Upload
others
View
4
Download
0
Embed Size (px)
Citation preview
NANOSTRUCTURED CARBON NANOTUBE SCHOTTKY JUNCTION SOLARCELLS
By
MAUREEN K. PETTERSON
A DISSERTATION PRESENTED TO THE GRADUATE SCHOOLOF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OFDOCTOR OF PHILOSOPHY
UNIVERSITY OF FLORIDA
2013
1
c© 2013 Maureen K. Petterson
2
To my family
3
ACKNOWLEDGMENTS
I would like to thank my adviser, Dr. Andrew Rinzler, for his unwavering support
and guidance throughout the past few years. His knowledge of both experimental and
theoretical aspects of physics research is comprehensive and inexhaustible; he was always
present to answer theoretical questions or help design and build experimental apparatus.
His ability to place contemporary research in historical perspective dissuades a myopic
view of graduate research, instead fostering an appreciation for the wide applicability of
experimental results. Dr. Rinzler gave me freedom to spend days, weeks, and months
tediously troubleshooting experiments and his encouragement and availability expedited
the successful results, while his patience and commiseration alleviated frustration over the
failures.
I’d like to thank my committee members for their ongoing support. Dr. Stanton and
Dr. Tanner were among the first professors I met within the physics department, and
I’m glad to have had their counsel over the past several years. Collaborations with Dr.
Hebard and his group have been enlightening and fruitful, culminating in some excellent
published work. Even before he was on my committee, Dr. Biswas lent a lively atmosphere
to the department without undermining the ethos of graduate research and discussions
with him regarding professional pursuits have been beneficial. I’m thankful to Dr. So for
his excellent objective input; his own research gave him particularly good insight into my
research projects.
I also owe a lot of gratitude to my labmates: Dr. Mitchell McCarthy, Dr. Bo Liu,
Dr. Rajib Das, Dr. Svetlana Vasilyeva, Dr. Max Lemaitre, Dr. Pooja Wadhwa, Dr. Evan
Donoghue, Dr. Po-Hsiang Wang, Yu Shen, Xiao Chen, Nan Zhao, Jie Hou, Matt Gilbert,
and Kyle Dorsey. Information gleaned from discussions with them greatly facilitated my
understanding of physics and chemistry. They were eager to help with any problems I
encountered and offered their expertise and guidance on many aspects of my experiments.
4
Not relegated to just professional cohorts, they also livened up the atmosphere and made
coming into lab an enjoyable experience.
I’d like to thank Darlene Latimer and Pam Marlin for wading through the academic
bureaucracy on my behalf and always ensuring I was on track to graduate. Darlene’s
genuine concern for all of the graduate students is heartwarming and her contributions to
the physics department are beyond measure. I’d like to thank Jay Horton, Tim Noland,
and the machine shop for providing their technical expertise to our lab and for fabricating
and fixing most (if not all) of our experimental apparatus. I’d also like to thank Pete
Axson and the rest of the electronics shop for keeping our solar simulator in working
order.
I’d like to thank my friends for offering unlimited and unconditional support,
encouragement, and advice; without whom I never would have appreciated the camaraderie
induced by sporting events or tolerated summers in Florida. I’d especially like to thank
Evan Donoghue for bestowing upon me the lessons he learned in his time during graduate
school and for helping me find the perfect balance between hard work and personal
development. Physicists are not generally known for their social aptitude or enthusiastic
inclusion of newcomers, but the friends I have gained through the department defy the
stereotypes and have made my graduate school years ones of both academic and personal
growth.
Finally, I’d like to thank my family. My parents, John Petterson and Loretta Kelley,
for supporting me emotionally, intellectually, and financially for the past three decades.
Both having experienced the trials and tribulations of earning an advanced degree, their
empathy and understanding was greatly appreciated. I owe much to my siblings, Alyssa
and Carey Petterson, for shaping my personality and ultimately putting me on the path to
being a successful physicist.
I’d like to acknowledge the National Science Foundation for funding support under
award ECCS 1232018.
5
TABLE OF CONTENTS
page
ACKNOWLEDGMENTS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
LIST OF TABLES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
LIST OF FIGURES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
LIST OF ABBREVIATIONS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
ABSTRACT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
CHAPTER
1 INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2 INTRODUCTION TO CARBON NANOTUBES . . . . . . . . . . . . . . . . . 21
2.1 History and Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212.2 Synthesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3 INTRODUCTION TO SOLAR CELLS . . . . . . . . . . . . . . . . . . . . . . . 30
3.1 Fundamentals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 303.1.1 Generation and Solar Spectrum . . . . . . . . . . . . . . . . . . . . 303.1.2 Recombination . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.1.2.1 Radiative . . . . . . . . . . . . . . . . . . . . . . . . . . . 333.1.2.2 Auger . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 333.1.2.3 Shockley Reed Hall . . . . . . . . . . . . . . . . . . . . . . 34
3.1.3 Characterization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 353.1.4 Series and Shunt Resistance . . . . . . . . . . . . . . . . . . . . . . 37
3.2 Theoretical Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 393.3 Types of Solar cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.3.1 P-N Junction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 413.3.2 Organic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 443.3.3 Photoelectrochemical Devices . . . . . . . . . . . . . . . . . . . . . 453.3.4 Multi-junction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 473.3.5 Schottky junction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 473.3.6 Inversion Layer Cells . . . . . . . . . . . . . . . . . . . . . . . . . . 48
4 INTRODUCTION TO SCHOTTKY BARRIERS . . . . . . . . . . . . . . . . . 50
4.1 Fundamentals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 504.1.1 Basic Schottky Model . . . . . . . . . . . . . . . . . . . . . . . . . . 504.1.2 Current Transport . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
4.1.2.1 Thermionic Emission . . . . . . . . . . . . . . . . . . . . . 524.1.2.2 Thermionic Field Emission and Field Emission . . . . . . . 524.1.2.3 Minority Carrier Injection . . . . . . . . . . . . . . . . . . 53
6
4.1.3 Beyond Schottky-Mott . . . . . . . . . . . . . . . . . . . . . . . . . 534.1.3.1 Fermi Level Pinning: Bardeen Model and Metal Induced
Gap States . . . . . . . . . . . . . . . . . . . . . . . . . . 554.1.3.2 Bond Polarization . . . . . . . . . . . . . . . . . . . . . . 56
4.2 Schottky Junction Solar Cells . . . . . . . . . . . . . . . . . . . . . . . . . 574.2.1 Historical Background . . . . . . . . . . . . . . . . . . . . . . . . . . 574.2.2 CNT on Silicon Schottky Junction Solar Cells . . . . . . . . . . . . 58
4.2.2.1 Experimental Details and Equipment . . . . . . . . . . . . 584.2.2.2 Electronic Gating . . . . . . . . . . . . . . . . . . . . . . . 604.2.2.3 Inversion Layer Modeling . . . . . . . . . . . . . . . . . . 60
5 NANOSTRUCTURING FOR ENHANCED LIGHT ABSORPTION . . . . . . . 66
5.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 665.2 Potassium Hydroxide Etching . . . . . . . . . . . . . . . . . . . . . . . . . 665.3 Silicon Nanowires . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
5.3.1 Procedure and Characterization . . . . . . . . . . . . . . . . . . . . 695.3.2 Integration in solar cells and initial performance . . . . . . . . . . . 71
5.3.2.1 Remote Gating . . . . . . . . . . . . . . . . . . . . . . . . 735.3.2.2 Passivation of Nanowire Sidewalls . . . . . . . . . . . . . . 745.3.2.3 SWNT film transfer on SiNW . . . . . . . . . . . . . . . . 76
5.3.3 Discussion of inversion layer in SiNWs . . . . . . . . . . . . . . . . . 785.4 Effect of Oxygen and Water on Device Performance . . . . . . . . . . . . . 79
5.4.1 Effect of ambient oxidation . . . . . . . . . . . . . . . . . . . . . . . 795.4.2 Reversible doping in ambient environment . . . . . . . . . . . . . . 815.4.3 Water vapor and oxygen contamination . . . . . . . . . . . . . . . . 82
5.4.3.1 CV measurements showing IL contamination . . . . . . . . 835.4.3.2 Exclusion on planar device . . . . . . . . . . . . . . . . . . 84
5.5 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
6 PASSIVATION OF SILICON . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
6.1 Atomic Layer Deposition of Al2O3 and HfO . . . . . . . . . . . . . . . . . 906.1.1 Al2O3 and HfO results . . . . . . . . . . . . . . . . . . . . . . . . . 93
6.2 Hydroquinone . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 966.3 Sulfur . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 986.4 Discussion and Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
7 ADDITIONAL PROJECTS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
7.1 TFSA Doping of Graphene-Si and Carbon Nanotube-Si Devices . . . . . . 1047.1.1 Graphene-Si Solar Cells . . . . . . . . . . . . . . . . . . . . . . . . . 1047.1.2 TFSA with carbon nanotubes . . . . . . . . . . . . . . . . . . . . . 107
7.2 Backside Doping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1087.3 Concluding Remarks and Path Forward . . . . . . . . . . . . . . . . . . . . 110
APPENDIX
7
A FULL SIMULATIONS FOR THE INVERSION LAYER CELL . . . . . . . . . 113
B SOLAR CELL PARAMETERS WITH INCREASING OXIDATION TIME . . . 114
REFERENCES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
BIOGRAPHICAL SKETCH . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
8
LIST OF TABLES
Table page
3-1 Current maximum efficiencies for various photovoltaic devices [1] . . . . . . . . . 49
4-1 Theoretical vs Experimental Schottky Barrier Heights: Barrier heights measuredat 300K, theoretical values determined from Schottky-Mott relation . . . . . . . 65
5-1 Performance for various film deposition techniques and thicknesses . . . . . . . . 88
6-1 Performance for ALD devices for VG = -1.0 V . . . . . . . . . . . . . . . . . . . 103
7-1 Performance summary for TFSA doped graphene and SWNT solar cells . . . . . 112
7-2 Performance for backside doped substrates . . . . . . . . . . . . . . . . . . . . . 112
9
LIST OF FIGURES
Figure page
2-1 Graphene lattice and geometric classifications for carbon nanotubes . . . . . . . 22
2-2 Band structure for carbon nanotubes . . . . . . . . . . . . . . . . . . . . . . . . 23
2-3 Density of states for semiconducting and metallic carbon nanotubes . . . . . . . 24
2-4 Electronic gating of SWNT film . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2-5 Dedoping of SWNT film during high temperature bake . . . . . . . . . . . . . . 29
3-1 The solar spectrum received both outside the Earth’s atmosphere and at thesurface of the Earth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3-2 Three recombination routes within semiconductors. . . . . . . . . . . . . . . . . 34
3-3 Maximum generated power density (blue box), defined by P = VMJM . . . . . . 36
3-4 Circuit equivalent showing series and shunt resistance. . . . . . . . . . . . . . . 38
3-5 The effects of series and shunt resistance on the J-V curve . . . . . . . . . . . . 39
3-6 The Shockley-Queisser limit showing the maximum theoretical efficiency as afunction of band gap for a single p-n junction solar cell. . . . . . . . . . . . . . 41
3-7 Schematic for a p-n junction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3-8 Band diagram of P-N junction with no bias, reverse bias, and forward bias . . . 43
3-9 Schematic of the bilayer and bulk heterojunction solar cells. . . . . . . . . . . . 45
3-10 Schematic of simple semiconductor/liquid junction solar cell showing redox reactionsoccurring both at the semiconductor surface and at a metal counterelectrode. . . 46
4-1 Schottky barrier band diagrams . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
4-2 Interface states shown in the realistic model of a Schottky junction. . . . . . . . 55
4-3 Carbon nanotube-silicon Schottky junction cell . . . . . . . . . . . . . . . . . . 59
4-4 Schematic and results for electronically gated SWNT-Si cell . . . . . . . . . . . 61
4-5 Schematic and performance for grid cell . . . . . . . . . . . . . . . . . . . . . . 62
4-6 Simulations showing inversion layer in silicon extending across entire surface inbetween carbon nanotube strips . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
5-1 KOH schematic and performance. . . . . . . . . . . . . . . . . . . . . . . . . . . 67
5-2 The mechanism for silicon nanowire growth . . . . . . . . . . . . . . . . . . . . 69
10
5-3 Silicon nanowires grown in an HF/AgNO3 solution . . . . . . . . . . . . . . . . 70
5-4 Orientation of silicon nanowires . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
5-5 Reflectance of the silicon nanowire substrates as compared to untextured silicon. 72
5-6 Initial performance of the SWNT-SiNW device . . . . . . . . . . . . . . . . . . 73
5-7 Schematic for remote gating and SEM of SWNT-SiNW active area . . . . . . . 74
5-8 J-V of a SWNT-SiNW device showing the effect of sidewall passivation via oxidationon the performance of the device. . . . . . . . . . . . . . . . . . . . . . . . . . . 75
5-9 J-V curves for VG = -1.0 V, 0 V, +1.0 V on the SiNW device. . . . . . . . . . . 77
5-10 Evolution of J-V curve with oxidation in ambient atmosphere. . . . . . . . . . . 80
5-11 Reversibility of the J-V curve upon alternating exposure to argon and ambientatmospheres . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
5-12 Cyclic voltammograms of the glassy carbon electrode in EMI-BTI ionic liquidat 50 mV
s. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
5-13 Stability of planar device with oxygen and water excluded by gating in inertatmosphere with VG=-1.0 V. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
5-14 Degradation of the planar SWNT-SiNW device upon exposure to atmospherewith VG=-1.0 V. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
6-1 ALD growth process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
6-2 SEM image of ALD deposition . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
6-3 J-V curves for the ALD Al2O3 coated SWNT/SiNW cell . . . . . . . . . . . . . 93
6-4 J-V curves for ALD SWNT-SiNW device vs device without ALD . . . . . . . . 94
6-5 J-V for the ALD HfO device showing a lowering JSC due to the high reflectanceof the device. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
6-6 Silicon substrate and hydroquinone molecule . . . . . . . . . . . . . . . . . . . . 97
6-7 J-V curve of the HQ treated planar cell before, during, and after electronic gatingwith EMI-BTI. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
6-8 J-V curves for sulfur and hydroquinone passivated devices . . . . . . . . . . . . 101
7-1 JV curve for the monolayer graphene device. . . . . . . . . . . . . . . . . . . . . 106
7-2 Schematic and performance for graphene PV cell. . . . . . . . . . . . . . . . . . 107
11
7-3 J-V curves showing effect of TFSA doping and subsequent gating on SWNT-Sidevice. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
7-4 Blistering on the surface of the silicon following a high temperature bake to dopethe backside. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
A-1 Modeling of the inversion layer at the silicon surface in the carbon nanotubegrid solar cell . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
B-1 FF, JSC , VOC , and PCE for a SWNT-SiNW device for various oxidation timesin the lab atmosphere. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
12
LIST OF ABBREVIATIONS
A Richardson Constant
ALD Atomic Layer Deposition
AM1.5G AirMass 1.5 Global
CNT Carbon Nanotube
CVD Chemical Vapor Deposition
Dn/p Electron or hole diffusion constant
DOS Density of States
EF Fermi energy
EMI −BTI 1-Ethyl-3-methylimidazolium bis(trifluoromethylsulfonyl)imide
FF Fill Factor
JM Current density at maximum power point
Jphoto Photocurrent density
JO Saturation current density/dark current
JSC Short circuit current density
MIGS Metal induced gap states
MIS − IL Metal-insulator-semiconductor inversion layer cell
PM Maximum power density
PCE Power conversion efficiency
φm Work function of metal
φBn0 Schottky barrier height to n-type semiconductor
φ0 Neutral level (above EV ) of interface states
∆ Potential across interfacial layer
χ Electron affinity of semiconductor
ψbi Built-in potential
δ Thickness of interfacial layer
q Electron charge
13
Qsc Space-charge density in semiconductor
Qss Interface-trap charge
QM Surface-charge density on metal
Dit Interface-trap density
ǫi Permittivity of interfacial layer (vacuum)
ǫs Permittivity of semiconductor
PLV Pulsed Laser Vaporization
RS Series Resistance
RSH Shunt Resistance
SBH Schottky Barrier Height
SRH Shockley-Reed Hall Recombination
SiNW Silicon Nanowire
SWNT Single-Wall Nanotube
µe/p Electron or hole mobility
VM Voltage at maximum power point
VOC Open Circuit Voltage
VG Gate Voltage
14
Abstract of Dissertation Presented to the Graduate Schoolof the University of Florida in Partial Fulfillment of theRequirements for the Degree of Doctor of Philosophy
NANOSTRUCTURED CARBON NANOTUBE SCHOTTKY JUNCTION SOLARCELLS
By
Maureen K. Petterson
August 2013
Chair: Andrew G. RinzlerMajor: Physics
This dissertation explores and exploits the physical processes uncovered during
experiments aimed at improving solar cell efficiency in a novel electronically gated solar
cell through surface texturing. Besides the increased device efficiency, the findings shed
light on the previous limitations in similar devices and may have implications for other
semiconductor based devices.
Silicon nanowires have long been known for their excellent antireflection properties,
but have suffered substantially from recombination at the surface. Here, we deposit a
disperse carbon nanotube network on the tips of a forest of vertical silicon nanowires and
exploit electronic gating in a novel Schottky junction solar cell. Previous experiments on
carbon nanotube- silicon solar cells made use of an ionic liquid to modulate the nanotube
Fermi level via electronic gating. This modulation changed the Schottky barrier height of
the device and decreased the carbon nanotube film resistance, leading to power conversion
efficiencies of up to 12% for a gate voltage of -0.75V. Further experiments uncovered
an additional mechanism in which the ionic liquid induced an inversion layer within the
silicon, greatly facilitating hole extraction by repelling electrons from the silicon surface
(and consequently reducing recombination). We exploit this induced inversion layer within
our silicon nanowire solar cells and show a greatly increased power conversion efficiency
exceeding 15%, the highest reported efficiency for silicon nanowire based devices to date.
15
We also investigate the physical and chemical processes responsible for degradation
in these devices. We show that contamination of the ionic liquid with oxygen or
water leads to redox reactions for gate voltages previously thought to be well within
the electrochemical window. We subsequently demonstrate that by excluding these
contaminants, stable performance of the electronically gated nanotube/silicon solar cell
is possible. Advanced passivation techniques are used to alleviate such degradation.
Specifically, deposition of aluminum oxide via atomic layer deposition was used to create
a high quality, conformal, dielectric layer that inhibits electrochemical reactions between
the ionic liquid and the silicon, leading to minimal reduction in performance as the gate
voltage is applied.
16
CHAPTER 1INTRODUCTION
Photovoltaics have been an extremely active area of research since the early 1970s, yet
only a few device structures have passed the test of affordability, longevity, and efficiency.
Volatile markets, decreasing resources, and a penchant for innovation fuel the continued
research into different novel architectures utilizing a variety of organic and inorganic
materials. Concomitant with the increase in solar cell efficiency is a deeper understanding
of the underlying physical processes present in such devices, something which also has
more general scientific value. The high efficiency of the solar cells presented in this thesis
was achieved by efforts to understand the underlying physics of the devices and using that
knowledge to improve light absorption while minimizing losses and degradation. The high
efficiencies were realized through multiple methods, as discussed in detail in Chapters 4-7.
A brief summary of the dissertation is presented below.
Chapter 2 starts off by discussing the theoretical background of single wall carbon
nanotubes (SWNT), with particular emphasis on the ability to modulate the Fermi level
of the SWNTs due to their low density of states. This modulation can be experimentally
verified by observing the change in transmittance of the film during electronic gating, as
demonstrated in the work of Dr. Zhihong Chen and Dr. Zhuangchun Wu, who led that
effort in the Rinzler group.[8]
Chapter 3 gives a brief introduction to solar cells. Testing, characterization, and
different types of solar cells are presented, along with some of the challenges confronting
researchers in their pursuit to develop high efficiency devices. Historical information on
solar cells is mentioned with the purpose of showing how work described later in this
thesis can address and solve problems encountered in the photovoltaic devices developed in
the 1970s.
Chapter 4 presets a more thorough description of Schottky junction solar cells,
with emphasis on the physics of Schottky barriers, including how surface preparation
17
affects device performance. The carbon nanotube-silicon Schottky junction solar cell
is introduced, along with the work done by Dr. Pooja Wadhwa in which such a device
was electronically gated. Modulating the gate voltage between the active area film and
a gate film modulates the SWNT Fermi level, changing the Schottky built-in potential
and modulating device performance. Lastly, the inversion layer grid cell is described. This
device, in which I made my first contributions to this class of devices, demonstrated a
new mechanism by which the ionic liquid used during electronic gating of the SWNT film
simultaneously forms an inversion layer within the silicon.[10] This allowed for efficient
collection of photogenerated carriers far from the SWNT gridlines, boosting the efficiency
of the device from 10.9% to 12%.
