POLYCRYSTALLINE TANDEM PHOTOVOLTAICS

A DISSERTATION
SUBMITTED TO THE DEPARTMENT OF MATERIALS SCIENCE AND
ENGINEERING
AND THE COMMITTEE ON GRADUATE STUDIES
OF STANFORD UNIVERSITY
IN PARTIAL FULFILLMENT OF THE REQUIREMENTS
FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY

Colin David Bailie
December 2015

© 2015 by Colin David Bailie. All Rights Reserved.
Re-distributed by Stanford University under license with the author.

This work is licensed under a Creative Commons AttributionNoncommercial 3.0 United States License.
http://creativecommons.org/licenses/by-nc/3.0/us/

This dissertation is online at: http://purl.stanford.edu/kx438zm4205

ii

I certify that I have read this dissertation and that, in my opinion, it is fully adequate
in scope and quality as a dissertation for the degree of Doctor of Philosophy.
Michael McGehee, Primary Adviser

I certify that I have read this dissertation and that, in my opinion, it is fully adequate
in scope and quality as a dissertation for the degree of Doctor of Philosophy.
Yi Cui

I certify that I have read this dissertation and that, in my opinion, it is fully adequate
in scope and quality as a dissertation for the degree of Doctor of Philosophy.
Hemamala Karunadasa

Approved for the Stanford University Committee on Graduate Studies.
Patricia J. Gumport, Vice Provost for Graduate Education

This signature page was generated electronically upon submission of this dissertation in
electronic format. An original signed hard copy of the signature page is on file in
University Archives.

iii

Acknowledgments
I would like to thank everyone who helped me reach the finish line.
First, I thank my family, who have supported me throughout my life in all pursuits. My
mother, who has seen her efforts in helping with grade school science fair projects blossom into
science and engineering projects far beyond such world-changing efforts as ”Which paper airplane
design flies the furthest”. My father, who has always provided the counterpoint by playing sports
with me. My sister, who has always been my greatest champion.
Second, I thank my groupmates and collaborators without whom I would not have managed to publish anything of consequence. Science and engineering today is truly a team effort. I also
thank my groupmates for distracting me from being in the laboratory 24/7. Without them, I may
never have purchased (and used) a surfboard, climbing gear, trail running shoes, a squash racquet,
backpacking gear, and a road bicycle.
Finally, I thank my advisor Michael McGehee. His advice has proved invaluable on countless occasions. I am eternally grateful for the time he has dedicated to helping me grow as a scientist
and as a person and am glad for the friendship that has developed.

iv

Published Works
Dye-sensitized solar cell works
C. D. Bailie, W. H. Nguyen, J. Burschka, T. Moehl, M. Gr¨atzel, M. D. McGehee, and A. Sellinger, ”Molecular engineering of organic dyes for improved recombination lifetime in solid-state
dye-sensitized solar cells,” Chemistry of Materials, vol. 25, pp. 15191525, apr 2013.
C. D. Bailie, E. L. Unger, S. M. Zakeeruddin, M. Gr¨atzel, and M. D. McGehee, ”Melt-infiltration of
spiro-OMeTAD and thermal instability of solid-state dye-sensitized solar cells,” Physical Chemistry
Chemical Physics, vol. 16, pp. 4864, jan 2014.

Co-authored
T. P. Brennan, J. R. Bakke, I.-K Ding, B. E. Hardin, W. H. Nguyen, R. Mondal, C. D. Bailie,
G. Y. Margulis, E. T. Hoke, A. Sellinger, M. D. McGehee, and S. F. Bent, ”The importance of
dye chemistry and TiCl4 surface treatment in the behavior of Al2 O3 recombination barrier layers
deposited by atomic layer deposition in solid-state dye-sensitized solar cells,” Physical Chemistry
Chemical Physics, vol. 14, pp. 12130, jul 2012.
K. E. Roelofs, T. P. Brennan, J. C. Dominguez, C. D. Bailie, G. Y. Margulis, E. T Hoke, M. D.
McGehee, and S. F. Bent, ”Effect of Al2 O3 recombination barrier layers deposited by atomic layer
deposition in solid-state CdS quantum dot-sensitized solar cells,” Journal of Physical Chemistry C,
vol. 117, pp. 5584-5592, feb 2013.
G. Y. Margulis, M. G. Christoforo, D. Lam, Z. M. Beiley, A. R. Bowring, C. D. Bailie, A. Salleo,
and M. D. McGehee, ”Spray Deposition of Silver Nanowire Electrodes for Semitransparent SolidState Dye-Sensitized Solar Cells,” Advanced Energy Materials, vol. 3, pp. 1657-1663, dec 2013.

v

W. H. Nguyen, C. D. Bailie, E. L. Unger, and M. D. McGehee, ”Enhancing the Hole-Conductivity
of Spiro-OMeTAD without Oxygen or Lithium Salts by Using Spiro(TFSI)2 in Perovskite and DyeSensitized Solar Cells,” Journal of the American Chemical Society, vol. 136, pp. 10996-11001, aug
2014.

Perovskite solar cell and tandem works
C. D. Bailie, M. G. Christoforo, J. P. Mailoa, A. R. Bowring, E. L. Unger, W. H. Nguyen, J.
Burschka, N. Pellet, J. Z. Lee, M. Gr¨atzel, R. Noufi, T. Buonassisi, A. Salleo, and M. D. McGehee,
”Semi-transparent Perovskite Solar Cells for Tandems with Silicon and CIGS,” Energy & Environmental Science, vol. 8, pp. 956-963, dec 2015.
C. D. Bailie, J. P. Mailoa, E. C. Johlin, E. T. Hoke, A. J. Akey, W. H.Nguyen, M. D. McGehee, and
T. Buonassisi, ”A 2-terminal perovskite/silicon multijunction solar cell enabled by a silicon tunnel
junction,” Applied Physics Letters, vol. 106, p. 121105, 2015.
C. D. Bailie, and M. D. McGehee, ”High-efficiency tandem perovskite solar cells,” MRS Bulletin,
vol. 40, pp. 681-686, aug 2015.

Co-authored
E. L. Unger, E. T. Hoke, C. D. Bailie, W. H. Nguyen, A. R. Bowring, T. Heum¨
uller, M. G. Christoforo, and M. D. McGehee, ”Hysteresis and transient behavior in current-voltage measurements of
hybrid-perovskite absorber solar cells,” Energy Environ. Sci., vol. 7, pp. 3690-3698, aug 2014.

vi

Contents
Acknowledgments

v

Published Works

vi

1 Introduction

1

1.1

State of the Photovoltaic Industry . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1

1.2

Physics of Solar Cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

3

1.2.1

Tandem Solar Cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

5

1.3

Tandem Architectures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

7

1.4

Perovskite Solar Cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

8

2 Monolithic Perovskite/Silicon Tandems

15

2.1

Published Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

15

2.2

Limitations of Monolithic Tandems . . . . . . . . . . . . . . . . . . . . . . . . . . . .

22

3 Mechanically-Stacked Tandems using Silver Nanowires
3.1

3.2

Published Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

28

3.1.1

Semi-Transparent Cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

28

3.1.2

Mechanically-Stacked Tandems . . . . . . . . . . . . . . . . . . . . . . . . . .

31

Stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

34

4 Mechanically-Stacked Tandems using ITO
4.1

4.2

26

36

Published Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

37

4.1.1

Deposition Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

37

4.1.2

Semi-Transparent Cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

38

4.1.3

Tandems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

39

4.1.4

Stability Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

40

Future of Mechanically-Stacked Tandems . . . . . . . . . . . . . . . . . . . . . . . .

42

vii

5 Cost-Modeling of Perovskite Solar Cells
5.1

44

Published Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

45

5.1.1

Model Assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

45

5.1.2

Single-Junction Perovskite Cost Model . . . . . . . . . . . . . . . . . . . . . .

47

5.1.3

Mechanically-Stacked Tandem Perovskite Cost Model . . . . . . . . . . . . .

49

5.1.4

LCOE Comparison Across Technologies . . . . . . . . . . . . . . . . . . . . .

51

6 Conclusions

53

7 Appendices

55

7.1

7.2

7.3

7.4

7.5

Monolithic Tandem Experimental Information . . . . . . . . . . . . . . . . . . . . . .

55

7.1.1

Silicon Sub-cell Fabrication Procedure . . . . . . . . . . . . . . . . . . . . . .

55

7.1.2

Perovskite Sub-Cell Fabrication Procedure . . . . . . . . . . . . . . . . . . . .

56

7.1.3

Multijunction Cell Testing Protocols . . . . . . . . . . . . . . . . . . . . . . .

58

7.1.4

Supplemental Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

59

Silver Nanowire Deposition Methods . . . . . . . . . . . . . . . . . . . . . . . . . . .

61

7.2.1

Lamination By Point Force . . . . . . . . . . . . . . . . . . . . . . . . . . . .

61

7.2.2

Lamination By Distributed Force . . . . . . . . . . . . . . . . . . . . . . . . .

62

Mechanically-Stacked Tandems Using Silver Nanowires Experimental Information . .

65

7.3.1

Perovskite Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

65

7.3.2

Silicon Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

67

7.3.3

Measurement Procedures . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

68

7.3.4

Supplemental Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

69

Mechanically-Stacked Tandems Using ITO Experimental Information . . . . . . . . .

71

7.4.1

Perovskite Device Fabrication Method . . . . . . . . . . . . . . . . . . . . . .

71

7.4.2

J-V Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

71

7.4.3

EQE Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

72

7.4.4

Stability Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

72

7.4.5

Supplementary Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

72

Cost-Modeling of Perovskite Solar Cells Model Information . . . . . . . . . . . . . .

76

7.5.1

Process flow for single-junction standard architecture . . . . . . . . . . . . . .

76

7.5.2

Tandem Performance Modeling . . . . . . . . . . . . . . . . . . . . . . . . . .

77

Bibliography

78

viii

List of Tables
1.1

LCOE of comparable electricity technologies. Data acquired from [1]. . . . . . . . . .

2.1

Photovoltaic parameters from 2-terminal perovskite/Si multijunction cell (hero device). 19

3.1

Performance metrics of semi-transparent and opaque perovskite devices. . . . . . . .

29

3.2

Performance metrics of mechanically-stacked tandems . . . . . . . . . . . . . . . . .

32

4.1

Photovoltaic parameters of a semi-transparent perovskite solar cell with ITO rear
electrode compared to an opaque perovskite solar cell with Al/Ag rear electrode. . .

4.2

39

Photovoltaic parameters of semi-transparent perovskite and mono-crystalline silicon
cells and the resulting tandem efficiency. . . . . . . . . . . . . . . . . . . . . . . . . .

4.3

3

39

Efficiency as a function of temperature extracted by averaging the data in Figure
4.4 over the entire time period. The temperature coefficient is extracted as a linear
fit of the data and normalized against the expected efficiency at 25 ◦ C. . . . . . . . .

ix

42

List of Figures
1.1

2015 utility scale benchmark system price from [2]. Module price is 0.65 $/W, balance
of systems price is 1.12 $/W for a total system price of 1.77 $/W. . . . . . . . . . . .

2

1.2

Average commercial efficiency over time. Figure from [3]. . . . . . . . . . . . . . . .

2

1.3

Absorption process in a semiconductor. Photons of energy greater or equal to the
bandgap energy are absorbed and create electron-hole pairs that thermalize down to
the bandgap energy. Photons of energy less than the bandgap energy are not absorbed
by the semiconductor. Figure adapted from Stanford University MatSci 302 lecture
slides taught by Prof. Michael McGehee. . . . . . . . . . . . . . . . . . . . . . . . . .

1.4

The terrestrial solar spectrum, AM1.5G, spans ultraviolet (UV), visible (VIS), and
infrared (NIR) light. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1.5

4
4

Theoretical efficiency of single-junction solar cells. Input spectrum is 6000 K blackbody. AM1.5G spectrum yields slightly different values with a theoretical maximum
of 33.7 %. Current silicon record from NREL record photovoltaics chart [4]. . . . . .

1.6

5

Essential function of single-junction vs. double-junction solar cells. Double-junction
(tandem) solar cells split the solar spectrum between two absorbers, resulting in a
higher potential efficiency. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1.7

6

Theoretical efficiency of double-junction solar cells. Input spectrum is 6000 K blackbody. AM1.5G spectrum yields slightly different values with a theoretical maximum
of 46.1 %. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1.8

6

Scalable tandem architectures. (Left) A mechanically-stacked tandem combines two
separately-fabricated solar cells. (Right) A monolithic tandem integrates two junctions into a single solar cell. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

x

8

1.9

Perovskite is the name of the mineral CaTiO3 but this term is being used for all
compounds with the general formula ABX3 that have the same crystal structure as
CaTiO3 or are derived from this structure. These materials consist of two cations,
the cation A is 12 fold coordinated by the anions X and the cation B 6-fold where
X can either be oxygen or a halide. The most prominent perovskite in solar cells
right now is methylammonium lead iodide. It is worth noting that the structure of
methylammonium lead iodide deviates from the ideal cubic perovskite structure as
the octrahedra become slightly tilted and the structure is consequently tetragonal. .

10

1.10 Perovskite solar cell architectures. (Left) an n-i-p architecture using a mesoporous
TiO2 scaffold. (Middle) an n-i-p architecture in a planar configuration. (Right) a
p-i-n architecture in a planar configuration

. . . . . . . . . . . . . . . . . . . . . . .

11

1.11 J-V curve of a perovskite solar cell. The relevant points of JSC , VOC , and PM AX are
shown. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

12

1.12 External quantum efficiency of a perovskite solar cell as a function of wavelength.
Unity represents perfect quantum efficiency. . . . . . . . . . . . . . . . . . . . . . . .

13

1.13 Schematic examples depicting the processes of transmission, reflection, and absorption
of light. Image from www.chroma.com. . . . . . . . . . . . . . . . . . . . . . . . . . .
2.1

14

(a) The device structure of a 2-terminal monolithically grown perovskite/Si multijunction solar cell with an n-type Si base. The polished SEM image is taken at 45◦
tilt to show the Ag nanowire mesh (500 nm scale bar). (b) Band diagram of the
perovskite/silicon cell interface showing the charge-transport mechanism around the
Si tunnel junction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2.2

16

(a) TEM image of the n++/p++ silicon tunnel junction interface after the dopant
activation annealing (left: 30 nm scale bar) and high-resolution TEM image of the
n++ layer, showing the partially crystalline nature of this layer (right: 5 nm scale
bar). (b) SIMS profile of the Si emitter and tunnel junction layer showing the sharp
doping profile at the tunnel junction interface. (c) Comparison of J-V profile for
identical Si cells with and without a tunnel junction, showing negligible effect of the
tunnel junction to the single-junction Si cell performance. . . . . . . . . . . . . . . .

xi

17

2.3

(a) J-V curve of the 2-terminal perovskite/silicon multijunction solar cell under AM1.5G
illumination. Forward and reverse-bias scan directions are shown with 5 s measurement delay per data point. Steady-state values for JSC , VOC , and MPP are depicted
as blue circles and are measured by averaging over 30 s after reaching steady state.
The VOC of 1.58 V is approximately the sum of the perovskite and Si cell VOC . (b)
Time-dependent output current of the multijunction cell near maximum power point
(1.20 V forward bias) showing that the output reaches steady state after a measurement delay of 30 s. (c) Total device reflection and EQE of the perovskite and Si
sub-cells of a typical perovskite/Si multijunction cell. The EQE spectra exhibit a low
blue EQE in the top perovskite sub-cell due to spiro-OMeTAD parasitic absorption,
and low red EQE in the bottom Si sub-cell due to both spiro-OMeTAD parasitic
absorption and the lack of good Si back surface passivation scheme. The perovskite
sub-cell EQE is corrected to match the measured JSC and the silicon sub-cell EQE is
reported as measured . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2.4

20

Perovskite is solution-processed onto a textured silicon wafer. The nominal thickness
of the perovskite layer is 250-500 nm while the thickness of silicon pyramids is 5-12
µm. More perovskite is deposited in the valleys between the pyramids while the peaks
are left completely bare of perovskite . . . . . . . . . . . . . . . . . . . . . . . . . . .

2.5

23

Result of a PC1D simulation. Blue curve represents the EQE of a silicon cell with
full texturing and an ideal double layer anti-reflection coating. Black curve represents
the EQE of a silicon cell without front surface texturing and the silicon surface in
direct contact with air. Red curve is the EQE of a silicon solar cell with a planar
front surface with a thin film layer stack with thicknesses and real refractive indices
similar to a perovskite solar cell showing an intermediate performance between the
two extremes. Numbers in the legend refer to the refractive index and thickness of
the thin film layers on top of the silicon. . . . . . . . . . . . . . . . . . . . . . . . . .

2.6

24

Representative spectra of the sun’s illumination of Denver, CO over a year. Figure
reproduced from [5]. In this plot, all light to the left of ∼700 nm is absorbed by the
perovskite and all light to the right of ∼700 nm is absorbed by the silicon (assuming
a modified perovskite bandgap). Note that for many of these spectra, a currentmismatch will occur between the perovskite and silicon. . . . . . . . . . . . . . . . .

xii

25

3.1

Mechanically-stacked tandem wiring configurations that result in two terminals exiting the module. (Left) The perovskite and silicon cells are wired in series to enforce
a current-matching condition. The total current in each cell can be matched by adjusting the size of the perovskite solar cell (the perovskite and silicon do not need to
be current-density matched). (Right) The perovskite cells are wired in series and the
silicon cells are separately wired in series. These two strings are wired together in
parallel to enforce a voltage-matching condition. A third alternative is to use a power
electronic circuit in either configuration to correct for any mismatches in current or
voltage between the strings of cells. . . . . . . . . . . . . . . . . . . . . . . . . . . . .

3.2

27

Perovskite device results. a) Current-voltage curves comparing best opaque vs. semitransparent perovskite devices. b) EQE of semi-transparent device and opaque device.
Note that the opaque device does not have AR coatings. c) Transmission through
semi-transparent perovskite with AR coatings. Peak transmission is 77 % around 800
nm.

3.3

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

30

Perovskite and CIGS/Si tandem results. a) Current-voltage and b) EQE of semitransparent perovskite cell, unfiltered CIGS cell, and CIGS cell filtered by the perovskite cell. c) IV curves and d) EQE of semi-transparent perovskite cell, unfiltered
TI-Si cell, and TI-Si cell with an infrared-optimized anti-reflection coating filtered by
the perovskite cell. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

3.4

NREL certification of a mechanically-stacked tandem. The tandem efficiency is the
sum of the individually measured cells, 17.9%. . . . . . . . . . . . . . . . . . . . . . .

4.1

32
33

Semi-transparent inverted perovskite device architecture. a) Energy level diagram.
b/c) Cross-sectional SEM and illustrative schematic of device architecture showing
ITO electrode encapsulation layer. . . . . . . . . . . . . . . . . . . . . . . . . . . . .

4.2

Performance comparison of the opaque and semi-transparent devices showing a) J-V

4.3

Mechanically-stacked perovskite/silicon tandem performance. a) J-V curves of tan-

curve of devices showing comparable FF and VOC and b) max power point tracking.

37
38

dem with the max power of the tandem calculated from the addition of the perovskite
and silicon cells. b) EQE of original mono-Si, perovskite, and filtered silicon solar cells. 40
4.4

Thermal stability of ITO-capped perovskite solar cells at 60 and 100 ◦ C compared to
opaque device with ZnO and Al/Ag. . . . . . . . . . . . . . . . . . . . . . . . . . . .

4.5

41

J-V, EQE, and Transmission of a possible semi-transparent perovskite solar cell based
on the best-in-class 20.1 % opaque solar cell. Semi-transparent perovskite cell efficiency is 17.5 % . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

4.6

43

J-V and EQE of the 25.6 % record silicon solar cell if filtered by the transmission
curve in Figure 4.5. Filtered silicon efficiency is 8.6 % . . . . . . . . . . . . . . . .

xiii

43

5.1

Single-junction perovskite module scaled process flow. . . . . . . . . . . . . . . . . .

46

5.2

Mechanically-stacked tandem module scaled process flow. . . . . . . . . . . . . . . .

47

5.3

Single-junction perovskite module step costs. . . . . . . . . . . . . . . . . . . . . . .

48

5.4

Single-junction perovskite compared to CdTe. . . . . . . . . . . . . . . . . . . . . . .

49

5.5

Mechanically-stacked tandem module step costs. . . . . . . . . . . . . . . . . . . . .

50

5.6

LCOE calculations for different perovskite scenarios assuming an operational lifetime
of 30 years, a 140 $/m2 BoS cost, and a 7 % internal rate of return. . . . . . . . . .

7.1

52

EQE curve of single-junction n-type silicon cells with and without a tunnel junction.
This curve shows negligible parasitic absorption in the tunnel junction for wavelength
λ > 500 nm. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

7.2

59

Transient (a) JSC and (b) VOC of the perovskite/silicon tandem cell as they stabilize
over the time. The measured steady-state values are JSC = 11.5 mA/cm2 and VOC
= 1.58 V, respectively. The 30 s settling time for the VOC is not shown as the cell
was stabilized at VOC for > 30 s prior to starting the measurement. . . . . . . . . .

7.3

59

EQE of a semi-transparent perovskite solar cell illuminated through either the n-side
(glass/TiO2 side) or through the p-side (AgNW/spiro-OMeTAD side). The glass
side EQE integrates to 17.3 mA/cm2 while the AgNW side EQE integrates to 11.4
mA/cm2 . Neither side has anti-reflection coatings. . . . . . . . . . . . . . . . . . . .

7.4

60

The light transmission through a 470-nm-thick doped spiro-OMeTAD film on glass.
The contribution of the glass is removed from this plot. The absorption features from
300-400 and 450-550 nm are readily visible in the EQE plot above. . . . . . . . . . .

xiv

60

7.5

Process for vacuum lamination of silver nanowires onto a solar cell. 1) The baseplate
of the transfer chamber designed to withstand the pressures of the process. At the
bottom of this baseplate is a connection to a vacuum source used to remove all air
pockets that can disrupt the uniformity of the applied pressure. 2) A porous metal
(aluminum) baseplate is used to distribute the vacuum evenly across the surface of the
porous plate to avoid plugging the vacuum source and prevent removal of air pockets.
3) An O-ring is used to provide an edge seal between the top and bottom halves of
the chamber to ensure proper pressure buildup. 4) The solar cell is placed in the
middle of the porous metal plate. 5) The silver nanowires on a plastic film are placed
face-down onto the solar cell. 6) A tape seal is placed in-between the O-ring and the
porous metal plate to allow the application of a diaphragm. 7) A plastic diaphragm
is attached to the tape and provides a barrier between the two halves of the transfer
chamber. Vacuum is pulled on the lower half and the diaphragm deforms around
the solar cell, providing a flat bubble-free surface onto which a uniform pressure can
be applied. 8) The top half of the transfer chamber is placed on top and bolted to
the bottom half through radial bolt holes in both chambers. The o-ring provides a
seal between the top half and ambient. A fitting on the side of the top half connects
to a positive pressure source such as a compressed gas canister. Positive pressure is
applied to the diaphragm, uniformly pressing the nanowires into the solar cell. . . .

