CIGSe Superstrate Solar Cells:
Growth and Characterization
of CIGSe Thin Films
on Transparent Conductive Oxides
vorgelegt von
Dipl.-Phys.
Marc Daniel Heinemann
geb. in Oldenburg
von der Fakult¨at IV - Elektrotechnik und Informatik
der Technischen Universit¨at Berlin
zur Erlangung des akademischen Grades
Doktor der Naturwissenschaften
-Dr.-rer.-nat-
genehmigte Dissertation
Promotionsausschuss:
Vorsitzender: Prof. Dr. K. Petermann
Gutachter: Prof. Dr. B. Rech
Gutachter: Prof. Dr. S. Schorr
Gutachter: Prof. Dr. M. Powalla
Tag der wissenschaftlichen Aussprache 13. Oktober 2015
Berlin 2016
ii
Abstract
The prospects and the limitations of Cu(In,Ga)Se2(CIGSe) solar cells in superstrate configura-
tion were studied in this work. The, compared to the standard substrate configuration, inverted
device structure, sets new requirements for the materials used in the device. CdS cannot be
used as the buffer layer between the CIGSe absorber and the ZnO window layer due to its low
thermal stability. The direct deposition of CIGSe onto ZnO is known to induce the formation
of a GaOxlayer at the CIGSe/ZnO interface, but its influence on the device was yet unclear.
The correlation between the interface and device properties of CIGSe/ZnO devices is the core
part of this work.
Interface analysis show that the GaOxlayer exhibits large impurities of Cu, which are known
to induce acceptor states in oxides. Also In was found as an impurity and is shown to reduce
the interface band gap, increasing the interface recombination. The amount of Cu and In within
the GaOxlayer depend critically on the CIGSe deposition process and best efficiencies were
achieved for a process type leading to the lowest Cu and In concentrations. A device model,
based on numerical methods, was set up and can explain several aspects of the superstrate
device’s behaviour. The performance limiting effect was found to be indeed deep acceptor states
within the GaOxand not a conduction band spike as was speculated earlier. Analysis of the
amorphous GaOxshow that the electron affinity is similar to CIGSe due to oxygen deficiency.
This understanding leads to new concepts to overcome the efficiency limitations. In this
work the use of doped amorphous oxide diffusion barriers is tested. Ga2O3deposited at low
temperatures and without intentional doping is shown to perform best. Nevertheless the devices
still suffer from acceptor states, in this case at the CIGSe/GaOxinterface.
The efficiency of the CIGSe/ZnO devices could be substantially increased to above 10 % by
doping the CIGSe layer with Na. This was shown to be a sensitive process step, as Na, tends to
accumulate within the GaOxlayer, and, similar to Cu, induces acceptor states within it. The
device degradation and the previously reported effect of forward-biasing is assumed to originate
from electro-migration of Na within the p/n-junction. A low-rate post-deposition of NaF could
reduce the Na concentration at the interface and leads to a stable device efficiency of up to 11 %.
Zn diffusion from the ZnO into the CIGSe absorber is shown to lower the p-type doping
and the electron lifetime within the CIGSe. Device simulations however indicate, that this does
not limit the device efficiency substantially and that Na reduces the negative effect of the Zn
contamination.
Further it is shown, that the Au back contact can be substituted by MoOx/Ag without
sacrificing the device efficiency. Simulations suggest that the high reflectivity and high scattering
of this type of back contact leads to a possible reduction of the CIGSe layer thickness down to
600 nm, while maintaining the same efficiency as a substrate device with a 1000 nm thick CIGSe
layer. Another advantage of the superstrate configuration was shown to be the defect annealing
within the ZnO during the CIGSe deposition. This increases the electron mobility and the
overall sub-band gap transparency.
Band gap engineering by Ga and S is shown to be well implementable, making the superstrate
configuration very attractive provided that a Cu tolerant buffer layer can be found in the future.
iii
Zusammenfassung
Die vorliegende Arbeit untersucht die Chancen und die Limitierungen von Cu(In,Ga)Se2(CIGSe)
Solarzellen in Superstratkonfiguration. Diese im Vergleich zur gew¨ohnlichen Substratkonfigura-
tion invertierte Schichtstruktur, stellt neue Anforderungen an die Materialieneigenschaften ihrer
Schichtkomponenten. Dies f¨uhrt unter anderem zur Untauglichkeit der generell verwendeten
CdS-Pufferschicht, aufgrund deren geringen thermischen Stabilit¨at. Bei der direkten Abschei-
dung von CIGSe auf ZnO ist bekannt, dass sich eine d¨unne GaOx-Schicht an der Grenzfl¨ache
bildet. Der Einfluss dieser Schicht war bisher ungekl¨art, weshalb die Grenzfl¨achenbildung und
deren Einfluss auf die Solarzelleneigenschaften hier eingehend untersucht wurde.
Ein numerisches Bauteilmodell wurde aufgestellt, welches die Solarzelleneigenschaften bei
unterschiedlichen Grenzfl¨acheneigenschaften konsistent erkl¨aren kann. Mit der Hilfe dieses Mod-
ells wird gezeigt, dass der typischerweise niedrige Wirkungsgrad durch tiefe Akzeptorzust¨ande
im GaOxverursacht wird und nicht, wie bislang angenommen, durch eine zu niedrige Elektrone-
naffinit¨at des GaOxs. Es konnte best¨atigt werden, dass amorphes Ga2O3eine Elektronenaffinit¨at
¨ahnlich zu der von CIGSe besitzt. Direkt an der Grenzfl¨ache durchgef¨uhrte Photoelektronen-
spektroskopiemessungen zeigen, dass die GaOx-Schicht sowohl mit Cu als auch mit In verun-
reinigt ist. Die Cu-Verunreinigung wird als Ursache f¨ur die Akzeptorzust¨ande vermutet, w¨ahrend
die In-Verunreinigung die Aktivierungsenergie der Rekombination an der CIGSe/GaOxGren-
zfl¨ache vermindert. Aus Tiefenprofilen der Elemente geht hervor, dass die Konzentration von
In und Cu in der GaOx-Schicht stark von dem verwendeten CIGSe-Abscheideprozess abh¨angen.
Die Schicht mit der geringsten In- und Cu-Verunreinigung f¨uhrt zu dem h¨ochsten Wirkungsgrad,
welcher jedoch weiterhin durch die Cu induzierten Akzeptorzust¨ande im GaOxlimitiert ist. Die
Zn-Diffusion aus dem ZnO in den CIGSe-Absorber hat sich als sch¨adlich aber nicht limitierend
herausgestellt.
Diese Erkenntnisse f¨uhrten zu neuen Ans¨atzen zur Effizienssteigerung. In dieser Arbeit
wurden dotierte amorphe Oxidschichten gepr¨uft, um die Cu-Diffusion zu reduzieren und die
Akzeptorzust¨ande zu kompensieren. Amorphes Ga2O3hat zu den besten Ergebnissen gef¨uhrt.
Die Limitierung durch Akzeptorzust¨ande, in diesem Fall an der CIGSe/Ga2O3Grenzfl¨ache,
bleibt jedoch bestehen.
Der Wirkungsgrad in den CIGSe/ZnO-Solarzellen konnte durch eine kontrollierte Na-Dotierung
wesentlich erh¨oht werden. Dabei hat sich herausgestellt, dass Na, ¨ahnlich wie auch Cu, Akzep-
torzust¨ande an der Grenzfl¨ache verursacht. Diese f¨uhren dar¨uber hinaus zu einer Degradation
der Solarzelle, angetrieben durch die Migration von Na im elektrischen Feld des p/n-¨
Ubergangs.
Eine NaF-Nachbehandlung mit geringer Diffusionsrate erm¨oglichte eine Dotierung der CIGSe-
Schicht ohne eine wesentliche Erh¨ohung der Na Konzentration an der Grenzfl¨ache zu verur-
sachen. Auf diese Weise konnte ein zeitlich stabiler Wirkungsgrad von 11.0 % erreicht werden.
Des Weiteren wird gezeigt, dass es m¨oglich ist, MoOx/Ag als R¨uckkontaktmaterialien zu
verwenden. Simulationen deuten darauf hin, dass die hohe Reflektivit¨at des R¨uckkontaktes eine
Reduzierung der CIGSe Schichtdicke von 1000 nm auf 600 nm erlaubt. Die CIGSe-Abscheidung
f¨uhrt zudem zu einem Ausheizen von Defekten im ZnO. Dies erh¨oht die Elektronenbeweglichkeit
im ZnO und dessen Transparenz unterhalb der Bandl¨ucke. Eine Aufweitung der Bandl¨ucke an
den Grenzfl¨achen durch Ga und S Gradienten wurde als praktisch anwendbar aufgezeigt, was
die Anwendung der Superstratkonfiguration als Ganzes attraktiv macht, vorausgesetzt eine Cu
tolerante Pufferschicht wird verf¨ugbar.
CONTENTS CONTENTS
Contents
Acronyms, Symbols and Notations v
Introduction v
1 CIGSe solar cell basics 7
1.1 Cu(In,Ga)Se2SolarCells............................ 7
1.1.1 Physics of a Solar Cell . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.1.2 The superstrate configuration . . . . . . . . . . . . . . . . . . . . . 9
1.1.3 Literaturereview ............................ 11
1.2 Thermodynamics of the interface reaction . . . . . . . . . . . . . . . . . . 12
1.2.1 Gibbsfreeenergy............................ 13
1.2.2 Diffusion................................. 15
1.3 Material Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.3.1 Cu(In,Ga)Se2(CIGSe) ......................... 16
1.3.2 ZnO ................................... 19
2 Experimental: Deposition and Characterisation 23
2.1 CIGSegrowth.................................. 23
2.2 TCOgrowth................................... 28
2.3 Metallization and device layout . . . . . . . . . . . . . . . . . . . . . . . . 30
2.4 Device Characterization . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.4.1 J−Vmeasurements.......................... 31
2.4.2 C−Vmeasurements.......................... 33
2.4.3 External Quantum Efficiency (EQE) . . . . . . . . . . . . . . . . . 36
2.4.4 Electron-Beam-Induced Current (EBIC) . . . . . . . . . . . . . . . 37
2.5 Material Characterization . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.5.1 X-ray Photoelectron Spectroscopy (XPS) . . . . . . . . . . . . . . . 38
2.5.2 X-ray Diffraction (XRD) . . . . . . . . . . . . . . . . . . . . . . . . 41
2.5.3 Glow Discharge-Optical Emission Spectroscopy (GDOES) . . . . . 42
3 Numerical Simulation 45
3.1 Controlling interface recombination in CIGSe devices . . . . . . . . . . . . 48
3.2 Acceptor states at the hetero-interface . . . . . . . . . . . . . . . . . . . . 51
v
vi CONTENTS
4 TCO evaluation 57
4.1 FTO,ITOandZnO .............................. 57
4.2 Zn(O,S) ..................................... 59
4.3 ZnOannealing.................................. 60
4.4 Summary: TCO evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
5 ZnO/CIGSe device and interface analysis 63
5.1 Interfaceformation ............................... 63
5.1.1 CGSe/ZnO interface formation . . . . . . . . . . . . . . . . . . . . 63
5.1.2 CISe/ZnO interface formation . . . . . . . . . . . . . . . . . . . . . 65
5.1.3 CIGSe/ZnO interface formation . . . . . . . . . . . . . . . . . . . . 67
5.1.4 InfluenceofNa ............................. 70
5.1.5 DiffusionofZn ............................. 76
5.1.6 Discussion................................ 77
5.2 DevicePorperties................................ 79
5.2.1 Interface composition . . . . . . . . . . . . . . . . . . . . . . . . . . 80
5.2.2 Deposition temperature . . . . . . . . . . . . . . . . . . . . . . . . 81
5.2.3 DiffusionofZn ............................. 83
5.2.4 Influenceofalkalis ........................... 85
5.2.5 Discussion................................ 89
5.3 Comparison with buffer free ZnO/CIGSe substrate devices . . . . . . . . . 94
5.4 Summary: ZnO/CIGSe interface . . . . . . . . . . . . . . . . . . . . . . . . 97
6 Back contact and degradation 99
6.1 Thebackcontact ................................ 99
6.2 Degradation...................................101
6.3 Summary: Back contact and degradation . . . . . . . . . . . . . . . . . . . 107
7 Device Modelling 109
7.1 Device model for superstrate solar cells . . . . . . . . . . . . . . . . . . . . 109
7.2 Device model for substrate solar cells . . . . . . . . . . . . . . . . . . . . . 116
7.3 Superstrate vs. Substrate . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
7.4 Summary: Device modelling . . . . . . . . . . . . . . . . . . . . . . . . . . 119
8 Strategies for efficiency improvement 121
8.1 Parameterevaluation..............................121
8.2 Definition of the ideal buffer layer . . . . . . . . . . . . . . . . . . . . . . . 123
8.3 Combinatorial material exploration . . . . . . . . . . . . . . . . . . . . . . 124
8.4 Amorphous Ga2O3buffer............................127
8.5 Summary of and Outlook for improvement strategies . . . . . . . . . . . . 130
9 Summary and Conclusion 133
10 Additional Information 137
CONTENTS vii
10.1 Amorphous Ga2O3characterization......................137
10.2Sulphurgradient ................................143
10.3Lightmanagement ...............................144
10.3.1ZnOannealing .............................145
10.3.2 Highly reflective back contact . . . . . . . . . . . . . . . . . . . . . 151
10.3.3 Ultra thin absorbers . . . . . . . . . . . . . . . . . . . . . . . . . . 154
10.3.4Summary ................................159
10.4CIGSegrowth..................................161
10.5Tandemconfiguration..............................163
List of Journal publications 165
Bibliography 182
Acknowledgement 183
Symbols and Abbreviations
∆EC/V,BC CBM/VBM gradient within
CIGSe at the back contact
IPIonization potential
∆EC/V,S CBM/VBM gradient within
CIGSe at the hetero-junction
χElectron affinity
∆EC,IF CBM offset at the hetero-
junction
Sn,h interface recombination veloc-
ity for electrons and holes
EFFermi level EgBand gap
NAAcceptor state density RSSeries resistance
EBBinding energy RPParallel resistance
αAbsorption coefficient φWork function
LnElectron diffusion length NDDonor state density
lDiode quality factor EF,n Electron quasi Fermi level
JSC Short circuit current density EF,h Hole quasi Fermi level
J0Saturation current density dSCR Space charge region width
VOC Open circuit voltage SP Stoichiometry Point
FF fill factor XRD X-ray Diffraction
ηPower Conversion Efficiency CIGSe Cu(In,Ga)Se2
VBM Valence Band Maximum DOS Density Of States
CBM Conduction Band Minimum SCR Space Charge Region
LLS laser light scattering CISe CuInSe2
ILR Infrared Light Reflectometry CGSe CuGaSe2
WLR White light Reflectometry PDT Post Deposition Treatment
PVD Physical Vapour Deposition PLD Pulsed Laser Deposition
GDOES Glow Discharge Optical Emis-
sion Spectroscopy
XPS X-ray Photoelectron Spec-
troscopy
TCO Transparent Conductive Ox-
ide
EBIC Electron Beam Induced Cur-
rent
SCAPS Solar cell Capacitance Simu-
lator
TEM Transmission Electron Mi-
croscopy
EQE External Quantum Efficiency
Introduction
The highest reported efficiency of CIGSe solar cells is currently at 21.7 % for devices in
substrate configuration and is expected to further increase in the future [1]. But despite
the highest efficiency of all thin film materials, it is not CIGSe, but CdTe which is leading
the thin film solar cell market. CdTe has lower production costs [2] partially due to
the easily up-scalable deposition process of CdTe and partially due to its superstrate
device structure. But what are the advantages of the superstrate configuration? To
answer this, Fig. 0.1a shows the breakdown of the production costs for a framed CIGSe
module in substrate configuration. The costs are dominated by the material costs and
by the depreciation, the latter depending strongly on the processing time per module [3].
A breakdown of the total CO2footprint in Fig. 0.1b shows that the two glass sheets
and the electricity used during the high temperature CIGSe deposition dominate the
CO2emission. Thus both factors, costs and CO2emission, can be strongly reduced by
eliminating one glass sheet and by reducing the processing time and the material usage.
This is where the superstrate configuration has its benefits. The thickness of the TCO
layer can be reduced in the superstrate configuration as it allows improving the electro-
optical properties due to the deposition on a flat substrate and a subsequent annealing.
The back contact can be designed for high light reflection, allowing the use of thinner
CIGSe absorber layers. Materials like Ag, Cu or Al would also allow a reduction of the
thickness of the back contact layer. Another advantage is that the encapsulate can be
opaque, which allows the elimination of one glass sheet and with it a quarter of the total
material costs of a module [4].
The drawback of the superstrate configuration is that the highest reported efficiency
for devices in superstrate configuration is a non-certified and unstable efficiency of 12.8 %,
which was reported in 2001 [5]. This is due to the difficulty to achieve a defined p/n-
junction at the high CIGSe deposition temperature. Chemical reactions between the
CIGSe and the buffer layer was shown to deteriorate the p/n-junction which leads to
lower efficiencies compared to the substrate-type device. For superstrate devices without
buffer layers the TCO window reacts with the CIGSe absorber, forming a GaOxinterfacial
oxide layer at the CIGSe/TCO interface [6]. However, the influence of this oxide layer on
the electric device properties is poorly studied and not yet understood. Also the influence
of the CIGSe deposition conditions onto the formation of this interfacial oxide layer was
not yet investigated. Sodium, which is key to achieve highly efficient CIGSe devices in
4CONTENTS
(a) (b)
Figure 0.1: Break down of a) the costs involved for the production of a CIGSe module [3]
and b) the CO2footprint of the production of an unframed CIGSe module [7].
general, was shown to lead to meta-stabilities in superstrate devices, whereas the deeper
relationship still remain in the dark [5].
The aim of this work is to develop an understanding for the correlation of the interface
chemistry at the CIGSe/TCO interface, the efficiency limitations and the degradation
mechanisms of superstrate solar cells. Pathways to overcome these limitations and
increase the efficiency will be discussed and partially tested. Proof for the above
mentioned advantages of the superstrate configuration will be given as well.
This thesis is organized in the following way. After introducing the basics (Ch. 1-3),
the most suitable TCO for CIGSe superstrate will be evaluated (Ch. 4), which will be
studied in depth (Ch. 5-6), followed by the device simulations (Ch. 7), from which new
strategies for efficiency improvement will be derived (Ch. 8).
Chapter 1: An introduction to the basics of a CIGSe solar cell and to its main active
materials. Literature review for the state of the art of CIGSe superstrate cells is reported.
Thermo-dynamical calculations are presented for possible interface reactions between a
TCO and CIGSe.
Chapter 2: Introduction to the experimental details of the fabrication and the char-
acterization of CIGSe devices as used in the course of this thesis. The basic knowledge
required to understand the characterization methods is given as well.
Chapter 3: Introduction to the numerical device simulations. Exemplary simulations
show the effects of defects at the interface between CIGSe and a TCO.
Chapter 4: Evaluation of different standard TCOs.
CONTENTS 5
Chapter 5: Analysis of the CIGSe/ZnO interface chemistry and its influence on the
electronic properties of the solar cell in superstrate and in substrate configuration.
Chapter 6: The effect of Cu deficiency on the back contact is studied. Degradation
mechanisms occurring at the interfaces to the TCO and the back contact are analysed.
Chapter 7: Device simulations of the J−Vand C−Vcurves of the studied superstrate
solar cell. Together with the results from Chapter 5 and 6 the efficiency limitations are
identified or confirmed. The device properties are correlated to the device properties of
the solar cells in substrate configuration.
Chapter 8: Discussion of different strategies for efficiency improvement. Application
of new buffer materials deposited by combinatorial material exploration.
Additional information:
Detailed opto-electronic and XPS analysis of amorphous GaOx.
Influence of sulphur gradient on the reduction of interface recombination.
Experimental and theoretical proof of potential optical advantages of the superstrate
configuration. ZnO annealing and etching as well as highly reflective back contacts
are studied.
Influence of the ZnO substrate on the morphology of the CIGSe morphology.
Information about CIGSe tandem solar cells.
Chapter 1
CIGSe solar cell basics
1.1 Cu(In,Ga)Se2Solar Cells
This first section will introduce the working principle of a solar cell, the energy band
alignment at the CIGSe/Oxide interface, the superstrate configuration and the state of
knowledge about it, the material basics and the thermodynamics at the CIGSe/Oxide
interface. More in depth information about general solar cells can be found in [8] and on
CIGSe solar cells in [9].
1.1.1 Physics of a Solar Cell
A solar cell is a device that converts the energy of light directly into electrical energy
by the photovoltaic effect [10]. During this process, an electron within a semi-conductor
becomes energetically excited by the absorption of a photon. For photon energies above
the semi-conductor band gap, the electron becomes excited from the valence band into
the conduction band and becomes spatially separated from the positively charged hole by
diffusion. Electron and hole selective contacts to the semi-conductor lead to the permanent
separation of the electrons and the holes, creating a potential difference between these
contacts which can be used to drive an electric current.
The electron selective contact in a CIGSe solar cells is realized by a p/n-junction and
the hole contact by a metal with a similar work function as CIGSe. A p/n-junction
develops at the interface between the p-type CIGSe material, which has a high density
of mobile holes, and a n-type material, like ZnO, which has a high density of mobile
electrons. Once both materials are brought into contact, the free electrons from the n-
type material diffuse into the CIGSe and recombine with the free holes in the CIGSe,
leading to a negatively charged region in the CIGSe and a positively charged region in
the n-type material, together this forms the so-called space charge region (SCR). Fig. 1.1
illustrates this situation. Due to the charge separation an electric field is generated, which
limits the diffusion of the electrons and holes leading to an equilibrium between drift and
diffusion. It should be noted at this point, that this field can also cause electro-migration
of highly mobile ions, like Cu+or Na+, which will become important later in this work.
81 CIGSe solar cell basics
Figure 1.1: Top: Band diagram of a CIGSe/ZnO solar cell calculated with SCAPS (top). The
CIGSe layer is 1
µ
m thick and the ZnO layer 200 nm. The conduction band minimum (CBM)
and the valence band maximum (VBM) is shown together with the electron and hole quasi
Fermi levels, EF,n and EF,h. The space charge region of the p/n-junction under 100 mWcm−2
illumination is marked with the dashed line. The charge separation at the hetero-junction lead
to an electric field (bottom). The local gradient of the quasi Fermi level, determine the flow
of the photo-generated charge carriers. Bottom: The charge distribution within the SCR and
the resulting electric field, which drives the electron into the n-type material.
The equilibrium between drift and diffusion of the charge carriers is equivalent to a flat
Fermi level throughout the device, which tells that the chemical potential of the electrons
is the same everywhere within the device. During illumination electron-hole pairs are
created and the Fermi distribution becomes different for electrons and holes. This requires
the introduction of the quasi-Fermi level for electrons, EF,n and the quasi-Fermi level for
holes, EF,h [11], as shown in Fig. 1.1. During illumination, the quasi-Fermi levels are not
flat within the CIGSe, thus electrons can lower their potential energy by moving to the
minimum of the electron quasi-Fermi level, which is located in the n-type material.
In case that the p-type material and the n-type material are different from each other,
the junction is called hetero-junction. Different electron affinities, χn/p, and/or different
band gaps, Eg,n/p, of these two materials lead to band offsets at the CBM or the VBM
respectively. This is shown in Fig. 1.2 for the example of CIGSe and ZnO. The electron
affinity is defined as the difference between the vacuum energy level and the CBM. If
χp< χnis valid, the offsets can be approximated by the following equations (Anderson’s
law [12]), shown here for the example of CIGSe and ZnO:
1.1 Cu(In,Ga)Se2Solar Cells 9
Figure 1.2: Band diagram of CIGSe and ZnO before and after contact. Shown are the
vacuum levels, the conduction band minima (CBM), the valence band maxima (VBM), the
work function (Φ), the electron affinity (χ), the ionization potential (IP), the band gap (Eg)
and the CBM and VBM offsets between CIGSe and ZnO.
∆EC,IF =χp−χn=χCIGSe −χZnO =4.5eV −4.6eV = −0.1eV,(1.1)
∆EV,IF =(χp+Eg,p)−(χp+Eg,n)=(χCIGSe +Eg,CIGSe)−(χZnO +Eg,ZnO) (1.2)
= (4.5eV + 1.2eV) −(4.6eV + 3.4eV) = −2.3eV.
Especially the conduction band offset ∆EC,IF has a strong influence on the recombi-
nation losses in CIGSe solar cells as will be discussed in Sec. 3.1 and Sec. 3.2.
Loss mechanisms during the conversion from light to electrical energy are thermal re-
laxation of the excited electron, as illustrated in Fig. 1.1, entropy generation, optical losses
as well as radiative and non-radiative recombination. The non-radiative recombination
losses are described in Sec. 3. The conversion efficiency is extracted from the plot of the
photo-generated current density over the photo-voltage, the so-called J−Vcurve, which
is explained in Sec. 2.4.1.
1.1.2 The superstrate configuration
This work studies CIGSe solar cells in the superstrate configuration. In this configuration,
the light passes through the substrate, which is commonly glass, onto which the active
layers are deposited. Thus, during normal operation the substrate is above the active
layers and the configuration is therefore called superstrate. This is shown in Fig. 1.3
together with the standard substrate configuration.
Within the substrate configuration the CIGSe is deposited onto the hole collecting
material, which is usually Mo. The p/n-junction is formed by the deposition of the n-
type materials CdS and i-ZnO. CdS is called the buffer layer between CIGSe and ZnO,
as it reduces the interface recombination velocity at the hetero-interface to CIGSe due to
10 1 CIGSe solar cell basics
Figure 1.3: Substrate (left) and superstrate (right) configuration of CIGSe solar cell devices
(not to scale). In the example for the substrate device, the window layer material is ZnO:Al/i-
ZnO, the buffer layer CdS or Zn(O,S), the back contact Mo or if optical transparency is required
ITO. In the example for the superstrate device, the window layer material is ZnO:Al, the buffer
layer i-ZnO and the back contact Au.
a better lattice match [13], better conduction band alignment [14] and a better chemical
compatibility [15] to CIGSe compared to ZnO. Between the window layer ZnO:Al and
the buffer layer CdS, a thin highly resistive layer of i-ZnO is used to reduce the impact
of shunts on the device performance [16].
In the superstrate configuration the CIGSe layer is deposited onto the electron col-
lecting window layer, which is ZnO:Al. In between the window layer and the CIGSe the
i-ZnO is inserted similar as in the substrate configuration. However, the buffer layer is
dismissed, since CdS as the buffer layer material is not suitable due to its low thermal
stability (see Tab. 1.2). The back contact is deposited onto the CIGSe absorber which
is typically Au. Mo is not used as the successful application requires the formation of
an interfacial MoSexlayer at the interface to CIGSe, but this forms only at high tem-
peratures. In addition, for optimum performance, the CIGSe deposition process should
be altered in order to invert the Ga gradient, which can be used to increase the charge
carrier collection efficiency.
From the technological point of view, the superstrate configuration leads to the follow-
ing advantages and disadvantages:
1. Increased quality of the transparent conductive oxide (TCO), since the TCO can
be deposited on a flat substrate and deposited at whatever temperature is optimal
for the specific TCO. In case that ZnO:Al is used, annealing in nitrogen was shown
to strongly increase the transparency and conductivity [17] (see Sec. 10.3.1 for more
details). Further, in [18] it was shown that the conductivity of ZnO:Al drops if
deposited on rough substrates like CIGSe surfaces. The rough surface leads to an
accelerated ZnO degradation [18].
2. The p/n-junction undergoes the high temperature CIGSe deposition step. This leads
1.1 Cu(In,Ga)Se2Solar Cells 11
to unwanted inter-diffusion (as Zn diffusion) and possible phase transformations (as
GaOxformation [6]). See Sec. 5.1 for more details.
3. The back contact can be designed to achieve low back contact recombination in
combination with high light reflectivity and strong light scattering. This would
allow the use of thinner CIGSe layers, which helps to lower the high material costs,
especially for indium. See Sec. 10.3.2 for more details.
4. Device encapsulation from the back side can be opaque and flexible, which may
lower the costs compared to a second glass sheet as used for substrate devices. The
disadvantage is, that the supporting substrate has to be transparent, which makes
it difficult to use flexible substrates like metal foils.
In addition, to achieve all chalcopyrite 4-terminal tandem devices, it is necessary to
prepare the wide-gap chalcopyrite in superstrate configuration. This will be discussed in
Sec. 10.5.
1.1.3 Literature review
The first efficient chalcopyrite superstrate devices were reported in 1994 by Nakada et
al. [19] and by Negami et al. [20], both with CdS as the buffer layer. The CdS was
observed to diffuse into the CISe absorber at deposition temperatures of 450
and above,
which limited the device efficiency to 8.1 %. Dismissing the CdS buffer and the use of
Ga containing CIGSe, lead to an increase in device efficiency above 10 %, reported in [5]
and [21].
In both reports, the CIGSe was deposited at 550
on top of an i-ZnO layer. The
formation of a thin layer of GaOxbetween the i-ZnO and the CIGSe was observed [6], as
well as meta stabilities when applying a voltage bias or light bias [21] [5].
The highest efficiency reported so far, 12.8 %, was achieved in 2001 by Nakada et al.
by applying a forward bias to the finished device, which enhanced the efficiency from
2 % to 12.8 % [5]. The Cu/(In+Ga) ratio was 0.44, which corresponds to the ordered
vacancy phase Cu(In,Ga)3Se5. Higher Cu contents lead to shunting of the device. 30 mg
of Na2S was deposited during the CIGSe deposition, leading to a Na concentration of
approximately 2 at.% within the CIGSe. In the experiments conducted for this thesis, it
was not possible to reproduce these results with similar experimental conditions.
The second report on superstrate devices with efficiencies above 10 % is from Haug et
al. [21] in 2002. An efficiency of 11.2 % for slightly Cu poor absorbers without any external
sodium supply. Extensive light soaking was required to increase the efficiency from 5 %
temporarily to above 11 %. The light soaking was observed to increase the charge carrier
density in the device [22].
Another interesting approach was reported by Minemoto et al. [23], who exchanged the
i-ZnO layer with ZnMgO which was shown to reduce the interface recombination. The
ZnMgO was deposited by RF co-sputtering without intentional heating. By depositing the
CIGSe absorber at a low temperature of 450
they achieved 9.0 % efficiency. Kaigawa et
12 1 CIGSe solar cell basics
Table 1.1: Overview of the published results on CIGSe superstrate solar cells. All reports
use i-ZnO buffer layers if not noted else wise.
Author PCE Speciality Stability
[5] Nakada 12.8 % ODC absorber, Na2S, forward bias unstable
[21] Haug 11.2 % light soaking unstable
[26] this work 11.0 % low-rate NaF PDT stable
[23] Minemoto 9.0 % ZnMgO buffer -
[19] Nakada 8.1 % CdS buffer -
[20] Negami 6.7 % CdS buffer -
[27] Shafarman 4.0 % ZnSe buffer -
[28] Kaigawa 3.8 % Cu(In,Ga)S2-
[29] Balboul 3.5 % CuGaSe2-
[30] Ikeda 2.9 % In2(S,Se)3buffer -
[31] Nguyen 2.1 % spray pyrolisis CIGSe -
al. prepared wide gap Cu(In,Ga)S2onto ZnO substrates and achieved 3.8 % by a two stage
process. Balboul et al. also processed wide gap CuGaSe2and reached 3.5 %. Shafarman
and co-workers currently work on superstrate CIGSe devices and have reported efficiencies
up to 5.1 % for i-ZnO buffer layers and 4.0 % for ZnSe buffer layers. Recently Ohm et al.
tried to grow CIGSe on ZnO nanorods, but due to strong interface recombination losses
the efficiency could not be increased above 2 % [24].
Back-wall superstrate devices, which have a transparent back contact and are illumi-
nated through this transparent back contact have been realized by Shafarman et al. with
an efficiency of 8.3 % [25]. The efficiency in this case was limited by the back contact
recombination. This approach is in this thesis not further pursued.
In summary, the best results have been obtained with i-ZnO/CIGSe junctions. How-
ever, intensive light- or voltage- soaking were required to achieve efficiencies above 10 %.
In both cases the formation of GaOxwere observed, which were assumed to induce a
CBM spike. However, only few studies on CIGSe superstrate devices were done and
the influence of the GaOxlayer, of sodium supply and of the biasing/ageing is not yet
understood.
1.2 Thermodynamics of the interface reaction
The difficulty of designing a defined p/n-junction is the difficulty to understand and to
control the thermodynamics at the interface between the p-type and the n-type material.
Thermodynamic processes like diffusion or chemical reactions at this interface can deter-
mine the electronic properties of the whole device, as it will be seen in the case of CIGSe
grown on ZnO in Sec. 5.1. The physics involved in these processes will be introduced in
this chapter. The tools which are available to control these processes are temperature,
time and chemical composition.
1.2 Thermodynamics of the interface reaction 13
1.2.1 Gibbs free energy
The “driving force” of all thermodynamic processes is the minimization of the Gibbs free
energy of a given isolated system. Therefore a reaction from one state of a system to
another state is favourable if ∆G < 0. The change of the Gibbs free energy, ∆G, is
defined as follows:
∆G= ∆H−T∆S, (1.3)
with T the temperature, ∆Sthe change of entropy and ∆Hthe change of enthalpy during
the potentially occurring reaction. Thus, ∆Gbecomes negative if ∆Hdecreases or ∆S
increases. The entropy is defined as the amount of energetically equivalent arrangements
of a given system. An increase of entropy therefore increases the disorder within the
system. Diffusion along a defined interface will lead to an increase in entropy, due to an
increase of disorder. The enthalpy is defined as H=U+pV , with Uthe internal energy,
pthe pressure and Vthe volume of the system. At constant pressure a change in enthalpy
is therefore equivalent to a change in volume or heat of the system. A chemical reaction
is mostly driven by a reduction of the enthalpy of the system, for example an increase of
inter-atomic binding energy. With ∆Sand ∆Hdefined as:
∆H=H(products)−H(reactants) (1.4)
∆S=S(products)−S(reactants),(1.5)
with S(products) and H(products) being the entropy and enthalpy of all products,
S(reactants) and H(reactants) of all reactants. Since the enthalpy is a potential en-
ergy and cannot be measured, it is set to 0 for single elements at the standard state of
293 K and 1 atm. The so-called “enthalpy of formation” ∆Hof a compound system, like
CuInSe2, is then referred to the standard enthalpy of the constituting materials, Cu, In
and Se. It follows for a phase transformation ∆H= ∆H(phase1) −∆H(phase2) and
∆S= ∆S(phase1) −∆S(phase2).
A reaction can spontaneously occur if ∆H < T∆Sis fulfilled. Thus, at higher tempera-
tures, the equilibrium between disorder and locally defined stable phases is pushed towards
disorder. To perform the exact calculations for a given system at temperatures different
from 293 K it is necessary to know the temperature dependence ∆Hand ∆S, which de-
pend on the heat capacity at constant pressure Cp(=∆H/∆T=a+bT+c/T2+∆T2) of the
compound. Further, starting from a certain concentration of different elements present
locally within a small volume a large amount of different phases have to be considered
which can possibly occur. Commercial programs like ChemSage [35] are able to calculate
the phase equilibriums at any given temperature from a given molar concentration of
starting materials or compounds. The advantage of these programs is the large database
of enthalpies, entropies and heat capacities for a large selection of different compounds.
Results of the calculations for the interface of ZnO with Cu(In,Ga)Se2or the binary com-
pounds Ga2Se3, In2Se3and Cu2Se at 1 bar and 900 K are shown below. It should be
noted however, that no temperature dependence of these reactions could be observed and
14 1 CIGSe solar cell basics
Table 1.2: Standard enthalpies of formation and standard entropies of different compounds
at 293 K and in solid state, taken from [32], [33] and [34]. The values for Cu(In,Ga)Se2are
interpolations of the CuInSe2and CuGaSe2values.
Compound ∆H / kJ mol−1S / J K−1mol−1
CuInSe2-262 158.0
CuGaSe2-313 147.0
Cu(In,Ga)Se2-273 160.0
ZnO -350 43.6
Ga2O3-1089 84.9
In2O3-926 104.1
SnO2-581 52.3
Ga2Se3-408.8 179.9
In2Se3-309 163.6
Cu2Se -65 129.7
ZnSe -170 77.6
CuGaO2-637 83.2
Cu2O -198 93
MoO3-745 77.8
CdS -177
calculations for 300 K lead to the same results.
4Cu(In, Ga)Se2+ 3ZnO ⇒Ga2O3+ 2CuInSe2+Cu2Se + 2ZnSe (1.6)
2CuGaSe2+ 3ZnO ⇒Ga2O3+Cu2Se + 2ZnSe (1.7)
2CuInSe2+ 3ZnO ⇒In2O3+Cu2Se + 2ZnSe (1.8)
2(In, Ga)2Se3+ 3ZnO ⇒Ga2O3+In2Se3+ 3ZnSe (1.9)
Ga2Se3+ 3ZnO ⇒Ga2O3+ 3ZnSe (1.10)
In2Se3+ 3ZnO ⇒In2O3+ 3ZnSe (1.11)
Cu2Se +ZnO ⇒Cu2Se +ZnO (1.12)
The higher enthalpy of formation of Ga2O3and In2O3compared to ZnO leads to a com-
plete cation exchange of the oxide and the selenide. If In and Ga are present simultane-
ously, then oxygen binds preferably the gallium atoms. It has to be kept in mind, that
these calculations only indicate preferable reaction mechanism. The change in free energy
for the reaction of ZnO with In2Se3is -528 kJ/mol which is very close to the change in
free energy for the reaction with Ga2Se3, which is -581 kJ/mol.
Also other TCOs like SnO2and In2O3are all predicted to selenize, once they are in
contact with Ga2Se3:
Ga2Se3+In2O3⇒Ga2O3+In2Se3(1.13)
2Ga2Se3+ 3SnO2⇒2Ga2O3+ 3SnSe2.(1.14)
1.2 Thermodynamics of the interface reaction 15
Even though the change in the free energy is only -53 kJ/mol for the reaction of Ga2Se3
with In2O3, it has been observed experimentally [36].
Only Cu2Se is predicted not to react with ZnO. At elevated temperature however, the
entropy becomes more important and diffusion of Cu into the ZnO is expected to occur.
Also, it has to be kept in mind, that these calculations do not take the reaction kinetics
into account, which potentially limits the cation exchange between the materials at low
temperatures. Thus even though the reactions are proposed to occur at room temperature
it is likely that they are limited by slow diffusion processes, which will be introduced in
the next section.
1.2.2 Diffusion
As described above, inter-diffusion between the elements of two adjacent layers, is driven
by an increase in entropy and a reduction of the free energy of the system. The change
of the concentration Nof a certain element at the position xover the time tdepends on
the concentration gradient of that element at that point [37]. In the one dimensional case
and a spatially constant diffusion constant D, this can be written as
∂N
∂t =D∂2N
∂x2.(1.15)
The diffusion constant describes the thermally activated hopping process of the elements
or ions through a given crystal lattice. It can be described as follows:
D(T) = D0exp(−EA/kT ),(1.16)
with D0being the diffusion constant at infinite temperature and EAthe activation energy
of the hopping mechanism. The hopping may occur between interstitial, vacancy or
occupied lattice sites. For example, in [38] the diffusion constant for Zn in polycrystalline
CIGSe and single crystal CISe was measured to be D01.3e-12 cm2s−1with EA1.24 eV. A
second, slower, diffusion mechanism was also identified. It was speculated that the faster
diffusion mechanism is via Znistates and the slower one via ZnCu.
The shape of the diffusion profile depends on the boundary conditions at the material
interfaces, on whether the material supply is limited or not. In the case of sodium doping,
the supply is limited if a thin layer of NaF is deposited on the CIGSe layer prior to the
annealing and it is unlimited if the NaF is deposited continuously with a low rate during
annealing. In the case of an unlimited supply the diffusion profile has the form of an
complementary error function:
N(x, t) = N0erf x
2pD(T)t!,(1.17)
with N0being the constant dopant concentration at the surface. In the case of limited
supply, the surface concentration is described by the initial concentration N0,i and N0(t) =
16 1 CIGSe solar cell basics
Figure 1.4: Normalized diffusion profile of a dopant diffusing into a material with a single
diffusion constant
−N0,i/(2pD(T)tπ). In this case the profile decays exponentially with the square of the
distance from the surface:
N(x, t) = N0(t) exp −x2
4D(T)t.(1.18)
The resulting diffusion profiles depend strongly on the temperature and the time the sys-
tem is hold at this temperature. Fig. 1.4 compares the diffusion profiles for unlimited and
for limited supply, both calculated with equal time, temperature and diffusion constants.
1.3 Material Characteristics
The materials which will be studied most within this thesis are CIGSe, ZnO and amor-
phous GaOx. However, very little literature exists on amorphous GaOx, why its electro-
optical properties are studied in Sec. 10.1.
1.3.1 Cu(In,Ga)Se2(CIGSe)
CIGSe is one of the most successful absorber materials for thin film photovoltaic applica-
tions due to its high absorption coefficient, low density of deep defects and the possibility
to engineer the band gap. This section will give a short overview of the crystal structure
and the electronic structure.
CIGSe structure
The semi-conductor Cu(In,Ga)Se2belongs to the I-III-VI compounds and crystallizes in
the chalcopyrite structure. The chalcopyrite is a superstructure of the ZnSe sphalerite
structure with ordered substitution of Zn with Cu, In and Ga. The lattice has a body
centred tetragonal order in which each Cu, In or Ga cation is tetrahedrally coordinated to
1.3 Material Characteristics 17
Figure 1.5: Tetragonal chalcopyrite structure of Cu(In,Ga)(Se,S)2shown together with the
lattice parameters a(5.75 ˚
A) and c(11.5 ˚
A) for Ga/(Ga+In)=0.23 and Se/(Se+S)=1 (left).
Image taken from [39]. Band gap values and relative position of the CBM and VBM for the
different chalcopyrite compounds compared to the CMB and VBM of CuInSe2(second from
left). Values taken from [40].
four Se anions and each Se anion is tetrahedrally coordinated to two Cu and two In/Ga
atoms. The different bonding strength of the weak covalent Se-Cu bond compared to
the partially ionic bond of Se-In or Se-Ga leads to a tetragonal distortion [41]. For the
Ga/(Ga+In) ratio of 0.23 the distortion is compensated and the lattice parameter cequals
twice the lattice parameter a. It was shown that poly-crystalline Cu(In,Ga)Se2with this
ratio, deposited via co-evaporation, exhibit a larger average grain size compared to other
Ga/(Ga+In) ratios [42].
The band structure of Cu(In,Ga)Se2is derived by density-functional (DFT) theory [43].
The valence band emerges from a hybridization of the Se 4porbitals and the Cu 3dorbitals.
A Cu deficiency therefore leads to a lowering of the valence band maximum [40]. Fig. 1.5
shows the lowered VBM of the ordered vacancy compound CuIn5Se8compared to the
CuInSe2compound. The conduction band is mainly derived from In 5s/ Ga 4sand
Se 4porbitals. A change in the Ga/In ratio leads to a change in the position of the
conduction band minimum (CBM) [44]. Fig. 1.5 shows the difference in the CBM of the
CuGaSe2and the CuInSe2compound. Since Se porbitals contribute to the valence and
conduction band, the substitution of Se with S influences the position of the conduction
and valence band [45], as shown in Fig. 1.5.
Defects in CIGSe
The relevant intrinsic defects within CuInSe2are listed in Tab. 1.3. The shallow defects
in CuInSe2are the acceptor VCu and the donor InCu, which have low formation energies
18 1 CIGSe solar cell basics
Table 1.3: Properties of the native defects and defects which are possibly induced due to the
diffusion of elements from ZnO into the CIGSe. The description ”shallow” refers to defects
energetically close to one of the bands and contribute to the conductivity, ”deep” refers to
localized states which act as recombination centers.
Origin Defect Type Position Ref.
in the band gap
VCu acceptor shallow [46]
VIn acceptor deep [46]
CuIn acceptor deep [46]
intrinsic VSe donor deep [46]
InCu donor shallow [46]
Cuidonor deep [46]
(VSe-VCu)* amphoteric shallow [46]
from OSe acceptor deep [47]
ZnO ZnIn acceptor shallow [15]
ZnCu donor shallow [15]
Alkali NaIn acceptor deep [46]
treatment NaCu neutral [46]
Naidonor [46]
for Cu poor compositions, whereas the defects related to Cu rich compositions like
CuIn or Cuiare deep defects which will degrade the electronic quality of the crystal.
Within Cu poor crystals it is energetically favourable when two VCu
−acceptors and one
InCu2+ donor form neutral defect pairs. This can lead to neutral chalcopyrite crystals
even for strong off-stoichiometric conditions. Small deviations from the 2:1 defect
ratio determine whether the crystal becomes p-type or n-type. Due to the very low
formation energy of the acceptor VCu the deviation is generally directed towards p-type.
Cu-poor CuInSe2can turn n-type by extrinsic doping as long as the Ga concentration is
relatively low. For high Ga concentrations the formation energy of VCu becomes negative
once the Fermi level increases to mid-gap, this increases the VCu concentration and
compensates the n-type doping. Selen vacanies, VSe are predicted to form amphoteric
defect pairs with VCu, changing from donor to acceptor depending on the Fermi level po-
sition. This is often assumed to be the origin of the observed metastabilities in CIGSe [48].
Extrinsic defects can lead to a n-type conductivity in CIGSe with low Ga content [15],
which could be beneficial at the CIGSe interface to the buffer material. CdCu induced
by a CdS buffer layer is a shallow donor in CuInSe2, which qualifies CdS as a buffer
layer in substrate devices. ZnCu is known to be a shallow donor as well, which qualifies
Zn containing buffer layers as alternatives to CdS. However, ZnCu is not as shallow as
CdCu. In CuInSe2, ZnCu is predicted to be 60 meV below the conduction band and for
1.3 Material Characteristics 19
CuGaSe2110 meV [15], which could introduce recombination centres in the chalcopyrite.
ZnIn is predicted to be a shallow acceptor lying within the valence band for CuInSe2and
70 meV above the valence band in CuGaSe2. Assuming a similar trend as calculated for
Cd [15], the formation energy for ZnCu is supposed to be smaller than for ZnIn under
p-type condition. This predicts a preferential occupation of Cu sites resulting in an
increase of donor states due to the presence of Zn. However, experimental observation
on the closely related CuInS2system do not confirm these predictions. Neutron powder
diffraction experiments showed a preferred occupation of In sites by Zn [49]. Enzenhofer et
al. [50] reported a preferred occupation of In sites in the bulk of CuInS2and a preferred
occupation of Cu sites at the Cu-poor surface. Similar results were obtained in [51].
Secondary ZnSe phases are unlikely to form, as Schorr et al. [49] found that a phase
separation occurs only for molar concentrations above 8 % of ZnSe in CuInSe2.
Selen vacancies can be occupied by oxygen atoms, transforming the donor states into
deep OSe acceptors. Interestingly, CuInSe1−xOxwas shown not to be stable and would
decompose into CuInSe2, In2O3and Cu2O [47]. Thus it is unclear whether oxygen acts
as an acceptor or if it leads to the precipitation of binary phases.
Sodium is known to increase the p-type doping in CIGSe, if present at low concentra-
tions [52]. It will preferably occupy Cu sites and NaCu is electrically inactive. However,
NaInCu neutralizes a double donor state, whereas NaVCu neutralizes a single acceptor.
NaVCu has a higher formation energy as NaInCu , which leads to the preferred occupation
of InCu states by Na. This is believed to be the origin of the increased p-type doping
density induced by sodium [47]. Photoluminescence measurements confirmed that the
enhanced p-type doping originates from the annihilation of donor states and not from the
formation of new acceptor states [53].
1.3.2 ZnO
ZnO is one of the most common TCOs used in photovoltaic devices, due to its wide
band gap, high conductivity and relatively low material costs. In CIGSe devices, ZnO
doped with Al or with Ga is usually used as the window layer and in superstrate devices
intrinsic ZnO sometimes form the p/n-junciton with CIGSe [54]. In this section the
crystal structure, the band structure, the intrinsic and the extrinsic defects of ZnO will
be summarized.
Structure
ZnO layers deposited on glass substrates crystallizes in the hexagonal wurtzite structure
[55]. The structure is shown in Fig. 1.6 together with the lattice parameters a and c.
