zheng, S. Kwon, and R. Chen, “Thermal transport in amorphous materials: a
review,” Semicond. Sci. Technol., vol. 31, no. 11, p. 113003, 2016.
[73] C. J. Shih et al., “Phase field modeling of excimer laser crystallization of thin silicon films on
amorphous substrates,” vol. 100, no. 5, p. 053504, 2006.
[74] H. Nagai et al., “Thermal Conductivity Measurement of Molten Silicon by a Hot-Disk Method in
Short-Duration Microgravity Environments,” Jpn. J. Appl. Phys., vol. 39, no. 3R, p. 1405, 2000.
[75] Corning Display Technologies, “EAGLE XG ® Slim Glass Product Information Sheet,” no. August.
pp. 1–2, 2013.
[76] Corning, “Order Corning Eagle XG.” Corning, pp. 1–5, 2019.
[77] Y. Okada and Y. Tokumaru, “Precise determination of lattice parameter and thermal expansion
coefficient of silicon between 300 and 1500 K,” J. Appl. Phys., vol. 56, no. June 1984, p. 314,
1984.
[78] T. Matsushita and H. Tayanaka, “Method for separating a device-forming layer from a base
body,” US RE42,830 E, 1998.
[79] R. B. Bergmann, T. J. Rinke, R. M. Hausner, M. Grauvogl, M. Vetter, and J. H. Werner, “Thin film
solar cells on glass by transfer of monocrystalline Si films,” Int. J. Photoenergy, vol. 1, no. 2, pp.
1–5, 1999.
[80] J. Govaerts et al., “Solar Cells from Epitaxial Foils: An Epifoil Epiphany,” Energy Procedia, vol.
- 111 -
8. Literature
77, pp. 871–880, 2015.
[81] V. Depauw, I. Gordon, G. Beaucarne, J. Poortmans, R. Mertens, and J. P. Celis, “Large-area
monocrystalline silicon thin films by annealing of macroporous arrays: Understanding and
tackling defects in the material,” J. Appl. Phys., vol. 106, no. 3, 2009.
[82] V. Depauw, Y. Qiu, K. van Nieuwenhuysen, I. Gordon, and J. Poortmans, “Epitaxy-free
monocrystalline silicon thin film: First steps beyond proof-of-concept solar cells,” Prog.
Photovoltaics Res. Appl., vol. 19, no. 7, pp. 844–850, 2011.
[83] J. I. Goldstein, D. E. Newbury, J. R. Michael, N. W. M. Ritchie, J. H. J. Scott, and D. C. Joy, Scanning
Electron Microscopy and X-Ray Microanalysis. New York, NY: Springer New York, 2018.
[84] H. J. G. Gunderson, “Stereology of arbitrary particles *,” J. Microsc., vol. 143, no. May 1985, pp.
3–45, 1986.
[85] H. J. G. Gunderson, “Notes on the estimation of the numerical density of arbitrary profiles : the
edge effect,” J. Microsc., vol. 111, no. 15, pp. 219–223, 1977.
[86] V. Howard, “Unbiased estimation of particle density in the tandem scanning reflected light
microscope,” J. Microsc., vol. 138, no. May, pp. 203–212, 1985.
[87] J. C. Harmand et al., “Atomic Step Flow on a Nanofacet,” Phys. Rev. Lett., vol. 121, no. 16, pp.
1–5, 2018.
[88] Z. Iqbal and S. Veprek, “Raman scattering from hydrogenated microcrystalline and amorphous
silicon,” J. Phys. C Solid State Phys., vol. 15, pp. 377–392, 1982.
[89] S. Veprek, F.-A. Sarott, and Z. Iqbal, “Effect of grain boundaries on the Raman spectra, optical
absorption, and elastic light scattering in nanometer-sized crystalline silicon,” Phys. Rev. B, vol.
36, no. 6, pp. 3344–3350, Aug. 1987.
[90] I. H. Campbell and P. M. Fauchet, “The effects of microcrystal size and shape on the one phonon
Raman spectra ofcrystalline semiconductors,” Semiconductors, vol. 58, no. 10, pp. 739–741,
1986.
[91] C. Smit, R. A. C. M. M. Van Swaaij, H. Donker, A. M. H. N. Petit, W. M. M. Kessels, and M. C. M.
Van de Sanden, “Determining the material structure of microcrystalline silicon from Raman
spectra,” J. Appl. Phys., vol. 94, no. 5, pp. 3582–3588, 2003.
[92] C. T. Trinh et al., “Assessment of Bulk and Interface Quality for Liquid Phase Crystallized Silicon
- 112 -
8. Literature
on Glass,” IEEE J. Photovoltaics, vol. 9, no. 2, pp. 364–373, 2019.
[93] M. Trahms et al., “All-Thin-Film Tandem Cells Based on Liquid Phase Crystallized Silicon and
Perovskites,” IEEE J. Photovoltaics, vol. 9, no. 3, pp. 621–628, 2019.
[94] S. Garud et al., “Laser firing in silicon heterojunction interdigitated back contact architecture
for low contact resistance,” Sol. Energy Mater. Sol. Cells, vol. 203, no. May, p. 110201, 2019.
[95] C. Thi Trinh et al., “Potential of interdigitated back-contact silicon heterojunction solar cells for
liquid phase crystallized silicon on glass with efficiency above 14%,”Sol. Energy Mater. Sol. Cells,
vol. 174, no. June 2017, pp. 187–195, 2018.
[96] D. Amkreutz, J. Haschke, S. Kuhnapfel, P. Sonntag, and B. Rech, “Silicon thin-film solar cells on
glass with open-circuit voltages above 620 mV formed by liquid-phase crystallization,” IEEE J.
Photovoltaics, vol. 4, no. 6, pp. 1496–1501, 2014.
[97] R. Varache, C. Leendertz, M. E. Gueunier-Farret, J. Haschke, D. Muñoz, and L. Korte,
“Investigation of selective junctions using a newly developed tunnel current model for solar cell
applications,” Sol. Energy Mater. Sol. Cells, vol. 141, pp. 14–23, 2015.
[98] A. Alkauskas, M. D. McCluskey, and C. G. van de Walle, “Tutorial: Defects in semiconductors-
Combining experiment and theory,” J. Appl. Phys., vol. 119, no. 18, p. 181101, 2016.
[99] D. V. Lang, “Deep-level transient spectroscopy: A new method to characterize traps in
semiconductors,” J. Appl. Phys., vol. 45, no. 7, pp. 3023–3032, 1974.
[100] D. K. Schroder, Semiconductor Material and Device Characterization, 3rd Edition, 3rd ed. IEEE
Press and Wiley, 2015.
[101] K. Dornich, N. Schüler, B. Berger, and J. R. Niklas, “Fast, high resolution, inline contactless
electrical semiconductor characterization for photovoltaic applications by microwave detected
photoconductivity,” Mater. Sci. Eng. B Solid-State Mater. Adv. Technol., vol. 178, no. 9, pp. 676–
681, 2013.
[102] J. A. Hornbeck and J. R. Haynes, “Trapping of Minority Carriers in Silicon. I. p-Type Silicon,” Phys.
Rev., vol. 97, no. 2, pp. 311–321, Jan. 1955.
[103] A. Ali, T. Gouveas, M. A. Hasan, S. H. Zaidi, and M. Asghar, “Influence of deep level defects on the
performance of crystalline silicon solar cells: Experimental and simulation study,” Sol. Energy
Mater. Sol. Cells, vol. 95, no. 10, pp. 2805–2810, 2011.
- 113 -
8. Literature
[104] S. A. McHugo, H. Hiesimair, and E. R. Weber, “Gettering of metallic impurities in photovoltaic
silicon,” Appl. Phys. A Mater. Sci. Process., vol. 64, no. 2, pp. 127–137, 1997.
[105] S. Weiss and R. Kassing, “Deep Level Transient Fourier Spectroscopy (DLTFS)---A technique for
the analysis of deep level properties,” Solid. State. Electron., vol. 31, no. 12, pp. 1733–1742,
1988.