Chapter 5 details my work on nanostructuring the silicon surface to improve
light absorption. This work was motivated by the results of the inversion layer device
discussed in Chapter 4. The ability to induce an inversion layer within the silicon no
longer constrains us to have the carbon nanotube film touching the entire silicon surface,
allowing exploration of alternative architectures. Silicon nanowires (SiNW), known for
their excellent anti-reflection properties, were integrated into a SWNT-SiNW device and
took full advantage of the ionic liquid-induced inversion layer along the nanowire sidewalls.
The greatly increased surface area of the nanowires required modifications to the solar
cells, specifically an increased surface area gate film to compensate for the ions needed
to induce the inversion layer, and a SWNT spray deposited film to improve connection
between the silicon nanowires and the carbon nanotubes. Integration of these two led to a
greatly improved power conversion efficiency of over 15%, the highest PCE for any silicon
nanowire device to date.
The latter part of Chapter 5 addresses the stability of these devices during electronic
gating. Experiments on the carbon nanotube-silicon Schottky junctions solar cells
showed a reduction of the PCE as the device was electronically gated. Characteristics
of degradation suggested that redox reactions were facilitating oxidation of the silicon
18
surface, forming a barrier to carrier extraction and decreasing performance. The greatly
increased surface area of the silicon nanowires, versus the planar devices, exhibited a
greater degradation (and hence more redox reactions), as evidenced by an increased
parasitic gate current. Experiments to test the electrochemical window of our ionic liquid
showed a substantial reduction of the window due to contamination of water and oxygen.
Aware that testing in ambient atmosphere would lead to immediate contamination,
I tested a planar device in a glovebox using dried ionic liquid and observed stable
performance over numerous hours. A dramatic reduction in gate currents indicating
negligible redox reactions conclusively demonstrated that the degradation was due to
contamination of water and oxygen at the silicon surface.
Chapter 6 describes my work to reduce degradation of the SWNT-SiNW devices
during electronic gating. Though simple encapsulation can eliminate degradation, I
explored both atomic layer deposition and chemical passivation as a means to elicit stable
performance. Hydroquinone and sulfur passivation led to improved device performance
prior to electronic gating, but ultimately proved to be incompatible with the ionic
liquid. Atomic layer deposition of aluminum oxide on the fully fabricated device limited
contact between the ionic liquid and the silicon surface during electronic gating without
sacrificing the inversion layer. This reduced contact limited redox reactions and lowered
the gate current by a factor of 60. Though the device was tested in the atmosphere with
”contaminated” ionic liquid, the ALD layer improved the stability of the devices and still
produced a high power conversion efficiency of 14.8%.
Finally, Chapter 7 discusses two side projects: graphene-silicon Schottky junction
solar cells and backside doping of the silicon substrates. The former demonstrated a
greatly enhanced power conversion efficiency upon introduction of the organic dopant,
bis(trifluoromethylsulfonyl)amide (TFSA). The improvement in efficiency is attributed to
an increase in the Schottky barrier height, decrease in series resistance, and the ability of
the TFSA to act as an anti-reflection layer. Lastly, backside doping was found to improve
19
the power conversion efficiency by limiting recombination at the back contact. A spin
on dopant deposited onto the backside of the silicon substrates produced a think, highly
doped region on the backside of the silicon. The power conversion efficiency of the planar
devices was improved from 9.7% to 13.4%.
20
CHAPTER 2INTRODUCTION TO CARBON NANOTUBES
2.1 History and Structure
First structurally interpreted in 1991 by Sumio Iijima at the Nippon Electric
Company (NEC), carbon nanotubes have been the focus of intense research and
development for the past two decades. Fundamental properties and novel applications
have been explored in the physical sciences since their discovery, while within the medical
and biological sciences much work has been done to incorporate carbon nanotubes into
various devices ranging from prosthetics to molecular transporters.[2–5] The wide range of
potential applications is derived from the unique electrical, physical, and optical properties
possessed by these fullerenes. A quasi-one dimensional structure, carbon nanotubes can be
thought of as a sheet of graphene rolled into a seamless tube with diameters approximately
1-10 nm and aspect ratios up to 105.
The electronic and optical properties of carbon nanotubes can be calculated from the
band structure of graphene due to the local structural similarity of the two. Graphene is a
simple two dimensional hexagonal lattice composed of sp2 bonded carbon atoms. A long,
narrow rectangular strip cut from this lattice and rolled up along the narrow dimension
(with bonds reformed across the seam) generates the structure of a single wall nanotube
(SWNT). Depending upon the orientation of the strip direction relative to the graphene
lattice SWNTs of three distinct structural classifications can form: armchair, zigzag, or
chiral. Aside from their geometric classifications, SWNTs can be subdivided into either
metallic or semiconducting types based on the nanotube n,m index (defined below).[6, 7]
Figure 2-1A shows the graphene lattice and corresponding unit cell with primitive
vectors, −→a1 and −→a2 along with the chiral vector−→C . Defined as
−→C = n−→a1 +m−→a2 for integer n, m, (m < n), (2–1)
21
A Graphene lattice B Carbon nanotube geometries
Figure 2-1. Graphene lattice and geometric classifications for carbon nanotubes. Figure A:Graphene lattice showing primitive vectors −→a1 and −→a2 that comprise the unitcell. Also shown is the chiral vector that determines nanotube type. Figure B:The three geometric types of carbon nanotubes. Reprinted with permissionfrom R. Saito, G. Dresselhaus, and M. S. Dresselhaus, Physical properties ofcarbon nanotubes (Imperial College Press, 1998)
the chiral vector is a linear combination of primitive vectors governing the directions in
which a graphene sheet can be rolled up, subsequently determining the specific properties
of the nanotube; n=m results in an armchair nanotube, m=0 corresponds to a zigzag
nanotube, and all other combinations of n,m result in a chiral nanotube. The conditions
for a metallic nanotube require that (n − m) = 3j, where j is an integer. Consequently,
all armchair nanotubes are metallic while zigzag and chiral nanotubes can be either type;
these selection rules lead to an overall 2:1 ratio of semiconducting to metallic nanotubes
across all allowed n, m.
The energy dispersion relations and density of states (DOS) for nanotubes are derived
by placing the appropriate circumferential boundary conditions on the energy dispersion
relation for graphene and solving for the allowed k values and associated energy states.
For a single sheet of graphene,
22
Eg2D(kx, ky) = ±t{
1 + 4 cos
(√3kxa
2
)
cos
(kya
2
)
+ 4 cos2(ky
a
)}1
2
, (2–2)
where t = -3.033eV and kx and ky correspond to the x and y axes in reciprocal space.[7]
The boundary conditions require ka to be multiples of π, with the exact dispersion
relation for a given nanotube depending on the chiral vector. Figure 2-2 shows the band
structure for both a metallic and semiconducting nanotube. Shown in Figure 2-3 is the
density of states for both a semiconducting and metallic nanotube, with the graphene
DOS overlayed as a dotted line.
Figure 2-2. Band structure for carbon nanotubes. Shown is a metallic a) (5,5) and b) (9,0)nanotube and c) a semiconducting (10,0) nanotube. Reprinted with permissionfrom Saito, et al. Physical properties of carbon nanotubes (Imperial CollegePress, 1998)
The DOS of graphene goes to exactly zero at the six Dirac points in the Brillioun
zone, making graphene a zero gap semiconductor. The cyclical symmetry of carbon
nanotubes restricts the allowed wave vectors along the circumference. Divergences of the
density of states, called Van Hove singularities, are due to the one dimensionality of the
carbon nanotubes and are the visible spikes in Figure 2-3.
23
Figure 2-3. The density of states for a semiconducting nanotube (left), and a metallicnanotube (right). The dotted line overlayed in both images is the density ofstates for graphene. Note the low DOS for the metallic nanotubes near theFermi energy. γ0 is 3.13 eV and corresponds to the C-C tight binding overlapenergy. Reprinted with permission from Saito, et al. Physical properties ofcarbon nanotubes (Imperial College Press, 1998)
Carbon nanotubes are notable for their relatively low density of states and easily
manipulated Fermi level. Both electronic gating and chemical charge transfer doping can
be utilized to shift the carrier concentration of the nanotubes, allowing integration into
devices that demand control over conductance.[8–10] This change in electronic populations
concomitant with a Fermi level modulation can be demonstrated by a change in the
film transmittance. Figure 2-4B shows the experimental set up for demonstrating such
modulation via electronic gating. Two carbon nanotube films are transferred to a quartz
substrate and then baked at 600C to dedope the films, after which the entire substrate
is submerged in a liquid electrolyte and a gate voltage is applied between the two films,
driving the ions in the electrolyte to either film in a response to Coluombic forces.
Inclusion of ionic liquid boosts the capacitance of the nanotube film, akin to inserting a
dielectric between a parallel plate capacitor. Figure 2-4A shows the transmittance data
taken with a UV-Vis spectrophotometer as a gate voltage between the test film and gate
film was held at incremental voltages from -1.8V to +1.8V. The S1, S2, and M1 peaks
corresponding to electronic transitions between Van Hove singularities. As a negative gate
24
voltage is applied, the test film becomes more p-doped as the Fermi level is pushed farther
from vacuum, resulting in an increased transmittance as electrons are depleted from the
Van Hove singularities that comprise the ground state for the corresponding transition.
As a positive gate voltage is applied, the trend reverses and the transmittance decreases.
This contradicts the intuitive idea that any applied gate voltage (positive or negative) will
result in an increased transmittance due to a mid-gap Fermi level at zero applied voltage.
In the idealized case, any applied gate voltage will decrease the number of available states
for electronic transitions, lowering the transmittance and leading to ambipolar behavior.
The experimental results seem to belie the theory. Though initially attributed to possible
contamination effectively p-doping the carbon nanotubes, we now believe this discrepancy
is attributed to electrons from the carbon nanotubes being donated to an oxygen/water
redox couple, as was studied in depth by Martel in 2009.[11] The hydrophilic quartz
substrate and water contaminated ionic liquid provide ample amounts of water and oxygen
to facilitate electrochemical reactions, leading to p-type behavior
A Electronic Gating B Experimental Setup
Figure 2-4. Electronic gating of SWNT film. Figure A: Change in transmittance of a 45nmthick SWNT film with various applied gate voltage. Figure B: Experimentalset-up for transmittance measurements.
25
2.2 Synthesis
Carbon nanotubes can be synthesized through four different methods: arc discharge,
pulsed laser vaporization (PLV), plasma torch, and chemical vapor deposition (CVD).
The initial procedure for carbon nanotube fabrication, arc discharge growth is achieved by
applying a potential between two graphite/carbon electrodes with the resultant discharge
heating the carbon targets, catalyzing the formation of carbon nanotubes and other
fullerenes.[12] PLV also uses a high energy beam to form carbon nanotubes, but employs a
green and infrared laser to ablate a carbon target.[13, 14] Environmental conditions within
the carbonaceous vapor affect the specific properties of the carbon nanotubes; temperature
relates to the diameter while the presence of metal catalysts particles facilitate formation
of single wall nanotubes instead of multi-walled nanotubes.[15–17] During the growth,
several individual carbon nanotubes adhere together via Van der Walls forces to form
nanotube bundles tens of nanometers in diameter. Plasma torch growth is another
permutation of growth via thermal decomposition of catalyst precursors in a carbonaceous
gas. In this method, a carbon source and a metal catalyst source are simultaneously fed
through a plasma torch, producing carbon nanotubes in the heated vapor.[18]
CVD growth comprises a few different methods: conventional CVD, plasma enhanced
CVD (PECVD), HiPCo (high-pressure conversion of carbon monoxide), Ferrocene
injection, and floating catalyst method, to name a few. The first generation of CVD
growth developed by Nikolaev, et al. used flowing carbon monoxide in conjunction with a
metal catalyst containing gas in a continuous flow reactor. Growth of carbon nanotubes
occurred though thermal decomposition of the metal catalyst within the heated carbon
monoxide flow, after which they would adhere to the sides of the quartz tube. [19].
Alternatively, metal catalyst particles can be pre-deposited onto a carrier substrate and
then placed into the growth chamber whereupon the carbon containing reactant gas would
be fed. Decomposition in the high temperature growth chamber would be followed by
carbon nanotube growth facilitated by the metal particles on the carrier substrate.[20–22]
26
The carbon nanotubes used for the experiments described in the following chapters
were synthesized by PLV and subsequently underwent an extensive purification process
to eliminate residual amorphous carbon, catalyst particles, and other non-nanotube
contaminants. A 2.4M nitric acid reflux removed amorphous carbon (and other forms
of carbon exhibiting relatively weak sp3 bonding) and metal impurities in addition to
p-doping the carbon nanotubes. Multiple centrifugations at 6000 RPM followed by
decantation of the acidic supernatant neutralized the carbon nanotube/acid solution and
allowed their dispersion in a 1% Triton-X surfactant solution. Next, crossflow filtration
eliminated reaction products and fine particulates by repeatedly passing the solution
through long, hollow fibers with small pores along the sidewalls. These small pores allow
the passage of small particulates while preventing the permeation of carbon nanotubes.
This filtration continued until the permeate is clear, indicating a majority of small
particles have been removed from the solution. An additional centrifugation (6000-10000
RPM), this time retaining the supernatant, separated out contaminants with a density
greater than that of the surfactant buoyed carbon nanotubes.. Lastly, a filtration though
a 650 µm membrane removed large particulates not broken down by the nitric acid reflux
or eliminated by previous steps, ultimately producing a purified, surfactant based carbon
nanotube suspension.[8, 23–25]
In the studies discussed in this work the SWNTs were typically used in the form
of thin transparent films. Such film formation proceeded as follows. Starting with the
purified solution, the carbon nanotubes are vacuum filtered onto a mixed cellulose
(MCE) membrane and copiously rinsed with deionized water to remove residual Triton-X
surfactant. After drying under an incandescent lamp, the carbon nanotube film is ready
to be transferred to a substrate or stored in an inert atmosphere until needed. To transfer
to a substrate (such as glass, ITO, PET, silicon, etc), the nanotube film is first placed
against the substrate (MCE side up), on top of which is placed a porous Teflon membrane
followed by a hydrated sheet of porous plastic. This assembly is sandwiched between
27
two aluminum plates that are clamped together to supply pressure between the carbon
nanotube film and the substrate. Upon being placed in a 100C oven, the water in the
porous plastic evaporates and wets the nanotube film. As the steam slowly diffuses out
of the assembly with continued heating, the nanotubes are brought into intimate contact
with the substrate and are thereby retained via Van der Walls forces. After several hours,
the assembly is removed and the substrate/film is placed in an acetone vapor bath to
dissolve the MCE membrane, leaving behind the carbon nanotube film. Subsequent liquid
acetone baths ensure complete removal of the cellulose, after which the substrate is placed
in an isopropol bath (a generally cleaner solvent than acetone, in which the latter is
miscible), removed, and thoroughly dried in a nitrogen stream.
As mentioned above, the carbon nanotubes are charge transfer p-doped during the
nitric acid reflux. Nitrogen based cations, NOx, intercalate the nanotube bundles and
sequester electrons, effectively shifting the Fermi level towards the valence band edge
and hole doping the nanotubes.[26] This results in a change in transmittance in both the
optical and IR regimes that can be observed with a UV/VIS/NIR spectrophotometer.
Baking at 600C provides enough thermal energy to de-adsorb the dopant species, resulting
in a lower conductivity and reduced transmittance at wavelengths corresponding to Van
Hove singularities, as shown in Figure 2-5.
Controlled doping combined with the ability to modulate the Fermi level via
electronic gating make it possible to tailor carbon nanotube’s electrical and optical
properties on an as-needed basis. The photovoltaic devices presented in this dissertation
take full advantage of this malleability and exhibit superior performance compared to
similar photovoltaic devices. Additionally, information regarding interactions between
the carbon nanotubes and other materials (crystalline semiconductors and electrolytes)
has been gleaned through experiments aimed at improving PV efficiency, elucidating
fundamental physical interactions within the device.
28
Figure 2-5. Dedoping of SWNT film during high temperature bake. The shift intransmittance for a 45 nm single walled nanotube film baked at 600C,indicating dedoping. The inset shows the density of states of a semiconductingand metallic nanotube, with the shaded regions indicating electronicpopulations (in this case, p-type doping).
29
CHAPTER 3INTRODUCTION TO SOLAR CELLS
3.1 Fundamentals
A rise in gas prices coupled with heightened environmental awareness has prompted
increasing amounts of research into alternative energy sources. Though much research
has been put into the advancement of various types of solar cells, silicon based devices
continue to dominate the market. Silicon is readily available, well understood, and a staple
of the electronics manufacturing industry. Combined with nuclear, wind, and hydro, solar
power promises to become one of the front contenders in the energy market. It’s modest
and localized installation requirements make it attractive for individuals who desire
more control over their energy use or for urban locations with minute amounts of spare
real estate. Other forms of alternative energy require vast tracts of land or are highly
geographic dependent, while solar power can be efficiently harvested most anywhere south
of the mid-latitudes. A single type of solar cell will not be sufficient to meet the needs of
every application, and as such continuing research in all types of solar cells is vital to the
future of alternative energy.
Unlike a battery, which supplies a constant voltage, a solar cell is more akin to a
current generator that is limited by the photon flux incident on the active area of the
device. Excluding those devices exhibiting multiple exciton generation, every photon
with energy above the bandgap that is absorbed within the active material generates an
electron-hole pair. The primary aims of R & D in this area are to both absorb all photons
hitting the solar cell and to collect all the photogenerated carriers. Table 3-1 at the end of
the chapter shows the current progress of various types of photovoltaic devices.
3.1.1 Generation and Solar Spectrum
Not all semiconducting materials are suitable for broadband absorption of the solar
spectrum; a large bandgap is ideal for collecting high energy photons, while a smaller
bandgap wastes much of the energy of higher energy photons as the photogenerated
30
electrons decay to the bottom of the conduction band by generating heat within the cell.
As such, small bandgap materials are generally relegated to an IR absorption layer within
tandem cells. The sun’s spectrum is that of a blackbody with an average temperature of
5800K, however the radiation received at the top of the Earth’s atmosphere differs from
that in space due to enhanced absorption and scattering from gas and water vapor and the
oblique angle at which the radiation hits the terrestrial surface. Shown in Figure 3-1 are
the relevant solar spectra at the top and bottom of earth’s atmosphere compared to a 5250
C black body spectrum..[27]
Figure 3-1. The solar spectrum received both outside the Earth’s atmosphere and at thesurface of the Earth. Image created by Robert A. Rohde for Global WarmingArt.
Comparisons of solar cell efficiency are only valid if all the devices have been tested
under the same conditions. In order to create consistency among reported performance,
the community has agreed upon a solar spectrum corresponding to the average radiation
received at the mid-latitudes, designated Air Mass 1.5 Global (AM1.5G). Equivalent to
roughly 100 mWcm2 , this spectrum approximates the solar irradiance at the Earth’s surface
when the sun is 48.2◦ off zenith and accounts for scattering by the atmosphere. AM0
31
is the solar irradiance at the top of the atmosphere (roughly equivalent to a blackbody
spectrum at 5800K), a higher fluence that overestimates photon flux for solar installations
in the United States.
A photon incident on a solar cell that is absorbed within the semiconducting
material will excite an electron to the conduction band, leaving behind a hole in the
valence band. Extraction of this electron only occurs if there is a driving force moving
the electron or hole towards the electrodes. Extraction of the charges is predicated on
transport of the charge to the electrodes, necessitating an internal mechanism to move the
photogenerated carriers. While both drift (due to an internal electric field) and diffusion
(due to concentration gradient) can contribute to the current, the former is more efficient
at quickly separating electron hole pairs and ferrying them to their irrespective electrodes.
This driving force is found within diodes, and the total current in a solar cell can be
represented by the Shockley diode equation with an additional term corresponding to the
photocurrent.[27]
Jtotal = Jphoto + Jdiode (3–1)
Jdiode = Jo
[
exp
(qV
kT
)
− 1
]
,
where Jo is referred to as the dark current and yields information regarding recombination
at the junction (a more detailed analysis of the Shockley equation is given in Chapter 3).
The binding energy of the electron hole pair varies depending on the absorber
material: in silicon it is only 14.7 meV, low enough for the pair to be dissociated at room
temperature. In contrast, the binding energy in organic materials is on the order of 0.5 -
1 eV, necessitating a mechanism for dissociation in order to separate and then extract the
photogenerated carriers.[28, 29]
3.1.2 Recombination
Recombination is the combining of a photogenerated electron and hole. It remains the
primary loss mechanism within solar cells; carriers that recombine cannot be extracted as
32
usable electrical energy and overall cell efficiency decreases. There are three recombination
types pertinent to solar cells: Radiative, Auger, and Shockley-Reed-Hall (SRH). The
dominant source of recombination is dependent on the type of material used: electron
hole pairs in direct band gap semiconductors are predisposed to combine radiatively,
while carriers in indirect band gap semiconductors are more likely to under go SRH
recombination. Auger recombination is possible in both types of semiconductors, though
typically negligible at most doping densities.[30] The type of band gap isn’t the only
contributing factor to recombination rates; crystallinity plays a large role as well.