The structure is composed of two inter-penetrating hexagonal-close-packed sub lattices
for each type of atom, which are displaced to each other by one atom along the c-axis.
Sputtered ZnO layers are polycrystalline and preferably grow in the [001] direction, which
is along the c-axis.
The energy band structure resulting from DFT calculations for wurtzite ZnO is shown
20 1 CIGSe solar cell basics
Figure 1.6: Unit cell of ZnO in the hexagonal wurtzite structure shown together with the
lattice constants a (3.25 ˚
A) and c (5.207 ˚
A)(left) [60]. Corresponding band structure obtained
from DFT calculations, corrected to the experimental band gap, whereas the grey stripe indi-
cate the calculated band gap (right) [56].
on the right hand of Fig. 1.6. The direct band gap of 3.44 eV was corrected to the ex-
perimental value [56]. The valence band is derived from O 2pand Zn 4sorbitals. The
conduction band is almost solely derived from Zn sorbitals [56]. Thus, alloying ZnO
with MgO to (Zn,Mg)O influences the position of the valence band maximum and the
conduction band minimum [57]. Alloying ZnO with ZnS to Zn(O1−x,Sx) shifts only the
valence band maximum for low sulphur concentrations, x < 0.3. The conduction band
is shifted as well for larger concentrations of sulphur [58]. The density of states (DOS)
for the conduction and valence bands are shown in Fig. 1.6 as well. The low DOS at the
CBM can cause an increase of the optical band gap, in case the ZnO becomes strongly
n-type doped as described by the Burstein-Moss effect [59].
Defects
The energetic position of the native defects in ZnO were calculated by Janotti et al [61].
It was shown, that they do not contribute to the n-type conductivity of ZnO. The oxygen
vacancies VOare deep donors and can only compensate p-type doping, but cannot lead to
n-type conductivity. The Zn interstitial Zniis a shallow donor, but has a high formation
energy under n-type conditions, which limits the density of Znidefects. The same holds
for Zinc antisites. Zinc vacancy VZn on the other side have low formation energies under
n-type conditions. They are deep acceptors and partially compensate the n-type doping.
The out-diffusion of Zn during the CIGSe deposition onto the ZnO would increase the
VZn density and could be a potential source of p-type doping.
Oxygen interstitials and anti-sites can also act as acceptors but they have very high
formation energies and are assumed to have no significant impact.
When growing CIGSe onto ZnO at temperatures between 400
and 600
, inter-
diffusion between these materials is likely to occur. Table 1.4 lists the possible generated
defects by this process. The diffusion of Ga and In into ZnO increases the n-type conduc-
1.3 Material Characteristics 21
Table 1.4: Properties of the native defects and defects which are possibly induced due to the
diffusion of elements from NaF doped CIGSe into the ZnO. The description ”shallow” refers
to defects energetically close to one of the bands and contribute to the conductivity, ”deep”
refers to localized states which act as recombination centers.
Origin Defect Type Position Ref.
in the band gap
VZn acceptor deep [61]
intrinsic VOdonor deep [61]
Znidonor shallow [62]
CuZn acceptor deep [63]
CIGSe GaZn donor shallow [64]
deposition InZn donor shallow [64]
SeOdonor deep [65]
NaZn acceptor deep [66]
Alkali KZn acceptor deep [66]
treatment Naidonor shallow [66]
Kidonor deep [66]
FOdonor shallow [62]
tivity, since they are common doping elements for ZnO by introducing the shallow donor
states GaZn and InZn [64]. Se is the least mobile atom within CIGSe, which makes the
diffusion of Se into the ZnO unlikely. SeOis assumed to be a deep donor [65] and should
have no significant influence on the ZnO properties for low concentration.
The diffusion of Cu into ZnO is more critical since Cu is incorporated preferably on
the Zn sites which leads to CuZn acceptor states [63]. P-type conductivity was not yet
achieved by Cu doping, since the CuZn is too deep. Still, rectifying ZnO homo-junctions
were claimed by doping ZnO with 3 at.% of Cu [67]. A problem of p-type doping ZnO
is the compensating nature of the native VOdeep donor, whose formation energy drops
and concentration rises with increasing p-type doping.
The diffusion of Na and K from the glass substrate or from the CIGSe absorber lead
to amphoteric defects within the ZnO. For low concentrations they are predicted to form
deep acceptor states due to the occupation of Zn sites, NaZn or KZn. However, once the
Fermi level moves towards mid-gap, the formation energy of the Naiand Kibecomes
reduced and the Naiand Kidonor states compensate the p-type doping [66].
Chapter 2
Experimental: Deposition and
Characterisation
2.1 CIGSe growth
The CIGSe layers studied in this work were deposited by co-evaporation of Cu-In-Ga-Se,
as this method allows a good controllability of the Ga-gradient and keeps the selenization
of the underlying substrate to a minimum compared to selenization processes of metallic
precursors. This section will shortly introduce the physical vapour deposition (PVD)
system, the process control and the deposition process it self.
PVD Chamber
The evaporation of Cu-In-Ga-Se was done within a vacuum system, similar to a Molecular
Beam Expitaxy (MBE) system, the so-called PVD-B chamber at the HZB, which is shown
in Fig. 2.1. The deposition was performed at a background pressure of 1e-7 mbar. The
evaporation sources were effusion cells with pyrolytic boron nitride (PBN) crucibles for
Cu, In, NaF and a vitreous carbon crucible for Ga. The source-substrate distance is
530 mm leading to a inhomogeneity of +/- 3 % in the deposited layer thickness across a
200x200 mm2deposition area. Selenium is supplied via a Valved Selenium Cracker Source
(VSCS) which optionally provides thermally cracked selenium molecules (not yet used)
and allows fast regulation of the Se flux. The substrate is heated by a resistive wire heater
with a PBN ceramic diffuser. A maximum substrate size of 200x200 mm2can be coated
and heated uniformly. A cooling shroud fed with water is used to cool the inside walls of
the PVD to limit re-evaporation of previously condensed material.
Process control
Apyrometer is used to measure the substrate surface temperature. It measures the
thermal radiation at a fixed wavelength. The radiation intensity increases with the tem-
perature to the power of four and linearly with the material emissivity. The latter re-
24 2 Experimental: Deposition and Characterisation
(a) (b)
Figure 2.1: a) Schematic sketch of the PVD. 1. Manipulator; 2. Motor for substrate rotation;
3. Substrate heater; 4. Thermo-element; 5. Rotating substrate holder; 6. Substrate shutter;
7. ILR and LLS; 8. LLS detector; 9. IR detected by Pyrometer; 10. ILR detector; 11. Quartz
crystal balance; 12.-14. Cu, In, Ga effusion cells; 15. Valved Se Cracker Source b) technical
drawing of the PVD-B from Createc, showing the cooling shroud (yellow) within the PVD [68].
2.1 CIGSe growth 25
quires a calibration of the measured temperature, which can be the softening point of
glass for example. Laser light scattering (LLS) is a common technique to control the
co-evaporation process of CIGSe [69] [70] [71]. For this technique, chopped laser light,
which is scattered from the growing film, is recorded by a Lock-In amplifier. The signal
intensity is influenced by the film thickness, due to interference effects, by the film rough-
ness, due to scattering, and by the refractive index of the growing film. Thus, changes
in one of these parameters can be detected by LLS. The stoichiometry points (SP1 and
SP2) are detected by the change of the refractive index induced by the segregation of
a CuSexphase on top of the Cu-rich CIGSe surface. A He-Ne laser with a wavelength
of 633 nm is used. Thus, the interference effect disappears the moment the band gap is
low enough to absorb the laser light. Infrared Laser Reflectometry (ILR) records
specular reflected monochromatic infrared light with λ=1040 nm. As the refractive index
of the growing material is approximately known, the number of the interference maxima
and minima are a measure for the film thickness during the deposition. White Light
Reflection (WLR) is a relatively new process control which was first reported by [72]
to get an exact value for the film thickness. Within this work the technique has been
refined to allow exact band gap, roughness and defect absorption of the growing film.
Within WLR the specular reflected white light is recorded with two diode arrays to cover
the wavelength range from 400 nm to 1600 nm. The reflection spectrum depends on the
thickness, roughness, band gap, Urbach energy and refractive index. As the refractive
index can be reasonable approximated from literature values it is possible to extract all
these parameters instantly. The technique is not yet applied to the CIGSe growth for
superstrate solar cells.
CIGSe thin film deposition
The deposition process for the CIGSe absorbers followed a similar routine as the well-
known 3-stage process [73], but is slightly modified to achieve a Ga profile, which is
inverted compared to the standard process to account for the superstrate structure. The
key of the 3-stage process is that the CIGSe layer is grown Cu-poor while it passes through
a Cu-rich regime only close to the end of the deposition process. The Cu-rich regime
induces a re-crystallization [74], which leads to reduced film stress [75] and crystallographic
disorder [76]. Further, the Ga profile can be well controlled by this process. The modified
3-stage process used in this work will be shortly explained in the following. During all
the three stages the Se deposition rate is kept constant at around 15 ˚
A/s.
1st stage: Deposition of a layered Ga2Se3/In2Se3/Ga2Se3/In2Se3stack at 300
substrate temperature. The thickness ratio of the Ga2Se3layers to the In2Se3de-
termines the minimum Ga/(Ga+In) ratio and therefore the minimum band gap of
the CIGSe device, as shown in Fig. 2.3. The layer stack has also been varied during
this work as will be discussed later.
2nd stage: Co-evaporation of Cu-Ga-Se. The temperature is increased with
40 K/min to typically 525
in stage 2A. During this stage, the increased tem-
26 2 Experimental: Deposition and Characterisation
Figure 2.2: Modified 3-stage process for a superstrate-type CIGSe deposition. The deposition
rates are chosen to lead to an inverted Ga profile compared to the standard 3-stage process.
The Se evaporation rate was constant throughout the process at 15 ˚
A/s, which corresponds to
a Se/In rate ratio of around 3. The LLS and ILR signals are used to control the deposition
process. The increase of the LLS signal in stage 2 indicates the first point of stoichiometry,
SP1, and the decrease the second point SP2.
2.1 CIGSe growth 27
(a) (b)
Figure 2.3: GDOES depth profile of Ga/(Ga+In) ratio resulting from the modified 3-stage
process at a) 550
and b) 420
substrate temperature. The Ga/(Ga+In) gradient is deter-
mined by the Ga rates and the deposition temperature. During deposition of the 420
layer,
the Ga rate was set too low in the 3rd stage, leading to a weak Ga gradient close to the surface.
perature leads to an equalization of the Ga/(Ga+In) ratio of the pre-deposited
stack from the 1st stage. The evaporated Cu diffuses into this stack leading to the
formation of Cu(In,Ga)5Se8phase, followed by the Cu(In,Ga)3Se5phase and then
the Cu(In,Ga)Se2phase in stage 2B and 2C. The Ga/(Ga+In) ratio of the growing
CIGSe layer increases with time due to the co-evaporation of Ga (GDOES depth
profile shown in Fig. 2.3). Once the stoichiometry point is reached a CuSexphase
starts to form on the CIGSe surface. CuSexhas a higher refractive index compared
to CIGSe and leads to an increase of the LLS signal once it segregates on the CIGSe
surface. This is used for the detection and defines the end of stage 2B. In stage 2C
only Cu-Se is evaporated until the CIGSe/CuSexstack reaches a Cu/(Ga+In) ratio
of around 1.1.
3rd stage: Co-evaporation of Ga-In-Se, with a Ga/In rate ratio of 4. The tem-
perature is kept constant at the temperature of the 2nd stage. The CuSexphase is
transformed to CIGSe and the second point of stoichiometry is marked by a decrease
of the LLS signal. The Ga-In-Se evaporation is continued until a Cu/(Ga+In) of
0.85 is achieved.
At the end of the 3rd stage, the Se evaporation rate is reduced from 15 ˚
A/s to 1 ˚
A/s
and the substrate is kept at the deposition temperature for 5 minutes, before the substrate
is cooled down with 20 K/min. At a substrate temperature of 250
the Se source is fully
closed.
NaF Post-Deposition Treatment (NaF PDT)
The sodium supply for the CIGSe layers were controlled by the deposition of the compound
material NaF onto the finished CIGSe layer. This was done by thermal evaporation
28 2 Experimental: Deposition and Characterisation
within the same PVD chamber used for the CIGSe deposition. However, in between both
depositions, the samples were brought in contact with ambient air for several seconds.
If not stated differently, the NaF PDT was performed as following: 10 nm of NaF were
deposited at a rate of 3 ˚
A/s at substrate temperature of 300
. The substrate temperature
was then hold for 10 minutes prior cooling down to room temperature within the vacuum
chamber. The term ”low-rate NaF PDT” refers in this thesis to the deposition process
during which NaF is deposited constantly for 10 minutes at a reduced deposition rate of
10 ˚
A/min. This is supposed to lead to a different diffusion profile within the CIGSe layer
(see Sec. 1.2.2). Generally the NaF PDT was not followed by any washing steps.
Ga profile
The resulting Ga profile from the above described deposition process at 525
and without
sodium supply is shown in Fig. 2.3a. The CIGSe bulk is characterised by a shallow Ga
gradient which supports the electron collection. The bulk gradient is mainly determined
by the Ga rates in stage 1 and 2. The CIGSe close to the surface is characterised by a
steep Ga gradient, mainly determined by the Ga rate in stage 3. Since the surface layer
forms the back contact, the Ga gradient helps to reduce the back contact recombination
losses by reducing the electron density at the back contact.
The profile strongly depends on the deposition temperature. For temperatures below
500
, the layered structure from the 1st stage becomes visible in the Ga profile, as seen
in Fig. 2.3b. This has the benefit, that the Ga content at the hetero-junction is increased
compared to the CIGSe bulk. The disadvantage is the notch in the Ga profile at 0.5
µ
m
film depth. It appears, that during the Cu-rich phase preferably In diffuses to the surface
and reacts with the CuSexphase, leading to the observed notch.
2.2 TCO growth
The ZnO layers used during this work were deposited by dynamical radio frequency (RF)
magnetron sputtering, while most of the new buffer layers were deposited with com-
binatorial pulsed laser deposition (PLD). Here, both deposition techniques are shortly
introduced.
Sputter deposition
Sputtering is a physical vapour deposition in which the target material ejects atoms
due to bombardment with accelerated Ar ions [77]. The Ar ions are confined within a
magnetic and electric field of a magnetron close to the target. To avoid charging of the
target, as often observed in insulating targets, an alternating electric field at the radio
frequency can be used. In dynamic sputtering the substrate moves back and forth below
the fixed target. The ejected target atoms condensate on the substrate and form a thin
film. On the way to the substrate, the atoms undergo scattering with Ar gas (which is
above 1e-3 mbar) and with background impurities (below 1-6e mbar). The total pressure
2.2 TCO growth 29
Table 2.1: Process parameter of the ZnO:Al and i-ZnO sputter deposition.
Parameter ZnO:Al i-ZnO
Sputter system VISS 300 S VISS 300 S
Target dopant concentration 1 wt% Al2O3-
Base pressure <4e-7 mbar <4e-7 mbar
Ar pressure 1.5
µ
bar 8
µ
bar
Ar/O2gas flow 0 1
Sputter power (target size 0.075 m2) 2.5 kW 1.5 kW
Substrate temperature 180
without heating
Substrate velocity 0.25 m/min 1.1 m/min
Runs (dynamic mode) 18 28
and the sputtering power control the energy of the particles reaching the substrate and
both have a strong influence on the film growth. Further, the film growth depends on
the substrate temperature. The process parameters used for the ZnO layers used in this
work, if not mentioned otherwise, are shown in Tab. 2.1. Prior to the ZnO deposition,
the glass substrates were cleaned and optionally a 185 nm thick SiOxdiffusion barrier to
avoid sodium diffusion was sputtered onto the glass.
Combi-PLD deposition
Combinatorial pulsed laser deposition (PLD) was used to fabricate TCO layers with
compositional gradients from a wide range of different materials. PLD uses laser ablation
[78] to create a plasma from a defined solid state target. This has the advantage, that
the chamber does not heat up during deposition and cross contamination from previous
depositions is less severe than in other deposition methods. A schematic drawing of
the PLD system is shown in Fig. 2.4. Material from up to six different targets can be
deposited onto a single substrate. Whereby the deposition of the different materials is
done sequentially, by switching the targets while the position of the laser beam is fixed. No
simultaneous depositions are possible. The target position is off-center to the substrate
center, leading to thickness gradient on the static substrate. Turning the substrate in
between the deposition steps, allows to create a material concentration gradient. The
laser ablating is induced by a pulsed KrF laser (248 nm), with an energy density of 300
mJ, a repetition rate of 10 Hz and a distance of 10 cm between the material source and
the sample substrate.
One deposition step consists of maximum 50 laser pulses, which ablate material from
one target, like Ga2O3, while the substrate is kept at a fixed position. For 50 pulses,
the deposited material thickness varies from around 1 nm to 0.25 nm, depending on the
position on the 5 cm x 5 cm large substrate. In the next deposition step material from a
different target can be deposited onto a different position on the substrate. The sequen-
tially deposited layers are supposed to inter-diffuse vertically, while horizontally, almost
any arbitrary material gradient can be achieved. The procedure is repeated until the
30 2 Experimental: Deposition and Characterisation
(a) (b)
Figure 2.4: a) Schematic drawing of the Pulsed Laser Deposition (PLD) system. The target
carousel can switch between six targets. The laser ablated material hits the sample off-center,
leading to a film thickness gradient. Turning the sample when switching the target creates
material gradient. b) A 5x5 cm glass/GZO/InGaOx/CIGSe/Au sample with a X-gradient
(thickness) and a Y-gradient (Ga content). The Au pixels define the active area of the solar
cell, leading 64 solar cells each with a slightly different buffer layers.
desired film thickness is reached.
2.3 Metallization and device layout
After the TCO and the CIGSe deposition, the back contacts were deposited onto the
CIGSe. As a standard back contact material Au is used. Au was thermally evaporated
within an electron beam evaporator. A strong electric field combined with a magnetic
field is used to accelerate and steer the electron beam, which is ejected from a tungsten
filament by thermionic emission, onto the Au target, which heats up and evaporates. The
Au vapour condensates onto the substrate leading to the formation of a thin Au film. A
shadow mask in front of the substrate was used to create defined back contacts in different
sizes, as shown in Fig. 2.5. The pressure during the evaporation was typically 2e-6 mbar.
The distance between the Au target and the rotating substrate was 20 cm, leading to
a slight heating of the substrate to approximately 60 during the process duration of
around 8 minutes. The layer thickness was 100 nm.
The metallization of the PLD samples was performed with a different shadow mask
to allow the analysis of the graded materials. 64 Au pixels, each 3x3 mm in size, were
deposited onto a 50x50 mm substrate, as shown in Fig. 2.4b.
2.4 Device Characterization 31
Figure 2.5: Layout of the Au back contacts on a 50 x 25 mm large substrate. Generally five
solar cells were defined by the Au back contacts with three different sizes. The substrate edges
were free of CIGSe and the two Au bars were in direct contact to ZnO.
2.4 Device Characterization
To understand the electronic processes which determine the performance of the solar cell,
it is necessary to study the recombination and the charging processes occurring at different
applied voltages. This can be done by measuring J−Vand C−Vcurves and describing
them analytically or numerically based on physical models. Further clarification regarding
the recombination losses can be gained from the quantum efficiency measurement (QE),
which measures the photo-current wavelength dependent, and from the electron beam
induced current (EBIC) measurement, which locally resolves the current generation. In
this section these four techniques will be introduced.
2.4.1 J−Vmeasurements
The J−Vcurve displays the current density Jversus the applied voltage Vof a device
under test in the dark or under illumination. For a standard rectifying p/n-junction,
fabricated from homogeneous semi-conductors, the characteristics of the J−Vcurve can
be described by the one-diode model following the equation [11]:
J=J0exp q(V−JRs(V))
lkT −1+V−JRs(V)
Rp−JphotoηJ(V),(2.1)
with the following parameters:
1. The voltage dependent series resistance Rs(V), describing the ohmic losses but also
barriers for current extraction. Can be therefore different in the dark and under
illumination. Shown in Fig. 2.6.
2. The parallel resistance Rp, describing the parasitic current pathways. Has to be
measured in the dark. Shown in Fig. 2.6.
3. The ideality factor l, describes the pathway of the recombination current. l= 1
for recombination at the interface or outside the space charge region. l= 2 for
recombination in the space charge region.
4. The reverse saturation current J0, describing the recombination losses.
32 2 Experimental: Deposition and Characterisation
5. The photo-current JphotoηJ(V), where Jphoto describes the maximum extractable
photo-current and ηJ(V) describes the voltage dependent efficiency for the extraction
of photo generated electron-hole pairs [9].
This equation can be solved analytically if the photo-current is not voltage dependent.
However, this is not the case for the solar cells studied in this work. This leads to ideality
factors above 2, which has no physical meaning within the simple diode model. Therefore,
the J−Vcurves shown in this work, will be analysed numerically as described in Sec. 3.
Still, it will be referred to the standard characteristics of a J−Vcurve of a solar cell, like
fill factor FF, open circuit voltage VOC, short circuit current JSC and finally the power
conversion efficiency η(PCE). These values are defined in Fig. 2.6. Further, analytical
expressions are useful to express qualitatively evaluable trends. A good example is the
open circuit voltage, which, under negligence of the shunt conductance, can be described
by [9]:
VOC =kT
qln (NA,CIGSe + ∆n)∆n
n2
i,(2.2)
where NA,CIGSe is the p-type doping concentration of CIGSe, ∆nis the excess charge
carrier density due to the illumination, niis the intrinsic charge carrier density. This
expression shows, that the VOC depends on the doping concentration and the excess
electron density within the CIGSe, which is defined by the illumination intensity and the
electron lifetime. The value for the band gap defines the intrinsic charge carrier density.
VOC can also be expressed in the following way [9]:
VOC =Eg
q−kT
qln JphotoηJ(VOC)
j0exp(−Ea/lkT),(2.3)
The VOC loss relative to Eg/q is mainly determined by the activation energy Eafor
recombination. Eacan be lowered by defect states within the band gap or by a conduction
band cliff at the interface (see Sec. 3.1).
Experimental conditions: The J−Vcurves presented within this thesis were mea-
sured at 25
, controlled by water cooling, with a Keithley 238 source measure unit. The
optional illumination from a halogen lamp was set to a light intensity of 100 mW/cm2and
the spectrum was modified to fit the AM1.5G spectrum. The solar simulator class was B.
The intensity calibration was performed with a calibrated Si solar cell. The uncertainty
of the absolute light intensity is estimated to be ±5%. Additional the uncertainty of
the active area is around ±5%, leading to a total error of ±7%. The voltage sweep was
performed from positive to negative voltages at a speed of 250 mV/s. No light-soaking
prior to the measurements was performed.
2.4 Device Characterization 33
Figure 2.6: Dark (dashed line) and illuminated J−Vcurve of a CIGSe superstrate solar
cell. The resistances Rpand Rsare measured on the dark J−Vcurve around the voltage
range marked with the thick lines. The VOC, the JSC, the JMPP and the VMPP are taken
from the illuminated J−Vcurve. The fill factor FF and the power conversion efficiency ηare
calculated as shown in the graph.
2.4.2 C−Vmeasurements
The capacitance is a measure for the charge stored within the device, studying the voltage
and frequency dependence of the capacitance can therefore give information about the
spatial and energetic distribution of localized charges such as defect levels within a semi-
conductor. The capacitance can be derived from the phase shift of an applied alternative
voltage bias and the current induced by it. The current response is determined by the
complex conductivity of the device, which is called the admittance Y[79].
Y(ω) = G(ω) + iωC(ω),(2.4)
with G(ω) being the conductivity and C(ω) the capacitance, both in principle dependent
on the frequency ωof the applied voltage. If no charging or discharging occurs, the
admittance equals the conductivity and the phase shift becomes zero. If only charging
and discharging occurs, the admittance equals the capacitance and the phase shift becomes
90
°
. By measuring the amplitude and phase shift of the current it is possible to calculate
the capacitance, which is defined as following:
C(ω) = AdQ
dV,(2.5)
where dQis the additional charge stored in the device by a small change of the voltage
dV.Ais the area of the device.
Space charge capacitance
When applying a small AC voltage (20 mV) oscillating around 0 V to a perfect p/n-
junction, the conductivity G(ω) is zero due to the absence of shunts and the capacitance
34 2 Experimental: Deposition and Characterisation
originates from charging and discharging shallow defect states with free charge carriers at
the edges of the space charge region. The capacitance can be then described similar to a
parallel plate capacitor with an inserted dielectric [79].
C=0A
dSCR
,(2.6)
where 0is the vacuum permittivity, the dielectric constant of the charge depleted semi-
conductor and dSCR the width of the space charge region. This formula only holds for
homo-junctions or for n+/p- or n/p+-hetero-junctions, where the space charge region
width within one side of the junction can be neglected. The space charge width for such
a junction can be described by the following equation [79]:
dSCR =s20(Vbi −V)
eNCV
,(2.7)
where eis the elementary charge, Vbi the built-in voltage (defined by the difference of the
Fermi levels in the pand nmaterial), Vthe applied DC voltage and NCV the shallow defect
density at the edge of the space charge region. For homogeneously doped semi-conductors
it is possible to derive NCV by combining Eq. 2.5 and Eq. 2.6. For non-homogeneous semi-
conductors like CIGSe the doping profile has to be measured, which is shown in the next
Section.
Profiling
In non-homogeneously doped semi-conductors, the change of the space charge region
ddSCR due to a small voltage bias dVdepends on the local doping density NCV(x) at
the edge of the space charge region dSCR. With the assumption that only shallow defects
contribute to the capacitance, an equation for ddSCR can be derived by extending Eq. 2.7
with ddSCR/dSCR and integrating over dSCR [80]:
ddSCR =0dV
edSCRNCV(dSCR).(2.8)
A change in ddSCR leads to a change of the capacitance as defined in Eq. 2.5:
dC
dV=−A0
dSCR2
ddSCR
dV.(2.9)
Inserting Eq. 2.8 into Eq. 2.9 leads to the equation for the position dependent shallow
defect density NCV(x):
NCV(dSCR) = −C3
e0A2dC/dV.(2.10)
By choosing small changes of the applied voltage dV , in a way that dC is linear to dV , the
expression dC/dV can be exchanged with ∆C/∆V, which is more practical in application.
It has to kept in mind, that these equations are only valid for semi-conductors without
deep defects, which is not necessarily the case for CIGSe. In the next section the influence
of deep defects will be discussed.
2.4 Device Characterization 35
Deep defect capacitance
The capacitance Cdof a homogeneously distributed deep defect whose charging and dis-
charging is fast enough to follow the change of the applied voltage is described by the
following equation [81]:
Cd=e2
kT Ndf(Ed)(1 −f(Ed)),(2.11)
where Ndis the defect density, f(Ed) the Fermi function at the energy level of the defect, k
the Boltzmann constant and Tthe temperature. Thus, the capacitance is highest, when
f(Ed)=0.5, which is the case when the Fermi level crosses the defect level. Thus the
position of the defect charging and discharging process does not occur at the edge of the
space charge region, but somewhere within the space charge region. Eq. 2.10 is no longer
valid, because of two effects. First, the deep defects increase the capacitance compared
to the capacitance defined in Eq. 2.5. Secondly, they reduce dSCR if they are charged, i.e.
if f(Ed)>0. It is, however, possible to freeze out the influence of deep defects to the
capacitance, since the average charging and discharging time depends on the temperature
and on how deep the defect lies within the band gap. The cut-off frequency ω0defines the
transition when the defect charging cannot follow the change of the applied AC voltage
any more. It is given for a defect, which interacts with the valence band by [81]:
ω0= 2cpNVeEa/kT ,(2.12)
with cpbeing the capture coefficient, NVthe density of states of the valence band and
Eathe activation energy, which is the difference of the defect level and the valence band.
Thus for high activation energies or for low temperatures, defect charging can eventually
not follow the change of the applied voltage and does not contribute any more to the
capacitance. The study of the frequency dependence of the defect capacitance is called
admittance spectroscopy.
Influence of series resistance
The influence of the solar cell series resistance Rsand parallel resistance 1/G on the
measured capacitance Cmhas to be kept in mind. The real space charge capacitance
differs from the measured capacitance as follows [82]:
Cm=C
(1 + RsG)2+ω2C2R2
s
.(2.13)
For high Rsor high G= 1/Rpvalues the measured capacitance is lower than the real
capacitance. Therefore these two conditions Rs<< Rpand Rs<< 1/ω2C2should always
be met. For Rp=1000 Ω, Rs=1 Ω , C=500 nF, which are common values for the solar cells
studied in this thesis, the measured capacitance is Cm=0.1 Cat f=1 MHz and Cm=0.99 C
at f=1 kHz. Therefore, all measurements shown in this work were done at f=1 kHz.
36 2 Experimental: Deposition and Characterisation
Experimental conditions: The C−Vcurves presented within this thesis were mea-
sured at room temperature, with an Agilent 4284A precision LCR meter. The solar cells
were contacted with metal probes, whose influence was corrected for prior to each mea-
surement. The devices were relaxed in the dark for 5 minutes before each measurement.
2.4.3 External Quantum Efficiency (EQE)
The external quantum efficiency (EQE) is a measure for the efficiency of the conversion
of incoming photons to extracted charge carriers in a solar cell. Thus, from the EQE it
is possible to calculate the short circuit current by a convolution of the EQE with the
solar spectrum. However, the short circuit current of the solar cells studied in this work
exhibit a ”pathologic” dependence on the incoming light intensity and therefore leads to
different values for the short circuit current in EQE and J−V. The more interesting
application of the EQE measurement in this case is the possibility to approximate the
collection length for the minority carriers, Ln, from the EQE spectrum [83]. The following
equation describes the external quantum efficiency [8]:
Qext(λ) = hc
eλ
jphoto(λ)
E(λ),(2.14)
where hc/λ describes the energy of the photon with the wavelength λ,ethe elementary
electric charge, jphoto(λ) the photo-current at the photon wavelength λand E(λ) the
energy of the incoming photons at the wavelengths λ. The photo-current can be described
by jphoto(λ) = (qLnG0)/(1 + αLn) [84], with G0being the generation rate at the surface
and αthe absorption coefficient. From this it follows:
Qext(λ) = hc
eλ
1
E(λ)
qLnG0(λ)
1 + α(λ)Ln
.(2.15)
By lumping together all constants in the equation, the equation becomes:
λ
Qext(λ)=α(λ)−1+Ln·const., (2.16)
from which the collection length can be graphically obtained by plotting λ/Qext over α−1.
Interpolation of λ/Qext to zero gives Ln.
Experimental conditions: The EQE spectra were recorded at room temperature. The
light was provided by a halogen lamp whose wavelength band width was narrowed by a
Czerny–Turner monochromator to 10 nm. The short circuit current was amplified by a
Stanford Research 560 amplifier and recorded with the lock-in technique (SR830). A
calibrated Si and GaAs solar cell was measured prior to the device under test to quantify
the external quantum efficiency.
2.4 Device Characterization 37
Figure 2.7: Schematic of an EBIC experiment on a cross section of a CIGSe superstrate
solar cell. The resulting current profile is shown schematically as well. The collection efficiency
within the space charge region, which is marked with dashed lines, is set to unity.
2.4.4 Electron-Beam-Induced Current (EBIC)
Another technique to study the electron-hole pair extraction within a solar cell is the
Electron-Beam-Induced Current (EBIC) measurement. The basic principle of this tech-
nique is shown in Fig. 2.7. The electron beam generated from a scanning electron micro-
scope, is used to create electron hole pairs within the solar cell. The electron beam is then
scanned along the cross section of the solar cell, which is at short circuit condition, while
the generated current is recorded and plotted over the lateral location aof the electron
beam. From this a photo-current profile IEBIC, as shown in Fig. 2.7, is obtained which can
be further analysed. The obtained current profile is a convolution of the charge carrier
collection function fc(x) of the solar cell and the lateral generation function g(x) of the
electron beam [85]:
IEBIC(a)=
∞
0
g(x, a)fc(x)dx. (2.17)
The generation function gis described in [86] as follows:
g(x)= A
RG
β(|a−x|),(2.18)
where A=(EbIb(1−Λ))/(eEeh) stands for the generation rate, depending on the electron
beam energy Eb, the beam current Ib, the electron-hole pair generation Eeh and the back
scatter coefficient Λ. RG(ρ, Eb) stands for the Gruen penetration depth depending on
the material density ρand Eb. The collection function fccan be obtained by solving the
differential equation fC(x)=L2
n∆fc(x), with the boundary conditions fc(xSCR) = 1 and
38 2 Experimental: Deposition and Characterisation
fc(xC) = exp(−(SC/De)x)) [87], which leads to:
fc(x)dx =
1
Lncosh x−xc
Ln−SC
Desinh x−xc
Ln
Sc
Desinh xc−xSCR
Ln+1
Lncosh xc−xSCR
Ln,(2.19)
where xSCR is the position of the space charge region edge within the CIGSe, xcthe
position of the CIGSe-Au back contact, Scthe back contact recombination velocity of
electrons, and Lnthe electron diffusion length.
Experimental conditions: The EBIC signal was recorded within a Gemini LEO 1530
Scanning Electron Microscope. To access the cross-section of the device, the device under
test was broken at a predefined place of fracture, immediately before the measurement.
No polishing of the cross-section was performed. The contact wires were fixed onto the
sample with silver paint. The current was amplified by an SR560. At an extractor voltage
at 10 kV and an extractor current of 260
µ
m, a probe current of 63 pA was measured at
no high current mode. The detected current was amplified and returned to the image
processing unit of the SEM.
Care has to be taken not to reach the high in-injection level, which occurs when
the density of injected electrons exceeds the doping density of CIGSe, as this leads to
a charge accumulation and a distortion of the built-in potential and subsequent of the
EBIC profile [88].
2.5 Material Characterization
Material characterization allows the correlation of the electrical properties with the struc-
tural and chemical properties of a sample. This can be used to determine the origin of
device limitations and to optimize the fabrication routine accordingly. In this section the
following techniques will be introduces: X-ray Photoelectron Spectroscopy (XPS), X-ray
Diffraction (XRD) and Glow Discharge-Optical Emission Spectroscopy (GDOES).
2.5.1 X-ray Photoelectron Spectroscopy (XPS)
The chemical reaction between the TCO and the CIGSe layer is studied by cleaving the
sample at the respective interface with subsequent XPS surface analysis of both sides.
XPS can provide information about the elemental composition at the surfaces and Auger
analysis can additionally reveal the chemical state of the detected atoms at the interface.
This section will give a brief overview of the XPS technique and the sample preparation.
Further in depth information can be found in [89].
Basic principle
The electron binding energy, EB, in core levels depend on the specific element, the electron
orbital and the chemical environment of the probed atom. Therefore, it is possible to
2.5 Material Characterization 39
identify the type and the state of an atom by measuring the binding energy of these
electrons. To do this the core electrons are excited by X-ray photons of a known energy
hν, which is sufficient to eject the electrons from the atom. Excess energy provides kinetic
energy for the electron, Ekin, which can be measured with an electron detector. Using the
following equation one can infer EBfrom Ekin [90]:
Ekin =Ehν −EB−Φspec,(2.20)
with Φspec being the work function of the spectrometer. A sketch of this process is shown
in Fig. 2.8a. An example of a XPS spectrum in shown in Fig. 2.8d, where the detected
electron intensity is plotted over their binding energy. Peaks correspond either to core
levels of a specific element or from Auger processes (sketched in Fig. 2.8c). The energy
of the Auger electrons depend on the energy difference between the core hole and the
energy level of the electron that fills this core hole. Note that this Auger process has a
much higher cross section then the radiative process, shown in Fig. 2.8b. However, the
ejected electrons, both photo-excited or Auger, experience strong scattering within the
probed material. This limits the information depth to a few nano meters, depending on
the kinetic energy of the electrons.
The radiative recombination process shown in Fig. 2.8b is exploited in techniques like
X-ray fluorescence spectroscopy (XRF) to identify elements in the material bulk. XRF is
used in this thesis to calibrate the GDOES measurements (see Sec. 2.5.3)
Quantification
Each isolated peak in the XPS spectrum shown in Fig. 2.8d corresponds to one type of
atom. The intensity, which is the integral of the peak, is proportional to the amount of
atoms present within the sample, thus it is possible to quantitatively analyse the XPS
measurement. However, the error of such a quantification can become quite large due
to the fact that other factors have influence on the peak intensity. In general the peak
intensity IAX of an electron ejected from an energy level X within an Atom A can be
described as follows:
IAX =σAX(hν, Z, X)·D(EAX)·LAX(γ)·J0·CA·λM(EAX)cosθ ·T(x, y, γ, θ, EAX),
(2.21)
where σAX(hν, Z, X) is the cross section for the photo-ionization process, which depends
on the photon energy hν, the element Z and the electron orbital X. D(EAX) is the detection
efficiency of the detector, which depends on the electron energy. LAX(γ) is the asymmetry
factor of the electron emission, which depends on the angle of the excitation to the electron
orbital X. J0is the intensity of the X-ray source and CAis the density of the atom A in
the sample. λM(EAX)cosθ is the average free path length of the electron, which depends
on the electron energy and the excitation angle θ.T(x, y, γ, θ, EAX) is the transmission
function, which mainly depends on the kinetic energy of the free electron. All these values
can be measured or calculated, which allows the determination of CAfrom the intensity
40 2 Experimental: Deposition and Characterisation
Figure 2.8: Schematic representation of the processes occurring during a) photo-ionisation
b) recombination and emission of a photon and c) recombination and emission of an Auger
electron [92]. d) XPS survey scan taken from a CIGSe absorber.
IAX. However, all the correction factors in Eq. 2.20 come along with an uncertainty. First,
the sample itself is the origin of the uncertainty, that is due to inhomogeneities of the atom
density, both in depth and along the surface. Further, uncontrolled contamination from
air on the sample surface can be expected. Second, the properties of the measurement
setup is not always perfectly known. And thirdly, the calculation for the asymmetry
factor, the photo-ionization cross section and the free path length are subject to errors.
This leads to a minimum error of 52 % [91]. The error can be further minimized if peaks
of similar kinetic energy are compared, since most of the correction factors mainly depend
on the kinetic energy. In this work an average error of 60 % is assumed for all quantitative
XPS values.
Sample preparation and experimental conditions
To enable the XPS analysis of the TCO/CIGSe interface in superstrate solar cells, it is
necessary to cleave the sample at this interface. It was found that this can be achieved
by thermal shock. To do this, a few mm thick aluminium plate is glued with silver epoxy
onto the CIGSe surface. Care has to be taken, that the epoxy does not spread to the
sample edges, since this will hinder the cleaving. After drying the epoxy, the resulting
glass/TCO/CIGSe/Ag/Al stack is inserted into a beaker filled with liquid nitrogen. Due
to the rapid cool down and the different thermal expansion coefficient of these layers,
stress builds up and the sample cleaves at the weakest point, which is the TCO/CIGSe
interface. However, this only works if an interfacial oxide layer like GaOxforms at the
interface, otherwise the stack cleaves at the Ag/Al interface.
The beaker, still filled with liquid nitrogen, is then loaded into a glove box and only
inside a nitrogen filled glove box, the cleaved sample is removed from the liquid nitrogen.
It is assumed, that despite the nitrogen atmosphere, a 1 nm thick C2H4layer accumulated
on the sample surface. Both sides are then fixed with graphite tape onto the XPS sample
holder and loaded into the XPS system.
The XPS measurements were performed in the CISSY tool, which is a XPS system
2.5 Material Characterization 41
connected to a glove box and a sputter deposition chamber. The X-ray tube uses a
magnesium and an aluminium anode, which have characteristic emission lines at 1253,6 eV
and 1486,6 eV respectively. The angle between the incoming X-rays and the entrance slit
of the spectrometer is 54,7◦. More detailed information on the system specification can
be found in [93].
The resulting measurement data was analysed by using the multi-peak fitting routine
of the analysis software IGOR [94]. First, the background is subtracted with a linear
function, then the peaks are fitted with Voigt functions. For quantitative analysis the
results were corrected according to the ionization cross section and mean free path of the
electron lines as well as the electron analyzer characteristics (as described by Eq. 2.21)
and possible surface contamination (1 nm thick C2H4).
Care is to be taken since the relative error of the at.% values is at least 50%. Since we
only compare similar samples the error is systematic in character since it mainly originates
from uncertainties of the mean free path length of the x-rays, the ionisation cross section
and the transmission function. Therefore it is still possible to interpret trends between
the samples.
2.5.2 X-ray Diffraction (XRD)
In this work X-ray diffraction (XRD) is used to identify crystalline phases and preferred
orientation of poly-crystalline thin films.
The periodicity within a crystal can be measured by the positive interference of X-
rays scattered at different, parallel planes of electron densities. The condition for such a
positive interference is given by Braggs law [95]:
nλ= 2dsin θ, (2.22)
where λis the X-ray wavelength, θthe Bragg angle, dthe distance between the parallel
planes and n is called interference order. If λand θare known, it is possible to calculate
d/n. Commonly θis varied during the XRD measurement, which allows to scan for
different lattice spacings present in the studied crystal. The angle is generally varied
by a goniometer, whereby only planes perpendicular to the substrate normal lead to
interference peaks. In poly-crystalline films this allows to determine the preferred crystal
orientation by comparing the peak intensities within the obtained pattern to a powder
pattern. When performing energy dispersive XRD (ED-XRD), the Bragg angle θis fixed,
but the detected wavelengths λis varied [96]. Commonly polychromatic photons from
synchrotron radiation is used as the X-Ray source.
The XRD peak width is limited by the effects from the X-ray source, the detector area,
and any deficiency in the focus, but for small crystalline domain sizes it is often limited by
the domain size. If this is the case, the domain size can be calculated from β, which is the
FWHM value (in radians) of the peak after subtracting the instrumental line broadening,
42 2 Experimental: Deposition and Characterisation
with the Scherrer formula [97]:
D=0.9∗λ
βcos(θ).(2.23)
Experimental conditions: In this work the X-rays were generated at a Cu anode,
whose Cu-KαX-ray radiation has a characteristic wavelength λof 1.5418 ˚
A. The used
Diffractometer was a PANalytical XPert MPD with a PIXcel linear detector, set up in the
Bragg-Brentano configuration to record the XRD patterns. The Cu anode was supplied
with 40 kV and a current of 30 mA. A description of the experimental ED-XRD details
can be found in [98].
2.5.3 Glow Discharge-Optical Emission Spectroscopy (GDOES)
Depth profiles of the elemental composition of CIGSe/TCO thin film stacks are very
useful to identify phase transitions at the interface as well as analysing the Ga depth pro-
file within the CIGSe layer. There are several techniques available to study the elemental
depth profiles: Energy-Dispersive X-ray spectroscopy in a Transmission or Scanning Elec-
tron Microscope (SEM/TEM-EDX), Time-Of-Flight Secondary Ion-Mass Spectrometry
(TOF-SIMS), X-ray Photoelectron Spectroscopy (XPS) in combination with sputtering
or, as described here, Glow Discharge-Optical Emission Spectroscopy (GDOES). In this
work GDOES is used due to its better sensitivity compared to SEM-EDX or XPS and the
fact, that it is more convenient in application compared to TOF-SIMS or TEM-EDX.
The basic principle
The principle of GDOES is to sputter material from the sample’s surface and to identify
the material by their characteristic optical emission, induced by the electronic excitation
within the sputtering plasma.
The experimental setup is separated into the optical detector and the glow discharge
source for the sputtering process [99]. The idea is to sustain an Ar discharge current
between anode and cathode, in which the sample acts as the cathode. The resulting Ar
plasma close to the sample sputters material from the sample surface and this material
gets ionized and excited within the Ar plasma. The form of the resulting sputter crater
determines the depth resolution and has to be carefully optimized for each material by
varying the plasma conditions (pressure, voltage and current in pulsed mode). The optical
emission of the sputtered material is focused onto a concave grating and the monochro-
matic light is then detected by a CCD array. The analysed spectral range lies between
110 and 800 nm, sufficient for most elements including light elements as carbon, but not
oxygen.
2.5 Material Characterization 43
Calibration
To correlate the measured intensities Ii,k of the optical emissions to the chemical com-
position of the film it is necessary to calibrate the measurement. This is shown in detail
in [99] and shortly summarized here. The calibration is done by analysing a film with a
known chemical composition and thickness, which must be similar to the sample under
test. This allows determining a constant emission yield Ri,k and the sputtering rate qjfor
a certain element iwithin the sample jat the wavelength k. Both values depend on the
plasma conditions, detection setup and material composition of the sample. Therefore the
calibration has to be done prior to every measurement. The concentration of the element
within the sample is described by:
ci=Ii,k
qjRi,k
(2.24)
To calculate the depth of the sputter crater it is necessary to know the mass density ρof
the film, which can be approximated by a sum of the pure element densities ρiand from
this the depth dof the sputter crater is determined:
d=X
m
mqj∆t
ρA (2.25)
, with m being the numbers of optical emission spectra recorded during the glow discharge
and A the area of the sputter crater.
Example
An example of a CIGSe depth profile measured with GDOES is shown in Fig. 2.9a. The
uncalibrated optical emission signals at certain wavelengths, characteristic to a specific
element, are shown in Fig. 2.9a. From Fig. 2.9a it is obvious, that the emission yields
vary strongly between the elements, as Na is for example only a trace element, but its
optical emission signal is as high as for Cu. For the calibration a reference cell with
constant Ga content is used, whose composition and thickness was determined from X-
Ray Fluorescence (XRF) analysis. Trace elements like Na and C are not considered during
the calibration. Oxygen is not considered as its concentration is negligible and because
the signal for oxygen at 130 nm is too noisy and is not shown here. The calibrated results
are shown in Fig. 2.9b. The surface appears to be covered by a few nm thick Se layer,
which is an artefact due to the unstable sputter conditions during the first seconds of
the measurement. After stabilization of the experimental conditions, the Se and the Cu
concentration profiles are flat throughout the film. The In and Ga profiles add up to a
constant (In+Ga) profile and the Cu/(In+Ga) ratio is 0.9 throughout the film. At the
interface to the molybdenum layer the emission signals of the CIGSe elements decrease
approximately exponential. The exact form depends on the crater formation and the
surface roughness of the CIGSe and the Molybdenum film. In [99] it is mentioned that
the sputter rate for Cu is slower compared to that of In and Ga, thus leading to columns
44 2 Experimental: Deposition and Characterisation
(a) (b)
Figure 2.9: a) Uncalibrated optical emission profiles of different elements measured at their
characteristic wavelength. Inlet: Normalized profiles in order to compare the decay of the
different profiles at the interface to Mo. b) Atomic percentage profiles of the CIGSE elements
and Mo, calibrated with a reference cell. Inlet: Typical sputter crater at different sputter time
steps measured with a profilometer.
of pure copper within the sputter crater. To check whether this is the case for the sputter
conditions used in this work, the signals of all CIGSe elements are normalized to the
copper signal close to the interface. If the sputter rate of Cu would differ from the other
elements, the decay of the Cu signal at the interface to Mo should also differ from the
decay of the other signals. As seen in the inlet of 2.9a this is not the case. Since all
elements decay similar to each other, it can be concluded, that the sputter rates of all
elements are similar to each other.
Experimental conditions: The spectrometer used in this work is the GDA 650 from
Spectruma with 2.5-mm anode diameter and argon as a discharge gas (3.5 hPa process
pressure, votlage 500 V). Due to the low sample conductivity pulsed RF plasma mode was
used with a frequency of 350 Hz and a low duty cycle. The resuting crater is shown in
the inlet of Fig. 2.9b.
Chapter 3
Numerical Simulation
CIGSe thin film solar cells are difficult to describe with analytical models due to their
non-ideal characteristics caused by charged interface defects, deep bulk defects, non-ohmic
back contacts and band discontinuities. Therefore numerical simulations can be the better
choice to reproduce or to evaluate the experimental results [100]. These numerical models
calculate the charge distribution and the drift/diffusion current based on the poisson
equation and continuity equation for electron and holes. The most known commercial
program is Sentaurus TCAD [101], which is able to simulate the electrical performance
of three dimensional device structures and is also suitable for testing advanced optical
concepts. Other freeware programs like AFORS-HET [102], ASA [103], PC1D [104] and
SCAPS [105] exist, which are able to simulate one dimensional structures. In this work
SCAPS was used to do the numerical simulations, since it is capable to compute AC
signals for the simulation of capacitance spectroscopy [106] and is designed to handle
graded semiconductors [107] with large band discontinuities, which is necessary for CIGSe
devices. This chapter will give an introduction to the mathematical framework of the
numerical simulation programs.