7.6

64

Optical density over time of perovskite film in dipping solution. The signal at 700 nm
was used to determine the growth rate of the perovskite, while the signal at 850 nm
was used to detect the presence of other optical phenomena (changes in scattering,
reflection, incident light intensity). . . . . . . . . . . . . . . . . . . . . . . . . . . . .

66

7.7

Transmission of the 12.4 Ω/ AgNW film with the PET substrate subtracted. . . .

69

7.8

Distribution of semi-transparent cell performance gathered from 3 batches where device procedures were largely kept constant between batches. Low-efficiency devices
exhibited shorting likely caused by too much pressure applied manually during the
AgNW electrode transfer. Medium-efficiency devices generally exhibited low photocurrent likely caused by incomplete transfer of the AgNW electrode due to too
little pressure applied manually during the electrode transfer. . . . . . . . . . . . . .

7.9

70

Example of difference in spectral response depending on AR coating applied to Si cell.
The regular TI-Si cell has a broadband AR coating. . . . . . . . . . . . . . . . . . .

70

7.10 J-V Curves showing problems with the ZnO/ITO interface. At room temperature,
a large extraction barrier is present which greatly limits fill factor. This barrier is
alleviated as temperature is raised; however the barrier reappears upon cooling. . . .

73

7.11 A 50 nm film of AZO nanoparticles displays high transmission. The AZO nanoparticles are annealed at either 75 or 150 ◦ C after deposition. . . . . . . . . . . . . . . . .

xv

73

7.12 Transmission through an ITO/Glass reference sample from the same sputter deposition batch as the best devices. The 500 nm layer of ITO was annealed at 100 ◦ C for
15 minutes to match the thermal processing profile of the perovskite samples and the
resistivity dropped from 11 to 10 Ω/. . . . . . . . . . . . . . . . . . . . . . . . . . .

74

7.13 Comparison of the EQE of semi-transparent and opaque perovskite solar cells. The
lack of a metal back reflector in the semi-transparent solar cells results in lower photocurrent. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

74

7.14 Transmission through the semi-transparent perovskite solar cell. Transmission is limited at longer wavelengths due to coherent reflections. . . . . . . . . . . . . . . . . .

75

7.15 Reflection of semi-transparent and opaque perovskite solar cells. Mitigating reflection
is an area of future work. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

75

7.16 Modeled EQE of perovskite and HIT-Si cells. . . . . . . . . . . . . . . . . . . . . . .

77

xvi

Chapter 1

Introduction
This thesis details the initial efforts on perovskite/Silicon and perovskite/CIGS tandem solar cells
by Colin Bailie and his advisor Prof. Michael McGehee at Stanford University in collaboration with
numerous academic and industrial partners.

1.1

State of the Photovoltaic Industry
The history of the photovoltaic industry can be divided generally into four generations.

In the first generation, silicon solar cells were primarily used for space applications (1955-1975). [6]
Then from 1975-1985, the initial large purchases of terrestrial solar cells placed the emphasis on
designing modules that could be deployed on land and reaching today’s warranty specifications. [6]
The third generation of the industry ended just 3-4 years ago, where prior to 2012 the total system
cost for utility scale installations was dominated by the cost of the module itself and primary focus
was on reducing module costs. [2] Enter the fourth generation, where the module cost represents a
decreasing portion of the total system cost (Figure 1.1). The remainder of the system cost, termed
the balance of systems (BoS), consists of inverters, labor, land, racking, and developmental costs
among others.

1

CHAPTER 1. INTRODUCTION

2

Figure 1.1: 2015 utility scale benchmark system price from [2]. Module price is 0.65 $/W, balance
of systems price is 1.12 $/W for a total system price of 1.77 $/W.
Even with the recent precipitous cost reductions in silicon modules [7], electricity from
solar power is not yet on par with comparable electricity technologies (Table 1.1). [1] With the
module no longer the driver of the total system cost, reducing the BoS cost is increasingly important.
Efforts in this area are multi-faceted including streamlining labor and installation, developing novel
racking systems, and improving inverters among others. A large opportunity also exists for the
module to greatly affect the overall system price as well. Because this system price is given in $/W,
increasing the performance of the panel (i.e power output) allows the module to affect the BoS cost
indirectly. The recent history of the average commercial module efficiency (Figure 1.2) shows a
slowly increasing efficiency with time with the average efficiency in 2015 ∼16 %.

Figure 1.2: Average commercial efficiency over time. Figure from [3].

CHAPTER 1. INTRODUCTION

3

Table 1.1: LCOE of comparable electricity technologies. Data acquired from [1].
Technology

LCOE (¢/kWh)

Coal
Natural Gas
Nuclear
Wind (Offshore)
Solar

9.5-14
7.5-10
10
7.4 (19.7)
12.5

In order to reach costs competitive with other electricity technologies by improving the
efficiency of the panel, the efficiency must be dramatically improved from the typical 16 % efficiency
today to 25-30 %. In this thesis, I will outline one potential path for reaching these targets of
efficiency and commercial viability.

1.2

Physics of Solar Cells

The goal of a solar cell has been articulated most simply by Prof. Dick Swanson.
• First, a solar cell should generate as many electron-hole pairs as possible from the incident
sunlight (i.e. absorb as much of the solar spectrum as possible and turn it into electrical
current).
• Second, coax as many electrons and holes as possible to go to the correct electrode (i.e extract
as much of the current as possible).
• Finally, do this at as high a voltage as possible (as high a difference in electromotive force
between the leads as possible)
A property in all semiconducting materials called a bandgap governs both the current and
voltage potential of a solar cell. The bandgap is defined by the top of the valence band (or ionization
potential, Φ) to the bottom of the conduction band (or electron affinity). Only photons with an
energy greater than or equal to the bandgap energy can be absorbed by the semiconductor (Figure
1.3). Because the solar spectrum consists of photons of a wide range of energy, mostly situated
between 1 and 4 electron-volts (eV) (Figure 1.4), the solar cell can generate more current (goal
#1) by decreasing the bandgap of the absorbing semiconductor. However, the bandgap also restricts
the voltage that the current can be extracted at. In practice it is impossible to extract the current
at a voltage higher than the bandgap, meaning that the bandgap should be as large as possible to
maximize goal #3. An optimization occurs because power (or efficiency) is the product of current
and voltage.

CHAPTER 1. INTRODUCTION

4

Figure 1.3: Absorption process in a semiconductor. Photons of energy greater or equal to the
bandgap energy are absorbed and create electron-hole pairs that thermalize down to the bandgap
energy. Photons of energy less than the bandgap energy are not absorbed by the semiconductor.
Figure adapted from Stanford University MatSci 302 lecture slides taught by Prof. Michael McGehee.

Figure 1.4: The terrestrial solar spectrum, AM1.5G, spans ultraviolet (UV), visible (VIS), and
infrared (NIR) light.

CHAPTER 1. INTRODUCTION

5

Single-junction solar cells use one absorbing semiconductor material. Every unconcentrated
commercial module uses single-junction solar cells. Because of the wide energy range of the solar
spectrum, these types of solar cells are theoretically limited to 33.7 % efficiency (Figure 1.5).
[8, 9] Any photon energy in excess of the bandgap energy is lost almost immediately as waste heat
in a process called thermalization (Figure 1.4). The practical efficiency limit for silicon, which
dominates the commercial market, is ∼25 % (depicted as a star in Figure 1.5). One way to push
beyond the limits of silicon solar cells is to use tandem solar cells.

Figure 1.5: Theoretical efficiency of single-junction solar cells. Input spectrum is 6000 K blackbody.
AM1.5G spectrum yields slightly different values with a theoretical maximum of 33.7 %. Current
silicon record from NREL record photovoltaics chart [4].

1.2.1

Tandem Solar Cells
Tandem (or multijunction) solar cells use multiple absorbing semiconductors to harvest the

solar spectrum. Using multiple absorbers allows each absorber to specialize in a portion of the solar
spectrum, as illustrated in Figure 1.6 instead of a single absorber that is responsible for absorbing
the entire solar spectrum and therefore has a fairly small bandgap and extracts current at a low
voltage. In a tandem, a large-bandgap solar cell first absorbs the high-energy portion of the solar
spectrum and extracts that current at a high voltage. This large-bandgap solar cell is transparent
to the low-energy portion of the solar spectrum which is absorbed by a small-bandgap solar cell
below that extracts the current at a lower voltage. Using two absorbers in this fashion pushes the
theoretical efficiency of 46.1 % with the correct choice of bandgaps (Figure 1.7).

CHAPTER 1. INTRODUCTION

6

Figure 1.6: Essential function of single-junction vs. double-junction solar cells. Double-junction
(tandem) solar cells split the solar spectrum between two absorbers, resulting in a higher potential
efficiency.

Figure 1.7: Theoretical efficiency of double-junction solar cells. Input spectrum is 6000 K blackbody. AM1.5G spectrum yields slightly different values with a theoretical maximum of 46.1 %.

CHAPTER 1. INTRODUCTION

1.3

7

Tandem Architectures
There are two scalable methods of making tandem solar cells, each with their own advan-

tages and disadvantages (Figure 1.8). A mechanically-stacked tandem combines two separatelyfabricated solar cells by simply putting them on top of each other. From a high-level practical view,
these types of solar cells are easy to make since both solar cells can be independently optimized for
efficiency then combined at the end to form a tandem cell, which accelerates prototyping. Because
the cells are independent of each other, there are few restrictions on the design (e.g. no tunnel
junction or recombination layer required, no current matching condition). Monolithic tandems are
the classic structure employed in most tandem designs from III-V tandems to organic tandems. In
a monolithic tandem, junctions are deposited on top of one another in order, generating a single
solar cell with multiple junctions contained within. To make this architecture function, a tunnel
junction or recombination layer is deposited between each junction to allow the flow of current
between junctions. Compared with mechanically-stacked tandems, which have three transparent
electrodes, monolithic tandems require only one transparent electrode. Because transparent electrodes are not perfectly transparent, this leads to a higher practical efficiency potential for monolithic
tandems than mechanically-stacked tandems. Monolithic tandems have drawbacks as well, requiring
a current-matching restriction on the system (the same amount of current must pass through both
junctions) as well as requiring more manufacturing steps to be processed serially, which affects yield
and makes prototyping more difficult.
A tandem solar cell works best with an ideal division of the solar spectrum between the
two absorbers (Figure 1.7). For the bottom absorber in the tandem, a commercially available
technology such as silicon (Si) or copper indium gallium diselenide (CIGS) provides a number of
advantages for the tandem. Both of these technologies have a bandgap around 1.1 eV, which is
smaller than the desired bandgap for a single-junction solar cell, around 1.3-1.4 eV. The fact that
these absorbers are not ideal in single-junction solar cells is what makes them perfect absorbers
for tandems. For the standard AM1.5G solar spectrum, the ideal distribution is to have a 1.1 eV
absorber and a 1.7-1.8 eV absorber [9]. Small-bandgap solar cells (with bandgaps <1.5 eV) are
much more developed than large-bandgap solar cells (bandgap >1.5 eV). Choosing a commercial
bottom-cell technology allows for a simplification of the problem and focus on the top cell alone.

CHAPTER 1. INTRODUCTION

8

Figure 1.8: Scalable tandem architectures. (Left) A mechanically-stacked tandem combines two
separately-fabricated solar cells. (Right) A monolithic tandem integrates two junctions into a single
solar cell.

1.4

Perovskite Solar Cells
The top cell technology used in this thesis is based on the metal-halide perovskite absorber.

First discovered in 1978 [10], this absorber was originally developed for solar cells as a sensitizer in
liquid dye-sensitized solar cells without much acclaim due to the tendency for the perovskite material
to dissolve in the liquid. [11] In 2012, researchers at Oxford University in the UK and at EPFL in
Switzerland developed this material for use in solid-state dye-sensitized solar cells. [12, 13]
The use of these materials in a solid state architecture shocked the academic community
of photovoltaics researchers. First, these cells were 10-12 % efficient, far beyond the previous solidstate dye-sensitized solar cell records (7.2-8.5 %) [14, 15]. Second, the devices exhibited extremely
high voltages (0.9-1.1 V) uncommon for a material deposited onto a high surface area scaffold by
solution-processing (and therefore amorphous, semi-crystalline, or polycrystalline in nature). Third,
the bandgap of the initial metal-halide perovskite composition used was around 1.6 eV, making
it a new large-bandgap solar technology (which is a category that lacks many high-performance
options) with a bandgap nearly ideal for the top cell in a tandem combination on top of either
Si or CIGS. This initial metal-halide composition was called ”methylammonium lead iodide” with

CHAPTER 1. INTRODUCTION

9

the chemical formula CH3 NH3 PbI3 (sometimes written as MAPbI3 with ”MA” as a shorthand for
”methylammonium”).
A perovskite describes a crystal structure (discovered first for the material calcium titanate
- CaTiO3 - and named after mineralogist Lev Perovski) that has the basic stoichiometry ABX3 . A is
a cation, which in this case is methylammonium. B is a metal, in this case lead. X is an anion, which
in this case is the halide element iodine. This metal-halide perovskite is formed by the combination
of two salts, PbI2 and MAI.
Discovery of this material set off a flurry of research activity in the area. It was discovered
that this material could be deposited an a vast variety of methods from single-step deposition methods from both stoichiometric [13] and non-stoichiometric solutions [15], two-step solution deposition
methods where the lead salt was deposited first from solution then the organic salt was introduced
by either dipping [16] or spinning [17], two-step mixed deposition methods where the lead salt was
deposited first from solution then the organic salt was introduced by vapor phase [18], evaporation
methods where both salts were deposited by vapor phase [19], and later, conversion methods where
formation of the full perovskite required a solvent rinse to activate the crystallization [20]. It was
also (re)discovered that the composition of the perovskite could change its optoelectronic properties. Specifically, while the MAPbI3 perovskite had a bandgap of 1.6 eV, substituting the iodine for
bromine (another halide element) allowed the bandgap to be continuously tuned up to 2.3 eV (the
bandgap of MAPbBr3 ). [21] In addition, switching the A site material could affect the bandgap.
Substituting the methylammonium for a slightly larger molecule formamidinium (chemical formula
HC(NH2 )2 ) shifts the tunable bandgap range by bromine substitution to 1.5-2.2 eV. [22] Substituting the methylammonium for the inorganic element cesium shifts the bandgap in the other direction,
to 1.7 eV for CsPbI3 [23] and 2.4 eV for CsPbBr3 [24]. The possibility exists for the metal site to
be replaced with elements other than lead (MASnI3 has a 1.2 eV bandgap for example) [25], though
these materials have severe instability to air exposure. So far these efforts have culminated in the
optimization of the perovskite cell to 20.1 % efficiency. [26]

CHAPTER 1. INTRODUCTION

10

Figure 1.9: Perovskite is the name of the mineral CaTiO3 but this term is being used for all
compounds with the general formula ABX3 that have the same crystal structure as CaTiO3 or
are derived from this structure. These materials consist of two cations, the cation A is 12 fold
coordinated by the anions X and the cation B 6-fold where X can either be oxygen or a halide.
The most prominent perovskite in solar cells right now is methylammonium lead iodide. It is worth
noting that the structure of methylammonium lead iodide deviates from the ideal cubic perovskite
structure as the octrahedra become slightly tilted and the structure is consequently tetragonal.
The perovskite can be considered an intrinsic material as deposited and no experimental
evidence exists that the perovskite can be controllably doped with either carrier type, making p-n
homojunctions and heterojunctions unavailable architectures at this time. Instead, the perovskite
solar cell consists of the perovskite material sandwiched between two heterojunctions, one p-type and
one n-type (Figure 1.10). This sandwich configuration can be either an n-i-p architecture where the
n-type heterojunction is deposited before the perovskite and the p-type heterojunction is deposited
after the perovskite or a p-i-n architecture where the p-type heterojunction is deposited before and
the n-type heterojunction is deposited after. The n-i-p architecture was the first to be developed
(a direct transition from dye-sensitized solar cells) in both mesoporous configurations (where the
n-type heterojunction forms a porous layer that the perovskite fills in) and planar configurations
(where each layer is solid and has a smooth surface). The mesoporous n-i-p configuration generally
uses titanium dioxide (TiO2 ) nanoparticles to form the porous layer [13]. This is a holdover from
dye-sensitized solar cells, yet the record 20.1 % efficient perovskite cell still uses this mesoporous
architecture. [26] The planar n-i-p structure has been demonstrated using a number of different ntype materials, including TiO2 , ZnO, SnO2 , and C60 . The list of p-type heterojunction layers is too
vast to list fully - the most common materials being the organic small molecule spiro-OMeTAD and
the polymer PTAA. The p-i-n architecture is a more recent development, using either a polymeric
layer (e.g. PEDOT:PSS or PTAA) or an inorganic layer (e.g. NiO) as the p-type layer upon which

CHAPTER 1. INTRODUCTION

11

the perovskite is deposited then the n-type layer from the n-type materials list above.

Figure 1.10: Perovskite solar cell architectures. (Left) an n-i-p architecture using a mesoporous
TiO2 scaffold. (Middle) an n-i-p architecture in a planar configuration. (Right) a p-i-n architecture
in a planar configuration
For the purposes of the remainder of this thesis, solar cells can be fully described by three
main experiments: Solar simulator I-V or J-V curves, external quantum efficiency, and basic optical
measurements of reflection, absorption, and transmission.
A solar simulator is a xenon arc lamp with filters between the lamp and the solar cell to help
the lamp closely resemble the sun within a small degree of error. The sun, by the AM1.5G standard,
illuminates the earth with an intensity of 100 mW/cm2 (or 1000 W/m2 ). The solar cell is attached
to a load (resistance) in series to complete the solar cell’s circuit. By changing this resistance, the
voltage drop over the solar cell can be modified. At different loads, the voltage drop across the
cell can be measured and the current running through the system can be measured, generating an
I-V curve. There are three important points on an I-V curve. At zero resistance, the solar cell is
in a short circuit condition and the current measured in this condition is called the ”short-circuit
current” (ISC ). At infinite resistance, the solar cell is in an open circuit condition and the voltage
measured across the solar cell in this condition is called the ”open-circuit voltage” (VOC ). At some
intermediate resistance, the product of current and voltage is maximized and this represents the
most power that can be extracted from the solar cell and is therefore called the ”maximum power
point” (MPP or PM AX ). To extract an efficiency value, the I-V curve must be converted to a J-V
curve by dividing the current by the area of the solar cell to obtain a current density (in units of
current/area) (Figure 1.11). This turns the ISC into a ”short-circuit current density” (JSC ) and
the power into a power density. The power density, when divided by the input power density of the
sun (100 mW/cm2 ) yields the efficiency of the solar cell (η), given in %. A fourth metric called the
fill factor (FF) is the ratio of the measured PM AX over the product of the JSC and VOC .

CHAPTER 1. INTRODUCTION

12

Figure 1.11: J-V curve of a perovskite solar cell. The relevant points of JSC , VOC , and PM AX are
shown.
The short-circuit current density is a product of the solar cell’s ability to absorb light,
extract that current under zero external resistance, and how much light is available from the sun to
absorb at each wavelength (energy) integrated over all wavelengths (energies). The product of the
amount of light absorbed at each wavelength and the probability of extracting the current generated
by that light is called the external quantum efficiency (EQE) (Figure 1.12). Integrating the EQE
with the solar spectrum generates a solar cell’s JSC . EQE is generally measured by connecting the
solar cell to an ammeter, placing it in a dark space, illuminating it with monochromatic light, and
measuring the amount of current generated compared to the amount of light shining on the cell.
The monochromatic light is generated by splitting a white light source into separate wavelengths
(similar to a prism) and selecting only one wavelength at a time.

CHAPTER 1. INTRODUCTION

13

Figure 1.12: External quantum efficiency of a perovskite solar cell as a function of wavelength.
Unity represents perfect quantum efficiency.
The interaction of light with the solar cell is extremely important in determining its efficiency. At all times, light has three options when interacting with matter. It can be reflected,
absorbed, or transmitted (Figure 1.13). For solar cells, it is ideal to have no external reflections
for light above the bandgap of the absorber (but internal reflections can be good), and to absorb
all of the light above the bandgap. For a tandem, these rules extend to the bandgap of the bottom
absorber in the tandem, meaning that the top absorber must be able to transmit the light of an
energy between the top and bottom absorber bandgaps so that it can be absorbed below. These
measurements are conducted in a spectrophotometer.

CHAPTER 1. INTRODUCTION

14

Figure 1.13: Schematic examples depicting the processes of transmission, reflection, and absorption
of light. Image from www.chroma.com.
With this knowledge of solar cells, tandems, and the metal-halide perovskite, we can construct the first prototypes of perovskite/silicon and perovskite/CIGS solar cells to learn the limitations of these architectures and the practical iterations necessary to make tandems a viable commercial photovoltaic technology and help photovoltaics reach cost competitiveness with other sources of
electricity. In this thesis I will discuss the initial prototypes of monolithic and mechanically-stacked
tandems, the practical advantages and limitations of both based on these prototypes, and model the
cost of scaled manufacturing.

Chapter 2

Monolithic Perovskite/Silicon
Tandems
Monolithic tandems are the traditional architecture for tandems and are a good place
to start to develop a perovskite/silicon tandem. This chapter is predominantly work published
in Applied Physics Letters in 2015 [27] followed by an unpublished section on the limitations of
monolithic tandem solar cells.