Amorphous silicon, which can possess hugely varying densities of defects and dangling
bonds, has carrier lifetimes on the order of 10−9 s, far shorter than the value for it’s
crystalline counterpart, which has a minority carrier lifetime of 2.5 x 10−3 s.[30, 31]
3.1.2.1 Radiative
Radiative recombination is a two body interaction in which an electron combines
with a hole and produces a photon. Conservation of crystal momentum dictates that
this process only occur in a direct band gap semiconductor where a direct band to
band transition may occur without an additional interaction. Hence, direct band gap
materials are prone to having high levels of recombination in the bulk, necessitating
thinner semiconductor layers in order to extract photogenerated carriers before they
recombine. This results in a relatively short diffusion length and smaller carrier lifetime.
One of the most popular direct bandgap semiconductors, GaAs, has a minority carrier
lifetime of approximately 10−8 s, several orders of magnitude shorter than that of silicon.
Consequently, direct band gap materials are often relegated to multi-junction solar cells.
Shown in Figure 3-2 are the schematics for the three types of recombination.
3.1.2.2 Auger
Unlike radiative recombination, Auger recombination is a three carrier interaction
that can occur in both direct and indirect band gap semiconductors. An electron and
hole recombine with a kinetic energy transfer to a third carrier which is subsequently
33
Figure 3-2. Three recombination routes within semiconductors. Adapted from Principlesof Semiconductor Physics, Van Zeghbroeck, B. unpublished.
excited to a higher energy level. The third carrier then relaxes down via phonon emission.
Although this process is permitted in all semiconductors, it requires high carrier densities
to contribute significantly to recombination, especially in direct band gap semiconductors.
Though the majority of recombination in indirect band gap semiconductors is due to
defects (mostly at the surface), a majority of recombination in the bulk can be attributed
to Auger if the defect density is low.
3.1.2.3 Shockley Reed Hall
Shockley Reed Hall (SRH) recombination is recombination of an electron and hole
that is catalyzed by a defect or trap state. This remains the dominant recombination
mechanism in indirect band gap semiconductors, as purely radiative recombination is
impossible and Auger is only likely with high carrier densities. Defects in the crystal
create new energy states, allowing an electron and hole to recombine radiatively or
non-radiatively while still conserving crystal momentum. Trap states can ”trap” carriers
for a finite amount of time, during which they can recombine with another carrier or
be thermally excited out of the trap. This recombination mechanism explains the long
diffusion length in indirect band gap monocrystalline semiconductors, as photogenerated
34
carriers are lost to recombination only at the surface. This allows relatively thick solar
cells modules capable of absorbing lower energy photons. Comparatively, amorphous and
polycrystalline silicon possess a high density of grain boundaries that act as defects and
facilitate recombination, limiting the optimal absorber thickness.
The recombination rate at the surface can be quantified by the recombination velocity
parameter, S:
Sn = vthσsNn (3–2)
where vth is the thermal velocity (typically ∼107 cms), σs is recombination cross section
(typically ∼10−15 cm2), and Nn is the density of trap states at the semiconductor surface.
A higher surface recombination velocity indicates increased carrier recombination.[30]
Highly quality passivated silicon can achieve surface recombination velocities as low as
0.25 cm2
vsfor undoped silicon.[32], as compared to 2590 cm2
vsfor polycrystalline silicon.[33]
3.1.3 Characterization
In lieu of attaching a variable resistive load to the solar cell to calculate efficiency, the
device is connected to a power source and the voltage is ramped from negative to positive
voltages. The current at each voltage is recorded and plotted as a J-V curve. Each point
in the fourth quadrant represents a particular load resistance and an identical J-V curve
can be generated in that quadrant by connecting the solar cell to a variable resistor and
recording the current and voltage at each resistance value. Though the device generates
power only in the fourth quadrant, biasing the device over a full range of voltages yields
important information about the solar cell in regards to both photogeneration and diode
behavior.
The four parameters used to evaluate a solar cell’s performance are power conversion
efficiency (PCE), short circuit current (JSC), open circuit voltage (VOC), and fill factor
(FF). The bias voltage where the photocurrent is equal and opposite to the diode current
(ie, Jtotal = 0) is called the open circuit voltage, VOC. The output current when Vbias= 0 V
is defined as the short circuit current density, JSC . An overall measure for the efficiency of
35
Figure 3-3. Maximum generated power density (blue box), defined by P = VMJM
a cell, PCE is defined as a ratio of the maximum power generated by the cell to the power
of the incident radiation on the active area of the device.
The short circuit current density is taken directly from the J-V curve and is
approximately equal to the photocurrent (the dark diode current is generally orders of
magnitude smaller). The VOC is defined as the voltage at which the photocurrent is equal
and opposite to the diode current, Jphoto = Jdiode:
Jphoto = JO
[
exp
(qVOC
kT
)
− 1
]
(3–3)
Solving for VOC yields:
VOC =kT
qln
(Jphoto
JO+ 1
)
(3–4)
The VOC is extracted directly from the J-V and can be used to calculate other
parameters, such as the dark current (and subsequently Schottky barrier height).This
equation elucidates the factors contributing to a high or low VOC , namely and increase in
36
short circuit current and/or a decrease in dark current. Increasing VOC by increasing the
light intensity is a well known phenomena exploited through solar concentrators, though
the semiconducting material must have high enough mobilities to avoid carrier saturation
(and inversion). The inverse logarithmic dependence on dark current points to a reduction
in junction recombination as another way to increase the VOC .
The fill factor is a measure of how much the solar cell functions as an ideal diode,
with a FF of 1 corresponding to a completely square shaped J-V curve in the fourth
quadrant. Table 3-1 in the beginning of the chapter shows high performing solar cells
possessing a FF of approximately 0.8-0.9.
FF =VMJM
VOCJSC(3–5)
VM and JM are the voltage and current density corresponding to the maximum power
point: the point on the J-V curve where the maximum power is generated by the solar
cell. This is found by graphing the power density vs voltage and solving for dPdV
= 0.
The power generated at the maximum power point is represented by the blue square in
Figure 3-3.
With those parameters defined, we can now solve for the PCE:
η =JMVM
Pincid
=JSCVOCFF
Pincid
(3–6)
Note that this is the efficiency at the maximum power point; the efficiency will be lower
at other equivalent loads in the fourth quadrant. Consequently, maximizing the power
extracted from a solar cell involves matching the load to the resistance at the maximum
power point.
3.1.4 Series and Shunt Resistance
Both series and shunt resistance have a detrimental effect on solar cell performance,
with the best performance extracted by minimizing the former and maximizing the latter.
37
Figure 3-4. Circuit equivalent showing series and shunt resistance.
The expression for the total current can be rewritten to include both series and shunt
resistance:
Jtotal = Jphoto − JOexp
[q(V + IRS)
ηkT
]
− V + IRS
RSH
(3–7)
Series resistance is the resistance encountered by carriers as they are extracted from
the device. Series resistance should ideally be as low as possible; high series resistances
lead to a lower FF and ultimately lower PCE. High series resistances can be due to
employing poorly conductive contacts or having insufficient electrical contact such that
the photogenerated carriers cannot be efficiently extracted. Shunt resistance is ideally as
high as possible, as it represents all current paths that carry the charges through a circuit
in parallel with the load, i.e., the photogenerated carriers do no useful work and do not
contribute to the overall efficiency. One possible contributor to low shunt resistance in
both p-n junctions and Schottky junctions is losing photogenerated carriers out the edge of
a device. A circuit schematic depicting series and shunt resistances is shown in Figure 3-4,
while Figure 3-5 shows the degradation of the J-V curve with large series and small shunt
resistances.
38
Figure 3-5. The effects of series and shunt resistance on the J-V curve. Optimal values areRseries = 0 ohm cm2 and Rshunt = ∞ ohm cm2.
3.2 Theoretical Limitations
No physical, biological, or chemical process is 100% efficient due to thermodynamic
limitations, and solar cells are no exception. In 1961 Shockley and Queisser published
their exhaustive calculation on the maximum theoretical efficiency of single p-n junction
solar cells, often called the detailed balance limit. This maximum efficiency is attributed
to three primary mechanisms: emission of blackbody radiation, spectrum losses, and
recombination.[34, 35] All objects emit blackbody radiation as a function of their
temperature. Solar cells operating at room temperature emit radiation corresponding
to 300K, and this emission accounts for loses of 7%.
Spectral losses are the losses of photon energy exceeding the band gap of the absorber
material. Creation of a single electron-hole pair only requires energy equal to the band
gap of the semiconductor, anything in excess is carried away by the charges as kinetic
energy, subsequently lost to phonons (heat) as the electrons relax to the bottom of the
conduction band and holes to the top of the valence band. Heating of the device can be
especially degrading to performance as the dark current is exponentially dependent on
temperature, leading to a lowered VOC and lowered PCE with an increase in temperature.
Mitigation of spectral losses is achieved by introducing multi-junction cells composed of
individual layers, each specifically tailored to collection photons of a particular frequency.
39
Lastly, up or down converters, luminescent layers that absorb high(low) frequencies and
re-emit low(high) frequency photons, can be utilized to better match the photon frequency
to the band gap and minimize energy loss and avoid heating. Up converters, typically
placed at the backside of the cell, are theorized to increase the maximum theoretical
efficiency to 47.6% for non-concentrated solar light, a noticeable improvement from the
Shockley-Quiesser limit. Down converters, which must be placed on the top of the cell,
reap benefit due to avoidance of thermalization of the photogenerated carriers, leading
to less heating of the device. Theoretical efficiency for these reach 38.6%, only a modest
improvement from the Shockley-Quiesser limit. [36]
Recombination is inevitable in solar cells and much effort is aimed at minimizing
losses. Photogenerated electrons and holes must travel to the electrodes to be extracted as
usable electrical energy, and poor construction of PV cells or using incompatible materials
lead to high rates of recombination. Even with careful control over surface properties
and use of compatible materials, recombination will still occur due to inherent properties
of solar cells. Differing effective masses of electrons and holes lead to different diffusion
lengths. A fast moving electron can collide with a slow moving hole leftover from a
previous photon absorption, recombining through one of the mechanisms discussed above.
An increase in photon flux correlates with an increase in the density of photogenerated
carriers, increasing the probability of recombination. Materials which possess a higher
minority carrier diffusion length are less likely to suffer from increased recombination at
higher photon fluxes.
Combining these three loss mechanisms together leads to the maximum theoretical
efficiency for a single p-n junction solar cell as a function of bandgap (Figure 3-6). The
maximum possible efficiency is 33.7% for a band gap of 1.34 eV. With a band gap of 1.12
eV, silicon can attain a peak efficiency of only 29%. Utilizing more complex architectures
such as tandem solar cells with concentrated light, increases the theoretical efficiency
40
Figure 3-6. The Shockley-Queisser limit showing the maximum theoretical efficiency as afunction of band gap for a single p-n junction solar cell.
efficiency dramatically; an infinitely layered solar cell under concentrated light can
theoretically reach 86%.[37]
3.3 Types of Solar cells
3.3.1 P-N Junction
The most prevalent solar cells are composed of a p-type and n-type semiconductor
brought into contact to form a p-n junction. P-N junctions can be both heterojunctions or
homojunctions, though in the case of the former, care much be taken to match the lattice
constants to minimize strain and defects at the junction. When a p-type semiconductor
is brought into contact with an n-type semiconductor, mobile charges rearrange in order
to establish equilibrium of Fermi levels. The offset in work function between the two
materials drives electrons in the n-type semiconductor towards the p-type semiconductor,
where they combine with the holes and leave the region with a net negative charge.
The n-side is left with a net positive charge. This continues until the resulting built-in
potential, Vbi, prevents any further migration of charge due to Coloumbic repulsion. The
built in potential and altered carrier density near the junction are manifested as a bending
of the semiconductor conduction and valence bands, with the region over which the band
bending occurs called the depletion region (Figure 3-7).[30, 38]
41
Figure 3-7. Schematic for a p-n junction. Shown is the space charge (depletion) region inboth p and n side. Electron transfer driven by Fermi level offset establishes abuilt in potential (yellow arrow) opposing further transfer of charge.
Solving Poisson’s equation yields the width of this region in both the p and n-type
semiconductor;
wp =1
Na
√√√√
2ǫsVbi
q(
1
Na+ 1
Nd
) , (3–8)
wn =1
Nd
√√√√
2ǫsVbi
q(
1
Na+ 1
Nd
) , (3–9)
where the total width is the sum of the two:
wtotal =
√
2ǫsq
(1
Na+
1
Na
)
Vbi (3–10)
where Vbi is the built in potential, and Nd and Na are the donor and acceptor densities,
respectively. Note that as the carrier density increases, the depletion width decreases; in
the case where one side is heavily doped, the depletion layer exists almost entirely in the
other side. This built in potential acts as a barrier to majority carriers while facilitating
transport of minority carriers, making solar cell efficiency of p-n based devices more
sensitive to minority carrier diffusion lengths.
42
Figure 3-8. Band diagram of P-N junction with no bias, reverse bias, and forward bias. Vbiis the built in potential and Vbias is the bias voltage. In reverse bias, thebarrier increases and the depletion layer expands. Under forward bias, thebarrier decreases and the depletion layer shrinks, leading to a higher current.
Under illumination, electron hole pairs are created within the bulk; UV photons tend
to be absorbed near the surface while IR photons are absorbed deeper in the substrate.
Minority carriers generated close to the depletion layer are carried by the built in potential
across the junction, where they become majority carriers and must diffuse to an electrode
to be extracted. Carriers generated farther from the depletion layer either diffuse to the
junction and are swept by the built in potential to the electrodes, or they recombine.
In p-n solar cells, the doped semiconductor material does not possess a high enough
conductivity to function as an electrode, necessitating metal grid lines evenly spaced
across the top of the device. These gridlines prevent incoming photons from reaching the
bulk, lowing the short circuit current and PCE commensurate with the percentage of
43
surface area covered by the metal. Increases in carrier lifetime can mitigate these losses by
allowing gridlines to be spaced further apart.
3.3.2 Organic
Organic based photovoltaics have become increasingly investigated due to low
temperature processing and the potential lower cost of materials, an easy way to reduce
the cost of manufacturing that would allow solar energy to be more competitive on the
open market. Organic molecules are flexible and can be tailored to possess the desired
electronic characteristics (e.g. tailored band gaps), two attributes distinctly lacking in
silicon. However, excitons in organic materials generally have binding energies on the
order of 0.5 - 1 eV1 .[39] Separation cannot occur within the bulk, but must occur at the
junction of the donor and acceptor materials.The strongly bound electron hole pairs have
a high probability of recombining, leading to diffusion lengths of tens of nanometers within
most organic photovoltaic materials, leading to recombination losses. A balance must be
struck between having sufficient absorber thickness to capture the majority of the light
incident on the device and minimizing the path to the junction.[29, 40, 41]
Two main architectures for organic solar cells are thin bilayer devices and bulk
heterojunction devices. The former relies on using thin layers of high conductivity p and
n-type conjugated polymers (mobilities comparable to amorphous silicon) to form a p-n
junction, while the latter uses a single solution mixture of the same polymers to minimize
the distance between the bulk and the junction.
The bilayer device, shown in Figure 3-9A, has the two layered donor and acceptor
materials sandwiched between electrodes. In most organic devices, the anode is constructed
of indium tin oxide (ITO) and the cathode of thin aluminum. Poly(ethylene-dioxythiophene)
(PEDOT:PSS) is often deposited on top of the ITO to minimize band offset. Due to the
1 Measurements of some organics, MEH-PPV and PPV have yielded binding energies from zero to ∼1eV. Discrepancies are possibly attributed to inaccurate treatment of electron-phonon coupling or otherelectronic effects.
44
A Bilayer Organic Solar Cell B Bulk Heterojunction Solar Cell
Figure 3-9. Schematic of the bilayer and bulk heterojunction solar cells.
short diffusion length of excitons, only those generated within 20nm of the junction
will be dissociated and collected, leading to relatively poor efficiencies. The bulk
heterojunction cell, shown in Figure 3-9B address this issue by having the acceptor
and donor material mixed together prior to deposition, improving efficiencies by combing a
thicker absorber layer with a donor-acceptor junction distributed throughout the bulk.
More recently there has been interest in organic-inorganic hybrid devices. In addition
to low temperature fabrication, these take advantage of the long diffusion lengths within
silicon and the excellent conductivity and antireflection properties of the polymer to
produce a solar cell with a PCE of 13%.[42]
The major barrier to large scale implementation of organic solar cells is their lack of
long term stability. Inorganic devices continue to work decades after installation, whereas
organic polymers suffer from both light induced degradation and oxygen/water vapor
degradation. Simple encapsulation solves the latter problem, but avoiding the degradation
associated with light involves using filters to block out selective wavelengths, which also
lowers the fraction of the solar spectrum being absorbed and turned into available energy.
[29]
3.3.3 Photoelectrochemical Devices
In the electrolyte of a photoelectrochemical cell, the electrochemical (Nernst)
potential of the incorporated redox couple sets the equilibrium distribution of the couple
45
Figure 3-10. Schematic of simple semiconductor/liquid junction solar cell showing redoxreactions occurring both at the semiconductor surface and at a metalcounterelectrode.
between its reduced and oxidized states. When the electrolyte comes into contact with
the semiconductor, the two exchange charge, simultaneously shifting the electrochemical
potential of the redox couple and the Fermi level of the semiconductor until they are
in equilibrium (thus establishing the depletion layer in the semiconductor). Under
illumination photocarriers are created in the semiconductor and charge of one sign is
repelled from the junction while charge of the other sign is driven to the junction where it
reacts with one member of the redox couple. That charge is then delivered by ion diffusion
to a counter electrode that comprises the second terminal of the cell. Though initially
promising, silicon based liquid junction (so called) solar cells suffer from electrochemical
reactions at the silicon surface, creating a high surface defect density in addition to
creating new species which would contaminate the electrolyte.
Another popular type of photoelectrochemical cell is the dye sensitized solar cell,
or Gratzel cell. These use as the semiconductor titanium dioxide which has too large a
bandgap to itself be useful as an absorber but this is ”sensitized” by the incorporation
of smaller bandgap dye molecules that are deposited onto the surface of the TiO2.
46
Otherwise they operate as discussed above with a redox couple ferrying charge to a
counterelectrode.[43] Similar to organic cells, Gratzel cells depend in high surface areas for
charge separation and must employ a bulk heterojunction structure to achieve appreciable
power conversion efficiencies.
3.3.4 Multi-junction
Multi-junction cells address the need to capture the incident radiation without
wasting above bandgap energy to non-radiative relaxation by having multiple layers.Tandem
devices are two, three, or more distinct layers, with the top layers absorbing shorter
wavelengths and the bottom cells absorbing at longer wavelengths. Due to their multiple
layers, these devices are not held to the Shockley-Quiesser limit imposed on single junction
Schottky and p-n junction solar cells, already achieving over 40% efficiency. The biggest
drawback is cost due to the complicated, multi-step processing and use of expensive
semiconducting materials, relegating these devices to specialized high value applications
(satellites and military).
3.3.5 Schottky junction
Schottky junction solar cells are similar to p-n junction solar cells with one of
the semiconductors replaced with a metal. The equilibration of the metal with the
semiconductor creates the depletion layer within the latter must generally be made so
thick that it is opaque. To get light to the semiconductor the metal is patterned as a
grid of narrow lines allowing the light to get into the semiconductor between the opaque
lines. Schottky junctions also benefit from being majority carrier devices fabricated
with low temperature processes, but up until the mid-1970s suffered from a lower open
circuit voltage than p-n junction solar cells.[44] This low VOC is partly due to the higher
dark current inherent in devices that rely on thermionic emission of majority carriers
for transport across the junctions (as in Schottky junctions). Additionally, surface
states that pin the Fermi level are also responsible for increased recombination and a
corresponding increase in dark current. A VOC and corresponding low PCE kept Schottky
47
junction devices from competing with p-n solar cells until Godfrey and Green developed a
17.6% metal-insulator-semiconductor (MIS) PV cell with an open circuit voltage of over
0.65V.[45, 46] This excellent performance was realized though a thin (<2nm) insulating
layer between the metal and semiconducting layer, passivating the silicon surface and
reducing recombination while negligibly affecting current transport. The carbon nanotube
- silicon solar cells presented in this dissertation are Schottky junction devices, and a much
more complete description of the underlying physics is presented in the following chapter.
3.3.6 Inversion Layer Cells
Inversion layer cells induce an inversion layer within the device to enhance performance.
The first of these devices was developed by RL Call in the early 1970s and was composed
of a grid based Schottky junction cell possessing larger spacing between the metal grid
lines and a thick insulating layer sandwiched between a transparent electrode and the
semiconductor. A voltage between the transparent electrode and the semiconductor
formed an inversion layer at the surface. This so called induced junction cell benefited
from efficient charge collection due to the surface depletion layer, which repelled the
majority carriers while attracting the minority carriers avoiding their recombination as
the minority carriers diffused to the widely spaced electrodes.[47] Call abandoned such
electronically induced junctions because of the challenges of getting pinhole free insulators
over the large areas needed but discovered that the deposition of certain insulators
simultaneously trapped charge of a sign that created the inversion layer without needing
to do so actively. Such cells were further refined by Godfrey and Green in the late ’70s.