Basic equations
The basic equations used in the numerical simulation are the Poisson equation and the
continuity equations. The Poisson equation describes the local electrostatic potential
Φ(−→
x , t) within the solar cell and for the one dimensional approximation of a solar cell
reads as follows [108]:
δ2Φ(x, t)
δx2=−q[n(x, t)−p(x, t)]
ε0εr
,(3.1)
where xstands for the position within the solar cell, tfor the time, qfor the elementary
charge, for the dielectric constant and n(x), p(x) for the electron and hole densities
respectively, including trapped charges within defects. Charged defects change the local
electrostatic potential and may improve or hinder the extraction of charge carriers. The
46 3 Numerical Simulation
Table 3.1: Baseline parameter used in this work for the numerical simulations, if not stated
otherwise. Charge carrier lifetimes and interface recombination velocities varied for most sim-
ulations and are not given here. The recombination velocity at the back contact was always
set to 1e+7 cm/s. Data taken from [100].
Symbol Property Unit CIGSe ZnO
Egband gap eV 1.18 3.3
χelectron affinity eV 4.5 4.6
NCCB effective density of states cm−32.2e+18 3.7e+18
NVVB effective density of states cm−31.8e+19 1.1e+19
νthnelectron thermal velocity cm/s 1e+7 1e+7
νthhhole thermal velocity cm/s 1e+7 1e+7
µnelectron mobility cm2/V s 1e+2 1e+2
µhhole mobility cm2/V s 2.5e+1 2.5e+1
rrelative permittivity (at 1kHz) 13.6 9
Rrad Radiative recombination coefficient cm3/s 1e-10 1e-10
Aabsorption coefficient prefactor 1/(cm√eV ) from file from file
continuity equations are [108]:
δn(x, t)
δt =Un(x, t) + 1
q
δJn(x,t)
δx ,(3.2)
δp(x, t)
δt =Up(x, t) + 1
q
δJp(x,t)
δx ,(3.3)
where Jn,p is the electron/hole current density and Un,p is the net generation rate of
electrons and holes, which is the difference between generation of free carriers and recom-
bination of free carriers. These equations ensure the conservation of energy. The drift
and diffusion currents can be calculated from the respective charge carrier densities and
potentials as follows [108]:
Jn=−qnµn
δφ
δx +qDn
δn
δx,(3.4)
Jp=−qpµp
δφ
δx +qDp
δp
δx,(3.5)
where µn,h stands for the charge carrier mobilities and Dn,h for the diffusion constants
of electrons and holes. At the boundaries of the device model, i.e. the metal contacts,
the potential values and the charge carriers densities are fixed. For a discrete position
xwithin the device, three non linear equations have to be solved, the poisson equation
and the two continuity equations. Since the 1 dimensional device structure is discretized
into N points, 3N equations have to be solved to get a solution for the full device. This
is usually done by starting with a first guess of the potential and the charge carrier
densities, then the potential is calculated from the poisson equation and corrected,
followed by the calculation of the charge carrier density from the continuity equations.
47
The whole procedure is repeated until convergence is achieved. The current can then be
calculated from the final potential and charge carrier distribution within the device.
Recombination in a p-type absorber via a single defect level is described by the
Shockley-Read-Hall recombination rate RSRH [109].
RSRH =(np −p2
i)
τpn+Ncexp ET−Ec
kBTL+τnp+NVexp EV−ET
kBTL,(3.6)
with pibeing the intrinsic hole density, τn,p the electron and hole lifetime, NVand NC
are the effective densities of states of the valence and conduction band, ETthe energy
level of the defect, ECthe energy at the conduction band minimum (CBM) and EVat
the valence band maximum (VBM). The exponential terms become large, and with it the
recombination rate low, when ETlies close to ECor EV, as this is the case for shallow
defects. It can further be shown [110], that for equal τnand τp,RSRH reaches a maximum
once n=pis fulfilled. In a p/n-junction this is fulfilled within the space charge region,
with the exact position depending on the doping densities and the voltage bias. Further,
low charge carrier lifetimes, τnor τp, increase the recombination rate. An estimate of the
single level electron lifetime τncan be calculated by [100]:
τn=1
σnνth,nNdef
,(3.7)
where Ndef is the defect density and σnthe effective cross-section of the defect for electrons.
The cross-section of ionized donor states is larger for electrons than for holes due to the
coulomb attraction. A simple approximation of the cross-section for a Coulomb-attractive
defect center is σn=q4/(16πε2
rk2
BT2) [111]. For single ionized donor states, it is in the
range of 10e-12 - 10e-13 cm2for electrons and for holes 10e-15 - 10e-16 cm2. Further
recombination pathways like radiative and Auger recombination are also implemented
into SCAPS but both effects only have a neglebile effect on the total recombination
current in the CIGSe devices studied here.
Interface recombination is handled in a very similar way to bulk recombination, by the
Pauwels-Vanhoutte theory [112], which is an extension of the Shockley-Read-Hall theory.
The difference is that the interface defect states can interact with the charge carriers from
both materials. Eq. 3.6, describing the SRH recombination rate, has to be extended to
describe the interaction between the defect state and two conduction band and two valence
bands. For example, due to the conduction band offset at the CIGSe/ZnO interface, the
interface defect interacts mainly with the holes from the VBM of CIGSe and with the
electrons from the CBM of ZnO. Further, the recombination rate becomes a recombination
velocity, since the defect density Ndef in Eq. 3.7 has the unit cm−2at the interface instead
of cm−3in the bulk.
Due to the different band gaps and electron affinities of the materials in hetero-junction
solar cells, charge carriers may have to overcome energetic barriers at interfaces. The
48 3 Numerical Simulation
charge carrier transport over a barrier ΦBis calculated with the formula for thermionic
emission [113]. The equation for the thermionic emission current of electrons, JTE,n, reads:
JTE,n =qνthNcexp ΦB
kBTLexp qV
kBTL−1,(3.8)
where TLis the lattice temperature and Vthe applied voltage bias. In addition, band
to band tunnelling, intra band tunnelling, tunnelling to interface defects and tunnelling
to contacts are all implemented into SCAPS, details can be found in [114].
The baseline parameters used for the SCAPS simulations in this thesis are shown in
Tab. 3.1.
3.1 Controlling interface recombination in CIGSe de-
vices
Lattice mismatch and elemental diffusion at a hetero-interface can lead to high interface
defect densities in hetero-junction solar cells. This may disqualify certain materials for
the application as a buffer layer in CIGSe devices. This section will present the impact
of interface recombination in CIGSe solar cells and how it can be reduced by device design.
As shown in Eq. 3.6, the bulk recombination rate at a single defect level depends on the
distance of the defect level ETto the CBM and the VBM. The higher the band gap of the
material the smaller the recombination rate at the defect, due to the reduced density of free
electrons and holes. The same holds for recombination at defects at the hetero-interface,
only that the band gap at the hetero-interface depends on the energetic position of VBM
and CBM of both materials. These depend on the electron affinity and the band gap of
the materials as described in Eq. 1.1. Thus, the buffer layer should have a lower electron
affinity but a higher band gap compared to CIGSe, in order not to reduce the interface
band gap. In the special case that i-ZnO is in direct contact to CIGSe, the conduction
band offset was shown to be slightly negative, which is called a ”cliff” like conduction band
offset. The ”cliff” like offset lowers the interface band gap and increases the band-to-band
recombination as well as the recombination via defect levels at the interface. An example
of a cliff like interface with a neutral mid-gap defect level (σn=σh=1e-13 cm2) is given in
Fig. 3.1a by the solid lines. In the following the parameters of the buffer layer are varied
to analyse the influence on the interface recombination losses. Since the open circuit
voltage, VOC (see Sec. 2.3), is the parameter which is most sensitive to recombination
losses (as long as no Fermi-level pinning is present) it will be used as a measure for the
recombination losses.
The simulated VOC for varying ∆EC,IF=χCIGSe-χbuffer is shown in Fig. 3.1b. In case,
that only band-to-band recombination occurs at the interface the CB offset can become as
small as -0.2 eV without experiencing any loss in VOC. For well passivated interfaces, with
a interface recombination velocity of SIF=1e+3 cm/s, already small negative offsets lead
3.1 Controlling interface recombination in CIGSe devices 49
(a) (b)
Figure 3.1: a) Band diagram of a CIGSe/buffer/ZnO solar cell at VOC condition. ∆EC,S
stands for the CBM gradient of the CIGSe close to the surface/interface, ∆EV,S for the
VBM gradient of the CIGSe close to the surface/interface, ∆EC,IF for the CBM offset at
the CIGSe/buffer interface. b) VOC depending on ∆EC,IF. No CBM or VBM gradient set
within the CIGSe, ∆EC,S= ∆EV,S=0. The doping level of the buffer is set 10x the value of
the CIGSe layer.
to a VOC loss. In case of high interface recombination velocities, with SIF=1e+6 cm/s,
a positive ∆EC,IF, also called a conduction band ”spike”, of at least +0.1 eV is required,
but the VOC is still reduced compared to the passivated interface.
To further reduce interface recombination losses it is required to increase the band gap
of CIGSe close to the hetero-junction. Fig. 3.2a shows the effect of increasing the CBM
of the CIGSe at the interface relative to the bulk value, which is called ∆EC,S in this
work. For the specific device model used for these calculations, it is possible to increase
the conduction band by up to 300 meV without introducing any loss in the short circuit
current JSC or the fill factor FF (not shown), while the VOC steadily increases due to the
reduced density of free electrons available for recombination at the interface. Still, the
effect is rather limited due to the unchanged density of holes within the CIGSe at the
interface. This requires a lowering of the CIGSe VBM.
The influence of lowering the VBM of the CIGSe at the interface compared to the bulk
value, ∆EV,S, is also shown in Fig. 3.2a. For a ∆EV,S of 150 meV present only at the first
10 nm from the interface is already sufficient to suppress most of the interface recombina-
tion losses. This can be achieved by anion substitution from selenium to sulphur, or by
a reduction of the Cu content on the surface (see Sec. 1.3.1). The width of the modified
CIGSe layer should be limited to the width of the space charge region and should not be
larger than 200 nm, otherwise it may create a barrier for the electron extraction, since the
p-type doping density is not reduced at the same time.
Another issue is the buffer doping density. Hetero-junctions from n-type and p-type
materials with similar doping densities, have equal electron and hole densities at the
50 3 Numerical Simulation
(a) (b)
Figure 3.2: a) VOC depending on the energy difference of the CBM and the VBM within
the bulk and the interface of the CIGSe (see Fig. 3.1a). The electron affinity of the buffer
is set 100 meV smaller than the electron affinity of the CIGSe bulk, ∆EC,IF=100 meV. SIF
was set to 1e+6 cm/s. b) VOC depending on the doping level of the 50 nm thick buffer
layer, with ∆EC,IF=100 meV. ∆EC,S=∆EV,S=0 was set for the red and the blue profile.
∆EC,S=∆EV,S=150 meV was set for the green profile. Interface recombination velocity was
set to SIF = 1e+6 cm/s.
interface, which according to Eq. 3.6 results in high recombination losses. To reduce the
hole concentration at the interface the type of majority charge carrier has to be inverted in
the CIGSe layer close to the hetero-interface, this is called an inverted junction. This can
be achieved by a high n-type doping density of the buffer layer. To reduce the interface
recombination it is therefore crucial that the doping density of the n-type buffer is larger
than the p-type doping of the CIGSe. The simulation results in Fig. 3.2b show how VOC
is influenced by the doping density of a 50 nm thick n-type buffer layer in the case of a
high interface recombination velocity (1e+6 cm/s) and no CIGSe band gap gradient close
to the hetero-interface. It shows, that the buffer doping should be at least 1e+18 cm−3.
The most trivial but at the same time most difficult way to reduce the interface
recombination losses is to reduce the amount of defect states. But even in the presence
of high defect densities the interface recombination losses can be almost completely
quenched, if the CIGSe interface band gap is increased by 300 meV and if the buffer
doping density is increased to above 1e+17 cm−3. The top profile in Fig. 3.2b shows this.
In summary, by engineering the material properties of the buffer layer and the CIGSe
at the hetero-interface it is possible to achieve high efficiencies even in the presence of
highly defective hetero-interfaces. This qualifies non-lattice matched materials as buffer
layers as long as their electron affinity is smaller compared to the bulk CIGSe and their
n-type doping density is higher compared the p-type doping density of CIGSe. However,
if the interface defects are acceptor type, they trap electrons and have a strong impact
on the band alignment, eventually leading to an electron extraction barrier. This will be
shown in the next section.
3.2 Acceptor states at the hetero-interface 51
(a) (b)
(c) (d)
Figure 3.3: Energy band diagrams at the hetero-interface region of a CIGSe/buffer/ZnO solar
cell, the voltage bias was set to +500 mV. The influence on the band alignment of acceptor
states NAat different energetic and spatial positions (green line) is shown in the graphs b), c)
and d).
3.2 Acceptor states at the hetero-interface
As shown earlier the quality of the CIGSe/buffer interface determines interface recombi-
nation losses, but also the bulk recombination can be influenced by the interface quality.
An electron barrier at the hetero-interface induced by a conduction band spike or due to
acceptor states at the interface lead to a slower electron extraction and hence a possible
higher bulk recombination loss. In this section both of these factors are studied exemplary
on a bulk limited CIGSe device model. The results presented here will become important
for the discussion of the device properties in Sec. 5.2 and Sec. 8.3. It should be noted
that interfacial donor defects cannot induce electron barriers in CIGSe devices and are
therefore not considered in this section.
Within the applied model, the electron diffusion length is set to a relatively low value
of 500 nm. No band gap gradings are set and the electron affinity of the buffer χbuffer was
52 3 Numerical Simulation
chosen to be the same as of the CIGSe layer (in Fig. 3.3 a CBM spike is shown only to
visualize the interface to the buffer layer). The cross section of the acceptor states were
chosen to be small in order to limit recombination induced by them.
The influence of the spatial position of the acceptor states on the energy band diagram
is shown in Fig. 3.3. The diagram shows the situation at a forward bias of +500 mV, close
to the VOC.χbuffer is chosen to induce a small spike for better illustration of the effects,
for the simulations shown in Fig. 3.4 it was set to the CIGSe value.
For the situation that no acceptor states are present at the interface the space charge
region expands into the CIGSe absorber and the inversion at the interface is strongly
pronounced as seen in Fig. 3.3a. The presence of a high density of interfacial acceptor
states removes the inversion within the CIGSe layer completely, independent of whether
the states are located at one of the buffer interfaces or in the buffer bulk, Fig. 3.3b-d.
This can have several effects. As shown earlier, the interface recombination losses
will increase, but also the electron collection will be reduced due to the missing space
charge region within the CIGSe. If the acceptor density is sufficiently high, the electron
quasi-Fermi level drops down towards the acceptor level. This leads to an increase of the
CBM at the position where the acceptor states are located and therefore to an electron
barrier. Such a barrier can reduce the extraction efficiency and increase the bulk and
interface recombination. In the following it will be shown how acceptor states at three
different locations effect the device performance.
Acceptor states at the interface between CIGSe and the buffer layer can lead to a
pinning of the Fermi level (in the dark) at the interface. The consequence of this is, that
the width of the space charge region within the CIGSe does not vary with the applied
voltage bias and is determined by the energetic position of the acceptor state. Fig. 3.3b
shows an exemplary band diagram at a voltage bias of +500 mV for an acceptor state
present 300 meV above the CIGSe VBM. It is seen that dSCR within the CIGSe becomes
almost zero. Interface states energetically located higher than 300 meV above the CIGSe
VBM would increase dSCR, whereas states closer to the valence band would further reduce
dSCR and possibly introduce a barrier. Fig. 3.4a shows the effect on the current collection.
Due to the reduced space charge region width the photo-current becomes smaller. The
open circuit voltage increases since the electron density within the CIGSe is strongly
reduced. A kink in the J−Vcurve can develop if a conduction band spike is present
or if the acceptor state is closer to the CIGSe valence band. The density of the defect
states have a strong influence on the measured capacitance as can be seen in Fig. 3.4b. A
characteristic of the Fermi level pinning is the flat profile of the capacitance profile.
A high concentration of acceptor states located within the buffer layer, lifts up the
conduction band relative to the electron quasi-Fermi level within the buffer. An electron
barrier develops with a similar energetic height for both, the electron extraction (photo-
current) and the electron injection (diode current) into the CIGSe. Thus, the J−Vcurve
in Fig. 3.4c exhibits a characteristic s-shape around the VOC. At higher voltage biases, the
series resistance does not increase. The C−Vcurve in Fig. 3.4d shows a strong increase of
the capacitance at forward bias, since at this condition the space charge region is confined
3.2 Acceptor states at the hetero-interface 53
(a) (b)
(c) (d)
(e) (f)
Figure 3.4: Simulated J−Vand C−V(1 kHz) curves at 293 K for different positions
and densities of acceptor states at the hetero-interfae of a CIGSe/buffer/ZnO solar cell. The
electron affinity of the buffer and of CIGSe were set to the same value. a)-b) Acceptor states
between CIGSe and buffer. c)-d) Acceptor states within buffer. e)-f) Acceptor states between
buffer and ZnO. Interface recombination velocity was set to zero.
54 3 Numerical Simulation
to the buffer layer and the space charge capacitance increases with 1/dSCR, see Eq. 2.6.
In case a high density of acceptor states is located at the interface between the buffer
layer and the TCO, the space charge region at forward bias is limited to the TCO. This
induces an electron barrier for electron extraction and injection, with the difference to
the previous case, that the barrier for electron injection is higher than for extraction.
This leads to a disturbed s-shape of the J−Vcurve as shown in Fig. 3.4e. The onset of
the current injection is pushed to higher voltages compared to the situation when the
acceptor states are located in the buffer layer. The capacitance is slightly smaller but
both situations lead to similar capacitance profiles.
Another cause for an electron barrier at the hetero-interface between CIGSe and the
buffer layer is a large conduction band spike, e.g. when χbuffer is 500 mV smaller compared
to χCIGSe. The effect on the J−Vcurve of such a barrier is shown in Fig. 3.5a. The
effect is similar to an increases series resistance, as the barrier height is comparable for
electron extraction as for injection. The barrier has no influence on the VOC since the
Fermi level is less influenced by the conduction band alignment. Further it only has a
negligible influence on the measured capacitance (Fig. 3.5b).
In the presence of a high density of acceptor states within the buffer layer, it was shown
above that a barrier can be present already for χbuffer =χCIGSe. As shown in Fig. 3.6a this
barrier increases strongly for a slightly reduced χbuffer. On the other hand, the barrier can
be reduced by increasing χbuffer above χCIGSe. Thus, in the case of high acceptor densities
at the interface, a slight cliff may be beneficial for the device performance.
In case of a high acceptor density within the buffer layer and a high interface
recombination velocity at the CIGSe-buffer interface, the negative effect of a conduction
band cliff sets in, as described in the previous section. The VOC decreases linear with
the decrease of χbuffer (Fig. 3.6b).
In summary, the shape of the J−Vand C−Vcurves are good indicators for the
existence and the location of interfacial acceptor states. It allows to differentiate between
interfacial electron extraction barriers induced by a spike in the CBM or by charged
acceptor states. This will become important during the study of the CIGSe superstrate
solar cells in the next Chapter.
3.2 Acceptor states at the hetero-interface 55
(a) (b)
Figure 3.5: Simulated J−Vand C−V(1 kHz) curves at 293 K. Variation of the buffer
electron affinity in the absence of interfacial acceptor states.
(a) (b)
Figure 3.6: Simulated J−Vcurves at 293 K. Variation of the buffer electron affinity in the
presence of interfacial acceptor states and a) low interface recombination velocity or b) high
interface recombination velocity.
Chapter 4
TCO evaluation
The basic requirements on a TCO used as the window layer in a CIGSe superstrate
device is first of all good optical transparency between 1.2 eV (CIGSe band gap) and 3 eV
of photon energy combined with a low sheet resistance, <10
W
/. Further, if the TCO is
forming the p/n-junction with CIGSe, the TCO electron affinity is important to reduce the
interface recombination losses (Sec. 3.1). However, it is difficult to predict which TCO
will be best suited, since little is known about the formation of the inter-facial phases
(Sec. 1.2) and their influence on the device performance [115]. Therefore this chapter
will compare the performance of different TCOs in CIGSe superstrate devices. In depth
interface analysis will be done for the most promising TCO in the subsequent chapter.
4.1 FTO, ITO and ZnO
The three most commonly used TCOs in photovoltaics are In2O3:Sn (ITO), SnO2:F (FTO)
and ZnO:Al (AZO). This section will compare their performance as the window layer in
CIGSe superstrate solar cells without buffer layer. The AZO was capped with a 100 nm
thick layer of i-ZnO, which therefore forms the interface with the CIGSe. All TCOs were
deposited on alkali containing glass substrates. The sheet resistance of the three TCOs,
measured before and after the CIGSe deposition, are listed in Tab. 4.1, together with the
VOC and the ηof the resulting solar cells.
Fig. 4.1a shows the J−Vcurve for the three TCOs upon which a CIGSe layer was
deposited at 560
via the modified three stage process as described in Sec. 2.1. The
most striking difference between the device performances is the difference in VOC. The
VOC of the ZnO device is 400 mV, the VOC of the ITO device 120 mV and the VOC of the
FTO device is 0 mV. The FTO device is not shunted though, as a rectifying behaviour at
negative voltage biases can be observed. The ITO device shows the largest short circuit
current, 28 mA/cm2and the lowest series resistance, 1.7
W
cm2, compared to the ZnO
device with 23 mA/cm2and 3.3
W
cm2. Still, ηof the ZnO device is the highest with
2.4 %, followed by the ITO device with 1 %. The sheet resistance of the TCOs after the
process were almost unchanged as shown in Tab. 4.1. Even for CIGSe depositions at lower
temperatures the highest device efficiencies were achieved by using ZnO as the window
58 4 TCO evaluation
(a) (b)
Figure 4.1: a) J−Vcurves of Au/CIGSe/TCO superstrate devices with In2O3:Sn (ITO),
SnO2:F (FTO) and i-ZnO/ZnO:Al (ZnO) as the TCOs. CIGSe deposited at 560
. In-
set: CBM and VBM levels of the different CIGSe/TCO stack. b) J−Vcurves of
Au/CIGSe/Zn(O,S)/ZnO:Al devices with varying sulphur content in the Zn(O,S) layer. CIGSe
deposited at 520
.
layer (not shown here).
Discussion The significant difference in the VOC of the devices with the different TCOs
is likely to originate from the different interface formation of each TCO with the CIGSe
layer. In [116] the formation of Ga2O3was observed for CIGSe growth at 550
on ZnO
and on In2O3, but not for SnO2. This may be a reason for the increased VOC of ITO and
ZnO. Further, in [117] the electron affinities χof these oxides are compared, where χfor
ZnO is listed as 4.5 eV, very similar to the one of CIGSe, 4.7 eV for ITO and 4.9 eV for
FTO. Thus, a conduction band cliff forms at the CIGSe interface to ITO and FTO. This
is shown schematically in the SCAPS band diagram of Fig. 4.1a. According to Fig. 3.1
in Sec. 3.1 a cliff like conduction band alignment leads to a strongly reduced VOC in the
presence of interface defects. For high interface recombination velocities, the drop in
VOC was shown to be equivalent to ∆EC,IF/q at the TCO/CIGSe interface. For FTO
the difference in VOC compared ZnO is 400 mV, which is exactly the difference in their
electron affinity. For ITO the difference in VOC is 280 mV compared to 200 meV difference
of their electron affinity. Thus, despite the different chemical interface formation the VOC
Table 4.1: Comparison of the standard TCOs and their performance in CIGSe superstrate
devices. Note that χZnO of 4.6 eV is used for the device simulations in this work.
TCO Rsq,before [Ω/]Rsq,after [Ω/]χ[eV] VOC [mV] η[%]
FTO 8 9 4.9 0 0
ITO 6 6 4.7 120 1
i-ZnO/ZnO:Al 6 6 4.5 400 2.4
4.2 Zn(O,S) 59
correlates well with the electron affinities of the TCOs, as expected in case of a high
interface recombination velocity (as shown in Fig. 3.1).
The higher series resistance of the ZnO device compared to the other two devices
is seen in Tab. 4.1 does not to originate from the degradation of the ZnO conductivity.
The formation of Ga2O3, as observed in the literature, may lead to the increased series
resistance.
Despite the poor efficiencies of all the devices with different TCOs, ZnO seems to be
the TCO with the highest potential for superstrate devices as it reproducibly lead to the
highest open circuit voltages and power conversion efficiencies. In Sec. 5 the interface
formation between ZnO and CIGSe with varying Ga content will be studied in order to
understand and optimize the system.
4.2 Zn(O,S)
In CIGSe substrate solar cells typically Zn(O,S) is used to replace the CdS buffer layer
with an oxide material. The advantage of Zn(O,S) over ZnO is the reduced electron
affinity, turning the conduction band cliff into a conduction band spike and reducing the
interface recombination. Good results for substrate devices were obtained for different
deposition methods, like sputtering [118], chemical bath deposition [119] and ALD [120].
In superstrate configuration it hasn’t been tested so far. Here, Zn(O,S) with different
O/S ratios was deposited via ALD onto i-ZnO/ZnO:Al substrates and used as the buffer
layer for superstrate CIGSe solar cells. Experimental details of the Zn(O,S) deposition
can be found in [121]. The Zn(O,S) layer was deposited by an alternating sequence of
ZnO and ZnS layers. The O/S ratio corresponds to the number of ZnO layers deposited
before one ZnS layer was deposited. The ratio 12:1 corresponds to a S/(S+O) in the
layer of approximately 0.16 and 6:1 to approximately 0.3. According to [122], this leads
approximately to an electron affinity reduction of 150 meV for the 12:1 ratio and 300 meV
for the 6:1 ratio. The CIGSe absorber was deposited at 520
via the modified three stage
process.
The resulting J−Vcurves are shown in Fig. 4.1b. The J−Vcurves of the i-ZnO
device exhibit a relatively poor fill factor but high open circuit voltage of 610 mV and a
short circuit current of 32 mA/cm2. The devices with Zn(O,S) exhibit a strong s-shape
character. The 12:1 Zn(O,S) layer shifts the photo-current decay and the injection current
onset by around 250 mV. The 6:1 Zn(O,S) layer shifts the photo-current decay by around
600 mV.
Discussion According to Sec. 3.2 the s-shape character of the Zn(O,S) devices originate
from a strong electron barrier at the hetero-interface. This is surprising, because the
electron affinity of the Zn(O,S) devices are approximately 4.45 eV and 4.3 eV, far below
4.1 eV which would be required to lead to such J−Vcurves (compare with Fig. 3.6a).
But in Fig. 3.6c it was shown, that such a s-shape profile can occur for lower electron
affinities, if a high density of acceptor states is present at the interface. In such a case,
60 4 TCO evaluation
(a)
std. Rsq neµn
ZnO
W
/sq cm−3cm2(V s)−1
as-dep. 8.0 4e+20 45
annealed 5.3 3.7e+20 60
(b)
Figure 4.2: a) Transmission curves, corrected for the reflectance, of standard as-deposited
ZnO:Al and of ZnO:Al annealed during the CIGSe deposition process at 525
. The corrected
transmission of the cap annealed ZnO:Al is shown for comparison. b) Sheet resistance of the
standard ZnO samples measured with a four point probe. Free charge carrier density and
electron mobility extracted from T and R spectra using a fitting procedure based on a Drude
approach [123]. It should be noted that the improved properties are only observed on sodium
free glass substrates.
it is beneficial to have a slight conduction band cliff at the hetero-interface, which is the
case for i-ZnO.
4.3 ZnO annealing
With ZnO being the best performing TCO, it is interesting to study the effect of annealing.
It is known that the opto-electronic properties of ZnO can be improved by annealing
the ZnO under conditions prohibiting oxygen loss or chemi-sorption of oxygen. A very
successful method was found to be annealing in nitrogen while capping the ZnO with
SiOx[17]. The application of such ZnO is attractive for CIGSe superstrate solar cells,
as the optical losses can be strongly reduced. A similar effect is observed in this work
simply due to the high-temperature CIGSe deposition onto the ZnO as shown in Fig. 4.2.
The CIGSe/ZnO stack was cleaved after the process, which allowed to measure the sheet
resistance and the optical transmission. It can be seen that both properties improve due
to the annealing which occurred during CIGSe deposition. The average light absorption
between 400 nm and 1100 nm is reduced from 10.7 % to 6.5 % which yields a photo-current
increase of 1.5 mA/cm2. At the same time does the sheet resistance decrease from 8
W
/sq
to 5.3
W
/sq, which allows a thickness reduction of the ZnO:Al layer by 30 %.
It should be noted, that this is only observable when the sodium diffusion from the glass
is blocked. A more detailed study of the ZnO annealing can be found in the appendix
in Sec. 10.3.1. It is shown, that the increased mobility can be measured via the Hall
technique and via the free carrier absorption. This proves that the improved properties
4.4 Summary: TCO evaluation 61
are induced by a bulk and not a grain boundary effect. It is assumed that the density of
bulk defects, induced by the dopants, are reduced.
4.4 Summary: TCO evaluation
1. Evaluation: Different oxides, with different electron affinities, were tested as the
window layer within the buffer-free superstrate solar cells. The highest efficiency
was achieved by using a i-ZnO/ZnO:Al double layer. The oxides with higher electron
affinity than ZnO lead to low VOC values due to high interface recombination. The
oxides with lower electron affinity lead to a strong electron barrier at the hetero-
interface, presumably due to the combination of a high acceptor defect density and
an increased electron affinity. Thus, even though ZnO leads to a slight conduc-
tion band cliff (50-100 meV) it leads reproducibly to the highest power conversion
efficiencies. Therefore the chemical and electronic properties of the CIGSe/ZnO
interface will be studied in the following sections.
2. ZnO annealing: The CIGSe deposition onto the ZnO layer was shown to improve
the electron mobility and the transparency of the ZnO, provided that no sodium
diffuses from the glass into the ZnO.
Chapter 5
ZnO/CIGSe device and interface
analysis
The p/n-heterojunction is the most important and the most sensitive part of a CIGSe solar
cell and in the superstrate configuration it is at the same time the least controllable inter-
face. As it was shown in Sec. 1.2, phase formations and diffusion processes are expected
to occur at the interface between CIGSe and ZnO. Conduction band misalignment and
lattice mismatch possibly lead to high interface recombination losses as shown in Sec. 3.1.
Therefore the first section studies the interface formation in detail. The next section will
correlate the solar cell performance to the different chemical compositions at the interface
and find the optimum process condition. Finally, a comparison with ZnO/CIGSe devices
in substrate configuration will be given to find a general conclusion for the ZnO/CIGSe
interface. In the following the expression CIGSe/ZnO will be used as an abbreviation for
ZnO:Al/i-ZnO/CIGSe.
5.1 Interface formation
This section studies the chemical interface formation of different chalcopyrites with ZnO.
First the experimental results will be described from Sec. 5.1.1 to Sec. 5.1.5, while the
discussion of all results takes place in Sec. 5.1.6.
5.1.1 CGSe/ZnO interface formation
A CGSe thin film was deposited via the 3-stage-process (see Sec. 2.1) on top of the i-
ZnO/AZO double layer coated with a 10 nm NaF precursor. The deposition temperature
was 330
in the first stage and 520
in the second and third stage. To study the
interface reaction between the CGSe and the ZnO layer, an SEM image and a GDOES
depth profile were recorded from the CGSe/ZnO sample (Fig. 5.1 and 5.2a).
The SEM image shows poly-crystalline growth of the CGSe layer on top of the ZnO,
with no correlation of their morphology. No inter-facial layer from the reaction of CGSe
with ZnO layer can be immediately identified. The GDOES elemental depth profile in
64 5 ZnO/CIGSe device and interface analysis
Figure 5.1: SEM image (InLens detector) of CGSe grown on ZnO at 520
. The top 100 nm
of the ZnO stack appears brighter in the SEM picture, when recorded with the InLens detector,
due to the lower doping density of the i-ZnO compared to the darker appearing AZO.
(a) (b)
Figure 5.2: a) GDOES elemental depth profiles of the CGSe/ZnO stack. The dashed lines
mark the approximate height of the sputter crater profile, which leads to the misleading ap-
pearance of an intermixed interface. It is around 370 nm. b) Depth profile of the Ga/Se and
Cu/Se ratios from the same sample. The increased Ga/Se ratio at the interface indicate that
Ga is partially bound to O instead of Se.
5.1 Interface formation 65
Fig. 5.2a shows the hetero-interface region. The edges of the hetero-interface are marked
with two dashed lines. The distance between the hetero-interface edges is defined by
the depth/homogeneity of the sputter crater profile (see inlet of Fig. 2.9b), the surface
roughness and the thickness of a potential interfacial layer. At the hetero-interface the
Cu, Ga and Se signal decrease similar to the elemental profiles shown in Sec. 2.5.3 and as
expected from a typical sputtering crater profile. The GDOES setup is not able to detect
oxygen in order to identify interfacial oxide layers, however, the profile of the Ga/Se ratio,
shown in Fig. 5.2b, is expected to rise once the Ga is partially bound to O instead of Se.
And indeed, within the interface region, the Ga/Se ratio increases. This increase could
also originate from the formation of a Ga2Se3phase, but this should lead to a simultaneous
change of the Cu/Se ratio, which cannot be observed. Therefore the Ga atoms at the ZnO
interface have to be bound partially to oxygen instead of selenium, forming an GaOxrich
layer. This is in accordance with the thermo-dynamical considerations in Section 1.2.
Further, the constant Cu/Se ratio within the interfacial region, indicates that the Cu
atoms around the interface are bound within the CGSe matrix. This shows, that the
composition close to the interface is not Cu depleted compared to the bulk, and that no
formation of CuOxoccurs at the interface.
In summary, only a thin layer rich in GaOxforms at the interface, which can be
identified from the increased Ga/Se GDOES signal ratio within the interfacial region, but
cannot be identified in the SEM picture.
5.1.2 CISe/ZnO interface formation
Similar to the CGSe/ZnO sample, CISe was deposited on top of the i-ZnO/AZO double
layer coated with a 10 nm NaF precursor at a substrate temperature of 550
. An SEM
image of the stack is shown in Fig. 5.3. The CISe layer grows poly-crystalline on top of
the ZnO, with no morphologic correlation to the ZnO layer. At the interface to the ZnO
a thin layer of around 30 nm thickness can be seen. The GDOES elementary depth profile
of the region around the CISe/ZnO interface is shown in Fig. 5.4a. The interface region
is marked with two dashed lines. At the interface, a strong increase of the In signal (in
Fig. 5.4a and of the In/Se signal (in Fig. 5.4b) can be observed. Similar to the CGSe/ZnO
case the formation of In2Se3at the interface can be excluded, since the Cu/Se ratio does
not decrease simultaneously. Thus, the In atoms close to the ZnO interface have to be
partially bound to oxygen instead of selenium, forming an InOxrich layer. Again, this
is in accordance to the thermo-dynamical considerations. The constant ratio of Cu/Se
before it drops into the noise level indicates that the Cu atoms close to the interface are
bound within the CISe matrix and that no formation of CuOxoccurs at the interface.
Compared to the CGSe/ZnO system, the interfacial oxide layer in the CISe/ZnO sys-
tem is thicker and can be seen in the SEM image and from the higher In/Se ratio compared
to the Ga/Se ratio in the GDOES depth profiles. In addition, the elemental GDOES depth
profiles differ in shape at the hetero-interface. At the CGSe/ZnO interface (Fig. 5.2a) the
Se and Cu signal decrease sharply followed by a longer tail, as expected for sharp inter-
faces (compare with Sec. 2.5.3). In the CISe/ZnO system the profiles do not show this
66 5 ZnO/CIGSe device and interface analysis
Figure 5.3: SEM image (SE detector) of CISe grown on ZnO at 520 . The magnified area
shows the InOxlayer at the CISe/ZnO interface.
(a) (b)
Figure 5.4: a) GDOES elemental depth profiles of the CISe/ZnO stack. The dashed lines
mark the approximate height of the sputter crater profile plus the InOxthickness, which is
around 470 nm. b) Depth profile of the In/Se and Cu/Se ratios from the same sample. The
increased In/Se ratio indicate the InOxlayer at the interface.
sharp signal change at the interface. Surface roughness and the sputter conditions were
similar for the CISe and CGSe samples and cannot explain this effect. This indicates,
that the interfacial InOxlayer is not a sharp interface but that it is heavily intermixed
with Se, Cu and Zn. Furthermore, the In/Se profile increases over the interfacial region
until it reaches the maximum value at the ZnO side, showing that the reaction front is at
the interface to ZnO, as expected from the thermo-dynamical calculations.
In the following this interfacial oxide layer will be named InOxfor simplicity.
5.1 Interface formation 67
Figure 5.5: SEM image (SE detector) of CIGSe grown on ZnO at 525 , with Au on top as
the back contact.
5.1.3 CIGSe/ZnO interface formation
In CIGSe/ZnO stacks the question arises whether the formation of InOxor GaOxdomi-
nates, or whether an (Iny,Ga1−y)Oxalloy forms. Fig. 5.5 shows an SEM image of a sample
fabricated with a modified three stage process at T = 525 without sodium addition,
as described in Sec. 2.1. The CIGSe layer grows poly-crystalline with no morphologic
correlation to the ZnO layer. No inter-facial layer can be directly identified. To study the
interface formation between CIGSe and ZnO TEM images were taken and XPS spectra
of cleaved samples were recorded.
TEM analysis Fig. 5.6a shows a HR-TEM image of the hetero-junction region. In the
region between the CIGSe and the ZnO an approximately 6 nm thick layer can be iden-
tified. Fig. 5.6b shows an interface image taken at a different, thinner sample area. The
interfacial layer is marked in red within this image. The layer thickness varies between
4 and 6 nm, thus around +/-20%. To identify the crystalline structure of the interfa-
cial layer a fourier transform (FFT) of this image was calculated at the interface, the
ZnO and the CIGSe area. The FFT images of the ZnO and the CIGSe areas show dis-
tinct peaks originating from the periodicity of the lattice planes, with dZnO=0.27 nm and
dCIGSe=0.32 nm. The FFT of the interface layer does not show any peaks characteristic
for a crystalline structure, which leads to the conclusion that the layer has an amorphous
structure. The chemical composition of the interface layer is obtained by a TEM-EDX
line scan over the interface at a thicker part of the TEM sample. The EDX profile is
shown in Fig. 5.7a. Clearly visible from this measurement is that the amorphous layer
consists of a segregation of Ga. No segregation of In or Cu is visible. The O profile is
shifted 4 nm from the Zn profile towards the CIGSe layer, indicating that the Ga is bound
to oxygen within a few nanometer thick layer. The FWHM of the Ga EDX signal peak
is 22 nm, which is broader than the width of the actual interfacial layer, since it is the
projection of a rough interface.
68 5 ZnO/CIGSe device and interface analysis
(a) (b)
(c)
Figure 5.6: a) HR-TEM image of the CIGSe/ZnO interface. The CIGSe is grown at 525
without additional NaF supply. b) HR-TEM image of the same CIGSe/ZnO interface recorded
at a thinner sample area used for EDX analysis. The interface layer between CIGSe and
ZnO is coloured in red. c) From left to right: FFT of the ZnO area, peak corresponding to
2.7 ˚
A(spacing of the (002) lattice planes). FFT of the interface area, no peaks visible. FFT of
the CIGSe area, peak corresponding to 3.5 ˚
A(spacing of the (112) lattice planes). The same
amount of pixels were used for all three FFTs.
(a) (b)
Figure 5.7: a) TEM-EDX line profile showing the elemental depth profiles at the CIGSe/ZnO
interface b) TEM-EDX map of the same interface region. The GaOxlayer is clearly visible
due to the increased Ga Kline signal at the GaOxlayer.
5.1 Interface formation 69
A two-dimensional EDX map of the Ga Signal from the same interface region is also
shown in Fig. 5.7b. The Ga-rich layer observed in the one-dimensional plot can be observed
along the hetero-interface. It covers the whole interface relatively homogeneous, with a
variation of the FWHM of the Ga EDX signal peak of around +/-20 % along the interface,
similar to the variation of the GaOxthickness observed in Fig. 5.6b.
XPS analysis Further insight into the elemental composition of the interfacial layer
was obtained by quantitative XPS measurements. To analyse the interface, the sample
was cleaved by thermal shock at the interface between CIGSe and ZnO, as described in
Sec. 2.5.2. The layer remaining on the glass substrate was the ZnO layer and the layer
attached to the silver epoxy the CIGSe layer. As seen in Tab. 5.1 the surface of the
ZnO layer contains all elements of ZnO and CIGSe. The quantitative analysis (details
in Sec. 2.5.2) shows that it mainly consists of Ga, O and Se with impurities of Cu, In
and Zn in the range of a few atomic percent. No sodium was detected at the interface.
The information depth of the Ga and Zn XPS signal is around 2 nm. Assuming that the
GaOxlayer is free of holes, the Zn signal does not originate from the ZnO underneath the
interfacial layer. The concentration of Zn within the interfacial layer may increase closer
to the ZnO layer though. The information depth varies for the different elements between
1.6 nm and 6.1 nm and the error of the quantitative results are around 60 %, thus the
chemical composition can only be approximated to (Iny,Ga1−y)(Oz,Se1−z)x:Zn,Cu , with
y and z close to 1. In the following the interfacial layer will be named GaOxwhenever the
In concentration is below 5 at.%, else it will be named (Iny,Ga1−y)Oxor InOxfor y=1.
The surface of the CIGSe layer consists mainly of Cu poor CIGSe with around 1 at.%
Zn and O contamination. The low Cu-content cannot be confirmed by the TEM-EDX
depth profile in Fig. 5.7a.
Table 5.1: Quantitative results from XPS measurements shown in atomic % of all elements
within a sample. The sample was cleaved at the CIGSe/ZnO interface. Care is to be taken
since the relative error of the atomic concentrations is between 50% and 60%.
inf. depth in ZnO ZnO surface inf. depth in CIGSe CIGSe surface
nm at.% nm at.%
Cu 2.4 3 3.1 13
In 4.6 5 5.8 25
Ga 1.6 44 2.1 6
Se 6.1 17 7.8 54
O 4.2 32 4.1 1
Zn 2.0 1 2.6 1
Na 4.2 0 2.3 0
70 5 ZnO/CIGSe device and interface analysis
Temperature dependence
In this section, the temperature dependence of the interfacial oxide formation is studied
with the help of GDOES depth profiling. All samples were grown similar to the sample
used for the TEM/XPS analysis and without external sodium supply. Figs. 5.8a-c show
the ratios of the cations to selenium for various CIGSe deposition temperatures, each
together with the Zn signal as a reference.
Interestingly, for the sample deposited at 420
, all cations, Cu, In and Ga, show a
slight increase of their ratio to Se at the interface. Following the arguments given in
the previous section, this originates from an anion substitution from selenium to oxygen
within a thin layer at the interface. In this case the interfacial layer consists of all elements
in similar concentration, forming a Cu(Iny,Ga1−y)(Oz,Se1−z)xlayer.
At the deposition temperature of 520
it is known from the TEM and XPS measure-
ments in the previous section, that a 6 nm thick GaOxlayer forms with impurities from
Cu, In, Zn and Se. The GDOES signal ratios in Fig. 5.8b show indeed that mainly the
Ga/Se ratio increases at the interface to ZnO, whereas The Cu/Se and the In/Se ratios
show only minor deviations.
If the temperature is further increased to 550
(Fig. 5.8c), the Ga/Se ratio at the
hetero-interface also increases further compared to the deposition at 520
. Indicating the
growth of a thicker GaOxlayer at higher deposition temperatures. Interestingly though,
at this temperature, the Cu/Se ratio at the hetero-interface increases noticeably as well,
indicating detectable Cu diffusion into the GaOxlayer at such elevated temperatures.
The Ga/Se ratios for the three different profiles are shown together in Fig. 5.8d. The
signal peak intensities of the ratios were calculated by fitting the signal peaks with Voigt
functions using the horizontal dashed line in Fig. 5.8d as the baseline. The peak area of
the Ga/Se ratio of the 420
sample is 25 %, and the intensity of the 550
sample 150 %
of the peak intensity of the 520
sample. In the previous section, the GaOxthickness
for a 520
sample was shown to be approximately 6 nm, leading to an estimated GaOx
thickness of 9 nm for the 550
sample. The interfacial oxide layer of the 420
sample,
consists of all elements and is therefore estimated to be around 3 nm thick.
5.1.4 Influence of Na
Since Na is an essential doping element for CIGSe it is important to study the influence of
Na on the inter-facial phase formation between ZnO and the different CIGSe with varying
Ga content.
NaF Precursor One way to supply sodium is by depositing a NaF precursor on the
substrate prior to the CIGSe deposition. Here, 10 nm of NaF were deposited at 300
on top of the ZnO layer. In the literature it is known, that if present during the growth
process, sodium slows down the Ga diffusion [124] and changes the preferential crystalline
orientation of the CIGSe layer [47]. This work however, concentrates on the impact on
the interface reaction in the CISe/ZnO and CGSe/ZnO system.
5.1 Interface formation 71
(a) (b)
(c) (d)
Figure 5.8: a) -c) Calibrated ratios of the cations to Se for different deposition temperature
as stated in the graphs. The Zn profile is shown as a reference, the dashed lines indicate the
interface with ZnO. d) Ga/Se ratios for the different temperatures, normalized to the same
Ga/Se value just before the ZnO interface. The dashed horizontal lines are used as the baselines
for the peak area calculations.
Figs. 5.9 and 5.10 show the GDOES elemental depth profiles of the resulting
CGSe/ZnO and CISe/ZnO stacks deposited with the NaF precursor layers in-between.
The depth profiles of the elements Cu, In, Ga, Se, Zn, Na and Au are shown. Au is not
present in the film and indicates the noise level. The GDOES signal intensities are not
transformed into at.% since there is no calibration standard for sodium, which is required
for this procedure. The interface region is marked with two dashed lines.
The shape of the sodium profile within the chalcopyrite is similar for both CISe and
CGSe. It shows a local maximum at the depth, approximately where the chalcopyrite
surface was Cu-rich during growth (compare with Sec. 1.3.1). Most of the sodium remains
at the hetero-interface though, where the shape of the profile shows two overlapping
peaks (well resolved in the CISe/ZnO stack in Fig. 5.9).
72 5 ZnO/CIGSe device and interface analysis
Figure 5.9: Uncalibrated GDOES depth profile of a CuInSe2/ZnO stack, fabricated with a
NaF precursor.
Figure 5.10: Uncalibrated GDOES depth profile of a CuGaSe2/ZnO stack, fabricated with
a NaF precursor. Indium is present as an impurity from the deposition chamber.
5.1 Interface formation 73
(a) (b)
Figure 5.11: a) Depth profile of the Ga/Se ratio and the Na signal of CGSe on ZnO with and
without NaF precursor. The increase in the Ga/Se signal at the interface relative to the bulk
indicates the presence of GaOx. Sodium catalyses the oxide formation. The dashed lines mark
the interface region. b) Depth profile of the In/Se ratio of CISe on ZnO with and without NaF
precursor.
Fig. 5.11 shows the sodium profiles together with the ratios of In/Se and Ga/Se of
the two stacks. The ratios at the ZnO interface are increased for both stacks with NaF
precursor. The increased ratio at the interface compared to the bulk can be explained by
the increased binding of In and Ga with O instead of Se at the interface (see Sec. 5.1.2).
A higher concentration of sodium at the interface clearly leads to a more pronounced
formation of InOxin CISe samples and GaOxin CGSe samples.
Further, the shape of the sodium profile resembles the shape of the Ga/Se and the
In/Se ratio, respectively. This indicates, that the sodium accumulates in the bulk of the
GaOxor InOxlayers and not on their interfaces to CIGSe and ZnO.