[106] R. A. Sinton and A. Cuevas, “A quasi-steady-state open-circuit voltage method for solar cell
characterization,” 16th Eur. Photovolt. Sol. Energy Conf., no. May, pp. 1–4, 2000.
[107] M. V. Peralta-Martinez and W. A. Wakeham, “Thermal conductivity of liquid tin and indium,”
Int. J. Thermophys., vol. 22, no. 2, pp. 395–403, 2001.
[108] G. W. . Kaye and T. . Laby, Tables of Physical and Chemical Constants. London: Longmans,Green
and Co, 1911.
[109] Y. He, D. Donadio, and G. Galli, “Heat transport in amorphous silicon: Interplay between
morphology and disorder,” Appl. Phys. Lett., vol. 98, no. 14, pp. 1–4, 2011.
[110] J. Y. Wang, Z. Wang, L. P. H. Jeurgens, and E. J. Mittemeijer, “Diffusion, Crystallization, and Layer
Exchange upon Low-Temperature Annealing of Amorphous Si/ Polycrystalline Al–Layered
Structures,” in Metal-Induced-Crystallization-Fundamentals-and-Applications, 1st ed., Z. Wang,
L. P. H. Jeurgens, and E. J. Mittemeijer, Eds. Pan Stanford Publishing, 2015, pp. 61–102.
[111] C. Ehlers, R. Bansen, T. Markurt, D. Uebel, T. Teubner, and T. Boeck, “Solution growth of Si on
reorganized porous Si foils and on glass substrates,” J. Cryst. Growth, vol. 468, pp. 268–271,
2017.
[112] R. J. Jaccodine, “Surface Energy of Germanium and Silicon Surface Energy of Germanium and
Silicon,” J. Electrochem. Soc., vol. 110, p. 524, 1963.
[113] R. H. Glaenzer and A. G. Jordan, “The electrical properties of dislocations in silicon,” Solid. State.
Electron., vol. 12, pp. 247–258, 1969.
[114] V. V. Kveder, Y. A. Osipyan, W. Schröter, and G. Zoth, “On the Energy Spectrum of Dislocations
in Silicon,” Phys. status solidi, vol. 72, no. 2, pp. 701–713, 1982.
[115] J. L. Hastings, S. K. Estreicher, and P. A. Fedders, “Vacancy aggregates in silicon,” Mater. Sci.
Forum, vol. 258–263, no. PART 1, pp. 509–514, 1997.
[116] G. Pfeiffer, T. Prescha, and J. Weber, “Transition metal deep level defects in silicon,” Jpn. J. Appl.
- 114 -
8. Literature
Phys., vol. 32, no. S1, pp. 239–241, 1993.
[117] R. H. Hopkins and A. Rohatgi, “Impurity effects in silicon for high efficiency solar cells,” J. Cryst.
Growth, vol. 75, no. 1, pp. 67–79, 1986.
[118] P. Pellegrino, J. Lalita, A. Halle, P. Le, C. Jagadish, and B. G. Svensson, “Annealing kinetics of
vacancy-related defects in low-dose MeV self-ion-implanted n -type silicon,” Phys. Rev. B, vol.
64, pp. 1–10, 2001.
[119] A. Hallén, B. U. R. Sundqvist, Z. Paska, B. G. Svensson, M. Rosling, and J. Tirén, “Deep level
transient spectroscopy analysis of fast ion tracks in silicon,” J. Appl. Phys., vol. 67, no. 3, pp.
1266–1271, Feb. 1990.
[120] B. G. Svensson, A. Hallén, and B. U. R. Sundqvist, “Hydrogen-related electron traps in proton-
bombarded float zone silicon,” Mater. Sci. Eng. B, vol. 4, no. 1–4, pp. 285–289, Oct. 1989.
[121] R. Radu, I. Pintilie, L. C. Nistor, E. Fretwurst, G. Lindstroem, and L. F. Makarenko, “Investigation
of point and extended defects in electron irradiated silicon - Dependence on the particle
energy,” J. Appl. Phys., vol. 117, no. 16, 2015.
[122] P. Rosenits, T. Roth, S. W. Glunz, and S. Beljakowa, “Determining the defect parameters of
thedeep aluminum-related defect center insilicon,” Appl. Phys. Lett., vol. 91, no. September,
2007.
[123] E. V. Monakhov, A. Ulyashin, G. Alfieri, A. Y. Kuznetsov, B. S. Avset, and B. G. Svensson,
“Divacancy annealing in Si: Influence of hydrogen,” Phys. Rev. B - Condens. Matter Mater. Phys.,
vol. 69, no. 15, pp. 1–4, 2004.
[124] A. Peaker et al., “High Resolution Laplace Deep Level Transient Spectroscopy A New Tool To
Study Implant Damage In Silicon,” 2002.
[125] S. M. Sze and K. K. Ng, Physics of Semiconductor Devices, 3rd ed. Wiley, 1969.
[126] S. Garud et al., “Toward High Solar Cell Efficiency with Low Material Usage: 15% Efficiency with
14 μm Polycrystalline Silicon on Glass,” Sol. RRL, vol. 4, no. 6, pp. 1–8, 2020.
[127] M. A. Easton, M. A. Qian, and D. H. StJohn, “Grain Refinement in Alloys: Novel Approaches,”
Ref. Modul. Mater. Sci. Mater. Eng., pp. 1–7, 2016.
- 115 -
9. List of figures
9.
L
ist
of
f
i
gure
s
F
ig
u
re
1:
Mo
tiv
a
ti
on fo
r
SSL
P
E-
gr
o
wt
h of S
i
on
gl
ass fo
r
P
V. (a) Con
ve
n
ti
ona
l
abso
r
b
er
s on
ly
us
e
a
f
r
a
cti
on of
t
h
e
ma
teri
a
l
i
n
ve
s
te
d
,
w
h
ile
mos
t
of
t
h
e
us
e
d ma
teri
a
l
i
s
electric
a
lly
i
na
ctive
du
e
t
o sun
lig
h
t
no
t
re
a
c
h
i
n
g
it
o
r
l
os
t
du
ri
n
g
sa
wi
n
g
. Th
e
re
ma
i
n
i
n
g
ma
teri
a
l
i
s on
ly
us
e
d fo
r
m
ec
han
ic
a
l
s
t
ab
ility
o
r
i
s
l
os
t
du
ri
n
g
s
ewi
n
g
(b) SSL
P
E off
er
s a bo
tt
om up app
r
oa
c
h
t
o
gr
o
w
h
ig
h-qua
lity
cry
s
t
a
lli
n
e
S
i
on an
amo
r
phous subs
tr
a
te
. Thus
,
gl
ass
c
an b
e
us
e
d fo
r
m
ec
han
ic
a
l
s
t
ab
ility,
w
h
ic
h
i
s mu
c
h
c
h
e
ap
er
t
han
so
l
a
r
-
gr
ad
e
S
i
(
c
) SSL
P
E
re
s
e
mb
le
s
t
h
e
f
l
oa
t
-
gl
ass p
r
o
ce
ss
w
h
ic
h p
r
om
i
s
e
s a h
ig
h s
c
a
li
n
g
po
te
n
ti
a
l
.
Du
ri
n
g
t
h
e
f
l
oa
t
-
gl
ass p
r
o
ce
ss a
gl
ass m
elt
i
s pou
re
d on
t
o a b
e
d of mo
lte
n m
et
a
l,
on
w
h
ic
h
t
h
e
gl
ass
c
oo
l
s and so
li
d
i
f
ie
s
,
i
nh
eriti
n
g
t
h
e
f
l
a
t
n
e
ss of
t
h
e
m
et
a
l
su
r
fa
ce
.
A
f
ter
suff
icie
n
t
so
li
d
i
f
ic
a
ti
on
,
t
h
e
gl
ass
c
an b
e
c
u
t
i
n
t
o
l
a
rge
sh
eet
s. ...................................................................................................................
7
F
ig
u
re
2
G
a-
A
s (a) and Sn-S
i
(b) a
re
e
x
e
mp
l
a
ry
ma
teri
a
l
s
y
s
te
ms fo
r
L
P
E.