These devices suffered from impermanence of the trapped charge and over time the
inversion layer would disappear.[44] The devices presented in this dissertation constitute in
some sense a rediscovery of the phenomena exploited by Call, Godfrey and Green.
48
Table 3-1. Current maximum efficiencies for various photovoltaic devices [1]
Type Voc (V) Jsc(mAcm2
)FF (%) Efficiency (%)
Monocrystalline Silicon (PERL) 0.706 42.7 82.8 25.0Polycrystalline Silicon 0.664 38.0 80.9 20.4Commercial Silicon - - - 13Monocrystalline GaAs 1.107 28.3 86.7 28.3Triple Junction GaInP/GaInAs/Ge 2.691 14.7 86.0 34.1Dye Sensitised 0.714 21.93 70.3 11.0Organic Tandem 0.899 16.75 66.1 12.1
49
CHAPTER 4INTRODUCTION TO SCHOTTKY BARRIERS
4.1 Fundamentals
Schottky junction solar cells are composed of a metal in contact with a semiconductor
and display rectifying behavior. Analogous to a p-n junction in which one side is
degenerately doped, the depletion layer exists solely on one side of the junction.
Semiconductor doping, Schottky barrier height, and interface dynamics affect the
transport of carriers within the device, affecting the functionality and PCE. The
highly conductive metal contact also functions as the electrode, eliminating the need
for additional gridlines to extract photogenerated carriers.
4.1.1 Basic Schottky Model
To first order, the Schottky barrier height formed at the interface between a metal
and semiconductor can be approximated by the Schottky-Mott model. Assuming a metal
of work function φm and a semiconductor of work function φn and electron affinity χ, upon
placing the two into electrical contact the energy difference between the work function
of the metal and the Fermi level of the semiconductor drives electrons to move from one
material to the other. Charge rearrangement continues until electrostatic equilibrium
is established, forming a potential gradient within the semiconducting material that
opposes any further transfer of electrons. This induces band bending of the conduction
and valence bands in the semiconductor, with the depletion layer width determined by
the spatial region over which the band bending occurs. A higher doping density yields
steeper band bending and a shorter depletion width. Shown in Figure 4-1 below are the
band profiles for an n-type semiconductor and a metal for which φm > χ, and for a p-type
semiconductor and metal for which φm < χ. For an n-type semiconductor, electrons
approaching the metal-semiconductor junction from the semiconductor side see a potential
of qVbi. Electrons approaching the junction from the metal see a potential barrier equal to
the Schottky barrier height. [30, 38, 48]
50
A N-Type Semiconductor
B P-Type Semiconductor
Figure 4-1. Schottky barrier band diagrams. Adapted from Ayalew, T, ”SiCSemiconductor Devices Technology, Modeling, and Simulation”, 2004
To first order, the barrier height is given by the Schottky-Mott relation:
qφSBH = q(φm − χ) for n-type semiconductor (4–1)
qφSBH = Eg − q(φm − χ) for p-type semiconductor (4–2)
4.1.2 Current Transport
Schottky barriers function as diodes, allowing negligible current under reverse bias
and exhibiting an exponential current with forward bias. Under reverse bias, the built
in potential increases and very little charge flows until breakdown. Under forward bias,
the band bending decreases as the metal is raised to a higher potential relative to the
semiconductor, resulting in an exponential increase of current, principally as carriers spill
over the barrier. In both cases the idealized Schottky barrier remains unchanged. Carrier
51
transport in Schottky diodes occurs through several mechanisms: thermionic emission,
thermionic field emission, field emission (tunneling), and minority carrier injection.[49]
4.1.2.1 Thermionic Emission
Thermionic emission, the primary transport mechanism for moderately doped
semiconductors and the one most relevant to the work presented in the following chapters,
is the transport of energetic carriers over the potential barrier at the junction. Thoroughly
investigated by Hans Bethe in the 1940s, the total current is a simple sum of the current
flowing from the semiconductor into the metal plus the current flowing in the opposite
direction.[50] The former is given by
Js→m = A∗T 2exp
(
−qφSBH
kT
)
exp
(qV
kT
)
(4–3)
A∗ =4πqm∗k2
h3
where V is the bias voltage and A is the Richardson constant which is dependent on the
effective mass (m∗)of the carriers. A = 120 Acm2K2 and
(A∗
A
)
n−Si= 2.1 for n-type silicon.
Current flowing from the metal into the semiconductor is independent of the bias voltage
and is given by
Jm→s = −A∗T 2exp
(
−qφSBH
kT
)
(4–4)
Summing the two equations yields
Jtotal = Jo
[
exp
(qV
kT
)
− 1
]
(4–5)
with Jo = A∗T 2exp(− qφSBH
kT
)as the saturation current density (often called dark current).
4.1.2.2 Thermionic Field Emission and Field Emission
For heavily doped (> 1017) semiconductors at low temperature, tunneling begins to
contribute significantly to carrier transport. Field emission is the tunneling of carriers
through the potential barrier at the semiconductor/metal junction. These carriers
52
generally possess little kinetic energy and lie near the Fermi level. Thermionic field
emission considers the tunneling of moderately thermally excited carriers across the
junction. Carries that already possess some thermal energy (though less than that
required to surmount the barrier) ”see” a smaller potential width than carriers that are
not thermally excited. A parameter for evaluating the transport regime is given as:
E00 ≡q~
2
√
N
m∗ǫs(4–6)
With kT ≫ E00, thermal emission is the dominant transport mechanism. If kT ≪ E00,
field emission (tunneling) is the primary mode of transport across the barrier. Finally, if
kT ≈ E00, then thermal field emission dominates. [49]
4.1.2.3 Minority Carrier Injection
Schottky junction diodes are primarily thought of as majority carrier devices due
to the extremely small contribution from minority carrier diffusion. At large forward
bias, however, drift of minority carriers becomes comparable to the thermal emission of
majority carriers over the barrier. The current contribution due to minority carriers is the
sum of the drift and diffusion processes:
J = qµnE︸ ︷︷ ︸
Drift
− qDdn
dx︸ ︷︷ ︸
Diffusion
(4–7)
Where µ is the mobility, n is the number concentration, D is the diffusion coefficient, and
dndx
is concentration gradient, and E is the electric field. The total current is limited by
minority carrier recombination within the depletion layer.
4.1.3 Beyond Schottky-Mott
Decades of experimental results have shown that the Schottky-Mott theory is not
an accurate measure of Schottky barrier heights for most metal-semiconductor interfaces,
motivating alternate theories to explain the physics and chemistry of these junctions.
53
Shown in Table 4-1 at the end of the chapter are the theoretical and measured Schottky
barrier heights for various semiconductor-metal interfaces.[51, 52]
Clearly the experimental results deviate from the theoretical values predicted by the
Schottky-Mott theory. For simplicity, an equation to quantitatively predict barrier heights
on n-type silicon has been adopted based on experimental results: qφSBH = 0.27qφm−0.52.
A more complete picture of the ”real” junction is shown in Figure 4-2. Included
are interface states occurring in the bandgap of silicon, denoted Qss, the origin of
which is explored in more detail below. It’s clear from this graphic that the underlying
physics of metal-semiconductor junctions is far more complicated than captured by the
Schottky-Mott model.
The graphic represents the junction without any external applied bias. Upon
introduction of an electric field, the combination of the applied field and the induced
image charges result in a small Schottky barrier lowering. Consequently, real junction are
somewhat bias dependent, with the amount the barrier is lowered given by
δφ =
√qǫm
4πǫs(4–8)
ǫm =
√
2qN |ψs|ǫs
and ψs is the surface potential. In addition to Schottky barrier lowering, devices can
suffer from leakage of carriers out the sides of the active area. This edge leakage often
occurs at the sharp corners of metal electrodes where highly concentrated electric field
lines facilitate tunneling. Mitigation of this effect is achieved in both p-n junctions and
Schottky junctions by using guard rings.
54
Figure 4-2. Interface states shown in the realistic model of a Schottky junction.
4.1.3.1 Fermi Level Pinning: Bardeen Model and Metal Induced Gap States
In 1947 John Bardeen published his paper on interface states at the semiconductor-metal
junction and their role in creating a barrier independent of the Schottky-Mott model. A
property of the semiconductor surface, interface states can affect the Schottky barrier
height dramatically, producing a barrier height completely independent of the metal work
function. The semiconductor surface can possess a very high density of surface states
- far higher than in the bulk. This leads to Fermi level pinning, a process in which the
55
extremely high density of surface states prevent the semiconductor Fermi level from
shifting in response to the metal contact, even after reaching equilibrium. Not only do
these surface states pin the Fermi level and create a Schottky barrier independent of
φm, but they also act as trap sites, facilitating recombination of electrons and holes.[53]
Nearly 30 years after Bardeen published his work on Fermi level pinning at semiconductor
surfaces, Tersoff published his theory exploring alternate sources of Fermi level pinning
in order to explain the discrepancy in the Bardeen model for predicting the SBH within
ionic semiconductors. At a free semiconductor surface (not an MS interface), the Fermi
level can be pinned by a relatively small amount of surface defects. The screening
length of these defect states within the semiconductor is relatively large, leading to a
correspondingly large surface dipole and shift in EF . At an MS interface, the charges in
the metal screen the surface defects, leading to a smaller local dipole and smaller shift in
EF . His theory, based on metal induced gap states (MIGS), posits that the continuum of
states existing in the metal at the MS interface ”leak” into the semiconductor, leading to
gap states that decay into the bulk and consequently pin the Fermi level. The barrier
height is the sum of both the dipole due to metallic screening of the MIGS and a
surface dipole, either of which may dominate depending on the bulk properties of the
semiconductor.[53]
4.1.3.2 Bond Polarization
A more contemporary theory pioneered by Raymond Tung looks to chemical bonds
formed across the metal-semiconductor junction as the source of interface states that cause
the barrier height to deviate from the Schottky-Mott approximation. All properties of
the Schottky junction, including barrier height and interface dipole, are a consequence of
the chemical bonds formed between the semiconductor and metal. The interface dipole
is simply due to the polarization of these bonds. This model assumes an extremely small
interface region and predicts the barrier height using the following equation:
56
φSBH,n = γB (φm − χs) + (1− γB)Eg
2(4–9)
γB = 1− e2NBdMS
ǫit|β|and β is the bond integral, NB is the density of chemical bonds, dMS is the distance
between the atoms at the metal surface from the atoms at the semiconductor surface, ǫit is
the permittivity of the interface, and κ accounts for charges ”hopping” from one atom to
another (metal-metal, metal-semiconductor, semiconductor-semiconductor).[54–56]
Though bond polarization is an excellent model for chemically active metals, carbon
nanotubes are not chemically reactive. Due to this lack of chemical reactivity of the
carbon nanotubes and the reluctance to form chemical bonds with substrates, the Bardeen
model provides more insight into the behaviors relevant to our solar cells.
4.2 Schottky Junction Solar Cells
4.2.1 Historical Background
Built by Charles Fritts in 1894, the first solar cell was a Schottky junction device
constructed from selenium sandwiched between gold and another metal. Solar cell research
and development activity remained low until the 1950s, when high quality silicon became
available, spurring advancements in homojunction structures. Initially engineered for
space applications, it wasn’t until the energy crisis in the 1970s that solar cell development
began to focus on production for commercial use. Both p-n junctions and Schottky
junction solar cells were extensively researched, with metal-insulator-semiconductor
inversion layer (MIS-IL) devices achieving 17.6%.[45, 46] In the decades since Godfrey
and Green’s high performing MIS structure, solar cell designs have expanded dramatically
by exploiting advances in polymer science for organic devices, using new technologies to
fabricate high quality thin film structures, developing novel light trapping techniques, and
creating hybrid structures exploiting the strengths of varied materials. Concomitant with
those developments was a greater understanding of the physical and chemical processes
57
pertinent to photovoltaics. A discussion of all PV types is outside the scope of this thesis,
but much of the work pioneered in the previous decades greatly influenced the design of
the devices presented below.
4.2.2 CNT on Silicon Schottky Junction Solar Cells
Carbon nanotubes were first incorporated into devices as conductive electrodes for
either supercapacitors or organic photovoltaics.[57, 58] In 2002, a carbon nanotube/polymer
solar cell with an efficiency of 0.04% was developed, catalyzing the use of carbon
nanotubes in photovoltaics.[59] Low performance in CNT/organic PV cells encouraged
development using inorganic materials, and in 2007 a double walled carbon nanotube -
silicon device with a PCE of 1.38% was published.[60]
SWNT-Si solar cells have achieved high efficiencies in recent years, due in part to
the malleability of the electronic and optical properties of carbon nanotubes. In 2010, it
was demonstrated that an ionic liquid could be used to electronically modulate the Fermi
level of the carbon nanotube film, changing the built in potential, junction dynamics, and
PCE.[9] The devices presented below have taken the original solar cells, shown below in
Figure 4-3 and further optimized the device performance by exploiting the effect of the
ionic liquid.
4.2.2.1 Experimental Details and Equipment
Substrate were diced from a 500 µm thick, <100>/<111>, n-Type silicon wafer
possessing either a 200nm or 1000 nm thick thermal oxide. The <100> wafers had a
resistivity of 0.5-0.7 Ohm-cm, while the <111> wafers were slightly less doped with a
resistivity of 4-10 Ohm-cm. Onto the surface of the oxide was defined a square 12x12 mm2
Au/Cr (60/10 nm) pad, possessing a 2x4 mm2 rectangular window at its center. This
Au/Cr pad served several functions: it provided an etch mask for a BOE etch of the oxide
in the window and served as the electrical contact to the nanotubes that were draped as
a thin film from the Au/Cr layer down across the silicon; and finally, it provided a literal
shadow mask, limiting the collimated, simulated solar radiation to the window area in
58
Figure 4-3. Carbon nanotube-silicon Schottky junction cell. Included are both the activearea film and the gate film
which the active area was defined. After a BOE etch removed the thermal oxide within
the gold framed window, a carbon nanotube film was transferred following the procedure
in Wu, et al..[8] Another film transferred to a secondary gold electrode functioned as the
gate film and was insulated from the silicon by the thermal oxide. Ohmic contact between
the Si wafer backside and a stainless steel sheet was made by a gallium-indium (Ga/In)
eutectic spread between the two.
Illumination was provided by a 150W xenon lamp (Oriel 6255) in an Oriel 6136
housing powered by a model 8500 power supply. An Oriel 81094 AM1.5G filter approximated
the solar spectral distribution. Light from the inhomogeneous source was focused into the
aperture of a 150 mm long, fused silica Homogenizing Rod (Edmund Optics P65-837) by
a 50 mm diameter fused silica lens with a 65 mm focal length. The output face of the
Homogenizing Rod was imaged in the horizontal focal plane of the sample by a 50 mm
diameter, 100 mm focal length fused silica lens after rotation by 90 degrees with a broad
band mirror (Newport 66225). The intensity at the sample plane was adjusted to 100 mWcm2
by translation of the 65 mm FL lens, cutting down on the fraction of the light entering the
59
Homogenizing Rod. Homogeneity of the light intensity over the 1 cm2 central region of
the homogenized beam at the sample plane was measured to be within 5%.
4.2.2.2 Electronic Gating
These devices differed from conventional Schottky junction solar cells by inclusion of
an ionic liquid 1-Ethyl-3-methylimidazolium bis(trifluoromethylsulfonyl)imide (EMI-BTI)
to alter the electronic properties of the nanotube film, which in turn modifies the
characteristics of the Schottky barrier and built in potential. The ionic liquid saturates
both the active area SWNT film and a secondary SWNT film. A gate voltage is applied
between the two films, polarizing the ions in the ionic liquid and increasing or decreasing
the carrier density of the SWNT film in the active window. This capacitive doping alters
the Fermi level of the film and subsequently changes the built in potential at the SWNT-Si
junction. Unlike photoelectrochemical cells where the ionic liquid actually transfers
charge between electrodes, there is no net current flow through the ionic liquid. EMI-BTI
has a large electrochemical window spanning nearly 5 V, allowing the application of
appreciable voltages between the two SWNT films without electrochemistry occurring at
the electrodes. This process, akin to charging an electrolytic capacitor, draws no current
once equilibrium is established at each gate voltage. To characterize the cell, a fixed gate
voltage is applied to the SWNT gate film and the voltage across the electrodes is swept
from -1V to 1V and the resulting J-V curve is analyzed to extract device performance.
Then another gate voltage is applied and a new J-V curve is generated. Previous results
demonstrated an optimized performance at a gate voltage of -0.75V; positive gate voltages
result in a smaller PCE, fill factor, and VOC . [9] The resulting J-V curves for gate voltages
between +/-0.75 V are shown in Figure 4-4B. The native device PCE without gating was
8.5%. Gating modulated the PCE between 3.6 and 10.9
4.2.2.3 Inversion Layer Modeling
The previous work by Dr. Wadhwa elucidated the mechanism by which the ionic
liquid improved the efficiency of the solar cells. The intended function of the electrolyte
60
A Schematic of the during electronic gating.
-0.4 -0.2 0.0 0.2 0.4 0.6 0.8-40
-30
-20
-10
0
10
20
30
40
50
60
Cur
rent
Den
sity
(mA
/cm
2 )
Voltage (V)
-0.75V -0.6V -0.45V -0.3V -0.15V 0V 0.15V 0.3V 0.45V 0.6V 0.75V
B Resultant J-V curves at various gate voltages.
Figure 4-4. Schematic and results for electronically gated SWNT-Si cell. Gate voltage isthe potential difference between the active area film and the gate film
was to change the Fermi level of the CNT film, increasing or decreasing the charge
transferred between the silicon and CNT in order to establish equilibrium of Fermi
levels in addition to increasing the conductivity of the film. However, an experimental
architecture in which the carbon nanotube film was pattered into thin strips via
photolithography demonstrated another mechanism at play. In this device, the carbon
nanotube strips measured 100 µm across and were evenly spaced every 300 µm, as shown
in Figure 4-5A below. The active area of the device has less than 30% film coverage.
As expected, the photocurrent prior to addition of ionic liquid was reduced due to the
decreased film coverage. Though reduced by a factor of 2, the photocurrent did not
decrease directly in proportion to the reduced SWNT film coverage; the long diffusion
lengths in single crystal silicon allows collection of photocarriers generated outside the
junction. Upon addition of ionic liquid the photocurrent increased dramatically, indicating
an alteration of the silicon surface in a way to facilitate charge collection. Applying a gate
voltage further improved device performance, consistent with previous planar devices. [10]
61
A The active area of the film com-prising of 2 x 4 mm silicon windowwith patterned carbon nanotube film.
B Figure B: J-V curve for the same device. Note reduced pho-tocurrent prior to addition of EMI-BTI.
Figure 4-5. Schematic and performance for grid cell.
We inferred, and simulations by J. Guo and J. Seol in the Electrical and Computer
Engineering department at the University of Florida confirmed, that the ionic liquid
spontaneously forms a layer of charge at the silicon surface where no nanotubes are
present, inducing an inversion layer in the silicon outside the SWNT-silicon junction. The
positive carriers that diffused into the depletion region traveled along the surface of the
silicon until they encountered a nanotube and were extracted. This is manifested as an
increased photocurrent and corresponding increase in efficiency. This behavior can be
qualitatively understood as follows. When the nanotubes and the n-Si are first placed
in intimate contact, the free energy of electrons in the n-Si (work function:φSi = 4.3 eV)
is reduced by their transfer to the carbon nanotubes (work function: φSWNT = 4.9 eV).
Such transfer stops when Coulombic restoring forces due to the charge imbalance raise
the local potential (the built in potential) to prevent further charge exchange, establishing
equilibrium. In the presence of electrolyte ions, the ions migrate to compensate the
transferred charge and thus permit the exchange of substantially more charge before
62
the equilibrium is reached. Additional electrons are transferred to the nanotubes from
the Si regions between the nanotube grid lines compensated by positive electrolyte ions
surrounding the nanotubes, while the positive charge left behind in the n-Si inversion layer
is compensated by negative electrolyte ions accumulated at the Si surface. The electrolyte
here serves much as it does in an electrolytic capacitor to raise the capacitance of the
system with a self-potential provided internally by the original Fermi level offset between
the nanotubes and the n-Si, or externally by the gate field. Shown in Figure 4-6 below
are the simulated results at a bias voltage of 0V for VG = -0.75V and VG = +0.75V. Full
simulation results are included in Appendix A. [61]
Exploitation of this discovery opened the doors for alternative architectures designed
to optimize device performance. Guo and Seol showed that the ionic-induced extension
of the depletion region occurred over hundreds of microns, allowing a reduction in
the fraction of nanotube film covering the silicon active area without sacrificing hole
extraction. The positive carriers that diffused into the depletion region traveled along the
surface of the silicon until they encountered a nanotube and were collected. The reduction
in the nanotube film area enhanced the number of photons being absorbed in the silicon
surface to increase both Jsc and the PCE.