NaF post deposition treatment (PDT) The NaF precursor strongly enhanced the
interface reaction between CIGSe and ZnO and appears not to be suitable for the fab-
rication of a superstrate solar cell with a very defined p/n-junction. An alternative for
controlled sodium incorporation to the NaF precursor is the NaF post deposition treat-
ment (PDT), which can be done at a temperature lower than the CIGSe deposition
temperature. This paragraph studies the influence of a NaF PDT on CIGSe/ZnO stacks
only, not separated for CISe and CGSe as it has been done for the NaF precursor.
10 nm of NaF were deposited after the CIGSe deposition by thermal evaporation onto
the CIGSe surface. The substrate temperature after the NaF deposition was kept constant
for 10 minutes before cooling down to room temperature. Fig. 5.12a shows the diffusion
profile of sodium within the CIGSe absorber for different PDT substrate temperatures.
For the PDT temperature of 100
it seems that no sodium diffuses into the absorber,
but remains on the CIGSe surface. The low concentration of sodium found in the bulk and
at the hetero-interface is most likely due to the sodium diffused from the glass substrate.
74 5 ZnO/CIGSe device and interface analysis
For the PDT temperature of 250
, the sodium is still mostly on the CIGSe surface, but
the concentration is increased in the CIGSe bulk with the highest concentration found
at the hetero-interface. The PDT at 400
leads to the highest sodium concentration
within the CIGSe bulk and especially at the hetero-interface. The interface concentration
is increased considerably compared to the interface concentration obtained by the PDT
at 250
.
The Ga/Se ratio profiles of these samples are shown in 5.12b. The Ga/Se ratio of
all samples slightly increases at the hetero-interface indicating the GaOxformation. No
change in the Ga/Se ratios can be observed for the samples with the different PDT
temperatures. This indicates, that the Na PDT does not lead to an increased oxidation
at the interface as it does for the NaF precursors. But similar as for the NaF precursor,
Na accumulates within the interfacial oxide layer as seen from the Na peak position in
the depth profile.
To get a more detailed view of the interfacial chemical composition before and after
the Na PDT done at 300
, XPS measurements on cleaved samples were performed.
As described in Sec. 2.5.2, the samples were cleaved in an inert atmosphere at the
CIGSe/ZnO interface. The sodium concentration is measured to be 6% on the ZnO
side and 4% on the CIGSe side. These values are most likely not the real sodium
concentrations within the layers, since sodium is prone to diffuse to the surface after
cleaving the sample. Further, the oxygen concentrations are increased for the ZnO and
the CIGSe surface compared to the sample without Na. This could be due to adsorption
of oxygen on the CIGSe surface during the cleaving process from the oxygen solved
within the liquid nitrogen. Whether or not the Ga is really more heavily oxidized due
to the increased Na concentration or if the oxygen is simply bound to Na on the surface
(i.e. to NaOH) can be verified by the analysis of the Ga LMM auger peak. The Ga
LMM auger peak is reported in the literature to be around 1060.5 eV for Ga atoms
bound to selenium and around 1057.5 eV if bound to oxygen [125]. Fig. 5.13 shows the
normalized Ga LMM Auger peaks of the ZnO and CIGSe surfaces before and after the
NaF PDT. The ZnO surface shows exactly the same auger peak profile for the surface
with and without sodium. The CIGSe surface does not show any indication for oxidized
Ga. Thus the increased oxygen concentration for the sodium containing surfaces are not
relevant and originate from a thin layer of possibly NaOH. No further oxidation of the
GaOxoccurs due to the PDT.
In summary, it is shown, that the sodium supply via NaF precursors leads to a strong
increase of the oxidation of the CIGSe layer at the interface to ZnO. The sodium remains
preferentially within this interfacial oxide layer. Sodium supply via PDT was effective
for temperatures above 100
, when sodium diffuses into the CIGSe bulk and also accu-
mulates within the interfacial oxide layer. This concentration at the interface increases
strongly for temperatures above 250
. Despite the presence of sodium no further oxida-
tion or growth of the GaOxlayer could be observed during the NaF PDT at temperatures
of 300
.
5.1 Interface formation 75
(a) (b)
Figure 5.12: GDOES depth profiles of a) Na and b) Ga/Se ratio from samples with NaF
post-depositions performed at different temperatures. The amount of sodium at the interface
depends critically on the temperature.
(a) (b)
Figure 5.13: Normalized Ga LMM auger peaks from XPS measurements of the cleaved
CIGSe/ZnO samples, with and without NaF PDT. a) ZnO side. Both spectra are fitted with
two Gaussian functions and a liner background level. b) CIGSe side.
76 5 ZnO/CIGSe device and interface analysis
5.1.5 Diffusion of Zn
During the formation of the oxide layer in-between the chalcopyrite and the ZnO layer
the following reaction occurs for the case of CGSe/ZnO (see Sec. 1.2):
4CuGaSe2+ 3ZnO →Ga2O3+ 3ZnSe +Cu2Se + 2CuGaSe2(5.1)
The resulting ZnSe and Cu2Se may remain as secondary phases at the interface or alloy
with the chalcopyrite bulk. Zn diffusion profiles within CISe and CGSe deposited at
520
, with and without NaF, are shown in Fig. 5.14. The average Zn concentration
within the absorber correlates very well with the intensity of the interface reaction of
the different chalcopyrites. From the SEM image in Fig. 5.3 it is known that CISe with
a NaF precursor forms an approximately 30 nm thick interfacial InOxlayer, from which
an average Zn concentration of around 2 at.% can be calculated within the CISe layer
(calculation details are given below). This fits well to the GDOES intensity of Zn, which
is between 1.2-2.5% within the CISe layer. This confirms that most of the ZnSe diffuses
into the absorber layer. If sodium is not present during the absorber growth, the interface
reaction is less pronounced and the Zn signal is around 1% in the CISe layer. The same
trend can be observed for CGSe with a generally lower Zn concentration due to the lower
interface reaction in the CGSe/ZnO system.
In the CIGSe/ZnO system, the interface reaction is even less pronounced and the Zn
concentration cannot be approximated from the GDOES profiles, since the Zn signal does
not rise above the noise level. But from the TEM image (Fig. 5.6) the GaOxthickness is
known to be around 6 nm. Taking the weight and molar density of ZnO (5.61 g/cm3and
81,4 g/mol [126]) and Ga2O3(6.44 g/cm3and 187,4 g/mol [127]) into account, one can
calculate the amount of oxygen atoms per cm3to be 4.13e+22 and 6.19e+22, respectively.
Thus, for a certain amount of oxygen atoms, their required volume decreases if they change
their binding partners from Zn to Ga. Around 9 nm of ZnO must be dissipated during the
formation of the 6 nm Ga2O3. Assuming, that all the Zn atoms, which are then bound to
Se diffuse into the CIGSe absorber (5.77 g/cm3and 336,29 g/mol for Ga/(Ga+In)=0.3
[128]) and distribute uniformly in this layer, the Zn concentration would be around 0.3
at.%. within the 2.8
µ
m thick CIGSe layer. The effect of the Zn concentration on the
electrical device properties are studied in Sec. 5.2.3.
The diffusion constant of Zn in CISe (without NaF precursor) can be approximated by
fitting Eq. 1.18 to the Zn profile close to the ZnO in Fig. 5.14 to DZn,CISe ≈1e-12 cm2/s,
similar to the 1.3e-12 cm2/s reported for Zn diffusion via Cu-sites in the literature [129].
It should be noted, that the Zn distribution is not completely independent of the native
chalcopyrite defect distribution, mainly VCu and InCu or GaCu [46]. This is especially
pronounced in the Zn profile of the CGSe samples, since the Zn signal increases towards
the CGSe surface, away from the source, the interface to ZnO. Interestingly, the Zn
concentration within the CGSe/NaF sample has a sudden increase at a sputter depth of
1.3
µ
m. It should be noted, that in these nominally pure CGSe samples, In was observed
as an impurity, shown in Fig. 5.10. Further it can be seen, that the Zn profile correlates
with the Na profile and anti-correlates with the In profile.
5.1 Interface formation 77
Figure 5.14: GDOES depth profiles of the Zn signal in different chalcopyrites deposited at
520
on ZnO. Calibrated with the XRF data, which gives approximate values for the Zn
concentration in the bulk. The Zn concentration in the bulk is strongly enhanced if sodium is
present during the chalcopyrite growth. In the CIGSe/ZnO stacks the Zn signal in the bulk is
within the noise level. The estimated value of 0.3 % is marked with the dashed line.
5.1.6 Discussion
The results of Sec. 5.1 are discussed in this section.
CISe vs. CGSe: CISe deposited at 520
on top of ZnO leads to the formation
of a 30 nm thick InOxlayer. In CGSe/ZnO stacks, an approximately 10 nm thick
GaOxlayer formed under the same experimental conditions. The reason for the
difference in the oxide thickness may originate from a higher reaction rate of CISe
with ZnO, but this is unlikely, since the two layers are separated the moment the
first closed layer of the interfacial oxide layer has formed. Thus the diffusion rates of
In/Ga and O through the interfacial oxide layer must determine the growth rate of
the layer. Since GaOxhas a smaller lattice parameter than InOx[130], the diffusion
constant of O is supposed to be smaller, which could explain the difference of the
interfacial oxide layer thickness for the CISe/ZnO and the CGSe/ZnO stack.
Temperature dependence: The chemical composition of the interfacial oxide
layer was shown to depend on the CIGSe deposition temperature. At 420
the
interfacial oxide layer consists of all elements in similar concentration, with a thick-
ness of approximately 3 nm. At 520
mainly GaOxforms at the interface and the
formation of CuOxis suppressed. However, the GaOxlayer has impurities of In and
Cu between 3-5 at.%, as known from XPS measurements. The layer thickness was
determined to be around 6 nm from HR-TEM images.
Interestingly, the thermodynamic calculations in Sec. 1.2 did not predict the forma-
tion of CuOxat any temperature, since the enthalpy of formation is lower than for
ZnO. Thus at 420
the system is limited by the reaction kinetics, possibly limited
due to the relatively low ionic mobility at 420
and has therefore not reached its
energetic minimum. It appears that the oxidation process occurs during the high
78 5 ZnO/CIGSe device and interface analysis
temperature deposition step in stage 2, when Cu, In and Ga are present at the inter-
face to ZnO. At the interface an anion exchange of Se and O leads to the oxidation
of Cu, In and Ga.
When the temperature is raised, the Ga ions gain sufficient mobility, which low-
ers the kinetic limitation and the Cu-O and the In-O bonds are replaced with the
energetically favoured Ga-O bonds. After the formation of the first GaOxmono-
layers, the growth should be limited by the diffusion rates of the different elements
through this initial layer. The ionic size of In3+ is 0.81 ˚
A and 0.77 ˚
A for Cu1+, both
larger than to 0.62 ˚
A for Ga3+ [131]. This indicates a lower diffusivity of In and
Cu ions compared to Ga ions through the GaOxlayer, which also explains the low
concentration of Cu and In within the 6 nm thick GaOxlayer.
For a temperature as high as 550
, an increased concentration of Cu was again
found to be bound to oxygen at the interface. Most likely this is due to the increased
diffusion of Cu into the GaOx, which was still low at 520
. The higher diffusivity
of Cu compared to In is due to the smaller ionic radius and the weaker covalent
bond of Cu-Se compared to the In-Se bond within CIGSe [41].
Impurities within GaOx:XPS interface analysis showed that the GaOxlayer
has large impurities of In, Zn, Cu and Se in the range of some at.% points. XPS
results suggest that the layer is anion poor. The influence of these impurities on the
electronic properties will be studied in the next chapter.
Influence of Na: It was shown that NaF precursors lead to an increased oxidation
of Ga or In and that the Na preferably accumulates within the interfacial oxide layer.
No further oxidation at the interface was observed for the samples post treated with
NaF, during which the sample is heated to 300
for only 10 minutes. Still, it
was found, that Na increasingly accumulates within the interfacial oxide layer, with
increasing temperature during the post treatment.
Thus it can be concluded, that the presence of sodium acts catalytic on the oxidation
process. But the oxidation process is negligible at low temperatures applied for a
short time, as for the NaF PDT. The catalytic effect is unlikely to be induced by an
increased diffusion rate of Ga and O, since sodium is known to reduce the mobility
of Ga [124]. It is more likely to be induced by a reduction of the Zn-O bonding
strength at the interface between ZnO and the interfacial oxide layer. This could be
triggered by the strong ionicity of the Na-O bond, which reduces the ionicity and
with it the bonding strength of a neighbouring the Zn-O bond, similar as argued
in [132] for surface reactions.
Thus, depending on the time and the temperature, the presence of sodium within
the GaOxcould lead to an increase of the net density of acceptor states in two
ways, first by reducing the amount of VOdonor states within the GaOx, secondly
by introducing acceptor states due to NaGa states.
5.2 Device Porperties 79
Zn diffusion: It was shown that Zn, which is set free from the ZnO layer during
the interface reaction, diffuses into the chalcopyrite bulk and that the concentration
of Zn depends on the thickness of the interfacial oxide layer. Due to the enhanced
formation of an interfacial oxide layer in the ZnO/CISe system, the Zn concentration
within CISe devices was estimated to be up to 2 at.% points. Within CGSe layers
a lower Zn concentration between 0.5 at.% and 1 at.% was found. The interfacial
oxide layer in the studied CIGSe devices was only 6 nm, from which an average Zn
concentration of 0.3 at.% was calculated.
The resulting Zn depth profile within the chalcopyrite bulk was shown not to be
distributed as expected from a single diffusion process as shown in Fig. 1.4. Rather
it is found, that the Zn distribution correlates with the Na distribution, especially
clearly observable in the CGSe layer. Na is iso-electronic to Cu, and Cu sites are
commonly believed to be the preferred occupation sites for Na in the literature
(see Sec. 1.3.1). Due to the strong off-stoichiometry of the CGSe layer, with a
Cu/(In+Ga) ratio of approximately 0.9, it can be assumed that the Cu vacancies
are not saturated by Na, Zn or In impurities. Thus, the observed correlation with
Na implies that Zn, in the region where it follows the shape of the Na profile,
preferably sits on VCu sites. The observed anti-correlation of the Zn profile with
the In impurity profile in the CGSe layers supports this assumption. Further away
from the ZnO interface, where the Zn concentration is lower, no correlation to the
In and Na profiles is found any more and it may be, that Zn is equally distributed
between the native defects. Thus, this analysis indicates a preferable VCu site
occupation of Zn atoms at high Zn concentrations. Zn can be both donor and
acceptor in CIGSe (see Sec. 1.3.1), depending on whether it sits on a Ga/In site or
a Cu site. The impact of the Zn impurities on the device is studied in Sec. 5.2.3.
In summary, all samples studied in this section showed an inter-facial oxide layer, which
developed during the high temperature process step in stage 2 and 3. The composition
was shown to depend on the chemical composition of the chalcopyrite at the interface, the
chalcopyrite deposition temperature and the NaF treatment. These three dependencies
can be used to engineer the interface. The next section will study the influence of the
different interfacial oxide layers on the device performance.
5.2 Device Porperties
In this section the information obtained on the structural and chemical properties of the
ZnO/CIGSe interface will be correlated with the electrical properties of ZnO/CIGSe/Au
devices. First, the experimental results will be presented, which will then be discussed
jointly at the end of this chapter.
80 5 ZnO/CIGSe device and interface analysis
(a) (b)
Figure 5.15: a) J−Vcurve of devices with CISe, CIGSe and CGSe deposited on ZnO. The
deposition process of the CIGSe, 1B started with stage 1B (In-Se) instead of 1A (Ga-Se). b)
J−Vcurve of devices with the same CIGSe absorbers, but different 50 nm thick annealed
precursor layers.
5.2.1 Interface composition
The elemental composition of the interfacial oxide was shown in the Sec. 5.1 to be depen-
dent on the Ga content of the CIGSe layer deposited on top of the ZnO. Thick layers
of InOxwere formed for CISe whereas thinner GaOxlayers were formed for CIGSe and
CGSe. Fig. 5.15a shows the J−Vcurves of devices with CISe, CGSe or CIGSe absorber
deposited on ZnO at 525
.
The CISe devices were observed to be generally shunted. Lock-in thermography mea-
surement showed, that no local shunts were present and thus it can be concluded that the
InOxinterfacial layer forms an ohmic contact between ZnO and CISe. Hot-point probe
was performed on the bare absorber, which showed that the CISe layer remains p-type
even with a few at.% contamination of Zn.
Devices made from CGSe do show a rectifying behaviour, but a relative low VOC
of 640 mV, compared to CGSe devices in substrate configuration whose VOC is around
900 mV. The J−V-curve shows a kink in the fourth quadrant, indicating some kind of
barrier for the charge carriers.
A similar kink can be observed for the CIGSe devices, when the deposition process
has started with stage 1A, which is Ga-Se. The VOC drops significantly if the deposition
process starts with stage 1B, which is In-Se. This results in an In rich chemical
composition around the interface to ZnO, possibly leading to the formation of (In,Ga)Ox
instead of GaOx.
To prove that the difference in the rectifying behaviour is due to the interfacial oxide
layer and not due to the different Ga content in the chalcopyrite bulk layers, thin buffer
layers were deposited prior to the absorber deposition. To analyse the chemical com-
5.2 Device Porperties 81
position of the buffer layers with XPS, first 10 nm thick In2Se3and Ga2Se3layers were
deposited onto ZnO substrates and annealed in vacuum at the CIGSe process temperature
of 520
. The XPS measurements on these samples showed the presence of (In,Ga)Oxfor
the In2Se3precursor and GaOxfor the Ga2Se3precursor. (In,Ga)Oxinstead of pure InOx
must have formed due to Ga contaminations from the deposition chamber. Knowing the
interface reaction to occur for these very thin buffer layers, now 50 nm of In2Se3, Ga2Se3
and Se were deposited onto fresh ZnO substrates and annealed at the CIGSe process
temperature of 520
. On top of these buffer layers, CIGSe was deposited, resulting in a
nominally identical bulk chemical composition but different interfacial chemical compo-
sition. The CIGSe process was standard as described in Sec. 2.1. The results are shown
in Fig. 5.15b. The Se precursor slightly reduces the VOC and the JSC compared to the
standard device in Fig. 5.15a. The device with the In2Se3precursor, which was shown to
be transformed to (In,Ga)Ox, was not shunted, as the pure CISe samples without any Ga
content. But it showed a low parallel resistance and VOC of only 55 mV. The sample with
the pre annealed Ga2Se3precursor was shown to form pure GaOxat the interface. This
leads to an increased VOC but a decreased JSC compared to the CIGSe sample without
precursor and very similar to the J−Vcurve of the CGSe sample. Clearly indicating
that the chemical composition at the interface is the determining factor for the shape of
the J−Vcurve.
5.2.2 Deposition temperature
Fig. 5.16a shows the J−Vcurves of samples with absorber layers deposited at different
temperatures. For temperatures below 500
a low open circuit voltage in combination
with a high short circuit current is usually observed. The open circuit voltage increases
for samples made at 525
, while the maximum photo-current remains the same. If the
deposition temperature is further increased, the voltage does not increase further but the
fill factor and the maximum photo-current decreases. For the sample fabricated at 560
the light and the dark J−Vcurve cross over in the first quadrant. The optimum CIGSe
deposition temperature was found to be around 525
.
The corresponding C−Vcurves are shown in Fig. 5.16b. All samples show a strong
increase of the capacitance when positively biased. Standard substrate devices generally
have capacitances below 50 nF/cm2at positive bias, while the capacitance of superstrate
devices go up to 600 nF/cm2. It should be noted, that the dashed lines in Fig. 5.16b
indicate that in this bias range the capacitance of the space charge could not be measured
correctly due to the high DC current present at positive voltages for the 475
sample,
and due to the degradation of the back contact (see Sec. 6.2), which leads to a second
diode in opposite direction, for the 560
sample. It should be noted, that the capacitance
at negative voltage bias is increased for the 560
sample compared to the other two. The
plot of the charge carrier density over the width of the space charge region, derived with
Eq. 2.10 and shown in the inlet of Fig. 5.16b, shows a constant value in the absorber bulk
for all three samples. It is around 6e+14 cm3for the samples made at 475
and 525
and slightly lower at around 4e+14 cm3for the 560
sample.
82 5 ZnO/CIGSe device and interface analysis
(a) (b)
Figure 5.16: a) dark and illuminated J−Vcurves of devices fabricated at different temper-
atures. b) C−Vcurve of the same devices recorded at 293 K and 1kHz. The dashed lines
indicate that the capacitance does not correspond to the space charge region due to secondary
effects.
For the rest of this work, devices fabricated at the optimum temperature of 525
are
studied, if not mentioned other wise.
Charge carrier diffusion length To understand the distorted shape of the J−V
curve in Fig. 5.16, it is useful to study the voltage and depth dependent charge carrier
collection within the device. EQE, EBIC and C−Vare used in this work to study the
charge carrier collection. Fig. 5.17a shows the EBIC image of the 525
device, recorded
at short circuit condition. The EBIC image shows the local current generation along a
cross section, randomly broken from a finished device. There is a slight dependence on the
grain structure, but a clear trend can be observed towards lower currents for increasing
distances from the CIGSe/ZnO interface. Fig. 5.17b shows a representative profile over a
single absorber grain. By fitting the Eq. 2.19 to the profile, it is possible to extract the
electron diffusion length within CIGSe, Ln, and the space charge region width, xSCR (see
2.4.4). The result for xSCR between 0 - 200 nm, whereas 0 nm lead to the best fit. The
result for Lnwithin the CIGSe is between 800 nm and 900 nm, with the best fit for 880 nm.
As a comparison to the EBIC results, the effective collection length, which is xSCR plus
Ln, can be extracted from the slope of the EQE spectra by Eq. 2.16. The wavelengths
range between 550 - 800 nm was used, since this is the range less influenced by the shape
of the ZnO absorption spectrum. Fig. 5.18 shows the EQE spectra for three different bias
voltages. The effective collection length extracted from the EQE spectrum recorded at
short circuit condition is 1000 nm. With xSCR = 220 nm as derived with Eq. 2.7 from
the C−Vmeasurement, the diffusion length Lnbecomes 780 nm. At a voltage bias of
-250 mV, the EQE increases in the near infrared region and the calculated diffusion length
remains at around 800 nm and xSCR increases to 800 nm according to the C−Vresults.
For the positive voltage bias, the diffusion lengths is calculated to 200 nm and the xSCR
5.2 Device Porperties 83
(a) (b)
Figure 5.17: a) Electron Beam Induced Current (EBIC) image of the 525
device, recorded
from a cross-section, and superimposed with the SEM image. A probe current of 63 pA was
measured at no high current mode (Gemini column, extractor voltage at 10 kV, extractor at
260
µ
m). b) Extracted line profile together with a fit of Eq. 2.19 performed with EBIC-view.
25 nm. Further, the whole EQE spectrum is lowered to smaller values even in the short
wavelength region, indicating a barrier present at forward bias.
5.2.3 Diffusion of Zn
During the reaction of CIGSe with ZnO, not only GaOxis formed but also free Zn atoms,
which diffuse into the absorber as shown in Fig. 5.14. It was estimated that the Zn
concentration in the best superstrate devices with a 6 nm thick GaOxlayer is around 0.3
at.%. As it is discussed in Sec. 1.3.1, this can lead to an increase in acceptor states or donor
states depending on whether the Zn atoms preferably sit on Cu or In/Ga sites. The effect
Figure 5.18: EQE spectra of the 525
device at different voltage biases. The electron
diffusion length within the CIGSe is calculated to be around 800 nm.
84 5 ZnO/CIGSe device and interface analysis
(a) (b)
Figure 5.19: a) IV curves of solar cells fabricated in substrate configuration
ZnO/CdS/CIGSe/ZnSe/Mo with different thickness values of ZnSe layers added as precur-
sor onto the Mo back contact for controlled Zn contamination. 2 nm ZnSe correspond to
0.05 at.% in the CIGSe bulk. The Zn doping of the CIGSe layer has strong influence on
the VOC.b) Corresponding C−Vcurves, recorded at 1 kHz and 293 K. The Zn doping of
the CIGSe layer increases the CIGSe doping level for low concentrations and decreases it for
high Zn concentrations. The inlet shows the relation between charge carrier density and Zn
concentration.
of Zn on CIGSe is now studied in substrate devices to enable the exact control of the Zn
content. To do this ZnSe buffer layers were deposited on Molybdenum substrates and on
top of that the CIGSe absorber, followed by CdS and ZnO. The CIGSe was deposited by a
three stage process at 520
, designed to lead to a flat Ga profile with Ga/(In+Ga)=0.27
and Cu/(In+Ga)=0.87. 2, 5 and 20 nm thick ZnSe layers were deposited by thermal
evaporation at 300
. Due to the similar mass density and average atomic mass of ZnSe
and CIGSe 20 nm of ZnSe can be calculated to correspond to 0.5 at.% of Zn within the
2
µ
m thick CIGSe layer. The 2 nm and 5 nm layers correspond to 0.05 at.% and 0.12 at.%.
The J−Vmeasurements of the Zn doped devices are shown in Fig. 5.19a. The main
influence of the Zn doping can be observed on the open circuit voltage. The VOC is
highest for the samples with low Zn concentration of 0.05 at.%, around 520 mV, whereas
the VOC values of the devices without Zn are lower at around 500 mV. The cells with a
high Zn concentration of 0.5 at.% show the lowest VOC, around 425 mV. Corresponding
to this, the effective charge carrier density, derived from C−Vmeasurements, follows
the same trend, as shown in Fig. 5.19b. The sample with 0.05 at.% has a high doping
density of 1e+15 cm−3compared to 1e+14 cm−3for the sample with 0.5 at.% Zn. To
analyse whether the Zn doping has an effect only on the charge carrier density or also on
the charge carrier lifetime, the VOC of representative samples is plotted in Fig. 5.20b over
a logarithmic scale of the charge carrier density. For a constant charge carrier lifetime
of the different samples, the measurements points should fall on a single line, since the
VOC is proportional to the exponential of the charge carrier density (see Eq. 2.2). The
5.2 Device Porperties 85
(a) (b)
Figure 5.20: a) Photo-luminescence spectra of CIGSe/ZnSe/Mo stacks with different ZnSe
thickness values b) Comparison of the VOC and the PL intensity with the doping density taken
from Fig. 5.19. The sample with high Zn concentration deviates from the VOC /IP L ∝ln(NCV )
line (black), obtained from Eq. 2.2 for a constant electron lifetime, due to increased non-
radiative recombination.
three samples with 0, 0.05 and 0.12 at.% Zn lie indeed on a single line. The sample with
0.5 at.% Zn (20 nm ZnSe) does deviate though. This could originate from a lower bulk
lifetime of this sample or from increased interface recombination.
In order to eliminate the influence of the interface recombination at the CdS in-
terface, the recombination in bare absorbers was studied by photo-luminescence (PL)
spectroscopy. The resulting PL spectra of the bare absorbers are shown in Fig. 5.20a.
The PL intensity is, similar to the VOC, proportional to the charge carrier density and
can also be lowered by non-radiative recombination in the bulk or at the interface. The
integrated intensity of these peaks is plotted in Fig. 5.20b. The same trend as for the
VOC can be observed for the PL intensity. The sample with 0.5 at.% Zn shows stronger
non-radiative recombination loss.
It should be further noted, that very similar results were obtained with CISe ab-
sorbers on ZnS buffer layers of different thickness, confirming that Zn leads to detrimen-
tal deep donor states if present in high concentrations. For the absorber studied here
(Ga/(In+Ga)=0.27 and Cu/(In+Ga)=0.87) the threshold concentration was found to be
approximately 0.25 at.% of Zn.
5.2.4 Influence of alkalis
Alkali treatments are usually performed on CIGSe absorbers to enhance the p-type doping
density. Sodium and Potassium are the most common choices of the alkali metals. Often
they are supplied via out-diffusion from alkali containing glass substrates through the
Mo layer. Since, in contrary to the Mo back contacts, ZnO layers are good diffusion
barriers for alkalis, the alkali dopants have to be supplied externally prior to the absorber
86 5 ZnO/CIGSe device and interface analysis
deposition by a precursor or after the CIGSe deposition by a post deposition treatment
(PDT). In order to suppress the formation of NaOH and KOH, the fluoride compounds
of the alkali metals, NaF and KF, are used. No negative influence of the fluorine atoms
on the CIGSe absorber has been shown so far [133].
Sodium Fluoride Treatment (NaF) Fig. 5.21a shows the J−Vcurves of superstrate
devices with different NaF treatments. During the post deposition treatments, 10 nm
of NaF were deposited on the CIGSe surface at a given temperature and annealed for
10 minutes at the same temperature. For the NaF precursor, 10 nm of NaF were deposited
onto the ZnO prior to the CIGSe deposition. The absorber was deposited via a modified
three-stage process starting with the co-evaporation of Ga-Se, as described in Sec. 1.3.1.
The NaF PDT at 100
leads to a J−Vcurve that is similar to the J−Vcurve of an
untreated device (see Fig. 5.16) but with a decreased series resistance (details in Sec. 6.1).
The JSC is 32 mA/cm2, the VOC 620 mV and the FF 28% leading to an overall ηof 5.5%.
If the temperature during the NaF PDT is increased to 300
,ηincreases to 10.1%, due
to an increase in JSC to 37 mA/cm2,VOC to 680 mV, and FF to 40%. The FF drops if
the NaF PDT is performed at temperatures higher than 300
. At 400
the FF drops to
20%, while the VOC remains 680 mV and the JSC drops to 25 mA/cm2. The device with
NaF deposited as a precursor prior to the CIGSe deposition has a low JSC of 5 mA/cm2
in combination with a high VOC of 740 mV and the injection current is pushed towards
higher voltages. This indicates a strong electrical barrier in the device.
Fig. 5.21b shows the corresponding C−Vcurves of the post-treated devices. The NaF
PDT at 100
leads to no change in the C−Vcurve compared to the untreated device
shown in Fig. 5.16b. If the capacitance is interpreted as a space charge capacitance, then
the high capacitance at positive bias translates with Eq. 2.7 into a 20 nm thick space charge
region. The free charge carrier density can be calculated with Eq. 2.10 to be 6e+14 cm−3
in the CIGSe bulk. The NaF PDT at 300
leads to C−Vcurves with high capacitances
for the whole studied voltage range. Leading to a SCR width of 10 nm at 0.5 V and 30 nm
at -0.5 V. The NaF PDT at 400
further increases the capacitance, lowering the SCR
width even further.
Beside the deposition temperature, the diffusion velocity of sodium into the absorber
may be important. In order to slow down the diffusion of sodium during the annealing
time a 10 nm thick Mo layer was deposited prior to the NaF PDT on the absorber surface.
This leads to a different sodium depth profile compared to the same NaF PDT performed
without Mo layer, shown in Fig. 5.22. The sodium concentration within the CIGSe bulk
reaches a similar value for both samples, but close to the CIGSe/ZnO interface and
especially at the interface, the concentration is considerably smaller for the sample with
the Mo layer on the CIGSe surface.
The J−Vand C−Vcurves of these two devices are compared to the untreated device
in Fig. 5.23a and b. Compared to the untreated device both show high JSC values of
37 mA/cm2. However, the VOC of the Mo diffusion barrier device is lowered from 670 mV
to 570 mV. Nevertheless ηis increased to 10.8% due to an increase of the FF from 40%
5.2 Device Porperties 87
(a) (b)
Figure 5.21: a) J−Vcurves of devices with CIGSe deposited at 525
and with NaF PDT
at performed at different temperatures b) C−Vcurves of the same devices recorded at 293 K
and 1 kHz.
Figure 5.22: GDOES depth profile of Na in CIGSe/ZnO stacks, with and without the depo-
sition of a 10 nm thick Mo diffusion barrier prior to the NaF PDT.
88 5 ZnO/CIGSe device and interface analysis
(a) (b)
Figure 5.23: a) J−Vcurves of devices with CIGSe deposited at 525
and with NaF PDT
at performed at different temperatures b) C−Vcurves of the same devices recorded at 293 K
and 1 kHz.
to 51%. Employing Eq. 2.10 the charge carrier density can be calculated from the C−V
curves in Fig. 5.23b. The Na PDT increased the charge carrier density in the CIGSe bulk
of the device with the Mo layer from 5e+14 cm−3to 9e+15 cm−3, and for the device
without Mo layer to 5e+16 cm−3.
In forward voltage bias, the capacitance of the Mo layer device is unchanged compared
to the sample without sodium. This indicates that the strong increase as observed for the
NaF treated sample without the Mo layer is most likely due to the presence of sodium at
the interface between CIGSe and ZnO.
Potassium Fluoride Post Deposition Treatment (KF PDT) The effect of potas-
sium in superstrate devices is shown in Fig. 5.24. All samples were from the same deposi-
tion run performed at 525
. The post deposition treatment was performed at nominally
300
. The deposition of potassium fluoride was performed at a high deposition rate of
several nm/s and lead to a slight kink in the fourth quadrant of the J−Vcurve. Very
similar to the J−Vcurve of the sample with the NaF PDT in Fig. 5.21. Not shown here,
but KF provided as a precursor leads to an even more pronounced kink within the J−V
curve. This indicates, that potassium, similar to sodium, induces an electron barrier at
the hetero-interface.
The pure NaF treatment performed in this sample variation was done with a very low
NaF deposition rate of 1 nm/min for 10 minutes. This leads to a J−Vcurve similar to
the one in Fig. 5.21 where the sodium diffusion rate was lowered by the additional Mo
layer.
The third sample became the same NaF PDT for only 5 minutes, followed by a
KF PDT with 1 nm/min also for 5 minutes. Compared to the pure NaF treatment,
the combined NaF and KF treatment leads to an increase of the VOC from 525 mV to
580 mV. ηof this device was 11.0 %, one of the highest measured efficiencies obtained
5.2 Device Porperties 89
Figure 5.24: J−Vcurves of devices with absorbers from the same deposition run performed
at 525
, but with different PDTs performed at 300
. The device with the NaF PDT followed
by the KF PDT reached the highest efficiency of 11%. The J−Vcharacteristics were obtained
from a one-diode model fit of the illuminated and dark curve (l, J0, Rs, Rp). The minimum
band gap was 1.15 eV (GDOES) and the Cu/(In+Ga) = 0.89.
in this work, indicating further potential of improvements by fine tuning the alkali post
deposition.
In summary, as shown before, low-rate deposition of sodium leads to an improved fill
factor above 50 %. The combination of NaF and KF treatment was shown to improve the
VOC similar as observed in substrate devices.
5.2.5 Discussion
VOC and the chemical interface composition
In Sec. 5.2.1 it was shown, that the VOC of the superstrate device depends critically on the
chemical composition of the CIGSe absorber directly at the interface to the ZnO. Pure
CISe absorbers were completely shunted due to the non-rectifying contact between the
InOxlayer, which formed at the interface to ZnO. Pure CGSe layers and CIGSe layers
with an annealed Ga2Se3precursor, were shown to have the highest VOC values, but low
JSC values. In CIGSe devices, the VOC was shown to be dependent on how much In was
present in the GaOxlayer, which depend on whether or not the deposition process was
started with In-Se or whether or not a In2Se3precursor was used.
As it will be shown later (in Fig. 8.4), the band gap of amorphous InGaOxvaries by 1 eV
depending on the In content, being the lowest for pure InOx. According to the common
anion rule, which was found to be a good approximation for most oxide semiconductors
[134], the valence band maximum is similar for all oxides but the conduction band varies
with varying cation species. Thus, the CBM of InOxis expected to be around 1 eV lower
than the CBM of GaOx. As discussed already in Sec. 4 this should lead to a conduction
band cliff at the interface for InOx, and a conduction band spike for GaOx. How this
difference can influence the VOC in a standard device was shown in Fig. 3.1. For highly
90 5 ZnO/CIGSe device and interface analysis
defective interfaces, the VOC drops almost linearly with the height of the conduction
band cliff, which lowers the activation energy for the interface recombination. Therefore
it can be assumed that larger concentrations of In within the interfacial oxide layer limit
the VOC due to a high interface recombination velocity and a cliff like conduction band
alignment.
A Cu poor surface, as often observed in substrate devices, could reduce the interface
recombination losses as discussed in Sec. 3.1. The XPS measurement presented in Sec. 5.1
indicate such a Cu-poor surface, but the error of the quantitative determination is large
and TEM-EDX as well as GDOES measurements could not confirm this. In the following
no Cu poor surface is assumed for the superstrate devices.
In summary, In contaminations within the GaOxlayer reduce the VOC due to an
increasingly cliff like conduction band offset.
FF and the Cu impurities within the GaOx
All superstrate samples studied in this Section, showed a strong capacitance increase
together with a decrease of the extracted photo-current at forward biases, leading to a
low FF and a low η. This behaviour was found in Sec. 3.2 to be characteristic for devices
with a high acceptor density around the buffer layer. In this model the capacitance is
interpreted as a pure SCR capacitance. The capacitances measured at forward bias would
translate into a dSCR of just 10-20 nm. With 5-10 nm of the SCR within the p-type region.
This is approximately the thickness of the GaOxlayer, seen in the TEM images in Fig. 5.6.
Such a short space charge region would limit the collection efficiency of the photo-current
and could explain the observed low fill factors of the J−Vcurves. However, such a small
space charge region would require a high density of acceptor states at the hetero-junction.
These acceptor states may originate from OSe within CIGSe, induced by an oxidation of
the CIGSe at the interface to ZnO or to the interfacial GaOxlayer. The OSe acceptor state
is reported to be 120-140 meV above the VBM of CIGSe [?]. Thermodynamic calculations
predict that the oxide binaries will segregate during the oxidation process [47], limiting
the OSe acceptor states to the interface between CIGSe and the oxide.
Another origin of interfacial acceptor states can be found from the XPS measurements
presented in Sec. 5.1.3. They show that the interfacial oxide layer consists of GaOxwith
large amounts of impurities of all other elements. Se and In are iso-electronic to O and Ga
respectively, and the defects SeOand InGa are charge neutral. They only lower the band
gap of the GaOxlayer [135] [136] [137], which was shown in Sec. 3.2 to have negligible
influence on the capacitance. However, Cu impurities are known to lead to acceptor states
within TCOs, as shown in Sec. 1.3.2. No literature exists on the doping effect of CuGa
states within Ga2O3, but it is known that at high concentrations of Cu, the p-type TCO
CuGaO2forms [138], in which the valence band is composed of the Cu 3d orbitals and
shifted by around 1 eV compared to Ga2O3. Also Zn is believed to form shallow ZnGa
acceptor states within Ga2O3[139], but for the amorphous GaZnO an increasing n-type
doping was observed for low Zn concentrations compared to the pure Ga2O3[140], with
5.2 Device Porperties 91
the Zn 4s orbitals forming the extended conduction band [141]. Therefore, it is not clear
whether or not Zn leads to acceptor states within amorphous GaOx. Experiments shown
in Sec. 8.3 indicate that Zn does not lead to p-type doping.
The XPS results further indicate, that the GaOxis oxygen poor which would induce
cation interstitials like Gaior anion vacancies like VO, both leading to deep donor
states [142]. The maximum charge carrier concentration reported previously for O-poor
Ga2O3is 1018 cm−3[143], which could be compensated by the large amount of impurities
from Cu .
In summary, CuGa or OSe are assumed to lead to acceptor states in the bulk at the
interface to GaOx, which in turn lowers the fill factor and limits the device efficiency.
VOC, FF and the deposition temperature
Besides the chemical composition of the interfacial layer, the deposition temperature of
the CIGSe absorber was shown to have a strong influence on the VOC as well as on the
FF. In Sec. 5.2.2 it was shown that absorbers deposited at 475
lead to low VOC values
below 200 mV and absorbers deposited at 560
lead to low fill factors.
The change of the interfacial composition with the deposition temperature was shown
in Fig. 5.1.3. For a deposition temperature as low as 420
, it was found that an ap-
proximately 2-3 nm thick interfacial oxide layer forms, with equal contribution of Cu, In
and Ga on the cation sites. Taking the arguments of the discussions above into account,
this material combination should lead to a p-type oxide, due to the Cu content, with an
electron affinity higher than that of CIGSe, due to the In content. This resembles the
situation illustrated in Fig. 3.6e for χ >4.5 eV. In this model the low VOC is induced by
the increased interface recombination due to the combination of high acceptor density and
high electron affinity of the buffer layer. Further, this situation could lead to an increased
tunnel current from the ZnO conduction band to the CIGSe valence band, if the buffer
layer is sufficiently thin.
Once the temperature is increased to around 520
, the GDOES and the XPS mea-
surements showed, that the contribution from Cu and In to the interfacial oxide layer is
strongly reduced and mainly GaOxis present at the interface. The low concentration of
In possibly leads to a reduced conduction band cliff at the hetero-interface, which leads
to a reduced interface recombination and an increased VOC. The Cu impurities induce a
high density of acceptor states in the interfacial oxide layer leading to the observed low
FF.
At temperatures above the optimum deposition temperature, at 560
, the GaOx
thickness increases further to approximately 9-10 nm. This is expected, as the Ga
diffusion through the GaOxlayer is enhanced at higher temperature, which was argued
to be the limitation of the interface reaction. The increased temperature also increases
the diffusion of the other elements and since Cu is the most mobile specie, an enhanced
diffusion of Cu into the GaOxis expected and is indeed observed for these high tem-
peratures. Cu is supposed to induce acceptor states in the oxide. And according to the
92 5 ZnO/CIGSe device and interface analysis
simulations shown in Fig. 3.4c an increase of the acceptor density leads to an increased
electron barrier and an increased capacitance, resulting in a kink in the fourth quadrant
of the J−Vcurve. The increased kink together with the increased capacitance was
indeed observed experimentally.
In summary, it is proposed that In reduces the activation energy for the interface
recombination and Cu induces acceptor states within the interfacial oxide layer. The
concentration of In and Cu within the GaOxlayer is the lowest at on optimum deposition
temperature of around 520
, leading to the highest device efficiencies.
Zn doping
As discussed in Sec. 1.3.1, Zn is amphoteric as it can act as a donor or acceptor depending
on which lattice site it occupies. ZnIn leads to acceptor states, whereas ZnCu leads to donor
states. Due to the overall Cu-poor composition of the CIGSe absorbers it is likely that
in this case Zn prefers the ZnCu states and thus reduces the p-type doping of the CIGSe
layer. This is in accordance with the observation in Sec. 5.1.5, which showed that the Zn
depth profile in CGSe correlates with the Na depth profile, whenever the Zn concentration
is high. The controlled doping experiments in Sec. 5.2.3 show indeed, that for a high Zn
concentration of 0.5 at.% the measured p-type doping is reduced when compared to the
Zn-free device. For low concentrations, below approximately 0.25 at.%, the measured
p-type doping in the CIGSe bulk was observed to increase. Thus, in the CIGSe studied
here, the Zn atoms appear to preferably occupy the In/Ga sites for low concentrations.
Further it was found from the J−Vand PL measurements that the electron lifetime,
and thus the recombination losses, are unchanged for Zn concentrations below 0.25 at.%
compared to undoped devices. But the recombination losses were increased for the sample
with 0.5 at.% Zn concentration. Since both measurements are sensitive to bulk and inter-
face recombination, both recombination channels could be increased due to the high Zn
contamination. But, as it was shown in Sec. 3, that the interface recombination generally
decreases when the bulk doping density decreases compared to that of the buffer layer.
Further, the increased recombination loss was observed for two totally different interfaces.
Thus it is more likely that bulk recombination has increased. Simulations with SCAPS
show that a donor level has to be located 200 - 300 meV below the CBM in order to in-
crease the recombination loss and lead to a VOC reduction of the observed 20 mV without
reducing the JSC. Thus it can be concluded, that ZnCu states are effective recombination
centres and that Zn preferably occupies the VCu states at concentrations above approxi-
mately 0.25 at.%. ZnIn is a shallow acceptor which does not act as a recombination centre
and is dominant for Zn concentration below approximately 0.25 at.%.
In Sec. 5.1.5, the Zn concentration for the sample prepared at 525
was approximated
to be 0.3 at.%. This is slightly above the critical threshold concentration for Zn atoms
(see inlet of Fig. 5.19b). Thus it seems that the amount of Zn present due to the
formation of a 6 nm thick GaOxlayer, could have a small negative influence on the charge
carrier density and charge carrier lifetime. It should have a more noticeable negative
5.2 Device Porperties 93
influence on the samples deposited at higher temperatures than 525
, which have thicker
GaOxlayers and therefore higher Zn concentration within the absorber. This could ex-
plain the lower charge carrier density and the lower VOC of the sample deposited at 560
.
In summary, Zn doping is only detrimental for the device when present at concen-
trations above 0.25 %. The Zn concentration for the optimum deposition temperature
is estimated to be only slightly above this value, and should therefore define no major
limitation for the state-of-the-art device efficiency.
Alkali treatment
As in substrate devices, it is found that Na and K doping increases the device efficiency,
in this case from around 5-6 % absolute to 10-11 %. The difference to substrate devices
is, that the alkali treatment increases the VOC and the JSC, as shown in Fig. 5.21a. An
increase of the VOC is the expected result of a NaF treatment due to the increased p-type
doping generally observed due to the reduction of InCu states by Na. The increase of JSC
is more unusual. This could originate from a Fermi-level pinning as illustrated in the
simulations shown in Fig. 3.4a. But the measured capacitance is not increased at negative
biases, which excludes such an option. Further, back contact recombination may play a
role, but the strong Ga gradient present in the films should not allow any back contact
recombination. Thus SRH bulk recombination is the most likely option. A cause of an
increased bulk recombination rate are the ZnCu donor states within the CIGSe. Further,
Na is known to preferably occupy Cu sites and the discussion of Fig. 5.10 revealed that
Na and Zn share similar sites for high concentrations of Zn as in the samples studied
here. Thus, the presence of Na could reduce the amount of ZnCu states, similar as it is
argued for the InCu states in the literature (Sec. 1.3.1). This would increase the p-type
doping and reduce the bulk recombination rate.
The effect of the NaF post-deposition is found to be highly dependent on the Na
deposition rate and temperature. A temperature of 300
is required to achieve sufficient
Na diffusion into the CIGSe. At higher temperatures the fill factor and short circuit
current become strongly reduced. In the previous Section it was shown, that Na tends to
accumulate at the hetero-interface the more the higher the PDT temperature is set. Here
the increasing Na content is correlated with an increasing capacitance. And an increasing
capacitance with a decreasing charge extraction was shown in Sec. 3.2 to originate from
an increase in acceptor states at the hetero-junction. The Na profile follows the shape of
the Ga/Se ratio, which indicates, that Na accumulates within the interfacial GaOxlayer
and not at the interface between CIGSe and GaOx. Thus it appears that Na introduces
acceptor states within the GaOxlayer.
In Sec. 1.3.2 it was already mentioned that NaZn leads to acceptor states within the
ZnO. No literature data is available for Ga2O3, but Na, like Cu, does introduce acceptor
states to most n-type oxides. Therefore it is likely that the high concentration of sodium
at the interface leads to a high concentration of acceptor states, which would narrow
94 5 ZnO/CIGSe device and interface analysis
the space charge region to ZnO and lead to the observed increase in the capacitance in
Fig. 5.21b.
It was found, that the concentration of Na at the hetero-interface can be reduced by
depositing a thin layer of Mo prior to the NaF PDT. The Mo layer is supposed to reduce
the diffusion rate of Na into the CIGSe, which leads to the beneficial depth profile, with
a low Na concentration at the hetero-interface while remaining a high Na concentration
in the bulk (Fig. 5.22 with Fig. 5.23). This leads indeed to the best fill factor. Reducing
the NaF rate to 1 nm/min was shown to have a very similar effect.
Thus it can be concluded from the correlation of the depth profiles, the C−Vcurves
and the J−Vcurves, that sodium leads to acceptor states within the GaOxlayer, which
lifts up the conduction band at the interface and therefore introduces an electron barrier
and lowers the extraction of the photo-current.
Potassium was tested as an alternative for sodium. Potassium leads to very good
PCEs in substrate devices, where it is supposed, that the potassium atoms sit on Cu sites
and are replaced by Cd atoms during the CdS deposition [144]. This leads to a n-type
doping of the surface and a reduced interface recombination. This mechanism is unlikely
to occur in superstrate devices though, since the deposition sequence is switched, first Zn
diffuses into the CIGSe and then K is introduced. And indeed, the use of potassium as a
precursor or supplied with a PDT, both lead to similar results as the use of sodium only.
