This document is the unedited Author’s version of a Submitted Work that was subsequently accepte
d for
publication in
A
C
S
N
a
n
o
, copyright © American Chemical Society after peer review.
To access the final edited
and published work see
http://pubs.acs.org/doi/full/10.1021/acs
nano.7b03148
1
High Photovoltaic Quantum Efficiency in Ultrathin van der Waals
Heterostructures
Joeson Wong
1
ǂ
,
Deep Jariwala
1,2
ǂ
,
Giulia Tagliabue
1,4
,
Kevin Tat
1
,
Artur R. Davoyan
1,2,3
,
Michelle C.
Sherrott
1,2
and
Harry A. Atwater
1,2,3,4
*
1
Department of Applied Physics and Material
s
Science, California Institute of Technology,
Pasadena, CA
-
91125, USA
2
Resnick Sustainability Institute, California Institute of Technology, Pasadena, CA
-
91125,
USA
3
Kavli
Nanoscience Institute, California Institute of Technology, Pasadena, CA
-
91125, USA
4
Joint Center for Artificial Photosynthesis, California Institute of Technology, Pasadena,
CA
-
91125, USA
* Corresponding author: Harry A Atwater (
haa@caltech.edu
)
ǂ
These authors contributed equally
A
BSTRACT
:
We report
experimental
measurements for
ultrathin
(< 15 nm)
van der Waals
heterostructures
exhibit
ing
external quantum
efficiencies
exceeding
50%
,
and show that these
structures
can achieve
experimental
absorbance > 90%.
By coupling
electromagnetic simulations
and experimental measurements
, w
e
show
that pn
WSe
2
/MoS
2
heterojunction
s
with vertical
carrier collection can have
internal photocarrier
collection efficiencies exceeding 70
%.
GRAPHICAL TABLE OF CONTENTS:
K
EYWORDS
:
van der Waals, heterojunction, photovoltaics,
quantum efficiency,
high efficiency,
MoS
2
, WSe
2
This document is the unedited Author’s version of a Submitted Work that was subsequently accepte
d for
publication in
A
C
S
N
a
n
o
, copyright © American Chemical Society after peer review.
To access the final edited
and published work see
http://pubs.acs.org/doi/full/10.1021/acs
nano.7b03148
2
Owing to their naturally passivated
basal planes
and
strong light
-
matter interactions,
transition metal
dichalcogenide
s
are of considerable interest
as
active elem
ents of optoelectronic
devices such as
light
-
emitting devices, photodetectors and
photovoltaic
s.
1,2
Ultrathin
transition
metal dichalcogenide
(TMD)
photovoltaic devices a few atom
ic layer
s in thickness
have been
realized
using
TMDs
such as
molybdenum disulfide (
MoS
2
)
and
tungsten diselenide (
WSe
2
).
3
–
8
C
omplete absorption of the solar spectrum is a challenge as the
thickness is
reduced to the
ultrathin limit
,
9
–
11
whereas efficient carrier collection
is challenging in thicker bulk TMD crystals.
The active layers i
n
conventional
photovoltaic
s
typically
range
from
a few
microns in direct gap
materials (
gallium arsenide
)
to a
hundred
microns thick
or more in indirect
gap materials
(silicon
)
.
12
Efficient ultrathin
and ultralight
(<100 g/m
2
)
photovoltaics ha
ve
long been sought for
many applications where weight and flexibility are important design considerations, such
as
applications
in
space
power systems, internet
-
of
-
things devices,
as well as
portable
and flexible
electronics
.
13
–
15
C
onventional photovoltaic materials are
mechanically
fragile
when
thinned
down
to the ultrathin (< 10 nm) regime, and interfacial reactions mean that a large fraction of
the crystal consists
of surface
-
modified regions rather than intrinsically bulk material. Surface
oxides and dangling bonds in ultrathin films often result in increased nonradiative recombination
losses, lowering photovoltaic efficiencies.
By contrast,
transition metal dicha
lcogenide
s
have
intrinsically high absorption and their layered crystallographic structures suggest the possibility
of achieving intrinsically passive basal planes in high quality crystals.
P
hotovoltaic
s
that
can
approach
the S
ho
c
kley
-
Queisser limit
,
16,17
ha
ve
two prerequisites:
first
,
that at open circuit, every
above
-
bandgap
photon
that
is
absorbed
is
extracted
as a
n
emitted
photon at the
band
-
ed
ge of the material,
i.e.
it has perfect external radiative efficiency
.