A
s
i
s so
lve
d
i
n
G
a m
elt,
o
r
S
i
i
n
Sn m
elt
and
t
h
e
m
i
x
t
u
re
(
1
)
i
s
c
oo
le
d (
2
) do
w
n so
t
ha
t
G
a
A
s o
r
ele
m
e
n
t
a
ry
S
i
p
reci
p
it
a
te
s...............
10
F
ig
u
re
3
SSL
P
E of S
i
i
s a
t
h
ree
-s
te
p p
r
o
ce
ss of subs
tr
a
te
p
re
pa
r
a
ti
on b
y
P
VD (a)
,
ox
i
d
e
re
mo
v
a
l
b
y
v
a
ri
ous m
et
hods (b) and
t
h
e
SSL
P
E
gr
o
wt
h-s
te
p
it
s
el
f (
c
).
U
n
like
L
P
E of
G
a
A
s
it
i
s a
c
h
ieve
d b
y
s
te
ad
ily
d
i
sso
lvi
n
g
a S
i
sou
rce
and p
reci
p
it
a
ti
on of S
i
cry
s
t
a
llite
s on
t
h
e
subs
tr
a
te
...........................................
10
F
ig
u
re
4
SSL
P
E u
tili
z
e
s a
te
mp
er
a
t
u
re
d
e
p
e
nd
e
n
t
so
lv
ab
ility
fo
r
t
h
e
d
e
pos
iti
on f
r
om so
l
u
ti
on.
I
f
t
op and
bo
tt
om
te
mp
er
a
t
u
re
s a
re
i
d
e
n
tic
a
l,
fo
rw
a
r
d and ba
ckw
a
r
d
re
a
cti
on a
re
i
d
e
n
tic
a
l
a
t
bo
t
h su
r
fa
ce
s..
13
F
ig
u
re
5:
T
y
p
ic
a
l
te
mp
er
a
t
u
re
c
on
tr
o
l
du
ri
n
g
p
r
o
ce
ss
i
n
g
i
n
t
h
e
i
nn
er
c
hamb
er
. H
e
a
t
up of
t
h
e
m
elt
(
I
)
,
s
e
d
i
m
e
n
t
a
ti
on mod
e
(
II
)
,
gr
o
wt
h mod
e
(
III
)
,
s
e
d
i
m
e
n
t
a
ti
on mod
e
(
I
V) and
c
oo
l
do
w
n (V). T
y
p
ic
a
l
i
s a
bas
e
te
mp
er
a
t
u
re
of
600
°
C a
t
t
h
e
sou
rce
wit
h
te
mp
er
a
t
u
re
d
i
ff
ere
n
ce
s fo
r
s
e
d
i
m
e
n
t
a
ti
on and
gr
o
wt
h
of
5
K
......................................................................................................................................................
13
F
ig
u
re
6:
Th
e
t
h
ree
d
i
ff
ere
n
t
m
et
hods app
lie
d fo
r
ox
i
d
e
c
on
tr
o
l
. (a) M
elt
-ba
ck
w
as
t
h
e
f
ir
s
t
su
cce
ssfu
lly
i
mp
le
m
e
n
te
d m
et
hod. H
ere,
i
ns
erti
on of
t
h
e
samp
le
du
ri
n
g
s
e
d
i
m
e
n
t
a
ti
on d
i
sso
lve
s
t
h
e
upp
er
l
a
yer
of
t
h
e
subs
tr
a
te
and
t
hus a
l
so
t
h
e
ox
i
d
e
. (b) HF-d
i
pp
i
n
g
etc
h
e
s
t
h
e
ox
i
d
e
s
electively
and pass
iv
a
te
s
t
h
e
su
r
fa
ce
. (
c
) S
c
ann
i
n
g
t
h
e
su
r
fa
ce
wit
h a
U
V-
l
as
er
h
e
a
t
s up
t
h
e
su
r
fa
ce
and b
ri
n
g
s
t
h
e
ox
i
d
e
t
o
re
a
cti
on
wit
h
t
h
e
c
hamb
er
a
t
mosph
ere
..............................................................................................................
15
F
ig
u
re
7
C
ritic
a
l
te
mp
er
a
t
u
re
subj
ecte
d
t
o amb
ie
n
t
p
re
ssu
re
i
n
re
f
ere
n
ce
t
o
t
h
e
c
ond
iti
ons
i
n
t
h
e
gr
o
wt
h
c
hamb
er
.
Alt
hou
g
h
t
h
e
c
hamb
er
i
s f
l
ush
e
d
wit
h H2 a
t
1050
mba
r,
t
h
e
pa
rti
a
l
p
re
ssu
re
of an
y
o
t
h
er
c
ompon
e
n
t
i
s
e
xp
ecte
d
t
o
c
o
rrel
a
te
t
o
t
h
e
v
a
c
uum p
re
ssu
re
b
e
fo
re
t
h
e
f
l
ood
i
n
g
a
r
ound
10
-
7 mba
r
e
x
e
mp
l
a
rily
. T
e
mp
er
a
t
u
re
s abo
ve
t
h
e
critic
a
l
v
a
l
u
e
s a
re
e
xp
ecte
d
t
o fa
v
o
r
t
h
e
p
r
odu
ct
s of
t
h
e
re
a
cti
on. ..........................................................................................................................................
22
- 116 -
9. List of figures
F
ig
u
re
8
Sho
w
s p
l
o
t
s of
e
qua
ti
on (
26
) so
lve
d
wit
h Ma
t
hLab fo
r
t
h
e
su
r
fa
ce
te
mp
er
a
t
u
re
o
ver
ti
m
e
(a)
and fo
r
t
h
e
te
mp
er
a
t
u
re
i
n
t
h
e
d
e
p
t
h of
t
h
e
sp
eci
m
e
n
,
a
t
t
h
e
e
nd of a pu
l
s
e
(
1
.
5
ns) and a
t
l
a
ter
ti
m
e
po
i
n
t
s (
3
.
0
ns and
9
.
0
ns).
N
a
t
u
r
a
lly,
t
h
e
d
i
s
tri
bu
ti
on
will
eve
n ou
t
wit
h
ti
m
e,
c
oo
li
n
g
t
h
e
ho
tter
and
w
a
r
m
i
n
g
t
h
e
c
oo
ler
s
ecti
ons. Ca
lc
u
l
a
ti
ons
were
don
e
wit
h a
l
as
er
i
n
te
ns
ity
of
I
0 =
0
.
4
W / m
²
and
t
h
e
abso
r
p
ti
on
c
o
e
ff
icie
n
t
=
1
.
47
×
10
-
9
1
/ m..........................................................................................
24
F
ig
u
re
9
S
ketc
h and band mod
el
of a s
i
mp
le
so
l
a
r
cell
wit
h a n-
ty
p
e
abso
r
b
er
(
lig
h
t
te
a
l
)
,
a p-
ty
p
e
e
m
itter
(da
rk
te
a
l
)
,
m
et
a
l
c
on
t
a
ct
s (
grey
) and fun
cti
ona
l
c
oa
ti
n
g
(b
l
u
e
). Th
e
e
n
ergy
d
i
ff
ere
n
ce
b
etwee
n
t
h
e
d
i
ff
ere
n
tly
dop
e
d s
e
m
ic
ondu
ct
o
r
s
c
aus
e
s b
e
nd
i
n
g
of
t
h
e
e
n
ergy
bands. Wh
e
n
lig
h
t
e
n
ter
s
t
h
e
abso
r
b
er
and
i
n
ter
a
ct
s
wit
h
electr
ons
,
t
h
ey
a
re
e
x
cite
d and
electr
on-ho
le
pa
ir
s a
re
cre
a
te
d. Th
e
po
te
n
ti
a
l
b
etwee
n n- and p-
ty
p
e
push
e
s pos
itive
electr
on ho
le
s
i
n
t
h
e
v
a
le
n
ce
band
t
o
t
h
e
f
r
on
t
c
on
t
a
ct
s and n
eg
a
tive
electr
ons
i
n
t
h
e
c
ondu
cti
on band
t
o
w
a
r
ds
t
h
e
ba
ck
s
i
d
e
c
on
t
a
ct,
cre
a
ti
n
g
electric
a
l
c
ha
rge
. ...................................................................................................................................