A kink developed in the fourth quadrant, strongly pronounced for the precursor and less
pronounced for the PDT. However, the combination of a 5 minute NaF PDT followed by
a 5 minute KF PDT lead to a PCE of 11%, compared to 9% for the pure NaF treatment
from the same deposition run. A speculative explanation could be, that K atoms replace
Cu atoms close to the hetero-interface, which increase the band gap and reduce the
interface recombination as observed in substrate devices [145]. In other deposition
runs an efficiency of 11% were also achieved for pure NaF treatments. Thus, the KF
treatment appears to be beneficial, but this is not yet sufficiently reproduced and secured.
In summary, The NaF and KF PDT lead to a strong increase in efficiency due an
increased bulk p-type doping and a reduced bulk recombination rate. Both effects pre-
sumably originate from the reduction of ZnCu and InCu states by NaCu or KCu. On the
other hand, the efficiency can be reduced by the alkalis if their concentration at the hetero-
interface is too high. They are argued to induce NaGa or KGa acceptor states within the
GaOxwhich leads to an electron barrier and a reduced fill factor.
5.3 Comparison with buffer free ZnO/CIGSe sub-
strate devices
To get the full picture of the ZnO/CIGSe interface, this section studies substrate devices
without CdS buffer layer. Several publications exist on ZnO/CIGSe/Mo devices with
5.3 Comparison with buffer free ZnO/CIGSe substrate devices 95
(a) (b)
Figure 5.25: a) J−Vcurves and b) C−Vcurve of substrate devices with CdS or i-ZnO
buffer. The i-ZnO was deposited via ALD at different deposition temperatures.
reported efficiencies around 12 % [146] [147]. The influence of the substrate temperature
during the i-ZnO deposition onto the CIGSe is studied here. The i-ZnO is deposited via
ALD and sputtering, the AZO window layer has been sputtered identically for all devices.
Fig. 5.25a shows the J−Vcurves of substrate devices with i-ZnO buffer layers and
one CdS reference device under illumination. The CIGSe absorber were deposited with
a standard three-stage process at 560
and a resulting Cu/(Ga+In) ratio of 0.85 and
Ga(Ga+In) ratio of 0.3. The CdS reference device exhibits the highest PCE of 14.2 %.
The ALD deposited i-ZnO at 120
leads to the second most efficient device, but the FF is
considerably lower compared to the CdS device. If the deposition temperature is increased
to 210
, the VOC starts decreasing from 590 mV to 410 mV. The same result, with only
slightly lower VOC, was obtained from an i-ZnO layer sputtered at 180
. If the deposition
temperature of the ALD i-ZnO is further increased to 300
, the VOC and PCE is reduced
to zero, indicating very high interface recombination currents. The corresponding C−V
curves are shown in Fig. 5.25b for the two devices deposited at 120
and 210
. The CdS
free devices show a strongly increased capacitance at all applied voltage biases compared
to the CdS device.
Discussion The data presented here shows that both deposition methods, ALD and
sputtering, lead to very similar device characteristics, whereas the deposition temperature
has an unexpected drastic influence. The low FF together with the increased capacitance
compared to the CdS device indicate that interfacial acceptor states are again responsible
for the lower FF, similar as it was found for the superstrate devices (Sec. 5.2.5). It was
however, not possible to find a device model which fits both the J−Vand C−Vcurves
sufficiently well. This is likely to be caused by the rough p/n-heterojunction in substrate
devices compared to superstrate devices, which reduces the straight forward correlation
of the C−Vand the J−Vcurves and cannot be represented by a one dimensional device
model any more.
96 5 ZnO/CIGSe device and interface analysis
Figure 5.26: PCE and VOC for ZnO/CIGSe devices in substrate and superstrate configura-
tion displayed versus the maximum temperature exposure of the ZnO/CIGSe interface during
fabrication. For substrate devices this is the ZnO deposition temperature, for superstrate de-
vices the CIGSe deposition temperature. The formation of GaOxin superstrate devices is
assumed to lead to the increase of PCE and VOC until it starts limiting the current at 560
.
Regarding the nature of the acceptor states, it is argued in [148], that the acceptor
states are induced by the VSe-VCu defect complex [48] within the CIGSe. This is unlikely,
since it can’t explain the temperature dependence. It could however however be assumed
that the acceptor states are induced by copper and sodium diffusion into the ZnO layer
or by OSe states at the CIGSe/ZnO interface. It should be noted that the very same
trend was observed in Cu2O/ZnO devices [149], which supports the assumption of Cu
being responsible for the acceptor states within ZnO. This interpretation would lead to a
conclusive model for the CIGSe/ZnO interface, in which the quality of the p/n junction
mainly depends on the temperature load that it experiences during fabrication.
The dependence of the power conversion efficiency on the temperature exposure of
the ZnO/CIGSe interface is shown in Fig. 5.26 for both configurations, substrate and
superstrate. ηand VOC appear to be correlated with the deposition temperature. It seems
a plausible assumption, that the acceptor type defect density from diffusion of copper and
sodium, increases with increasing temperature. The resulting increased interface defect
density reduces ηand VOC due to interface recombination, until, at a certain temperature,
a thin layer of GaOxforms at the interface. The GaOxlayer reduces the electron density
at the interface which reduces the interface recombination, leading again to an increase
of ηfor increasing temperature. At a certain temperature the Cu diffusion into the GaOx
occurs, which again increases the acceptor density at the interface and decreases the
electron collection and with it also η.
In summary, it is proposed that the diffusion of Cu and/or Na into ZnO or GaOxis a
general problem independent of the device configuration. The acceptor states induced by
Cu/Na reduce electron collection and increase interface recombination.
5.4 Summary: ZnO/CIGSe interface 97
5.4 Summary: ZnO/CIGSe interface
1. GaOxformation: At the interface between CIGSe and ZnO a thin layer of an
(Iny,Ga1−y)(Oz,Se1−z)xalloy forms with strong Cu and Zn contaminations. The
best efficiencies were reached for low In and Cu concentrations. The Ga rich alloys
were named GaOxfor simplicity. The In and Cu content was shown to depend on
the CIGSe deposition process. For low In content the process should start with
Ga-Se evaporation and the deposition temperature should be above 500
. For
low Cu content the deposition temperature should be below 550
. The optimum
temperature was found to be 525
.
2. Influence of sodium: Under certain conditions, Na was shown to be able to con-
siderably increase the device efficiency. In case it is present during CIGSe growth it
catalyses the GaOxformation. If supplied by a post-deposition it tends to accumu-
late at the CIGSe/GaOxinterface without further catalysing the GaOxformation.
However, when present in high concentration at the hetero-interface it induces an
electron barrier and leads to the reversible degradation of the device. Very low-rate
NaF post-deposition or diffusion through a thin Mo layer was shown to reduce the
Na concentration at the interface while keeping the concentration high within the
bulk.
3. Influence of Zn diffusion: Zn contamination was found to be proportional to
the thickness of the interfacial oxide layer. A systematic doping study suggests
that for low concentrations below 0.25 % the acceptor state ZnIn/Ga is dominant,
whereas for higher concentrations the ZnCu donor state is dominant. GDOES depth
profiles supports this assumption. The ZnIn/Ga acceptor state appears to be shallow,
whereas the ZnCu donor state leads to a reduction of the charge carrier lifetime. In
the superstrate device fabricated at the optimum temperature the Zn concentration
was estimated to be 0.3 %. This is likely to reduce the device efficiency, especially
at CIGSe deposition temperature above 525
. The electron diffusion length was
measured to be around 800 nm.
Chapter 6
Back contact and degradation
The back contact is the second critical interface in CIGSe devices. In superstrate devices
Au is generally used as the back contact material, which was shown to form an ohmic
contact to CIGSe. But so far no reports exist on the influence of potentially Cu poor
interfaces or on the degradation of the Au/CIGSe interface. Further no highly reflective
and cheap alternative back contact has been reported to date. This chapter will study
these objects as well as the degradation of the p/n-junction, which will be linked to the
electro-migration of sodium and copper in the electric field of the p/n-junction.
6.1 The back contact
The quality of the back contact depends on the difference of the metal and the CIGSe
work function. In substrate devices the formation of MoSexat the interface between
Molybdenum and CIGSe establishes an ohmic contact to CIGSe [150]. In superstrate
devices the back contact is deposited at room temperature on top of the CIGSe layer.
The formation of MoSexdoes not occur and as an alternative to molybdenum, gold is
generally used as the back contact, which has one of the highest work functions (5.3 eV)
of all metals. Still a Cu-poor back surface of the CIGSe could increase the CIGSe work
function leading to a non-ohmic contact.
Cu-poor CIGSe surface The CIGSe growth process, in which only Ga-Se is deposited
during the last stage, may induce a Cu-poor back surface. This would lower the VBM
and increase the work function at the interface to Au (see Sec. 1.3.1). To pronounce
this effect, the last stage of the growth process, during which no copper is offered, was
extended. Instead of 200 nm as in the standard process, 400 nm of Ga-Se are deposited
at the end of the process.
Fig. 6.1 shows the J−Vcurves of a device with a normal third stage and with an
extended third stage, both without external sodium supply. The J−Vcurve of the
standard device shows an increased series resistance starting at 800 mV. This occurs if
the back contact is not perfectly ohmic due to different work functions of the metal and
100 6 Back contact and degradation
(a) (b)
Figure 6.1: a) J−Vcurves of ZnO/CIGSe/Au stacks with a standard 3rd stage, an ex-
tended 3rd stage and an etched surface. The CIGSe is deposited at 525
.b) Corresponding
schematic energy band diagram at a forward bias of 600 mV. The Ga-rich back surface becomes
increasingly Cu-poor the thicker it gets, increasing the hole barrier. The dashed bars indicate
the different back surfaces of the three samples studied here.
the CIGSe. The resulting space charge region results in a slight barrier for the hole
extraction which increases the series resistance, a so-called roll-over.
The J−Vcurves of the device with the extended third stage shows a reduced fill factor
and a stronger roll-over starting from VOC. The strong roll-over observed for this device
can be explained by an increased hole barrier at the back contact due to the Cu-poor back
surface. To prove this, the top 500 nm from the surface was etched by using bromine-
methanol, to remove the Cu-poor surface completely. The J−Vcurve of the etched
sample is also shown in Fig. 6.1. The fill factor remains unchanged, therefore this should
be a bulk or ZnO/CIGSe interface effect. However, the roll-over disappears completely.
This proves that if the surface becomes Cu-poor, its work function increases, and a
roll-over at forward bias develops. If the Cu-poor surface is etched, the back contact
with Au becomes ohmic. The slight roll-over often observed for the standard superstrate
devices therefore indicate a slight Cu deficiency at the back surface.
To measure the barrier between Au and CIGSe a test structure glass/Mo/CIGSe/Au
was fabricated with the CIGSe deposited by the standard modified three-stage process at
525
. As shown in the previous paragraph the surface of the CIGSe absorber is slightly
Cu-poor, which is the origin of the hole barrier. As it can be seen in Fig. 6.2a the J−V
curve shows a roll-over in forward bias and in reverse bias. To quantify the barrier heights
the J−Vcurve was simulated with SCAPS (Fig. 6.2 a). From C−Vmeasurements this
was calculated to be 1e+15 cm−3and therefore the work function within the CIGSe is
5.4 eV. The barrier between Mo and CIGSe, which is the difference in their work function,
is calculated to be 140 meV. From this the work function of MoSexis determined to be
5.26 eV. The barrier at the CIGSe/Au interface is 160 meV. Assuming a maximum work
6.2 Degradation 101
(a) (b)
Figure 6.2: a) J−Vcurves of Mo/CIGSe/Au stack. The CIGSe is deposited at 525
on
top of the Mo layer. The Au is deposited at around 50
. The dotted red line shows the
simulation result from which the work functions of MoSexand the lowering of the VBM at
the back surface, EV,BC, are derived. The work function of Au is assumed to be 5.3 eV b)
Mo/CIGSe/Au stack with CIGSe doped with sodium.
function of 5.3 eV for Au, the lowering of the CIGSe VBM at the interface to Au has to
be at least 60 meV.
Once sodium is supplied to the CIGSe absorber via PDT the p-type doping increases,
which increases the thermionic emission over the barrier at the contacts, leading to an
ohmic contact between CIGSe with MoSexand Au, despite the Cu-poor surface. The
J−Vcurve is shown in Fig. 6.2b. The series resistance derived from this J−Vcurve is
0.12
W
/cm2.
6.2 Degradation
Most CIGSe devices are known to increase their performance on a short time scale due to
light soaking effects [151], which is assumed to change the hole population of the defects
[22], but also degrade on a long time scale due to back contact or TCO degradation [152].
The stability of CIGSe superstrate devices has not been studied so far. In this section the
degradation of the p/n junction and of the Au back contact are studied. It is shown, that
the back contact degradation can be hindered by storage in an inert atmosphere, whereas
the degradation of the p/n junction is independent of the atmosphere during storage.
Degradation of the back contact
Degradation of the back contact can be a problem in the superstrate configuration, since
the back contact is in contact with humidity and oxygen from air, which may lead to
oxidation. Further, the metal may diffuse into the CIGSe absorber. Fig. 6.3 shows the
degradation of the series resistance Rswith and without external sodium supply. Without
102 6 Back contact and degradation
(a) (b)
Figure 6.3: a) Dark J−Vcurves of a Mo/CIGSe/Au stack. The CIGSe is deposited at
525
on top of the Mo layer. The dotted red line shows the J−Vcurve of the same device
after 1 week storage in air. b) same as a), but the CIGSe absorber has received a standard
NaF PDT (see Sec. 2.1).
external sodium supply and freshly prepared devices the contact to Au is ohmic, which is
seen from the linear slope of the J−Vcurve at forward bias. This was previously shown
to be the case as long as the surface is not Cu poor. After storage in air for one week, a
roll-over in forward bias developed, which tells that the contact to Au has degraded. In
case sodium is present, the contacts to Au and Mo are both ohmic for the fresh device,
after 1 week of storage the contact to Au degraded as well. However, the sodium doped
device shows a much stronger increase of the Rsup to a value of 36
W
cm2compared
to 3
W
cm2for the non-doped device. This shows that sodium catalyses the degradation
process.
GDOES measurements were done to reveal if the origin of degradation is due to an
increasingly Cu-poor back surface, possibly induced by electro-migration of Cu. Fig. 6.4a
shows the Cu/(In+Ga) and the Ga/Se ratio of three samples which received the absorber
simultaneously in the same CIGSe deposition run. They were stored in nitrogen until the
NaF PDT was performed. The time between the absorber deposition and the NaF PDT
was different for each samples. After the NaF PDT 100 nm of Au was deposited and they
were stored in air. The GDOES measurements of all three samples were done at the same
day. The time between the GDOES measurement and the NaF treatment was 2 days, 3
days and 2 months.
The Cu/(In+Ga) ratio shows that the surface is indeed copper depleted for all three
samples, but it is independent on the date of the NaF PDT and is therefore not the
cause of degradation. The Ga/Se ratio however increases with increasing time difference
between the PDT and the measurement. As shown in Sec. 5.1, the increased Ga/Se ratio
is a sign for the formation of GaOxat the Au/CIGSe interface. Only the sample which
received the NaF treatment two days before the measurement did not show any sign of
GaOxformation. GaOxhas a large bang gap of 4.8 eV and is therefore assumed to block
6.2 Degradation 103
(a) (b)
Figure 6.4: a) Ratios of the GDOES signals for a CIGSe layer deposited at 525
. The time
between the NaF PDT and the GDOES measurement was varied. b) Depth profile of the Au
concentration. The tail is due to the sputter crater and not due to Au diffusion.
holes very efficiently. The results in Sec. 5.1 showed that sodium acts as a catalyzer for the
oxidation of Ga at high temperatures. At high oxygen, gallium and sodium concentrations,
this seems to occur even at room temperature.
The Au depth profile is shown in Fig. 6.4b. The shape of the profile does not originate
from diffusion but from the sputter crater and surface roughness (see Sec. 2.5.3). No
increased Au concentration within the CIGSe layer can be observed two month after the
Au deposition compared to two days after the Au deposition.
Thus, the cause of the degradation of the series resistance can be found in the formation
of GaOxat the interface between CIGSe and Au. A high concentration of sodium at the
interface catalysis the oxide formation. The deposition of a thicker layer of a metal, which
sticks better to CIGSe than Au, may reduce the degradation of the back contact.
Degradation of the p/n junction
If sodium is supplied via PDT at temperatures around 300
or above, an additional
degradation pathway to the degradation of the back contact can be observed even if the
device is stored in an inert atmosphere. Fig. 6.5a shows the J−Vcurve of a sodium
doped device after 3 days of storage in an inert atmosphere, without illumination and at
open circuit condition. During this time the J−Vcurve did not develop a roll-over, as
observed for the back contact degradation, but a kink in the fourth quadrant. Further it
can be observed, that the capacitance increases at all applied voltages.
The degradation process was found to be reversible, by applying a positive voltage to
the device. Fig. 6.5a shows the J−Vand the C−Vcurves of two solar cells from the
same substrate, but one was stored at open circuit condition and one at forward bias of
+1 V, both in air. The biased solar cell developed no kink. However, the short circuit
104 6 Back contact and degradation
Table 6.1: Quantitative results from XPS measurements shown in atomic % of all elements
within a cleaved CIGSe/ZnO sample, post-treated with NaF, fresh and degraded. The samples
were cleaved at the CIGSe/ZnO interface. Care is to be taken since the relative error of the
atomic concentrations is between 50% and 60%. Still, the trends between the samples are
qualitatively correct. The high concentration of Na and O are due to NaOH formation at the
surface.
ZnO side ZnO side CIGSe side CIGSe side
fresh degraded fresh degraded
Cu 1 1 13 11
In 2 2 24 26
Ga 27 30 12 10
Se 8 7 42 36
O 56 53 4 11
Zn 1 1 0 0
Na 6 5 4 11
current decreased, compared to the fresh device. Interestingly, the capacitance increases
proportional to the loss in photo-current at each voltage. Once this cell is stored at open
circuit condition for a few hours, the kink does develop as well. This indicates that the
degradation process is triggered by the electric field at the p/n junction of the device
(Fig. 1.1 shows the electric field).
XPS measurement were done on identically prepared fresh and degraded samples,
cleaved at the CIGSe/ZnO interface. The results are shown in Tab. 6.1. The chemical
composition of the CIGSe surface and the ZnO surface hardly changes during degradation.
But the concentration of sodium and oxygen actually changes. During degradation over
time, the sodium concentration within the GaOxslightly decreased whereas it increased
considerably in the CIGSe surface. The oxygen concentration follows the trend on both
surfaces. It was shown earlier in Sec. 5.1.4, that the concentration of oxygen at the surface
depends on the concentration of sodium due to the formation of NaOH. But nevertheless,
the change in the sodium concentration indicates, that the sodium likely diffuses during
degradation.
To see whether the applied voltage can have an influence on the Na distribution,
GDOES measurements were performed before and after the application of +1 V forward
bias to the device. The results are shown in Fig. 6.6a. Indeed the Na profile changed
due to the applied voltage, which proves that electro-migration of Na within CIGSe is
possible. The change is only visible at the back contact, where the concentration of Na
is high.
The Na migrates deeper into the absorber during the application of the positive bias,
which creates an electric field at the back contact as seen in Fig. 6.6b. This observed
sodium migration at the back contact is assumed to have no impact on the device, as the
exact doping profile at the back contact is not important for the device performance. But
it shows that sodium in principle does migrate within the CIGSe in the presence of an
6.2 Degradation 105
(a) (b)
(c)
η/ % η%
fresh 3 days
without Na 5.6 5.6
NaF PDT 10.1 4.9
NaF PDT + Mo 10.8 10.8
(d)
Figure 6.5: a) Effect of degradation and forward bias on the J−Vcurves of a sample with a
NaF post-deposition treatment at 300
. The photo-current decreases over time independent
of the storage condition, whereas the kink does not show up if the sample is stored under a
forward bias voltage of +1 V. b) C−Vcurves of the same devices recorded at 293 K and
1 kHz. c) J−Vcurves of a sample with a NaF post-deposition treatment at 300
and a Mo
diffusion barrier, freshly prepared and after three days. . d) Table of the device efficiencies
when freshly prepared and when stored for 3 days in the dark, without voltage bias and in
nitrogen atmosphere.
106 6 Back contact and degradation
(a) (b)
Figure 6.6: a) To prove the sodium electro-migration: Depth profile of sodium within the
same sample before and after applying +1 V. The sample was fabricated by the standard
superstrate CIGSe process with a NaF PDT. The back contact material was gold, which
induces Schottky junction under forward bias. The electric field within this junction leads to
the migration of sodium into the absorber. The degradation of the device probably occurs
due to the sodium migration at the p/n-junction at 0 V. b) band diagram of a standard
Au/CIGSe/i-ZnO/Al:ZnO stack, without interface acceptor states, at +1 V bias.
electric field. At the p/n-junction the concentration is too low to detect small changes in
the concentration by GDOES, but the band diagram in Fig. 6.6b shows that at +1 V the
Na+ions should be pushed into the GaOx.
In summary, Na present at the CIGSe/ZnO interface induces over time an increasing
barrier for electron extraction. During the degradation a diffusion of Na from the GaOxto
the CIGSe was observed from XPS measurements. The degradation process was partly re-
versible by applying an electric field. GDOES depth profiles could prove that Na migrates
within an electric field inside the CIGSe. Under operating conditions the photo-voltage
works against the degradation mechanism. It should be noted that a change in the GaOx
layer thickness could not be detected.
Discussion It was shown that degradation of superstrate devices occur under two con-
ditions, first, if stored in an oxygen containing atmosphere and second, if the device
contains sodium. Regarding the first point, in the presence of oxygen, a thin layer of
GaOxforms between the Au contact and the CIGSe absorber. This is likely to be caused
by cracks in the thin Au back contact, which allow oxygen to diffuse to the Au/CIGSe
interface. The presence of sodium was shown to accelerate this process.
If sodium is also present at the interface between ZnO and CIGSe, the p/n-junction
appears to degrade over time if stored unbiased and in the dark. It was shown that
the charge extraction efficiency was lowered during this process, most likely due to the
formation of an electron barrier at the interface.
In Sec. 5.2, it was argued, that the experimentally observed kink in the J−Vcurve of
6.3 Summary: Back contact and degradation 107
superstrate devices is due to NaGa states within the GaOxlayer. Thus it seems logically,
that the increasing kink during the degradation process is induced by sodium migration
to the interface, which increases the acceptor density within the GaOxlayer.
However, this migration seems to be induced by the electric field of the p/n-junction,
as it does not occur at an applied voltage of +1 V which cancels the electric field of the
p/n-junction. And the electric field of the p/n-junction would push the Na+ions, just
like holes, into the CIGSe away from the GaOxlayer. That raises the question, why does
the kink increase if the source of the kink, which is believed to be Na, diffuses out of the
narrow p/n-junction, which means out of the GaOxlayer.
This could only occur, if the NaGa states are transformed into VGa states within the
GaOxlayer. This would increase the valency of the acceptor state from -2 to -3, thus
increasing the negative charge within the GaOxand with-it the electron barrier.
No literature values for the mobility of Na+ ions directly in amorphous GaOxexists,
but for the Na2O–6Ga2O3system. It shows very high mobilities of Na, comparable with
those of liquids [153]. Assuming again, that the p/n-junction is limited to the GaOx, the
electric field in the dark without voltage bias is about 6 MV/cm within the sodium doped
GaOx, which is close to the estimated breakdown field of 8 MV/cm [154] for Ga2O3, and
strong enough to cause electro-migration of sodium or even copper ions. The migration
of Cu+within CISe due to an electric field was also observed in [155].
Applying a positive bias to the device was shown to hinder the degradation, or even
reverse the degradation. This can be explained by the reduction of the electric field at
the p/n-junction which stops or reverses the sodium migration.
However, the positive bias could not hinder the decrease of the short circuit current
over time together with the capacitance increase measured around zero voltage and at
negative biases. As shown in Sec. 3.2 and as it will be confirmed in the following Sec. 7.1
this can be explained by a p+ layer within the CIGSe at the interface to GaOx. Thus it
is likely that sodium from the CIGSe bulk diffuses to the interface and forms a p+ layer
at the CIGSe interface to GaOx.
To avoid the Na induced degradation mechanism it is therefore important to keep the
Na concentration at the interface as small as possible. Devices which were shown to have
little Na at the GaOxinterface, as the ones with a Mo diffusion barrier deposited prior
to the NaF PDT (Sec. 5.2.4), did not degrade over time. Also, the low-rate deposition
of NaF lead to low Na concentration at the interface and therefore stable and efficient
devices. An example is shown in Fig. 6.5c.
Damp-heat tests of encapsulated devices have not been performed so far, which has to
be done in the future.
6.3 Summary: Back contact and degradation
1. Back contact: A Cu-poor back surface was found to be present in most devices,
which was shown to induce a hole barrier at the back contact, leading to a roll-over
108 6 Back contact and degradation
in the J−Vcurve. Na doping decreases the barrier at the back contact.
Alternative back contact metals were found to be only Pt. The insertion of MoO3-x
allowed the use of Ag, which is highly reflective and more cost effective than Au
or Pt. Device simulations indicate the potential of a CIGSe thickness reduction by
40% due to the high reflectivity of Ag in comparison to Mo.
2. Degradation of the back contact: GaOxformation at the CIGSe/Au interface
was observed after prolonged storage in air. Cracks in the thin Au back contact
may allow oxygen to diffuse to the Au/CIGSe interface. The GaOxlayer introduces
a hole barrier due to the large band gap, which results in a strong roll-over of the
J−Vcurve. The presence of sodium at the CIGSe/Au interface was found to
accelerate this process.
3. Meta-stability/Degradation of the p/n-junction: A reversible degradation
mechanism was found to reduce the fill factor of the J−Vcurves. Based on XPS
and GDOES measurements, this is assumed to be induced by sodium migration out
of the GaOxlayer. The Na migration is triggered by the strong electric field of
the p/n-junction which is confined to the GaOxlayer. NaGa states are transformed
into VGa states, which have a valency of -3, thus increasing the negative charge
within the GaOxand with-it the electron barrier. This effect could explain the
observed dependence of the device performance on the light and voltage bias soaking
previously observed on superstrate devices in the literature [22] [5].
A non-reversible degradation is the reduction of the short-circuit current, which is
also assumed to originate from Na migration, but from the CIGSe bulk to the CIGSe
interface region. Creating an additional p+ layer at the interface. Both effects only
appear for high concentrations of Na at the interface. Low-rate NaF PDT reduces
the sodium concentration at the hetero-interface leading to stable devices.
Chapter 7
Device Modelling
This chapter sets up a 1-D device model to describe the properties of the different su-
perstrate devices observed so far. A device model for a substrate device fabricated under
comparable conditions will be used as a comparison at the end of this chapter. Based on
these models strategies to overcome the limitations will be discussed in Chapt. 8.
7.1 Device model for superstrate solar cells
To verify the speculations of the previous sections and to find solutions for the existing
limitations of the superstrate solar cell, a device model has to be found which can explain
the effects of the temperature dependence, Na doping and the ageing effects. This section
will set up a model for the non-degraded, non NaF treated device fabricated at 525
.
This model will then be used to find the relevant parameters, which can describe the
effects of Na and ageing.
Table 7.1: Source of the simulation parameters used to simulate the J−Vcurve of the
standard device fabricated at 525
without external Na supply.
measured from literature from model
∆EV,BC 0.06 eV χCIGSe 4.5 eV χGaOx4.55 eV
NA,CIGSe 5e+14 cm−3χZnO 4.6 eV ND,i-ZnO 1.58e+19 cm−3
Ln,CIGSe 800 nm ND,ZnO:Al 1e+20 cm−3NA,GaOx1.27e+19 cm−3
Egprofile Fig. 2.3 µn,CIGSe 10 cm2/Vs NA,CIGSeIF 5e+14 cm−3
dGaOx6 nm µhCIGSe 2.5 cm2/Vs NA,GaOx/ZnO 0 cm−2
RS0.7 Ω cm−2αCIGSe [100] Sn,hCIGSe 1.0e+6 cm/s
RP1 kΩ cm−2DOSCB 2.2e+18 cm−3
DOSVB 1.8e+19 cm−3
φAu 5.3 eV
110 7 Device Modelling
Standard device
Tab. 7.1 lists the important parameters used for the SCAPS device model for the standard
device, a non-degraded, non NaF treated device fabricated at 525
. Many parameters
are known from measurements or from the literature. The unknown parameters are listed
in the column ”from model”.
A precise quantitative solution can however not be expected, since the simulation does
not take 2 dimensional effects and grain boundaries into account. Even though the 2
dimensional variations are strongly reduced for superstrate devices compared to substrate
devices, where the roughness of the p/n-junction is determined by the CIGSe surface
roughness, the thickness variation of the GaOxlayer does also introduce a certain error.
Further, small errors in the literature or the experimental values will induce errors to the
fitting parameters. However, such a 1-D model based on many literature values is still a
valid estimation to the reality, if it is possible to find a parameter set, which reproduce
general trends observed in the J−Vcurves and in the C−Vcurves.
To find this parameter set, it is a good strategy to first fit the C−Vprofile with only
the relevant fitting parameters. These are the free charge carrier densities in the GaOxand
ZnO layers plus deep acceptors in the bulk and at the interfaces of the n-type materials
GaOxand ZnO and deep donors in the p-type material CIGSe. Once approximate values
are obtained from the C−Vcurve fitting, they are fixed for the J−Vcurve fitting.
For the J−Vcurve only the interface recombination velocity, Sn,hCIGSe, and the electron
affinity of the GaOxlayers are left as the free fitting parameters. The bulk and interface
defects within CIGSe are set as neutral in the model. The CIGSe layer is separated into
three layers within the model, one fixed bulk layer with NA,CIGSe and two interface layers,
which can differ from the bulk properties due to Cu deficiency and/or increased sodium
content (will be stated when applied).
Modelling C−Vdata To derive the charge distribution within the sample deposited
at the optimum temperature of 525
without external Na supply, the C−Vcurve is
fitted as described above. Fig. 7.1a shows the C−Vcurve with the typical strong capac-
itance increase under forward bias. Simulations assuming any kind of bulk defect in the
CIGSe, distributed homogeneously or close to the hetero-interface, could not reproduce
any qualitatively similar C−Vcurves. However, it is possible to simulate the C–Vcurve
almost perfectly, by assuming an acceptor state density of around 1e+19 cm−3within the
GaOxand setting the shallow donor state concentration to a similar value within the i-
ZnO. This leads to the simulated C−Vcurve shown in Fig. 7.1a. It should be noted, that
acceptor states at the interface between GaOxand ZnO, with a density of around 6e+12
cm−2, can lead to a similar shape of the experimental C−Vcurve. Whereas acceptor
states at the CIGSe/GaOxinterface, like OSe, could not reproduce the C−Vprofile. A
combination of defects at the CIGSe/GaOxinterface and the GaOxbulk could however
qualitatively reproduce the profile. Thus, it can be concluded that the high capacitance
at forward bias originates from acceptor states dominantly located within the GaOxlayer
or at the GaOx/ZnO interface. For the simulation results, the energetic position of the
7.1 Device model for superstrate solar cells 111
(a) (b)
Figure 7.1: a) Experimental and simulated C−Vcurves, recorded at 293 K and 1 kHz, from
a standard superstrate device fabricated at 525
without external Na supply. The simulation
with the best fit assumes a high acceptor state density NA,GaOxwithin the GaOxlayer. b)
Energy band diagram at a positive voltage bias of 200 mV resulting from the fit parameters
listed in Figure a). The inlet shows a magnification of the p/n-junction, where the 6 nm thick
and charge depleted GaOxlayer is marked by the two dashed lines.
acceptor states has to be at least 900 meV below the CBM of GaOxto be below the
quasi-Fermi level for electrons, which ensures permanent occupation. In this model the
acceptor level is arbitrarily set to 1.2 eV below the CBM of GaOx, as shown in Fig. 7.1b.
In this model, the electrons from the ZnO occupy the acceptor states within the GaOx.
If the density of acceptor states is similar to the shallow donor concentration in ZnO, the
SCR of the p/n-junction will be limited to the GaOxand the i-ZnO. If the acceptor density
is lower, the SCR can reach into the CIGSe. However, if the density is higher, the Fermi
level within the GaOxwill be pushed towards the acceptor level, resulting in a lift of the
CBM, which could induce an electron barrier. Both situations are shown in Fig. 7.2b and
in Fig. 7.1b for equal densities and +200 mV. If the acceptor density is equal or larger
compared to the donor density the SCR extends into the CIGSe only at negative voltage
biases. This is the reason for the low capacitance observed only at reverse bias together
with the efficient charge extraction as observed experimentally from the EQE spectra. At
forward bias, the capacitance increases, the SCR is limited to the GaOxand the charge
extraction is reduced.
It should be kept in mind, that the simulation is one dimensional and that e.g. spatial
variations of the GaOxthickness and GaOxdoping inhomogeneities on the nm-scale are
not taken into account. The TEM-EDX mapping image of the GaOxlayer in Fig. 5.7b
in Sec. 5.1.3 shows, that the local thickness varies between 5 - 10 nm. Similar variations
in the doping density can be assumed. Thus, the fitting value for the acceptor density
presented here is an average value, and it has to be assumed, that this value varies locally.
112 7 Device Modelling
(a) (b)
Figure 7.2: a) The dashed red line is the simulated J−Vcurve with the parameters obtained
from the fit to the C−Vcurve shown in Fig. 7.1. The interface barrier is overestimated, likely
due to local variations in the GaOxproperties. The green solid line shows the simulation result
for a reduced acceptor density, which reproduces the experimental curve well (black coarsely
dashed line). b) Energy band diagrams resulting from the two different acceptor densities used
for the simulation in Figure a).
(a) (b)
Figure 7.3: a) The same simulated J−Vcurve as shown in Fig. 7.2a, together with the
interface recombination current (S=1e+6 cm/s) and bulk recombination current (Ln=0.8
µ
m).
The dashed bars indicate the maximum photo-current. b) VOC and FF in dependence of
χGaOx. The model predicts χCIGSe = 4.55 eV for the CIGSe deposition temperature of 525
.
At a lower deposition temperature, the χGaOxis reduced due to the In content within the
GaOx, leading to lower VOC values.
7.1 Device model for superstrate solar cells 113
Modelling J−Vdata To simulate the J−Vcurve only two unknowns are left, since
the charge distribution and thus the doping profile was obtained from the C−Vfit.
As argued above and shown in Tab. 7.1 the electron affinity of GaOx(χGaOx) and the
interface recombination velocity (Sn,h) are still free fit parameters. The resulting best fit
to the experimental J−Vcurve is shown as the red dashed line in Fig. 7.2a. Independent
of the choice of the two free fit parameters, the electron transport barrier at the GaOx
is simply too high to achieve a good fit to the experimental data. But, as stated in the
previous paragraph, the GaOxacceptor density and therefore the barrier height is likely
to vary locally. Such a variation would create preferable electron transport pathways,
similar to point-contacts on the nm scale. If the GaOxacceptor density is lowered from
1.63e+19 cm−3to 1.27e+19 cm−3to take this effect into account, the barrier over the GaOx
is reduced as shown in Fig. 7.2b. Together with χGaOx=4.55 eV and Sn,h=1e+6 cm/s for
the free fit parameter, a very good fit to the experimental J−Vcurve is possible, as shown
in Fig. 7.2a. However this requires a reduction of NA,GaOxby 20 % for the illuminated
J−Vcurve and 30 % for the dark J−Vcurve (not shown). The recombination currents
for the simulated J−Vcurve are shown in Fig. 7.3a, separated into interface and bulk
recombination. The simulated curves are only shown until J= 0 because SCAPS were
running into problems for the non-ohmic back contacts under injection condition where
J > 0.
According to these simulation results, the open circuit voltage and the fill factor
are mainly defined by the bulk recombination rate, which is strongly enhanced under
forward bias due to the extraction barrier induced by the non-depleted acceptor states in
the GaOx. The voltage dependent profile of the bulk recombination therefore correlates
well with the corresponding capacitance profile. The voltage dependent collection length
obtained from the EQE spectra support this result. The short circuit current is reduced
due to bulk and interface recombination. The interface recombination remains about
constant for all voltages, due to the low electron density within the GaOxlayer.
Impact of the electron affinity The electron affinity for the GaOx,χGaOx, was de-
termined to be 4.55 eV, leading to a small cliff at the CIGSe/GaOxinterface. However,
lower CIGSe deposition temperatures increase the In content and with-it χGaOx. Fig. 7.3b
shows the effect of different electron affinities on VOC and the fill factor.
If χGaOxis increased the CBM cliff increases, which increases the electron density
at the interface and therefore the interface recombination current. The limiting factor
for the interface recombination in this case is the hole density within the CIGSe at the
interface. The hole density increases with increasing positive voltage and therefore the
recombination current increases under forward voltage bias as seen in Fig. 7.3b. Thus, for
a cliff type interface between CIGSe and the interfacial oxide layer, the VOC limiting factor
is the interface recombination. Thus, this model can explain the observed significant drop
in VOC for GaOxlayers with increased electron affinity due to increased In content, like
the samples deposited at low temperatures.
114 7 Device Modelling
Table 7.2: Values derived from the C–V curve simulations: acceptor and donor state den-
sities within the GaOx, the CIGSe close to the interface, the CIGSe bulk and ZnO at the
interface. Values derived from the J–Vcurve simulations: Diffusion length within the CIGSe,
Ln,CIGSe, and interface recombination velocity at the CIGSe/GaOxinterface, SIF. It should
be noted, that the values for NAGaOxwere reduced by 10-20% for the illuminated J−V
curve simulations.
sample NAGaOxNACIGSe NACIGSe NDi-ZnOxLnSIF
bulk at interface bulk bulk
cm−3cm−3cm−3cm−3
µ
m cm/s
No Na 1.6e+19 5.0e+14 5.0e+14 2e+19 0.8 1.0e+6
Na @ 300
+ Mo 1.6e+19 9.0e+15 9.0e+15 2e+19 1.6 3.5e+6
Na @ 300
2.3e+19 1.0e+18 5.0e+16 1e+20 1.6 4e+5
- Stored at 0 V 2.8e+19 2.0e+18 5.0e+16 1e+20 1.6 4e+5
- Stored at +1 V 2.3e+19 2.0e+18 5.0e+16 1e+20 1.6 4e+5
Na @ 400
3.7e+19 1.1e+18 5.0e+16 1e+20 1.6 4e+5
In case of conduction band spike at the interface (when χGaOx<4.5 eV) photo-generated
electrons from the CIGSe accumulate at the interface and lead to an enhanced recom-
bination rate at flat band condition around 0 V. At forward bias the accumulation is
lowered due to the lifted conduction band at the interface. This explains the lowered
photo-current observed for devices with an annealed Ga2Se3precursor (see Fig. 5.15b) or
which were deposited at higher temperatures, as the GaOxlayer in these devices have a
decreased electron affinity due to the decreases In content.
Influence of sodium
The device model with the parameters listed in Tab. 7.2 is now used to simulate the C−V
curves of the sodium containing devices.
All experimental C−Vcurves for different sodium treatments can be well reproduced
by adjusting the acceptor densities within the GaOxand the CIGSe, as seen in Fig. 7.4a
and Fig. 7.4c. The adjusted parameters are summarized in Tab. 7.2. In all cases the
NaF PDT increases the charge carrier density in the CIGSe bulk considerably. Whereas
the CIGSe charge carrier density at the interface and the acceptor density within the
GaOxis only increased for the samples without Mo diffusion barrier. And this increase is
more pronounced for the NaF PDT at 400
compared to the PDT at 300
. The trend
of the acceptor densities is similar to the experimentally observed trend of the sodium
concentration at the interface as seen in Fig. 5.12a and 5.22. This shows that the sodium
concentration is directly related to the acceptor density at the hetero-interface. Further,
the device model leads to an increased i-ZnO doping for the samples with an increased
sodium concentration at the interface. This could be due to the catalytic effect of sodium,
which creates more oxygen vacancies within the ZnO close to the hetero-interface or due
to Na interstitials within the ZnO.
7.1 Device model for superstrate solar cells 115
(a) (b)
(c) (d)
(e) (f)
Figure 7.4: Simulated and experimental C−Vand J−Vcurves of CIGSe/ZnO devices
fabricated at 525
. The C−Vsimulations are based on the model of deep acceptor defects
within the GaOxlayer. For the J−Vsimulations the results of the C−Vfits were used, free
fit parameters were the diffusion length and the interface recombination velocity. a) and b)
show the results for different NaF PDT temperatures.c) and d) show the results for the Mo
diffusion barrier. e) and f) show the curves of the degraded samples.
116 7 Device Modelling
The simulated J−Vcurves for the sodium treated devices are shown in Fig. 7.4b
and Fig. 7.4d. The results of the C−Vsimulations were used for the doping and defect
distribution within the device. Due to the above discussed effect of the two-dimensional
variation of the GaOxthickness and doping density, the value were again lowered by
around 10-20 % to achieve a better fit to the experimental J−Vcurves. Further, the
charge carrier density in the interface near CIGSe layer was set to the same value as in
the bulk to enable fitting of the J−Vcurves. The diffusion length and the interface
recombination velocity were free fitting parameters, whereas the GaOxelectron affinity
was fixed at 4.55 eV. The fit results for these two parameters are shown in Tab. 7.2.
According to the fit results, the CIGSe layers, which contains sodium in the bulk, show
an increased electron diffusion length of 1.6
µ
m. Plus, the two samples, which have a high
concentration of sodium at the interface, show a reduced interface recombination velocity.
Leading to the conclusion, that within this model, sodium reduces the recombination rate
within the bulk as well as at the interface.
Degradation
The influence of the degradation under a voltage bias of +1 V on the C−Vcurve is shown
in Fig. 7.4e. It can be well reproduced with the model for the NaF treated device if the
charge carrier density is increased within the CIGSe in the direct vicinity of the hetero-
interface with GaOx. The exact values are given in Tab. 7.2. This creates an additional
p+ layer within the CIGSe, which leads to an additional electron barrier, constant for
all the voltages applied in the experimental range, explaining the reduced short circuit
current. The simulated J−Vcurve with the input data from the C−Vfit gives a very
good fit to the experimental J−Vcurve, the degradation has no influence on the electron
diffusion length, GaOxelectron affinity or interface recombination velocity.
The influence of the degradation without any voltage bias on the C−Vcurve is
also shown in Fig. 7.4e. To reproduce this curve an additional parameter needs to be
adjusted. The acceptor density within the GaOxhas to be increased from 2.3e+19 cm−3
to 2.8e+19 cm−3. This creates a stronger voltage dependent barrier for electron extraction
and leads to the additional decrease of the fill factor. Thus, in both cases the applied
model for the C−Vdata allows very good predictions for the J−Vdata, without any
additional parameter variation.
The reversible degradation process is the change of the acceptor density within the
GaOxand the non-reversible part is the due to the development of a p+ layer within the
CIGSe close to the hetero-interface. Both processes presumably due to Na+migration
(see Sec. 6.2).
7.2 Device model for substrate solar cells
To compare the properties of the superstrate device with them of a standard substrate
device this section sets up a model for a substrate device with CdS as the buffer layer,
7.2 Device model for substrate solar cells 117
(a) (b)
Figure 7.5: a) Experimental and simulated J−Vcurve of a substrate type device under
illumination, shown together with the bulk and interface recombination currents from the sim-
ulation (diffusion length 1.9
µ
m, IF rec. Vel. 2e+4 cm/s, mid-gap acceptor). b) Corresponding
band diagram at 0 V voltage bias.
(a) (b)
Figure 7.6: a) Simulated J−Vcurve from Fig. 7.5 with lowered VBM at the CIGSe surface
due Cu depletion, with increased electron lifetime of the CIGSe and with both effects com-
bined. b) Corresponding band diagram at 0 V voltage bias shows the lowered VBM due to Cu
depletion, potentially induced by the KF treatment.
but otherwise fabricated by identical tools and an equivalent deposition process. The
experimental and simulated J−Vcurve of a typical in house fabricated device is shown
in Fig. 7.5a. The charge carrier density of the CIGSe was obtained from C−Vmeasure-
ments performed at room temperature and 1 kHz. The band gap profile was measured
with GDOES. The properties for CdS, ZnO and Molybdenum were chosen as shown in
Tab. 3.1 and in [100]. The recombination rate is described with two parameters, a neutral
bulk defect and an acceptor type interface defect. Both defects were set as free fitting
parameters. From this an electron diffusion length of 1.9
µ
m (corresponds to a lifetime of
27 ns) and an interface recombination velocity of 2e+4 cm/s is obtained. Fig. 7.5a also
shows the recombination currents in the bulk and at the interface. This shows that within
this device model, the JSC is limited by the bulk recombination and the VOC is limited
118 7 Device Modelling
Table 7.3: Comparison of the solar cell characteristics of a substrate and a superstrate device.
Property unit Substrate Superstrate
NA,CIGSe cm−31e+16 9e+15
Ln
µ
m 1.9 1.6
SIF cm/s 2.0e+4 3.5e+6
RsΩcm−20.4 0.7
RpkΩcm−21.1 1.0
xSCR@500mV nm 230 10
η% 16.3 10.6
by the interface recombination. The efficiency of this device is 16 %. From time resolved
PL measurements charge carrier lifetimes between 10 and 100 ns are generally obtained.
If the lifetime within this device model is adjusted to 100 ns (corresponds to 3.6
µ
m) a
PCE of 18 % can be achieved, which is indeed often achieved experimentally. Therefore
it seems likely that the absorber bulk defects are the origin of the observed variations in
the device performance of in-house substrate devices.
To further increase ηand VOC it is necessary to reduce the interface recombination,
which can be done by changing the buffer chemistry, or by engineering the absorber surface
prior to the buffer deposition. The latter is done for the current record efficiency devices
by an additional KF treatment. The KF PDT depletes the surface from cupper [144]
resulting in a lowered VBM at the interface [43]. This can be modelled by changing the
properties of the first 10 nm of the absorber within the device model. The results are
shown in Fig. 7.6 together with the resulting band alignment at the interface.
7.3 Superstrate vs. Substrate
This section compares a standard substrate device with a superstrate device. The Na
supply for the substrate device was from the soda lime glass and for the superstrate device
from a low-rate NaF PDT, which was shown to lead to record efficiencies of up to 11 %.
Both absorber had a minimum band gap of approximately 1.15 eV. The experimental
J−Vcurves together with the corresponding energy band diagrams (at 500 mV) are
shown in Fig. 7.7. The solar cell characteristics are compared in Tab. 7.3.
The solar cell characteristics in Tab. 7.3 show, that despite the Zn contamination in the
superstrate device, both devices are quite comparable in terms of CIGSe doping, NA,CIGSe,
and electron diffusion length, Ln. This is another indication, that the concentration of
the harmful ZnCu states is reduced in the presence of Na. Further, the series and parallel
resistances, Rsand Rp, of both cells are comparable, with a lower series resistance of the
substrate device due to the metal grid on top of the TCO.
The difference in the efficiency, can be found in the interface recombination velocity,
SIF, and the width of the space charge region, xSCR, at forward bias. The small xSCR for
the superstrate cells is due to the high acceptor density at the GaOxinterfacial layer. This
7.4 Summary: Device modelling 119
(a)
(b) (c)
Figure 7.7: a) J−Vcurve of a substrate and a superstrate solar cell. b-c)) Energy band
diagram of the two devices at 500 mV voltage bias, obtained from the device modelling in
Sec. 7.2 and in Sec. 7.1.
prevents the type inversion observed for the substrate device in Fig. 7.7b. In combination
with the high interface recombination velocity this leads to a lower VOC for the superstrate
device. Further the charged acceptor states in the GaOxlower the fill factor due to a
reduced xSCR.
7.4 Summary: Device modelling
A device model was found which can reproduce the experimental C−Vand J−Vcurves
of ZnO/CIGSe/Au superstrate devices for different Na treatments and different ageing
conditions. The results were compared to the device model for a standard substrate
device. In the following a summary of the simulation results is given:
1. Acceptor states at the hetero-interface: The observed low J−Vcurve
fill factor of superstrate devices was successfully correlated to interfacial acceptor
120 7 Device Modelling
states. The model confirms the assumption of deep acceptor states (like CuGa)
within the GaOxlayer and/or at the GaOx/ZnO interface. Acceptor states at the
CIGSe/GaOxinterface (like OSe) may be present but could not describe the C−V
profile sufficiently well.