2.1

Published Work

In this prototype, we develop a 2-terminal perovskite/Si multijunction architecture on an n-type
Si solar cell, as shown in Figure 2.1a. We adopt a device area of 1 cm2 . We first start with a
double-side polished <100> n-type float zone Si wafers (15 Ω-cm, 300 µm thickness). The front
side of the wafer is then coated with a silicon nitride (SiNX ) film, which protects the planarity of
the Si front surface during the subsequent random pyramidal texturing step on the back surface of
the wafer. [28] After removing the SiNX protective layer using hydrofluoric acid (HF), we implant
boron on the planar front surface and phosphorus on the textured back surface of the wafer. We
then simultaneously form the p-type B emitter and n-type P back surface field (BSF) by drive-in
annealing in an N2 ambient.

15

CHAPTER 2. MONOLITHIC PEROVSKITE/SILICON TANDEMS

16

Figure 2.1: (a) The device structure of a 2-terminal monolithically grown perovskite/Si multijunction solar cell with an n-type Si base. The polished SEM image is taken at 45◦ tilt to show the Ag
nanowire mesh (500 nm scale bar). (b) Band diagram of the perovskite/silicon cell interface showing
the charge-transport mechanism around the Si tunnel junction.
After the emitter and BSF formation, we form the silicon-based band-to-band tunnel junction on top of the p-type emitter. This interlayer is necessary to facilitate carrier recombination
(holes from the n-type Si base passing through the p-type emitter and electrons from the perovskite
layer passing through its TiO2 electron transport layer respectively, as shown in Figure 2.1b). We
form an n++/p++ tunnel junction by depositing heavily doped n++ hydrogenated amorphous silicon (a-Si:H) using plasma-enhanced chemical vapor deposition (PECVD). The 30 nm-thick a-Si:H
layer is deposited at 250◦ C at a pressure of 200 mTorr (55 sccm of SiH4 gas and 50 sccm of 1%
PH3 in H2 gas) and plasma power density of 0.13 W/cm2 , and subsequently annealed in N2 ambient
at 680◦ C for 15 minutes to activate the dopants. [29–31] It is known that interdiffusion of dopant
species during the device fabrication process (such as the dopant activation anneal) may degrade the
tunnel junction conductivity. Accordingly, a 2-3 nm-thick intrinsic a-Si layer is inserted between the
p++ emitter and the n++ amorphous Si layer during the PECVD process (pressure of 150 mTorr,
55 sccm of SiH4 gas with a plasma power density of 0.16 W/cm2 ) to mitigate possible dopant interdiffusion. [31] After the dopant-activation anneal, the amorphous layers are partially crystallized as
shown in the transmission electron microscopy (TEM) image in Figure 2.2a. [29] Using secondary
ion mass spectrometry (SIMS), we then show that the dopant concentration on the n++/p++ Si
interface after the dopant activation anneal is in the order of 1019 -1020 cm−3 , which is suitable to
form a high-quality interband tunnel junction (Figure 2.2b). [31]

CHAPTER 2. MONOLITHIC PEROVSKITE/SILICON TANDEMS

17

Figure 2.2: (a) TEM image of the n++/p++ silicon tunnel junction interface after the dopant
activation annealing (left: 30 nm scale bar) and high-resolution TEM image of the n++ layer,
showing the partially crystalline nature of this layer (right: 5 nm scale bar). (b) SIMS profile of
the Si emitter and tunnel junction layer showing the sharp doping profile at the tunnel junction
interface. (c) Comparison of J-V profile for identical Si cells with and without a tunnel junction,
showing negligible effect of the tunnel junction to the single-junction Si cell performance.
We further confirm the functionality of the Si-based interband tunnel junction by fabricating single-junction n-type Si solar cells out of the tunnel-junction substrates. We apply an 80
nm-thick SiNX anti-reflection coating (ARC) on the planar front surface of the Si cell, as well as front
finger and back metalization using a Ti/Pd/Ag stack. We show in Figure 2.2c that the addition of
the tunnel junction on top of the n-type solar cell slightly reduces the short-circuit current (JSC ),
but the interband tunnel junction has a negligible effect on the series resistance. The measured
series resistances (RS ) of a cell with and without a tunnel junction are 1.03 Ω-cm2 and 1.08 Ω-cm2 ,
respectively. The RS of the cell without a tunnel junction is larger than the cell with a tunnel junction because the RS addition from the tunnel junction itself is smaller than the sample-to-sample
RSE variability of our c-Si cell fabrication process.
The efficiency of the planar single-junction Si cells are 13.8 % (without tunnel junction),
and 13.2 % (with tunnel junction). This efficiency is lower than commercial averages in part because
we designed our device to be bottom sub-cell compatible with a top perovskite sub-cell (function
under perovskite-filtered light), not a single-junction device functioning under an AM1.5G spectrum.
Some of these intentional design considerations are: (1) No surface texturing for light trapping
is applied because a planar front surface simplifies deposition of the top perovskite sub-cell; (2)
No p-type front surface passivation scheme is applied on the emitter because the same technique
cannot be implemented on the n-type portion of the tunnel junction. In addition to these intentional
design choices, the full-area back surface field (BSF) passivation only provides moderate back surface
passivation and should be upgraded in the future. A front surface passivation scheme which can be
decoupled from the tunnel junction formation needs to be developed. Dedicated clean furnaces for
emitter formation and back surface passivation are also necessary to make more efficient bottom Si
sub-cells.

CHAPTER 2. MONOLITHIC PEROVSKITE/SILICON TANDEMS

18

While the thin layer of Si composing the interband tunnel junction on the top of our cell
degrades the blue response (300-400 nm light) of our device and reduces JSC from 31.6 mA/cm2 to
31.0 mA/cm2 , in practice this effect is negligible, as only perovskite-filtered light with wavelength λ
> 500 nm is incident on the bottom Si sub-cell in the tandem. [32] The external quantum efficiency
(EQE) of our bottom Si sub-cells is shown in Figure 7.1, confirming the negligible parasitic absorption of the tunnel junction λ > 500 nm. It is worth noting that our interband tunnel junction
has low parasitic absorption because it is made from partially crystallized Si, which is an indirect
band gap material. This is in contrast with III-V-based interband tunnel junctions where the tunnel
junction layer thicknesses need to be minimized to reduce the parasitic absorption. [33]
Based on the knowledge that our interband tunnel junction has negligible impact on the
bottom Si sub-cells series resistance, we next fabricate the full monolithic perovskite/Si multijunction
solar cell. We first form 1.1x1.1 cm2 square patterns spaced 1.4x1.4 cm2 apart using photolithography. Afterwards, we form a 1.1x1.1 cm2 mesa on the silicon tunnel junction substrate by etching away
a 300 nm-thick Si layer (the entire n++ tunnel junction and most of the p++ emitter) using reactive
ion etching (RIE) to reduce the dark current in the final perovskite/Si multijunction device. We metalize the textured back side of our bottom Si sub-cell with a Ti/Pd/Ag/Pt metal stack to create the
negative contact for our multijunction cell. Then, we deposit a 30 nm-thick TiO2 layer on the planar
n++ c-Si front surface using atomic layer deposition (ALD) of a tetrakis(dimethylamido)titanium
(TDMAT) precursor (Cambridge NanoTech Savannah ALD tool, 150 ◦ C substrate temperature, 80

C precursor temperature, 440 mTorr base pressure, and 20 sccm N2 carrier gas). To achieve the

desired TiO2 thickness, we do 604 cycles of pulsing H2 O vapor for 0.02 s, waiting for 7 s, pulsing
TDMAT vapor for 0.2 s, and waiting for 7 s. This TiO2 layer serves as the n-type heterojunction
contact for the top perovskite sub-cell. It is known that TiO2 is a good electron-selective contact for
c-Si because of its conduction-band alignment, which also eases electron transport from the TiO2
layer into the n++ Si tunnel junction layer. [34, 35] After removing the TiO2 that remains on the
back metal (deposited during ALD) using a dilute (10%) HF solution, we cut the substrate into
1.4x1.4 cm2 pieces with the 1.1x1.1 cm2 tunnel junction mesa in the middle.
To build the perovskite sub-cell, an approximately 300-nm-thick mesoporous TiO2 layer
is first deposited by spin coating and then sintered at 500 ◦ C. The perovskite is then deposited in
a two-step conversion method modified from the procedure developed by Burschka et al. [16] An
organic p-type heterojunction contact, spiro-OMeTAD, is deposited by spin-coating on top of the
perovskite. This layer is chemically doped with 8 mol% spiro-(TSFI)2 as developed by Nguyen
et al. [36] and additionally includes the organic additives tert-butylpyridine (tBP) and lithium
bis(trifluoromethanesulfonyl)imide (LiTFSI). The top silver nanowire (AgNW) electrode is deposited
using the procedure developed by Bailie et al. [32] AgNWs are sprayed from solution onto a plastic
(PET) film, and then transferred from the PET to the spiro-OMeTAD layer via mechanical transfer
through the application of pressure via a ball bearing. The silver nanowire electrode before transfer

CHAPTER 2. MONOLITHIC PEROVSKITE/SILICON TANDEMS

19

has a sheet resistance of 9 Ω/ with a peak transmission of 89.5%. A 300-nm-thick silver contact
pad is deposited around the edge of the AgNW area by thermal evaporation to increase the geometry
of current collection as well as provide a means of making mechanical contact to the AgNW mesh,
resulting in active device area of 1x1 cm2 . To reduce incident-light reflection, a 111-nm-thick lithium
fluoride (LiF) layer is deposited by thermal evaporation.
The J-V curve of our 2-terminal perovskite/Si multijunction solar cell under AM1.5G illumination is shown in Figure 2.3a. Due to the hysteresis often observed in organic-inorganic halide
perovskite-based solar cells, it is important to be rigorous with the J-V characterization. [37] For
our device, we use a 5 s delay after each 100 mV voltage step before measuring the current and show
the performance of the device under both forward and reverse bias. However, we still find hysteresis
to be evident at this slow scan rate and further find that the current requires up to 30 s at a given
voltage to settle to a steady-state value. We therefore choose instead to measure the steady-state
values of the three critical points on the J-V curve: open circuit, short circuit, and the maximum
power point, depicted as blue circles in Figure 2.3a. We find the steady-state photocurrent at
short circuit (JSC ) to be 11.5 mA/cm2 , the steady-state voltage at open circuit (VOC ) to be 1.58
V (Figure 7.2), and the steady-state maximum power under load (MPP) to be 13.7 % at 1.20 V
bias as shown in Figure 2.3b. These three values result in a fill factor (FF) of 0.75. It is important
to note that the transient output current (and hence pseudo-efficiency) can be significantly overestimated relative to its steady-state output at MPP without a sufficient settling time. [37] The 1
cm2 cell was aperture-masked to ensure a correct illumination area. These results are summarized
in Table 2.1.
Table 2.1: Photovoltaic parameters from 2-terminal perovskite/Si multijunction cell (hero device).
VOC
JSC
FF
Efficiency
Aperture Area

1.58 V
11.5 mA/cm2
0.75
13.7 %
1 cm2

CHAPTER 2. MONOLITHIC PEROVSKITE/SILICON TANDEMS

20

Figure 2.3: (a) J-V curve of the 2-terminal perovskite/silicon multijunction solar cell under
AM1.5G illumination. Forward and reverse-bias scan directions are shown with 5 s measurement
delay per data point. Steady-state values for JSC , VOC , and MPP are depicted as blue circles and
are measured by averaging over 30 s after reaching steady state. The VOC of 1.58 V is approximately
the sum of the perovskite and Si cell VOC . (b) Time-dependent output current of the multijunction
cell near maximum power point (1.20 V forward bias) showing that the output reaches steady state
after a measurement delay of 30 s. (c) Total device reflection and EQE of the perovskite and Si
sub-cells of a typical perovskite/Si multijunction cell. The EQE spectra exhibit a low blue EQE
in the top perovskite sub-cell due to spiro-OMeTAD parasitic absorption, and low red EQE in the
bottom Si sub-cell due to both spiro-OMeTAD parasitic absorption and the lack of good Si back
surface passivation scheme. The perovskite sub-cell EQE is corrected to match the measured JSC
and the silicon sub-cell EQE is reported as measured
The VOC of our 2-terminal perovskite/Si multijunction has been measured as high as 1.65
V in some of our devices. This result is encouraging, especially because the VOC is approximately
the sum of the VOC for the perovskite top sub-cell and the bottom Si sub-cell illuminated through
a separate semi-transparent perovskite device on FTO [32] (approximately 1.05 V and 0.55 V,
respectively). We speculate that the tandem perovskite sub-cell’s VOC benefits from being in contact
with silicon rather than FTO. Mechanical lamination of the AgNW electrode was previously found
to be highly dependent on pressure, with too much pressure causing shorting. [32] However, we did
not observe shorting of the AgNW electrode in the tandem. This may be because the silicon emitter
is not as conductive as FTO, and therefore local shorts do not affect the full device area.
The slow current-dynamics and corresponding hysteresis observed in the tandem resemble
the sluggish dynamics of our perovskite solar cells and suggest that the perovskite sub-cell limits the
current of the tandem. [37] To investigate further, we illuminate the tandem with a white light LED,
which emits only in the visible spectrum, placing the silicon sub-cell in a current-limiting regime. In
this regime, we observe the current to settle within milliseconds as expected for silicon solar cells and
do not observe hysteresis. These findings are substantiated by EQE measurements (Figure 2.3c) of
the individual sub-cells. Notably, the perovskite junction EQE suffers at wavelengths shorter than
550 nm.
Our low tandem JSC of 11.5 mA/cm2 is caused by the fact that the perovskite illumination in our multijunction configuration comes from the p-side of the device, the opposite direction

CHAPTER 2. MONOLITHIC PEROVSKITE/SILICON TANDEMS

21

from conventional perovskite devices. To understand the directional dependence of illumination
on the perovskite sub-cell, we illuminate a semi-transparent single-junction perovskite solar cell as
developed by Bailie et al. [32] from both the standard illumination direction - through the TiO2
heterojunction, and the reverse illumination direction - through the spiro-OMeTAD heterojunction,
corresponding to the illumination direction of the top perovskite sub-cell in the monolithic tandem.
When illuminated through the TiO2 heterojunction the EQE of the semi-transparent cell integrates
to 17.3 mA/cm2 , whereas when illuminated through the spiro-OMeTAD heterojunction the EQE
integrates to 11.4 mA/cm2 (Figure 7.3). This device did not have anti-reflection coatings to simplify analysis, therefore the photocurrent values are 0.5 to 1.0 mA/cm2 lower than if anti-reflection
coatings were used. We attribute the lower current under illumination from the spiro-OMeTAD-side
to parasitic absorption by the doped spiro-OMeTAD layer. From absorption measurements of doped
spiro-OMeTAD on glass (Figure 7.4) we estimate that the absorbed flux of the AM1.5G spectrum
in this layer is 6.4 mA/cm2 from 300-750 nm, though this value may change slightly in a full device
stack due to optical interference between the device layers. We find that parasitic absorption by
spiro-OMeTAD also reduces the photocurrent available to the bottom Si sub-cell, absorbing the
equivalent of 2.0 mA/cm2 of infrared photons from 750-1200 nm.
The performance of our perovskite/Si multijunction cell is current-limited by the photocurrent of the top perovskite sub-cell to 13.7 % efficiency at steady state. In part, the limited
performance of the perovskite sub-cell is due to parasitic absorption in the spiro-OMeTAD layer.
The parasitic absorption can be reduced with a thinner spiro-OMeTAD layer. At present, the spiroOMeTAD layer thickness is optimized to planarize a rough perovskite top surface. With smoother
perovskite films, the spiro-OMeTAD layer can achieve planarization with a thinner layer. The
parasitic absorption may be completely removed by replacement of spiro-OMeTAD with an alternate p-type heterojunction contact that simultaneously exhibits both good conductivity and low
parasitic absorption. The perovskite sub-cell is additionally limited by quality of the perovskite
absorber. Our single-junction perovskite cells with gold back contacts fabricated as control devices
in a similar fashion to the perovskite sub-cell achieve at best 13.5 % efficiency. Optimization of
deposition conditions, precursor materials, and annealing protocols along with replacement or reduction of the spiro-OMeTAD layer is expected to yield a perovskite top sub-cell equivalent to the
record single-junction perovskite cell, which currently stands at 20.1 %. [38]
For the silicon bottom sub-cell, three improvements are envisioned: first, apply a backsurface field and excellent surface passivation to the back of the bottom Si sub-cell; second, fabricate
the silicon sub-cell in dedicated furnaces; third, decouple the front surface passivation scheme from
the tunnel junction formation. These three improvements are expected to yield a bottom Si sub-cell
with a VOC of 660-720 mV and a matched tandem JSC of 18-19 mA/cm2 upon illumination through
the top perovskite sub-cell.
We expect, with these changes designed to improve the sub-cells to the match the highest

CHAPTER 2. MONOLITHIC PEROVSKITE/SILICON TANDEMS

22

quality devices available today, the monolithic tandem would have a VOC of 1.84 V, a JSC of 19
mA/cm2 , a FF of 0.83, and a corresponding efficiency of 29.0 %. Ultimately, it has been suggested
that these monolithic tandems can surpass 35% efficiency through careful photon management. [39]
In summary, we have demonstrated a 1 cm2 , 2-terminal organic-inorganic halide perovskite/Si multijunction solar cell. This monolithic integration for n-type bottom Si sub-cell architecture is enabled by a Si-based tunnel junction fabricated directly on top of the bottom Si
sub-cell emitter, and by incorporating a semi-transparent silver nanowire-based electrode for the
top perovskite sub-cell. We obtained a multijunction device VOC as high as 1.65 V, which is the
expected sum of the perovskite and filtered silicon single-junction VOC s, demonstrating the potential of this approach. The best 2-terminal multijunction prototype device efficiency is 13.7 %.
This value is low compared to the record efficiency for perovskite or Si cells, in part because this
tandem prototype does not yet have best-in-class perovskite and Si layers. The parasitic absorption
in the spiro-OMeTAD hole-transport layer on the top of the perovskite sub-cell results in a tandem
current-limited by the top sub-cell. Further improvements of the multijunction perovskite/Si device
can be achieved by replacing the spiro-OMeTAD layer with wider band gap hole transport material, improving the quality of the perovskite absorber, use of dedicated furnaces for the Si sub-cell
fabrication, and by implementing better surface passivation scheme on the front and back side of
the bottom Si sub-cell. These improvements can yield a 29.0 % efficient tandem, with the ultimate
efficiency potential of these monolithic tandems surpassing 35 %.

2.2

Limitations of Monolithic Tandems
In the previous section, a 13.7 % efficient perovskite/silicon monolithic tandem had a high

VOC and FF but was highly limited by a large current mismatch between the perovskite (11.5
mA/cm2 ) and the silicon (14.7 mA/cm2 ). Parasitic absorption in the spiro-OMeTAD caused the
mismatch, and replacing the spiro-OMeTAD with a more appropriate window layer material with
no absorptions below 3 eV would result in a current-matched system generating

∼15

mA/cm2 .

Assuming no other changes to the system, this new tandem with a better window layer would be
∼18

% efficient.
As in the earlier section, we believe with the best possible silicon and perovskite materials

available today the tandem could be improved electronically to a VOC of 1.84 V and a FF of 0.83.
With a JSC of 15.0 mA/cm2 , this would produce a tandem efficiency of 22.9 %. This is still a far cry
from the 30 % that we would like the tandem to be. To get to 30 %, a matched photocurrent of 19
mA/cm2 is necessary (along with some minor improvements in VOC or FF). 19 mA/cm2 represents
an efficient division and collection of the solar spectrum between the two sub-cells.
Reaching such a high matched photocurrent may be impossible to reach in practice due to
a fundamental and potentially incompatible difference between silicon solar cells and the solution

CHAPTER 2. MONOLITHIC PEROVSKITE/SILICON TANDEMS

23

deposition of the perovskite. Through a half-century of device optimization, the best silicon solar
cells have texturing on the front and back surfaces of the silicon wafer. Silicon, an indirect-bandgap
semiconductor, absorbs the infrared light at energies close to its bandgap very weakly. Texturing the
surfaces of the wafer promotes internal reflection of the infrared light, allowing it to bounce around
inside the wafer until it has a high probability of being absorbed by the silicon.
The perovskite is solution-processed, which is an inherently planarizing method of deposition. This means that if the perovskite is deposited on a rough surface, more perovskite will be
deposited in the valleys and less perovskite will be deposited on peaks. In the case of silicon, the
texturing of the wafer produces pyramids that have a height around 5-12 µm. The perovskite, on
the other hand, is ideally between 250 and 500 nm thick. With a full order of magnitude difference
between the perovskite thickness and the roughness of the surface, the consequence is that the peaks
of the silicon pyramids are left completely bare of perovskite (Figure 2.4). Exposed silicon, when
the tandem is finished with an electrode on top, represents a shorting pathway that bypasses the
perovskite cell. This shunt reduces the effectiveness of the perovskite top cell, and will limit the
tandem’s efficiency. One possible way around this limitation is to vacuum-deposit the perovskite
and heterojunction layers (vacuum deposition is a conformal method of deposition) - but this may
offset the low tooling cost and high throughput that were major advantages of the solution-processed
metal-halide perovskite.

Figure 2.4: Perovskite is solution-processed onto a textured silicon wafer. The nominal thickness
of the perovskite layer is 250-500 nm while the thickness of silicon pyramids is 5-12 µm. More
perovskite is deposited in the valleys between the pyramids while the peaks are left completely bare
of perovskite
If the perovskite cannot be deposited onto a textured silicon surface, the silicon wafer must

CHAPTER 2. MONOLITHIC PEROVSKITE/SILICON TANDEMS

24

be polished on top. Losing some of this light-trapping in the silicon will hurt the infrared response
in the silicon solar cell, making current-matching more difficult.
A further complication is the optical effect of depositing thin film layers onto a silicon
solar cell (Figure 2.5). With a high refractive index ∼3.9, silicon requires an anti-reflection coating
in order to prevent reflections from the air-silicon interface (a silicon solar cell without an antireflection coating is shown as the black curve in Figure 2.5, and a silicon solar cell with an ideal
double-layer anti-reflection coating is shown as the blue curve in Figure 2.5). In the monolithic
tandem, the perovskite must be an anti-reflection coating for the silicon as well as optimized for
optical and electrical properties of the layers to make the perovskite cell efficient. As optimized for
a single-junction perovskite solar cell, if these layers are deposited onto a silicon solar cell, the result
is the red curve in Figure 2.5 which shows an intermediate performance between the two extremes.
Changes in the perovskite layers can improve the anti-reflection properties of the stack, but will
more than likely come as a sacrifice in the performance of the perovskite sub-cell.