30
F
ig
u
re
10
s
ketc
h of
t
h
e
p
r
odu
ce
d p
r
o
t
o
ty
p
e
d
evice
s and band mod
el
s a
l
on
g
on
e
l
a
ter
a
l
and
tw
o
ho
ri
zon
t
a
l
c
u
t
s. Du
e
t
o
t
h
e
c
onf
ig
u
r
a
ti
on of
t
h
e
a
ctive
l
a
yer
(n-
ty
p
e
) b
ei
n
g
gr
o
w
n on a s
ilic
on
w
af
er,
fun
cti
ona
li
za
ti
on has
t
o b
e
don
e
jus
t
f
r
om
t
h
e
sunn
y
s
i
d
e
of
t
h
e
cell
. E
lectric
a
l
c
on
t
a
ct
s a
re
re
a
li
z
e
d
wit
h m
et
a
l
re
p
re
s
e
n
ti
n
g
t
h
e
ba
ck
s
i
d
e
and TCO on
t
h
e
p-
ty
p
e
e
m
itter
re
p
re
s
e
n
ti
n
g
t
h
e
f
r
on
t
s
i
d
e
of
t
h
e
cell
. Th
e
p-
ty
p
e
subs
tr
a
te
pass
iv
a
te
s
t
h
e
ba
ck
s
i
d
e
of
t
h
e
abso
r
b
er,
push
i
n
g
ba
ck
e
x
cite
d
electr
ons
,
letti
n
g
t
h
e
m d
ri
f
t
l
a
ter
a
lly
. Th
e
n
t
h
ey
a
re
c
o
llecte
d b
y
t
h
e
m
et
a
l
c
on
t
a
ct
s. Ho
le
s a
re
c
o
llecte
d b
y
t
h
e
e
m
itter
l
a
yer
and
c
ons
e
qu
e
n
tly
t
h
e
TCO
l
a
yer
. ....................................................................................
32
F
ig
u
re
11:
S
ketc
h of
t
h
e
cl
us
ter
t
oo
l,
ha
r
bo
ri
n
g
mos
t
i
mpo
rt
an
tly
t
h
e
P
VD
c
hamb
er
and
t
h
e
SSL
P
E
gr
o
wt
h
c
hamb
er
. Th
ey
a
re
c
onn
ecte
d b
y
an au
t
oma
te
d hand
li
n
g
s
y
s
te
m
,
t
ha
t
a
ll
o
w
s fo
r
tr
anspo
rt
t
h
r
ou
g
h
v
a
c
uum b
y
tw
o
r
obo
tic
a
r
ms and
tw
o
l
oad
l
o
ck
s.
A
d
et
a
ile
d d
e
s
cri
p
ti
on
i
s
give
n
i
n
t
h
e
te
x
t
.
35
F
ig
u
re
12
Las
er
b
e
am (pu
r
p
le
)
i
n a
tec
hn
ic
a
l
d
r
a
wi
n
g
of
t
h
e
op
tic
a
l
s
y
s
te
m and
i
n p
ri
n
ci
p
le
. (a)
l
as
er
sou
rce,
(b) m
irr
o
r
1,
(
c
)
iri
s d
i
aph
r
a
g
m
,
(d)
le
ns
1
(fo
c
us
le
n
gt
h f
1
=
250
mm)
,
(
e
) a
tte
nua
t
o
r,
(f) m
irr
o
r
2,
(
g
)
le
ns
2
(fo
c
us
le
n
gt
h f
2
=
400
mm)
,
(h) s
c
an h
e
ad
,
(
i
) m
irr
o
r
3,
(j) sp
eci
m
e
n
,
(
k
)
electr
on
ic
s.......
40
F
ig
u
re
13:
Pre
pa
r
a
ti
on and usa
ge
of
RP
S subs
tr
a
te
s as p
re
s
e
n
te
d
i
n
t
h
i
s
w
o
rk
.
A
c
on
ve
n
ti
ona
l
S
i
(
111
)
w
af
er
(a)
i
s p
re
pa
re
d b
y
mas
k
lit
ho
gr
aph
y
and an a
rr
a
y
of ho
le
s
i
s app
lie
d (b).
By
ann
e
a
li
n
g,
a
mono
cry
s
t
a
lli
n
e
l
a
yer
fo
r
ms (
c
)
t
o m
i
n
i
m
i
z
e
su
r
fa
ce
e
n
ergy
and po
re
s
c
oa
le
s
ce
(d) fo
r
t
h
e
sam
e
re
ason.
E
ve
n
t
ua
lly
a f
ree
s
t
and
i
n
g
fo
il
of
1
µ
m
i
s fo
r
m
e
d (
e
) Th
e
fo
il
i
s
t
h
e
n us
e
d as subs
tr
a
te
i
n
t
h
e
SSL
P
E
p
r
o
ce
ss
,
w
h
ile
s
till
a
tt
a
c
h
e
d
t
o
t
h
e
pa
re
n
t
w
af
er
(f) and
c
u
t
of (
g
).
A
f
ter
s
e
pa
r
a
ti
on
,
t
h
e
fo
il
c
an b
e
us
e
d
fo
r
so
l
a
r
cell
p
re
pa
r
a
ti
on and
t
h
e
subs
tr
a
te
w
af
er
i
s po
li
sh
e
d
t
o b
e
re
us
e
d f o
r
RP
S p
re
pa
r
a
ti
on (h).
46
- 117 -
9. List of figures
F
ig
u
re
14
D
i
ff
ere
n
t
i
ma
gi
n
g
a
rti
fa
ct
s
i
n SEM as
t
h
ey
o
cc
u
rre
d
i
n
t
h
i
s
re
s
e
a
rc
h. a)
l
as
er
spo
t
s on a S
i
(
111
)
su
r
fa
ce
. Th
e
i
n
ci
d
e
n
ce
an
gle
of
t
h
e
b
e
am
le
ads
t
o da
rk
-
t
o-
lig
h
t
gr
ad
ie
n
t
of
t
h
e
un
tre
a
te
d samp
le,
and
an
i
n
ver
s
e
on
e
i
n
t
h
e
l
as
er
tre
a
te
d a
re
as. b) (
100
)-o
rie
n
te
d S
i
cry
s
t
a
llite
. Th
e
e
d
ge
s off
er
mo
re
poss
i
b
ilitie
s fo
r
t
h
e
electr
ons
t
o
e
s
c
ap
e
t
h
e
bu
lk,
ma
ki
n
g
t
h
e
e
d
ge
s app
e
a
r
b
rig
h
ter
.
c
) p
l
a
te
-
like
and
s
t
o
c
has
tic
mo
r
pho
l
o
gy
. Th
e
(
111
) su
r
fa
ce
s (
t
op
rig
h
t
) app
e
a
r
mu
c
h da
rker
t
han su
r
fa
ce
s o
rie
n
t
a
te
d
d
i
ff
ere
n
tly
. d) s
ee
d
l
a
yer
af
ter
m
elt
-ba
ck
wit
h
e
xpos
e
d subs
tr
a
te
gl
ass. S
i
n
ce
gl
ass
i
s no
t
c
ondu
cti
n
g
electr
ons a
w
a
y
f
r
om
t
h
eir
i
n
ci
d
e
n
ce
po
i
n
t
it
b
ec
om
e
s
c
ha
rge
d. Fu
rt
h
er
i
n
c
om
i
n
g
electr
ons a
re
re
f
lecte
d
,
le
ad
i
n
g
t
o b
rig
h
t
a
rti
fa
ct
s
wit
h an op
tic
a
l
e
ff
ect
app
e
a
ri
n
g
like
lig
h
t
re
f
lecti
on on b
e
nd
su
r
fa
ce
s. ................................................................................................................................................
50
F
ig
u
re
15
(a) numb
er
of
vi
s
i
b
le
cry
s
t
a
llite
s
wit
h
i
n a squa
re
of
25
µ
m af
ter
ox
i
d
e
gr
o
wt
h. Subs
tr
a
te
s
w
h
ere
HF-
etc
h
e
d and pass
iv
a
te
d.