63
A Model Parameters
B Simulations at VG = -0.75V and VG = +0.75V
Figure 4-6. Simulations showing inversion layer in silicon extending across entire surface inbetween carbon nanotube strips. Note the reduction in the inversion layer at+0.75V; at more positive bias voltages, this results in a reduced collection ofcarriers, as shown in previous J-V curves for gated devices (Figure 4-4B)
64
Table 4-1. Theoretical vs Experimental Schottky Barrier Heights: Barrierheights measured at 300K, theoretical values determined fromSchottky-Mott relation
Semiconductor Metal Theoretical SBH (eV)1 Experimental SBH (eV)2
n-Si Al 0.01-0.2 0.81n-Si Au 1.05-1.4 0.83n-Si Pt 1.1-1.9 0.9n-GaAs Al 0.03-0.2 0.93n-GaAs Au 1-1.38 1.05n-GaAs Pt 1-1.8 0.981 The theoretical value is given as a range as the work function of metals differs
depending on crystallographic orientation.2 Highest measured values.[62, 63]
65
CHAPTER 5NANOSTRUCTURING FOR ENHANCED LIGHT ABSORPTION
5.1 Overview
Texturing of photovoltaics to improve photon absorption has been aggressively
investigated for several decades as a means to improve solar cell efficiency.[28, 64–72] In
2000, Li et al. demonstrated the feasibility of using a simple metal-assisted chemical etch
to create porous silicon, opening the doors to facile production of optically absorbent
substrates with a high surface area.[73] While fabrication of porous structures using
chemical etching is a simple and low cost process, there is little control over the size and
uniformity of the resulting features (though spatial periodicity can be detrimental due to
diffraction losses). CVD growth is often utilized for applications where complete control
over diameter and depth is needed. In 2005, Peng et al. demonstrated a 9.31% efficient
silicon nanowire, p-n homojunction solar cell.[74] Continued innovation in nanostructured
devices has culminated in an 18.2% efficient porous silicon p-n homojunction cell.[75]
Considering the reflectivity of the devices presented in the previous chapter (silicon
+ SWNT + ionic liquid) is roughly 20% over wavelengths greater than the band gap of
silicon, optimizing light absorption can result in significant increases of the PCE. Methods
for texturing the silicon to enhance absorption are presented in this chapter, along with
complications inherent to alteration of the semiconductor substrate.
5.2 Potassium Hydroxide Etching
An aqueous solution of potassium hydroxide (KOH) is a basic solution frequently
used to etch pyramidal structures in silicon. At room temperature, the ratio of etch
rates between <100> and <111> planes is roughly 100:1 for a 30% KOH solution. This
anisotropy leads to regular pyramidal structures with the exposed faces corresponding
to the <111> plane.[76] Though the pyramids are excellent at reducing reflection from
the silicon surface, the pointed tops would provide little surface for contact to the SWNT
film and prompted fabrication of pyramidal grooves instead. Fabrication consisted of first
66
defining the areas to be etched through photolithography on a <100> substrate with
a 200 nm thermal oxide. SiO2 is largely unaffected by KOH, permitting its use as an
etch mask. A BOE etch removes the thermal oxide in the regions not protected by the
photoresist, after which the photoresist is removed and the substrate is then placed in the
KOH for the desired length of time. Following a DI rinse and N2 dry, the remaining oxide
is stripped with BOE and a SWNT film is transferred to the silicon following standard
procedures and subsequently tested. The resulting grooves were 10 µm wide, 14 µm deep,
and extended across the entire active area.
A Resultant grooves from KOHetch
-0.4 -0.2 0.0 0.2 0.4 0.6 0.8-40
-30
-20
-10
0
10
20
30
40
50
60
Cur
rent
Den
sity
(mA
/cm
2 )
Voltage (V)
Trenches No IL Trenches -0.75V No Trenches No IL No Trenches -0.75V
B J-V curves for KOH device
Figure 5-1. KOH schematic and performance.
Similar to the grid film, there is a reduced photocurrent without the ionic liquid and
an increased photocurrent with addition of the EMI-BTI, as shown in Figure 5-1. As
with planar devices, a marked improvement in efficiency occurs for negative gate voltages.
The maximum PCE obtained was 11.4%, an improvement of 12% over a planar, full film
<100> device (10.6% PCE). The fill factor remained roughly the same: 0.77 for the KOH
etched device versus 0.76 for the full film planar device. The increased performance can
67
therefore be attributed to the increased JSC resulting from the enhanced absorption of the
grooved structure. Note, however, that the overall PCE is still lower than that achieved
by the grid film. The dimensions of the grooves are far larger than the wavelength of
visible light, leading to a relatively poor increase in absorption compared to other textured
devices. Additionally, the periodic spacing of the grooves leads to diffraction at oblique
angles, making these structures inferior for light collection. Nonetheless, this result
reinforces the improved performance attained with a reduction in SWNT-Si contact and
encouraged continued experiments in surface texturing.
5.3 Silicon Nanowires
Due to their excellent absorption and easily tailored geometries, silicon nanowires
arrayed in vertical forests on a substrate have been extensively studied for integration
into photovoltaic devices. As in commercially available solar cells, p-n homojunctions
of various designs are found within silicon nanowire based devices. Fabrication involves
taking a doped semiconducting nanowire forest supported on a bulk substrate and using
a high temperature diffusion process to dope the surface of the nanowires. This doping
process can be optimized to dope just the nanowire, leaving the supporting substrate
with the original doping type, or the doping can be done coaxially, creating a radial
p-n junction.[72, 77] The first generation p-n nanowire solar cells suffered from high
series resistance, reduced light absorption due to metal finger electrodes, and diffraction
at oblique angles due to the regular spacing of nanowires.[72, 78, 79] Aside from p-n
homojunctions, silicon nanowires have been integrated into photoelectrochemical cells
and even devices utilizing a carbon nanotube top layer.[80, 81] The latter, however, were
poorly conceived and did not likely operate as photoelectrochemical cells as claimed,
but rather likely along the lines of the mechanisms discussed here with a PCE of only
1.3%. Despite impressive progress in the efficiency of silicon nanowire cells in general, the
maximum efficiency for any silicon nanowire based devices has not yet passed 10%.[78, 82,
83] The general thought in the community for the reason behind this limit is the challenge
68
of controlling the trap states (and resulting recombination) at the large surface area
created in the nanowire surfaces. Given the grid SWNT film results we reasoned that we
could avoid recombination via the electrolyte inversion layer and thus resolved to pursue
such Si nanowire devices.
5.3.1 Procedure and Characterization
Fabrication of the silicon nanowires was achieved via a simple chemical etch adapted
from KQ Peng, et al..[80, 84] Substrates were prepared by evaporating a gold electrode
frame surrounding an open window to become the active cell area, and then painting
photoresist everywhere save for the active area window. The photoresist protects the
substrate from the aggressive etch while the gold electrode also functions as an etch mask
to keep the nanowire ”growth” contained within the active area. The substrates were
placed in 6:1 BOE to remove the thermal oxide layer, rinsed with DI water, dried in an N2
stream, and submerged into a 4.6M/0.02M HF/AgNO3 solution for a varying amount of
time. At room temperature, a 4 minute yielded 1 µm long nanowires.
Figure 5-2. The mechanism for silicon nanowire growth. A: Silver ions adsorb ontosurface. B: Oxidation of silicon. C: Etching of SiO2 and sinking of silverparticle. Adapted from KQ Peng et al.
The etching solution produces silicon nanowires via a three step process, as shown in
Figure 5-2 silver ions (dissociated from the AgNO3) adsorb onto the silicon surface and
oxidize the silicon directly underneath the silver particle. This locally oxidized region of
silicon is etched away by the HF, creating a spatial vacancy that the silver particle sinks
into. The reduced silver undergoes a redox reaction and returns to its oxidized state,
upon which the process repeats until the reaction is quenched. The resulting structure is
69
Figure 5-3. Silicon nanowires grown in an HF/AgNO3 solution
dependent on both the molar concentration of the two active chemicals (HF and AgNO3)
and the length of etch, with the concentrations listed above producing the high aspect
ratio nanowires. Alternative wet chemical etches also explored involve electroless metal
deposition or sputtering/evaporation of molecular silver/gold, respectively, onto the
substrate and using an oxidant solution (typically H22O2 and HF) to catalyze redox
reactions at the metal conglomerates. The hole produced by the oxidation of the metal is
injected into the valence band of the silicon, oxidizing the region directly below the metal.
This alternative has the advantage of controlling metal deposition using photolithography
to control the etching geometry. Although these structures were initially considered
for integration into our SiNW devices, SEM images showed less uniform growth of
nanowires with this method. Additionally, a comprehensive review of silicon nanowires
in photovoltaic cells concluded that nanowire forests fabricated through in HF/AgNO3
solution had a higher VOC than those fabricated through other wet chemical etches.[80].
These two factors led us to use silicon nanowires fabricated in a HF/AgNO3 solution.
Immediately upon removal from the etchant solution, residual silver residing at
the bottom of the nanowires or extending across the nanowire surface are removed in
an 8M nitric acid bath, followed by two DI baths and a gentle N2 dry. The nanowires
70
A Silicon Nanowires on <100> crystallographicorientation.
B Silicon Nanowires on <111> crystallographicorientation.
Figure 5-4. Orientation of silicon nanowires.
agglomerate upon removal from the last DI bath, as shown in Figure 5-3. While growth
was initially performed on both <100> and <111> silicon substrates, SEM images showed
a distinct difference between the wires formed on different crystallographic orientation
substrates; <100> gave an isotropic distribution of vertical nanowires, while <111>
formed well defined regions of nanowires at other angles corresponding to the easily
etched <100> and <110> crystallographic planes, as shown in Figure 5-4. It has been
demonstrated that the selective etch orientation is related to oxidant concentration, with
low concentrations leading to etching along <100> planes and high concentrations leading
to vertical etching independent of crystalline orientation.[85]
Reflectance from the nanowires near normal incidence to the substrate on both
<100> and <111> were measured using a UV/VIS/NIR spectrophotometer from
400-1200nm. Though it was far lower than that for untextured silicon, the <111> still
exhibited higher reflectance than <100>, most likely due to the angled nanowires.
5.3.2 Integration in solar cells and initial performance
The silicon nanowires exhibit a very low reflectance over photon energies greater
than the band gap of silicon, but single-wall carbon nanotube on silicon nanowire
(SWNT-SiNW) devices suffer from a reduced contact between the silicon and SWNT
71
400 500 600 700 800 900 1000 11000.00
0.05
0.10
0.15
0.20
0.25
0.30
0.35
0.40
0.45
0.50
0.55
0.60
Reflectivity
Wavelength (nm)
Bare silicon w\IL 20nm SWNT on Si w/IL 20nm SWNT on <100> SiNW w/IL 20nm SWNT on <111> SiNW w/IL
Figure 5-5. Reflectance of the silicon nanowire substrates as compared to untexturedsilicon.
film. Initial devices performed relatively poor. In fact, there is practically no photocurrent
in the absence of the ionic liquid and the PCE is only 0.02%. Addition of the ionic liquid
yields a maximum gated PCE of 4.6%, relatively low compared to other devices, but
the increase emphasized how enhanced absorption coupled with the inversion layer were
able to partially compensate for the minimal SWNT-SiNW junctions at the tips of the
nanowires.
Though this efficiency is poor in comparison to the planar gated devices, it was good
enough to encourage further investigation and optimization. An increase in the gate
electrode film area commensurate with the increase in silicon surface area was imperative
for inducing the inversion layer in the nanowire sidewalls. Employing a gate film of a size
used in the planar devices evidently limited the capacitance and partially explained the
poor performance. Additionally, employing better transfer methods to enhance contact
between the carbon nanotube film and the silicon nanowires would further improve carrier
collection. Finally, the increased surface area provided more recombination centers,
72
Figure 5-6. Initial performance of the SWNT-SiNW device. Note that the blue curve isthe illuminated J-V prior to addition of ionic liquid. The inset shows minutepower generation prior to the addition of the ionic liquid.
making the performance highly sensitive to the surface properties of the silicon and
encouraged experiments aimed at passivating the silicon surface.
5.3.2.1 Remote Gating
The need for a larger capacitance gate electrode and desire to avoid taking up
(potentially useful) area on the face of the solar cell necessitated the development of
a remote gate electrode that still retained a high surface area while occupying a small
volume. This gate electrode consisted of a coiled Pt wire onto which a thick layer of
SWNTs had been deposited, placed within a 2 mm inner diameter polyethylene tube. The
small tube was filled with the viscous EMI-BTI electrolyte, rendered immobile though
capillary forces. The SWNTs on the Pt wire amounted to a 1cm x 1cm, 1 µm thick
film, over 2 orders of magnitude larger than the previous gate films. The end of this
gate electrode was touched to the EMI-BTI electrolyte drop (over the Au pad to avoid
shadowing light from the active area) connecting the electrolyte reservoirs. Note that this
remote gate electrode improves on the previous design where the gate electrode occupied
front surface real estate of the Si (thus, in principle, precluding that areas availability for
73
light capture). A schematic cross-section of the device architecture (not to scale) and the
wiring diagram for testing is shown in Figure 5-7A.
A Experimental set-up for electronic gating withthe remote gate
B SEM image of SWNT-SiNW activearea, showing the nanotube filmlaying across the SiNW.
Figure 5-7. Schematic for remote gating and SEM of SWNT-SiNW active area
5.3.2.2 Passivation of Nanowire Sidewalls
The benefits of a thin native oxide passivation layer in planar SWNT-Si solar cells
was noted in the supporting information of Wadhwa et al. and studied in some detail for
double walled carbon nanotube/planar-Si cells by Jia et al.[9, 86] Such passivation is also
critical for the silicon nanowire devices. If the oxide layer becomes too thick, however,
it presents a tunneling barrier that degrades the cell performance. An initially poor
performance of SWNT-SiNW cells tested immediately after deposition of the nanotube
layer suggested that the native oxide had grown too thick during the device fabrication
steps. Accordingly, a brief BOE etch of the SiNWs through the porous SWNT network
was implemented to strip away the oxide, followed by an oxide regrowth in the ambient
air. Figure 5-8 shows J-V curves for a SWNT-SiNW cell (no EMI-BTI) as a function of
time in ambient air following the BOE etch. The initial measurements exhibited very poor
74
performance; however, the short circuit current density, JSC , open circuit voltage, VOC,
and fill factor, FF, were all seen to improve with time up to 96 hr, after which the trends
reversed. The series resistance, RS, obtained from the slope at the highest forward bias,
was found to grow monotonically while still being low enough at 96 hr that the native
device performance was maximized at that time. Plots of the solar cell parameters with
increasing oxidation time are shown in Appendix B. Electrolyte gating was also found to
be optimized following such 96 hr oxidation. It has been shown that water plays an active
role in silicon oxidation, so it was reasoned that its exclusion by the hydrophobic ionic
liquid used during gating would avoid further silicon oxide formation once the electrolyte
was added.
-0.4 -0.2 0.0 0.2 0.4 0.6 0.8
-40
-30
-20
-10
0
10
20
30
40
50
60
Cur
rent
Den
sity
(mA/
cm2 )
Voltage (V)
post BOE etch 5 hrs 24 hrs 30 hrs
Figure 5-8. J-V of a SWNT-SiNW device showing the effect of sidewall passivation viaoxidation on the performance of the device.
The effect of gating with the EMI-BTI electrolyte at gate voltages of +1.0, 0, and
-1.0 V under 100 mWcm2 , AM1.5G illumination are shown in Figure 5-9. The gate voltage
75
induced modulation of the SWNT Fermi level relative to that of the n-Si, modulating
the built-in potential (Vbi) at the junction is indicated by the dramatic shift of the open
circuit voltage (VOC) from 0.15 V (at VG = +1.0 V) to 0.58 V (at VG = -1.0 V). At the
gate voltage of VG = +1.0 V, positive ionic charge in the electrolyte at the silicon surface
attracts majority carriers (electrons, in the n-Si) to the surface and into the nanowires.
Screened by these excess majority carriers from the positive ionic charge, photogenerated
holes can also approach the silicon surface, resulting in an enhanced surface recombination.
Combined with the simultaneous decrease in the built-in potential in the SiNWs at
their junctions with the nanotubes, the recombination losses lead to a fill factor that is
essentially zero. At the gate voltage of VG = -1.0 V, negative ions in the electrolyte at
the silicon surface repels the majority carriers, creating an inversion layer at the surface
which limits surface recombination in a major fraction of the SiNWs. Combined with the
enhanced Vbi in the SiNW at the SiNW/SWNT junctions, the fill factor increases to 0.76
and maximizes the cell performance. The 35 mAcm2 short circuit current density here is much
greater than that in the planar, gated SWNT/Si cells (mAcm2 ), consistent with the additional
light absorption due to the vertical SiNW array.
5.3.2.3 SWNT film transfer on SiNW
Two distinct methods for depositing the SWNT layer were explored: ultrasonic
spraying from an ethanol suspension and transfer of a pre-formed SWNT film made
by the filtration route.[8] Purely sprayed SWNT layers had to be made substantially
thicker than what is seen in Figure 5-7B to attain low resistance continuity to the Au/Cr
electrode. In our experience, however, photons absorbed in the nanotubes contribute
little, if at all, to the power generation, so that thicker nanotube layers degraded cell
performance.[10] A good compromise was to spray a thin layer of nanotubes followed by
the transfer of a 10 nm thick filtration fabricated film. The roughly optimized quantity
of nanotubes deposited by the combined method had a surface nanotube concentration of
1.3 µgcm2 , approximately equivalent to that in a 20 nm thick, entirely filtration formed and
76
-0.4 -0.2 0.0 0.2 0.4 0.6 0.8
-40
-30
-20
-10
0
10
20
30
40
50
60
Cur
rent
Den
sity
(mA/
cm2 )
Voltage (V)
VG = -1V
VG = 0V
VG = +1V
Figure 5-9. J-V curves for VG = -1.0 V, 0 V, +1.0 V on the SiNW device.
transferred film; however, the transfer of a 20 nm thick film without the sprayed layer did
not yield devices that performed as well as the partly sprayed, partly transferred layers.
Table 5-1 at the end of the chapter compares the performance of several SWNT/SiNW
cells at VG = -1.0 V for which the principle differences were the deposition method and
thickness of the SWNT layer. Device D (J-V curve shown in Figure 5-9) was the best, for
which the power conversion efficiency was 15.1%. Devices with a thicker (and hence more
light absorbing) net SWNT layer, (with correspondingly reduced light transmission into
the Si) exhibited poorer performance.
To explain the differences between the wholly transferred versus the partially
spray deposited films we note that thicker filtration fabricated SWNT films possesses
a greater mechanical stiffness. When such a film is transferred across vertical nanowires
with varying heights, the stiffness limits the films ability to conform over short length
scales, preventing contact to the shorter nanowires. This motivated the use of mixed
77
sprayed/transferred films and is consistent with the data in Table 5-1. Indeed it is this
ability of the nanotubes to touch and extract photocurrent from virtually every nanowire
tip, while providing a direct (non-tortuous), low impedance pathway to the gold electrode
(along with the gate induced inversion that avoids surface recombination) that explains
the dramatically improved performance in these cells over other silicon nanowire based
cells reported to date.
5.3.3 Discussion of inversion layer in SiNWs
The improvement garnered from the inversion layer in planar cells is impressive,
though predictable; the long diffusion length in single crystal silicon ensures adequate
collection of carriers with an inversion layer stretching hundreds of microns long. The
passivating native oxide layer coupled with the inversion layer prevent recombination at
the surface. The silicon nanowires, however, are known to have extremely high rates of
recombination. Indeed, this has historically been one of the limiting factors in efficiency
as the high surface area and long path to the electrode result in significant loss of carriers.
While oxidizing the substrate for numerous hours partially passivates the nanowire
sidewalls, performance is still extremely poor due to recombination. The induction of ionic
liquid immediately improves performance and allows carriers to then diffuse thought the
single crystal silicon nanowires until they reach a carbon nanotube. The geometries of the
silicon nanowires are such that it’s possible the entire nanowire is inverted. Simulations
presented in the previous chapter showed the inversion layer reaching a depth of over
1 µm, while the silicon nanowires are only tens of nanometers wide. Were the entire
nanowire to be inverted and there existed no potential gradient, there is no driving force
to repel the electron from the nanowire, eventually leading to recombination. It is thus
possible that further improvements could be obtained with thicker nanowires. Efforts to
experimentally test this are presently part of a collaboration with the CNMS at Oak Ridge
National Labs.
78
5.4 Effect of Oxygen and Water on Device Performance
Encapsulation of photovoltaics is universal; oxygen, water, and other atmospheric
contaminants shorten the lifetime of such devices. Especially vulnerable are organic
solar cells, which also suffer from UV light induced degradation.[29] Though silicon solar
cells typically function for decades, deleterious interactions with the environment require
encapsulation to ensure stability. The devices presented above showed some degradation
over time during continued testing with the lab atmosphere. Experiments to pinpoint the
source of degradation were carried out.