Figure 2.5: Result of a PC1D simulation. Blue curve represents the EQE of a silicon cell with
full texturing and an ideal double layer anti-reflection coating. Black curve represents the EQE
of a silicon cell without front surface texturing and the silicon surface in direct contact with air.
Red curve is the EQE of a silicon solar cell with a planar front surface with a thin film layer stack
with thicknesses and real refractive indices similar to a perovskite solar cell showing an intermediate
performance between the two extremes. Numbers in the legend refer to the refractive index and
thickness of the thin film layers on top of the silicon.
If the processing issues with the monolithic tandem can be overcome to manufacture a 30
% efficient solar cell, it must be put in a module and operated outside in the field. In the lab, a solar
cell is designed to operate under a standardized spectrum AM1.5G (Figure 1.4). Unlike laboratory
conditions, the solar spectrum changes constantly, with a different shape depending on the time
of day, time of year, latitudinal position, elevation, and whether the sun is obscured by clouds or
the sky is clear. These changes produce solar spectra with a wide variety of shapes - an example

CHAPTER 2. MONOLITHIC PEROVSKITE/SILICON TANDEMS

of Denver, CO shown in Figure 2.6. In Figure 2.6, light to the left of
by the perovskite and light to the right of

∼700

25

∼700

nm is absorbed

nm is absorbed by the silicon. With a tandem

current-matched according to the AM1.5G spectrum, the tandem will also be current-matched in
the field with an average photon energy (APE) ∼1.9 eV. This implies a current-mismatch between
the perovskite and silicon for all other spectra. Therefore, the monolithic tandem, while 30 % in a
laboratory condition, is unlikely to be near this efficiency for much of the time in the field.

Figure 2.6: Representative spectra of the sun’s illumination of Denver, CO over a year. Figure
reproduced from [5]. In this plot, all light to the left of ∼700 nm is absorbed by the perovskite and all
light to the right of ∼700 nm is absorbed by the silicon (assuming a modified perovskite bandgap).
Note that for many of these spectra, a current-mismatch will occur between the perovskite and
silicon.

Chapter 3

Mechanically-Stacked Tandems
using Silver Nanowires
The other scalable tandem architecture as illustrated in Figure 1.8 is the mechanicallystacked tandem. This architecture overcomes many of the limitations of the monolithic tandem
architecture. First, because the perovskite and silicon cells are fabricated independently of each
other, many of the original design optimizations can be maintained. For the silicon cell, this specifically means the silicon wafer can be textured on both sides, recovering the long-wavelength EQE lost
when switching to a planar front surface. For the perovskite solar cell, it is generally designed to be
deposited onto a smooth sheet of TCO-coated glass rather than a rough silicon surface. Second, the
perovskite is no longer coherently connected to the silicon, and anti-reflection/index-matching coatings on the perovskite cell and the silicon cell as well as an appropriate intermediate layer in-between
the cells can effectively minimize reflections. Third, the mechanically-stacked tandem has versatility
in wiring allowing it to recover potential loss in performance from changes in the environment that
the monolithic tandem cannot.
A common critique of the mechanically-stacked tandem is that the cells are often tested in
the laboratory in a four-terminal configuration. This means that the perovskite and silicon/CIGS
solar cells are tested (electrically) independently of each other. If no changes to this design are
made, then the module will also have four terminals (wires) exiting the junction box. Since a
single-junction module has two wires, the extra wires in the tandem will require extra inverters
and complicate the installation process to a degree that installing the tandem is too expensive for
utility-scale deployment, regardless of its efficiency. Fortunately, the mechanically-stacked tandem
can be wired in multiple ways that result in two terminals exiting the module.
Figure 3.1 shows two potential wiring configuration that result in two terminal exiting
the module. In Figure 3.1(Left), The perovskite and silicon cells are wired in series to enforce

26

CHAPTER 3. MECHANICALLY-STACKED TANDEMS USING SILVER NANOWIRES

27

a current-matching condition. The total current in each cell can be matched by adjusting the size
of the perovskite solar cell (the perovskite and silicon do not need to be current-density matched).
However, this current-matching condition is one of the limitations of the monolithic tandem and
should be avoided if possible. In Figure 3.1(Right), the perovskite cells are wired in series and
the silicon cells are separately wired in series. These two strings are wired together in parallel to
enforce a voltage-matching condition. A voltage-matching condition is much less sensitive to the
shape of the solar spectrum. However, the voltage of solar cells is sensitive to the temperature of the
module and the temperature-dependence of the cells are highly dependent upon the bandgap of the
absorber. A third alternative is to use a power electronic circuit in either configuration to correct
for any mismatches in current or voltage between the strings of cells.

Figure 3.1: Mechanically-stacked tandem wiring configurations that result in two terminals exiting
the module. (Left) The perovskite and silicon cells are wired in series to enforce a current-matching
condition. The total current in each cell can be matched by adjusting the size of the perovskite solar
cell (the perovskite and silicon do not need to be current-density matched). (Right) The perovskite
cells are wired in series and the silicon cells are separately wired in series. These two strings are
wired together in parallel to enforce a voltage-matching condition. A third alternative is to use a
power electronic circuit in either configuration to correct for any mismatches in current or voltage
between the strings of cells.
When making a semi-transparent perovskite solar cell, two transparent electrodes are required. The first is deposited onto the glass sheet, which is already a highly-optimized process, and
the second is deposited onto the perovskite cell. The perovskite cell has a number of design requirements for the transparent electrode that make depositing the electrode onto the perovskite difficult.
First, the perovskite is thermally sensitive and if left uncovered the methylammonium iodide salt
with decompose and evaporate at temperatures as low as 60 ◦ C given enough time. Second, the perovskite is solvent-sensitive. Because the perovskite is formed from two salts and deposited from polar
solvents, it is sensitive to the solution-deposition of layers on top of it that are deposited from either
water or polar organic solvents. Third, the heterojunction layers on either side of the perovskite are

CHAPTER 3. MECHANICALLY-STACKED TANDEMS USING SILVER NANOWIRES

28

often organic layers and with the methylammonium molecule, it is possible that the perovskite itself
can be considered as an organic in some situations. The typical transparent electrode is formed by
sputtering of a transparent conducting oxide (TCO). Sputtering is well known to damage organic
layers, and so TCOs are likely incompatible with direct deposition onto the perovskite or organic
heterojunction layers.
The first solution that we developed for compatibility with the perovskite was to use silver
nanowires that were initially sprayed onto a plastic film then mechanically transferred at room
temperature without exposing the perovskite to raised temperature or solvent. The next section is
work published in Energy and Environmental Science in 2015 [32], followed by a section concerning
the stability of these cells. Specifics of the nanowire lamination procedure can be found in the
Appendix.

3.1
3.1.1

Published Work
Semi-Transparent Cells
Mechanically-stacked tandems require a semi-transparent top cell, as illustrated in Figure

3.2. The perovskite solar cell architecture used in this study is similar to that developed by Burschka
et al [16]. We use a mesoporous titanium dioxide (TiO2 ) layer infiltrated with the perovskite and contacted on either side by electron-selective (compact TiO2 ) and hole-selective (2,2,7,7-tetrakis(N,N-dip-methoxyphenylamine)-9,9-spirobifluorene, spiro-OMeTAD) contacts (see Perovskite Methodology
subsection). For compatibility with these existing electron- and hole-selective contacts, we use the
MAPbI3 perovskite rather than the optically ideal MAPbBrI2 . MAPbBrI2 was also not chosen due
to a photo-instability observed in this material [40]. The transparent front electrode is fluorine-doped
tin oxide (FTO) coated glass. Typically, a perovskite solar cell is opaque with an approximately
100-nm-thick metal back electrode of either Au or Ag. This metal back electrode provides a lowresistance electrical contact and a reflective surface, giving the perovskite a second chance to absorb
any light that was not absorbed on the first pass. To enable the transparency required to make a
mechanically-stacked tandem, we needed a transparent top electrode.
The technical constraints that the top transparent electrode must meet are stringent. The
electrode must be highly transparent in the critical 600-1000 nm window where the perovskite
is not absorbing all of the light and the bottom cell has significant external quantum efficiency
(EQE). The sheet resistance of the transparent electrode should be at most 10 Ω/ [41] because the
transparent electrode must have high lateral conductivity to minimize resistive loss when carrying
the large current density generated in the perovskite cell. Perhaps most importantly, this electrode
must be applied after deposition of the spiro-OMeTAD layer onto a temperature- and solventsensitive perovskite solar cell without damaging it. For these reasons, high-performance transparent
conductive oxides widely used in industry cannot be directly sputtered onto a perovskite solar cell

CHAPTER 3. MECHANICALLY-STACKED TANDEMS USING SILVER NANOWIRES

29

without a buffer layer. An electrode meeting these criteria has not been demonstrated before now.
We use a silver nanowire (AgNW) mesh electrode which has been shown in other cases to have a low
sheet resistance and high optical transmission [42–44], and develop a new method of depositing this
electrode onto our perovskite cell in a room-temperature solvent-free process. This AgNW electrode
serves as the linchpin for our mechanically-stacked tandem architecture.
The current-voltage curves and metrics of the semi-transparent perovskite cells and opaque
control devices are shown in Figure 3.2 and Table 3.1. The loss in absorption in the perovskite
due to the removal of the opaque metal back electrode was offset by reduced reflection from the glass
surface by the AR coating, yielding comparable JSC between the semi-transparent and opaque cells.
We note that if the opaque cell had an AR coating, it would have approximately 0.5 mA/cm2 higher
photocurrent. We control our measurements for hysteresis in accordance with a paper by Unger et
al [37]. We found a 5 s delay time between stepping the voltage and measuring current necessary to
achieve steady state and remove any semblance of hysteresis. This procedure for removing hysteresis
was corroborated and confirmed by NREL when a device was sent for certification. Shadow masks
were used to define the illuminated area of a device. Opaque devices were illuminated through a
0.12 cm2 mask and semi-transparent devices were illuminated through a 0.39 cm2 mask.
Table 3.1: Performance metrics of semi-transparent and opaque perovskite devices.

Semi-Transparent Perovskite
Opaque Electrode Perovskite

JSC (mA/cm2 )
17.5
17.5

VOC (mV)
1025
982

FF (-)
0.710
0.740

Efficiency (%)
12.7
12.7

CHAPTER 3. MECHANICALLY-STACKED TANDEMS USING SILVER NANOWIRES

30

Figure 3.2: Perovskite device results. a) Current-voltage curves comparing best opaque vs. semitransparent perovskite devices. b) EQE of semi-transparent device and opaque device. Note that
the opaque device does not have AR coatings. c) Transmission through semi-transparent perovskite
with AR coatings. Peak transmission is 77 % around 800 nm.

CHAPTER 3. MECHANICALLY-STACKED TANDEMS USING SILVER NANOWIRES

31

Figure 3.2c shows that the transmission through the semi-transparent device peaks at 77%
around 800 nm, the center of the 600-1000 nm transmission window that is critical for tandems.
Much of the transmission loss is due to parasitic absorption in the FTO electrode, AgNW electrode,
and spiro-OMeTAD layer. Uniquely, our semi-transparent device has both a high below-bandgap
transmission and a high efficiency. Previous semi-transparent devices have had to sacrifice one of
these metrics to achieve the other. [43, 45] There remains significant room for improvement in the
transmission. Low-temperature processes would allow for fabrication of the perovskite cell on ITO,
which is more transparent than FTO. A more transparent hole-transporter than spiro-OMeTAD,
which in its oxidized form absorbs light throughout the visible and infrared [36], would also improve
transmission.

3.1.2

Mechanically-Stacked Tandems
We have made tandems with both Si and CIGS as bottom cells. Both have a bandgap

around 1.1 eV, which is sub-optimal for a single-junction solar cell but optimal for a double-junction
tandem [46], and are commercially successful solar technologies. We use a 17.0 % laboratory-scale
CIGS device made using previously reported procedures. [47–50] Although CIGS cells with 21 %
efficiency can be made, we chose a cell with a more modest efficiency for this demonstration to
illustrate how cells that can be made at scale could be enhanced with a perovskite top cell. The
current-voltage curves and external quantum efficiency of the semi-transparent perovskite solar cell,
the CIGS solar cell and the CIGS solar cell underneath the perovskite solar cell are shown in Figure
3.3. To arrive at the efficiency of the 4-terminal tandem, the efficiency of the semi-transparent
perovskite cell is added to the efficiency of the CIGS solar cell when underneath the perovskite cell.
With our 12.7 % semi-transparent perovskite cell, we improve the 17.0 % CIGS cell to 18.6 % in a
tandem (Figure 3.3 a,b/Table 3.2) as measured in-house.

CHAPTER 3. MECHANICALLY-STACKED TANDEMS USING SILVER NANOWIRES

32

Figure 3.3: Perovskite and CIGS/Si tandem results. a) Current-voltage and b) EQE of semitransparent perovskite cell, unfiltered CIGS cell, and CIGS cell filtered by the perovskite cell. c) IV
curves and d) EQE of semi-transparent perovskite cell, unfiltered TI-Si cell, and TI-Si cell with an
infrared-optimized anti-reflection coating filtered by the perovskite cell.
Table 3.2: Performance metrics of mechanically-stacked tandems
JSC (mA/cm2 )
17.5

VOC (mV)
1025

FF (-)
0.710

Efficiency (%)
12.7

TI-Si - Unfiltered
TI-Si w/ IR-ARC - Filtered
Tandem w/ Perovskite + Ti-Si

29.3
11.1

582
547

0.667
0.704

11.4
4.3
17.0

UMG-Si - Unfiltered
UMG-Si w/ IR-ARC - Filtered
Tandem w/ Perovskite + UMG-Si

27.9
9.4

590
553

0.705
0.698

11.6
3.6
16.3

CIGS - Unfiltered
CIGS - Filtered
Tandem w/ Perovskite + CIGS

31.2
10.9

711
711

0.768
0.788

17.0
5.9
18.6

Semi-Transparent Perovskite

CHAPTER 3. MECHANICALLY-STACKED TANDEMS USING SILVER NANOWIRES

33

Figure 3.4: NREL certification of a mechanically-stacked tandem. The tandem efficiency is the
sum of the individually measured cells, 17.9%.

CHAPTER 3. MECHANICALLY-STACKED TANDEMS USING SILVER NANOWIRES

34

Perovskite solar cells are already efficient enough to upgrade the performance of silicon solar cells made with low-quality silicon using the polycrystalline tandem approach. Here, we explore
lower-quality sources of Si including cast multicrystalline silicon (mc-Si) wafers made from feedstock
with high impurity content recycled from the top 10 % of other cast multicrystalline ingots (TI-Si)
(Figure 3.3 c,d/Table 3.2) and cast mc-Si wafers grown using 4.5 N (99.995 % pure) upgraded
metallurgical-grade Si (UMG-Si) (Table 3.2) instead of the more expensive Siemens-grade polysilicon (9 N, or 99.9999999 % pure). Low-quality Si sources generally are not commercially viable
today in single-junction devices because the material cost advantage of low-quality Si is offset by the
reduction in performance due to impurities and crystal defects. We improve an 11.4 % low-quality
Si cell to 17.0 % as a tandem, a remarkable relative efficiency increase of nearly 50 %. Such a drastic improvement in efficiency has the potential to redefine the commercial viability of low-quality
Si. Another tandem cell was sent to NREL for certification using a 17 % silicon bottom cell. The
tandem was certified as 17.9 % efficient (Figure 3.4).
When making tandems as opposed to single-junction devices, some design parameters
change for the bottom cell. The tandem relaxes the design constraints for both the Si and CIGS
top layers. For example, the 20-50 nm CdS window layer used in commercial CIGS devices results
in a photocurrent loss of 0.5 mA/cm2 due to reduced EQE from 400-550 nm caused by parasitic
light absorption in the CdS layer. However, this does not affect the EQE of a tandem because the
400-550 nm light is already absorbed in the top cell, decoupling the optimization of the electronic
and optical properties of the CdS layer. In single-junction Si cells, there is a strict tradeoff of the
series resistance vs. EQE from 400-550 nm due to minority carrier recombination in the emitter
layer. As the bottom cell in a tandem, the emitter thickness or doping can be increased without an
EQE penalty. In this work, the Si has a 35 Ω/ phosphorus-diffused emitter as opposed to 100 Ω/
in industry [51]. Lower sheet resistance in the emitter layer means bus bar spacing can be increased,
reducing shading losses. The design parameters also change for the optimal anti-reflection coatings
used in a tandem. All commercial solar cells use AR coatings to improve the transmission into the
solar cell. For a single-junction cell, the AR coating on top of the cell is optimized for a broad
spectral range from 400-1100 nm and necessarily suffers in performance at the edges of this range.
However, for the bottom cell in a tandem, the AR coating is optimized for a much narrower spectral
range between 800 and 1100 nm, and can maintain a much higher performance through this narrower
spectral range. Full consideration of the different design parameters between single-junctions and
tandems such as these examples could yield further improvements in the future.

3.2

Stability
With the silver nanowire lamination method, we successfully developed a transparent elec-

trode deposition process that did not damage the perovskite. While this allowed for the prototyping

CHAPTER 3. MECHANICALLY-STACKED TANDEMS USING SILVER NANOWIRES

35

of efficient mechanically-stacked tandems, this solution did not provide sufficient stability of the
perovskite cell. The silver nanowire electrode is a mesh of metal on top of the solar cell, implying
that much of the perovskite cell is uncovered by metal. These gaps between the silver nanowires
allowed raised temperatures to cause evaporation of the methylammonium iodide as well as ingress
of moisture from ambient air that degraded the perovskite. Additionally, direct contact of silver
(also true of most metals) with the perovskite (spiro-OMeTAD is not a good ionic diffusion barrier)
corrodes the silver to silver iodide [52], which is an insulator. The combination of these issues resulted in a lifetime for these semi-transparent cells in ambient air under illumination of only a few
hours. These short lifetimes motivated the search for a new transparent electrode solution.

Chapter 4

Mechanically-Stacked Tandems
using ITO
The application of a mesh transparent electrode on the perovskite created a solar cell with a
limited lifetime due to holes in the mesh as well as corrosion of the metal by halides in the perovskite.
The solution to these problems is to deposit a transparent electrode that is a solid layer and is not
composed of metal. Indium tin oxide (ITO) is the industry standard transparent electrode [53] and
is already known to have good moisture barrier properties [54].
Sputtering ITO onto perovskite cells has been attempted before [55] but with limited
success. The solution to the sputtering damage to organics during deposition of the ITO so far has
been to thermally evaporate a layer of MoOX . Even so, these devices have suffered from a lower VOC
than their opaque metal electrode counterparts. The alternative solution we came up with was to
coat the top of the perovskite cell with oxide nanoparticles. These oxide nanoparticles were deposited
from an alcohol solution that was compatible with the perovskite and then provided a buffer layer to
the sputtering of the ITO on top. By using doped oxide nanoparticles, ohmic contact could be made
between the heterojunction layer, the doped oxide nanoparticles, and the ITO, ensuring efficient
operation of the semi-transparent solar cell. The ITO provided a solid capping layer, stabilizing
the perovskite solar cell at high temperatures and because it was not a metal, was not corroded by
halides in the perovskite.
The next section is work submitted for publication to Advanced Materials [56] followed by
a section on the future of mechanically-stacked tandems.

36

CHAPTER 4. MECHANICALLY-STACKED TANDEMS USING ITO

4.1
4.1.1

37

Published Work
Deposition Method
Inverted perovskite solar cells, in which the perovskite is fabricated on top of a hole-selective

contact, can achieve a high power conversion efficiency >18 % [57]. Briefly, we spin poly(3,4ethylenedioxythiophene) polystyrene sulfonate (PEDOT:PSS) onto ITO-coated glass for a smooth,
hydrophilic p-type heterojunction layer. The perovskite is then deposited using a technique developed by Zhang et al. [58], followed by phenyl-C61 -butyric acid methyl ester (PCBM) as an n-type
heterojunction layer. See the Experimental section for more details. We solution deposit nanoparticles of the wide bandgap semiconductor ZnO to serve as an electron-selective buffer layer, protecting
the underlying organic PCBM and perovskite layers. The deep valence level of ZnO enables its use
as a selective contact, Figure 4.1a. ZnO has been demonstrated by Guo et al. [59] and later shown
by You et al. [60] to favorably affect the ambient stability of a perovskite solar cell. A 50 nm layer
of ZnO nps on the PCBM layer and an Al rear electrode enables a stabilized power output of 13.5
% (Figure 4.2b).

Figure 4.1: Semi-transparent inverted perovskite device architecture. a) Energy level diagram.
b/c) Cross-sectional SEM and illustrative schematic of device architecture showing ITO electrode
encapsulation layer.

CHAPTER 4. MECHANICALLY-STACKED TANDEMS USING ITO

38

Figure 4.2: Performance comparison of the opaque and semi-transparent devices showing a) J-V
curve of devices showing comparable FF and VOC and b) max power point tracking.
We sputter 500 nm of ITO directly onto the ZnO-np capped films to achieve a low sheet
resistance of 9.9 Ω/ after annealing at 100 ◦ C. The transmission spectrum of identically processed
ITO on a glass is shown in Figure 7.12. Figure 4.1b shows an SEM cross-section of the ITO
electrode on top of the perovskite. The open-circuit voltage (VOC ) is greater than 0.9 V, indicating
that ZnO successfully acts as a buffer layer to prevent sputter damage of the organic PCBM layer.
However, a large interfacial barrier exists between the ZnO and ITO layers, preventing carrier extraction. This barrier is temporarily removed when operating the device at an elevated temperature
of 70 ◦ C, as shown in Figure 7.10. However, the barrier returns as the device cools. We speculate
that the barrier arises from the misaligned work functions of ZnO and ITO. We introduce aluminum
doped (2 mol %) zinc oxide (AZO) nanoparticles to eliminate the extraction barrier, allowing the
device to operate optimally at 25 ◦ C.