18
Amani
et al
.
ha
ve recently
dem
onstrated that
superacid
-
treated monolayers of MoS
2
and WS
2
exhibit
internal radiativ
e efficiency
> 99%
,
19
suggesting that
th
e
condition
of very high external
radiative
efficiency might
be
satisfied
in transition metal
di
chalcogenides
.
The second
prerequisite
is
that
at
short circuit, the photovoltaic device
must convert every incident
above
-
bandgap
photon into
an extracted electron,
i.e.
it has external quantum
efficiency (EQE)
approaching unity
.
To understand the path to high EQE, we can deco
n
volute
the
external quantum efficiency
into
the product of two terms:
the
absorbance
and internal quantum efficiency (IQE).
H
igh EQE
device
s
exhibit both high absorption and
internal quantum efficiency
,
i.e.
,
carrier generation and
collection efficiency per absorbed photon
.
T
o date
, reports
of
van der Waals
based photovoltaic
devices have not considered
both of
these
concepts
and
separately
ev
aluate
d
them as
criteria
for high efficiency photovoltaics.
Coupling
electromagnetic simulations with absorption and EQE measurements
enables
quantitative
characterization of
few
-
atomic
-
layer thickness optoelectronic devices
in
van der
Waals heterostructures.
In this paper,
we dem
onstrate
external quantum efficiencies > 50%
(Figure 1(a)), indicating
that van der Waals heterostructures
have
considerable
potential
for
efficient photovoltaic
s
.
We show that
high EQE
result
s from
bot
h
high
optical
absorption
and
This document is the unedited Author’s version of a Submitted Work that was subsequently accepte
d for
publication in
A
C
S
N
a
n
o
, copyright © American Chemical Society after peer review.
To access the final edited
and published work see
http://pubs.acs.org/doi/full/10.1021/acs
nano.7b03148
3
efficient
electronic
charge carrier collection
.
We
analyze
the optical response using
electromagnetic
simulations
to
expla
in
how near
-
unity absorption
can be
achieved
in
heterostructures (Figure 1(b))
.
W
e find that experimental absorption results
for
van der Waals
heterostructures match well
with
these
electromagnetic simulations. Thus
we can
separate
optical
absorption
and electronic
transport
to quantitative
ly
compare
the
ir
effects
on charge
collection
efficiency
for
both
pn
heterojunction
s
and
Schottky junction
s
(Figure 1 (c))
.
In addition,
we
analyze the
role
of
few
-
layer graphene as a transparent top contact
(Figure 1 (d)).
Finally, w
e
outline
important considerations
for designing
high
efficiency
photovoltaic devices
.
B
y
simultaneously maximizing both external radiative efficiency and external quantum efficiency in
a single device, van der Waals materials
based photovoltaic devices
could in principle achieve
efficiencies close to
the S
hockley
-
Quei
sser limit
for their bandgaps
.
RESULTS AND DISCUSSION
Optoelectronic device characteristics
We
analyzed the optoelectronic device characteristics of a
high
-
performance
device
consisting of a
vertical van der Waals heterostructure device of 0.6 nm
thick
few
-
layer graphene
(
FLG
)
/9 nm WSe
2
/3 nm MoS
2
/Au
(see
Supporting
Information S1 for optical and photocurrent
images)
.
It
s optoelectronic and device characteristics are shown in Figure 2. First, we find that
this
device
exhibit
s an
EQE > 50% (
Figure 2 (a)) w
ith absorbance greater than 90% from
approximately
500 nm to 600 nm. Spectral features such as the exciton resonances of MoS
2
and
WSe
2
are well reproduced in the external quantum efficiency
spectrum.
In addition, w
e observe
a
maximum
single
-
wavelength powe
r conversion efficiency
(PCE)
of 3.4% under
740 W/cm
2
of
633 nm
laser
illumination
(Figure 2
(b))
.
Since
the high
-
performance device is
electrically in
parallel with other devices, typical
macroscopically large spot size (
~
cm) AM 1.5G
illumination
measurements wou
ld yield device
characteristics
substantially different from the high
-
performing one
. Thus, w
e estimate
d
the AM 1.5G
performance
using
extracted device
parameters
of
a diode fit
under laser illumination
(
see
Supporting
Information S2
for de
tails).
We
estimate the AM 1.5G PCE of this device to be
~
0
.
4%
.