5.4.1 Effect of ambient oxidation
Our initial foray into passivation started with ambient grown oxide. The passivating
effect of silicon oxide has been utilized on the first SWNT-Si devices fabricated by Dr.
Wadhwa and were explored more in depth due to the results in the preceding chapter.[9,
10] Investigations into the kink that is seen in both SWNT-Si cells and MIS solar cells
showed a complex connection between cell performance and oxide thickness at the
SWNT-Si junction. Immediately upon addition of the ionic liquid, a hump shows up
in the J-V curve for all solar cells tested. This was initially attributed to a modulation
of the interface dipole between the carbon nanotube film and the silicon, but further
experiments with the inversion layer cell indicated another possible mechanism: ions
from the ionic liquid situated at the silicon surface within the interstitial regions between
carbon nanotube bundles create a field that either enhances or counteracts the built in
potential. As the depletion layer decreases with an increase in bias voltage, the effect
of these ions increases. Furthermore, the gate voltage affects the relative populations of
cations versus anions within these interstitial regions. Though the kink can be caused by
the mechanisms explained above, this same feature also arises from a sub-optimal oxide
thickness at the silicon-oxide junction.[10, 87] Planar, full film solar cells were constructed
using both <100> and <111> silicon and tested continuously for up to 49 hours to test
the degradation of the cell over time. No gating was performed; only the atmosphere was
79
exposed to the active area. As shown in the plot below, a kink is formed between 7 and 22
hours.
-0.4 -0.2 0.0 0.2 0.4 0.6 0.8-30
-20
-10
0
10
20
30
40
Cur
rent
Den
sity
(mA
/cm
2 )
Voltage
5 min 15 min 30 min 1 hr 2 hr 7hr 22 hr 49 hr
Figure 5-10. Evolution of J-V curve with oxidation in ambient atmosphere.
The kink feature occurs for both little to no oxidation, and for too much oxidation.
We hypothesize that when the silicon surface is freshly etched, it possess shallow trap
states that inhibit carrier extraction and lead to a saturation in current density when the
bias voltage is close to the built in potential. As the bias voltage is increased, the carriers
have enough energy to be excited out of the traps and the J-V curve exhibits exponential
behavior following the Shockley diode equation. Conversely, too much oxide results in
a barrier. At room temperature, the oxidation rate slows dramatically in an ambient
environment after the first few nanometers as subsequent oxidation requires the oxygen
to diffusion through the already formed silicon oxide to reach the silicon as stated in the
Deal-Grove model. This leads to the formation of a kink that progressively becomes more
severe as the oxide layer grows.
80
5.4.2 Reversible doping in ambient environment
In 2009, Martel et al. explored the inability to n-dope carbon nanotube-silicon
devices. It was demonstrated that reduction-oxidation reactions were occurring between
the carbon nanotubes and water vapor present on the substrate surface. Though the
substrate (silicon oxide insulating layer on top of silicon) itself was not participating in the
redox couple, its hydrophilic surface facilitated such reactions through adsorption of water.
These redox reactions transferred electrons from the carbon nanotubes into H2O/O2
redox couple, effectively p-doping the carbon nanotubes.[11] Efforts to n-dope the carbon
nanotubes resulted in immediate transfer of electrons to the redox couple. Replacing the
silicon oxide with a hydrophobic dielectric, such as parylene, diminished this effect and
allowed electron conductance within the carbon nanotubes. We believe this same effect is
occurring in our solar cells during testing in ambient conditions. A reversible reduction in
VOC has been observed during characterization in an inert (argon) atmosphere relative to
testing in the ambient lab environment.
After a BOE etch to remove native oxide, the devices were placed in a small vacuum
chamber that was evacuated and subsequently filled with argon gas. J-V curves were
taken while the substrate sat in this environment for 2 hours under illumination by a high
intensity fiber optic lamp. The resulting J-V curve is characteristic of a device without
residual water vapor in the active area and without a passivating oxide layer, i.e. there is
still a kink due to surface trap states. The initial VOC is lower than what is normally seen
in the SWNT-Si planar solar cells, consistent with both Martel’s conclusions and also a
lack of a passivating layer. The chamber door was opened and the substrate exposed to
ambient atmosphere for numerous hours. The VOC steadily increased, albeit at a slower
rate than usual due to the initial exclusion of water vapor (thus slowing the rate of oxide
growth) and saturated at 16 hours. The chamber was then closed, pumped out, and
backfilled with argon gas. A notable decrease in the VOC resulted, indicating a shift in
81
nanotube Fermi level. When the chamber was opened back to ambient atmosphere, the
VOC increased to its previous value.
-0.2 0.0 0.2 0.4 0.6-5
-4
-3
-2
-1
0
1
2
3
4
5
Cur
rent
Den
sity
(mA
/cm
2 )
Voltage (V)
post BOE Argon 2 hrs Atmosphere 16 hrs Argon 3 hrs Atmosphere 2 hrs Argon overnight
Figure 5-11. Reversibility of the J-V curve upon alternating exposure to argon andambient atmospheres. The legend is ordered vertically based on the order ofmeasurements.
This reversible modulation implicates chemical reactions at the silicon/nanotube
surface in the presence of water or oxygen. Shown in Figure 5-11 are the J-V curves
for the device in both argon and ambient atmospheres. Concurrent with the change in
VOC was a development of a kink when water and oxygen were excluded. This can be
explained as the water having a moderate electrolytic affect on the device similar (but
more subtle) than what is seen with the EMI-BTI. Though this redox initially seems to be
beneficial to device performance by increasing the open circuit voltage, it became apparent
in subsequent experiments that the moisture adsorbed onto the silicon surface oxidizes the
silicon during electronic gating with the EMI-BTI, causing a degradation in performance.
5.4.3 Water vapor and oxygen contamination
Electrolyte gated cells do suffer a serious problem analogous to one that plagued
initially very promising liquid junction solar cells: chemical reactions at the silicon surface
82
degrade cell performance.[88] In the gated cells such degradation was accelerated by the
applied gate voltage so that when held, even for a few minutes at VG= 1.0 V, the J-V
curves began to exhibit an increasing series resistance and decreasing fill factor. Such
characteristics for the degradation suggested a continued growth of the oxide layer between
the SWNTs and the silicon, implying that water/oxygen had access to the junction despite
the hydrophobicity of the electrolyte. This deleterious effect is even more pronounced in
the SiNW devices, presumably attributable to the high surface area available for such
chemical reactions.
5.4.3.1 CV measurements showing IL contamination
To fully exclude water from the devices during electronic gating and ensure that
minimal electrochemical reactions were occurring at the silicon surface, measurements
confirming the wide electrochemical window of the EMI-BTI were carried out. Although
the as-received, EMI-BTI electrolyte was always stored and sampled from an inert
atmosphere glove box (argon, H2O, O2 each < 0.1 ppm), cyclic voltammetry measurements
on the electrolyte performed within the glove box (glassy carbon working-electrode, Ag
wire pseudo reference, Pt counter-electrode) revealed an electrochemical window of only
2.7 V, greatly reduced from its literature reported window of 4.4 V, but consistent with
being contaminated with water.
”Drying” the ionic liquid involved using activated molecular sieves (1:1 mixture of
3A and 4A) to trap water molecules. The sieves were submerged in the ionic liquid for 4
hours, after which cyclic voltammograms were carried out to check the electrochemical
window. As expected, the CV measurements showed a marked improvement after drying,
with the electrochemical window increasing by roughly 600 mV. Removing the dried
EMI-BTI from the glove box and repeating the CV measurement in ambient atmosphere
showed a gradual shrinking of the electrochemical window over the course of a few hours
as the ionic liquid became re-contaminated with water.
83
Figure 5-12. Cyclic voltammograms of the glassy carbon electrode in EMI-BTI ionic liquidat 50 mV
s: black line before treatment; red line after drying over the
molecular sieves. Black and red dotted lines specify the electrochemicalwindows of the EMI-BTI before and after drying, respectively.
The peaks in the CV measurement indicate reduction and oxidation reactions. It’s
unclear exactly what species are being created in the ionic liquid, and applying varying
potentials outside the electrochemical window produce different compounds with their
own associated redox potentials, leading to chemical reactions inside of the EMI-BTI
electrochemical window and degrading device performance. This is apparent in the black
curve in Figure 5-12. The spike at -0.25V is an oxidation reaction corresponding to the
reduction that occurred as the voltage was swept beyond -1.25 V.
5.4.3.2 Exclusion on planar device
Knowing that water contamination of the ionic liquid could lead to electrochemical
reactions inside the theoretical electrochemical window, we proceeded to fully gate a
device in a dry, inert atmosphere. After transferring a SWNT film, a final BOE etch was
84
followed by optimal oxidation in ambient atmosphere. To terminate further oxidation
and to evaporate residual surface water, the device was placed into the argon glove box
where it was stored for 4 days. Concurrently, a sample of the electrolyte was thoroughly
dried in an activated molecular sieve. At the end of this time the active cell area was
saturated with the thoroughly dried electrolyte and J-V measurements were periodically
recorded under illumination, in the glove box, with the gate voltage initially maintained
at a constant VG = -0.75 V. No degradation in any of the J-V characteristics was observed
even after 5 hours at this gate voltage. The gate voltage was subsequently raised to VG
= -1.0 V for an additional 5 hours with still no degradation observed. The gate voltage
was turned off and the device left in the glove box overnight. The next day a gate voltage
of -1.0 V was again applied, with no change in the J-V curve, as shown in Figure 5-13.
The device was subsequently moved into the laboratory ambient atmosphere and retested.
Degradation became noticeable within one hour of exposure to the ambient atmosphere
(at VG = 1.0 V), becoming progressively worse with further exposure, as shown in
Figure 5-14 These experiments strongly implicate water as the source of the degradation
in ambient atmosphere and indicate that by avoiding it such rapid degradation can be
overcome in the gated cells.
Interestingly, the VOC of the planar device was higher outside of the glove box, both
before and after exposure to the inert atmosphere. This reversible behavior is thought to
be caused by oxygen and water redox reactions with the carbon nanotube/silicon junction,
as discussed in Section 5.4.2. Exposure to the atmosphere, even after ionic liquid is
applied, results in an increase in VOC, indicating contamination of water and/or oxygen at
the junction. This theory is corroborated by the degradation in the J-V curve consistent
with water/oxygen contamination.
85
-0.4 -0.2 0.0 0.2 0.4 0.6 0.8-30
-20
-10
0
10
20
30
40
Cur
rent
Den
sity
(mA
/cm
2 )
Voltage (V)
-1V 1hr 2hr 3hr 4hr 5hr next day
Figure 5-13. Stability of planar device with oxygen and water excluded by gating in inertatmosphere with VG=-1.0 V.
5.5 Concluding Remarks
While metal assisted chemical etching of nanowires is a simple, low cost procedure,
the resultant nanowires are rather fragile. Their small diameter not only makes them
susceptible to breakage or agglomeration, but possibly allows complete inversion under
electronic gating. Current collaborations with Oak Ridge National Laboratory aim to
produce silicon nanowire devices with controlled pitches and diameters optimized for
optical absorption. This controlled growth should produce robust, uniform nanowires that
resist agglomeration. The ability to randomly space the nanowires inhibits diffraction and
further increases the optical absorption and uniform heights might eliminate the need for a
sprayed SWNT film.
Producing a record high 15.1% silicon nanowire device is a major advance over
the previous 10%, but only practical if the performance is stable. The propensity for
recombination at the nanowire sidewalls inhibit carrier collection while the high surface
area and severe topology exacerbate any chemical reactions with the ionic liquid. Locally
86
-0.4 -0.2 0.0 0.2 0.4 0.6 0.8-30
-20
-10
0
10
20
30
40
Cur
rent
Den
sity
(mA
/cm
2 )
Voltage (V)
Post glovebox 1hr 3hrs 5hrs Next day
Figure 5-14. Degradation of the planar SWNT-SiNW device upon exposure to atmospherewith VG=-1.0 V.
strong electric fields can catalyze local redox reactions, producing varied chemicals
deleterious to the device. Proof of stability with water exclusion was a critical first step in
proving the viability of these devices, but improved passivation is also imperative to avoid
long term degradation.
87
Table 5-1. Performance for various film deposition techniques and thicknesses
Device VOC (V) JSC(mAcm2
)FF Efficiency (%) Notes
A 0.58 32.5 0.74 13.9 20 nm transferredB 0.58 32.0 0.73 13.5 5 nm sprayed/25 nm transferredC 0.58 32.5 0.71 13.2 45 nm transferredD 0.58 35.0 0.76 15.1 10 nm sprayed/10 nm transferred
88
CHAPTER 6PASSIVATION OF SILICON
The need for excellent passivation of semiconductors has been extensively researched
in regards to both MIS and P-N junction devices. Inherent to p-n homojunctions is
a near perfect boundary when the junction is formed by high temperature diffusion
on a crystalline substrate. Heterojunctions and Schottky junctions suffer from abrupt
boundaries that facilitate recombination through defects and mid gap trap states. As
such, much effort has been put into various chemical and mechanical methods aimed at
creating defect free surfaces and junctions.[32, 89–91] Passivating silicon with an insulating
layer, such as silicon nitride, is a facile method to reduce recombination, but often requires
thicknesses inhibiting charge transport across the junction.[92] The challenge with the
silicon nanowire devices is to passivate the tips in such a way that holes can still be
transferred to the carbon nanotube film, while still providing robust passivation along
the nanowire sidewalls to limit recombination. Limitations on thickness constrain the
passivation layer between the SWNTs and SiNW to be less than a couple nanometers,
while the passivation of the sidewalls must not interfere with the ionic liquid induced
inversion layer.
Our experiments have shown increased degradation within the nanowire devices
as compared to the planar devices upon electronic gating. The planar devices, after
being held at a gate voltage of -0.75V for 20 minutes, showed a small decrease in power
efficiency: 10.12% to 10.06%, a 0.6% decrease. The nanowire based device degraded from
15.1% to 14.4% during the same period at a gate voltage of -1.0 V, an overall decrease
of 4.6%. The increase in surface area simultaneously creates more sites for recombination
of photogenerated carriers and adverse electrochemical reactions between the silicon and
ionic liquid. Excluding water from the junction mitigates the latter, but providing a high
quality passivation layer could help alleviate both problems. Presented below are several
methods of passivation achieved through chemical treatments and atomic layer deposition.
89
Figure 6-1. ALD growth process. 1. Hydroxide terminated substrate is exposed toreactant Trimethylaluminum, which chemiadsorbs onto the substrate,producing methane as a byproduct. 2. System is purged, removing unreactedprecursors and chemical byproducts. 3. Oxidant (water vapor) is fed into thesystem, where it reacts with the methyl ligand, forming OH groups. 4. Systemis again purged of excess species, and the process repeats.
6.1 Atomic Layer Deposition of Al2O3 and HfO
One means to avoid ambient water adsorption is to encapsulate the cells in an inert
atmosphere as must presently be done for other water/oxygen sensitive systems (e.g.
organic solar cells). Alternatively, a thin dielectric barrier layer coating the nanotubes
and SiNWs at the junctions may be sufficient to prevent oxidation due to the water
entrained in the electrolyte. In an attempt to create such a barrier we turned to
atomic layer deposition (ALD) of Al2O3. ALD deposited aluminum oxide has received
increasing interest as a silicon surface passivation layer since the availability of commercial
ALD systems.[93] The layer by layer deposition of vapor phase reactants (sequentially
trimethylaluminum and water) implies a conformal coating even through the pre-deposited
nanotube layer.
ALD is a self limiting process that deposits a monolayer by monolayer conformal
coating of by alternating reactant gasses. In between each reactant gas, the chamber is
purged to vacate any reactants not chemisorbed onto the substrate. Both thermal and
90
plasma/ozone ALD form conformal layers, though the use of unstable and reactive species
in the latter often create small defects in the coating and is only utilized for substrates
that are sensitive to high temperatures.[94] Exposure to ozone for even a couple minutes
causes extreme sidewall damage to the nanotubes, resulting in a substantial loss of
conductivity. Consequently, efforts were made to avoid plasma while still using a relatively
low deposition temperatures. The acid purified nanotubes we use are p-doped by it which
increases the built-in potential against n-type silicon so low temperatures were preferred to
prevent nanotube dedoping.
Figure 6-1 shows the four main phases of ALD Aluminum Oxide growth. Once loaded
into the ALD system and contained within an inert environment, the first reactant gas,
trimethylaluminum (TMA), is fed into the chamber. Reactions with the (OH)− on the
substrate result in chemiadsorbtion of methly terminated aluminum (bonded to oxygen),
creating methane gas as a byproduct. A purge removes any unreacted TMA along
with the methane. Water vapor is then fed into the chamber where it reacts with the
methyl terminated aluminum, releasing methane and resulting in a hydroxide terminated
aluminum. Another purge removes the excess water and methane, leaving the substrate
in a similar chemical state as it was in the first step. The process then repeats, each time
growing a single layer of aluminum oxide.
Prior to growth of the Al2O3 the SWNT/SiNW device underwent a final BOE
etch, followed by oxidation in ambient for the time that optimized device performance
(96 hr). Al2O3 was grown for 110 reactant cycles at a substrate temperature of 80
C.[95] Ellipsometry performed on such a film deposited on a flat silicon chip under these
conditions gave a film thickness of 8.8 nm (coverage in the first several cycles is incomplete
so the initial growth does not give complete layers). Figure 6-2 shows an SEM image of
the Al2O3 coated device. Bright spots in the image are enhanced secondary emission from
where the SiNW tips underlie the dielectric coated nanotubes.
91
Figure 6-2. SEM image of ALD deposition. ALD Al2O3 on SWNT-SiNW device. Brightspots are enhanced secondary emission from the tips of the underlying SiNWtips.
Measurement of parasitic gate currents can quantify the reactions occurring at
the Si surface, some portion of which should correspond to deleterious redox reactions
(other electrolyte or impurity reactions that don’t degrade the SWNT/Si interface can
also occur). For SWNT-SiNW devices without the ALD dielectric coating, the steady
state gate current at VG = -1.0 V in the non-dried electrolyte was typically 2.7 µA. For
the dielectric coated device this was reduced by a factor of 60 to 45 nA. Unfortunately,
this was still a factor of 110 greater that observed for the device measured in the glove
box using the dried electrolyte for which the gate current was 0.4 nA, and while the
rate of degradation of the coated device was greatly reduced over the uncoated device,
degradation was evident over the course of several hours (measured in the ambient
lab atmosphere in the non-dried EMI-BTI electrolyte). This implies that the ALD
layer remains permeable to water at the thickness used. A thicker layer may prevent
this, although a hydrophobic coating (e.g. Parylene) may be preferred to the naturally
hydrophilic oxide in such an application.
92
6.1.1 Al2O3 and HfO results
Figure 6-3 shows the J-V curves for the device before and after electrolyte addition
at gate voltages for the latter of zero and -1.0 V. At VG = -1.0 V the open circuit voltage,
short circuit current density and fill factor were, VOC = 0.62 V, JSC = 33.4 mAcm2 , FF =
0.73 resulting in a PCE of 14.8%. The slightly lower JSC and PCE over the uncoated
device is likely due to an increased light scattering due to the Al2O3.
-0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0-40
-30
-20
-10
0
10
20
30
40
Cur
rent
Den
sity
(mA
/cm
2 )
pre EMI-BTI VG= 0 V VG= -1 V
Figure 6-3. J-V curves for the ALD Al2O3 coated SWNT/SiNW cell. Without electrolyte(red) and with the electrolyte at the indicated gate voltages (black, blue). Theenhanced photocurrent on addition of the electrolyte is attributed to refractiveindex matching reducing the scattering.
The ALD results demonstrated good passivation of the SiNWs. The quantitative
reduction in the gate current indicating reduced redox occurring at the silicon/nanotube
junction is corroborated by the decreased degradation of the J-V curve, shown in
Figure 6-4. The decrease in performance for a non-ALD SiNW device is shown for
comparison, albeit the measurements are taken for different times (48 hours vs 72 hours),
93
-0.5 0.0 0.5-40
-30
-20
-10
0
10
20
30
40
No ALD -0.75V initial No ALD -0.75V 72hrs No ALD -0.75V 72hrs hold for 20min ALD -0.75V initial ALD -0.75V 48hrs ALD -0.75V 48hrs hold for 60min
Jsc
(mA
/cm
^2)
Voltage (V)
Figure 6-4. J-V curves for ALD SWNT-SiNW device vs device without ALD. Included isthe degradation occurring both with applied gate voltage and time spent inambient without the gate voltage applied. The ”No ALD -0.75V 72 hours” and”ALD -0.75V 48 hours” graphs indicate the device was sitting in ambientwithout the gate voltage on (but ionic liquid still in place) for the specifiedamount of time.
so an exact comparison is not provided. provided. The Al22O33 coating clearly helps but
it is clearly also not sufficient to overcome the electrochemical degradation. While it is
possible that more water had diffused into the device at 72 hours, and hence could exhibit
a higher degradation due to enhanced redox, we do not think that is the primary factor for
the increase in degradation relative to the ALD device.