2. Influence of Na: The influence of Na was correlated with an increase of the
charge carrier density in the CIGSe bulk, together with an increase of the density of
deep acceptor states at the hetero-junction. The optimum NaF PDT temperature
is therefore determined by the optimum balance between these two effects. Further,
Na reduces the recombination rate in the CIGSe bulk and at the hetero-interface
within this model.
3. CB alignment: The device model predicts a conduction band cliff of 50 meV
between CIGSe and the interfacial GaOxlayer. Still, the presence of the GaOx
layer reduces the interface recombination, since it lowers the electron density at the
hetero-interface. However, the cliff increases with increasing In content within the
interfacial oxide layer, which was shown to decrease the VOC due to a high interface
recombination velocity. Potential buffer layers leading to a CBM spike are predicted
to further increase the electron barrier and with it reducing the fill factor.
4. Origin of degradation: The simulation of the C−Vdata revealed that the re-
versible part of the degradation (meta-stability) is due to an increase of the acceptor
density within the GaOxlayer. The non-reversible part is due to an increase of the
acceptor density within the CIGSe at the interface. This is in accordance with
the model presented in Sec. 6.2, that the reversible part is due to Na migration in
the p/n-junction and the non-reversible part is due to Na migration outside of the
p/n-junction.
5. Comparison to substrate devices: The comparison suggests, that the electron
diffusion length is in NaF treated superstrate absorber is not strongly reduced due to
the Zn contamination. But the high interface recombination velocity in combination
with the high acceptor density at the hetero-interface reduces the FF and the VOC
of the superstrate device compared to the substrate device.
Chapter 8
Strategies for efficiency improvement
In the course of this thesis, the CIGSe/ZnO system has been optimized to a power
conversion efficiency of up to 11 %, mainly by experimental optimization of the CIGSe
deposition routine and the sodium supply. Here, the device model which was set up in
the previous chapter, will be used to evaluate which of the device parameters inhibits
the highest potential for further efficiency improvement. A proposition to change this
parameter will be followed up experimentally by applying new oxide buffer layers.
8.1 Parameter evaluation
The first obvious parameter to examine within the model is the minority carrier dif-
fusion length. The diffusion length in the cells studied here are eventually limited by
the Zn contamination, even though Na doping seems to lower this effect. Further, the
diffusion length could be improved by reducing the background impurity level during the
CIGSe fabrication. However, the simulations show, that an increase of the electron diffu-
sion length from 1.6
µ
m to 3.2
µ
m leads to an efficiency increase of only 0.5 %, from 11.0 %
to 11.5 %.
Another way to increase the efficiency is to reduce the interface recombination
losses. Within the model, this can be achieved most efficiently, by lowering the VBM
by 100 meV within the 10 nm of the CIGSe next to the hetero-interface (compare with
Sec. 3.1). An increase of ηto 13.3 % is predicted by the model. As shown in Fig. 1.5b
this can be achieved by reducing the Cu concentration at the hetero-interface. Substrate
devices exhibit the advantage of a Cu-poor surface induced by a surface reconstruction
[156] or induced by post treatments [144]. In superstrate devices it is difficult to engineer
a Cu poor interface due to the high mobility of Cu within CIGSe. However, optimisation
of a KF PDT may enable a Cu depletion at the hetero-interface in the future. It may
also be possible to exploit the high mobility and deplete the interface from Cu by electro-
migration [157] at negative biases. However, the cells studied here showed improvement
during positive biasing, due to the electro-migration of sodium. Another option to reduce
the VBM is the exchange of Se atoms locally with S atoms. In Fig. 8.1 and in more detail
122 8 Strategies for efficiency improvement
Table 8.1: Pathway to 20 % efficiency, by optimizing certain parameters within the device
model.
Steps Varied parameter from to efficiency
0 none - - 11.0
1 el. diff. length, LL,n 1.6
µ
m 3.2
µ
m 11.5
2 VBM offset, ∆EV,S 0 eV 0.1 eV 13.3
3 Acceptor density NA,GaOx1.2e+19 cm−32e+18 cm−315.5
4n-type doping, ND,GaOx1e+17 cm−32e+19 cm−317.1
1, 2 combined 15.2
2, 4 combined 18.4
1, 4 combined 19.3
1, 2, 3 combined 19.3
1, 2, 4 combined 20.2
1, 2, 3, 4 combined 20.3
in Sec. 10.2 it is shown, that it is indeed possible to confine the sulphur atoms close to the
hetero-interface by the deposition of an In2S3precursor prior to the CIGSe deposition.
However, the increased sulphur content also causes an increased CBM of the chalcopyrite,
which, in the presence of interfacial acceptor states, further increases the electron barrier
and therefore lowers the device efficiency.
The combined effect of an increased diffusion length and a reduced VBM (by Cu
depletion) at the hetero-interface would lead to a PCE of 15.2 %.
The simulations presented in Chapter 7 point out though, that the main limitation
of the superstrate solar cell devices originates from the high density of acceptor type
defects within the interfacial oxide layer. Reducing the acceptor defect density by one
order of magnitude, to 1.2e+18 cm−3, leads to a simulated efficiency of 15.5 %. However,
this would require a reduced diffusion of Cu and Na into the GaOxlayer. An alternative
would be to enhance the n-type doping of the interfacial oxide layer to 2e+19 cm−3,
which would increase the efficiency of 17.1 %.
An overview of the different efficiency improvements due to the different optimization
processes and their combinations are listed in Tab. 8.1. It should be noted, that in case a
buffer layer is found which can suppress the effect of the interfacial acceptor states (step
4 in Tab. 8.1), it will be easy to reduce the interface recombination (step 2) by applying a
sulphur gradient, leading to an efficiency of 18.4 %. Combined with an increased diffusion
length (3.2
µ
m) an efficiency of 20.2 % could be achieved, despite the high density of
acceptor defects within the buffer layer.
8.2 Definition of the ideal buffer layer 123
(a) (b)
Figure 8.1: a) J−Vcurve of devices with differently thick In2Se3precursors and a CIGSe
layer deposited by the modified 3-stage process at 475
.b) The corresponding GDOES depth
profile shows the confined S gradient and that Ga prefers to accumulate at the S-rich regions.
A solution to overcome this problem is presented in Sec. 10.2.
8.2 Definition of the ideal buffer layer
Thus, to achieve efficiencies above 20 %, a new material has to be found, which remains
its n-type character even in the presence of copper and sodium. Further it also has to
block the diffusion into the ZnO layer underneath. The additional criteria are high band
gap, suitable electron affinity and chemical resistance against the formation of GaOx
during the deposition of CIGSe.
Criteria for an ideal buffer layer material in a superstrate CIGSe device:
1. n-type despite Cu and Na doping at the interface,
2. ND>1e+18 cm−3,
3. Electron affinity, χ, between 4.25 eV and 4.50 eV,
4. Band gap above 3 eV,
5. Amorphous, or better lattice matched to CIGSe ,
6. High formation enthalpy, non-reactive with CIGSe and ZnO at 560
,
7. Diffusion barrier for Cu and Na.
Finding such a thermally stable and wide band gap material, in which Cu and Na do
not introduce acceptor states is difficult. They require a formation enthalpy larger than
the formation enthalpy of Ga2O3. Otherwise Cu-contaminated Ga2O3will form at the
interface to CIGSe, which was shown to limit the device efficiency. Further, an amorphous
124 8 Strategies for efficiency improvement
structure is generally preferred, since poly-crystalline layers allow Cu diffusion along the
grain boundaries. Thus, all oxides which crystallize above 400
, which is the minimum
CIGSe deposition temperature, are unlikely to perform well. To perform as a better
diffusion barrier than amorphous Ga2O3, the ionic radii of the cations should be smaller
compared to Ga3+, like Si2+ or Al3+. However, materials like amorphous SiOxor AlOx
have too large conduction band offsets to CIGSe [134]. 3 nm thick SiOxlayers have been
tested and didn’t allow current transport. AlOxhas an even smaller electron affinity. The
same for MnO.
However, a possible candidate for a material which is highly stable and a good diffusion
barrier, would be amorphous Ga2O3itself. It was shown in this work that GaOxforms
intrinsically at the CIGSe/TCO interface and that Cu and Na do introduce acceptor
states within it. But it is not clear, which defect density will be induced by Cu and Na if
Ga2O3is deposited purposefully prior to a CIGSe deposition at low temperatures. It may
be possible to compensate the acceptor states by intentional n-type doping of the Ga2O3
with Sn or Si [158] on the cation site or Cl and F [142] on the anion site. Charge carrier
densities as high as 1e+19 cm−3for Si [159] and 1.44e+19 cm−3for Sn [160] as a dopant
have been reported for crystalline β-Ga2O3. This would be sufficient to compensate
the acceptor states induced by Cu and Na. The electron affinity of crystalline β-Ga2O3
is reported to be 4 eV [143], which would require alloying Ga2O3with oxides of lower
electron affinity, like In2O3or SnO2. Unfortunately, no reports on the electron affinity
or on the doping of amorphous Ga2O3exists. Therefore this will be studied in depth in
Sec. 10.1. However, recently amorphous Ga2O3was successfully applied as a buffer layer
to CIGSe [?] and Cu2O [161] solar cells in substrate configuration and in CdTe solar cells
in superstrate configuration [162].
The question remains, whether or not the purposefully deposited Ga2O3layer will
have the suitable electron affinity and whether the Cu diffusion remains a problem. The
following sections will try to answer these questions.
8.3 Combinatorial material exploration
By knowing the desired properties of the optimum buffer layer material in CIGSe super-
strate devices, it would be ideal to predict the perfect material by quantum mechanical
computations [163]. However, the time frame of this work did not allow this approach.
An alternative option is to start with a good guess for a suitable material, like Ga2O3,
and tweak its properties to these desired by alloying with other materials. In Sec. 8 amor-
phous Ga2O3was identified as a base material, which will be alloyed with other elements
whose conduction band is made of sorbitals, like Zn, Sn or In. The spherical symmetry
of these orbitals is beneficial, as it makes the delocalized electronic transport less sensitive
to structural disorder compared to por dorbitals [164].
A very useful strategy to do this sufficiently fast is to produce graded samples, which
allow fast screening of the fabricated alloys. This can be achieved with combinatorial
pulsed laser deposition (PLD, see Sec. 2.2), in which the co-deposition of different materi-
8.3 Combinatorial material exploration 125
als, from spatially separated sources, allows to create very defined compositional gradients
onto the substrate.
Experiment
Based on the considerations given in the beginning of this section, the base material was
chosen to be Ga2O3, which is alloyed with TiO2, ZnO, SnO2, In2O3and ZnSnO3to change
the electron affinity and charge carrier density. The alloys were deposited on a 5 cm x 5 cm
large alkali-free glass substrate covered with GZO. Generally the layers were deposited
with two gradients, the overall layer thickness in X-direction and the Ga2O3content in
Y-direction. The deposition temperature was set to 100
to achieve amorphous layers,
and set to 500
to achieve crystalline films. The oxygen partial pressure was set to 1.3e-
5 mbar for the depositions at 500
and increased to 6.5e-3 mbar for the depositions at
100
to ensure the formation of transparent films. The doping with Sn and Ti is aiming
to increase the charge carrier density and compensate potential acceptor states induced by
Cu and Na during the subsequent CIGSe deposition. Additional materials with suitable
electron affinities for CIGSe, like ZnMgO:Ga, SrTiO3and TiO2:Nb, were prepared by PLD
on GZO substrates. All films were prepared together with the group ”Process Technology
and Advanced Concepts” at the National Renewable Energy Laboratory (NREL).
The CIGSe layer were deposited by M. Contreas at NREL. The deposition process was
a different three-stage process from the one explained in Sec. 2.1, due to different PVD
system specifications. In the first stage evaporation of In-Ga-Se was set with a constant
rate, followed by a Cu-In-Ga-Se at a constant rate, followed by In-Ga-Se in the 3rd stage.
The temperature was set to 450
. An SEM image of the absorber can be seen in Fig. 10.8.
The process was not optimized and lead to poor absorber quality.
The back contacts were realized by thermal evaporation of 64 Au pixels (3x3 mm)
distributed over the 5x5 cm CIGSe coated substrate. This allowed a mapping of the
electrical properties of the graded compositions.
Results
The performance results of the solar cells with the different oxides as the buffer layer are
shown in the table in Fig. 8.2. It should be noted, that only the best performing solar
cell of the 64 solar cells on each sample is shown. The results originate from two CIGSe
depositions, which lead to different absorber qualities, resulting in different values for
the JSC but identical VOC values (due to Fermi-level pinning at the interface, will be
discussed later). For both deposition runs, 10 nm thick amorphous Ga2O3deposited at
100
, leads to highest efficiencies. The following paragraph will shortly describe the
performance results of the different alloys.
Similar as in the work of Nakada et al. [5] the pure i-ZnO leads to non-rectifying devices
for the Cu/(In+Ga) ratio of about 0.85 used in this experiment. This was observed
reproducibly on 128 solar cells on two substrates and not only for i-ZnO, but also for
126 8 Strategies for efficiency improvement
(a) (b)
Material Gradient max. VOC max. PCE
ZnMgO:Ga Ga,Mg 0 0
i-ZnO d 0 0
ZnO:Ga - 0 0
SrTiO3d 0 0
TiO2:Nb Nb, d 60 0.06
(Zn,Ga)OxGa, d 0 0
(Sn,Ga)OxGa, d 500 0.02
(Zn,Sn,Ga)Ox(am) Zn, Sn 580 0.6∗
(Ti,Ga)OxGa, d 500 0.9∗
(Zn,Sn,Ga)Ox(cr) Zn, Sn 450 1.2∗
(In,Ga)Ox:Sn Ga, d 420 2.5
Ga2ZnO4Ga, d 500 2.6
Ga2O3(am) d 500 3.5
Ga2O3:Sn (am) Sn, d 500 2.9
Ga2O3:Ti (am) Ti, d 580 1.6
Ga2O3(cr) d 660 0.4∗
(c)
Figure 8.2: J−Vcurves of the best result of the 64 solar cells on each sample with a new
buffer layer. CIGSe absorbers from a) the first CIGSe deposition run b) the second CIGSe
deposition run. c) Overview of the J−Vcurve parameters. The first group are Ga2O3-free
layers, the second group are Ga2O3alloy layers, the last group are pure Ga2O3layers.
* The efficiency values marked by the star are around two times lower compared to the others,
due to a qualitatively worse CIGSe layer. However, the VOC values are comparable.
8.4 Amorphous Ga2O3buffer 127
ZnMgO:Ga, ZnO:Ga and (Zn,Ga)Ox. Thus, it is unlikely, that shunting causes the non-
rectifying behaviour. A possible explanation could be a tunnelling contact between these
oxide layers and the CIGSe. Based on the device model described in Sec. 7.1, this requires
a high electron density in the TCO and high acceptor density at the TCO/CIGSe interface.
The samples with buffer layers based on Ga2O3rich alloys do show a rectifying be-
haviour, and exhibit a VOC larger than zero. Ga2O3is known to reduce the charge carrier
density and reduce the electron affinity within alloys, which makes a tunnelling contact
to CIGSe more unlikely. The highest VOC is achieved by the crystalline Ga2O3layer,
which suffers from a poor JSC though. The crystalline Ga2O3and also the crystalline
ZnSnGaOxfilms lead to poor adhesion of the CIGSe layer onto the substrate.
Alloying Ga2O3with Ti leads to J−Vcurves similar to those with pure Ga2O3, but
exhibit a slightly lower short circuit current. The same occurs when doping Ga2O3with
Ti in the concentration range below 1 at.%. The optical measurements, showed that
alloying with Ti at 200
does not effect the band gap value, despite the fact, that the
band gap for pure TiO2is 3.2 eV. Pure TiO2samples lead to non-rectifying J−Vcurves.
Alloying Ga2O3with Zn leads to non-rectifying devices. Interestingly this is not the case
for the buffer layer deposited from a single Ga2ZnO4target, which leads to very similar
J−Vcurves as the pure Ga2O3layers, again with a slightly lower JSC. Alloying Ga2O3
with Sn leads to a strong electron barrier not allowing any current transport within the
studied voltage range. Doping with Sn, in the concentration range below 1 at.% leads
to a reduction of the photo-current. Alloying Ga2O3with Sn and Zn also reduced the
photo-current and for higher Ga concentration (above 50 at.%) a strong electron barrier
develops. Similar behaviour was observed for the amorphous and crystalline ZnSnGaOx
layers. Alloying Ga2O3with In leads to a reduction of the VOC while the JSC remains
the same as for the pure Ga2O3. The details are shown in Fig. 8.4d.
In summary, amorphous and undoped Ga2O3was found to perform best as a buffer
layer compared to the other oxides testes in this experiment. Doping and alloying Ga2O3
with Ti or Sn lead to lower JSC values. Alloying with In leads to lower VOC values. All
materials not based on Ga2O3lead to non-rectifying devices.
The positive result is, that the efficiency of 3.5 % was achieved without sodium addition
and without optimisation of the CIGSe deposition process. The next section will therefore
analyse the properties of the Ga2O3layer and the corresponding solar cells to understand
the limitations.
8.4 Amorphous Ga2O3buffer
To understand the limitations of the ZnO/a-Ga2O3/CIGSe device presented in the previ-
ous section it is necessary to study electronic properties of the amorphous Ga2O3. This
will be done in detail in Sec. 10.1 and some of the obtained results are shown here. Fig. 8.3
shows the charge carrier density, ND,GaOx, and the electron affinity, χGaOx, of a-Ga2O3
in dependence of the a-Ga2O3deposition temperature. The charge carrier density was
128 8 Strategies for efficiency improvement
(a) (b)
Figure 8.3: a) Charge carrier density within the Ga2O3layers, measured with C−Vtechnique
at 1 kHz and 298 K, b) Comparison of the electron affinities of CIGSe and ZnO (both taken
from literature) with the assumed electron affinities of the different Ga2O3layers. Based on the
literature value for the crystalline phase (4 eV) and the assumption that the decreasing optical
band gap with decreasing deposition temperature is due to an increasing electron affinity (see
Sec. 10.1). The error of determining χis estimated to be +/−100 meV. Best conduction band
alignment is achieved at a Ga2O3deposition temperature between 355 and 405
.
measured by C−Vat 1 kHz, therefore it has to be interpreted as the maximum charge
carrier density, since at this frequency deeper defects can contribute to the capacitance
which do not contribute to the free carrier density.
For deposition temperatures below 320
both properties, charge carrier den-
sity and electron affinity, are independent of the deposition temperature, with
ND,GaOx=2e+17 cm−3and χGaOx=4.7 eV. Interestingly both properties change once the
deposition temperature becomes larger than 320
,ND,GaOxincreases and χGaOxdecreases
(Fig. 8.3). As will be discussed in depth in Sec. 10.1, this is attributed to an oxygen va-
cancy defect band present in the low temperature deposited Ga2O3films. At 405
the
values are ND,GaOx=5e+17 cm−3and χGaOx=4.4 eV. This comes close to the optimum
properties for a buffer layer in CIGSe superstrate devices.
Fig. 8.4a show the effect of the Ga2O3deposition temperature on the device perfor-
mance. The device with the Ga2O3layer deposited at a temperatures of 360
exhibits
an increased series resistance and a strongly reduced photo-current compared to the de-
vices with Ga2O3layer deposited at lower temperatures. This indicates a barrier at the
hetero-interface, likely induced by the reduced χGa2O3. However, according to the results
shown in Fig. 8.3b χGa2O3should not differ more than 300 meV from χCIGSe, which would
be required to induce an electron barrier. The reason for the electron barrier induced by a
8.4 Amorphous Ga2O3buffer 129
(a) (b)
(c) (d)
(e) (f)
Figure 8.4: a) J−Vcurves of CIGSe superstrate devices with a-Ga2O3buffer layers deposited
at varying temperatures and 1.3e-3 mbar O2partial pressure. b) J−Vcurves for different
thickness but constant temperature of 200
.c) Simulated J−Vcurves for different χGa2O3
and different position of the acceptor states with the density NA.d) Experimental C−Vcurve.
e) Optical band gap of a-(GayIn1−y)2O3buffer layers with varying y value deposited at 100
and 1.3e-3 mbar O2partial pressure. f) J−Vcurves of CIGSe devices with a-(GayIn1−y)2O3
buffer layers and varying y value.
130 8 Strategies for efficiency improvement
small change of χGa2O3requires a high density of charged acceptor states at the interface,
similar as already observed for the CIGSe/ZnO system. The C−Vprofile is a flat curve
at a high capacitance value of around 100 nF for all applied voltages (shown in Fig. 8.4d).
This is typical for Fermi level pinning at the CIGSe/TCO interface as shown in Fig. 3.3b.
A fit to the experimental J−Vcurves indeed requires a high acceptor density at the
CIGSe/Ga2O3interface. The simulation results are shown in Fig. 8.4c. Slight changes
of χGa2O3can well reproduce the experimentally observed variations, when assuming a
high acceptor density at the CIGSe/Ga2O3interface (NA=7e+12 cm−2, 0.35 eV above
VBMCIGSe,σn,h=1e-15 cm2).
Another interesting observation is the dependence on the layer thickness. Fig. 8.4b
shows the J−Vcurves of the Ga2O3layer with varying thickness values. The 10 nm
thick layer leads to the best results, thicker layer increase the series resistance due to
the low conductivity of the Ga2O3layer. Surprisingly, layers thinner than 10 nm lead
to a strong electron barrier, which becomes noticeable from the s-shaped J−Vcurves.
Also the C−Vcurves (not shown) show a strong increase of the capacitance at forward
bias, which indicate that the s-shape is due to acceptor states at the Ga2O3interface.
The asymmetry of s-shape can indeed be simulated by the model of acceptor states at
the buffer/TCO interface discussed in Sec. 3.2 and shown in Fig. 3.3d and Fig. 3.4f. The
simulation results in Fig. 8.4c are based on acceptor states at the Ga2O3/ZnO interface,
with NA=3e+12 cm−2, 1 eV below CBMZnO,σn,h=1e-19 cm2.
To fully control χGa2O3it is interesting to alloy Ga2O3with In2O3. Fig. 8.4d shows
that the band gap of the amorphous (In,Ga)2O3decreases almost linearly from 4 eV to
3 eV with increasing In content. According to the anion rule, the VBM remains constant
for similar oxides independent of the cations, thus the electron affinity can be increased
by 1 eV due to alloying Ga2O3with In2O3. This is confirmed by the reduction of the VOC
with increasing In content (Fig. 8.4e ), which tells increased interface recombination due
to an increased CBM cliff.
8.5 Summary of and Outlook for improvement
strategies
The parameter evaluation of the superstrate devices studied in this thesis shows that the
research priority should be on novel buffer layers, which reduce the acceptor defect density
at the hetero-interface. This section tested new oxide buffer layers, which were chosen by
considerations based on their electron affinity and chemical stability. A summary of the
obtained results is given here:
1. Combinatorial material exploration: Different oxide materials, many of them
based on Ga2O3alloys, were tested as the buffer layer in CIGSe devices. Amor-
phous Ga2O3led to the best device performance. Doping and alloying attempts did
not improve the performance. This indicates, that additional mid-gap states are
introduced by the dopants, resulting in an increased interface recombination veloc-
8.5 Summary of and Outlook for improvement strategies 131
ity. New experiments at elevated temperatures could help to improve the doping
efficiency.
2. Amorphous Ga2O3:It was shown that the electron affinity and the charge carrier
density of amorphous Ga2O3can be adjusted in the range of 700 meV by the depo-
sition temperature and by alloying with In2O3. In principle this allows a material
design, which comes close to the criteria for an optimal buffer layer.
3. Acceptor states: Despite the well suited properties of amorphous Ga2O3, device
simulations suggest, that the devices utilizing amorphous Ga2O3as the buffer layer,
suffer from acceptor type defects positioned mainly at the CIGSe/Ga2O3interface.
For a Ga2O3layer thickness below 10 nm, the device model suggests acceptor states
at the Ga2O3/ZnO interface. This would indicate the Cu diffusion to the ZnO
interface for thin Ga2O3layers.
4. Sulphur gradient: It was shown, that an In2S3precursor can be used to introduce
a sharp sulphur gradient at the hetero-interface. However, as long as the acceptor
density at the hetero-interface is not compensated, an additional band gap increases
the electron barrier at the hetero-interface further. Nevertheless, such a well defined
sulphur gradient may allow highly efficient devices once a suitable new buffer layer
is found.
Outlook: Two options are suggested to overcome the limitations by future research
work. First, compensating the acceptor states by increasing the n-type doping of the
amorphous Ga2O3to above 1e+19 cm−3, or secondly, to find a material which is tolerant
to Cu and a Cu diffusion barrier at the same time.
The first option is to sufficiently increase the n-type doping in amorphous Ga2O3.
One way to do this is to increase the deposition temperature to 425
, as it was shown
in Fig. 8.3a. At this temperature the layer is still amorphous but the n-type doping is
increased. At the same time this decreases the electron affinity to around 4.2 eV, which
can be shifted back to 4.5 eV by alloying Ga2O3with In2O3to (In0.3Ga0.7)2O3. The
higher In content in combination with the increased deposition temperature leads to a
better dopability of the oxide. Instead of using Sn as the dopant, Si should lead to
more shallow donor states which would allow more efficient n-type doping [159]. Thus
amorphous (In0.5Ga0.5)2O3:Si deposited at 425
has the potential to reach charge carrier
densities of 2e+19 cm−3and is proposed as a buffer layer material for new experiments.
However, a general problem of oxides is, that the valence band is formed by O 2p
orbitals, which are energetically close to the Cu 3dorbitals, leading to a hybridization of
the O 2pand the Cu 3dorbitals to binding and non-binding states. The binding states lie
deep within the band gap, where they form a band structure as in the p-type CuGaO2,
or, if Cu is present as an impurity, they induce a compensation of the n-type doping.
Thus it may be necessary to switch from oxides to nitrides. Within integrated circuits
the use of nitrides as TiN, TaN or TiZrN as a Cu diffusion barrier is standard [165].
TiN, TaN and GaN have suitable work functions [166] and can become n-type doped by
132 8 Strategies for efficiency improvement
nitrogen vacancies. In case of GaN, the formation of GaOxwas reported to occur only at
temperature above 540 [167]. Thus, this could be a successful future research direction.
In summary, the outlook of superstrate CIGSe devices depends on whether or not
a suitable buffer layer can be found. However, if this is possible, the ability to produce
very thin CIGSe absorber with a steep S and Ga gradient at the front and back interface
respectively, should lead to highly efficient devices. The following design of the future
CIGSe solar cell in superstrate configuration is proposed:
Figure 8.5: Energy band diagram and schematic illustration of the proposed ideal super-
strate solar cell. A Ga gradient at the back contact reduces back contact recombination, S
gradient at the front reduces interface recombination. A thin CIGSe absorption layer reduces
the In consumption. The buffer layer is adjusted by the In/Ga ratio to the optimal CBM
alignment. Annealed ZnO and MoO3-x/Cu or MoO3-x/Ag back contact lead to low electrical
and optical losses. The simulated PCE is 21 % for an assumed electron diffusion length of
5m, compensated interfacial acceptor states and a high interface recombination velocity of
1e+6 cm/s.
Chapter 9
Summary and Conclusion
The growth of CIGSe absorbers onto TCOs is studied to achieve efficient CIGSe solar
cells in the superstrate configuration. It is desired to reveal the factors which are limiting
the device efficiency till date. In this work interface analysis by XPS and GDOES
measurements, combined with device analysis by numerical simulations were used to
establish this correlation for CIGSe layers co-evaporated onto ZnO layers. The results
indicate that a general limiting factor of the superstrate devices is the Cu diffusion into
the oxide materials.
Thermodynamic calculations predict the formation of Ga2O3at the CIGSe/ZnO in-
terface during the deposition of CIGSe onto ZnO or any other suitable oxides. Further,
during the high temperature deposition process, Cu is shown to diffuse from the CIGSe
layer into the oxide layer. From literature, it is known that Cu impurities induce acceptor
states within ZnO and most other oxide materials. In this work, general device simula-
tions have shown that these states if present at the hetero-junction, reduce the current
collection efficiency and that the devices are very sensitive on the electron affinity of the
oxide facing the CIGSe.
For the detailed interface analysis, ZnO was chosen as the oxide since it led to the
best performing devices owing to its slightly larger electron affinity compared to CIGSe.
The analysis shows that the composition of the interfacial layer between ZnO and CIGSe
depends critically on the CIGSe deposition process. At low deposition temperatures, a
3 nm thin (Cu,In,Ga)(O,Se)xalloy forms. The Cu content induces acceptor state, the In
content increases the electron affinity which reduces the VOC. The In and Cu content can
be reduced by increasing the deposition temperature, leading to a 6 nm thick GaOxlayer,
which increases the VOC and the device efficiency. However, the current collection is still
limited by an extraction barrier for electrons caused by Cu induced acceptor states. At
temperatures above 525
, the Cu diffusion into the GaOxincreases, further reducing the
current collection efficiency.
The formation of the interfacial oxide layer causes Zn diffusion into the CIGSe absorber.
It is shown that the effect of the Zn contamination is detrimental for a concentration above
0.25 %, which is the case for devices fabricated at 525
. The electron diffusion length
134 9 Summary and Conclusion
within the Zn contaminated devices is measured to be smaller compared to substrate
devices. However, Na doping counteracts the negative effect of Zn contamination and
increases the diffusion length to a value comparable to those in substrate devices.
Na was also found to catalyse the GaOxformation if supplied as a precursor to the
CIGSe growth. If it is supplied by a post-deposition it accumulates at the CIGSe/GaOx
interface, without further catalysing the GaOxformation. However, it induces additional
acceptor states, which further reduce the current collection efficiency if present at high
densities. The presence of Na at the hetero-junction further induces a reversible degrada-
tion of the device, which reduces the fill factor of the J−Vcurves. Based on XPS and
GDOES measurements, this is assumed to originate from Na migration out of the GaOx
layer. The Na migration is triggered by the strong electric field of the p/n-junction which
is confined to the GaOxlayer. This could explain the need for voltage/light soaking of
the previously reported superstrate devices.
Low-rate NaF post-deposition or diffusion through a thin Mo layer reduce the Na
concentration at the interface while keeping the concentration in the bulk sufficiently
high. This leads to stable efficiencies of up to 11 %, observed for the first time without the
need of light or voltage biasing. Efficiencies above 10 % were also achieved by employing
a MoO3-x/Ag back contact, which allows the substitution of the expensive Au contact.
The back contact was shown to be slightly non-ohmic due to the Cu-poor back surface in
superstrate devices. However, the presence of Na reduces its detrimental effect.
A 1-D model of the full device allowed a comparison of the solar cell parameters with a
standard substrate device. The highest potential for efficiency improvement was found to
be the reduction of the interfacial acceptor density, followed by a reduction of the interface
recombination velocity.
To find a new suitable oxide buffer layer, which can solve these issues, a combinatorial
material deposition approach was used to alloy Ga2O3with other oxides to find the
material with the optimum electron affinity, doping density and chemical stability. The
best results were obtained with amorphous Ga2O3as the buffer layer material, whose
electron affinity was found to be comparable to the one of CIGSe. However, interfacial
acceptor states were found to be present at the interface between CIGSe and GaOxand
the n-type doping density of the Ga2O3layer was not sufficient to compensate these.
To overcome this problem, future research should focus on new buffer layers, as for
example the development of an amorphous Ga2O3layers with doping densities above
1e+19 cm−3. First experiments presented in this thesis indicate that this can be achieved
by increasing the Ga2O3deposition temperature, alloying it with In2O3and doping it
with Si. Alternatively, the material system for the buffer layer should switch from oxides
to nitrides, as Cu may be less detrimental within nitrides.
Once a suitable new buffer layer is found, the application of a well defined sulphur
gradient, as shown in this work, should reduce the interface recombination to achieve
highly efficient devices. The speculated cost benefits of the superstrate configuration
were shown to be realistic, as the ZnO annealing will allow 30 % thinner ZnO:Al layers
while increasing the photo-current by 1.5 mA/cm2and the highly reflective back contact
135
will allow a CIGSe thickness reduction of 40 %. These factors could lead to a considerable
reduction of material costs, electricity costs and process time of CIGSe modules.
Chapter 10
Additional Information
In this Chapter a more in depth study of the amorphous Ga2O3, the sulphur gradient,
the ZnO annealing, the MoO3−xback contact and the CIGSe thickness reduction will be
given. Beyond this, light scattering from etched ZnO layers and the CIGSe growth on
ZnO will be addressed. Finally the paper on white light reflectometry (WLR) is shown.
10.1 Amorphous Ga2O3characterization
Amorphous Ga2O3lead to the highest device efficiency of the buffer layers tested within
the combinatorial material exploration. This is especially remarkable, since the electron
affinity for crystalline Ga2O3is given in the literature to be around 4 eV [?], which is
not suitable for CIGSe devices. Thus, in this section, the energy levels of the conduction
and valence bands are approximated via a combination of transmission/reflection (T/R)
spectroscopy, XPS and J−Vmeasurements. The oxygen content in the sample was
measured by XPS. Further, Au/Ga2O3/GZO Schottky devices were fabricated to measure
the charge carrier density and the Schottky barrier height between Au and Ga2O3. The
Ga2O3layers were deposited by PLD on substrates which exhibited a temperature gradi-
ent. The oxygen partial pressure was set to 2e-5 mbar for the samples with a temperature
gradient from 425
until 300
and to 1.3e-3 mbar for the samples with a temperature
gradient from 160
until 100
. Thickness of the deposited layers were 10 nm on GZO
substrates for the XPS measurements, 200 nm thick layers on GZO substrates for the
Schottky devices and 200 nm thick layers on Quartz substrates for the T/R measurements.
Results
Band diagram Fig. 10.1a shows the square of the absorption coefficient of Ga2O3layers,
obtained from the T/R measurements. For direct band gaps, the value of the band gap
Egcan be extracted from a linear extrapolation of α2to zero [168]. It is found, that the
band gap correlates strongly with the substrate temperature, at 425
the band gap is
4.5 eV and at 320
around 4 eV. For temperature below 320
the band gap remains at
138 10 Additional Information
4 eV, for temperatures above 450
, the Ga2O3crystallizes and the band gap becomes
4.7 eV, which is the literature value for crystalline Ga2O3[169].
Fig. 10.1c shows the valence band XPS spectrum for a sample deposited at 425
and
a sample deposited at 100
. The onset, at the binding energy of around 4 eV, originates
from electrons energetically located at the VBM. Thus, the binding energy at the onset
gives the distance between the valence band maximum and the Fermi level EF. For the
425
sample this difference is 4.5 eV and for the 100
sample it is 4.15 eV. These values
are very similar to the optical band gap of 4.5 eV and 4.0 eV, respectively, which tells,
within the uncertainty of the measurement, that the EFis located close to the CBM. The
GaOxmust therefore have a high degree of n-type doping.
This is indeed confirmed by the charge carrier density extracted from C−Vmeasure-
ments at 1 kHz and shown in Fig. 10.2a. For substrate temperatures around 300
and
below, the measured charge carrier density within the Ga2O3is around 3e+17 cm−3. It
increases with increasing temperature until 1e+18 cm−3at 425
. The difference between
the Fermi level and the CBM can be calculated from the measured charge carrier density,
which is 20 meV for 1e+18 cm−3and 50 meV for 3e+17 cm−3.
The small discrepancy observed between the optical band gap and the XPS data for
VBM-EFcan be explained by the surface sensitivity from the XPS measurement compared
to the T/R measurement. XPS probes only the first few nanometres and the band gap
at the surface is likely to differ from the band gap in the bulk. The general trend of the
VBM-EFvalue is however the same as for Eg, as can be seen in Fig. 10.1d. Both values
start increasing at substrate temperatures above 320
.
If the band gap change is due to a change of the electron affinity, which is the energetic
distance of the CBM to the vacuum level, the work function will change accordingly, which
is the difference between the Fermi level and the vacuum level, and with-it the Schottky
barrier between Ga2O3and Au will change. And indeed, for Ga2O3layers deposited at an
increased temperature the J−Vcurves of the Au Schottky devices in Fig. 10.1b show a
strong shift of the injection current to larger voltages, which has to be due to an increased
Schottky barrier. Thus it can be concluded that the change in the band gap is due to a
change of the electron affinity.
The highest reported charge carrier density for the crystalline β-Ga2O3is also
1e+18 cm−3. To check whether the observed change in the free carrier density is cor-
related to the oxygen vacancies, XPS measurements were performed and the ratio of the
Ga 2p and the O 1s peak intensities are plotted against the deposition temperature in
Fig. 10.2a. Two plateaus can be observed, one at lower oxygen concentration for the lower
temperatures and one at higher oxygen concentration for the higher temperatures. The
transition temperature is 320
, which is also the temperature where the band gap, Fermi
level and charge carrier density start to rise. Fig. 10.2b correlates band gap and the charge
carrier density with the oxygen to gallium ratio. It can be approximately described with
an exponential function.
10.1 Amorphous Ga2O3characterization 139
(a) (b)
(c) (d)
(e)
Figure 10.1: a) Square of the absorption coefficient of Ga2O3layers deposited at different
temperatures. The linear extrapolation until α2=0 gives the optical band gap, Eg.b) J−V
curves of Au/Ga2O3Schottky devices, with the Ga2O3layers deposited at different temper-
atures c) XPS spectra of two Ga2O3layers. The VBM onset (Ip-EF) varies with different
deposition temperatures. d) Ip-EFand Egplotted over the deposition temperature. Both
follow the same trend, showing that mainly the conduction increases with increasing deposi-
tion temperature. e) Charge carrier density within the Ga2O3layers measured with C−V
technique at 1 kHz and 298 K.
140 10 Additional Information
(a) (b)
Figure 10.2: a) Relative oxygen concentration between the samples given by the ratio of the
XPS intensities of Ga 2p and O 1s peaks. b) Correlation of the O to Ga ration with the band
gap and the charge carrier density.
Sn doping To compensate the potentially present acceptor states created by the Cu
and Na diffusion during the CIGSe deposition it may be necessary to further n-type dope
the Ga2O3. The best reported results with a charge carrier density of 1.44e+19 cm−3
have been achieved by doping with Sn [160]. No reports exist so far on doping amor-
phous Ga2O3. Fig. 10.3 shows the measured charge carrier density, extracted from C−V
measurements at 1 kHz, plotted over the Sn concentration. The measurements were per-
formed on a sample deposited at 100
with a Sn Gradient varying from 0.2 % until
0.9 %. The Sn concentration was estimated from the ratio of the SnO2to Ga2O3ablation
pulses, weighted with the deposition thickness per pulse for SnO2and Ga2O3. The mea-
sured charge carrier density increases almost linear with increasing Sn concentration from
2e+17 cm−3as measured for undoped Ga2O3until 1e+18 cm−3at a Sn concentration of
0.6 %. For Sn concentrations above 0.6 % the charge carrier density drops again, possibly
due to the formation of SnO2phases.
Figure 10.3: Charge carrier density within the Ga2O3layers measured with C−Vtechnique
at 1 kHz and 298 K.
10.1 Amorphous Ga2O3characterization 141
Discussion
(a) (b)
Figure 10.4: a) Energy band diagram for Ga2O3layers deposited at substrate temperatures
of 450
and 100
. Values for Ip,Egand EFwere obtained from XPS, T/R and C−V
measurements, respectively. The error of determining χis estimated to be +/−100 meV.
b) Comparison of the electron affinities of CIGSe and ZnO with the electron affinities of the
different Ga2O3layers. Best conduction band alignment is achieved at a Ga2O3deposition
temperature around 400
.
From the optical measurements it is clear that the band gap increases with increasing
deposition temperature and the Schottky devices have shown that this shift is due to a
decreased electron affinity. The increase of the electron affinity with increasing deposition
temperature goes along with an increase of the charge carrier density. Both properties
are correlated with the oxygen to gallium ratio, shown in Fig. 10.2b. It appears, that
the band gap and the charge carrier density increases exponentially with the oxygen to
gallium ratio until a certain ratio for the crystalline films. It is likely, that this ratio
equals the point of stoichiometry, which is O/Ga=1.5. From this it would follow, that
the band gap drops exponentially with the oxygen deficiency until it reaches a plateau for
O/Ga=1.42. Thus 3 % oxygen deficiency would be sufficient to reduce the band gap to
4 eV. This could be induced by a oxygen vacancy defect band. Theoretical calculations
have estimated the VOdonor state to be around 1 eV below the conduction band [142].
Thus, it is likely that a defect band created by VOdonor states decreases the band gap
by 700 meV compared to the stoichiometric crystalline film. Similar as it was reported
for MoO3non-stoichiometric films [170].
From literature it is known, that the electron affinity of crystalline Ga2O3lies at around
4 eV with the same optical band gap as observed in this study [?]. As it was found here,
Ga2O3layers deposited at lower temperatures exhibit smaller band gaps together with
142 10 Additional Information
smaller Schottky barriers to Au. From this correlation, it has to be assumed, that the
band gap reduction is due to an increase of the electron affinity. The band diagram for the
crystalline layer (from literature) and for the low temperature deposited amorphous layer
is shown in Fig. 10.4a. This band diagram is valid under the assumption, that the band
gap reduction is solely due to an electron affinity reduction. Fig. 10.4b shows the electron
affinities for different temperatures in comparison to the literature value for CIGSe and
ZnO.
The shift of the optical band gap of oxygen poor amorphous films were observed once
before in [171], although from 5 eV to 4.8 eV. The films were deposited by electron-beam
evaporation. In crystalline films, substoichiometry was also identified as the origin for
strong sub-band gap absorption leading to opaque films [160] . This was argued to be due
to the formation of Ga2O phases. The presence of Ga2O was identified by a shift of the
O 1s XPS peak together with a strong sub-band gap absorption, leading to opaque films.
Increasing the temperature to above 800
was argued to lead to the evaporation of the
Ga2O phase, since it has a low melting point of Ga2O of 650
. If the oxygen partial
pressure is set to values below 1 mbar at low temperatures around 100
, opaque films
were observed as well.
Further, the charge carrier density derived from the C-V analysis showed an increase
towards stoichiometry. Simulations with SCAPS show that donor defects below 700 meV
do not contribute to the measured capacitance. Thus, the increase of the charge carrier
density must originate from a different source. Conductivity in Ga2O3is commonly
explained by background impurities like hydrogen or silicon [142]. Doping the oxygen
poor and amorphous Ga2O3with Sn lead to an increase in a maximum increase of the free
charge carrier density up to 1e+18 cm−3for a Sn concentration of around 0.6 at.%. This
is the same value as obtained for the stoichiometric undoped films prepared at higher
deposition temperatures. It cannot be excluded, that Sn was present as a background
impurity for the undoped films. However, the strong correlation of the charge carrier
density, the band gap and the oxygen concentration indicate a certain mechanism,
independent of the extrinsic dopants, responsible for these changes. Further studies need
to be performed to fully understand this process.
Anyhow, the change of the electron affinity with the deposition temperature is
important for the application in solar cells, since the difference of the electron affinities
at hetero-interface should be small to allow current transport, while keeping the interface
recombination low. Fig. 10.4b compares the electron affinities of CIGSe and ZnO with
them of the different Ga2O3layers. The deposition temperature should be set around
400
to achieve optimum band alignment with CIGSe. These results show that
amorphous Ga2O3does not necessarily induces a charge transport barrier due to a CBM
spike at the interface to CIGSe devices. How the electron affinity will be influenced by
the CIGSe deposition is not know, however, the best fit to the experimental J−Vcurves
in Sec. 8.4 is achieved for χGa2O3=4.5 eV for the low temperature deposited (T<320
)
Ga2O3. It is marked in Fig. 10.4b as a red bar. The value fits very well to the results
obtained here for the pure Ga2O3layers deposited by PLD.
10.2 Sulphur gradient 143
The properties of the GaOxforming during the CIGSe deposition on ZnO are difficult
to compare to the results given here, since the formation occurs completely different from
the formation during a PLD deposition. Also the In impurities reduce the electron affinity.
The device model used for the J−Vcurve fitting in Sec. 7, lead to a value of 4.55 eV for
the electron affinity.
10.2 Sulphur gradient
This section will explore the possibility of reducing the interface recombination by a local
sulphur enrichment at the hetero-interface.
It was previously shown that low temperature CIGSe depositions lead to samples that
are limited by interface recombination. As discussed in Sec. 3.1 one strategy to reduce the
interface recombination losses is by increasing the band gap of the CIGSe at the hetero-
interface, e.g. by sulphurization. The effect of the sulphurization on the energy band
diagram is shown in Fig. 1.5b, the CBM is increased while the VBM is reduced. However,
the device modelling in Sec. 7 indicated that in the current state-of-the-art superstrate
solar cells the application of a sulphur enrichment at the hetero-interface cannot lead to
an increased efficiency since the reduced CBM will not be sufficiently compensated by
the band bending of the space charge region, leading to an increased electron barrier and
increased bulk recombination. Thus, the successful application of a Cu(In,Ga)(S,Se) layer
at the hetero-interface requires an inverted CIGSSe at the p/n-junction, requiring new
effective n-type buffer layers. Still, in order to analyse the potential of superstrate devices,
it is important to know whether or not it is possible to create a local sulphur gradient at
the hetero-interface.
Here, it is tested if an In2S3precursor layer, deposited onto the i-ZnO/ZnO:Al sub-
strate prior to the CIGSe deposition, can achieve a sharp sulphur gradient. The In2S3is
deposited at 200
by thermal evaporation from an In2S3powder. The layer thickness
were 10 nm and 70 nm. The CIGSe absorber is deposited at 475
by the modified three
stage process as described in Sec. 2.1. The resulting J−Vcurves are shown in Fig. 10.5a.
As expected from the device model, the In2S3buffer layer lead to an increased elec-
tron barrier, reducing the short circuit current and the series resistance, if the sulphur
remains at the interface. Both effects are more pronounced for the thicker In2S3precur-
sor. However, this indicates that the sulphur atoms remain localized at the interface.
Fig. 10.5b shows the GDOES depth profile. Two effects can be observed indeed. First,
sulphur atoms diffuse only slightly into the CIGSe absorber, leading to a very effective
sulphur gradient directly at the hetero-junction. Second, Ga accumulates at the sulphur
rich areas. While the first effect is the desired effect, the second could lead to situation,
where the CBM is increased too much due to the increases presence of S and Ga at the
hetero-interface.
Too reduce this effect, a new CIGSe deposition approach is tested. Followed by the
optional 50 nm of In2S3precursor deposition, 900 nm of In2Se3are deposited. Onto this
144 10 Additional Information
(a) (b)
Figure 10.5: a) J−Vcurve of devices with differently thick In2Se3precursors and a CIGSe
layer deposited by the modified 3-stage process at 475
.b) The corresponding GDOES depth
profile.
stack, Cu-Ga-Se is deposited at 420
. Due to the different diffusion constants of Cu and
Ga, this leads to a depth profile as shown in Fig. 10.6b. While Cu is homogeneously dis-
tributed throughout the absorber, the Ga concentration at the hetero-interface is reduced
to almost zero. By increasing the deposition temperature the Ga concentration could be
increased as desired. The low temperature deposition leads to a low power conversion
efficiency of 0.7 % for the sample without the In2S3precursor and 1.1 % for the sample
with the precursor. In this case, the sulphur gradient still induces an electron barrier, but
also reduces the recombination losses, which leads to a higher VOC. The generally poor
performance is due to the missing GaOxinterfacial oxide layer.
Discussion It is expected that a sulphur gradient at the hetero-interface will reduce
the interface recombination for well behaving devices with a inverted CIGSe layer close to
the hetero-interface. New buffer layer will hopefully allow this in the future. As expected
from the device model for the state-of-the-art superstrate solar cell, an enhanced sulphur
concentration at the interface leads to an increased electron barrier and a decreased de-
vice efficiency. But, nevertheless, it was shown that it is possible to design a local sulphur
gradient at the hetero-interface by applying a In2S3precursor. The strong Ga accumu-
lation due to the sulphur was reduced by a new deposition routine which allows good
control over the Ga concentration at the sulphur rich interface. This is important, since
the accumulation of S and Ga at the hetero-junction will induce an electron barrier.