This value is
presently too
low to be useful
for photovoltaics, but the high EQE values reported here indicate promise for high efficiency
devices, when device engineering efforts are able to also achieve correspondingly high open
circuit voltages in
van der Waals
based photovoltaic
s.
Further measurements were performed
at different laser powers
under 633
nm laser
illumination (Figure 3
), yielding various power
-
dependent characteristics. Examination of the
short
-
circuit current
퐼
푠푐
yielded nearly linear dependence on laser power,
as expected in ideal
photovoltaic devices. The dashed blue line represents the fit to the expression
퐼
푠푐
=
퐴
푃
휏
, where
퐴
is a constant of proportionality
,
푃
is the incident power,
and
휏
represents the degree of
nonlinearity in this device (
휏
=
1
is t
he linear case)
.
20
W
e find that
휏
=
0
.
98
in our device,
indicating nearly linear behavior under short circuit condi
tions.
In addition, in an ideal
photovoltaic device, the open circuit voltage is expected to grow logarithmically with the input
This document is the unedited Author’s version of a Submitted Work that was subsequently accepte
d for
publication in
A
C
S
N
a
n
o
, copyright © American Chemical Society after peer review.
To access the final edited
and published work see
http://pubs.acs.org/doi/full/10.1021/acs
nano.7b03148
4
power, since
푉
표푐
=
푛
푘
푏
푇
푞
ln
(
퐽
퐿
퐽
푑푎푟푘
+
1
)
≈
푛
푘
푏
푇
푞
ln
(
퐽
퐿
퐽
푑푎푟푘
)
for large illumination current densities
퐽
퐿
. Here,
퐽
푑푎푟푘
is
the dark current density
,
푛
is the ideality factor,
푘
푏
is the Boltzmann constant,
푇
is the temperature of the device, and
푞
is the fundamental unit of charge, so that
푘
푏
푇
푞
≈
0
.
0258
푉
at room temperature.
In
Figure 3 (b)
we see
that the exp
erimental data match well
with the diode
fit
(dashed black line, see
Supporting
Information S2
for fitting details)
, suggest
ing
an ideality factor of
푛
=
1
.
75
and a dark current density
퐽
푑푎푟푘
=
0
.
65
mA/(cm
2
) assuming a
30
휇
m
×
30
휇
m
device
area
.
Also,
since the power conversion efficiency (PCE) is given as
푃퐶퐸
=
퐽
푠푐
푉
표푐
퐹퐹
/
푃
푖푛
,
where
퐽
푠푐
is the short circuit current density,
푉
표푐
is the open circuit voltage,
퐹퐹
is
the fill fraction, and
푃
푖푛
in the incident power density,
we would expect the power conversion
efficiency to scale
roughly
logarithmically
as well
. This is
true
for
laser powers up to
~
7
40
W
/
c
m
2
(
Figure 3 (c)
). However,
for larger
input power,
the PCE decreases with increasing power
. Such a
drop in PCE can be
attributed to series resistance
s
in the device, either at the contacts or
at
the
junction. This
is corroborated by the match
between the experimental data (dots) and the
fitted
expression (
dashed line
)
, yielding the
diode
fitting parameters in the lower ri
ght ha
nd corner of
the plot
in Figure 3
(c).
The fit for the
푉
표푐
was simultaneously done with the PCE
, therefore
yielding the same set of parameters and a good match between experiment and extracted device
parameters.
Finally, we observed a decrease in
the EQE at 633
nm with increasing power
(Figure
3 (d
))
. Using the above fitted parameters, series resistance can
only
be used to partially explain
a decrease in the EQE at higher powers.
Thus, the
additional
decrease in EQE at hi
gher powers
may be due to
the onset of
carrier
density
-
dependent
non
radiative processes such as Aug
er or
biexcitonic recombination
which are
not accounted for in the diode fit used above
, where the
dark current is fixed for all powers
.
Absorption in van der Waals heterostructur
es
We first
investigat
e
the
absorption
and optical properties of
van der Waals
heterostructures
.
We formed a heterostructure
composed of
hexagonal boron nitride (
hBN
)
/
FLG/WSe
2
/
MoS
2
/
Au
. The
composite
heterostructure
has various regions (
inset of
Figure 4
(a)),
corresponding to different vertical heterostructures.
Given the sensitivity of the performance of
van der Waals materials to different environmental conditions and device fabrication
procedures
,
21
t
h
e samples fabricated here allow us to study optical and electronic
features of
different
heterostructures in a systematic
manner
by probing specific
heterostructures
fabricated on the same
monolithic substrate
.