Noting that Hafnium Oxide (HfO) possesses a dielectric constant twice that of Al2O3,
we also deposited a 25 nm thick AD HfO layer onto SWNT-SiNW substrates (already
oxidized for 96 hours in the lab atmosphere). It was reasoned that the thicker layer
would further reduce electrolyte access to the silicon surface while still providing the
same capacitance as the Al2O3. Device stability, specifically gate currents, were markedly
94
-0.4 -0.2 0.0 0.2 0.4 0.6 0.8-40
-30
-20
-10
0
10
20
30
40
50
60
Cur
rent
Den
sity
(mA
/cm
2 )
Voltage (V)
25 nm HfO ALD 8.8 nm Al2O3
Figure 6-5. J-V for the ALD HfO device showing a lowering JSC due to the highreflectance of the device.
smaller than for the 8.8 nm Al2O3. The gate current at VG = -1.0 V was 13 nA, as
opposed to the 60 nA for the thinner aluminum oxide ALD layer, indicating that a thicker
ALD inhibits redox reactions by limiting contact between the ionic liquid and the silicon
surface. The thicker ALD layer wasn’t without drawbacks; JSC was only 30 mAcm2 due to the
higher scattering induced by the thicker HfO layer, which had an associated haze visible to
the naked eye. Reduced degradation in the J-V curve of the HfO device was quantifiable
- After 30 minutes of continuously held gate voltage at -1V, the reduction in PCE was
measured to be 0.98%, compared to a degradation of 2.1% for the 8.8 nm Al2O3, which
was held at a lower gate voltage of -0.75V.1
1 Some of the measurements made on different devices were not identical, leading to discrepancies whentrying to compare degradation. Care was taken to try and compare results under similar gating condi-tions. Early devices, unfortunately, were not as extensively tested as later devices, resulting in less data forcomparison.
95
To assess the ability of the ALD layer to protect the silicon from redox reactions
with the ionic liquid and to fully evaluate the effect of the deposited dielectric, we also
deposited Al2O3 onto a planar SWNT-Si device. As for the previous devices, the solar
cell was allowed to oxidize in the ambient atmosphere for 2 hours prior to ALD growth,
which was found to be the optimum time for planar devices. All solar cell parameters
improved in the gated ALD device as compared to non-ALD devices. Most notable is the
increase in JSC both with and without ionic liquid, indicating the ALD layer acts as an
index matching layer that reduces reflection. With a refractive index of 1.55[96], the Al2O3
is well below the refractive index of silicon, n= 3.96, and similar to refractive index of the
ionic liquid, n = 1.42.2 Data are shown in Table 6.4 at the end of the chapter.
The ALD devices clearly exhibit superior performance relative to the non ALD
devices, though there is still some degradation. Current efforts are aimed at testing the
ALD devices in the glove box with the dried EMI-BTI. Complete exclusion of water
in conjunction with high quality passivation should afford excellent means to avoid all
sources of degradation. Despite the promising results of incorporating ALD into the
device, we also explored other methods of producing the same stability. Photovoltaics
must be stable in addition to inexpensive in order to compete on the open market,
motivating a search to find a facile, scale-able, and inexpensive means to protect the
modules from degradation.
6.2 Hydroquinone
Hydroquinone, an organic compound commonly used as a bleaching agent and
photographic developer, has recently been shown to produce ideal Schottky junctions
when used to treat to a silicon substrate.[97] Molecules attached to the dangling silicon
bonds create a surface dipole, changing the effective electron affinity of the semiconductor.
Through use of different additives, the length of the dipole can be altered, leading to a
2 Taken from TCI Chemicals website: http://www.tcichemicals.com/eshop/en/us/commodity/E0599/
96
A Silicon surface showing danging bonds afterremoval of oxide layer.
B Hydroquinone moleculewith attached methoxy (CH3)group, forming a surface dipoleand simultaneously passivatingthe surface.
Figure 6-6. Silicon substrate and hydroquinone molecule. Reprinted with permission fromR. Har-Lavan, et al., AIP Advances 2, 012164 (2012).
tunable effective electron affinity and subsequent change in Schottky barrier height. In the
paper by Har-Lavan, et al., it was shown that methanol yields that largest negative dipole,
yielding an effective electron affinity of 3.05 eV. This created a barrier height of nearly 1
eV, remarkable in that it is comparable to the band gap of silicon. Due to the dipole, the
silicon at the junction became strongly inverted, leading to carrier transport dominated
by minority carrier generation and recombination. Most important for our purposes, the
surface of the silicon was passivated, with stability shown over several days (as opposed to
hydrogen termination which is stable up to a few hours in ambient).
The preparation was modified slightly from the procedure done by Har-Lavan, et al.
A SWNT-Si device was prepared following the standard procedure, finishing with a final
BOE etch to remove any native oxide growth and leaving the silicon hydrogen terminated
at the surface. The substrate was placed in a 0.01M Hydroquinone in methanol solution
for 3 hours in a dark, ambient environment. The original paper called for sonication
in ethyl acetate, but the violence of such a process would have damaged the carbon
nanotube film. Treating the substrate prior to transferring the carbon nanotube film
produced devices with a higher series resistance and poor fill factor, indicating adverse
reactions between the hydroquinone and solvents used in the transfer process. As such,
97
HQ treatments were done after the film had been transferred, after which the substrate
was dipped into boiling dichloromethane for 30 seconds to remove any excess molecules
not firmly adsorbed onto the surface. This was followed by drying in an N2 stream and
construction of the backside contact.
Shown in Figure 6-7 are the J-V curves for the hydroquinone device. Initial
performance is comparable to that achieved without HQ treatment, with a VOC of
0.56V, FF =0.74 , JSC = 23.6 mAcm2 , and PCE of 9.73%. Addition of EMI-BTI to the active
area results in immediate degradation, with severe, irreversible degradation after electronic
gating, indicating that HQ passivation is not compatible with inclusion of ionic liquid.
Gated results for VG = -0.75V are VOC of 0.52V, FF =0.21 , JSC = 8.6 mAcm2 , and PCE
of 0.93%. Though the previous experiments on water and oxygen exclusion indicate that
the problem could be attributed to additional chemical interactions between the ionic
liquid and water or the hydroquinone and water, ultimately forming several compounds
deleterious to device stability. Stability measurements of the hydroquinone passivated
devices without electronic gating are presently under investigation.
6.3 Sulfur
Sulfur has been used to successfully passivate GaAs <100> and InP substrates and in
2007 it was shown that sulfur provides an excellent passivation for <100> silicon, resulting
in near ideal Schottky barriers with both low and high work function metals.[98] Though
the adherence of a foreign molecule to dangling silicon bonds forms an interface dipole,
experiments show the effect of this interface dipole on the SBH to be negligible relative
to the reduced Fermi level pinning achieved through a reduction of surface states. With
an electronegativity of 2.58, sulfur attracts electrons when covalently bonded to silicon
(electronegativity of 1.9),[99] resulting in a surface dipole in opposition to the built in
potential in the silicon at the SWNT-Si junction. Nonetheless, the passivation afforded by
this chemical treatment reduces the surface recombination velocity and subsequently the
98
-0.4 -0.2 0.0 0.2 0.4 0.6 0.8-40
-30
-20
-10
0
10
20
30
40
50
60
Cur
rent
Den
sity
(mA
/cm
2 )
Voltage (V)
pre EMI-BTI 0V -0.75V 0V post gating
Figure 6-7. J-V curve of the HQ treated planar cell before, during, and after electronicgating with EMI-BTI.
dark current, negating any decrease in the Vbi and leaving VOC largely unaffected, shown
in Figure 6-8A.
Early experiments with sulfur passivation following the procedure in Ali et al.
resulted in poor performance due to low FF and high RS. In light of research conducted
by Aibin, et al., we concluded that the time spent in the sulfur solution allowed growth
of large crystallites, thereby contaminating the silicon surface with large particulates and
impeding charge collection. Shortening the time in solution by a factor of 4 produced a
uniform monolayer of sulfur, passivating the surface without creating a physical barrier to
hole extraction.[100]
Similarly to the HQ experiments, sulfur passivation was done both prior to and after
SWNT film transfer, with optimal results for treatment after film transfer. After the
prepared devices underwent a final BOE etch to remove native oxide, they were placed in
99
a 0.33M/2.4M NH4)2S/NH4OH solution at 60C for 5 minutes. This was followed by two
DI baths of 5 minutes each, and an N2 dry. J-V curves were taken immediately after the
back contact had been assembled. Initial results showed excellent performance relative
to unpassivated devices. Addition of EMI-BTI and subsequent gating indicated negative
interactions of the sulfur with the ionic liquid. Initial gating curves were excellent, but
continued application of the gate voltage resulted in irreversible degradation, as shown in
Figure 6-8A.
A likely culprit in the poor performance is the presence of water, which was shown in
Chapter 4 to degrade device performance. Further experiments performed in a glove box
using dried ionic liquid will be used to check if this is correct.
6.4 Discussion and Summary
Given that encapsulation or a better optimized passivation layer should prevent the
electrochemical degradation, we consider other aspects of our present devices that could
limit their performance (suggesting means to increase their PCEs beyond the present
15%). One limitation concerns excess recombination at the back contact. It has long been
known that a back surface field induced by appropriate doping of the Si at the device back
contact can reduce recombination there, with corresponding improvements in the device
performance.[101] Another factor also limiting the PCE in our present design is their
geometry. In the present construction the Si wafer thickness (550 µm) is large relative
to the active area width (2 mm) meaning that photocarriers created near the edge of
the active window region have an appreciable cross-section for escape out the sides of
that region, thus contributing to the losses. Capturing those carriers could significantly
boost the device PCE. Quite recently PCEs comparable to what we demonstrate here
were obtained from planar, chemically charge transfer doped nanotube/Si solar cells
exploiting a TiO2 antireflection coating.[102] The broad-band reflectance due to that
coating was not as low as our vertical NW arrays , exhibiting a minimum of 5% at 600
nm, but rising smoothly on either side of the minimum to over 10% at 500 nm and 800
100
-0.4 -0.2 0.0 0.2 0.4 0.6 0.8-40
-30
-20
-10
0
10
20
30
40
50
60
Cur
rent
Den
sity
(mA
/cm
2 )
Voltage (V)
Post BOE etch to backside 4 hours in ambient EMI-BTI -0.75V EMI-BTI -0.75V 20 minutes
A Performance of the sulfur passivated device (post SWNT film transfer)
-0.4 -0.2 0.0 0.2 0.4 0.6 0.8-40
-30
-20
-10
0
10
20
30
40
50
60
Cur
rent
Den
sity
(mA
/cm
2 )
Voltage (V)
HQ pre EMI-BTI HQ at -0.75V S pre EMI-BTI S at -0.75V
B Hydroquinone versus Sulfur passivation upon electronic gating.
Figure 6-8. J-V curves for sulfur and hydroquinone passivated devices.
101
nm, respectively and to over 20% at the extremes of the relevant solar spectrum (400-1100
nm). The comparable performance, despite our reduced reflectance, suggests that their
devices exhibited lower losses which could be due to their more optimized geometry.
They used thinner (400 µm) wafers and larger active cell area (15 mm2 vs. our 8 mm2)
consistent with a reduced carrier leakage out the sides of their active region. Combing the
results presented here with future strategies for further improvement bode well for rapid
advancement of nanotube-silicon devices since their initial development in 2007.[60]
102
Table 6-1. Performance for ALD devices for VG = -1.0 V
Device ALD type and thickness Voc (V) Jsc(mAcm2
)FF Efficiency (%)
SiNW 8.8 nm Al2O3 0.62 33.2 0.73 14.84SiNW 25 nm HfO 0.58 29.7 0.72 9.74Planar 8.8 nm Al2O3 0.6 28 0.8 13.45
103
CHAPTER 7ADDITIONAL PROJECTS
7.1 TFSA Doping of Graphene-Si and Carbon Nanotube-Si Devices
Graphene has long been used as an electrode for organic photovoltaics, [103, 104]
but only in the past couple years has it been demonstrated in Schottky junction solar
cells.[105, 106] The first generation of devices achieved a PCE of approximately 1.7%,
as the as prepared graphene introduced high levels of resistance into the device which
reduced both the FF and JSC while increase RS. Nonetheless, we decided to integrate
graphene into our architecture in an attempt to mitigate the deleterious effects of excessive
oxide growth at the junction and to limit redox reactions between the silicon and ionic
liquid. With it’s hexagonally close packed carbon atoms and relatively low permeability
of water vapor and oxygen, the graphene lattice is suitable for hindering silicon oxidation
at the surface of the active area. This should stabilize the series resistance and hinder the
formation of the kink that accompanies an oxide induced barrier. Unless highly doped, a
single monolayer of graphene is too resistive act as an electrode, resulting in the low PCE
reported above. Additionally, a graphene-CNT hybrid device could offer the best of both
worlds: retention of the optimal oxide thickness and a low series resistance. Additionally,
graphene would be able to screen the silicon surface from the ions in ionic liquid, avoiding
redox reactions and leading to more stable performance with electronic gating.
7.1.1 Graphene-Si Solar Cells
The fabrication process for the graphene-Si device was modified slightly due to
the impermeability of the graphene; oxygen and water vapor cannot easily penetrate
the lattice to readily oxidize the silicon surface. The SWNT devices underwent a final
BOE etch to remove native oxide, followed by oxidation in ambient atmosphere for 2
hours until the JV curve exhibited stable, maximum performance. This process is not
practical for graphene devices, as the BOE cannot permeate the graphene to etch the
underlying oxide (though a tear in the graphene could allow etching of the underlying
104
oxide, reaction products could not diffuse out). Our modified fabrication process involved
BOE etching the gold framed active area window and exposing the substrate to the
ambient atmosphere for 2 hours to facilitate oxide growth before transfer of the graphene.
A drop of isopropanol (IPA) was placed on the window and a PMMA supported graphene
sheet, synthesized and provided by Dr. Max Lemaitre and Dr. Art Hebard, was placed
on top of the IPA (graphene side against the silicon). Pressure to adhere the graphene
to the silicon was supplied by aluminum blocks. After 4 hours at room temperature, the
IPA had evaporated and the graphene was held tightly to the substrate via Van der Walls
forces. An acetone vapor bath followed by several acetone soaks removed the PMMA
membrane. Fabrication of the back contact and characterization proceeded following
standard procedures.
As expected, the device exhibited a higher series resistance relative to the carbon
nanotube devices. Though graphene has high mobility its carrier density is low resulting
in a resistivity that makes for a high series resistance. Prior to addition of the EMI-BTI,
PCE was 1.9%. Addition of EMI-BTI and subsequent gating at -0.75V increased the
efficiency to 4.6%, an overall increase of over 140%. While this huge improvement
indicates facile manipulation of graphene’s electronic properties, the performance was
several factors below that of the carbon nanotube-silicon devices. We hypothesized
that a hybrid-graphene-SWNT device would exhibit better performance though a
confluence stability from the graphene layer and improved conductivity from the carbon
nanotube layer. Unfortunately the results for the hybrid device were worse than for the
graphene-only device, presumably due to silicon surface states formed during the excessive
amount of processing involved in transferring graphene and SWNTs. The JV curves are
shown in Figure 7-1 and cell characteristics shown in line one of Table 7-1.
In collaboration with Dr. Xiaochang Miao and Dr. Seffattin Tongay, a graphene-Si
solar cell was fabricated and then doped with Trifluoromethanesulfonyl-amide (TFSA,
(CF3SO2)2NH). Extensive research on dopants for graphene has been done the past several
105
-0.4 -0.2 0.0 0.2 0.4 0.6 0.8-40
-30
-20
-10
0
10
20
30
40
50
60
Cur
rent
Den
sity
(mA
/cm
2 )
Voltage (V)
pre EMI-BTI VG = 0 V VG = -0.75V
Figure 7-1. JV curve for the monolayer graphene device.
years, with TFSA (alternatively, TFSI) proving to be highly p-doping and stable due
to its hydrophobic nature.[107] Graphene devices were fabricated following the method
outlined above, with PCEs of 1.5-3%. A 20 mM solution of TFSA in nitromethane was
spun coated onto the graphene at 1000-1500 RPM for 1 minute, after which the devices
were characterized. Immediate improvement occurred for all solar cell parameters, as
summarized in Table 7-1 at the end of the chapter and shown in Figure 7-2B. The PCE
jumped to 8.6% due to a dramatic increase in JSC , VOC , and FF. The strong charge
transfer doping achieved with the TFSA proved to be fairly stable for the graphene device,
resulting in negligible efficiency decreases over several days.[106, 108]
The TFSA doping increased the Schottky barrier height by approximately 0.11eV,
from 0.79eV to 0.89eV, resulting in an increase of the built in potential and a reduction
of dark current due to reduced recombination at the junction. Additionally, the external
quantum efficiency increased by roughly 30% over visible wavelengths, indicating a more
106
A Architecture of the graphene cell.Identical to the SWNT-Si planar deviceswithout the on chip gate film.
B J-V curves showing effect of TFSA doping
Figure 7-2. Schematic and performance for graphene PV cell.
efficient collection of carriers most likely attributed to the increased built in potential. The
thin layer of TFSA potentially acts as an antireflection layer, as darkening of the active
area window post TFSA treatment was noticeable.
7.1.2 TFSA with carbon nanotubes
Given the excellent results with the TFSA doped graphene device, we set out to
incorporate the same treatment for the carbon nanotube films. Though our as-purified
SWNTs are already p-doped, the TFSA provides a higher level of doping as evidenced
by the increased VOC and smaller RS. Following the standard procedure for fabricating
SWNT-Si devices, the device was oxidized in the ambient atmosphere for 2 hours until
the JV curve revealed stable performance, after which the TFSA was spun coated onto
the device. Immediate testing revealed enhanced performance relative to the undoped
and ungated performance. Unlike the graphene device, the TFSA doping of the SWNTs
appeared to be transitory and a slow degradation in all parameters was noticed over
time. While we do not know the exact cause of the discrepancy between the stability
of the graphene devices vs. the carbon nanotube devices, we believe it is attributed to
the reduced, but still finite, access of atmospheric water to the line contact between the
107
nanotubes and the silicon, permitting oxidation there. The TFSA dopant was shown to
be highly incompatible with the ionic liquid (as expected due to their chemical reactivity),
and as such this method of doping would only be viable for ungated encapsulated devices.
Shown in Figure 7-3 are the JV curves for the TFSA doped carbon nanotube device.
-0.4 -0.2 0.0 0.2 0.4 0.6 0.8-40
-30
-20
-10
0
10
20
30
40
50
60
Cur
rent
Den
sity
(mA
/cm
2 )
Voltage (V)
Post BOE etch and oxidation Post TFSA TFSA at 50 min EMI-BTI VG = -0.75V
Figure 7-3. J-V curves showing effect of TFSA doping and subsequent gating on SWNT-Sidevice.
7.2 Backside Doping
Several experiments were performed attempting backside doping of the silicon
substrates to create a back surface field there. This strongly doped region has been used
extensively in p-n junction solar cells to reduce recombination.[109] An n+ - n junction
exhibits similar properties of a p-n junction: in both cases carriers diffuse from one region
to another until electrostatic equilibrium is reached. The resulting electric field serves
to sweep carriers of opposite charges in opposite directions. When employed on the
backside of photovoltaic cells, this secondary built in potential serves to reduce carrier
108
recombination at the back contact. The doping procedure is fairly straightforward: a
phosphorus spin on dopant (Filmtronix P509) is spun coated at 2500 RPM onto the
backside of the silicon substrate and baked at 1150 C in ambient atmosphere for 2
hours (ramping at 1 degree C per minute). The high temperature allows the dopant to
diffuse into the silicon, creating a shallow, gradated region of n+ doping. Preliminary
testing yielded a sheet resistance of 1.5-3 ohm/square on the backside of the wafers
(versus 20-60 ohm/square for the undoped wafers) and a modest increase in PCE was
obtained. The wafers used were also thinned to 250 µm before doping the backside, a
modification that should minimize recombination by reducing the cross-section for carrier
leakage out the sides of the active area. Additionally, the diffusion length of electrons in
lightly doped n-type silicon is on the order of hundreds of microns (depending on silicon
grade, growth parameters, and doping density) so reducing the path of travel should
result in an increased photocurrent. This improvement in comes at a price: slightly less
photoconversion from photons near the band gap of silicon. The absorption length of
silicon for light with energy 1.1eV is approximately 6.6 mm, with higher energy photons
having a shorter absorption length.[110] However, the irradiance of infrared photons in the
solar spectrum is relatively low, so the net change in photons absorbed within the silicon is
only marginally affected.
The higher VOC is attributed to a reduction in recombination, which subsequently
lowers the dark current and increases the open circuit voltage. Additionally, less
recombination of electron hole pairs generated near the backside increased JSC, as shown
in Table 7-2 at the end of the chapter. Though this method was highly successful at
increasing the PCE, the process was only viable on silicon with a thick (>1 µm) oxide
layer. The high temperature bake formed blisters on the surface of the substrate, creating
shorts within the device, shown in Figure 7-4. Altering the temperature and ramp rate
were unsuccessful at preventing the defects. Future projects should look to alternative
methods of implanting a highly doped backside region to reliably boost performance.