10.3 Light management
As discussed in the first chapter, the big advantage of the superstrate structure is the
advantageous light management due to the possibility to engineer the back contact, im-
10.3 Light management 145
(a) (b)
Figure 10.6: a) J−Vcurve of devices with and without an In2Se3precursors and a CIGSe
layer deposited by In-Se followed by Cu-Ga-Se at 420
.b) The corresponding GDOES depth
profile.
plement light scattering by ZnO etching and improving the ZnO transparency by annealing
prior to the absorber deposition. This Chapter studies these three issues.
10.3.1 ZnO annealing
Annealing is supposed to improve the electrical and optical properties of ZnO. The best
results reported in the literature so far, were achieved by capping the film with a-Si or
SiOxduring annealing in a nitrogen atmosphere [17]. The question for the application in
CIGSe superstrate devices is, whether vacuum annealing is sufficient and what happens
to the ZnO during the deposition of CIGSe at high temperatures without any capping
layer. To give an answer to this, first, the effect of vacuum annealing is studied, then the
influence of the CIGSe deposition and then the applicability of the cap annealed ZnO.
Vacuum annealed ZnO
To evaluate the effect of vacuum annealing, as deposited ZnO:Ga layers were annealed
at 250
and at 500
for 10 minutes in ultra high vacuum. The ZnO films were RF
sputtered at 50
from a Ga:ZnO target with 2 at.% Ga. The film thickness was 1200 nm.
The mobility and free charge carrier density was measured by the Hall technique and also
extracted from the optical transmission and reflection data, by fitting the spectra with an
empirical model based on a Drude approach. The fitting procedure was developed from
Pflug et al. within the RigVM environment [123]. The electrical and optical properties
of the as-deposited and the annealed films are shown in Tab. 10.1.
Results The conductivity increases due to the annealing and with increasing annealing
temperature the conductivity increases further. The results of the Hall measurements and
the optical measurements show that the mobility increases together with the conductivity
146 10 Additional Information
whereas the free charge carrier density slightly drops. The optical analysis also shows a
slightly decreasing density of free charge carriers with increasing annealing temperature.
Similar results were obtained by annealing in an oxygen atmosphere of 1.3e-4 mbar.
Only at an oxygen pressure of 1.7e-3 mbar the conductivity decreased due to a strong
reduction in free carrier density.
In addition to the observed increase in mobility a strong increase in the transparency
of the ZnO in the region between the band gap absorption and the free carrier absorption
can be observed. The optical band gap increases from 3.51 to 3.57 during the annealing.
The average transparency in the range of 400-1000 nm is increased from 77 % to 87 %.
For a typical band gap for CIGSe of 1.2 eV, the gain in photo-current is calculated to be
4 mA/cm2due to the increased transparency after the annealing.
The XRD pattern of the two films are shown in Fig. 10.8a. No change of the structural
properties can be identified from the XRD pattern, which only shows the (002) peak. The
(002) growth direction is preferentially orientated perpendicular to the glass substrate, as
can be seen in the SEM image in Fig. 10.8b. Using the formula of Scherrer (Eq. 2.23 [97]),
the domain size D can be calculated to 22 nm. Whereas an average grain diameter of
around 60 nm can be estimated from the SEM picture in Fig. 10.8b.
Discussion The beneficial effect of vacuum annealing on ZnO was observed earlier in
the literature. It was argued that the origin of the increased conductivity is due to an an
increase in charge carrier density from an increased oxygen vacancies density [172] or due
to the activation of the Ga dopants [173]. The annealing effect described in both works
did not come along with a reduced absorption, only a blue shift of the absorption edge
due to the Burstein-Moss effect. This can be ruled out in the here presented case, since
the charge carrier density does not increase during annealing.
An increase of mobility and transparency was observed by Yu et al. [174]. They argued,
on the basis of temperature dependent Hall-effect measurements, that the mobility in as-
deposited film is limited by grain boundary scattering, whereas the mobility in vacuum
annealed sample was limited by thermal lattice vibration scattering. They observed an
Table 10.1: Electrical and optical properties of 1.2
µ
m thick Ga:ZnO samples prepared via
RF sputtering on sodium free glass (Eagle 2000). The transparency is given for the ZnO/air
interface and not corrected for the reflectivity. It is given for the wavelength range of 400-
1000 nm. The charge carrier density and the mobility is also calculated from the T and R
spectra with the drude model.
property Rsq GT neµnneµn
unit
W
S/cm %cm−3cm2(V s)−1cm−3cm2(V s)−1
method 4 point v-d-Pauw T/R Hall from Hall T/R T/R
as deposited 5.7 1590 77 6.25e+20 15.9 5e+20 16.7
250
anneal 3.9 2322 84 5.56e+20 26.1
500
anneal 3.3 2746 87 5.96e+20 28.8 4e+20 21.5
10.3 Light management 147
(a) (b)
Figure 10.7: a) Transmittance corrected for the reflection, T/(1 −R), of the as deposited
and the annealed ZnO layers on glass. The AM1.5g photon flux spectrum is shown as well.
The typical band gap of CIGSe is marked by a dashed line. The gain in transparency leads to
a gain in photocurrent of 4 mA/cm2b) Square of the absorption coefficient spectrum of the as
deposited and the annealed ZnO layer. The optical band gaps are approximated to be 3.51 eV
and 3.57 eV by a linear extrapolation until α2= 0.
(a) (b)
Figure 10.8: a) XRD pattern of the as deposited and the annealed ZnO layers on glass. Only
the (002) peak is visible within the recorded angle range of 2θ= 20 −50
°
. No change of the
pattern occurs due to the annealing. b) SEM image of the annealed ZnO covered with CIGSe.
148 10 Additional Information
increase in the dominant (002) XRD peak due to the annealing. An increase of the domain
size is not observed in the here presented work, thus grain boundary scattering may not
be the origin of the here presented annealing effect. Even though the amount of charged
defects at the grain boundaries may change during the annealing, which could improve
the current transport along the grains. Other reports like [175] [176] observed all of the
effects also observed in this work, but did not give any explanation for the observations.
Ruske et al. observed an increased mobility and reduced sub-band gap absorption during
the annealing of capped ZnO layers inside a nitrogen atmosphere [17]. They showed that
the sub-band gap absorption can be described with Urbach tails, which becomes reduced
during the annealing [177]. It is assumed, that this is due to the reduction of structural
disorder, partially caused by the dopant atoms.
In the layers studied here, the increased conductivity after the annealing is solely due to
a higher mobility of the free charge carriers. This explains the reduced absorption in the
infrared region, since the improved charge carrier mobility narrows the plasma absorption
band. The annealing effect also comes along with a reduction of the light absorption
close to the band gap and in the visible light range. This may be due to a reduction of
the absorption from defect states. The fact, that the mobility increase was measured by
Hall and by the optical technique further indicate, that the origin is a bulk effect, as the
reduction of structural disorder as argued by Sch¨onau et al.
In summary, the work presented here, an increased mobility in combination with re-
duced sub-band gap absorption is observed after the vacuum annealing. It is likely that
the origin is the reduction of structural disorder induced by the Ga and Al dopants. The
structural disorder could induce a strong tailing of the valence band, which leads to an
increased sub band gap absorption and due to the scattering at the charged defects to an
reduced mobility for the as-deposited films. The annealing leads to a better structural
order which reduces these two effects.
ZnO annealing during CIGSe growth
The CIGSe deposition onto the ZnO layer has a similar annealing effect as the above
described vacuum annealing. The sub band gap absorption of the ZnO decreases and
the mobility increases, as shown in Fig. 10.9. The mobility increase at a constant charge
carrier density was derived from a fitting procedure based on the Drude approach, and
is therefore induced by a bulk and not grain boundary effect. Similar as for the vacuum
annealing this is likely to be induced by a reduction of structural disorder. The average
light absorption between 400 nm and 1100 nm is reduced from 10.7 % to 6.5 % which yields
a photo-current increase of 1.5 mA/cm2.
However, if sodium can diffuse into the ZnO during the CIGSe growth, the positive
effect of the annealing cannot be observed. Fig. 10.12 shows the transmission and the
photo-luminescence before and after the CIGSe deposition. Both sub-band gap absorption
and photo-luminescence increase during the CIGSe deposition. This may be induced by
sodium diffusion from the glass into the ZnO, forming the optical active defect NaZn [178]
[179].
10.3 Light management 149
(a)
std. Rsq neµn
ZnO
W
/sq cm−3cm2(V s)−1
before 8.0 4e+20 45
after 5.3 3.7e+20 60
(b)
Figure 10.9: a) Transmission curves, corrected for the reflectance, of standard ZnO on
sodium free substrate before and after the CIGSe deposition process at 525
. The corrected
transmission is also shown for the cap annealed ZnO for comparison. b) Free charge carrier
density and electron mobility of the standard ZnO samples extracted from T and R spectra
using a fitting procedure based on a Drude approach, developed from Pflug et al. [123]
Non-vacuum annealed ZnO
Annealing outside of the vacuum usually leads to a degraded charge carrier density. How-
ever, in [17] it is reported, that the annealing step can be performed in a nitrogen at-
mosphere, if a thin capping layer of a-Si or SiOxis deposited onto the ZnO:Al prior to
the annealing in order to prevent the oxygen chemisorption. Following to the annealing
step the capping layer is etched from the ZnO surface, resulting in ZnO:Al layers with
high mobilities of up to 67 cm2/Vs. This procedure is already successfully exploited in
a-Si:H/
µ
c-Si:H solar cells [180], which improved the efficiency by an absolute gain of 0.7 %.
In this section it is studied, whether or not such cap annealed ZnO layers are suitable for
CIGSe superstrate devices.
The J−Vcurve of the best performing cap annealed ZnO layers in CIGSe superstrate
devices fabricated at the standard temperature of 525
is shown in Fig. 10.10a. Com-
pared to the standard as deposited ZnO, the VOC is reduced by 150 mV. It is important
to note that the cap annealed ZnO is deposited on sodium free Corning Eagle XG glass,
whereas the standard ZnO is deposited on sodium containing soda lime glass. This leads
to an accumulation of sodium at the hetero-interface as it is seen in the GDOES depth
profiles shown in Fig. 10.11. Further, the Ga/Se ratio in Fig. 10.11 reveals, that at the
interface between the CIGSe and the cap annealed sodium free ZnO less GaOxforms
compared to the standard ZnO. In Sec. 5.1.4 it was shown, that the sodium catalyses the
GaOxformation, which in turn was shown in Sec. 5.3 to reduce the interface recombina-
tion. This explains the reduced VOC, which is solely due to the absence of sodium in the
glass substrate.
The photo-current at negative voltage bias is increased by 4 mA/cm2compared to
the standard ZnO with sodium. As it is seen in Fig. 10.12a, this is due to the higher
150 10 Additional Information
(a) (b)
Figure 10.10: J−Vcurves of CIGSe/ZnO devices with standard and annealed ZnO and a)
CIGSe layer deposited at 525
b) CIGSe layer deposited at 560
.
(a)
Figure 10.11: GDOES depth profile of the Ga/Se ratio and the uncalibrated sodium signal.
The Ga/Se ratio increases at the interface, indicating the formation of GaOx. The presence of
sodium catalyses the interfacial oxide formation.
(a) (b)
Figure 10.12: a) Transmission curves and b) PL spectra of standard and annealed ZnO
before and after the CIGSe deposition process at 525
. The peak at 650 nm originates from
the second harmonic
10.3 Light management 151
transparency of the cap annealed substrate. The lower sub-band gap absorption of the
cap annealed ZnO layer is likely to be due to reduced defect absorption as discussed in
the previous chapter for vacuum annealed ZnO. To check this, the PL spectrum of both
ZnO layers were recorded and displayed in Fig. 10.12b. The cap annealed ZnO has a
pronounced band to band PL peak with minor defect induced PL intensity in the sub-
band gap region. The PL spectrum of the standard ZnO shows a strong contribution of
the defect PL peaks in the sub-band gap region with only little band to band radiative
recombination. Thus it is likely that the lower sub-band gap absorptivity of the cap
annealed film originate from a reduced defect concentration. As discussed in the previous
chapter, this is believed to originate from structural disorder possibly induced by the
dopants.
By cleaving the sample with the technique described in Sec. 2.5.2 it was possible to
measure the change in transmission and PL spectra due to the deposition of a CIGSe
layer at 525
. The results are shown in Fig. 10.12a and b. The cap annealed ZnO
shows an increased band to band PL signal, which indicates further improved structural
order due to the high temperature CIGSe deposition. The infrared absorption is slightly
reduced, indicating, that the charge carrier density become reduced during the CIGSe
deposition, similar as for the vacuum annealed sample in the previous section. The sub-
band gap absorption and sub-band gap PL intensity is slightly increased after the CIGSe
deposition, which shows that the CIGSe deposition has a negative influence on the ZnO
layer. However, the average light absorption between 400 nm and 1100 nm is still reduced
from 6.5 % to 2.1 % which yields a photo-current increase of 1.6 mA/cm2compared to the
sodium free ZnO annealed during the CIGSe deposition.
Summary
ZnO annealing in vacuum was shown to increase the electron mobility and decrease the
defect absorption in ZnO layers. The increase in mobility was shown not to originate from
structural changes or reduced grain boundary scattering, but from the reduced defect con-
centrations in the ZnO bulk. The same effect is on observed due to the CIGSe deposition
onto as-deposited ZnO layers. However, the best ZnO properties were obtained from the
cap annealed ZnO, even though the CIGSe deposition slightly deteriorates these. The
vacuum annealing has the advantage, that it does not require any additional processing
step, as the ZnO gets annealed during the CIGSe deposition. If sodium diffused from the
glass into the ZnO layer, the annealing effect was not beneficial, but the defect absorption
actually increased.
10.3.2 Highly reflective back contact
A major advantage of the superstrate configuration is the possibility to design the
back contact in order to increase the reflectivity and decrease the recombination losses.
This allows reducing the absorber thickness, which would increase the production speed
and reduce the material costs. Possibly, it could even increase the efficiency, due to
152 10 Additional Information
(a) (b)
Figure 10.13: a) Calculated reflection at the interface between CIGSe and MoSex/Mo,
MoO3-x/Ag and Au. b) Simulated EQE spectra for a 500 nm thick CIGSe device with different
back contacts and without recombination in the CIGSe bulk. A gain in short circuit current
of 1.5 mA/cm−2can be achieved due to the highly reflective MoO3-x/Ag back contact.
better charge carrier collection if high doping levels limit the SCR width within the CIGSe.
CIGSe devices in the substrate configuration utilize molybdenum as the back contact
material, since it has the tendency to form a thin layer of MoSexduring the CIGSe
deposition, which is beneficial for the back contact quality [150]. The disadvantage of the
Mo/MoSe stack is the very low reflectivity.
The calculated light reflectivity at the interface between CIGSe and the MoSex/Mo
stack is shown in Fig. 10.13. The n and k values for CIGSe were taken from [181] and
for the metals from [182]. The Reflectivity is between 20 and 40 %, which leads to a
reduced EQE in the infrared region for thin absorber layers, as shown in Fig. 10.13b. The
reduced EQE translates into a short circuit current loss of 1.5 mA/cm−2for a 500 nm
thin CIGSe layer, compared to a device with a highly reflective contact like Au. The high
material costs of Au makes it no option for industrial application though. As shown in
Fig. 10.14 alternative metals to Au are only Pt and Ni, with Pt being as expensive as
Au and Ni being less reflective. Other highly reflective metals, like Al, Ag and Cu are
not well suitable due to their high low work function of 4.2, 4.5 and 4.7, respectively.
Copper has the further disadvantage of high diffusivity, which leads to shunted devices.
Aluminium oxidises very quickly, leading to a highly resistive contact. Leaving silver as
the best option. Devices fabricated with Ag back contacts were tested, but they suffered
from a strong Schottky contact which suppressed any photovoltaic activity.
Hole transport layers, like PEDOT:PSS or MoO3-x, typically used in organic solar
cells were also tested and surprisingly, MoO3-x in combination with Au lead to the best
devices. In organic solar cells MoO3-x is a standard material for hole extraction [183].
It is highly resistive, but 5-10 nm layer thickness were shown to be sufficient to achieve
ohmic contacts with ITO (WF 4.7 eV) [184] [183] and also Ag (WF 4.5 eV) [185]. The
10.3 Light management 153
Figure 10.14: J−Vcurves of ZnO/CIGSe stacks with different metals as the back contact,
without sodium doping. The dashed line is for a ZnO/CIGSe/Au device doped with sodium.
calculated reflectivity for the MoO3-x/Ag stack is also shown in Fig. 10.13a.
So far, only the MoO3-x/ITO stack was tested for semi-transparent CIGSe devices [184].
But the MoO3-x reacted with CIGSe to GaOxduring the deposition of CIGSe on top of
it. The MoO3-x/Ag stack has not yet been used for CIGSe solar cells. Here it is tested
for application as a highly reflective back contact in CIGSe superstrate devices. The
MoO3-x was thermally evaporated from a MoO3powder source onto two CIGSe coated
ZnO substrates at room temperature. One CIGSe layer was NaF post treated directly
after the CIGSe deposition. Reference cells were fabricated with Au back contacts, directly
after the CIGSe deposition, whereas, prior to the MoO3-x/Ag deposition, both CIGSe
layers were stored for 3 months in vacuum with several short vacuum breaks in between.
No further treatments were performed to remove potential oxidation of the surface on the
CIGSe layer.
The J−Vcurves of devices with Au and MoO3-x/Ag as the back contact are shown in
Fig. 10.15a. The devices are slightly Cu-poor at the back contact interface, which leads
to a high ohmic contact above 4
W
cm2for the devices fabricates without sodium. The
series resistance is slightly higher for the device with MoO3-x/Ag, probably induced by
the high resistive MoO3-x layer [186]. To increase the back contact quality it was shown
in Sec. 6.1, that it is necessary to dope the CIGSe surface with sodium. However, it was
not possible to perform the NaF PDT after the MoO3-x deposition, as it was done with
the 10 nm Mo layer described in Sec. 5.2.4, since the sodium did not diffuse through the
MoO3-x layer, even at 300
. Fig. 10.15b shows the J−Vcurve of the sample with the
NaF PDT performed prior to the MoO3-x deposition. Again, the MoO3-x layer increases
the series resistance by 0.5
W
cm2compared to the Au reference device. The MoO3-x/Ag
contact does not reduce the shunt resistance or the short circuit current. The variations
in the FF and the VOC are relatively small and likely to originate from typical sample
quality fluctuations and not from the MoO3-x/Ag contact.
154 10 Additional Information
(a) (b)
Figure 10.15: Experimental J−Vcurves of superstrate CIGSe solar cells with Au and
MoO3-x/Ag as the back contact, a) without any sodium present b) with NaF PDT at 300
for both samples.
10.3.3 Ultra thin absorbers
Fig. 10.16a displays the simulation results of the PCE for different values of the absorber
thickness. The device simulation was set-up with a Ga gradient towards the back contact,
the gradient increases over 100 nm CIGSe the conduction band minimum by 250 meV.
Further, no conduction band spike or cliff at the hetero-junction is set and the bulk
electron diffusion length is set to 2.3
µ
m. The charge carrier density is 1e+16 cm−3.
The Schottky barrier at the back contact was set to 200 meV with a high recombination
velocity of 1e+7 cm/s for electrons and holes. The back contact reflectivity was set to
95 %. The corresponding energy band diagram is shown in Fig. 10.16b. The absorption
coefficient for the different band gaps were taken from [187]. For 2.8
µ
m thick CIGSe layers
this setup results into a PCE of 18.6 %, limited by bulk and interface recombination. The
CIGSe thickness given in Fig. 10.16a refers to the CIGSe layer with no conduction band
gradient. The 100 nm thick graded layer, has to be added to the total thickness. It is
not varied during the thickness variation, since a too sharp increase of the Ga content
could lead to an increased defect density [188]. The diffusion length was set to 500 nm in
this layer, to account for the defects created by the Ga gradient. The contribution to the
absorption of this layer is low due to the increased band gap compared to the bulk.
The simulation results for the above described setup, show, that the PCE decreases
with decreasing CIGSe thickness. For thickness values between 500 nm and 2
µ
m this
is due to the increasing recombination losses at the back contact, despite the 250 meV
electron barrier at the back contact induced by the Ga Gradient. For thickness values
below 500 nm the PCE decreases sharply due to the poor light absorption yield. If the back
contact is perfectly passivated, which can be partially achieved with AlOxas reported in
[189], the recombination velocity for electrons is zero. This enhances the PCE considerably
for all thickness values below 3
µ
m.
10.3 Light management 155
(a) (b)
(c)
Figure 10.16: a) Dependence of the power conversion efficiency on the CIGSe layer thickness.
The recombination at the back contact is set to zero. Incomplete absorption and back contact
recombination leads to a reduced PCE for thin layers. b) Thickness dependence of the efficiency
for devices with a MoO3−x/Ag back contact and differently strong Ga gradients at the back
surface. c) Energy band diagram, which shows the Ga gradient at the back contact, which
reduces the back contact recombination and increases the electron collection.
156 10 Additional Information
The maximum efficiency of 18.9 % for the MoO3-x/Ag device, in the absence of back
contact recombination, is reached for the total CIGSe thickness of 1
µ
m. For thinner
layers, the light absorption is incomplete and for thicker layers the bulk recombination
is increased due to the small SCR at a charge carrier density of 1e+16 cm−3. For the
MoSex/Mo stack in substrate devices the maximum PCE is lowered to 18.8 % and shifted
to 1.4
µ
m due to the reduced back contact reflectivity. At a total CIGSe thickness of
600 nm the PCE is reduced to only 18.6 % for the superstrate device and 18.0 % for the
substrate device. For layers thinner than 500 nm, the PCE drops sharply. To reach an
efficiency of 18.6 %, the thickness of the CIGSe layer has to be increases to 1.0
µ
m.
It is questionable whether the back contact can be fully passivated in a real device.
But the back contact recombination can be also reduced by increasing the conduction
band gradient ∆EV,BC at the back contact from 250 meV to 500 meV. It even increases
the PCE further due to the improved electron collection from the back. Such a steep Ga
gradient is difficult to realize in substrate devices due to the inter-diffusion of In and Ga
during the deposition. In superstrate devices though it was shown in Sec. 2.1 that it can
be realized. This increases the maximum PCE to 19.0 % for a total thickness of 1
µ
m and
for 600 nm to 18.75 %.
The PCE could be increased even further due to the removal of the implemented
Schottky contact (not shown here), which improves the VOC value.
Experimentally, superstrate devices with thin and thick absorbers are difficult to com-
pare due to the different interface formation. The chemical composition at the interface
during the growth differs as well as the duration of the growth process. This leads to dif-
ferent charge carrier collection efficiencies and different VOC values. Fig. 10.17a shows the
J−Vcurve of a standard device with a 2.8
µ
m thick CIGSe layer and the same device with
a 0.75
µ
m thick CIGSe layer. The photo current at negative voltages drops by 1.5 mA/cm2
in the thinner device, which is exactly the expected value from the simulation. The VOC
and the FF are difficult to compare though, since the interface properties are not directly
comparable. The FF of the thin device is slightly better, which compensates the loss in
VOC. Still, the PCE for both devices are the same, as expected from the simulation.
Light scattering
Another option, to reduce the CIGSe thickness, is to increase the light pathway within
the absorber by light scattering. This can happen at the front or at the back contact.
The back contact has a given roughness due to the CIGSe surface roughness. This is
typically around 60 nm (RMS) for the co-evaporated layers prepared during this work.
The amount of diffuse scattering at the back contact can be calculated with the help of
the simple scalar scattering theory [190] [191] as follows:
Rdiffuse =Rtot ·exp(−2πσrms2nCIGSe(λ)cos(θ)
λ2
),(10.1)
10.3 Light management 157
(a) (b)
Figure 10.17: J−Vcurve of superstrate devices with CIGSe layers fabricated at 520
and
with a) different thickness b) plain and untreated ZnO as well as HCl etched ZnO.
with σrms describing the RMS surface roughness, nCIGSe(λ) the wavelength dependent
refractive index of CIGSe, θthe incoming angle to the surface normal. Equation 10.1 is
only valid if diffraction effects dominate the light reflection, which is the case for surfaces
whose facets are smaller or comparable to the wavelengths of the reflected light. For
typical CIGSe layers this is given for light in the visible and infrared region. The percent-
age of diffuse reflected light at the back contact is given in Fig. 10.19a. For the typical
roughness of 60 nm the diffuse reflected light is 90 % in the infrared region. Assuming
a cos2angle distribution of the reflected light, the average optical path of photons with
λ= 1200 nm would increase by 30 % for one way through the absorber. The minimum
thickness of 500 nm could be further reduced to 440 nm in order to achieve 19.1 % with
the in the previous section described model. The MoSex/Mo reference would lead to a
PCE of 18.3 %. To reach 19.1 % a thickness of 710 nm would be required.
In substrate devices the CIGSe surface roughness leads to light scattering at the front
contact. The equation to describe the diffuse transmission is similar to Eq. 10.1, but the
refractive index in the enumerator of the experiential is exchanged by half of the difference
of the refractive index of ZnO and CIGSe:
Tdiffuse =Ttot ·exp(−2πσrms(nCIGSe(λ)−nZnO(λ))cos(θ)
λ2
) (10.2)
Fig. 10.19b shows the percentage of diffuse transmittance of the total transmittance for
different values of the surface roughness. For the typical CIGSe roughness of 60 nm the
diffuse transmittance is 6 % and 94 % of the infrared light is transmitted specular. Thus,
the average optical pathway increases only 2 % for one way through the absorber, which
does not reduce the minimum absorber thickness noticeably.
The fraction of diffuse transmittance through the ZnO can be controlled in superstrate
devices by controlling the ZnO roughness prior to the CIGSe deposition. This is a standard
158 10 Additional Information
(a) (b)
Figure 10.18: a) Percentage of diffuse reflected light from the total reflected light at the
CIGSe/Au interface in superstrate devices. b) Percentage of diffuse transmitted light from the
total transmitted light at the ZnO/CIGSe interface in substrate devices.
procedure in amorphous silicon solar cells and usually achieved by HCl etching [192].
Achieving strong scattering for long wavelength photons with energies close to the band
gap of CIGSe devices is more difficult, due to the lower band gap and the lower refractive
index. To describe the scattering from a HCL etched ZnO it was found in [193] that the
exponent within the exponential function of Eq. 10.2 has to be changed from 2 to 3, to
account for the different roughness profile of the etched ZnO compared to the natural
CIGSe surface profile. This leads to the following formula:
Tdiffuse =Ttot ·exp(−4πσrms(nCIGSe(λ)−nZnO(λ))cos(θ)
λ3
) (10.3)
With this equation the measured diffuse transmittance of a HCL etched ZnO substrate
can be fitted to obtain the RMS roughness. Fig. 10.18a shows the measured and the cal-
culated diffuse transmission spectra, which are in very good agreement. The HCl etching
was performed for 60 s in a 1 wt.% HCl solution, as done in [194]. A RMS roughness of
46 nm is obtained from the fit. The calculated diffuse transmittance at the ZnO/CIGSe
interface in the infrared only increases by 0.2 %. Fig. 10.18b shows the dependence of the
diffuse transmittance at the CIGSe/ZnO interface. In order to get sufficient scattering to
reduce the layer thickness, the ZnO RMS roughness has to be above 100 nm for CIGSe
superstrate devices. This high roughness is likely to lead to shunting of the only 450 nm
thick CIGSe layer. Further it will increase the series resistance. The sheet resistance of
the 1
µ
m thick test sample had a roughness of 46 nm and the sheet resistance increased
from 6
W
/to 13
W
/due to the etching. This requires thicker ZnO layers, which in
turn increase the absorption losses within the ZnO and with it the PCE of the device
independent of the CIGSe thickness. The positive effect of the increased light scattering
is therefore exceeded by the negative effects from the increased series resistance or light
10.3 Light management 159
(a) (b)
Figure 10.19: a) The total diffuse transmitted light at the HCl etched ZnO interface with air,
fitted the function defined in Eq. 10.3. b) Percentage of the light which is diffuse transmitted
at the HCl etched ZnO interface with CIGSe.
absorption.
Experimentally, the J−Vcurves of superstrate devices with etched ZnO layers, show
very low fill factors and PCEs. Fig. 10.17b shows one example. Most likely this is caused
by a combination of increased series resistance and increased interface recombination, due
to the increased surface area. The parallel resistance, obtained from the dark J−Vcurve,
is however not reduced due to the 46 nm RMS roughness of the ZnO layer.
10.3.4 Summary
ZnO annealing was shown to enable a photo-current increase of up to 3 mA/cm2for cap
annealed ZnO and 1.5 mA/cm2for ZnO annealed during the CIGSe deposition, compared
to a as-deposited ZnO layer. At the same time it reduces the sheet resistance by 30 %.
Both effects were linked to the reduced structural disorder in the annealed films.
It was shown that MoO3-x/Ag back contacts can replace the Au contact in CIGSe
superstrate solar cells. Due to the high reflectivity of this contact it is possible to reduce
the total CIGSe thickness to 1
µ
m while even increasing the efficiency compared to thicker
devices. The increased efficiency is due to better charge carrier collection. Substrate
devices were shown to experience lower efficiencies already below 1.4
µ
m due to the poor
back contact reflection.
It was estimated that the total thickness of superstrate devices can be reduced to
540 nm in superstrate configuration while remaining the efficiency at 18.75 % compared to
17.8% in the substrate configuration. This is not only due to the improved light reflection
at the back contact, but also due to the enhanced light scattering naturally occurring in
superstrate devices. Increasing the light scattering further due to etching the ZnO prior to
the CIGSe deposition was not successful, due to increased interface recombination losses.
It is questionable whether TCO roughening is principally beneficial in CIGSe solar cells,
160 10 Additional Information
since efficient scattering requires a TCO RMS roughness around 100 nm due the low band
gap and refractive index of CIGSe.
10.4 CIGSe growth 161
10.4 CIGSe growth
To study the effect of an amorphous substrate on the CIGSe growth, this section will
compare a CIGSe layer grown on a crystalline ZnO layer and on an amorphous Ga2O3
layer. The difference of the ZnO and the CIGSe crystal structure and their lattice
constants is supposed to introduce crystallographic defects at the interface, as dangling
bonds for example, but may also induce increased stress within the CIGSe film or
influence the texture and the micro-structure. The ZnO surface is dominated by the
(002) plane, whose inter-atomic distance is 3.24 ˚
A. The inter-atomic distance in the
dominant (112) plane in CIGSe is between 2.4 and 2.6 ˚
A. Thus, no epitaxial growth
of CIGSe onto such textured ZnO is expected. As shown in the previous sections, no
transfer of texture and micro-structure of the ZnO to the CIGSe layer was observed
for the deposition temperature of 520
. However, at this temperature the formation
of an amorphous interfacial layer was observed. The amorphous layer forms during
the high-temperature deposition stage. Here, the CIGSe was deposited at a lower
temperature, to reduce the formation of this amorphous interfacial layer.
Experimental: The CIGSe was deposited by the modified 3-stage process at a
temperature of 470
. The buffer layer was chosen to be a crystalline i-ZnO layer and a
20 nm thick amorphous Ga2O3layer. The i-ZnO/ZnO:Al stack was deposited on soda lime
glass with a SiOXalkali barrier, the Ga2O3/ZnO:Ga stack was deposited on alkali free
glass. SEM images were taken and XRD patterns (in Bragg-Brentano geometry) recorded
Results: The SEM images of the resulting CIGSe layers are shown in Fig. 10.20a
and b. The obvious difference is the that the CIGSe layer grown on the i-ZnO exhibits
smaller grains compared to the CIGSe layer grown on the amorphous Ga2O3layer.
Further, it appears that the structure of the ZnO grain boundaries are transferred to the
grain boundary structure of the first 100 nm CIGSe layer. This cannot be observed for
the layer grown on the Ga2O3, which exhibits larger grains, with the same size close to
the substrate as in the bulk. The XRD pattern of both samples are shown in Fig. 10.20c,
which shows a different texture for both samples. The CIGSe deposited on i-ZnO has
a stronger (112) orientation compared to the CIGSe deposited on the amorphous layer.
The same was observed for CIGSe layers deposited at 450
.
Discussion: The CIGSe layers compared in this study show a different average grain
size and a different texture even though they were deposited in the same deposition run.
One reason could be the different glass and ZnO used as the substrate, which can lead to
different stress within the CIGSe film. One way to reduce the stress would be to increase
the grain growth [195], which could explain the different average grain sizes. But this
cannot explain the correlation of the micro-structure between the CIGSe layer, close to
the ZnO interface, and the ZnO layer. This indicates a structural correlation between the
ZnO and the CIGSe, which is lost once an amorphous interfacial layer develops. CIGSe
layers deposited at higher temperatures exhibit a more pronounced amorphous interfacial
162 10 Additional Information
layer and do not show such a correlation of the micro-structure.
(a) (b)
(c)
Figure 10.20: SEM image of a Cu(In0.7,Ga0.3)Se2layer grown at 470
on top of a a)
i-ZnO/AZO and b) a-Ga2O3/GZO substrate. c) XRD pattern of both samples.
10.5 Tandem configuration 163
10.5 Tandem configuration
To achieve all chalcopyrite 4-terminal tandem devices, it is necessary to prepare the
wide-gap chalcopyrite in superstrate configuration. Therefore, this section gives a short
overview on the different tandem configuration and their material requirements.
Figure 10.21: Monolithic (two-terminal) tandem device in substrate configuration (left) and
in superstrate configuration (right), both electrically connected by a tunnel-junction (hole
transport layer with ITO). Mechanically stacked (four-terminal) tandem device (center) me-
chanically connected with EVA foil (not to scale).
A mechanically stacked tandem device, as shown in the center image of Fig. 10.21,
requires a wide band gap semi-transparent superstrate device together with a narrow
band gap substrate device. The semi-transparent back contacts can be realized by using
ITO as the back contact [36]. These two devices are laminated together and contacted
separately, therefore they are also called 4-terminal device. For a narrow band gap of
1.16 eV for CIGSe, the wide gap material for the superstrate device should be between
1.7 eV and 2.0 eV to reach maximum theoretical efficiencies above 38 % [196]. Thus, a
combination of CIGSe and CGSe (Eg=1.70 eV) would be suitable.
An alternative tandem design is the monolithic, or 2-terminal, connection of the narrow
and wide band gap device. The distribution of optimum band gaps are narrower in this
case, due to the requirement of current matching in monolithic tandem devices. The
optimum combination would be 1.10 eV for the CIGSe bottom cell in combination with a
1.70 eV eV for the CGSe top cell. When choosing the substrate configuration, as shown
164 10 Additional Information
in the left image of Fig. 10.21, a wide gap CGSe solar cell has to be grown on top of
a narrow gap CIGSe bottom solar cell. This does not require the use of a superstrate
device, but the CIGSe bottom solar cell has to withstand a high temperature step during
the deposition of the top cell. The CdS/CIGSe interface was shown to be stable only until
300
[197], thus the CdS has to be exchanged with a thermally stable buffer layer just
as in superstrate devices. The monolithic tandem in superstrate configuration is shown
in the right image of Fig. 10.21. In this case the narrow gap bottom cell has to be grown
on top of the wide gap top cell.
In summary, independent of the choice for the tandem design, whether monolithic or
mechanically stacked, a thermally stable buffer layer has to be found in order to realize
an all chalcopyrite tandem solar cell with CGSe as the wide gap material. However, the
maximum efficiency for CGSe is so far limited to 11.2 % [198] and the highest wide gap
chalcopyrite superstrate device is reported to be only 3.8 % [28]. Further, parasitic light
absorption in CGSe devices were shown to be high [199]. This makes an all chalcopyrite
tandem unlikely at the moment and other wide gap materials like perovskites are more
favourable. Common PbI2based perovskite solar cells were shown to be stable until
140
[200].
List of Publications
First-authorship
M. D. Heinemann, V. Efimova, R. Klenk, B. Hoepfner, M. Wollgarten, T. Unold, H.-W.
Schock and C. A. Kaufmann, ”Cu(In,Ga)Se2Superstrate Solar Cells: Prospects and Lim-
itations”, Progress in Photovoltaics: Research and Applications (2014), published online.
M. D. Heinemann, D. Greiner, T. Unold, R. Klenk, H. Schock, R. Schlatmann and C.A.
Kaufmann, ”The Importance of Sodium Control in CIGSe Superstrate Solar Cells”, IEEE
Journal of Photovoltaics 5, 2015, Issue 1, 378-381
M.D. Heinemann, F. Ruske, D. Greiner, AR Jeong, M. Rusu, B. Rech, R. Schlatmann
and C.A. Kaufmann, “Light management in Cu(In,Ga)Se2 superstrate solar cells”, Solar
Energy Materials and Solar Cells (submitted).
M.D. Heinemann, J. Berry, G. Teeter, T. Unold and D. Ginley, “Oxygen deficiency and
Sn doping of amorphous Ga2O3“, Applied Physics Letters (accepted).
M. D. Heinemann, R. Mainz, H. Rodriguez-Alvarez, D. Greiner, C.A. Kaufmann and T.
Unold, ”In-situ Analysis of Cu(In,Ga)Se2 Absorber Growth: Band Gap Energy, Defect
Absorption, Growth Rate and Roughness”, Advanced Energy Materials (submission in
progress).
Co-authorship:
C. Kaufmann, D. Greiner, H. Rodriguez-Alvarez, A. Weber, M.D.Heinemann, J. Lauche,
M. Klaus, C. Genzel, H. W. Schock and R. Mainz, ”Co-evaporation of Cu(In, Ga)Se2 at
low temperatures: An In-Situ x-ray growth analysis ” IEEE 39th PVSC proceedings, 2013,
3058 - 3061
W. Ohm, W. Riedel, ¨
U. Ask¨uunger, M. D. Heinemann, C. A. Kaufmann, J. L. Garcia,
V. Izquierdo, X. Fontane, T. Goislard, M. C. Lux-Steiner, ”An Overview of Technological
Aspects of Cu(In,Ga)Se2Solar Cell Architectures Incorporating ZnO Nanorod Arrays
physica status solidi (a) 212 (2015), Nr. 1, 76-87.
J. K. Larsen, S.-Y. Li, J. J. Scragg, Y. Ren, C. H¨agglund, M.D. Heinemann, S. Kret-
zschmar, T. Unold, C. Platzer-Bj¨orkman, ”Interference effects in photoluminescence spec-
tra of CZTS and CIGS thin films”, Journal of Applied Physics, 118, 2015.
166 10 Additional Information
S. C. Siah, R. E. Brandt, L. T. Schelhas, K. Lim, J. D. Perkins, R. Jaramillo, M. D.
Heinemann, D. Chua, J. Wright, C. U. Segre, R. G. Gordon, M. F. Toney, T. Buonassisi,
”Dopant activation in Sn-doped Ga2O3 investigated by synchrotron-based X-ray absorp-
tion spectroscopy ” Applied Physics Letters, not yet published.
R. Mainz, H. Rodriguez-Alvarez, M. Klaus, D. Thomas, J. Lauche, A. Weber, M. D.
Heinemann, S. Brunken, D. Greiner, C. A. Kaufmann, T. Unold, H.-W. Schock, and C.
Genzel, ”Sudden Stress Relaxation in CuInSe2 Films during Cu-Se Deposition Revealed
by Real-time X-ray Analysis” Physical Review B 92, 2015, 155310 1-8.
R. Mainz, H. Rodriguez-Alvarez, D. Greiner, M. D. Heinemann, H. Stange, M. Klaus, C.
Genzel, H.-W. Schock, C. A. Kaufmann, ”Formation of Cu(In,Ga)Se2 thin-film solar cell
absorbers by multi-stage co-evaporation studied by real-time X-ray analysis” manuscript
in process
BIBLIOGRAPHY 167
Bibliography
[1] Jackson, Philip ; Hariskos, Dimitrios ; Wuerz, Roland ; Kiowski, Oliver ; Bauer,
Andreas ; Friedlmeier, Theresa M. ; Powalla, Michael: Properties of Cu (In, Ga) Se2
solar cells with new record efficiencies up to 21.7%. In: physica status solidi (RRL)-Rapid
Research Letters 9999 (2014)
[2] Raugei, M ; Bargigli, S ; Ulgiati, S: Energy and life cycle assessment of thin film
CdTe photovoltaic modules. In: Proceedings of the 20th European Photovoltaic Solar
Energy Conference, 2005, S. 6–10
[3] Dimmler, Bernhard: CIGS and CdTe based thin film PV modules, an industrial
r/evolution. In: Photovoltaic Specialists Conference (PVSC), 2012 38th IEEE IEEE,
2012, S. 002494–002499
[4] Aboudi, M: PV bill of materials: Costs and market outlook through 2015: An excerpt
from the GTM research report PV bill of materials outlook, 2010–2015. In: PV News 3
(2011), S. 3–9
[5] Nakada, T. ; Mise, T.: High-efficiency superstrate type cigs thin film solar cells with
graded bandgap absorber layers
[6] Terheggen, M. ; Heinrich, H. ; Kostorz, G. ; Haug, F.J. ; Zogg, H. ; Tiwari, AN:
Ga2O3 segregation in Cu(In,Ga)Se2/ZnO superstrate solar cells and its impact on their
photovoltaic properties. In: Thin solid films 403 (2002), S. 212–215
[7] Wild-Scholten, MJ d. ; Schottler, M: Solar as an environmental product: Thin-film
modules–production processes and their environmental assessment. In: Solar Energy 2012
(2013), S. 2011
[8] W¨
urfel, Peter ; W¨
urfel, Uli: Physics of solar cells: from basic principles to advanced
concepts. John Wiley & Sons, 2009
[9] Scheer, Roland ; Schock, Hans-Werner: Chalcogenide Photovoltaics: Physics, Tech-
nologies, and Thin Film Devices. John Wiley & Sons, 2011
[10] Green, Martin A.: Solar cells: operating principles, technology, and system applications.
(1982)
[11] Nelson, Jenny: The physics of solar cells. Bd. 57. World Scientific, 2003
[12] Anderson, RL: Germanium-gallium arsenide heterojunctions. In: IBM Journal of Re-
search and Development 4 (1960), Nr. 3, S. 283–287
[13] Abou-Ras, D ; Kostorz, G ; Romeo, A ; Rudmann, D ; Tiwari, AN: Structural and
chemical investigations of CBD-and PVD-CdS buffer layers and interfaces in Cu (In, Ga)
Se 2-based thin film solar cells. In: Thin Solid Films 480 (2005), S. 118–123
[14] Minemoto, Takashi ; Matsui, Takuya ; Takakura, Hideyuki ; Hamakawa, Yoshi-
hiro ; Negami, Takayuki ; Hashimoto, Yasuhiro ; Uenoyama, Takeshi ; Kitagawa,
168 BIBLIOGRAPHY
Masatoshi: Theoretical analysis of the effect of conduction band offset of window/CIS
layers on performance of CIS solar cells using device simulation. In: solar Energy Mate-
rials and solar Cells 67 (2001), Nr. 1, S. 83–88
[15] Persson, Clas ; Zhao, Yu-Jun ; Lany, Stephan ; Zunger, Alex: n-type doping of
CuInSe2 and CuGaSe2. In: Physical Review B 72 (2005), Nr. 3, S. 035211
[16] Ishizuka, S ; Sakurai, K ; Yamada, A ; Matsubara, K ; Fons, P ; Iwata, K ;
Nakamura, S ; Kimura, Y ; Baba, T ; Nakanishi, H u. a.: Fabrication of wide-gap
Cu (In 1- xGax) Se 2 thin film solar cells: a study on the correlation of cell performance
with highly resistive i-ZnO layer thickness. In: Solar energy materials and solar cells 87
(2005), Nr. 1, S. 541–548
[17] Ruske, F. ; Roczen, M. ; Lee, K. ; Wimmer, M. ; Gall, S. ; Hupkes, J. ; Hrunski, D.
;Rech, B.: Improved electrical transport in Al-doped zinc oxide by thermal treatment.
In: Journal of Applied Physics 107 (2010), Nr. 1, S. 013708–013708
[18] Greiner, D ; Gledhill, SE ; K¨
oble, Ch ; Krammer, J ; Klenk, R: Damp heat
stability of Al-doped zinc oxide films on smooth and rough substrates. In: Thin Solid
Films 520 (2011), Nr. 4, S. 1285–1290
[19] Nakada, Toklo ; Okano, Nobukata ; Tanaka, Yoshlyukl ; Fukuda, Hlroyukl ; Ku-
nioka, Akio: Superstrate-type CuInSe 2 solar cells with chemically deposited CdS win-
dow layers. In: Photovoltaic Energy Conversion, 1994., Conference Record of the Twenty
Fourth. IEEE Photovoltaic Specialists Conference-1994, 1994 IEEE First World Confer-
ence on Bd. 1 IEEE, 1994, S. 95–98
[20] Negami, Takayuki ; Nishitani, Mikihiko ; Ikeda, Mitsusuke ; Wada, Takahiro: Prepa-
ration of CuInSe 2 films on large grain CdS films for superstrate-type solar cells. In: Solar
energy materials and solar cells 35 (1994), S. 215–222
[21] Haug, F.J. ; Zogg, H. ; Tiwari, AN: 11% efficiency on CIGS superstrate solar cells with-
out sodium precursor. In: Photovoltaic Specialists Conference, 2002. Conference Record
of the Twenty-Ninth IEEE IEEE, 2002, S. 728–731
[22] Haug, F.J. ; Rudmann, D. ; Zogg, H. ; Tiwari, AN: Light soaking effects in
Cu(In,Ga)Se2 superstrate solar cells. In: Thin solid films 431 (2003), S. 431–435
[23] Minemoto, Takashi ; Harada, Shinya ; Takakura, Hideyuki: Cu(In,Ga)Se2
superstrate-type solar cells with Zn1xMgxO buffer layers. In: Current Applied Physics
12 (2012), Nr. 1, S. 171 – 173
[24] Ohm, Wiebke ; Riedel, Wiebke ; Ask¨
unger,¨
Umit ; Heinemann, Marc D. ; Kaufmann,
Christian A. ; Garcia, Juan L. ; Izquierdo, Victor ; Fontan´
e, Xavier ; Goislard,
Thomas ; Lux-Steiner, Martha C. u. a.: An overview of technological aspects of Cu
(In, Ga) Se2 solar cell architectures incorporating ZnO nanorod arrays. In: physica status
solidi (a) 212 (2015), Nr. 1, S. 76–87
BIBLIOGRAPHY 169
[25] Larsen, Jes K. ; Simchi, H ; Xin, P ; Kim, K ; Shafarman, WN: Backwall superstrate
configuration for ultrathin Cu (In, Ga) Se2 solar cells. In: Applied Physics Letters 104
(2014), Nr. 3, S. 033901
[26] Heinemann, MD ; Greiner, D ; Unold, T ; Klenk, R ; Schock, H-W ; Schlatmann,
R ; Kaufmann, CA: The Importance of Sodium Control in CIGSe Superstrate Solar
Cells. In: Photovoltaics, IEEE Journal of 5 (2015), Nr. 1, S. 378–381
[27] Larsen, Jes K. ; Xin, Peipei ; Shafarman, William N.: Formation of Ga2O3 barrier
layer in Cu (InGa) Se2 superstrate devices with ZnO buffer layer. In: MRS Proceedings
Bd. 1538 Cambridge Univ Press, 2013, S. mrss13–1538
[28] Kaigawa, R ; Funahashi, K ; Fujie, R ; Wada, T ; Merdes, S ; Caballero, R ;
Klenk, R: Tandem solar cells with Cu(In,Ga)S2 top cells on ZnO coated substrates. In:
Solar Energy Materials and Solar Cells 94 (2010), Nr. 11, S. 1880–1883
[29] Balboul, M.R ; Jasenek, A ; Chernykh, O ; Rau, U ; Schock, H.W: CuGaSe2-based
superstrate solar cells. In: Thin Solid Films 387 (2001), Nr. 1–2, S. 74 – 76
[30] Ikeda, Shigeru ; Kamai, Ryo ; Lee, Sun M. ; Yagi, Tetsuro ; Harada, Takashi ;
Matsumura, Michio: A superstrate solar cell based on In 2 (Se, S) 3 and CuIn (Se, S)
2 thin films fabricated by electrodeposition combined with annealing. In: Solar Energy
Materials and Solar Cells 95 (2011), Nr. 6, S. 1446–1451
[31] Nguyen, Duy-Cuong ; Takehara, Kenji ; Ryo, Toshihiro ; Ito, Seigo: Back Contact
Materials for Superstrate CuInS2 Solar Cells. In: Energy Procedia 10 (2011), Nr. 0, S. 49
– 54
[32] Knacke, Ottmar ; Kubaschewski, Oswald ; Hesselmann, Klaus: Thermochemical
properties of inorganic substances. Bd. 1. Springer Berlin, 1991
[33] Marr´
on, D F. ; Lehmann, S ; Lux-Steiner, M C.: Growth of isolated and embedded
Cu-containing chalcopyrite clusters and nanocrystals by dry processing. In: Physical
Review B 77 (2008), Nr. 8, S. 085315
[34] Guillemoles, Jean-Fran¸cois ; Kronik, Leeor ; Cahen, David ; Rau, Uwe ; Jasenek,
Axel ; Schock, Hans-Werner: Stability issues of Cu (In, Ga) Se2-based solar cells. In:
The Journal of Physical Chemistry B 104 (2000), Nr. 20, S. 4849–4862
[35] Eriksson, Gunnar ; Hack, Klaus: ChemSage—a computer program for the calculation
of complex chemical equilibria. In: Metallurgical Transactions B 21 (1990), Nr. 6, S.