A
f
terw
a
r
ds
,
ti
m
e
fo
r
ox
i
d
e
gr
o
wt
h
w
as
give
n.
P
a
rticle
c
oun
ti
n
g
w
as
don
e
manua
lly
a
cc
o
r
d
i
n
g
t
o
[84]
–
[86]
. (b)
gree
n ba
r
s
re
p
re
s
e
n
t
fu
lly
c
oun
t
ab
le
pa
rticle
s
,
re
d ba
r
s
re
p
re
s
e
n
t
e
x
cl
ud
e
d on
e
s
,
and pu
r
p
le
ba
r
s
re
p
re
s
e
n
t
cry
s
t
a
llite
s
w
h
ic
h
were
i
d
e
n
ti
f
ie
d as pa
r
as
itic
pa
rticle
s
wit
h a s
i
z
e
und
er
n
e
a
t
h
5
µ
m. L
i
n
e
s
i
n (a)
re
p
re
s
e
n
t
t
h
e
i
n
cl
us
i
on (
gree
n) and
e
x
cl
us
i
on (
re
d)
li
n
e
s.
I
nd
ic
a
ti
on and
c
oun
ti
n
g
w
as p
er
fo
r
m
e
d manua
lly
o
ver
a
5
x
5
gri
d and
t
h
e
r
an
ge
s of da
t
a found
a
re
sho
w
n
i
n b). a) sho
w
s a d
et
a
il
of F
ig
u
re
25c,
gr
o
wt
h af
ter
15
m
i
n add
iti
ona
l
ox
i
d
e
gr
o
wt
h .........
51
F
ig
u
re
16:
S
et
up of
t
h
e
i
n s
it
u TEM
e
xp
eri
m
e
n
t
d
e
p
icte
d as a s
ketc
h (a) and
wit
h
lig
h
t
m
icr
os
c
op
y
(b).
A
f
ter
ann
e
a
li
n
g,
t
h
e
fo
r
m
er
a-S
i
i
s fu
lly
cry
s
t
a
lli
z
e
d as
vi
s
i
b
le
i
n b
rig
h
t
f
iel
d TEM (
c
). .........................
53
F
ig
u
re
17:
R
aman sp
ectr
a of amo
r
phous and
cry
s
t
a
lli
z
e
d s
ilic
on. Th
e
da
rker
li
n
e
s sho
w
t
h
e
sp
ectr
a af
ter
d
well
ti
m
e
s of
8
m
i
n
,
t
h
e
lig
h
ter
li
n
e
s of
5
s and
t
h
e
lig
h
te
s
t
af
ter
jus
t
1
s. Fo
r
d
well
ti
m
e
s of
1
s
,
b
l
u
e
and
re
d a
re
no
t
d
i
s
ti
n
g
u
i
shab
le
.............................................................................................................
58
F
ig
u
re
18:
Ex
e
mp
l
a
ry
r
a
w
R
aman da
t
a (d
well
ti
m
e
1
s) and
t
h
e
G
auss
i
an f
it
us
e
d fo
r
mapp
i
n
g
. ........
58
F
ig
u
re
19:
D
i
ff
ere
n
t
app
r
oa
c
h
e
s fo
r
p
l
o
tti
n
g
of
t
h
e
G
auss
i
an f
it
s.
J
us
t
p
l
o
tti
n
g
t
h
e
p
e
a
k
h
eig
h
t
o
vere
mphas
i
z
e
s s
tr
on
g
s
ig
na
l
s and
t
hus z-pos
iti
on of
t
h
e
cry
s
t
a
l
s (a).
U
s
i
n
g
t
h
e
wi
d
t
h of
t
h
e
f
it
re
p
re
s
e
n
te
d b
y
t
h
e
s
t
anda
r
d d
evi
a
ti
on
σ
reve
a
l
s mo
re
cry
s
t
a
lli
n
e
a
re
as
,
bu
t
no suff
icie
n
t
c
on
tr
as
t
b
etwee
n
t
h
e
m
i
s a
c
h
ieve
d (b).
Be
s
t
mo
r
pho
l
o
gy
c
on
tr
as
t
i
s
give
n b
y
a quo
tie
n
t
of
t
h
e
tw
o
,
c
omb
i
n
i
n
g
ge
n
er
a
l
cry
s
t
a
l
l
o
c
a
ti
on
wit
h
t
h
eir
z-pos
iti
on........................................................................................
60
F
ig
u
re
20
p
r
o
ce
ss
i
n
g
of p
r
o
t
o
ty
p
e
so
l
a
r
cell
s. On
t
o a s
ilic
on subs
tr
a
te
(a) SSL
P
E s
ilic
on
i
s
gr
o
w
n (b). Th
e
e
m
itter
(
c
)
i
s app
lie
d b
y
CVD and
t
h
e
TCO
c
on
t
a
ct
(d) b
y
spu
tteri
n
g
.
A
spa
cer
fo
il
(
e
) p
reve
n
t
s TCO and
e
m
itter
f
r
om
etc
h
i
n
g
(f). M
et
a
l
c
on
t
a
ct
s a
re
app
lie
d b
y
P
VD (
g
) and su
r
p
l
us m
et
a
l
i
s
re
mo
ve
d
t
o
get
h
er
wit
h
t
h
e
po
ly
m
er
spa
cer
(h)..................................................................................................................
62
- 118 -
9. List of figures
F
ig
u
re
21:
TEM b
rig
h
t
f
iel
d
i
ma
ge
s of
t
h
e
cry
s
t
a
lli
za
ti
on du
ri
n
g
t
h
e
i
n s
it
u
e
xp
eri
m
e
n
t
.
Alre
ad
y
cry
s
t
a
lli
z
e
d s
ee
d
i
s
c
o
l
o
re
d
re
d fo
r
vi
s
i
b
ility
. C
ry
s
t
a
lli
za
ti
on
i
n
iti
a
li
z
e
s a
t
t
h
e
i
n
ter
fa
ce
ti
n-s
ilic
on af
ter
an
i
n
c
uba
ti
on
ti
m
e
of
5
m
i
n.......................................................................................................................
64
F
ig
u
re
22:
SEM
i
ma
ge
s of d
i
ff
ere
n
t
c
ha
r
a
cteri
s
tic
mo
r
pho
l
o
gie
s
w
h
ic
h d
evel
op du
ri
n
g
SSL
P
E
gr
o
wt
h of
S
i
. Th
e
s
t
o
c
has
tic
o
rie
n
t
a
ti
on
i
s mos
t
of
te
n obs
erve
d (a).
A
dd
iti
ona
lly
a p
l
a
te
-
like
mo
r
pho
l
o
gy
c
an b
e
obs
erve
d (b)
,
p
r
odu
ci
n
g
l
a
rger
cry
s
t
a
llite
s
w
h
ic
h fo
r
m
r
ound spo
t
s
wit
h sha
re
d
l
a
ter
a
l
o
rie
n
t
a
ti
on. Th
e
l
a
tter
c
an b
e
obs
erve
d b
e
s
t
du
ri
n
g
b
i
moda
l
gr
o
wt
h (
c
)
,
w
h
ere
t
h
e
tw
o majo
r
mo
r
pho
l
o
gie
s
c
o
e
x
i
s
t
(
c
o
l
o
ri
z
e
d
,
ma
ge
n
t
a
:
p
l
a
te
-
like,
te
a
l:
s
t
o
c
has
tic
). .................................................................................
67
F
ig
u
re
23:
Co
l
o
ri
z
e
d SEM
i
ma
ge
s of
t
h
e
gr
o
wt
h
l
a
yer
du
ri
n
g
SSL
P
E. Th
e
f
ir
s
t
r
o
w
(a) sho
w
s an o
verview
and
t
h
e
s
ec
ond
r
o
w
(b) sho
w
s
c
o
rre
spond
i
n
g
d
et
a
il
s. Th
e
c
o
l
umns (
1
t
o
6
)
re
p
re
s
e
n
t
d
i
ff
ere
n
t
ti
m
e
s
t
amps f
r
om
1
s
t
o
1
h as
i
nd
ic
a
te
d. Th
ree
phas
e
s
c
an b
e
obs
erve
d
:
su
r
fa
ce
nu
cle
a
ti
on and pa
rti
a
l
d
i
sso
lvi
n
g
(
1
and
2
)
,
fo
r
ma
ti
on of p
l
a
te
s and
re
f
i
n
i
n
g
of s
t
o
c
has
tic
a
re
as (
2
and
3
)
,
gr
o
wt
h and
s
t
ab
ili
za
ti
on of
t
h
e
cry
s
t
a
l
s fo
r
m
e
d b
e
fo
re
. ..........................................................................................