4.1.2

Semi-Transparent Cells
The J-V curves of the opaque (Al/Ag electrode) and semi-transparent (ITO electrode)

devices are shown in Figure 4.2a. The AZO nanoparticles allow operation of the semi-transparent
devices at room temperature with stabilized power efficiency of 12.3% (Figure 4.2b). The semitransparent device displays a high fill factor of 0.77 with a VOC of 0.95 V and JSC of 16.5 mA/cm2
(Table 4.1). The semi-transparent device with a top electrode area of 6 mm x 11 mm is illuminated
through a 4.5 mm x 8.75 mm aperture mask and the opaque device with a top electrode area of
4 mm x 5 mm is illuminated through a 3 mm x 4 mm aperture mask. We scan the devices in a
positive to negative voltage direction with a step size of 50 mV and a delay time between points of
0.2 s (Figure 4.2a). We validate the J-V measurements by operating the devices at their maximum
power point (Figure 4.2b). The performance of the sputtered ITO electrode is validated by the
comparable FF and voltage in the opaque and semi-transparent devices. The drop in current density

CHAPTER 4. MECHANICALLY-STACKED TANDEMS USING ITO

39

in the semi-transparent device can be explained by the lack of a metal electrode back-reflector to
increase light absorption. This decrease primarily occurs at longer wavelengths as shown by EQE
in Figure 7.13.
Table 4.1: Photovoltaic parameters of a semi-transparent perovskite solar cell with ITO rear
electrode compared to an opaque perovskite solar cell with Al/Ag rear electrode.
JSC * (mA/cm2 ) VOC (mV)
Semi-Transparent Perovskite
16.5
952
Opaque Electrode Perovskite
18.8
938
*obtained from EQE integration of AM1.5G spectrum
**obtained from maximum power point tracking

4.1.3

FF (-)
0.77
0.77

Efficiency** (%)
12.3
13.5

Tandems
We test the efficacy of our semi-transparent device in a mechanically-stacked tandem con-

figuration with a mono-crystalline silicon (mono-Si) bottom cell. The results are summarized in
Figure 4.3 and Table 4.2. Alone, the efficiency of the mono-Si is 17.0 %. The silicon cell (6 mm x
13 mm) is limited in VOC due to excess shaded area from the aperture mask and the top electrode.
With the semi-transparent perovskite solar cell stacked in front of the mono-Si cell, the efficiency
improves to 18.0 % (12.3 % + 5.7 %) with a JSC of 13.3 mA/cm2 from the filtered Si bottom cell.
Strong coherent reflections (see Figure 7.15) limit the transmission through the perovskite cell and
the JSC of the filtered silicon cell.
Table 4.2: Photovoltaic parameters of semi-transparent perovskite and mono-crystalline silicon
cells and the resulting tandem efficiency.
JSC (mA/cm2 ) VOC (mV) FF (-)
Perovskite
16.5
952
0.774
Silicon
38.3
587
0.754
Filtered Silicon
13.3
562
0.762
Mechanically-Stacked Tandem
*obtained from maximum power point tracking

Efficiency* (%)
12.3
17.0
5.7
18.0

CHAPTER 4. MECHANICALLY-STACKED TANDEMS USING ITO

40

Figure 4.3: Mechanically-stacked perovskite/silicon tandem performance. a) J-V curves of tandem
with the max power of the tandem calculated from the addition of the perovskite and silicon cells.
b) EQE of original mono-Si, perovskite, and filtered silicon solar cells.

4.1.4

Stability Measurements
Recent work has demonstrated that exposure to moisture results in the formation of both

mono- and di-hydrates of the CH3 NH3 PbI3 material, which slowly degrades into PbI2 [61–63]. This
process is accelerated by heat, which induces egress of methylammonium iodide, and electric fields,
which induce drift of methylammonium and iodine. [64] It is thus crucial to exclude water from reaching the perovskite layer. Other recent work has suggested that the material is thermally unstable
even in inert conditions, and that heating to just 85 ◦ C in a nitrogen atmosphere results in methylammonium decomposition and egress. [65] We examine the thermal stability of both semi-transparent
and opaque devices in ambient atmosphere under AM1.5G illumination using a maximum power
point tracking program. The perovskite in both cases is capped with PCBM then AZO nanoparticles followed by either Al/Ag (opaque) or ITO (semi-transparent). The opaque device has an AZO
nanoparticle barrier similar to the recently published study by You et al. [60] that demonstrated
improved ambient stability compared to devices without a nanoparticle layer. At 100◦ C, we find
that the nanoparticle layer is an insufficient barrier to prevent degradation, despite its reported
effectiveness as a moisture barrier. This result agrees with Conings et al. [65], suggesting that the
methylammonium iodide evolves as a gas (likely as CH3 NH2 and HI), degrades the remaining material into PbI2 and corrodes the metal electrode with the evolved HI vapor. The results make it
clear that storage (in dark conditions) or operation at room temperature is an insufficient metric
to represent the stability of devices operated in the field; exposure to increased temperatures is
necessary to fully evaluate layers used to improve the stability of perovskite cells.
The semi-transparent device, with an amorphous ITO capping layer, is operated at the
maximum power point under full AM1.5G illumination at 60 ◦ C and 100 ◦ C overnight in an ambient
atmosphere without additional encapsulation and shows no measurable degradation (Figure 4.4).

CHAPTER 4. MECHANICALLY-STACKED TANDEMS USING ITO

41

We use a AAA class solar simulator without any UV filtration. Because the cells do not degrade, we
conclude the ITO capping layer has prevented moisture ingress, methylammonium egress, and halide
corrosion of the metal electrode. We should also note that the lack of any deterioration in device
performance is strong support for the notion that the ionic motion responsible for hysteresis does
not necessarily induce degradation in operating devices as long as they are well encapsulated [64,66]
Additionally, the organic charge extraction layers are also protected by the ITO electrode barrier.

Figure 4.4: Thermal stability of ITO-capped perovskite solar cells at 60 and 100 ◦ C compared to
opaque device with ZnO and Al/Ag.
In addition to improved thermal stability, we extract efficiency values at multiple temperatures to estimate the efficiency temperature coefficient of the semi-transparent solar cell (Figure
4.3). We measure the temperature coefficient as -0.22 %(rel)/◦ C (Table 4.3), a significant improvement over the -0.34 %(rel)/◦ C for a commercial CdTe module[38].

CHAPTER 4. MECHANICALLY-STACKED TANDEMS USING ITO

42

Table 4.3: Efficiency as a function of temperature extracted by averaging the data in Figure 4.4
over the entire time period. The temperature coefficient is extracted as a linear fit of the data and
normalized against the expected efficiency at 25 ◦ C.

4.2

Temperature (◦ C)
35
60
100

Efficiency (%)
12.20
11.67
10.46

Temperature coefficient

-0.22 %(rel)/◦ C

Future of Mechanically-Stacked Tandems
Through the use of silver nanowire and ITO transparent electrodes, we have shown the abil-

ity to make 12-13 % efficient semi-transparent perovskite solar cells and 18-19 % efficient mechanicallystacked tandems using commercial-grade CIGS and silicon solar cells that were 17 % efficient alone.
The semi-transparent perovskite solar cells had 70-80 % peak transmission of sub-bandgap light.
With the ITO electrode, the solar cell was made perfectly stable to operation at 100 ◦ C over the
course of 10 h.
While a strong starting point from which to develop, these demonstrated mechanicallystacked tandems are far from the 30 % efficiency that we would like. Part of the difference can be
made up by using the best available silicon and perovskite solar cells available today. If we were to
take the metrics from the record 20.1 % perovskite solar cell and record 25.6 % silicon solar cell [38]
and make the perovskite solar cell semi-transparent, and not change the transmission of our semitransparent perovskite solar cell made in the n-i-p architecture except to flatten the transmission
curve, we would have a much more efficient tandem.
The 20.1 % record perovskite cell has a bandgap

∼1.5

eV and the JSC corresponds to a

perovskite thickness around 400 nm using perovskite n&k values modified from [67]. In a single-pass
solar cell (semi-transparent), the JSC would drop to 21.72 mA/cm2 . From the J-V curve of the 20.1
% cell [38], I extract the ideal diode parameters I0 = 0.65 nA, RS = 0.35 Ω, RSH = 2000 Ω, and n
= 2.364. Using the new JSC of 21.72 mA/cm2 , the semi-transparent efficiency is 17.5 % (Figure
4.5). We take the transmission curve of the cell to be 77 % broadband of the light not absorbed
by the perovskite. This flattens out the transmission curve of Figure 3.2, which is possible if the
parasitic absorption of the spiro-OMeTAD layer in the infrared has been corrected.

CHAPTER 4. MECHANICALLY-STACKED TANDEMS USING ITO

43

Figure 4.5: J-V, EQE, and Transmission of a possible semi-transparent perovskite solar cell based
on the best-in-class 20.1 % opaque solar cell. Semi-transparent perovskite cell efficiency is 17.5 %
The 25.6 % record silicon cell from [68] has a fitted ideal diode with parameters I0 =
15.0 fA/cm2 , RS = 0.364 Ω, RS H = 9782.9 Ω, n = 1.019. Unfiltered, the silicon solar cell has an
integrated EQE of 41.8 mA/cm2 . When filtered by the perovskite cell in Figure 4.5 (with 77 %
transmission), the integrated EQE of the silicon cell drops to 14.58 mA/cm2 . Using this new JSC ,
the filtered silicon cell efficiency is 8.6 % (Figure 4.6)

Figure 4.6: J-V and EQE of the 25.6 % record silicon solar cell if filtered by the transmission curve
in Figure 4.5. Filtered silicon efficiency is 8.6 %
Adding the best-in-class perovskite and silicon solar cells, a mechanically-stacked tandem
cell with 26.1 % efficiency could be made. From this point, 30 % efficiency is in reach through two
parallel efforts: raising the bandgap of the perovskite to 1.7 eV (with similar improvements in VOC )
and improving the transmission of the semi-transparent perovskite solar cell to 85-90 % from the 77
% today.

Chapter 5

Cost-Modeling of Perovskite Solar
Cells
With the prospect of efficient and stable perovskite solar cells and mechanically-stacked
perovskite/silicon tandems, cost modeling can be used to determine the opportunity for commercialization of these technologies. Further, we can estimate whether this new material, especially in the
mechanically-stacked tandem architecture, can make photovoltaics cost-competitive with the other
technologies described in the introduction Table 1.1.
One of the most thorough publicly-available analyses of the requirements to reach cost
competitiveness with other electricity technologies is the SunShot Vision Study technical report [69]
commissioned and published by the U.S. Department of Energy in 2012. The technical and cost
goals set forth by the study were to reach 1.00 $/W installed cost for utility scale solar by the
year 2020 with the suggestions that reaching this target would require a module efficiency > 25 %
and a module cost < 0.50 $/W. These requirements have been updated recently in an article by
Rebecca Jones-Albertus [70] and refined to be defined as a goal of 6 ¢/kWh for the levelized cost
of electricity (LCOE), which is a metric more directly comparable across electricity technologies.

These targets are likely to be reached with additional development over the next few years, with
the International Technology Roadmap for Photovoltaics [7] projecting these efficiencies and LCOE
targets will be met before 2020. This target of 6 ¢/kWh can be viewed as a minimum to reach costcompetitiveness for moderate grid penetration in areas with moderate solar insolation over the year.

Further reductions in the LCOE for solar enable deeper grid penetration in areas with moderate
and high solar insolation as well as open up markets with low solar insolation, making continued
development beyond the SunShot target highly attractive.
To model the cost of the perovskite solar cells, we develop a scalable process flow and make
the assumption that the perovskite can be scaled at high yield. We then account for all material

44

CHAPTER 5. COST-MODELING OF PEROVSKITE SOLAR CELLS

45

input costs, tooling depreciation, and labor and utilities costs in order to arrive at a bottom-up cost
for the perovskite and perovskite/silicon tandem.
The next section is work being prepared for submission to Energy and Environmental
Science [71].

5.1
5.1.1

Published Work
Model Assumptions
To model the single junction perovskite solar cell, we adapt the most common perovskite

cell architecture (Figure 5.1) termed the standard [13] architecture for large area deposition (see
supplemental information for more detail including all layer thicknesses and material costs). The
standard architecture for perovskite cells involves deposition on a glass substrate beginning with
an n-type heterojunction, followed by the perovskite, then a p-type heterojunction layer. 20 nm
of TiO2 is employed as the n-type heterojunction deposited via printing onto a fluorine-doped tin
oxide (FTO) glass sheet followed by crystallization at 500 ◦ C. The perovskite is printed from a
stoichiometric solution of lead(II) iodide (PbI2 ) and methylammonium iodide (MAI) followed by a
100 ◦ C anneal to form 500 nm of the MAPbI3 perovskite. We assume a thicker perovskite layer
than is currently optimized in literature with the expectation that this layer will grow in thickness
as the ability to deposit a high quality perovskite crystal improves. We choose to forego the use of
a porous TiO2 architecture in the model as the literature is methodically moving away from this
architecture and towards thin film architectures more typically utilized in commercial photovoltaics.
50 nm of spiro-OMeTAD, is printed on top of the device followed by the bottom contact. During
operation, the glass sheet is illuminated in a superstrate configuration, implying illumination through
the heterojunction layer deposited directly onto the glass. We model a typical serial-connected thin
film solar module architecture by employing laser scribes and insulator fill to isolate individual cells
and connect them serially. We choose aluminum as the back metal, cognizant of the problem of
metal corrosion by the halogen in the perovskite layer with the expectation that an appropriate
ion barrier will be developed and to estimate what are believed to be a reasonable best-case cost
scenario for a perovskites cell. We choose to seal the perovskite with both a front and back glass
sheet as a likely solution to the environmental instability of the perovskite layer. See appendix for
more details on process flow.

CHAPTER 5. COST-MODELING OF PEROVSKITE SOLAR CELLS

46

Figure 5.1: Single-junction perovskite module scaled process flow.
To model the mechanically-stacked tandem we deviate from a bottom-up cost model to
simplify the model. We choose to use the silicon module target set by the Sunshot program which
is 80 $/m2 at 20 % efficiency [70]. We follow the same perovskite process as Figure 5.1 with the
exception of step 10 where ITO is used as a transparent electrode instead of aluminum (Figure
5.2). The complete semi-transparent perovskite half-module replaces the top glass sheet in the
silicon module. We ignore the step of wiring the two half-modules together or any power electronics
required for field deployment at this time.

CHAPTER 5. COST-MODELING OF PEROVSKITE SOLAR CELLS

47

Figure 5.2: Mechanically-stacked tandem module scaled process flow.
For the tandem performance, we consider the case where the perovskite has reached 20 %
efficiency. HIT-Si bottom cells are modeled after industry standard cells (22 % efficiency). We use
publicly available commercial data and representative quantum efficiency for the silicon technology,
make simple optical assumptions (no interfacial reflections, 3 % reflection broadband from top glass
surface, 5 % absorption loss broadband in each transparent electrode, no sub-bandgap absorption
in the top cell - for a total sub-bandgap transmission of 87 % - and single-pass absorption in the
perovskite), and ignore any deviations from the ideal diode equation for the sub-cells as a function of
light intensity (see supplemental information for details). We assume the bandgap of the perovskite
is tunable, finding a bandgap of 1.74 eV is ideal for the HIT-Si bottom cell. Given a 20 % perovskite
top cell, a perovskite/HIT-Si module would be 30 %.

5.1.2

Single-Junction Perovskite Cost Model
The specific breakdown of perovskite manufacture represented in $/m2 is depicted in Fig-

ure 5.3. The total cost is 53.28 $/m2 for the standard architecture. We can aggregate the specific
costs in three categories: active layers, cell integration, and module assembly. The active layers
consist of the heterojunction layers and perovskite absorber layer. The cell integration category
consists of the TCO layer, P1, P2, and P3 laser scribes, insulator fill, and aluminum electrode.
The module assembly category is defined as the front and back glass, bus bars, EVA and edge seal,
junction box, inspection, testing, and binning. For the standard architecture, the active layers cost
5.58 $/m2 (10.5 % of total), the cell integration costs 15.65 $/m2 (29.4 %), and the module assembly
costs 32.02 $/m2 (60.2 %).

CHAPTER 5. COST-MODELING OF PEROVSKITE SOLAR CELLS

48

Figure 5.3: Single-junction perovskite module step costs.
For the perovskite module, the cost is dominated by module assembly costs that define
a significant cost floor for the perovskite module despite the use of relatively inexpensive absorber
layers. These module assembly costs are largely unavoidable unless flexible packaging is developed
that is compatible with the environmental instability of the perovskite. The cost is interpreted in
the common cost metric, $/W(DC) for a range of efficiencies from 12 to 22 % yielding a potential
cost range from 0.44 $/W(DC) to 0.24 $/W(DC) . As shown in Figure 5.4, there is a $/W(DC)
advantage for the perovskite compared to CdTe given a consistent module efficiency of 16 %.

CHAPTER 5. COST-MODELING OF PEROVSKITE SOLAR CELLS

49

Figure 5.4: Single-junction perovskite compared to CdTe.

5.1.3

Mechanically-Stacked Tandem Perovskite Cost Model
The specific breakdown of the tandem process flow in Figure 5.2 represented in $/m2 is

depicted in Figure 5.5. As a baseline, the silicon module is taken to be 80 $/m2 of which we assume
7 $/m2 is related to the front glass sheet. We remove the 7 $/m2 glass sheet from the 80 $/m2 silicon
module to result in a 73 $/m2 silicon half-module (includes EVA, edge seal, junction box, testing
costs). The perovskite half-module (steps 1-12 in Figure 5.2) totals 32 $/m2 , bringing the full
mechanically-stacked tandem to 105 $/m2 . This translates to 0.35 $/W if the module reaches 30 %
efficiency.

CHAPTER 5. COST-MODELING OF PEROVSKITE SOLAR CELLS

Figure 5.5: Mechanically-stacked tandem module step costs.

50

CHAPTER 5. COST-MODELING OF PEROVSKITE SOLAR CELLS

5.1.4

51

LCOE Comparison Across Technologies
While $/W(DC) is easy to calculate and readily interpretable, it does not account for the

degradation rate of the solar cell. The degradation rate of the module determines its usefulness and
economic value as a solar technology both in terms of the total energy yield of the module deployed
in the field as well as how the module is financed by lending institutions. High degradation rates are
a leading obstacle to the commercialization of perovskite solar cells [64]. To account for degradation
we calculate the levelized cost of electricity (LCOE) over a 30 year operational lifetime. LCOE
is calculated in the same method as laid out by Jones-Albertus et al [70]. We assume a constant
140 $/m2 BoS cost (therefore the BoS cost in $/W is inversely proportional to efficiency) and a
7 % internal rate of return [70] using the Department of Energy’s System Advisor Model (SAM)
to perform the annual solar harvesting and financial calculations. The LCOE target for the solar
industry as set forth by the SunShot program is 6 ¢/kWh, depicted in Figure 5.6 as a horizontal

black line. Figure 5.6 shows the LCOE for technologies with a variety of initial $/W(DC) values. A

horizontal line denotes the 6 ¢/kWh target. Depending on the efficiency of the perovskite cell, the

degradation rate of the a system employing a perovskite module to reach 6 ¢/kWh LCOE varies.

By this method of calculation, the minimum perovskite efficiency to reach 6 ¢/kWh is 16 % with 0
% annual degradation. An 18 % module can afford an annual degradation rate of 0.9 % to reach

6 ¢/kWh, 20 % module - a 1.8 % rate, 22 % module - a 2.7 % rate if the operational lifetime of
the system is 30 years regardless of degradation rate. These limits for the degradation rate are far

slower than what has been achieved to date. The mechanically-stacked tandem, despite being more
expensive from a $/m2 and $/W perspective, has a substantially lower LCOE as a full system than
the other technology scenarios considered. Namely, the high efficiency of the tandem has a large
effect on the system BoS cost, resulting in a < 5 ¢/kWh LCOE for a degradation rate up to 1.4
%/yr and a < 6 ¢/kWh LCOE for a degradation rate up to 3.3 %/yr.

CHAPTER 5. COST-MODELING OF PEROVSKITE SOLAR CELLS

52

Figure 5.6: LCOE calculations for different perovskite scenarios assuming an operational lifetime
of 30 years, a 140 $/m2 BoS cost, and a 7 % internal rate of return.
The scenarios modeled in this work, in particular the mechanically-stacked perovskite/silicon
tandem, have the potential to drive the system cost for utility-scale photovoltaics to 4-6 ¢/kWh
LCOE. This cost range will make utility-scale photovoltaics directly competitive with other sources
of electricity outlined in Table 1.1.

Chapter 6

Conclusions
The future of the photovoltaic industry is unwritten. With worldwide shipments around
40 GW/year today and needing to reach 100-250 GW/year in the future [7], new innovations are
necessary to bridge this gap. In this thesis, I have discussed the initial experimental efforts in a new
class of photovoltaics: polycrystalline tandems using metal-halide perovskites.
The metal-halide perovskite is a unique material class that fills a hole in available semiconductor materials - a high-bandgap, polycrystalline (solution-processable) semi-conductor with small
energetic losses. The combination of these properties has the potential to disrupt a solar industry
that is striving to improve the efficiency of their products while continuing to lower the cost of manufacture. The perovskite bandgap, potentially tunable between 1.5 and 2.3 eV, can be beneficial to
technologies that are sub-optimal in single-junction configurations. These include silicon, the dominant technology in the industry, as well as technologies needing help to reach mass-commercialization
such as CIGS and CZTS.
Tandems can be made successfully in both monolithic and mechanically stacked architectures, as demonstrated by the proof-of-concept prototypes in this thesis. The initial learning while
making these prototypes has led to some insights into the benefits and limitations of these architectures. The monolithic tandem requires substantial re-optimization of both the silicon cell and
the perovskite cell and possibly the development of new layer materials for better window layer
or index-matching properties. Even if these issues are overcome, the current-matching condition
between the silicon and perovskite limits the field performance of these tandems since the solar
spectrum is not constant. Furthermore, the serialized manufacturing (by adding 4-8 steps onto the
silicon cell) may result in a low yield for the finished product. The low yield was certainly more
true in these initial prototypes for monolithic tandems than for the mechanically-stacked option.
The mechanically-stacked tandem has the flexibility necessary to allow the silicon and perovskite
cells to undergo only minimal (or no) re-optimization then the flexibility to wire them into a module
in a configuration that better maintains the laboratory performance when placed in the field. The

53

CHAPTER 6. CONCLUSIONS

54

lone drawback of the mechanically-stacked tandem is the extra transparent electrodes, which make
reaching efficiency targets more difficult.
The path forward for perovskite polycrystalline tandems is clear. With the realistic efficiency potential of a mechanically-stacked tandem made with record perovskite and silicon materials
surpassing 26 % efficiency, efforts towards improving the transparency of the perovskite cell and raising the perovskite’s bandgap should push the tandem efficiency past 30 % within another year or
two of effort. Through bottom-up cost-modeling, the cost of these tandems can significantly reduce
the cost of solar if the module efficiency target of >25 % is reached. After this, the major barriers
to commercialization are scaling the perovskite while maintaining efficiency, doing so at high yield,
and demonstrating solutions to the stability problems of the perovskite. Efforts in all three phases
are already underway by multiple teams around the world and can hopefully be combined into the
tandem structure and processes to form a successful product that leads to ubiquitous solar energy.