1013–1023
[36] Nakada, Tokio ; Hirabayashi, Yutaka ; Tokado, Takehito ; Ohmori, Daiske ; Mise,
Takahiro: Novel device structure for Cu(In,Ga)Se2 thin film solar cells using transparent
conducting oxide back and front contacts. In: Solar Energy 77 (2004), Nr. 6, S. 739 – 747
[37] Heitjans, Paul ; K¨
arger, J¨org: Diffusion in condensed matter: methods, materials,
models. Springer Science & Business Media, 2006
170 BIBLIOGRAPHY
[38] Bastek, Jens ; Stolwijk, Nico ; Wuerz, Roland ; Albert, Juergen ; Sadewasser,
Sascha: Diffusion of Zn in solar-grade Cu (In, Ga) Se 2 and single-crystal CuInSe 2 thin
films. (2012)
[39] Schorr, S ; Stephan, C: www.helmholtz−berlin.de/forschung/oe/em/kristallographie/arbeitsgebiete/photovoltaic/point−
defect −analysisen.html. (2012)
[40] Wei, Su-Huai ; Zunger, Alex: Band offsets and optical bowings of chalcopyrites and
Zn-based II-VI alloys. In: Journal of Applied Physics 78 (1995), Nr. 6, S. 3846–3856
[41] Maeda, Tsuyoshi ; Wada, Takahiro: Characteristics of chemical bond and vacancy
formation in chalcopyrite-type CuInSe2 and related compounds. In: physica status solidi
(c) 6 (2009), Nr. 5, S. 1312–1316
[42] Abou-Ras, D ; Caballero, R ; Kaufmann, CA ; Nichterwitz, M ; Sakurai, K ;
Schorr, S ; Unold, T ; Schock, HW: Impact of the Ga concentration on the microstruc-
ture of CuIn1–xGaxSe2. In: physica status solidi (RRL)-Rapid Research Letters 2 (2008),
Nr. 3, S. 135–137
[43] Jaffe, JE ; Zunger, Alex: Electronic structure of the ternary chalcopyrite semiconduc-
tors CuAl S 2, CuGa S 2, CuIn S 2, CuAl Se 2, CuGa Se 2, and CuIn Se 2. In: Physical
Review B 28 (1983), Nr. 10, S. 5822
[44] Wei, Su-Huai ; Zhang, SB ; Zunger, Alex: Effects of Ga addition to CuInSe on its
electronic, structural, and defect properties. In: Applied physics letters 72 (1998), S. 3199
[45] Jaffe, JE ; Zunger, Alex: Anion displacements and the band-gap anomaly in ternary
AB C 2 chalcopyrite semiconductors. In: Physical Review B 27 (1983), Nr. 8, S. 5176
[46] Zhang, SB ; Wei, Su-Huai ; Zunger, Alex ; Katayama-Yoshida, H: Defect physics
of the CuInSe 2 chalcopyrite semiconductor. In: Physical Review B 57 (1998), Nr. 16, S.
9642
[47] Wei, S.H. ; Zhang, SB ; Zunger, A.: Effects of Na on the electrical and structural
properties of CuInSe. In: Journal of applied physics 85 (1999), S. 7214
[48] Lany, Stephan ; Zunger, Alex: Light- and bias-induced metastabilities in Cu(In,Ga)Se2
based solar cells caused by the (V(Se)-V(Cu)) vacancy complex. In: Journal of Applied
Physics 100 (2006), Nr. 11, S. 113725
[49] Schorr, S ; Tovar, M ; St¨
ußer, N ; Bente, K: Investigation of structural anomaly
and metal ordering in the solid solution 2ZnS–CuInS2 by neutron diffraction. In:
Physica B: Condensed Matter 350 (2004), Nr. 1–3, Supplement, S. E411 – E414. –
¡ce:title¿Proceedings of the Third European Conference on Neutron Scattering¡/ce:title¿
[50] Enzenhofer, T. ; Unold, T. ; Schock, H. W.: Doping induced structural changes in
CuInS2 thin films and the effects on optical and electrical properties. In: physica status
solidi (a) 203 (2006), Nr. 11, S. 2624–2629
BIBLIOGRAPHY 171
[51] Schmidt, SS ; Abou-Ras, D. ; Unold, T. ; Eisenbarth, T. ; Wilhelm, H. ; Grimm, A.
;Klaer, J. ; Schock, H.W.: Effect of Zn incorporation into CuInS2 solar cell absorbers
on microstructural and electrical properties. In: Journal of Applied Physics 110 (2011),
S. 064515
[52] Nakada, Tokio ; Iga, Daisuke ; Ohbo, Hiroki ; Kunioka, Akio: Effects of sodium on Cu
(In, Ga) Se2-based thin films and solar cells. In: JAPANESE JOURNAL OF APPLIED
PHYSICS PART 1 REGULAR PAPERS SHORT NOTES AND REVIEW PAPERS 36
(1997), S. 732–737
[53] Kimura, Ryuhei ; Nakada, Tokio ; Fons, Paul ; Yamada, Akimasa ; Niki, Shigeru ;
Matsuzawa, Takeo ; Takahashi, Kiyoshi ; Kunioka, Akio: Photoluminescence prop-
erties of sodium incorporation in CuInSe2 and CuIn3Se5 thin films. In: Solar energy
materials and solar cells 67 (2001), Nr. 1, S. 289–295
[54] Nakada, T. ; Kume, T. ; Mise, T. ; Kunioka, A.: Superstrate-Type Cu (In, Ga) Se2
thin film solar cells with ZnO buffer layers. In: Japanese journal of applied physics 37
(1998), S. L499–L501
[55] Yamamoto, Takashi ; Shiosaki, Tadashi ; Kawabata, Akira: Characterization of ZnO
piezoelectric films prepared by rf planar-magnetron sputtering. In: Journal of Applied
Physics 51 (1980), Nr. 6, S. 3113–3120
[56] Erhart, Paul ; Albe, Karsten ; Klein, Andreas: First-principles study of intrinsic
point defects in ZnO: Role of band structure, volume relaxation, and finite-size effects. In:
Physical Review B 73 (2006), Nr. 20, S. 205203
[57] Rao, G V. ; Sauberlich, F ; Klein, A: Influence of Mg content on the band alignment at
CdS/(Zn, Mg) O interfaces. In: Applied Physics Letters 87 (2005), Nr. 3, S. 032101–032101
[58] Persson, Clas ; Platzer-Bj¨
orkman, Charlotte ; Malmstr¨
om, Jonas ; T¨
orndahl,
Tobias ; Edoff, Marika: Strong valence-band offset bowing of ZnO 1-x S x enhances
p-type nitrogen doping of ZnO-like alloys. In: Physical review letters 97 (2006), Nr. 14,
S. 146403
[59] Burstein, Elias: Anomalous optical absorption limit in InSb. In: Physical Review 93
(1954), Nr. 3, S. 632
[60] Laboratory, OHBA’s: http ://www.geocities.jp/ohbalabobpage/structure6.html.
(2012)
[61] Janotti, Anderson ; Walle, Chris G. d.: Native point defects in ZnO. In: Physical
Review B 76 (2007), Nr. 16, S. 165202
[62] Zhang, SB ; Wei, S-H ; Zunger, Alex: Intrinsic n-type versus p-type doping asymmetry
and the defect physics of ZnO. In: Physical Review B 63 (2001), Nr. 7, S. 075205
[63] M¨
uller, G ; Helbig, R: ¨
Uber den Einbau von Kupfer in ZnO-Einkristallen. In: Journal
of Physics and Chemistry of Solids 32 (1971), Nr. 8
172 BIBLIOGRAPHY
[64] Minami, Tadatsugu ; Sato, Hirotoshi ; Nanto, Hidehito ; Takata, Shinzo: Group III
impurity doped zinc oxide thin films prepared by RF magnetron sputtering. In: Japanese
journal of applied physics 24 (1985), Nr. 10A, S. L781
[65] Sun, Luwei ; He, Haiping ; Liu, Chao ; Ye, Zhizhen: Highly efficient orange emission in
ZnO: Se nanorods. In: Journal of Applied Physics 108 (2010), Nr. 12, S. 124313
[66] Park, CH ; Zhang, SB ; Wei, Su-Huai: Origin of p-type doping difficulty in ZnO: The
impurity perspective. In: Physical Review B 66 (2002), Nr. 7, S. 073202
[67] Hsu, Cheng-Liang ; Gao, Yi-Dian ; Chen, You-Syuan ; Hsueh, Ting-Jen: Vertical p-
Type Cu-Doped ZnO/n-Type ZnO Homojunction Nanowire-Based Ultraviolet Photode-
tector by the Furnace System with Hotwire Assistance. In: ACS applied materials &
interfaces 6 (2014), Nr. 6, S. 4277–4285
[68] Greiner, D: PVD-B Anleitung. In: HZB intern 2014 (20), S. 1
[69] Scheer, R ; Neisser, A ; Sakurai, K ; Fons, P ; Niki, S: Cu (In1- xGax) Se2 growth
studies by in situ spectroscopic light scattering. In: Applied physics letters 82 (2003), Nr.
13, S. 2091–2093
[70] Kaufmann, CA ; Neisser, A ; Klenk, R ; Scheer, R: Transfer of Cu (In, Ga) Se¡ sub¿
2¡/sub¿ thin film solar cells to flexible substrates using an in situ process control. In: Thin
Solid Films 480 (2005), S. 515–519
[71] Sakurai, Keiichiro ; Neumann, Timo ; Hesse, Raik ; Abou-Ras, Daniel ; Jablon-
ski, Piotr ; Neisser, Axel ; Kaufmann, Christian ; Niki, Shigeru ; Scheer, Roland ;
Schock, Hans-Werner: Characteristics of scattered laser light signals from Cu (In, Ga)
Se 2 films. In: Thin solid films 515 (2007), Nr. 15, S. 6222–6225
[72] Hesse, Raik ; Schock, Hans-Werner ; Abou-Ras, Daniel ; Unold, Thomas: Method for
the in-situ determination of the material composition of optically thin layers. Dezember 25
2012. – US Patent 8,338,194
[73] Contreras, Miguel A. ; Tuttle, John ; Gabor, Andrew ; Tennant, Andrew ;
Ramanathan, Kannan ; Asher, Sally ; Franz, Amy ; Keane, James ; Wang, L ;
Scofield, John u. a.: High efficiency Cu (In, Ga) Se 2-based solar cells: processing of
novel absorber structures. In: Photovoltaic Energy Conversion, 1994., Conference Record
of the Twenty Fourth. IEEE Photovoltaic Specialists Conference-1994, 1994 IEEE First
World Conference on Bd. 1 IEEE, 1994, S. 68–75
[74] Barreau, N ; Painchaud, T ; Couzini´
e-Devy, F ; Arzel, L ; Kessler, J: Recrystal-
lization of CIGSe layers grown by three-step processes: A model based on grain boundary
migration. In: Acta Materialia 58 (2010), Nr. 17, S. 5572–5577
[75] R. Mainz, M. Klaus D. Thomas J. Lauche A. Weber M. D. Heinemann S. Brunken D.
Greiner C. A. Kaufmann T. Unold H.-W. S. H. Rodriguez-Alvarez ; Genzel, C.: Sudden
Stress Relaxation in CuInSe2 Films during Cu-Se Deposition Revealed by Real-time X-ray
Analysis. In: Not yet published
BIBLIOGRAPHY 173
[76] G¨
utay, Levent ; Regesch, David ; Larsen, Jes K. ; Aida, Yasuhiro ; Depredurand,
Val´erie ; Siebentritt, Susanne: Influence of copper excess on the absorber quality of
CuInSe2. In: Applied Physics Letters 99 (2011), Nr. 15, S. 151912
[77] Wolf, Bernhard: Handbook of ion sources. CRC press, 1995
[78] Chrisey, Douglas B. ; Hubler, Graham K.: Pulsed laser deposition of thin films. (1994)
[79] Sze, Simon M. ; Ng, Kwok K.: Physics of semiconductor devices. John Wiley & Sons,
2006
[80] Hilibrand, J ; Gold, RD: Determination of the impurity distribution in junction diodes
from capacitance-voltage measurements. In: RCA review 21 (1960), Nr. 2, S. 245–252
[81] Eisenbarth, Tobias: Identifikation von Defekten und Metastabilit¨aten in Cu (In, Ga)
Se 2-D¨unnschichtsolarzellen, Freie Universit¨at Berlin, Diss., 2010
[82] Wiley, John D. ; Miller, GL: Series resistance effects in semiconductor CV profiling.
In: Electron Devices, IEEE Transactions on 22 (1975), Nr. 5, S. 265–272
[83] Rech, Bern ; Unold, Thomas: Photovoltaik Praktikum: Herstellung einer Solarzelle.
Helmholtz Znetrum Berlin, 2012
[84] Arora, ND ; Chamberlain, SG ; Roulston, DJ: Diffusion length determination in p-n
junction diodes and solar cells. In: Applied Physics Letters 37 (1980), Nr. 3, S. 325–327
[85] Nichterwitz, Melanie ; Caballero, Raquel ; Kaufmann, Christian A. ; Schock,
Hans-Werner ; Unold, Thomas: Generation-dependent charge carrier transport in Cu
(In, Ga) Se 2/CdS/ZnO thin-film solar-cells. In: Journal of Applied Physics 113 (2013),
Nr. 4, S. 044515–044515
[86] Rechid, J ; Kampmann, A ; Reineke-Koch, R: Characterising superstrate CIS solar
cells with electron beam induced current. In: Thin solid films 361 (2000), S. 198–202
[87] Donolato, C: An alternative proof of the generalized reciprocity theorem for charge
collection. In: Journal of Applied Physics 66 (1989), Nr. 9, S. 4524–4525
[88] Haney, Paul M. ; Yoon, Heayoung P. ; Koirala, Prakash ; Collins, Robert W. ;
Zhitenev, Nikolai B.: Electron beam induced current in the high injection regime. In:
arXiv preprint arXiv:1410.4435 (2014)
[89] Briggs, David ; Seah, Martin P.: Practical surface analysis by Auger and X-ray pho-
toelectron spectroscopy. In: D. Briggs, & M. P. Seah,(Editors), John Wiley & Sons,
Chichester 1983, xiv+ 533 (1983)
[90] Stefan, H¨ufner: Photoelectron Spectroscopy: Principles and Applications. 1995
[91] Seah, MP: The quantitative analysis of surfaces by XPS: a review. In: Surface and
Interface Analysis 2 (1980), Nr. 6, S. 222–239
[92] H¨
opfner, Britta: Diffusion und Phasenbildung bei der Abscheidung von Chalkopyritab-
sorbern auf transparenten oxidischen Leitern, Freie Universit¨at Berlin, Diss., 2014
174 BIBLIOGRAPHY
[93] Lauermann, Iver ; B¨
ar, Marcus ; Fischer, Christian-Herbert: Synchrotron-based spec-
troscopy for the characterization of surfaces and interfaces in chalcopyrite thin-film solar
cells. In: Solar Energy Materials and Solar Cells 95 (2011), Nr. 6, S. 1495–1508
[94] Pro, Igor: ; WaveMetrics. In: Inc.: Lake Oswego, OR (1996)
[95] Borchardt-Ott, Walter: Kristallographie. Bd. 4. Springer, 1987
[96] K¨
ampfe, Bernd ; Luczak, Falk ; Michel, Bernd: Energy Dispersive X-Ray Diffraction.
In: Particle & Particle Systems Characterization 22 (2005), Nr. 6, S. 391–396
[97] Scherrer, Paul: Bestimmung der Gr¨osse und der inneren Struktur von Kolloidteilchen
mittels R¨ontgenstrahlen. In: Nachrichten von der Gesellschaft der Wissenschaften zu
G¨ottingen, mathematisch-physikalische Klasse 1918 (1918), S. 98–100
[98] Rodriguez-Alvarez, Humberto ; Weber, Alfons ; Lauche, Jakob ; Kaufmann, Chris-
tian A. ; Rissom, Thorsten ; Greiner, Dieter ; Klaus, Manuela ; Unold, Thomas
;Genzel, Christoph ; Schock, Hans-Werner u. a.: Formation of CuInSe2 and Cu-
GaSe2 Thin-Films Deposited by Three-Stage Thermal Co-Evaporation: A Real-Time X-
Ray Diffraction and Fluorescence Study. In: Advanced Energy Materials 3 (2013), Nr. 10,
S. 1381–1387
[99] Efimova, Varvara: Study in analytical glow discharge spectrometry and its application in
materials science, Diss., 2011
[100] Gloeckler, M ; Fahrenbruch, AL ; Sites, JR: Numerical modeling of CIGS and CdTe
solar cells: setting the baseline. In: Photovoltaic Energy Conversion, 2003. Proceedings of
3rd World Conference on Bd. 1 IEEE, 2003, S. 491–494
[101] Sentaurus, TCAD: User Guide, Synopsys, 2005. In: URL: http://www. synopsys.
com/Tools/TCAD/Pages/default. aspx
[102] Stangl, R ; Kriegel, M ; Schmidt, M: AFORS-HET, Version 2.2, a numerical computer
program for simulation of heterojunction solar cells and measurements. In: Photovoltaic
Energy Conversion, Conference Record of the 2006 IEEE 4th World Conference on Bd. 2
IEEE, 2006, S. 1350–1353
[103] Pieters, BE ; Krc, J ; Zeman, M: Advanced numerical simulation tool for solar cells-
ASA5. In: Photovoltaic Energy Conversion, Conference Record of the 2006 IEEE 4th
World Conference on Bd. 2 IEEE, 2006, S. 1513–1516
[104] Clugston, Donald A. ; Basore, Paul A.: PC1D version 5: 32-bit solar cell modeling on
personal computers. In: Photovoltaic Specialists Conference, 1997., Conference Record of
the Twenty-Sixth IEEE IEEE, 1997, S. 207–210
[105] Burgelman, Marc ; Nollet, Peter ; Degrave, Stefaan: Modelling polycrystalline
semiconductor solar cells. In: Thin Solid Films 361 (2000), S. 527–532
[106] Burgelman, Marc ; Decock, Koen ; Khelifi, Samira ; Abass, Aimi: Advanced elec-
trical simulation of thin film solar cells. In: Thin Solid Films (2012), Nr. 0, S. –
BIBLIOGRAPHY 175
[107] Burgelman, Marc ; Marlein, Jonas: Analysis of graded band gap solar cells with
SCAPS. In: Proceedings of the 23rd European Photovoltaic Solar Energy Conference,
WIP, Munchen, 2008, S. 2151–2155
[108] Koster, LJA ; Smits, ECP ; Mihailetchi, VD ; Blom, PWM: Device model for the
operation of polymer/fullerene bulk heterojunction solar cells. In: Physical Review B 72
(2005), Nr. 8, S. 085205
[109] Sah, Chih-Tang ; Noyce, Robert N. ; Shockley, William: Carrier generation and
recombination in pn junctions and pn junction characteristics. In: Proceedings of the IRE
45 (1957), Nr. 9, S. 1228–1243
[110] Nichterwitz, Melanie: Dissertation. (2012)
[111] Bube, Richard: Fundamentals of solar cells: photovoltaic solar energy conversion. Else-
vier, 2012
[112] Pauwels, HJ ; Vanhoutte, G: The influence of interface state and energy barriers on
the efficiency of heterojunction solar cells. In: Journal of Physics D: Applied Physics 11
(1978), Nr. 5, S. 649
[113] Rhoderick, EH: Metal-semiconductor contacts. In: IEE Proceedings I (Solid-State and
Electron Devices) 129 (1982), Nr. 1, S. 1–14
[114] Selberherr, Siegfried: Analysis and simulation of semiconductor devices. Springer
Wien; New York, 1984
[115] Nakada, Tokio: Microstructural and diffusion properties of CIGS thin film solar cells
fabricated using transparent conducting oxide back contacts. In: Thin Solid Films 480–481
(2005), Nr. 0, S. 419 – 425
[116] Nakada, T. ; Kume, T. ; Kunioka, A.: Superstrate-type CuInSe2-based thin film solar
cells by a low-temperature process using sodium compounds. In: Solar energy materials
and solar cells 50 (1998), Nr. 1-4, S. 97–103
[117] Delahoy, Alan E. ; Guo, Sheyu: Transparent conducting oxides for photovoltaics. In:
Handbook of Photovoltaic Science and Engineering, Second Edition (2011), S. 716–796
[118] Grimm, A ; Kieven, D ; Lauermann, I ; Lux-Steiner, M C. ; Hergert, F ;
Schwieger, R ; Klenk, R: Zn (O, S) layers for chalcoyprite solar cells sputtered from a
single target. In: EPJ Photovoltaics 3 (2012)
[119] Hubert, C ; Naghavi, N ; Roussel, O ; Etcheberry, A ; Hariskos, D ; Menner,
R ; Powalla, M ; Kerrec, O ; Lincot, D: The Zn (S, O, OH)/ZnMgO buffer in thin
film Cu (In, Ga)(S, Se) 2-based solar cells part I: Fast chemical bath deposition of Zn
(S, O, OH) buffer layers for industrial application on Co-evaporated Cu (In, Ga) Se2 and
electrodeposited CuIn (S, Se) 2 solar cells. In: Progress in Photovoltaics: Research and
Applications 17 (2009), Nr. 7, S. 470–478
176 BIBLIOGRAPHY
[120] Platzer-Bjorkman, C ; Torndahl, T ; Abou-Ras, Daniel ; Malmstrom, J ;
Kessler, John ; Stolt, Lars: Zn (O, S) buffer layers by atomic layer deposition in
Cu (In, Ga) Se 2 based thin film solar cells: band alignment and sulfur gradient. In:
Journal of Applied Physics 100 (2006), Nr. 4, S. 044506–044506
[121] Merdes, Saoussen ; Ziem, Florian ; Lavrenko, Tetiana ; Walter, Thomas ; Lauer-
mann, Iver ; Klingsporn, Max ; Schmidt, Sebastian ; Hergert, Frank ; Schlatmann,
Rutger: Above 16% efficient sequentially grown Cu (In, Ga)(Se, S) 2-based solar cells with
atomic layer deposited Zn (O, S) buffers. In: Progress in Photovoltaics: Research and Ap-
plications (2015)
[122] Kieven, D. ; Grimm, A. ; Lauermann, I. ; Lux-Steiner, M.C. ; Palm, J. ; Niesen, T.
;Klenk, R.: Band alignment at sputtered ZnSx O1–x/Cu (In, Ga)(Se, S) 2 heterojunc-
tions. In: physica status solidi (RRL)-Rapid Research Letters (2012)
[123] Pflug, Andreas ; Sittinger, Volker ; Ruske, Florian ; Szyszka, Bernd ; Dittmar,
Georg: Optical characterization of aluminum-doped zinc oxide films by advanced disper-
sion theories. In: Thin Solid Films 455 (2004), S. 201–206
[124] Rudmann, D ; Bilger, G ; Kaelin, M ; Haug, F-J ; Zogg, H ; Tiwari, AN: Effects of
NaF coevaporation on structural properties of Cu (In, Ga) Se 2 thin films. In: Thin Solid
Films 431 (2003), S. 37–40
[125] Iwakuro, Hiroaki ; Tatsuyama, Chiei ; Ichimura, Shoji: XPS and AES studies on
the oxidation of layered semiconductor GaSe. In: Japanese Journal of Applied Physics 21
(1982), Nr. 1R, S. 94
[126] Klingshirn, C: ZnO: From basics towards applications. In: physica status solidi (b) 244
(2007), Nr. 9, S. 3027–3073
[127] Patnaik, Pradyot: Handbook of inorganic chemicals. Bd. 28. McGraw-Hill New York,
2003
[128] Tinoco, T ; Rinc´
on, C ; Quintero, M ; P´
erez, G S.: Phase diagram and optical
energy gaps for CuInyGa1- ySe2 alloys. In: physica status solidi (a) 124 (1991), Nr. 2, S.
427–434
[129] Bastek, J ; Stolwijk, NA ; Wuerz, R ; Eicke, A ; Albert, J ; Sadewasser, S: Zinc
diffusion in polycrystalline Cu (In, Ga) Se2 and single-crystal CuInSe2 layers. In: Applied
Physics Letters 101 (2012), Nr. 7, S. 074105
[130] Janowitz, Christoph ; Scherer, Valentina ; Mohamed, Mansour ; Krapf, Alica ;
Dwelk, Helmut ; Manzke, Recardo ; Galazka, Zbigniew ; Uecker, Reinhard ; Irm-
scher, Klaus ; Fornari, Roberto u. a.: Experimental electronic structure of In2O3 and
Ga2O3. In: New Journal of Physics 13 (2011), Nr. 8, S. 085014
[131] Janiak, Christoph ; Klap¨
otke, Thomas M. ; Meyer, Hans-J¨urgen ; Riedel, Erwin:
Moderne Anorganische Chemie. de Gruyter, 1999
BIBLIOGRAPHY 177
[132] Kronik, L. ; Cahen, D. ; Schock, H.W.: Effects of sodium on polycrystalline Cu (In,
Ga) Se2 and its solar cell performance. In: Advanced Materials 10 (1998), Nr. 1, S. 31–36
[133] Granath, Karin ; Bodeg˚
ard, Marika ; Stolt, Lars: The effect of NaF on Cu (In, Ga)
Se 2 thin film solar cells. In: Solar Energy Materials and Solar Cells 60 (2000), Nr. 3, S.
279–293
[134] Klein, Andreas: Energy band alignment at interfaces of semiconducting oxides: A review
of experimental determination using photoelectron spectroscopy and comparison with the-
oretical predictions by the electron affinity rule, charge neutrality levels, and the common
anion rule. In: Thin Solid Films 520 (2012), Nr. 10, S. 3721 – 3728
[135] Wenckstern, H von ; Splith, D ; Purf¨
urst, M ; Zhang, Z ; Kranert, Ch ; M¨
uller,
S ; Lorenz, M ; Grundmann, M: Structural and optical properties of (In, Ga) 2O3 thin
films and characteristics of Schottky contacts thereon. In: Semiconductor Science and
Technology 30 (2015), Nr. 2, S. 024005
[136] Oshima, Takayoshi ; Fujita, Shizuo: Properties of Ga2O3-based (Inx Ga1–x) 2O3 alloy
thin films grown by molecular beam epitaxy. In: physica status solidi (c) 5 (2008), Nr. 9,
S. 3113–3115
[137] Guo, Weiyan ; Guo, Yating ; Dong, Hao ; Zhou, Xin: Tailoring the electronic structure
of b-Ga 2 O 3 by non-metal doping from hybrid density functional theory calculations. In:
Physical Chemistry Chemical Physics 17 (2015), Nr. 8, S. 5817–5825
[138] Yanagi, Hiroshi ; Kawazoe, Hiroshi ; Kudo, Atsushi ; Yasukawa, Masahiro ; Hosono,
Hideo: Chemical design and thin film preparation of p-type conductive transparent oxides.
In: Journal of Electroceramics 4 (2000), Nr. 2-3, S. 407–414
[139] Chang, Pai-Chun ; Fan, Zhiyong ; Tseng, Wei-Yu ; Rajagopal, A ; Lu, Jia G.: β-Ga
2 O 3 nanowires: Synthesis, characterization, and p-channel field-effect transistor. In:
Applied Physics Letters 87 (2005), Nr. 22, S. 222102–222102
[140] Zhao, Junliang ; Sun, Xiao W. ; Tan, Swee T.: Bandgap-engineered Ga-rich GaZnO
thin films for UV transparent electronics. In: Electron Devices, IEEE Transactions on 56
(2009), Nr. 12, S. 2995–2999
[141] Orita, Masahiro ; Ohta, Hiromichi ; Hirano, Masahiro ; Narushima, Satoru ; Hosono,
Hideo: Amorphous transparent conductive oxide InGaO3 (ZnO) : a Zn4s conductor. In:
Philosophical magazine B 81 (2001), Nr. 5, S. 501–515
[142] Varley, JB ; Weber, JR ; Janotti, A ; Walle, CG Van d.: Oxygen vacancies and
donor impurities in β-Ga 2 O 3. In: Applied Physics Letters 97 (2010), Nr. 14, S. 142106–
142106
[143] Mohamed, M ; Irmscher, K ; Janowitz, C ; Galazka, Z ; Manzke, R ; Fornari, R:
Schottky barrier height of Au on the transparent semiconducting oxide β-Ga 2 O 3. In:
Applied Physics Letters 101 (2012), Nr. 13, S. 132106–132106
178 BIBLIOGRAPHY
[144] Chiril a, Adrian ; Reinhard, Patrick ; Pianezzi, Fabian ; Bloesch, Patrick ; Uhl,
Alexander R. ; Fella, Carolin ; Kranz, Lukas ; Keller, Debora ; Gretener, Christina
;Hagendorfer, Harald u. a.: Potassium-induced surface modification of Cu (In, Ga)
Se2 thin films for high-efficiency solar cells. In: Nature Materials 12 (2013), Nr. 12, S.
1107–1111
[145] Pistor, P ; Greiner, D ; Kaufmann, CA ; Brunken, S ; Gorgoi, M ; Steigert, A ;
Calvet, W ; Lauermann, I ; Klenk, R ; Unold, T u. a.: Experimental indication for
band gap widening of chalcopyrite solar cell absorbers after potassium fluoride treatment.
In: Applied Physics Letters 105 (2014), Nr. 6, S. 063901
[146] Delahoy, AE ; Ruppert, A. ; Contreras, M.: Charging and discharging of defect
states in CIGS/ZnO junctions. In: Thin solid films 361 (2000), S. 140–144
[147] Igalson, M ; Platzer-Bj¨
orkman, Charlotte: The influence of buffer layer on the
transient behavior of thin film chalcopyrite devices. In: Solar energy materials and solar
cells 84 (2004), Nr. 1, S. 93–103
[148] Serhan, Jaafar ; Djebbour, Zakaria ; Favre, Wilfried ; Migan-Dubois, Anne ; Darga,
Arouna ; Mencaraglia, Denis ; Naghavi, N ; Renou, Gilles ; Guillemoles, J-F ;
Lincot, D: Investigation of the metastability behavior of CIGS based solar cells with
ZnMgO–Zn (S, O, OH) window-buffer layers. In: Thin Solid Films 519 (2011), Nr. 21, S.
7606–7610
[149] Nishi, Yuki ; Miyata, Toshihiro ; Minami, Tadatsugu: The impact of heterojunction
formation temperature on obtainable conversion efficiency in n-ZnO/p-Cu 2 O solar cells.
In: Thin Solid Films 528 (2013), S. 72–76
[150] Wada, T ; Kohara, N ; Nishiwaki, S ; Negami, T: Characterization of the Cu (In, Ga)
Se 2/Mo interface in CIGS solar cells. In: Thin Solid Films 387 (2001), Nr. 1, S. 118–122
[151] Gostein, Michael ; Dunn, Lawrence: Light soaking effects on photovoltaic modules:
Overview and literature review. In: Photovoltaic Specialists Conference (PVSC), 2011
37th IEEE IEEE, 2011, S. 003126–003131
[152] Sundaramoorthy, R ; Pern, FJ ; DeHart, C ; Gennett, T ; Meng, FY ; Contreras,
M ; Gessert, T: Stability of TCO window layers for thin-film CIGS solar cells upon
damp heat exposures: part II. In: SPIE Solar Energy+ Technology International Society
for Optics and Photonics, 2009, S. 74120J–74120J
[153] Foster, LM ; Chandrashekhar, GV ; Scardefield, JE ; Bradford, RB: Phase
Diagram of the System Na2O
·
Ga2O3-Ga2O3 and its Relation to the System Na2O
·
Al2O3-
Al2O3. In: Journal of the American Ceramic Society 63 (1980), Nr. 9-10, S. 509–512
[154] Higashiwaki, Masataka ; Sasaki, Kohei ; Kuramata, Akito ; Masui, Takekazu ; Ya-
makoshi, Shigenobu: Gallium oxide (Ga2O3) metal-semiconductor field-effect transistors
on single-crystal beta-Ga2O3 (010) substrates. In: Applied Physics Letters 100 (2012),
Nr. 1, S. –
BIBLIOGRAPHY 179
[155] Soltz, D. ; Dagan, G. ; Cahen, D.: Ionic mobility and electronic junction movement in
CuInSe2. In: Solid State Ionics 28 (1988), S. 1105–1110
[156] Liao, Dongxiang ; Rockett, Angus: Cu depletion at the CuInSe 2 surface. In: Applied
physics letters 82 (2003), Nr. 17, S. 2829–2831
[157] Gartsman, Konstantin ; Chernyak, Leonid ; Lyahovitskaya, Vera ; Cahen, David
;Didik, Vladimir ; Kozlovsky, Vitaliy ; Malkovich, Roald ; Skoryatina, Elena ;
Usacheva, Valentina: Direct evidence for diffusion and electromigration of Cu in CuInSe
2. In: Journal of applied physics 82 (1997), Nr. 9, S. 4282–4285
[158] V´
ıllora, Encarnaci´on G ; Shimamura, Kiyoshi ; Yoshikawa, Yukio ; Ujiie, Takekazu
;Aoki, Kazuo: Electrical conductivity and carrier concentration control in β-Ga 2 O 3
by Si doping. In: Applied Physics Letters 92 (2008), Nr. 20, S. 202120–202120
[159] M¨
uller, Stefan ; Wenckstern, Holger von ; Splith, Daniel ; Schmidt, Florian ;
Grundmann, Marius: Control of the conductivity of Si-doped β-Ga2O3 thin films via
growth temperature and pressure. In: physica status solidi (a) 211 (2014), Nr. 1, S. 34–39
[160] Orita, Masahiro ; Hiramatsu, Hidenori ; Ohta, Hiromichi ; Hirano, Masahiro ;
Hosono, Hideo: Preparation of highly conductive, deep ultraviolet transparent β-Ga
2 O 3 thin film at low deposition temperatures. In: Thin Solid Films 411 (2002), Nr. 1,
S. 134–139
[161] Minami, Tadatsugu ; Nishi, Yuki ; Miyata, Toshihiro: High-efficiency Cu2O-based
heterojunction solar cells fabricated using a Ga2O3 thin film as n-type layer. In: Applied
Physics Express 6 (2013), Nr. 4, S. 044101
[162] Noufi, Rommel ; Zweibel, Kenneth: High-efficiency CdTe and CIGS thin-film solar
cells: highlights and challenges. In: Photovoltaic Energy Conversion, Conference Record
of the 2006 IEEE 4th World Conference on Bd. 1 IEEE, 2006, S. 317–320
[163] Pilania, Ghanshyam ; Wang, Chenchen ; Jiang, Xun ; Rajasekaran, Sanguthevar ;
Ramprasad, Ramamurthy: Accelerating materials property predictions using machine
learning. In: Scientific reports 3 (2013)
[164] Chiang, HQ ; Wager, JF ; Hoffman, RL ; Jeong, J ; Keszler, Douglas A.: High
mobility transparent thin-film transistors with amorphous zinc tin oxide channel layer. In:
Applied Physics Letters 86 (2005), Nr. 1, S. 013503
[165] Lee, Chiapyng ; Kuo, Yu-Lin: The evolution of diffusion barriers in copper metallization.
In: JOM 59 (2007), Nr. 1, S. 44–49
[166] Eller, Brianna S. ; Yang, Jialing ; Nemanich, Robert J.: Electronic surface and dielec-
tric interface states on GaN and AlGaN. In: Journal of Vacuum Science & Technology A
31 (2013), Nr. 5, S. 050807
[167] Li, Nola ; Park, Eun-Hyun ; Huang, Yong ; Wang, Shenjie ; Valencia, Adriana ;
Nemeth, Bill ; Nause, Jeff ; Ferguson, Ian: Growth of GaN on ZnO for solid state
lighting applications. In: SPIE Optics+ Photonics International Society for Optics and
Photonics, 2006, S. 63370Z–63370Z
180 BIBLIOGRAPHY
[168] Pankove, Jacques I.: Optical processes in semiconductors. Courier Corporation, 2012
[169] Tippins, HH: Optical absorption and photoconductivity in the band edge of β-Ga 2 O 3.
In: Physical Review 140 (1965), Nr. 1A, S. A316
[170] Deb, SK ; Chopoorian, JA: Optical Properties and Color-Center Formation in Thin
Films of Molybdenum Trioxide. In: Journal of Applied Physics 37 (1966), Nr. 13, S.
4818–4825
[171] Al-Kuhaili, MF ; Durrani, SMA ; Khawaja, EE: Optical properties of gallium oxide
films deposited by electron-beam evaporation. In: Applied physics letters 83 (2003), Nr.
22, S. 4533–4535
[172] Fang, GJ ; Li, DJ ; Yao, B-L: Effect of vacuum annealing on the properties of transparent
conductive AZO thin films prepared by dc magnetron sputtering. In: Physica status solidi
(a) 193 (2002), Nr. 1, S. 139–152
[173] Sheu, Jinn-Kong ; Shu, KW ; Lee, Min-Lum ; Tun, Chun-Ju ; Chi, Gou-Chung: Effect
of thermal annealing on Ga-doped ZnO films prepared by magnetron sputtering. In:
Journal of The Electrochemical Society 154 (2007), Nr. 6, S. H521–H524
[174] Yu, Xuhu ; Ma, Jin ; Ji, Feng ; Wang, Yuheng ; Zhang, Xijian ; Ma, Honglei: Influence
of annealing on the properties of ZnO: Ga films prepared by radio frequency magnetron
sputtering. In: Thin Solid Films 483 (2005), Nr. 1, S. 296–300
[175] Guill´
en, C ; Herrero, J: Optical, electrical and structural characteristics of Al: ZnO
thin films with various thicknesses deposited by DC sputtering at room temperature and
annealed in air or vacuum. In: Vacuum 84 (2010), Nr. 7, S. 924–929
[176] Ma, Quan-Bao ; Ye, Zhi-Zhen ; He, Hai-Ping ; Zhu, Li-Ping ; Huang, Jing-Yun ; Zhang,
Yin-Zhu ; Zhao, Bing-Hui: Influence of annealing temperature on the properties of
transparent conductive and near-infrared reflective ZnO: Ga films. In: Scripta Materialia
58 (2008), Nr. 1, S. 21–24
[177] Sch¨
onau, S ; Ruske, F ; Neubert, S ; Rech, B: Analysis of Urbach-like absorption
tails in thermally treated ZnO: Al thin films. In: Applied Physics Letters 103 (2013), Nr.
19, S. 192108
[178] Meyer, BK ; Sann, J ; Zeuner, A: Lithium and sodium acceptors in ZnO. In: Super-
lattices and Microstructures 38 (2005), Nr. 4, S. 344–348
[179] Kouklin, N ; Gupta, A ; Omari, M: Transition metal-doped ZnO nanowires: en route
towards multi-colour light sensing and emission applications. INTECH Open Access Pub-
lisher, 2010
[180] Neubert, Sebastian ; Wimmer, Mark ; Ruske, Florian ; Calnan, Sonya ; Gabriel,
Onno ; Stannowski, Bernd ; Schlatmann, Rutger ; Rech, Bernd: Improved conversion
efficiency of a-Si:H/
µ
c-Si:H thin-film solar cells by using annealed Al-doped zinc oxide as
front electrode material. In: Progress in Photovoltaics: Research and Applications 22
(2014), Nr. 12, S. 1285–1291
BIBLIOGRAPHY 181
[181] Orgassa, K ; Rau, U ; Schock, HW ; Werner, JH: Optical constants of Cu (ln, Ga)
Se/sub 2/thin films from normal incidence transmittance and reflectance. In: Photovoltaic
Energy Conversion, 2003. Proceedings of 3rd World Conference on Bd. 1 IEEE, 2003, S.
372–375
[182] Winsemius, Pieter ; Van Kampen, FF ; Lengkeek, HP ; Van Went, CG: Temperature
dependence of the optical properties of Au, Ag and Cu. In: Journal of Physics F: Metal
Physics 6 (1976), Nr. 8, S. 1583
[183] Shrotriya, Vishal ; Li, Gang ; Yao, Yan ; Chu, Chih-Wei ; Yang, Yang: Transition
metal oxides as the buffer layer for polymer photovoltaic cells. In: Applied Physics Letters
88 (2006), Nr. 7, S. 073508
[184] Simchi, Hamed ; McCandless, Brian E. ; Meng, T ; Boyle, Jonathan H. ; Shafar-
man, William N.: MoO3 back contact for CuInSe2-based thin film solar cells. In: MRS
Proceedings Bd. 1538 Cambridge Univ Press, 2013, S. 173–178
[185] Kyaw, AKK ; Sun, XW ; Jiang, CY ; Lo, GQ ; Zhao, DW ; Kwong, DL: An inverted
organic solar cell employing a sol-gel derived ZnO electron selective layer and thermal
evaporated MoO 3 hole selective layer. In: Applied Physics Letters 93 (2008), Nr. 22, S.
221107–221107
[186] Zhao, DW ; Liu, P ; Sun, XW ; Tan, ST ; Ke, L ; Kyaw, AKK: An inverted organic
solar cell with an ultrathin Ca electron-transporting layer and MoO 3 hole-transporting
layer. In: Applied Physics Letters 95 (2009), Nr. 15, S. 153304–153304
[187] Paulson, PD ; Stephens, SH ; Shafarman, WN: Analysis of Cu (InGa) Se2 alloy film
optical properties and the effect of Cu off-Stoichiometry. In: MRS Proceedings Bd. 865
Cambridge Univ Press, 2005, S. F1–4
[188] Dietrich, Jens ; Abou-Ras, Daniel ; Rissom, Thorsten ; Unold, Thomas ; Schock, H
;Boit, Christian: Compositional Gradients in Cu (In, Ga) Se Thin Films for Solar Cells
and Their Effects on Structural Defects. In: Photovoltaics, IEEE Journal of 2 (2012), Nr.
3, S. 364–370
[189] Hsu, W-W ; Chen, JY ; Cheng, T-H ; Lu, SC ; Ho, W-S ; Chen, Y-Y ; Chien, Y-J ; Liu,
CW: Surface passivation of Cu (In, Ga) Se¡ inf¿ 2¡/inf¿ using atomic layer deposited Al¡
inf¿ 2¡/inf¿ O¡ inf¿ 3¡/inf¿. In: Applied Physics Letters 100 (2012), Nr. 2, S. 023508–023508
[190] Bennett, HEJ ; Porteus, JO: Relation between surface roughness and specular re-
flectance at normal incidence. In: JOSA 51 (1961), Nr. 2, S. 123–129
[191] Carniglia, CK: Scalar scattering theory for multilayer optical coatings. In: Optical
Engineering 18 (1979), Nr. 2, S. 182104–182104
[192] Kluth, O ; Rech, B ; Houben, L ; Wieder, S ; Sch¨
ope, G ; Beneking, C ; Wagner,
H ; L¨
offl,A;Schock, HW: Texture etched ZnO: Al coated glass substrates for silicon
based thin film solar cells. In: Thin solid films 351 (1999), Nr. 1, S. 247–253
182 BIBLIOGRAPHY
[193] Zeman, M ; Van Swaaij, RACMM ; Metselaar, JW ; Schropp, REI: Optical modeling
of a-Si: H solar cells with rough interfaces: Effect of back contact and interface roughness.
In: Journal of Applied Physics 88 (2000), Nr. 11, S. 6436–6443
[194] H¨
upkes, J¨urgen ; Owen, Jorj I. ; Pust, Sascha E. ; Bunte, Eerke: Chemical Etching
of Zinc Oxide for Thin-Film Silicon Solar Cells. In: ChemPhysChem 13 (2012), Nr. 1, S.
66–73
[195] Thomas, Diana: In situ Spannungs-und Strukturanalyse von Molybd¨an-und CuInS2-
D¨unnschichten mittels R¨ontgendiffraktion, Universit¨atsbibliothek der Technischen Univer-
sit¨at Berlin, Diss., 2012
[196] Nell, Matthias E. ; Barnett, Allen M.: The spectral pn junction model for tandem
solar-cell design. In: Electron Devices, IEEE Transactions on 34 (1987), Nr. 2, S. 257–266
[197] Kijima, Shunsuke ; Nakada, Tokio: High-temperature degradation mechanism of Cu
(In, Ga) Se2-based thin film solar cells. In: Applied physics express 1 (2008), Nr. 7, S.
075002
[198] Ishizuka, Shogo ; Yamada, Akimasa ; Fons, Paul J. ; Shibata, Hajime ; Niki, Shigeru:
Structural tuning of wide-gap chalcopyrite CuGaSe2 thin films and highly efficient solar
cells: differences from narrow-gap Cu (In, Ga) Se2. In: Progress in Photovoltaics: Research
and Applications 22 (2014), Nr. 7, S. 821–829
[199] Schmid, M ; Krc, Janez ; Klenk, R ; Topic, Marko ; Lux-Steiner, M C.: Optical
modeling of chalcopyrite-based tandems considering realistic layer properties. In: Applied
Physics Letters 94 (2009), Nr. 5, S. 053507
[200] Supasai, T ; Rujisamphan, N ; Ullrich, K ; Chemseddine, A ; Dittrich, Th:
Formation of a passivating CH3NH3PbI3/PbI2 interface during moderate heating of
CH3NH3PbI3 layers. In: Applied Physics Letters 103 (2013), Nr. 18, S. 183906
Acknowledgement
First of all I would like to thank my wife Anna, who supported me throughout the course of
this PhD with her cheerfulness :)
I want to thank Christian Kaufmann, Thomas Unold and Hans-Werner Schock, who made
it possible for me to work in their groups and learn from their experience and knowledge.
Especially I would like to thank Christian, who was a great supervisor throughout the three
years. Special thanks also to Bernd Rech and Thomas Unold for making it possible for me to
join NREL for six months. At NREL I would like to thank David Ginley, Joe Berry and Miguel
Contreas for the support during this time. In addition I want to thank Susan Schorr, Michael
Powalla and Bernd Rech for reviewing this thesis.
At HZB I want to thank:
·Jakob Lauche and Dieter Greiner for keeping the PVD up and running,
·Britta H¨opfner and Jan Alsmeier for the great help with the XPS measurements,
·Daniel Abou-Ras and Norbert Sch¨afer for facilitating the SEM,
·Lars Steinkopf, Jo Klaer, Jan Schniebs and Karsten Prietzel for all the help inside the lab,
·Roland Mainz for facilitating the in-situ EDXRD measurements,
·Anja Scheu and Dieter Greiner for optimizing and running the GDOES measurements,
·Markus Wollgarten for doing the TEM images,
·Reiner Klenk and Susanna Harndt for the discussions and providing the In2S3evaporator,
·Carola Kirch and Michael Kirsch for the ZnO depositions,
·Justus Just for providing the EBIC fitting software,
·Alexander Redinger for preparing ZnS and ZnSe layers,
·Rutger Schlatmann, Sebastian Schmidt, Stephan Brunken and Dieter Greiner for proof reading,
·Sergej Levcenco for all the entertaining discussions on optics and capacitances,
·Joachim Liebich for the LabView support,
·Dieter Greiner for performing the Drude model fits to the optical data,
·Sebastian Fichter for introducing me to the program ChemSage,
·Steffen Kretzschmar for the support in the optical lab.
Also I want to thank the whole former team of the E-IT institute for the very pleasant three
years inside and outside the HZB.
Last but not least special thanks to my friends, especially Jan, Tobi and Thorsten, as well as my
family, especially my parents, my sister and my grandparents, to remind me of the life outside
of the PhD while keeping me motivated at the same time.