68
F
ig
u
re
24
tw
o
t
h
e
o
rie
s of
t
h
e
o
rigi
n of p
l
a
te
-
like
cry
s
t
a
llite
pa
tc
h
e
s.
I
n
eit
h
er
c
as
e, t
h
e
pa
tc
h mus
t
sha
re
a un
iti
n
g
r
oo
t
.
It
on
ly
i
s of qu
e
s
ti
on a
cry
s
t
a
l
nu
cle
a
te
s ou
t
of
t
h
e
s
ee
d f
ir
s
t
(a-
1
) o
r
i
f a
r
oo
t
d
evel
ops
l
a
ter
a
lly
i
n
t
h
e
s
ee
d f
ir
s
t
(b-
1
) .
I
n
t
h
e
f
ir
s
t
c
as
e,
t
h
e
cry
s
t
a
llite
on
t
h
e
su
r
fa
ce
w
ou
l
d o
rigi
na
te
t
h
e
pa
tc
h
(a-
2
) and on
ly
af
terw
a
r
ds
cry
s
t
a
lli
z
e
t
h
e
s
ee
d und
er
n
e
a
t
h.
I
n
t
h
e
s
ec
onda
ry
c
as
e,
a
l
a
rger
cry
s
t
a
l
i
n
t
h
e
s
ee
d (b-
2
)
w
ou
l
d
gr
o
w
ou
t
and b
ec
om
e
p
r
onoun
ce
d as a su
r
fa
ce
pa
tc
h
l
a
ter
.
B
o
t
h
c
as
e
s
re
su
lt
i
n s
i
m
il
a
r
mo
r
pho
l
o
gie
s (a/b-
3
) ............................................................................................................................
71
F
ig
u
re
25:
SEM
i
ma
ge
s of HF-
tre
a
te
d samp
le
s. Subs
tr
a
te
s
w
h
ere
HF-
etc
h
e
d and pass
iv
a
te
d.
A
f
terw
a
r
ds
,
ti
m
e
fo
r
ox
i
d
e
gr
o
wt
h
w
as
give
n.......................................................................................
72
F
ig
u
re
26:
R
aman sp
ectr
a
relev
an
t
t
o
t
h
i
s
w
o
rk
(d
well
ti
m
e
8
m
i
n).
N
o
t
ab
ly,
som
e
sp
ectr
a a
re
no
t
d
i
s
ti
n
g
u
i
shab
le
f
r
om ano
t
h
er
. Th
i
s
c
on
cer
ns SSL
P
E S
i
and a
re
f
ere
n
ce
w
af
er
and a-S
i
s
ee
ds
w
h
ic
h
i
s
p
r
om
i
s
i
n
g
fo
r
t
h
eir
cry
s
t
a
lli
n
e
p
r
op
ertie
s.
U
np
r
a
ctic
a
lly
fo
r
t
h
e
c
ha
r
a
cteri
za
ti
on
i
s
t
h
e
i
d
e
n
tic
a
l
app
e
a
r
an
ce
of
gl
ass and
t
h
e
amo
r
phous s
ee
d obs
erve
d
t
h
r
ou
g
h
gl
ass. ..............................................
77
F
ig
u
re
27:
Co
rrel
a
ti
on of s
t
anda
r
d d
evi
a
ti
on and
ce
n
tr
a
l
w
a
ve
numb
er
d
irectly
a
t
t
h
e
b
egi
nn
i
n
g
of
cry
s
t
a
lli
za
ti
on and af
ter
1
h.
N
o
t
ab
ly
s
t
anda
r
d d
evi
a
ti
on d
ecre
as
e
s and
ce
n
tr
a
l
w
a
ve
numb
er
i
n
cre
as
e
s
on a
ver
a
ge
.
B
o
t
h
v
a
l
u
e
s
get
mo
re
homo
ge
nous
wit
h
cry
s
t
a
lli
za
ti
on.
U
nf
it
da
t
a po
i
n
t
s a
re
d
i
s
c
a
r
d
e
d
reg
a
r
d
i
n
g
t
h
e
r
u
le
s us
e
d fo
r
t
h
e
mapp
i
n
g
d
e
s
cri
b
e
d
i
n
t
h
e
te
x
t
..........................................................
77
F
ig
u
re
28:
D
evel
opm
e
n
t
of
cry
s
t
a
llite
s du
ri
n
g
gr
o
wt
h.
A
cry
s
t
a
lli
n
e
n
etw
o
rk
d
evel
ops du
ri
n
g
1
h of
gr
o
wt
h and
eve
n
t
ua
lly
le
ads
t
o ma
tri
x
i
n
ver
s
i
on b
etwee
n amo
r
phous and
cry
s
t
a
lli
n
e
ma
teri
a
l
. .......
78
- 119 -
9. List of figures
F
ig
u
re
29
c
on
tig
uous and homo
ge
nous
gr
o
wt
h a
c
h
ieve
d b
y
l
as
er
tre
a
t
m
e
n
t
(bo
tt
om ha
l
f).
I
n
t
h
e
un
tre
a
te
d a
re
a (upp
er
ha
l
f) on
ly
s
c
a
rce
and
i
nhomo
ge
nous
cry
s
t
a
llite
s a
re
found.............................
83
F
ig
u
re
30
i
n
te
ns
ity
map of
t
h
e
unadjus
te
d b
e
am (a) and
t
h
e
adjus
te
d b
e
am (b).
A
G
auss
i
an
app
r
ox
i
ma
ti
on of bo
t
h (
gree
n
c
u
rve
i
n
t
h
e
p
r
oj
ecti
ons) sho
w
s no
t
ab
le
d
evi
a
ti
ons – m
e
asu
re
d
v
a
l
u
e
s
i
n
t
h
e
ce
n
ter
a
re
h
ig
h
er
and bo
t
h spo
t
s ha
ve
p
r
onoun
ce
d shou
l
d
er
s un
like
t
h
e
G
auss
i
ans us
e
d.
Nevert
h
ele
ss
G
auss
i
ans
will
b
e
us
e
d fo
r
a
ge
n
er
a
li
z
e
d d
e
s
cri
p
ti
on of
t
h
e
b
e
ams. (a)
Aver
a
ge
s
t
anda
r
d
d
evi
a
ti
on
σ
=
96
.
0
µ
m
,
elli
p
ticity
(
σ
x
-
σ
y)/
σ
x
=
0
.
72
(b)
Aver
a
ge
s
t
anda
r
d d
evi
a
ti
on
σ
=
46
.
4
µ
m
,
elli
p
ticity
(
σ
x
-
σ
y)/
σ
x
=
0
.
91
.....................................................................................................................
84
F
ig
u
re
31
a-S
i
su
r
fa
ce
tre
a
te
d
wit
h an adjus
te
d
,
h
ig
h
e
n
ergy
l
as
er
pu
l
s
e
. a)
t
h
e
SEM
i
ma
ge
sho
w
s a
cr
a
ter wit
h a
w
a
ll
of a
cc
umu
l
a
te
d ma
teri
a
l
su
rr
ound
i
n
g
it
. Ou
t
s
i
d
e
of
t
h
e kr
a
ter
a
re
a
,
d
i
ff
ere
n
t
shad
i
n
g
s
a
re
vi
s
i
b
le
on
t
h
e
su
r
fa
ce
. b)
t
h
e
R
aman map
w
as p
r
o
ce
ss
e
d
wit
h
t
h
e
a
lg
o
rit
hm p
re
s
e
n
te
d b
e
fo
re
.