Chapter 7

Appendices
7.1
7.1.1

Monolithic Tandem Experimental Information
Silicon Sub-cell Fabrication Procedure
We first start with a double-side polished <100> n-type float zone silicon (Si) wafer (15

Ω-cm, 300 µm thickness). The front side of the wafer is then coated with a 300 nm-thick silicon
nitride (SiNX ) film, which protects the planarity of the Si front surface during the subsequent random
pyramidal texturing step (3% weight KOH solution in DI mixed with isopropanol (6:1 volume), 80◦ C
etch for 20 minutes) on the back side of the wafer. After removing the SiNX protective layer using
hydrofluoric acid (HF), we clean the sample using RCA cleaning procedure (RCA1 = 10 minute,
80◦ C dip in 5:1:1 NH4 OH:H2 O2 :H2 O, RCA2 = 10 minute, 80 ◦ C dip in 5:1:1 HCl:H2 O2 :H2 O):
deionized (DI) water dip, HF dip, RCA1 clean, DI water dip, HF dip, DI water dip, RCA2 clean,
DI water dip, HF dip, DI water dip, N2 drying. We then implant boron on the planar front surface
(11 B with 1.8x1015 cm−2 dose, 6 keV implantation energy) and phosphorus on the textured back
surface (31 P with 41015 cm−2 dose, 10 keV implantation energy) of the wafer. After cleaning the
wafer again using RCA cleaning procedure, we dip the wafer in dilute HF solution for oxide removal,
clean it with DI water and dry it with N2 . Then we simultaneously form the p-type B emitter and
n-type P back surface field (BSF) by drive-in annealing at 960 ◦ C in an N2 ambient for 30 minutes.
After the emitter and BSF formation, we clean the wafers again with RCA cleaning procedure, and then perform dilute HF oxide removal, DI water dip, and N2 drying. We then form
and create our n++/p++ tunnel junction interface by depositing heavily doped n++ hydrogenated
amorphous silicon (a-Si:H) using plasma-enhanced chemical vapor deposition (PECVD). We first
deposit a 2-3 nm-thick intrinsic a-Si layer on top of the p++ emitter using the PECVD process
(temperature of 250 ◦ C, pressure of 150 mTorr, 55 sccm of SiH4 gas with a plasma power density
of 0.16 W/cm2 ). Afterwards, a 30 nm-thick a-Si:H layer is deposited at 250 ◦ C at a pressure of 200

55

CHAPTER 7. APPENDICES

56

mTorr (55 sccm of SiH4 gas and 50 sccm of 1% PH3 in H2 gas) and plasma power density of 0.13
W/cm2 . Subsequently, the Si sample is annealed in N2 ambient at 680 ◦ C for 15 minutes to activate
the dopants and partially crystallize the amorphous layer.
After the tunnel junction formation, we form 1.1x1.1 cm2 square-shaped mesa spaced
1.4x1.4 cm2 apart. This is done using photolithography. We deposit positive photoresist on both
sides of the wafer (Shipley 1813 photoresist spun at 4000 rpm for 40 s, baked at 115 ◦ C for 1 minute).
The front side undergoes mesa patterning and is exposed for 4.5 s, and developed in CD-26 developer
for approximately 1 minute), while the back side is left unexposed to protect the BSF on the back
side during the subsequent mesa formation step. The mesa formation is then done on the front
side of the Si wafer using reactive ion etching (RIE) to etch 300 nm of the Si layer, removing the
n++ tunnel junction and most of the p++ emitter outside the square-shaped mesa. The remaining
photoresist mask is then removed using solvent clean (3 minute sonication in acetone, 3 minute
sonication in IPA, rinsing with DI water and drying with N2 ).
We perform dilute HF oxide removal, DI water dip, and N2 drying. We then form the
back metal by electron-beam evaporation. A layer stack of Ti/Pd/Ag/Pt with layer thicknesses
of 20/20/300/30 nm is chosen (Ti for adhesion, Pd for metal diffusion barrier, Ag for electrical
conduction, and Pt for corrosion protection during perovskite sub-cell processing) followed by rapid
thermal anneal (400◦ C in N2 for 5 minutes) to improve metal adhesion.
We again perform dilute HF oxide removal, DI water dip, and N2 drying before depositing
30 nm-thick TiO2 using atomic layer deposition (Cambridge NanoTech Savannah ALD tool, 150

C substrate temperature, 80◦ C precursor temperature, 440 mTorr base pressure, and 20 sccm N2

carrier gas. To achieve the desired TiO2 thickness, we do 604 cycles of pulsing H2 O vapor for 0.02 s,
waiting for 7 s, pulsing TDMAT vapor for 0.2 s, and waiting for 7 s. After the TiO2 ALD deposition,
we deposit photoresist on the top for TiO2 protection (Shipley 1818 photoresist spun at 4000 rpm
for 40 s, baked at 115 ◦ C for 1 minute). Dilute HF with 10% concentration in DI is then used to
dissolve the TiO2 layer which got deposited on the back metal during the ALD process. Finally, the
tunnel junction substrates are then laser-scribed from the back to form 1.4x1.4 cm2 substrates. After
mechanical cleaving, we have 1.4x1.4 cm2 tunnel junction substrates with 1.1x1.1 cm2 square-shaped
mesa (active junction area) in the middle and TiO2 layer on the planar front surface protected by
photoresist.

7.1.2

Perovskite Sub-Cell Fabrication Procedure
The tunnel junction substrates were then shipped with a protective photoresist layer. The

substrates were prepared for further processing by sonication in acetone for 5 minutes, sonication
in IPA for 5 minutes, rinsing with DI water and drying with N2 , 5 minutes of UV-ozone cleaning,
and sintering for 30 minutes at 450 ◦ C. After cooling down, mesoporous TiO2 films were spun onto
the TiO2 surface at 4000 rpm for 30 s and sintered at 450 ◦ C. The spin-coating solution was a 1:3

CHAPTER 7. APPENDICES

57

dilution of 18-NRT TiO2 paste (Dyesol) in ethanol.
All previous steps were performed in ambient atmosphere. The remainder of device fabrication was performed in a N2 glovebox with <5 ppm O2 and H2 O. The TiO2 substrates were dried
by heating to 500 ◦ C with a hot air gun for 30 minutes and immediately brought into the glovebox.
A 1.3 M PbI2 solution was prepared by dissolving PbI2 (Aldrich, 211168) into anhydrous DMF
(Acros, 32687) and stirring on a hotplate at 100 ◦ C. The DMF was filtered through a 200 nm PTFE
filter (Pall, 4552) prior to adding to the solution in order to remove particulates. Methylammonium
iodide (MAI) was purchased from Dyesol and used as received. A solution of 10 mg MAI per 1 mL
anhydrous IPA (Acros, 61043) was prepared and allowed to dissolve at room temperature. A pure
IPA rinse solution was prepared as well. The IPA was filtered through a 20 nm PTFE filter prior to
adding to the solutions in order to remove particulates.
After the TiO2 substrates were cooled to room temperature, 100 µL of the 100 ◦ C PbI2 /DMF
solution was pipetted onto the substrate and spun at 6500 rpm for 90 s. The resulting film was dried
for 30 minutes on a 70◦ C hotplate. After cooling, the films were dipped in the MAI/IPA solution
for 15 minutes. Films were then rinsed in IPA, dried by spinning at 4000 rpm for 30 s, and placed
back on the 70 ◦ C hotplate for 30 minutes. After cooling, 75 µL of a spiro-OMeTAD (Lumtec,
LT-S922) solution was spun on top at 4000 rpm for 30 s. The spiro-OMeTAD solution was 163
mM spiro-OMeTAD in anhydrous chlorobenzene (Sigma-Aldrich, 284513). The spiro-OMeTAD was
dissolved by placing on a hotplate at 70 ◦ C for more than 30 minutes. 534 mM of tert-butyl pyridine
(Aldrich, 142379) and 86 mM of Li-TFSI (Aldrich, 15224) dissolved as 520 mg/mL in anhydrous
acetonitrile (Acros, 61096) were added to the spiro-OMeTAD solution. In this study, 8 mol% of the
spiro-OMeTAD was spiro-OMeTAD(TFSI)2 , resulting in 16 % of spiro-OMeTAD molecules being
chemically oxidized to ensure conductivity in the spiro-OMeTAD layer. The spiro-OMeTAD(TFSI)2
was synthesized as reported elsewhere in literature. After the spiro-OMeTAD solution was prepared,
it was filtered through a 20 nm Al2 O3 filter (Whatman, 6809-3102) to remove any aggregates and
particulates. Films were then removed from the glovebox and stored overnight in a desiccator at
20% RH.
An AgNW film on PET was transferred on top of the spiro-OMeTAD film similar to
previously reported procedures. A change was made to use a spring-loaded ball bearing to better
control the applied pressure to the AgNW film. The spring was set to 500g force at full depression
of the ball into the housing. After transfer of the AgNW electrode, 300 nm Ag was thermally
evaporated through a patterned shadow mask around the edges of the device leaving 1x1 cm2 active
area in the middle of the 1.1x1.1 cm2 mesa. These bars of silver helped reduce the unnecessary
series resistance in the AgNW electrode by not limiting the current collection to one geometrical
direction. The nearly completed device was lights-soaked under visible illumination for 10 minutes
then stored in a desiccator for 12 hours before applying the anti-reflective coating. A 111 nm-thick
LiF anti-reflective coating was then added to the device. This thickness was optimized to provide

CHAPTER 7. APPENDICES

58

anti-reflection for the visible spectrum.

7.1.3

Multijunction Cell Testing Protocols
Current-voltage characteristics for the multijunction cells were measured using a Keithley

model 2400 digital source meter and a Newport Oriel model #94023A solar simulator. The solar simulator irradiance was characterized and compared to the AM1.5G spectral standard. The
perovskite sub-cell’s spectral mismatch factor was calculated as 0.990 using the EQE data from
Figure 2.3c. The silicon sub-cell’s spectral mismatch factor was calculated as 0.965 using the EQE
data from Figure 2.3c. Neither the solar simulator intensity nor the calculated efficiencies were
increased to account for this small spectral mismatch factor. Consequently, the reported currents
and power conversion efficiencies in this manuscript are likely slightly conservative. Samples were
illuminated through a 1 cm2 aperture area. A 5 s delay time at each voltage step was used to try
and minimize hysteresis. Longer delay times were impractical due to the limitations of the testing
software. Because transient hysteretic behavior was observed for up to 30 s after setting a voltage,
steady-state values of JSC , VOC , and MPP were determined by setting the voltage condition for 30
s then averaging data for the next 30 s. The cell was illuminated through 1 cm2 aperture mask to
ensure a correct illumination area.
External quantum efficiency (EQE) for the tandem was recorded as a function of the
wavelength using a Model SR830 DSP Lock-In Amplifier (Stanford Research Systems). A 100 W
tungsten lamp (Newport) was used to provide an excitation beam, which was focused through a
Princeton Instruments SpectraPro 150 monochromator. To measure the EQE of the perovskite subcell, an 870 nm infrared LED light source (Enfis) illuminated the sample to force the top perovskite
sub-cell into a current-limiting regime. The monochromated signal was chopped slowly at 1.2 Hz
to allow for settling of the signal. At each wavelength, the program waited for a delay time of 30 s
then data was collected for 5 s. The time constant on the lock-in amplifier was 1 s. We found that
the infrared chromatic bias caused an approximately 500 mV forward bias in the silicon sub-cell,
applying a 500 mV reverse bias to the perovskite sub-cell to maintain short circuit conditions. It
has been shown that applying a reverse bias to many perovskite solar cells prior to and during
photocurrent measurements results in a uniform decrease in the measured EQE spectrum, possibly
due to a temporarily reduced current collection efficiency resulting from ion drift. To estimate the
EQE spectrum of the perovskite sub-cell under broadband AM 1.5 illumination where these nonideal operating conditions are not present, the perovskite sub-cell EQE was scaled by a constant
factor to match the measured JSC . To measure the EQE of the bottom Si sub-cell, a 465 nm visible
LED light source (Enfis, 7 mW/cm2 ) illuminated the sample to force the bottom Si sub-cell into a
current-limiting regime. The monochromated signal was chopped quickly at 500 Hz to overcome the
capacitive impedance of the perovskite sub-cell. The Si EQE is reported as measured.

CHAPTER 7. APPENDICES

7.1.4

59

Supplemental Figures

Figure 7.1: EQE curve of single-junction n-type silicon cells with and without a tunnel junction.
This curve shows negligible parasitic absorption in the tunnel junction for wavelength λ > 500 nm.

Figure 7.2: Transient (a) JSC and (b) VOC of the perovskite/silicon tandem cell as they stabilize
over the time. The measured steady-state values are JSC = 11.5 mA/cm2 and VOC = 1.58 V,
respectively. The 30 s settling time for the VOC is not shown as the cell was stabilized at VOC for
> 30 s prior to starting the measurement.

CHAPTER 7. APPENDICES

60

Figure 7.3: EQE of a semi-transparent perovskite solar cell illuminated through either the nside (glass/TiO2 side) or through the p-side (AgNW/spiro-OMeTAD side). The glass side EQE
integrates to 17.3 mA/cm2 while the AgNW side EQE integrates to 11.4 mA/cm2 . Neither side has
anti-reflection coatings.

Figure 7.4: The light transmission through a 470-nm-thick doped spiro-OMeTAD film on glass.
The contribution of the glass is removed from this plot. The absorption features from 300-400 and
450-550 nm are readily visible in the EQE plot above.

CHAPTER 7. APPENDICES

7.2

61

Silver Nanowire Deposition Methods

We first form our AgNW transparent electrode on a flexible polyethylene terephthalate (PET) film
by spray deposition. AgNWs were deposited using a custom built spray deposition system. 4.5 mg
of silver nanowires (Blue Nano BL35L - 35 nm diameter, 15 µm length or ACS Agnws-60 - 60 nm
diameter - 30µm length) were dispersed in 40 mL of methanol, mixed in a vortex mixer for 30 s, and
placed in an ultrasonic bath for 10 s. This value was chosen such that the transmission/conductivity
tradeoff of the AgNW electrode maximizes the power conversion efficiency of the tandem. The
dispersion was delivered at a rate of 4 mL/min by a syringe pump to an atomizing nozzle (Spray
Systems Co. 1/4JN-SS+SU11DF-SS). The nozzle was positioned 72 mm above the substrate surface
and was supplied with nitrogen gas at 50 psi. The PET was taped to a computer controlled X-Y
motion stage under the nozzle and moved at 20 cm/s in a pattern such that the nozzle above
uniformly covered a 4.5 cm x 25 cm rectangular area during deposition. . The deposition was
carried out on 1-5 mil thick Polyethylene terephthalate (PET). The choice of PET as a substrate for
the AgNW film is important for the transfer lamination technique for reasons described below. The
spray deposition was carried out at the elevated temperature of 60 ◦ C to increase the conductivity
of the AgNW electrode. PET has a glass transition temperature of about 70 ◦ C and so higher
deposition temperatures and further annealing, although desirable, could not be used here.
This spray deposition process yielded a 50 x 300 mm of AgNW transparent electrode on
PET. This was then cut with scissors into pieces approximately matching the perovskite substrate
size. The AgNW film on PET was stored at ambient laboratory conditions for two weeks before
transfer lamination.

7.2.1

Lamination By Point Force
The patterned AgNW film on PET was placed facedown onto the nearly completed per-

ovskite device so that the AgNWs were in contact with the top spiro-OMeTAD layer. A 0.17 mm
thick glass coverslip was placed on top of the PET substrate. The AgNWs were transfer laminated
from the PET to the perovskite solar cell by applying approximately 500 g of downward force onto
the coverslip through a single 1/4-inch diameter ball bearing.
The 500 g force was applied by one of three methods. First, A metal ball bearing was
controlled by hand to apply the correct pressure. Manual control of pressure was found to often
result in overpressure, causing shorts in the solar cell by pressing the nanowires too far into the film.
Second, a metal ball bearing on a spring was controlled by hand to apply the correct pressure. The
spring was set to apply 500g of force when fully compressed. As manual control over position is
much better than manual control over force, this method proved much more reproducible. Third,
the metal ball bearing was enclosed in a housing on rails where the height of the ball and therefore
the compression on the spring could be set without relying on manual control.

CHAPTER 7. APPENDICES

62

The bearing was selectively rolled over the active area of the perovskite device so that the
AgNW film was completely and uniformly donated from the PET to the top spiro-OMeTAD layer
of the perovskite solar cell. The rolling action of the bearing reduces lateral shear force on the PET
preventing any movement of the donor PET substrate relative to the acceptor perovskite solar cell.
Lateral movement here causes discontinuities in the laminated AgNW film severely degrading its
conductivity. The flexibility and softness of the PET substrate allows the AgNW film to conform
to the surface of the perovskite device during transfer despite any dust or other imperfections that
may be present on either the surface of the AgNW film or the surface of the spiro-OMeTAD layer.
This, coupled with the relatively small contact point of the ball bearing ensures complete transfer
lamination of the AgNW film to the perovskite device without damaging the mechanically sensitive
nanostructured AgNW film in the presence of dust or other imperfections.
The 500 g transfer force was found to be sufficient to ensure the AgNWs were completely
donated from the PET but not too much that they were forced through the spiro-OMeTAD layer
causing bridges/shunts across it. AgNW bridges across the spiro-OMeTAD layer lead to increased
recombination since the spiro-OMeTAD layer can no longer effectively block electrons and in extreme
cases shunting of the device if the AgNWs bridge through the TiO2 as well. The coverslip serves two
purposes. First, it isolates the lateral movement of the ball bearing from the PET, which prevents
cracks, or discontinuities in the transferred AgNW film as described above. Second, it serves to
increase the area over which the force from the ball bearing is applied to the PET, thus reducing the
pressure felt by the AgNWs during the transfer process. This reduced pressure is a further safeguard
against AgNWs bridging through the spiro-OMeTAD layer.

7.2.2

Lamination By Distributed Force
Lamination of the silver nanowires by point force was highly variable. The glass coverslip

was necessary to the process in order to distribute the force appropriately from the point source.
While this was often fine, the rigid glass coverslip did not conform well to dust particles and other
sources of macroscopic roughness on the surface of the perovskite device. These particles caused
”keep-out” zones to form around the particles where no silver nanowires were deposited. While these
”keep-out” zones could be reduced in size by pressing harder, the increased pressure often caused
shorting of the solar cell.
The solution to the ”keep-out” zones, aside from removing all particulates from the system,
was to remove the need for the glass coverslip. Applying pressure directly rather than applying a
force would bypass the need for distribution of an applied force. To this end, I designed and built a
system to apply a gas pressure uniformly onto the perovskite solar cells to laminate the nanowires
(Figure 7.5)
Vacuum/pressure lamination is a strategy common to the manufacture of composite parts,
especially with non-standard shapes such as airplane wings or boat hulls. The idea is to use the

CHAPTER 7. APPENDICES

63

vacuum to remove air bubbles from the system and pressure (sometimes) to press parts together
then heat is generally applied to cure a resin and adhere the composite layers together. A similar
system was used here, using vacuum to remove air bubbles and a positive pressure source to press
the nanowires into the solar cell uniformly. Heat was not necessary for this application. A pressure
∼3000

psi was applied to the top half of the chamber as the transfer pressure.
This method proved much more reproducible than the point force lamination and was much

less prone to shorting. However, the diaphragms used were not well designed for this system and
often broke under the high pressures and the sharp edges of the glass.
While this is a batch process and therefore would likely have been expensive to scale, a
compromise between the 0-dimensional (point source) application of force and the 2-dimensional
(pressure source) application of force could be the use of a 1-dimensional application of force such
as using a roller. Here, dust and other particulates would be an issue still, but by using a soft roller
the applied pressure may still be uniform without the use of a pressure distribution layer (such as
the glass coverslip).

CHAPTER 7. APPENDICES

64

1.

2.

3.

4.

5.

6.

7.

8.

Figure 7.5: Process for vacuum lamination of silver nanowires onto a solar cell. 1) The baseplate
of the transfer chamber designed to withstand the pressures of the process. At the bottom of this
baseplate is a connection to a vacuum source used to remove all air pockets that can disrupt the
uniformity of the applied pressure. 2) A porous metal (aluminum) baseplate is used to distribute
the vacuum evenly across the surface of the porous plate to avoid plugging the vacuum source and
prevent removal of air pockets. 3) An O-ring is used to provide an edge seal between the top and
bottom halves of the chamber to ensure proper pressure buildup. 4) The solar cell is placed in the
middle of the porous metal plate. 5) The silver nanowires on a plastic film are placed face-down
onto the solar cell. 6) A tape seal is placed in-between the O-ring and the porous metal plate to
allow the application of a diaphragm. 7) A plastic diaphragm is attached to the tape and provides a
barrier between the two halves of the transfer chamber. Vacuum is pulled on the lower half and the
diaphragm deforms around the solar cell, providing a flat bubble-free surface onto which a uniform
pressure can be applied. 8) The top half of the transfer chamber is placed on top and bolted to the
bottom half through radial bolt holes in both chambers. The o-ring provides a seal between the top
half and ambient. A fitting on the side of the top half connects to a positive pressure source such
as a compressed gas canister. Positive pressure is applied to the diaphragm, uniformly pressing the
nanowires into the solar cell.