109
Figure 7-4. Blistering on the surface of the silicon following a high temperature bake todope the backside.
7.3 Concluding Remarks and Path Forward
The improvements discussed in the previous chapters indicate the potential for
further improvement of the SWNT-SiNW devices by increasing both efficiency and
longevity. The ALD proved to be an excellent barrier between the silicon and the ionic
liquid without sacrificing the inversion layer needed to reduce recombination. Though
degradation did eventually occur in the ALD devices, parasitic gate currents indicate that
the electrochemical reactions were greatly reduced compared to the non-ALD SiNW based
devices. Additionally, it was demonstrated that water vapor and oxygen contamination at
the surface of the device were responsible for the decrease in performance during electronic
gating, necessitating testing within an inert atmosphere to elicit stable performance.
Lastly, a highly doped region at the back contact was effective at reducing recombination
and boosting PCE. Incorporating these three mechanisms into photovoltaic designs
promises to deliver devices that achieve a PCE above 15%, but also stable.
Future projects are currently underway, including a collaboration with Oak Ridge
National Laboratory to develop more ordered silicon nanopillar devices. These new
architectures will be more robust than the nanowire. The ability to selectively control
diameter will allow us to explore the charge collection efficiency when the entire nanopillar
is either completely or partially inverted. Though beyond the scope of this thesis, the
110
ability of the carbon nanotube film to conform to the substrate surface would make it an
idea candidate for forming Schottky junctions with amorphous silicon. A short diffusion
length due to multiple grain boundaries necessitate excellent connection between the
electrode and the silicon. The SWNT film within our Schottky junction devices also
acts as a transparent electrode, making it an ideal candidate for a-Si based devices. The
flexibility of the carbon nanotube film compliments the malleability of amorphous silicon,
raising the possibility of a flexible, thin film solar cell.
The vast majority of experimental breakthroughs realized at the academic level
never reach the consumer market. As daunting and discouraging as it may seem to the
budding scientist, no experiment is completely disconnected from either contemporary
innovations or the research conducted in the past. The primary goal of science is to
increase the knowledge of how the world works, disseminate that knowledge to those who
are interested, and ultimately use that knowledge to advance society. The advances made
during these experiments in photovoltaic cells will hopefully benefit other scientists and
contribute to technological innovation.
111
Table 7-1. Performance summary for TFSA doped graphene and SWNT solar cells
Device Voc (V) Jsc(mAcm2
)FF Efficiency (%) Notes
Graphene 0.42 14.2 0.32 1.9Graphene w/TFSA 0.54 25.3 0.63 8.6Graphene w/ EMI-BTI 0.52 26.84 0.34 4.6 VG = -0.75VSWNT 0.51 24.0 0.72 8.97SWNT w/TFSA 0.54 26.0 0.75 10.43SWNT w/TFSA and EMI-BTI 0.45 25.65 0.21 2.45 VG = -0.75V
Table 7-2. Performance for backside doped substrates
Device Voc (V) Jsc(mAcm2
)FF Efficiency (%) Notes
<111> 500 µm 0.54 25 0.74 9.98 VG = -0.75V<111> 250 µm 0.52 26.7 0.65 9.74 VG = -0.75V<111> 250 µm doped 0.58 29.6 0.79 13.42 VG = -0.75V
112
APPENDIX AFULL SIMULATIONS FOR THE INVERSION LAYER CELL
Figure A-1. Modeling of the inversion layer at the silicon surface in the carbon nanotubegrid solar cell
113
APPENDIX BSOLAR CELL PARAMETERS WITH INCREASING OXIDATION TIME
Figure B-1. FF, JSC , VOC , and PCE for a SWNT-SiNW device for various oxidation timesin the lab atmosphere.
114
REFERENCES
[1] M. A. Green, K. Emery, Y. Hishikawa, W. Warta, and E. D. Dunlop, Progress inPhotovoltaics: Research and Applications 20, 12 (2012).
[2] T. J. Webster, M. C. Waid, J. L. McKenzie, R. L. Price, and J. U. Ejiofor,Nanotechnology 15, 48 (2004).
[3] S. Polizu, O. Savadogo, P. Poulin, and L. Yahia, Journal of nanoscience andnanotechnology 6, 1883 (2006).
[4] A. Guiseppi-Elie, C. Lei, and R. H. Baughman, Nanotechnology 13, 559 (2002).
[5] N. W. S. Kam, T. C. Jessop, P. A. Wender, and H. Dai, Journal of the AmericanChemical Society 126, 6850 (2004).
[6] R. Saito, G. Dresselhaus, and M. S. Dresselhaus, Physical properties of carbonnanotubes (Imperial College Press, 1998) gb98064666.
[7] M. S. Dresselhaus, G. Dresselhaus, and P. C. Eklund, Science of fullerenes and
carbon nanotubes (Academic Press, 1996) 95020629.
[8] Z. Wu, Z. Chen, X. Du, J. M. Logan, J. Sippel, M. Nikolou, K. Kamaras, J. R.Reynolds, D. B. Tanner, A. F. Hebard, and A. G. Rinzler, Science 305, 1273 (2004).
[9] P. Wadhwa, B. Liu, M. A. McCarthy, Z. Wu, and A. G. Rinzler, Nano Letters 10,5001 (2010), doi: 10.1021/nl103128a; M3: doi: 10.1021/nl103128a; 14.
[10] P. Wadhwa, G. Seol, M. K. Petterson, J. Guo, and A. G. Rinzler, Nano Letters 11,2419 (2011), doi: 10.1021/nl200811z; M3: doi: 10.1021/nl200811z; 14.
[11] C. M. Aguirre, P. L. Levesque, M. Paillet, F. Lapointe, B. C. StAntoine,P. Desjardins, and R. Martel, Advanced Materials 21, 3087 (2009).
[12] S. Iijima, Nature 354, 56 (1991), m3: 10.1038/354056a0; 10.1038/354056a0.
[13] T. Guo, P. Nikolaev, A. Thess, D. Colbert, and R. Smalley, Chemical PhysicsLetters 243, 49 (1995).
[14] A. Thess, R. Lee, P. Nikolaev, H. Dai, P. Petit, J. Robert, C. Xu, Y. H. Lee, S. G.Kim, and A. G. Rinzler, Science-AAAS-Weekly Paper Edition 273, 483 (1996).
[15] C. M. Schauerman, J. Alvarenga, B. J. Landi, C. D. Cress, and R. P. Raffaelle,Carbon 47, 2431 (2009).
[16] M. Yudasaka, R. Yamada, N. Sensui, T. Wilkins, T. Ichihashi, and S. Iijima, TheJournal of Physical Chemistry B 103, 6224 (1999).
[17] S. Bandow, S. Asaka, Y. Saito, A. Rao, L. Grigorian, E. Richter, and P. Eklund,Physical Review Letters 80, 3779 (1998).
115
[18] C.-K. Chen, W. L. Perry, H. Xu, Y. Jiang, and J. Phillips, Carbon 41, 2555 (2003).
[19] P. Nikolaev, M. J. Bronikowski, R. K. Bradley, F. Rohmund, D. T. Colbert,K. Smith, and R. E. Smalley, Chemical physics letters 313, 91 (1999).
[20] W. Li, S. Xie, L. X. Qian, B. Chang, B. Zou, W. Zhou, R. Zhao, and G. Wang,Science (New York, NY) 274, 1701 (1996).
[21] H. Zhu, C. Xu, D. Wu, B. Wei, R. Vajtai, and P. Ajayan, Science 296, 884 (2002).
[22] H. Cheng, F. Li, G. Su, H. Pan, L. He, X. Sun, and M. Dresselhaus, AppliedPhysics Letters 72, 3282 (1998).
[23] A. Rinzler, J. Liu, H. Dai, P. Nikolaev, C. Huffman, F. Rodriguez-Macias, P. Boul,A. H. Lu, D. Heymann, and D. Colbert, Applied Physics A: Materials ScienceProcessing 67, 29 (1998).
[24] Y. Q. Xu, H. Peng, R. H. Hauge, and R. E. Smalley, Nano letters 5, 163 (2005).
[25] H. Hu, B. Zhao, M. E. Itkis, and R. C. Haddon, The Journal of Physical ChemistryB 107, 13838 (2003).
[26] R. Graupner, J. Abraham, A. Vencelov, T. Seyller, F. Hennrich, M. M. Kappes,A. Hirsch, and L. Ley, Phys.Chem.Chem.Phys. 5, 5472 (2003).
[27] J. Nelson and M. Ratner, Physics Today 57, 71 (2004).
[28] C. C. Koch, Nanostructured materials: processing, properties and potential applica-
tions (Noyes Publications / William Andrew Publishing, 2002) 2001096788.
[29] H. Hoppe and N. S. Sariciftci, Journal of Materials Research 19, 1924 (2004).
[30] S. M. Sze and K. K. Ng, Physics of semiconductor devices (Wiley-Interscience,2007).
[31] H. Okamoto, H. Kida, S. Nonomura, K. Fukumoto, and Y. Hamakawa, Journal ofApplied Physics 54, 3236 (1983).
[32] E. Yablonovitch, D. Allara, C. Chang, T. Gmitter, and T. Bright, Physical ReviewLetters 57, 249 (1986).
[33] M. Imaizumi, T. Ito, M. Yamaguchi, and K. Kaneko, Journal of Applied Physics81, 7635 (1997).
[34] W. Shockley and H. J. Queisser, Journal of Applied Physics 32, 510 (1961).
[35] T. Tiedje, E. Yablonovitch, G. D. Cody, and B. G. Brooks, Electron Devices, IEEETransactions on 31, 711 (1984).
[36] M. PECKERAR, Washington DC , 213 (1975).
116
[37] A. D. Vos, Journal of Physics D: Applied Physics 13, 839 (1980).
[38] S. O. Kasap, Principles of electronic materials and devices, Vol. 81 (McGraw-HillNew York, NY, 2006).
[39] M. Knupfer, Applied Physics A 77, 623 (2003).
[40] S. Gunes, H. Neugebauer, and N. S. Sariciftci, Chemical Reviews-Columbus 107,1324 (2007).
[41] J.-M. Nunzi, Comptes Rendus Physique 3, 523 (2002).
[42] S. Jeong, E. C. Garnett, S. Wang, Z. Yu, S. Fan, M. L. Brongersma, M. D.McGehee, and Y. Cui, Nano letters 12, 2971 (2012).
[43] U. Bach, D. Lupo, P. Comte, J. Moser, F. Weissrtel, J. Salbeck, H. Spreitzer, andM. Grtzel, Nature 395, 583 (1998).
[44] D. L. Pulfrey, Electron Devices, IEEE Transactions on 25, 1308 (1978).
[45] R. Godfrey and M. Green, Applied Physics Letters 31, 705 (1977).
[46] R. Godfrey and M. Green, Applied Physics Letters 34, 790 (1979).
[47] R. Call, (1973).
[48] N. W. Ashcroft and N. D. Mermin, Saunders College, Philadelphia 1 (1976).
[49] C. Crowell and S. Sze, Solid-State Electronics 9, 1035 (1966).
[50] H. A. Bethe, Theory of the boundary layer of crystal rectifiers (RadiationLaboratory, Massachusetts Institute of Technology, 1942).
[51] A. Cowley and S. Sze, Journal of Applied Physics 36, 3212 (1965).
[52] J. Andrews and M. Lepselter, Solid-State Electronics 13, 1011 (1970).
[53] J. Bardeen, Physical Review 71, 717 (1947).
[54] R. T. Tung, Physical review B 64, 205310 (2001).
[55] R. T. Tung, Physical Review Letters 84, 6078 (2000).
[56] R. Tung, Physical Review Letters 52, 461 (1984).
[57] M. W. Rowell, M. A. Topinka, M. D. McGehee, H. J. Prall, G. Dennler, N. S.Sariciftci, L. Hu, and G. Gruner, Applied Physics Letters 88, 233506 (2006).
[58] A. D. Pasquier, H. E. Unalan, A. Kanwal, S. Miller, and M. Chhowalla, AppliedPhysics Letters 87, 203511 (2005).
[59] E. Kymakis and G. Amaratunga, Applied Physics Letters 80, 112 (2002).
117
[60] J. Wei, Y. Jia, Q. Shu, Z. Gu, K. Wang, D. Zhuang, G. Zhang, Z. Wang, J. Luo,and A. Cao, Nano letters 7, 2317 (2007).
[61] W. Chen, G. Seol, A. G. Rinzler, and J. Guo, Applied Physics Letters 100, 103503(2012).
[62] E. H. Rhoderick and R. H. Williams, Metal-semiconductor contacts (Clarendon PressOxford, 1988).
[63] G. Myburg, F. Auret, W. Meyer, C. Louw, and M. V. Staden, Thin Solid Films325, 181 (1998).
[64] Y. F. Chang, Q. R. Chou, J. Y. Lin, and C. H. Lee, Applied Physics A: MaterialsScience Processing 86, 193 (2007).
[65] C. M. Hsu, S. T. Connor, M. X. Tang, and Y. Cui, Applied Physics Letters 93,133109 (2008).
[66] M. J. Huang, C. R. Yang, Y. C. Chiou, and R. T. Lee, Solar Energy Materials andSolar Cells 92, 1352 (2008).
[67] K. Peng and J. Zhu, Journal of Electroanalytical Chemistry 558, 35 (2003).
[68] K. Peng, M. Zhang, A. Lu, N. B. Wong, R. Zhang, and S. T. Lee, Applied PhysicsLetters 90, 163123 (2007).
[69] K. Q. Peng, X. Wang, X. Wu, and S. T. Lee, Applied Physics Letters 95, 143119(2009).
[70] F. Wang, H. Y. Yu, X. Wang, J. Li, X. Sun, M. Yang, S. M. Wong, and H. Zheng,Journal of Applied Physics 108, 024301 (2010).
[71] J. Zhu, Z. Yu, G. F. Burkhard, C. M. Hsu, S. T. Connor, Y. Xu, Q. Wang,M. McGehee, S. Fan, and Y. Cui, Nano Letters 9, 279 (2008).
[72] E. Garnett and P. Yang, Nano letters 10, 1082 (2010).
[73] X. Li and P. Bohn, Applied Physics Letters 77, 2572 (2000).
[74] K. Peng, Y. Xu, Y. Wu, Y. Yan, S. Lee, and J. Zhu, Small 1, 1062 (2005).
[75] J. Oh, H.-C. Yuan, and H. M. Branz, Nature nanotechnology 7, 743 (2012).
[76] H. Seidel, L. Csepregi, A. Heuberger, and H. Baumgrtel, Journal of theElectrochemical Society 137, 3612 (1990).
[77] E. C. Garnett and P. Yang, Journal of the American Chemical Society 130, 9224(2008).
[78] X. Li, Current Opinion in Solid State and Materials Science 16, 71 (2012).
118
[79] E. C. Garnett, M. L. Brongersma, Y. Cui, and M. D. McGehee, Annual Review ofMaterials Research 41, 269 (2011).
[80] K. Peng, X. Wang, and S. T. Lee, Applied Physics Letters 92, 163103 (2008).
[81] Q. Shu, J. Wei, K. Wang, H. Zhu, Z. Li, Y. Jia, X. Gui, N. Guo, X. Li, C. Ma,and D. Wu, Nano Letters 9, 4338 (2009), doi: 10.1021/nl902581k; M3: doi:10.1021/nl902581k; 14.
[82] N. Guo, J. Wei, Q. Shu, Y. Jia, S. Song, Y. Xu, H. Wang, P. Li, H. Zhu, andK. Wang, Applied Physics A 107, 911 (2012).
[83] L. He, C. Jiang, H. Wang, and D. Lai, ACS Applied Materials Interfaces 4, 1704(2012).
[84] K. Peng, J. Hu, Y. Yan, Y. Wu, H. Fang, Y. Xu, S. T. Lee, and J. Zhu, AdvancedFunctional Materials 16, 387 (2005).
[85] Z. Huang, T. Shimizu, S. Senz, Z. Zhang, N. Geyer, and U. Gosele, The Journal ofPhysical Chemistry C 114, 10683 (2010).
[86] A. Cao, Physical Chemistry Chemical Physics (2012).
[87] P. Viktorovitch, G. Pananakakis, G. Kamarinos, and R. Basset, in Electron Devices
Meeting, 1977 International, Vol. 23 (1977) pp. 62–62, iD: 1.
[88] J. F. Gibbons, G. W. Cogan, C. M. Gronet, and N. S. Lewis, Applied PhysicsLetters 45, 1095 (1984).
[89] T. Saga, NPG Asia Materials , 96 (2010).
[90] H. Angermann, T. Dittrich, and H. Flietner, Applied Physics A: Materials ScienceProcessing 59, 193 (1994).
[91] A. G. Aberle, Solar Energy Materials and Solar Cells 65, 239 (2001).
[92] T. Lauinger, J. Schmidt, A. G. Aberle, and R. Hezel, Applied Physics Letters 68,1232 (1996).
[93] G. Dingemans and E. Kessels, Journal of Vacuum Science Technology A: Vacuum,Surfaces, and Films 30, 040802 (2012).
[94] R. L. Puurunen, Journal of Applied Physics 97, 121301 (2005).
[95] B. Dlubak, P. Kidambi, R. Weatherup, S. Hofmann, and J. Robertson, AppliedPhysics Letters 100, 173113 (2012).
[96] M. Groner, F. Fabreguette, J. Elam, and S. George, Chemistry of Materials 16, 639(2004).
119
[97] R. Har-Lavan, O. Yaffe, P. Joshi, R. Kazaz, H. Cohen, and D. Cahen, AIPAdvances 2, 012164 (2012).
[98] M. Y. Ali and M. Tao, Journal of Applied Physics 101, 103708 (2007).
[99] D. R. Lide and T. J. Bruno, CRC handbook of chemistry and physics (CRC PressILlc, 2012).
[100] H. Aibin, W. Wenwu, and X. Qiuxia, Journal of Semiconductors 30, 084001 (2009).
[101] J. G. Fossum, Electron Devices, IEEE Transactions on 24, 322 (1977).
[102] E. Shi, L. Zhang, Z. Li, P. Li, Y. Shang, Y. Jia, J. Wei, K. Wang, H. Zhu, andD. Wu, Scientific Reports 2 (2012).
[103] J. Wu, H. A. Becerril, Z. Bao, Z. Liu, Y. Chen, and P. Peumans, Applied PhysicsLetters 92, 263302 (2008).
[104] X. Wang, L. Zhi, and K. Mllen, Nano letters 8, 323 (2008).
[105] X. Li, H. Zhu, K. Wang, A. Cao, J. Wei, C. Li, Y. Jia, Z. Li, X. Li, and D. Wu,Advanced Materials 22, 2743 (2010).
[106] X. Miao, S. Tongay, M. K. Petterson, K. Berke, A. G. Rinzler, B. R. Appleton, andA. F. Hebard, Nano letters 12, 2745 (2012).
[107] S. M. Kim, Y. W. Jo, K. K. Kim, D. L. Duong, H.-J. Shin, J. H. Han, J.-Y. Choi,J. Kong, and Y. H. Lee, ACS nano 4, 6998 (2010).
[108] S. Tongay, K. Berke, M. Lemaitre, Z. Nasrollahi, D. Tanner, A. Hebard, andB. Appleton, Nanotechnology 22, 425701 (2011).
[109] S. Eglash, “Competition improves silicon-based solar cells,” (2009).
[110] M. A. Green and M. J. Keevers, Progress in Photovoltaics: Research andApplications 3, 189 (1995)
120
BIOGRAPHICAL SKETCH
Maureen Petterson was born in San Francisco, California in 1983. The youngest of
three children, she found support and encouragement to explore the universe through
her parents, who both believed in inquiry based learning and always forced her to
think instead of giving easy answers. She received a Bachelors of Science in Physics
(Astrophysics) from the University of California, Santa Cruz in 2006. Though her degree
focused on Astrophysics, a senior year thesis in Medical Physics working under Dr.
Hartmut Sadrozinski led to an interest in more hands on research. After graduation
she worked at the Santa Cruz Institute for Particle Physics for two years, an experience
that was invaluable for how it shaped her into an analytic and patient researcher and
ultimately cemented her decision to pursue an advanced degree in physics. After sending
in her graduate school applications, she left the United States and spent the next 6
months backpacking around Southeast Asia, the Middle East, and Eastern Europe with
Griffiths’ Quantum Mechanics in tow. She enrolled at the University of Florida in the
fall of 2008. Active within the physics community, she was the Graduate Student Council
representative for the physics department, served on the physics Graduate Student
Advisory Committee, and participated in numerous outreach events. In her second year
she started working with Dr. Andrew Rinzler on carbon nanotube based photovoltaics, a
field in which she could combine both gratifying research and exciting physical principles
with a desire to impact and contribute to society in a meaningful and positive way.
121