D
i
ff
ere
n
t
e
x
e
mp
l
a
ry
sp
ectr
a a
re
sho
w
n
i
n
c
).
t
h
e
Kr
a
ter
a
re
a
i
s
cry
s
t
a
lli
z
e
d
,
as
well
as
t
h
e
d
irect
su
rr
ound
i
n
g
s of
t
h
e
w
a
ll
a
re
a. C
r
a
ck
s
wit
h
i
n
t
h
e
s
ilic
on a
re
vi
s
i
b
le
t
h
r
ou
g
hou
t
t
h
e
cry
s
t
a
lli
z
e
d a
re
a.
c
)
sho
w
s
e
x
e
mp
l
a
ry
R
aman sp
ectr
a ob
t
a
i
n
e
d a
t
t
h
e
i
nd
ic
a
te
d
l
o
c
a
ti
ons
1
t
o
4
. .....................................
84
F
ig
u
re
32
te
mp
er
a
t
u
re
a
c
h
ieve
d
i
n an
i
nf
i
n
ite
s
i
ma
l
ele
m
e
n
t
d
e
p
e
nd
i
n
g
on
l
as
er
i
n
te
ns
ity
i
n
t
ha
t
s
eg
m
e
n
t
. Th
e
c
h
e
m
ic
a
l
c
omp
r
opo
rti
ona
ti
on
re
a
cti
on of s
ilic
on d
i
ox
i
d
e
t
a
ke
s p
l
a
ce
s s
t
a
rti
n
g
a
t
T =
750
°
C (
gree
n h
ig
h
lig
h
ti
n
g
)
,
and
t
h
e
m
elti
n
g
of a-S
i
a
t
T =
1147
°
C (
re
d h
ig
h
lig
h
ti
n
g
).
B
o
t
h
re
a
cti
ons
a
re
e
ndo
t
h
er
m
ic
and
re
qu
ire
c
ons
i
d
er
ab
ly
mo
re
e
n
ergy
t
han jus
t
h
e
a
ti
n
g
t
h
e
v
o
l
um
e
w
h
ic
h
re
su
lt
s
i
n
a s
l
o
wer
te
mp
er
a
t
u
re
i
n
cre
m
e
n
t
o
ver
t
h
e
e
n
ergy
r
an
ge
i
n
w
h
ic
h h
e
a
ti
n
g
and a
re
a
cti
on
t
a
ke
s p
l
a
ce
.
...............................................................................................................................................................
85
F
ig
u
re
33
s
i
mu
l
a
te
d
te
mp
er
a
t
u
re
s
re
su
lti
n
g
f
r
om
l
as
er
tre
a
t
m
e
n
t
wit
h an
i
d
e
a
lly
g
auss-shap
e
d
l
as
er
b
e
am. Th
e
c
h
e
m
ic
a
l
c
omp
r
opo
rti
ona
ti
on
re
a
cti
on s
t
a
rt
s a
t
T =
750
°
C (
gree
n h
ig
h
lig
h
ti
n
g
) and m
elti
n
g
a
t
T =
1147
°
C (
re
d h
ig
h
lig
h
ti
n
g
)
,
bo
t
h damp
e
n
i
n
g
h
e
a
ti
n
g
of
t
h
e
subs
tr
a
te
. a) D
i
ff
ere
n
t
t
o
t
a
l
spo
t
e
n
ergie
s
wit
h a
c
ons
t
an
t
b
e
am
wi
d
t
h (
σ
=
100
µ
m). b) D
i
ff
ere
n
t
b
e
am
wi
d
t
hs
σ
wit
h a
c
ons
t
an
t
t
o
t
a
l
pu
l
s
e
e
n
ergy
(E =
3
.
0
µJ
).
c
) and d) d
e
p
ict
t
h
e
e
n
ergy
d
i
s
tri
bu
ti
ons
le
ad
i
n
g
t
o
t
h
e
re
sp
ective
te
mp
er
a
t
u
re
p
r
of
ile
s
i
n a) and b)..........................................................................................................
86
F
ig
u
re
34
i
n
ter
a
cti
on of a sma
ll,
c
o
rrecte
d b
e
am
wit
h
t
h
e
s
ee
d su
r
fa
ce
and on
t
h
e
subs
e
qu
e
n
t
SSL
P
E
gr
o
wt
h.
I
n
ter
a
cti
on a
re
a d
e
p
icte
d as a) SEM
i
ma
ge
of
t
h
e
s
ee
d
,
b)
R
aman map of
t
h
e
s
ee
d
,
and
c
) SEM
i
ma
ge
of
t
h
e
re
su
lti
n
g
SSL
P
E mo
r
pho
l
o
gy
. Comp
le
m
e
n
t
a
ry
R
aman sp
ectr
ums d) and
l
a
ter
a
l
p
l
o
t
of
t
h
e
cry
s
t
a
lli
n
e
R
aman p
e
a
k
e
) a
re
p
r
o
vi
d
e
d. Th
e
mo
lte
n and
recry
s
t
a
lli
z
e
d
ce
n
ter
sho
w
a sa
ti
sfa
cti
on
i
n
te
ns
ity
of
t
h
e
cry
s
t
a
lli
n
e
R
aman p
e
a
k
and
le
ads
t
o p
reve
n
te
d
gr
o
wt
h du
ri
n
g
SSL
P
E.
P
a
rti
a
l
cry
s
t
a
lli
za
ti
on
wit
hou
t
m
elti
n
g
do
e
s no
t
h
i
nd
er
gr
o
wt
h and
t
h
e
l
as
i
n
g
suppo
rt
s SSL
P
E......................
88
- 120 -
9. List of figures
F
ig
u
re
35
te
mp
er
a
t
u
re
d
e
p
e
nd
e
n
t
b
1
c
o
e
ff
icie
n
t
s (a-
1
and b-
1
) and
t
h
eir
Fou
rier
tr
ansfo
r
ma
ti
on (a-
2
and b-
2
)
reve
a
li
n
g
e
n
ergy
and
e
ff
ective
cr
oss s
ecti
on of
t
h
e
d
e
f
ect
s ..................................................
91
F
ig
u
re
36
C
ell
bu
il
d-up and da
t
a of
cell
s bu
il
d f
r
om SSL
P
E-
e
p
il
a
yer
s
,
gr
o
w
n
eit
h
er
on S
i
(
111
) (a
,
b and
c
-
1
) o
r
on an
RP
S(
111
) (a
,
b and
c
-
2
). (a) pho
t
o
gr
aphs of
t
h
e
cell
a
rr
a
y
s
wit
h
t
h
e
quas
i
-ba
ck
s
i
d
e
c
on
t
a
ct
c
o
veri
n
g
a
ll
i
na
ctive
f
r
on
t
a
re
a. Th
e
a
ctive
a
re
a
i
s
c
o
vere
d
wit
h
I
TO and
i
s us
e
d as f
r
on
t
-s
i
d
e
c
on
t
a
ct
.
(b)
J
-V
c
u
rve
s of on
e
cell
of
t
h
e
re
sp
ective
w
af
er
and (
c
)
re
f
lectivity
and
e
x
ter
na
l
quan
t
um
e
ff
icie
n
cy
wit
h
t
h
e
re
su
lti
n
g
i
n
ter
na
l
quan
t
um
e
ff
icie
n
cy
.....................................................................................
93
F
ig
u
re
37
summa
ry
of
re
su
lt
s
c
on
cer
n
i
n
g
mo
r
pho
l
o
gy
d
evel
opm
e
n
t
du
ri
n
g
t
h
e
gr
o
wt
h of a
po
lycry
s
t
a
lli
n
e
SSL
P
E
l
a
yer
and d
y
nam
ic
s
ee
d
l
a
yer
cry
s
t
a
lli
za
ti
on. (a) b
e
fo
re
i
ns
erti
on
i
n
t
o
ti
n
,
t
h
e
s
ee
d
l
a
yer
i
s fu
lly
amo
r
phous and
l
o
c
a
te
d on a subs
tr
a
te
gl
ass (b)
i
n
t
h
e
mom
e
n
t
of
i
ns
erti
on
,
nu
cle
a
ti
on of
su
r
fa
ce
SSL
P
E
cry
s
t
a
llite
s
i
n h
ig
h d
e
ns
ity
t
a
ke
s p
l
a
ce
. Th
e
su
r
fa
ce
of
t
h
e
s
ee
d
l
a
yer
cry
s
t
a
lli
z
e
s
i
mm
e
d
i
a
tely
. Som
e
cry
s
t
a
lli
z
e
d d
e
nd
rite
s
re
a
c
h
t
h
e
i
n
ter
fa
ce
of
gl
ass and a-S
i
(
c
)
cry
s
t
a
lli
za
ti
on of
t
h
e
s
ee
d
c
omm
e
n
ce
s
,
wit
h d
e
nd
rite
s
gr
o
wi
n
g
l
on
ger
and b
r
oad
er
wit
h
i
n
t
h
e
a-S
i
.