CHAPTER 7. APPENDICES

7.3

65

Mechanically-Stacked Tandems Using Silver Nanowires
Experimental Information

7.3.1

Perovskite Methodology
Pilkington TEC15 FTO glass (1.6 and 2.2 mm) was patterned by selective etching with Zn

powder (J.T. Baker, 4282-01), 4 M HCl (Fisher, A144-212), and mechanical abrasion with a cotton
swab. The glass was cleaned by sonication in a diluted Extran solution (EMD, EX0996-1), acetone
(EMD, AX0115-1), and isopropanol (IPA) (EMD, PX1835P-4). After 20 min of UV-ozone, the glass
was heated to 500 ◦ C on a hotplate. A 1:10 dilution of titanium diisopropoxide bis(acetylacetonate)
(Aldrich - 325252) in ethanol (Sigma-Aldrich, 187380) was repeatedly sprayed from an airbrush
nozzle to achieve a

∼50

nm thick films of TiO2 on top of the FTO. After cooling down, the glass

was immersed in a 70 ◦ C batch of 40 mM TiCl4 (Sigma-Aldrich, 208566) in ultrapure water (J.T.
Baker, 6906-02) for 30 min. The glass was then rinsed in DI water and dried on a 70 ◦ C hotplate
for 15 min. After cooling down, mesoporous TiO2 films were spun onto the TiO2 /FTO surface at
4000 rpm for 30 s and sintered at 450 ◦ C. The spin-coating solution was a 1:3 dilution of 18-NRT
TiO2 paste (Dyesol) in ethanol.
All previous steps were performed in ambient atmosphere. The remainder of device fabrication was performed in a nitrogen glovebox with < 5 ppm O2 and H2 O. The TiO2 substrates were
dried by heating to 500 ◦ C with a hot air gun for 30 min and immediately brought into the glovebox. A 1.3 M PbI2 solution was prepared by dissolving PbI2 (Aldrich, 211168) into anhydrous DMF
(Acros, 32687) and stirring on a hotplate at 100 ◦ C. The DMF was filtered through a 200 nm PTFE
filter (Pall, 4552) prior to adding to the solution in order to remove particulates. Methylammonium
iodide (MAI) was synthesized according to a previously reported procedure. [12] A solution of 10
mg MAI per 1 mL anhydrous IPA (Acros, 61043) was prepared and allowed to dissolve at room
temperature. A pure IPA rinse solution was prepared as well. The IPA was filtered through a 20
nm PTFE filter prior to adding to the solutions in order to remove particulates.
After the TiO2 substrates were cooled to room temperature, 100 µL of the 100 ◦ C PbI2 /DMF
solution was pipetted onto the substrate and spun at 6500 rpm for 90 s. The resulting film was
translucent yellow and dried for 30 min on a 70 ◦ C hotplate. After cooling, the films were dipped
in the MAI/IPA solution. The films were monitored optically for the formation of the perovskite
(Figure 7.6). The signal at 700 nm was used to determine the growth rate of the perovskite,
while the signal at 850 nm was used to detect the presence of other optical phenomena (changes
in scattering, reflection, incident light intensity). When the derivative of both signals matched, the
formation was considered complete and the perovskite film was removed, rinsed in IPA, dried by
spinning at 4000 rpm for 30 s, and placed back on the 70 ◦ C hotplate for 30 min. At this point,
the films were a translucent brown. After cooling, 75 µL of a spiro-OMeTAD (Lumtec, LT-S922)
solution was spun on top at 4000 rpm for 30 s. The spiro-OMeTAD solution was 59 mM (for opaque

CHAPTER 7. APPENDICES

66

devices) and 163 mM spiro-OMeTAD (for semi-transparent devices) in anhydrous chlorobenzene
(Sigma-Aldrich, 284513). The spiro-OMeTAD was dissolved by placing on a hotplate at 70

C for

>30 min. 193 mM of tert-butyl pyridine (Aldrich, 142379) and 31 mM of Li-TFSI (Aldrich, 15224)
dissolved as 520 mg/mL in anhydrous acetonitrile (Acros, 61096) were added to the 59 mM spiroOMeTAD solution. For the 163 mM spiro-OMeTAD solution, the amount of tert-butyl pyridine and
Li-TFSI was kept consistent with respect to the concentration of the spiro-OMeTAD. In this study,
8 mol% of the spiro-OMeTAD was spiro-OMeTAD(TFSI)2 [36], resulting in 16% of spiro-OMeTAD
molecules being chemically oxidized to ensure conductivity in the spiro-OMeTAD layer. After the
spiro-OMeTAD solution was prepared, it was filtered through a 20 nm Al2 O3 filter (Whatman,
6809-3102) to remove any aggregates and particulates. Films were then removed from the glovebox
and stored overnight in a desiccator at 20% RH.

Figure 7.6: Optical density over time of perovskite film in dipping solution. The signal at 700
nm was used to determine the growth rate of the perovskite, while the signal at 850 nm was used
to detect the presence of other optical phenomena (changes in scattering, reflection, incident light
intensity).
For the opaque electrode devices, 100 nm Au was thermally evaporated through a patterned
shadow mask to form the back electrode. For the semi-transparent devices, an AgNW film on PET
(details in following section). After transfer of the AgNW electrode, 100 nm Ag was thermally
evaporated through a patterned shadow mask around 3 edges of the devices. These bars of silver
helped reduce the unnecessary series resistance in the AgNW electrode by not limiting the current
collection to one geometrical direction. The nearly completed device was stored in a desiccator for
12 hours before applying the anti-reflective coatings. LiF anti-reflective coatings were then added to
the semi-transparent devices. 133 nm LiF was deposited onto the glass surface. This was optimized
to provide anti-reflection for the broad solar spectrum from 400-1100 nm. 176 nm LiF was deposited
onto the AgNW mesh. This thickness was optimized to provide anti-reflection for the infrared

CHAPTER 7. APPENDICES

67

spectrum from 800-1100 nm.
7.3.1.1

AgNW Stability
To make these devices more stable, we suggest developing a surface coating on the wires

to make them more stable. Our laboratory environment has normal oxygen levels as well as a small
partial pressure of sulfur gas added to the environment. While we have no direct proof of a coating
on the silver nanowires, we believe either an oxide or a sulfide layer was formed on the surface of the
wires. Silver oxide and sulfide are much more stable than metallic silver in the presence of iodine
(which readily forms silver iodide) and improve the resistance of the electrode to corrosion. We see
that electrodes stored in the laboratory environment prior to transfer onto a device are much more
stable than electrodes stored in an inert nitrogen atmosphere. We set aside one of these devices for
long-term testing and it showed no change in performance over a month of storing in the dark in
desiccated air. There is certainly room to improve on this procedure by intentionally introducing
an oxide or sulfide layer onto the nanowires or by coating the surface with other materials (e.g.,.
ligands, a gold coating, etc.)

7.3.2

Silicon Methodology
We fabricate our silicon cells from multicrystalline silicon wafers with high impurity content.

Instead of using expensive polysilicon made using the Siemens process, the feedstock for our cast
multicrystalline Si wafers are 4.5 N (99.995% pure) upgraded metallurgical silicon (UMG-Si) or recycled silicon from the top 10% of a cast multicrystalline industrial ingot (TI-Si). These ∼200 µm-thick
wafers are subjected to saw-damage removal wet etch in CP4 solution (15:5:2 HNO3 :CH3 COOH:HF)
for 2 cycles of 2 minute etch, resulting in

∼180

µm thick wafers. The wafers are then cleaned us-

ing RCA1 (5:1:1 H2 O:NH4 OH:H2 O2 ) and RCA2 (5:1:1 H2 O:HCl:H2 O2 ) solutions at 70 ◦ C, 10 min
each, to remove organic and metallic surface contaminants. The wafers are then loaded into a Tystar
POCl3 diffusion furnace at 700 ◦ C for phosphorus emitter formation. After ramping up the furnace
temperature to 865 ◦ C, POCl3 gas is flowed into the furnace for 12 minutes, followed by 6 minutes
of N2 purge. After waiting for an extra time of 10 minutes, we purge the furnace chamber with O2
for 7 minutes followed by temperature ramp down of ∼3 ◦ C/minute. The samples are then unloaded
at 500 ◦ C.
At the end of the diffusion process, phosphorus emitter with diffusion depth of 200-300
nm and emitter sheet resistance of 35 Ω/ are formed on both sides of the wafer. We remove the
phosphosilicate glass (PSG) layer formed on the wafers surface by dipping them in buffered oxide
etch (BOE 5:1) solution for approximately 30 s, then deposit SiNx anti-reflection coating (ARC)
using plasma-enhanced chemical vapor deposition (PECVD). In this intermediate step, the standard
Si cells have an ARC thickness of 82 nm, while the cells optimized for infrared light (800-1100 nm)
have an ARC thickness of ∼125 nm. We remove the backside emitter by SF6 reactive ion etch (RIE)

CHAPTER 7. APPENDICES

with a rate of

∼1

µm/min to etch

68

∼1

µm deep. After, we deposit 1 µm thick Al on the backside

of the wafer using e-beam evaporation, followed by rapid thermal annealing (RTA) at 900 ◦ C for
30 s in N2 atmosphere to form aluminum back surface field (Al BSF) at the back of the wafer.
After this annealing step, the thickness of the ARC layers shrink down to 78 nm (optimized for
standard AM1.5 illumination) and 115 nm (optimized for IR response). We defined our finger area
of the cells using photolithography, and etched SiNX on the corresponding area using RIE. We then
deposit 20/20/300 nm Ti/Pd/Ag metal stack as the finger, for which Ti acts as the adhesion layer,
Pd acts as the diffusion barrier, and Ag acts as the conductive layer. Lift-off process in acetone
is used to form the metal fingers, followed by RTA at 400 ◦ C for 5 minutes in Ar atmosphere for
metal adhesion. The cells are finally cut out of the wafers by laser scribing followed by mechanical
cleaving.

7.3.3

Measurement Procedures

7.3.3.1

IV Measurements
Current-voltage characteristics of the perovskite cells were measured using a Keithley model

2400 digital source meter. The irradiation source was a 300 W xenon lamp (Oriel). The lamp was
calibrated against the integrated photocurrent obtained by EQE. The voltage was swept in the
direction of open circuit to short circuit. A 5 s delay time at each voltage step before taking data
removed any transient hysteretic behavior of the perovskite devices. [37] For consistency, Si and
CIGS cells were measured with the same sweep parameters.
The semi-transparent perovskite was illuminated through a 0.39 cm2 aperture area. The
total area of the CIGS and Si is also 0.39 cm2 , to minimize leakage current. The opaque perovskite
cell was illuminated through an 0.12 cm2 aperture area.
7.3.3.2

Tandem Measurement Method
The EQE of the top cell, bottom cell, and bottom cell filtered by the top cell are individually

determined. The solar simulator is calibrated against the expected photocurrent of the cell from
the EQE measurement. A J-V curve returns the essential metrics of the cell’s performance. The
procedure is repeated for all three cases. In a 4-terminal configuration, the performance of the
individual cells is added together.
7.3.3.3

EQE Measurements
For the perovskites, the external quantum efficiency (EQE) was recorded as a function of the

wavelength using a Model SR830 DSP Lock-In Amplifier (Stanford Research Systems) without light
bias. A 100 W tungsten lamp (Newport) was used to provide an excitation beam, which was focused
through a Princeton Instruments SpectraPro 150 monochromator and chopped at approximately 2

CHAPTER 7. APPENDICES

69

Hz. At each wavelength, a delay time of 3 s was used to allow the signal to settle, and afterwards
data was collected for 4 s. The time constant on the lock-in amplifier was 1 s.
For the Si and CIGS cells, the EQE measurement was sped up because these cells have
much faster settling times than the perovskites. The excitation beam was chopped at 300 Hz, the
delay time was reduced to 0.1 s, data was collected for 1 s, and the time constant of the lock-in
amplifier was 30 ms. To measure the EQE of the bottom cell in the tandem, the above procedure
for the Si and CIGS cells was repeated with the perovskite top cell placed in front of the Si or CIGS
cell to filter the incoming light. The EQE and IV measurements were made immediately following
the evaporation of the LiF anti-reflective coatings.

7.3.4

Supplemental Figures

Figure 7.7: Transmission of the 12.4 Ω/ AgNW film with the PET substrate subtracted.

CHAPTER 7. APPENDICES

70

Figure 7.8: Distribution of semi-transparent cell performance gathered from 3 batches where device
procedures were largely kept constant between batches. Low-efficiency devices exhibited shorting
likely caused by too much pressure applied manually during the AgNW electrode transfer. Mediumefficiency devices generally exhibited low photocurrent likely caused by incomplete transfer of the
AgNW electrode due to too little pressure applied manually during the electrode transfer.

Figure 7.9: Example of difference in spectral response depending on AR coating applied to Si cell.
The regular TI-Si cell has a broadband AR coating.

CHAPTER 7. APPENDICES

7.4

71

Mechanically-Stacked Tandems Using ITO Experimental
Information

7.4.1

Perovskite Device Fabrication Method
We acquired 10 Ω/ ITO coated glass from Xin Yan Technology. Prior to device fabrica-

tion, 105 nm of MgF2 are thermally evaporated as an anti-reflection coating onto the glass. We find
low deposition rate to be important to prevent islanding. After cleaning, Al 4083 PEDOT:PSS from
Clevios is spun onto the ITO surface at 4000 rpm for 30 seconds and subsequently dried at 135 ◦ C
for 15 minutes. Next, a 40 wt% solution of 3:1 molar ratio methylammonium iodide (MAI, Dyesol)
to lead(ii) acetate (PbAc2 , Sigma) is dissolved in anhydrous n,n-dimethylformamide (DMF, Acros).
Hypophosphorous acid (HPA) from Sigma Aldrich is added to the solution with a molar ratio of
0.75 % HPA/PbAc2 . The solution of MAI and PbAc2 is spun at 2000 rpm in a dry air box with
>25 ppm water until a color change occurs from clear to light brown, at approximately 60 seconds.
The films are dried at room temperature until they are uniformly brown and then annealed at 100

C for 5 minutes in dry air. A 2 wt% solution of PCBM (Solenne bv) in anhydrous DCB is spun at

3000 rpm for 45 seconds and annealed for 5 minutes at 90 ◦ C. The samples are then taken out of the
glovebox and either AZO (Sigma) or ZnO (Sigma) nanoparticle inks with 15 nm average particle
diameter dispersed in IPA (2.5 wt%) are spun on at 4000 rpm. Two layers of AZO nanoparticles
are spun sequentially to produce an approximately 50 nm thick layer. The ZnO and AZO films are
dried at 75 ◦ C for 5 minutes.
For the opaque devices, 100 nm of Al and 125 nm of Ag were thermally evaporated through
a patterned shadow mask with a 0.2 cm2 aperture size as the back electrode. We observe an
improvement in air stability with the addition of the Ag layer.
For the semi-transparent devices, 500 nm of ITO is sputtered onto the device on the day
following perovskite deposition. For sputtering, we use a base pressure < 5x10−6 torr, deposition
pressure of 2x10−3 , a power density of 8 W/in2 , and 5 % oxygen partial pressure. The devices are
post-annealed at 100 ◦ C for 15 minutes. 300 nm of Ag was evaporated around the perimeter of the
device, leaving a 0.6 cm2 aperture size open. Finally, 150 nm of MgF2 is thermally evaporated as
an anti-reflection coating.

7.4.2

J-V Measurements
Current-voltage measurements were performed using a Keithley model 2400 digital source

meter and 300 W xenon lamp (Oriel) solar simulator was used for irradiation. Due to the coherent
reflections in the perovskite causing a significant mismatch between a KG5 reference diode and the
spectral response of the perovskite cell, the lamp was calibrated against the integrated photocurrent
calculated by EQE. J-V curves are taken from forward to reverse bias. Reported values for efficiency

CHAPTER 7. APPENDICES

72

were obtained by operating the cells at steady state max power and averaging over a minimum of
60 seconds.
The opaque cells and semi-transparent cells were illuminated through a 0.12 and 0.39 cm2
aperture area, respectively.

7.4.3

EQE Measurements
The external quantum efficiency (EQE) was recorded as a function of the wavelength using

a Keithley model 236 without light bias. A 100 W tungsten lamp (Newport) was used to provide an
excitation beam, which was focused through a Princeton Instruments SpectraPro 150 monochromator and chopped at approximately 0.6 Hz. At each wavelength, data was collected for and averaged
for 5 s.
To measure the EQE of the bottom cell in the tandem, the above procedure for the Si was
repeated with the perovskite top cell placed in front of the Si or CIGS cell to filter the incoming
light.

7.4.4

Stability Measurements
To test the thermal stability of the semi-transparent devices, the cells were placed on a

hot plate, glass side down, and illuminated from the top with a Newport Oriel model # 94023A
solar simulator and monitored using a Keithley model 2400 digital source meter. For opaque device
thermal stability, the devices were sandwiched between a second glass slide before being placed on
the hot plate to prevent scratching of the perovskite device, while still allowing light to enter through
the glass side.

7.4.5

Supplementary Figures

CHAPTER 7. APPENDICES

73

Figure 7.10: J-V Curves showing problems with the ZnO/ITO interface. At room temperature,
a large extraction barrier is present which greatly limits fill factor. This barrier is alleviated as
temperature is raised; however the barrier reappears upon cooling.

Figure 7.11: A 50 nm film of AZO nanoparticles displays high transmission. The AZO nanoparticles are annealed at either 75 or 150 ◦ C after deposition.

CHAPTER 7. APPENDICES

74

Figure 7.12: Transmission through an ITO/Glass reference sample from the same sputter deposition batch as the best devices. The 500 nm layer of ITO was annealed at 100 ◦ C for 15 minutes to
match the thermal processing profile of the perovskite samples and the resistivity dropped from 11
to 10 Ω/.

Figure 7.13: Comparison of the EQE of semi-transparent and opaque perovskite solar cells. The
lack of a metal back reflector in the semi-transparent solar cells results in lower photocurrent.

CHAPTER 7. APPENDICES

75

Figure 7.14: Transmission through the semi-transparent perovskite solar cell. Transmission is
limited at longer wavelengths due to coherent reflections.

Figure 7.15: Reflection of semi-transparent and opaque perovskite solar cells. Mitigating reflection
is an area of future work.

CHAPTER 7. APPENDICES

7.5

76

Cost-Modeling of Perovskite Solar Cells Model Information

7.5.1

Process flow for single-junction standard architecture

1. APCVD of 250 nm SnO2 :F
2. Wash front glass
3. Print 20 nm TiO2 compact layer. Print a mixture of ethanol and titanium isopropoxide using a
slot die coater. Uses 0.27 g titanium isopropoxide and 3.33 g ethanol per m2 of glass. Catalog
price for titanium isopropoxide is $0.0724 /g and catalog price for denatured ethanol is $0.0085
/g
4. Anneal TiO2 layer. 5 min anneal at 500 ◦ C
5. Print 500 nm perovskite layer. Print a 1 M stoichiometric solution of lead(ii) iodide (PbI2 )
and methylammonium iodide (MAI) in n,n-dimethylformamide (DMF) using a slot die coater.
Uses 2.94 g PbI2 , 1.02 g MAI, and 6 mL DMF per m2 of glass. Catalog price for PbI2 is $0.952
/g, catalog price for MAI is $0.892 /g, and catalog price for DMF is $0.02 /mL.
6. P1 laser scribe through SnO2 :F, TiO2 and perovskite layers to define cells
7. Insulator fill for cell isolation
8. Print 50 nm spiro-OMeTAD hole transport layer. Print a 72 mg/mL solution of spiro-OMeTAD
in chlorobenzene using a slot die coater. Uses 0.092 g of spiro-OMeTAD and 1.3 mL chlorobenzene per m2 of glass. Futures quote for bulk spiro-OMeTAD is $10 /g, catalog price for
chlorobenzene is $0.08/mL.
9. P2 laser scribe down to SnO2 :F to enable series connection
10. Deposit 1 µm aluminum.
11. P3 mechanical scribe down to spiro-OMeTAD and edge isolation for cell isolation
12. Electrically connect cells by solder-welding metal ribbon busbars and conductive adhesive tape
13. Feed bus bar ribbons through pre-drilled hole in tempered back glass. Bond the cells and the
busbar assembly to the back class with EVA and edge seal.
14. Solder the busbar ribbons to the junction-box leads, and bond the J-box to the back glass
with potting agent and tape.

CHAPTER 7. APPENDICES

77

15. Light soaking followed by high voltage isolation (Hi-Pot), ground continuity, and solar simulator JV testing.
16. Visual inspection and module binning and packing.

7.5.2

Tandem Performance Modeling

HIT-Si initial data acquired from Panasonic HIT VBHN285SJ40 285W panel datasheet. Quantum
efficiency represented from published wafer results [72], scaled to match the photocurrent density of
the panel. To determine the filtered performance, cell represented by the ideal diode equation with
the following parameters: J0 = 7.5E-11 A/cm2 , JP H = 38.54E-3 A/cm2 , RS = 0.36 Ω-cm, RSH =
1.4E10 Ω-cm, n = 1.4.
Perovskite absorption represented by published complex refractive index data [67]. Optical
effects include no interfacial reflections, 3 % reflection from top glass surface, 5 % absorption loss
broadband in each transparent electrode, no sub-bandgap absorption in the top cell, and single-pass
absorption in the perovskite. The published data on the 1.6 eV MAPbI3 perovskite is compressed in
energy space to approximate higher-bandgap perovskite materials. A perovskite thickness of 1000
nm is chosen to achieve high optical density in single-pass absorption. A perovskite bandgap of 1.74
eV is current matched with the HIT-Si cell. Perovskite sub-cell represented as 20 % efficiency and
is assumed to not suffer losses upon integration with the silicon sub-cells.

Figure 7.16: Modeled EQE of perovskite and HIT-Si cells.