I
n
t
h
e
s
e
a
re
as
,
s
t
o
c
has
tic
a
lly
o
rie
n
t
a
te
d
cry
s
t
a
llite
s p
er
s
i
s
t
. Th
eir
numb
er
i
s
re
du
ce
d b
y
Os
w
a
l
d
ri
p
e
n
i
n
g,
w
h
ile
t
h
e
re
ma
i
n
i
n
g
cry
s
t
a
llite
s
e
n
l
a
rge
. Som
e
v
o
l
um
e
of
t
h
e
s
ee
d
i
n su
r
fa
ce
a
re
a do
e
s
cry
s
t
a
lli
z
e
l
a
ter
a
lly,
and
t
ha
t
l
a
ter
a
l
cry
s
t
a
lli
za
ti
on
i
s
re
p
re
s
e
n
te
d
i
n
l
a
rger,
f
l
a
t
SSL
P
E
cry
s
t
a
llite
s on
t
h
e
su
r
fa
ce
. (d) Os
w
a
l
d
Ri
p
e
n
i
n
g
and
gr
o
wt
h on
t
h
e
su
r
fa
ce
c
on
ti
nu
e
s
,
w
h
ile
cry
s
t
a
lli
za
ti
on s
t
a
te
of
t
h
e
s
ee
d s
t
ab
ili
z
e
s as s
ee
n
i
n (
e
).....................................................................................................................................................
102
- 121 -
10. List of tables
10. L
ist
of
t
ab
l
e
s
Tab
le
1:
D
e
ns
ity
of S
i
and S
i
O2
w
h
ic
h o
rigi
na
te
s
el
f-pass
iv
a
ti
n
g
b
e
ha
vi
o
r
of S
i
....................................
14
Tab
le
2
Examp
le
s f
r
om
liter
a
t
u
re
and
t
h
i
s
w
o
rk
sho
wi
n
g
t
h
e
h
ig
h
r
an
ge
of
re
su
lt
s
i
n
t
h
e
cry
s
t
a
lli
za
ti
on
of a-S
i
on
gl
ass d
e
p
e
nd
i
n
g
on
v
a
ri
ous p
r
o
ce
ss pa
r
am
eter
s
wit
h d
:
l
a
yer
t
h
ick
n
e
ss
;
T
:
op
er
a
ti
on
te
mp
er
a
t
u
re;
t
i
n
c
i
n
c
uba
ti
on
ti
m
e;
and
t
cry
s
t
cry
s
t
a
lli
za
ti
on
ti
m
e
...........................................................
26
Tab
le
3
ma
teri
a
l
pa
r
am
eter
s us
e
d
i
n
t
h
e
s
i
mu
l
a
ti
on ............................................................................
42
Tab
le
4:
Pr
op
ertie
s of Co
r
n
i
n
g
Ea
gle
X
G
gl
ass
®
as us
e
d
i
n fo
r
t
h
i
s
w
o
rk
. Th
e
h
ig
h-p
er
fo
r
man
ce
f
l
oa
t
ma
teri
a
l
i
nh
i
b
it
s
gre
a
t
c
h
e
m
ic
a
l
and
t
h
er
ma
l
re
s
i
s
t
an
ce
.
A
dd
iti
ona
lly,
it
s
t
h
er
ma
l
e
xpans
i
on
c
om
e
s
cl
os
e
t
o
t
ha
t
of S
i
and
t
hus m
i
n
i
m
i
z
e
s
t
h
er
ma
l
s
tre
ss
ge
n
er
a
ti
on b
etwee
n
l
a
yer
s ..............................
42
Tab
le
5:
Pr
o
ce
ss
i
n
g
pa
r
am
eter
s du
ri
n
g
P
VD
gr
o
wt
h of
t
h
e
s
ee
d
l
a
yer
.
P
a
r
am
eter
s ma
y
v
a
ry
d
e
p
e
nd
e
n
t
on
t
h
e
fun
cti
ona
lity
of
t
h
e
c
hamb
er
and
l
ab
e
n
vir
onm
e
n
t
(p
re
ssu
re,
h
e
a
ter
l
amps
,
ev
apo
r
a
ti
on
r
a
te
)
w
h
ile
o
t
h
er
s a
re
s
et
f
r
om
t
h
e
op
er
a
t
o
r
t
o adjus
t
o
r
t
o
re
a
c
h
cert
a
i
n
e
xp
eri
m
e
n
t
a
l
g
oa
l
s (
electr
on b
e
am
c
u
rre
n
t
and
v
o
lt
a
ge,
ev
apo
r
a
ti
on
r
a
te,
f
i
na
l
t
h
ick
n
e
ss of
t
h
e
l
a
yer
).....................................................
44
Tab
le
6:
h
e
a
t
c
ondu
ctivity
c
o
e
ff
icie
n
t
s
i
nf
l
u
e
n
ci
n
g
t
h
e
cry
s
t
a
lli
za
ti
on sp
ee
d
i
n SSL
P
E ........................
67
Tab
le
7
Su
r
fa
ce
e
n
ergie
s of
t
h
e
t
h
ree
l
o
we
s
t
M
iller
i
nd
e
x
e
d fa
cet
s of S
i
[112]
...................................
71
Tab
le
8
S
t
a
ge
s of mo
r
pho
l
o
gy
d
evel
opm
e
n
t
of
t
h
e
s
ee
d .....................................................................
80
Tab
le
9
electric
a
l
da
t
a as a
lre
ad
y
pub
li
sh
e
d
i
n
[30]
. Th
e
h
ig
h m
e
asu
re
d
li
f
eti
m
e
s ma
ke
t
h
e
ma
teri
a
l
a
c
and
i
da
te
fo
r
so
l
a
r
cell
bu
il
d-up............................................................................................................
90
Tab
le
10
c
ha
r
a
cteri
s
tic
s of
t
h
e
b
e
s
t
p
r
o
t
o
ty
p
e
so
l
a
r
cell
s fo
r
e
a
c
h subs
tr
a
te
ty
p
e
..............................
92
Tab
le
11:
A
ssum
e
d o
rigi
ns of
t
h
e
d
e
f
ect
s found
i
n mono
cry
s
t
a
lli
n
e
SSL
P
E s
ilic
on.
All
rec
o
r
d
e
d d
e
f
ect
s
c
an b
e
e
xp
l
a
i
n
e
d b
y
d
i
ff
ere
n
tly
p
r
onoun
ce
d
v
a
c
an
cy
c
omp
le
x
e
s
,
re
sp
ectively
d
i
s
l
o
c
a
ti
ons.
I
f
t
h
e gr
o
wt
h
m
et
hod o
rigi
na
te
s
t
h
e
d
e
f
ect,
s
i
m
il
a
ritie
s a
re
likely
. E
3
and
P
H
R
-V
3
w
ou
l
d
re
p
re
s
e
n
t
ann
e
a
le
d s
t
a
te
s
of o
t
h
er
d
e
f
ect
s
,
and
t
h
e
i
n
cre
as
e
d
c
ap
t
u
re
cr
oss s
ecti
on of
t
h
e
d
i
s
l
o
c
a
ti
ons
c
ou
l
d s
te
m f
r
om ann
e
a
li
n
g
as
well
. V
2
i
s
k
no
w
n
t
o app
e
a
r
i
n
c
o
e
x
i
s
te
n
ce
wit
h E
3
.........................................................................
95
- 122 -