Bibliography
[1] “Levelized Cost and Levelized Avoided Cost of New Generation Resources in the Annual Energy
Outlook 2015,” 2015.
[2] D. Chung, C. Davidson, R. Fu, K. Ardani, R. Margolis, D. Chung, C. Davidson, R. Fu, K. Ardani, and R. Margolis, “U.S. Photovoltaic Prices and Cost Breakdowns : Q1 2015 Benchmarks
for Residential , Commercial , and Utility-Scale Systems U.S. Photovoltaic Prices and Cost
Breakdowns : Q1 2015 Benchmarks for Residential , Commercial , and Utility-Scale Systems,”
no. September, 2015.
[3] M. Munsell, “Could First Solar’s Thin Film Beat Silicon PV on Efficiency?,” greentechmedia,
2014.
[4] “NREL Efficiency Chart Rev. 08-06-2015,”
[5] H. Liu, Z. Ren, Z. Liu, A. G. Aberle, T. Buonassisi, and I. M. Peters, “The realistic energy yield
potential of GaAs-on-Si tandem solar cells: a theoretical case study,” Optics Express, vol. 23,
p. A382, apr 2015.
[6] M. A. Green, “Silicon photovoltaic modules: a brief history of the first 50 years,” Progress in
Photovoltaics: Research and Applications, vol. 13, pp. 447–455, aug 2005.
[7] “International Technology Roadmap for Photovoltaic 2014 Results,” 2015.
[8] W. Shockley and H. J. Queisser, “Detailed Balance Limit of Efficiency of p-n Junction Solar
Cells,” Journal of Applied Physics, vol. 32, no. 3, p. 510, 1961.
[9] S. P. Bremner, M. Y. Levy, and C. B. Honsberg, “Analysis of Tandem Solar Cell Efficiencies
Under AM1.5G Spectrum Using A Rapid Flux Calculation Method,” Progress in Photovoltaics:
Research and Applications, no. November 2007, pp. 225–233, 2008.
[10] D. Weber, “CH3NH3PbX3, a Pb(II)-System with Cubic Perovskite Structure,” Zeitschrift f¨
ur
Naturforschung, vol. 33b, no. August 1978, pp. 1443–1445, 1978.

78

BIBLIOGRAPHY

79

[11] A. Kojima, K. Teshima, Y. Shirai, and T. Miyasaka, “Organometal halide perovskites as visiblelight sensitizers for photovoltaic cells.,” Journal of the American Chemical Society, vol. 131,
pp. 6050–1, may 2009.
[12] M. M. Lee, J. Teuscher, T. Miyasaka, T. N. Murakami, and H. J. Snaith, “Efficient Hybrid
Solar Cells Based on Meso-Superstructured Organometal Halide Perovskites,” Science, vol. 338,
no. November, pp. 643–647, 2012.
[13] H.-S. Kim, C.-R. Lee, J.-H. Im, K.-B. Lee, T. Moehl, A. Marchioro, S.-J. Moon, R. HumphryBaker, J.-H. Yum, J. E. Moser, M. Gr¨atzel, and N.-G. Park, “Lead Iodide Perovskite Sensitized
All-Solid-State Submicron Thin Film Mesoscopic Solar Cell with Efficiency Exceeding 9%,”
Scientific Reports, vol. 2, pp. 1–7, aug 2012.
[14] J. Burschka, A. Dualeh, F. Kessler, E. Baranoff, N.-L. Cevey-Ha, C. Yi, M. K. Nazeeruddin,
and M. Gr¨
atzel, “Tris(2-(1H-pyrazol-1-yl)pyridine)cobalt(III) as p-Type Dopant for Organic
Semiconductors and Its Application in Highly Efficient Solid-State Dye-Sensitized Solar Cells.,”
J. Am. Chem. Soc., vol. 133, pp. 18042–5, nov 2011.
[15] I. Chung, B. Lee, J. He, R. P. H. Chang, and M. G. Kanatzidis, “All-solid-state dye-sensitized
solar cells with high efficiency,” Nature, vol. 485, pp. 486–489, may 2012.
[16] J. Burschka, N. Pellet, S.-J. Moon, R. Humphry-Baker, P. Gao, M. K. Nazeeruddin, and
M. Gr¨
atzel, “Sequential deposition as a route to high-performance perovskite-sensitized solar
cells,” Nature, vol. 499, pp. 316–319, jul 2013.
[17] Z. Xiao, C. Bi, Y. Shao, and Q. Dong, “Efficient, High Yield Perovskite Photovoltaic Devices
Grown by Interdiffusion of Solution-Processed Precursor Stacking Layers,” Energy & . . . , 2014.
[18] Q. Chen, H. Zhou, Z. Hong, S. Luo, H.-S. Duan, H.-H. Wang, Y. Liu, G. Li, and Y. Yang,
“Planar Heterojunction Perovskite Solar Cells via Vapor-Assisted Solution Process,” Journal
of the American Chemical Society, vol. 136, pp. 622–625, dec 2013.
[19] M. Liu, M. B. Johnston, and H. J. Snaith, “Efficient planar heterojunction perovskite solar cells
by vapour deposition,” Nature, vol. 501, pp. 395–398, sep 2013.
[20] N. J. Jeon, J. H. Noh, Y. C. Kim, W. S. Yang, S. Ryu, and S. I. Seok, “Solvent engineering
for high performance inorganic-organic hybrid perovskite solar cells,” Nature Materials, vol. 13,
no. July, pp. 897–903, 2014.
[21] J. H. Noh, S. H. Im, J. H. Heo, T. N. Mandal, and S. I. Seok, “Chemical Management for
Colorful, Efficient, and Stable Inorganic-Organic Hybrid Nanostructured Solar Cells.,” Nano
letters, vol. 13, pp. 1764–1769, mar 2013.

BIBLIOGRAPHY

80

[22] G. E. Eperon, S. D. Stranks, C. Menelaou, M. B. Johnston, L. M. Herz, and H. J. Snaith,
“Formamidinium lead trihalide: a broadly tunable perovskite for efficient planar heterojunction
solar cells,” Energy & Environmental Science, vol. 7, no. 3, p. 982, 2014.
[23] G. E. Eperon, G. M. Patern`
o, R. J. Sutton, A. Zampetti, A. A. Haghighirad, F. Cacialli, and
H. J. Snaith, “Inorganic caesium lead iodide perovskite solar cells,” J. Mater. Chem. A, vol. 3,
no. 39, pp. 19688–19695, 2015.
[24] M. Kulbak, D. Cahen, and G. Hodes, “How Important Is the Organic Part of Lead Halide
Perovskite Photovoltaic Cells? Efficient CsPbBr 3 Cells,” The Journal of Physical Chemistry
Letters, vol. 6, pp. 2452–2456, jul 2015.
[25] N. Noel and S. Stranks, “Lead-Free Organic-Inorganic Tin Halide Perovskites for Photovoltaic
Applications,” Energy & . . . , 2014.
[26] C. Kim, S. Ryu, J. Seo, and S. I. Seok, “High-performance photovoltaic perovskite layers
fabricated through intramolecular exchange,” Science, no. May, pp. 1–8, 2015.
[27] J. P. Mailoa, C. D. Bailie, E. C. Johlin, E. T. Hoke, A. J. Akey, W. H. Nguyen, M. D. McGehee,
and T. Buonassisi, “A 2-terminal perovskite/silicon multijunction solar cell enabled by a silicon
tunnel junction,” Applied Physics Letters, vol. 106, p. 121105, 2015.
[28] P. Singh, R. Kumar, M. Lal, S. Singh, and B. Das, “Effectiveness of anisotropic etching of silicon
in aqueous alkaline solutions,” Solar Energy Materials and Solar Cells, vol. 70, pp. 103–113,
dec 2001.
[29] Y. Wang, X. Liao, Z. Ma, G. Yue, H. Diao, J. He, G. Kong, Y. Zhao, Z. Li, and F. Yun,
“Solid-phase crystallization and dopant activation of amorphous silicon films by pulsed rapid
thermal annealing,” Applied Surface Science, vol. 135, pp. 205–208, sep 1998.
[30] N. M. M. F. Vilela, A. Freundlich, A. Bensaoula, “High performance In0.53Ga0.47As tunnel
junctions grown by Chemical Beam Epitaxy,” Proc. 5th Int. Conf. Indium Phosphide Relat.
Mater, vol. 227, 1993.
[31] Niu Jin, Sung-Yong Chung, A. Rice, P. Berger, P. Thompson, C. Rivas, R. Lake, S. Sudirgo,
J. Kempisty, B. Curanovic, S. Rommel, K. Hirschman, S. Kurinec, P. Chi, and D. Simons,
“Diffusion barrier cladding in Si/SiGe resonant interband tunneling diodes and their patterned growth on PMOS source/drain regions,” IEEE Transactions on Electron Devices, vol. 50,
pp. 1876–1884, sep 2003.
[32] C. D. Bailie, M. G. Christoforo, J. P. Mailoa, A. R. Bowring, E. L. Unger, W. H. Nguyen,
J. Burschka, N. Pellet, J. Z. Lee, M. Gr¨atzel, R. Noufi, T. Buonassisi, A. Salleo, and M. D.

BIBLIOGRAPHY

81

McGehee, “Semi-transparent Perovskite Solar Cells for Tandems with Silicon and CIGS,” Energy & Environmental Science, vol. 8, pp. 956–963, dec 2015.
[33] G. J. Bauhuis, P. Mulder, and J. J. Shermer, “Ultra-thin, high performance tunnel junctions
for III-V multijunction cells,” Progress in Photovoltaics: Research and Applications, vol. 22,
pp. 656–660, 2014.
[34] S. Avasthi, W. E. McClain, G. Man, A. Kahn, J. Schwartz, and J. C. Sturm, “Hole-blocking
titanium-oxide/silicon heterojunction and its application to photovoltaics,” Applied Physics
Letters, vol. 102, no. 20, p. 203901, 2013.
[35] J. Jhaveri, S. Avasthi, K. Nagamatsu, and J. C. Sturm, “Stable low-recombination nSi/TiO¡inf¿2¡/inf¿ hole-blocking interface and its effect on silicon heterojunction photovoltaics,”
in 2014 IEEE 40th Photovoltaic Specialist Conference (PVSC), pp. 1525–1528, IEEE, jun 2014.
[36] W. H. Nguyen, C. D. Bailie, E. L. Unger, and M. D. McGehee, “Enhancing the HoleConductivity of Spiro-OMeTAD without Oxygen or Lithium Salts by Using Spiro(TFSI) 2 in
Perovskite and Dye-Sensitized Solar Cells,” Journal of the American Chemical Society, vol. 136,
pp. 10996–11001, aug 2014.
[37] E. L. Unger, E. T. Hoke, C. D. Bailie, W. H. Nguyen, A. R. Bowring, T. Heum¨
uller, M. G.
Christoforo, and M. D. McGehee, “Hysteresis and transient behavior in currentvoltage measurements of hybrid-perovskite absorber solar cells,” Energy Environ. Sci., vol. 7, pp. 3690–3698,
aug 2014.
[38] M. A. Green, K. Emery, Y. Hishikawa, W. Warta, and E. D. Dunlop, “Solar cell efficiency tables
(Version 45),” Progress in Photovoltaics: Research and Applications, vol. 23, pp. 1–9, jan 2015.
[39] P. Loper, B. Niesen, S.-J. Moon, S. Martin de Nicolas, J. Holovsky, Z. Remes, M. Ledinsky, F.-J.
Haug, J.-H. Yum, S. De Wolf, and C. Ballif, “Organic-Inorganic Halide Perovskites: Perspectives
for Silicon-Based Tandem Solar Cells,” IEEE Journal of Photovoltaics, vol. 4, pp. 1545–1551,
nov 2014.
[40] E. T. Hoke, D. J. Slotcavage, E. R. Dohner, A. R. Bowring, H. I. Karunadasa, and M. D.
McGehee, “Reversible photo-induced trap formation in mixed-halide hybrid perovskites for
photovoltaics,” Chem. Sci., vol. 6, pp. 613–617, nov 2015.
[41] M. W. Rowell and M. D. McGehee, “Transparent electrode requirements for thin film solar cell
modules,” Energy Environ. Sci., vol. 4, no. 1, pp. 131–134, 2011.
[42] J.-Y. Lee, S. T. Connor, Y. Cui, and P. Peumans, “Solution-Processed Metal Nanowire Mesh
Transparent Electrodes,” Nano Letters, vol. 8, pp. 689–692, feb 2008.

BIBLIOGRAPHY

82

[43] Z. M. Beiley, M. G. Christoforo, P. Gratia, A. R. Bowring, P. Eberspacher, G. Y. Margulis,
C. Cabanetos, P. M. Beaujuge, A. Salleo, and M. D. McGehee, “Semi-Transparent Polymer
Solar Cells with Excellent Sub-Bandgap Transmission for Third Generation Photovoltaics,”
Advanced Materials, vol. 25, pp. 7020–7026, dec 2013.
[44] G. Y. Margulis, M. G. Christoforo, D. Lam, Z. M. Beiley, A. R. Bowring, C. D. Bailie, A. Salleo,
and M. D. McGehee, “Spray Deposition of Silver Nanowire Electrodes for Semitransparent
Solid-State Dye-Sensitized Solar Cells,” Advanced Energy Materials, vol. 3, pp. 1657–1663, dec
2013.
[45] X. Wu, J. Zhou, A. Duda, J. C. Keane, T. Gessert, Y. Yan, and R. Noufi, “13.9%-efficient
CdTe polycrystalline thin-film solar cells with an infrared transmission of ˜50%,” Progress in
Photovoltaics: Research and Applications, vol. 14, pp. 471–483, sep 2006.
[46] T. J. Coutts, K. a. Emery, and J. Scott Ward, “Modeled performance of polycrystalline thin-film
tandem solar cells,” Progress in Photovoltaics: Research and Applications, vol. 10, pp. 195–203,
may 2002.
[47] M. Contreras, J. Tuttle, A. Gabor, A. Tennant, K. Ramanathan, S. Asher, A. Franz, J. Keane,
L. Wang, J. Scofield, and R. Noufi, “High efficiency Cu(In,Ga)Se2-based solar cells: processing
of novel absorber structures,” in Proceedings of 1994 IEEE 1st World Conference on Photovoltaic Energy Conversion - WCPEC (A Joint Conference of PVSC, PVSEC and PSEC), vol. 1,
pp. 68–75, IEEE, 1994.
[48] “US Patent No. 5,441,897 (15 August 1995) and US Patent No. 5,436,204 (25 July 1995).”
[49] M. A. Contreras, M. J. Romero, B. To, F. Hasoon, R. Noufi, S. Ward, and K. Ramanathan,
“Optimization of CBD CdS process in high-efficiency Cu(In,Ga)Se2-based solar cells,” Thin
Solid Films, vol. 403-404, pp. 204–211, feb 2002.
[50] K. Ramanathan, M. a. Contreras, C. L. Perkins, S. Asher, F. S. Hasoon, J. Keane, D. Young,
M. Romero, W. Metzger, R. Noufi, J. Ward, and A. Duda, “Properties of 19.2% efficiency
ZnO/CdS/CuInGaSe2 thin-film solar cells,” Progress in Photovoltaics: Research and Applications, vol. 11, pp. 225–230, jun 2003.
[51] “International Technology Roadmap for Photovoltaic (ITRPV) 2013 Results,” Tech. Rep.
March, 2014.
[52] Y. Kato, L. K. Ono, M. V. Lee, S. Wang, S. R. Raga, and Y. Qi, “Silver Iodide Formation in
Methyl Ammonium Lead Iodide Perovskite Solar Cells with Silver Top Electrodes,” Advanced
Materials Interfaces, vol. 2, sep 2015.

BIBLIOGRAPHY

83

[53] K. Ellmer, “Past achievements and future challenges in the development of optically transparent
electrodes,” Nature Photonics, vol. 6, no. 12, pp. 808–816, 2012.
[54] B. Henry, A. Erlat, A. McGuigan, C. Grovenor, G. Briggs, Y. Tsukahara, T. Miyamoto,
N. Noguchi, and T. Niijima, “Characterization of transparent aluminium oxide and indium
tin oxide layers on polymer substrates,” Thin Solid Films, vol. 382, pp. 194–201, feb 2001.
[55] P. L¨
oper, S.-J. Moon, S. Mart´ın de Nicolas, B. Niesen, M. Ledinsky, S. Nicolay, J. Bailat,
J.-H. Yum, S. De Wolf, and C. Ballif, “Organicinorganic halide perovskite/crystalline silicon
four-terminal tandem solar cells,” Phys. Chem. Chem. Phys., vol. 17, pp. 1619–1629, dec 2015.
[56] K. A. Bush, C. D. Bailie, Y. Chen, T. Leijtens, A. R. Bowring, F. Moghadam, and M. D.
McGehee, “Thermal and Environmental Stability of Semi-Transparent Perovskite Solar Cells
for Tandems Enabled by a Sputtered ITO Electrode,” Advanced Materials, Submitted, 2015.
[57] C. Bi, Q. Wang, Y. Shao, Y. Yuan, Z. Xiao, and J. Huang, “Non-wetting surface-driven highaspect-ratio crystalline grain growth for efficient hybrid perovskite solar cells,” Nature Communications, vol. 6, p. 7747, jul 2015.
[58] W. Zhang, M. Saliba, D. T. Moore, S. K. Pathak, M. T. H¨orantner, T. Stergiopoulos, S. D.
Stranks, G. E. Eperon, J. a. Alexander-Webber, A. Abate, A. Sadhanala, S. Yao, Y. Chen, R. H.
Friend, L. a. Estroff, U. Wiesner, and H. J. Snaith, “Ultrasmooth organicinorganic perovskite
thin-film formation and crystallization for efficient planar heterojunction solar cells,” Nature
Communications, vol. 6, p. 6142, jan 2015.
[59] F. Guo, H. Azimi, Y. Hou, T. Przybilla, M. Hu, C. Bronnbauer, S. Langner, E. Spiecker,
K. Forberich, and C. J. Brabec, “High-performance semitransparent perovskite solar cells with
solution-processed silver nanowires as top electrodes,” Nanoscale, vol. 7, no. 5, pp. 1642–1649,
2015.
[60] J. You, L. Meng, T.-B. Song, T.-F. Guo, Y. M. Yang, W.-H. Chang, Z. Hong, H. Chen, H. Zhou,
Q. Chen, Y. Liu, N. De Marco, and Y. Yang, “Improved air stability of perovskite solar cells
via solution-processed metal oxide transport layers,” Nature Nanotechnology, pp. 1–8, oct 2015.
[61] J. Yang, B. D. Siempelkamp, D. Liu, and T. L. Kelly, “Investigation of CH 3 NH 3 PbI
3 Degradation Rates and Mechanisms in Controlled Humidity Environments Using in Situ
Techniques,” ACS Nano, vol. 9, pp. 1955–1963, feb 2015.
[62] J. A. Christians, P. A. Miranda Herrera, and P. V. Kamat, “Transformation of the Excited
State and Photovoltaic Efficiency of CH 3 NH 3 PbI 3 Perovskite upon Controlled Exposure to
Humidified Air,” Journal of the American Chemical Society, vol. 137, pp. 1530–1538, feb 2015.

BIBLIOGRAPHY

84

[63] A. M. A. Leguy, Y. Hu, M. Campoy-Quiles, M. I. Alonso, O. J. Weber, P. Azarhoosh, M. van
Schilfgaarde, M. T. Weller, T. Bein, J. Nelson, P. Docampo, and P. R. F. Barnes, “Reversible
Hydration of CH 3 NH 3 PbI 3 in Films, Single Crystals, and Solar Cells,” Chemistry of
Materials, vol. 27, pp. 3397–3407, may 2015.
[64] T. Leijtens, G. E. Eperon, N. K. Noel, S. N. Habisreutinger, A. Petrozza, and H. J. Snaith,
“Stability of Metal Halide Perovskite Solar Cells,” Advanced Energy Materials, vol. 5, oct 2015.
[65] B. Conings, J. Drijkoningen, N. Gauquelin, A. Babayigit, J. D’Haen, L. D’Olieslaeger, A. Ethirajan, J. Verbeeck, J. Manca, E. Mosconi, F. D. Angelis, and H.-G. Boyen, “Intrinsic Thermal
Instability of Methylammonium Lead Trihalide Perovskite,” Advanced Energy Materials, vol. 5,
aug 2015.
[66] X. Li, M. Tschumi, H. Han, S. S. Babkair, R. A. Alzubaydi, A. A. Ansari, S. S. Habib, M. K.
Nazeeruddin, S. M. Zakeeruddin, and M. Gr¨atzel, “Outdoor Performance and Stability under
Elevated Temperatures and Long-Term Light Soaking of Triple-Layer Mesoporous Perovskite
Photovoltaics,” Energy Technology, vol. 3, no. 6, pp. 551–555, 2015.
[67] P. L¨
oper, M. Stuckelberger, B. Niesen, M. Filipic, S.-j. Moon, J.-h. Yum, M. Topic, S. D. Wolf,
and C. Ballif, “Complex Refractive Index Spectra of CH3NH3PbI3 Perovskite Thin Films Determined by Spectroscopic Ellipsometry and Spectrophotometry,” Journal of Physical Chemistry
Letters, vol. 6, pp. 66–71, 2015.
[68] K. Masuko, M. Shigematsu, T. Hashiguchi, D. Fujishima, M. Kai, N. Yoshimura, T. Yamaguchi, Y. Ichihashi, T. Mishima, N. Matsubara, T. Yamanishi, T. Takahama, M. Taguchi,
E. Maruyama, and S. Okamoto, “Achievement of More Than 25% Conversion Efficiency With
Crystalline Silicon Heterojunction Solar Cell,” IEEE Journal of Photovoltaics, vol. 4, pp. 1433–
1435, nov 2014.
[69] “SunShot Vision Study,” Tech. Rep. February, 2012.
[70] R. Jones-Albertus, D. Feldman, K. Horowitz, and M. Woodhouse, “Technology Advances
Needed for Photovoltaics to Achieve Widespread Grid Price Parity,” Progress in Photovoltaics,
Submitted, 2015.
[71] C. D. Bailie, L. L. Tinker, J. P. Mailoa, T. Buonassisi, M. Woodhouse, and M. D. McGehee, “Modeling the cost of metal-halide perovskite and perovksite/silicon tandem photovoltaic
modules,” Energy & Environmental Science, Manuscript in Preparation.
[72] M. Taguchi, A. Yano, S. Tohoda, K. Matsuyama, Y. Nakamura, T. Nishiwaki, K. Fujita, and
E. Maruyama, “24.7% Record efficiency HIT solar cell on thin silicon wafer,” IEEE Journal of
Photovoltaics, vol. 4, no. 1, pp. 96–99, 2014.

Master your semester with Scribd & The New York Times

Special offer for students: Only $4.99/month.

Master your semester with Scribd & The New York Times

Cancel anytime.