This is enabled by the small spot size of our laser,
which additionally allows us to properly normalize the spectral response without artificially
including geometric
factors (see Methods for details).
As an example, consider the optical
response
at
the
loca
tion of the blue dot in
the inset
of
Figure 4
(a
). T
he vertical heterostructure
there
is composed of 1.5 nm FLG/
4 nm WSe
2
/
5 nm
MoS
2
/Au. This location can be probed
spectrally for its abso
rption characteristics (Figure 4
(a
)),
This document is the unedited Author’s version of a Submitted Work that was subsequently accepte
d for
publication in
A
C
S
N
a
n
o
, copyright © American Chemical Society after peer review.
To access the final edited
and published work see
http://pubs.acs.org/doi/full/10.1021/acs
nano.7b03148
5
revealing near
-
unity absorption in van der Waals heterostructures.
The peaks at
~
610,
~
670,
and
~
770 nm correspond to the
resonant excitation
of
the
MoS
2
B
exciton
, MoS
2
A
exciton
, and
WSe
2
A exciton, respectively
.
22
On the other hand, the
broad
mode at
~
550 nm corresponds to
the photonic mode that leads to near
-
unity absorption
.
23
Measurements of the a
bsorption can
be corroborated with
electromagnetic simul
ations
, unveiling
both the accuracy between
simulation and experiment
al
results as well as the fraction of photon flux absorbed into individual
layers of the
heterostructure stack (Figure 4
(b
)).
Despite the near
-
unity absorption observed in
the heterostructure
stack
, there is parasitic absorption in both the underlying gold substrate and
in the few
-
layer graphene that accounts for
~
20% of the total absorbance. Such parasitic
absorption can be red
uced
by using a silver back
reflector, as shown in Figure 4
(c) and Figure 4
(d)
.
W
e
find that the simulated and measured absorbance is
also
in good agreement
for the case
of a silver back
reflector
. Thus, the optical response of a van der Waals heterostruc
ture can be
modelled accurately using
full wave
electromagnetic simulations
and
our method of
measurement yields accurate and reliable results
.
To note,
the
s
ubwavelength dimension of the
total
heterostructure
thickness is
critical for
achieving near
-
unity absorption. Indeed
,
the entire stack can be treated as a single effective
medium, where
small
phase shifts are present between layers and therefore the material
discontinuities are
effectively imperceptible to the incident
light
(
see
Supporting
Infor
mation S3
for details).
Ultimately, the v
an der Waals heterostructure
-
on
-
metal
behaves a
s a
si
n
g
le
absorbing
material with effective medium optical properties
. T
herefore, as previously
demonstrated, near
-
unity absorption
at different wavelengths can be
ac
hieved for a
semiconducting layer with
the appropriate thickness
23,24
(
~
10
-
15
nm total thickness
for TMD
heterostructures)
.
C
arrier collection in
van der Waals
semiconductor junctions
As discussed
above
,
another criterion for high EQE is
efficient carrier collection.
Given the
large
exciton binding energies in TMDs (
~
50
-
100 meV in the bulk)
,
25,26
the
large
internal electric
field
at the
semiconductor
hetero
junction may play a role in exciton dissociation and subsequent
c
arrier collection.
C
harge carrie
r separation in TMDs
can be accomplished using either a
pn
junction
or a
Schottky junction
, and w
e find t
hat a pn
heterojunction dramatically enhances the
E
QE
when compared with a Schottky junction.
The heterostructure
described in Figure 4
(a) and (b)
can
be probed as
an optoelectronic
device with the formation of a top
electrode (see inset of Figure 5
(a)).
Since t
he back
reflector
(gold)
can
simultaneously
serve as a back contact to the entire vertical heterostructure
,
we can
use this scheme
to
compare
the electronic performance of
various vertical heterostructures
.
G
iven the work
function between WSe
2
(p
-
type)
and Au, it
is expected that a Schottky
junction
27
will form between the two materials (See Figure 1 (c)),
whereas
WSe
2
(p
-
type)
on top of
MoS
2
(n
-
type)
is expected to form a pn
heterojunction
.
4
High spati
al resolution scanning photocurrent
This document is the unedited Author’s version of a Submitted Work that was subsequently accepte
d for
publication in
A
C
S
N
a
n
o
, copyright © American Chemical Society after peer review.
To access the final edited
and published work see
http://pubs.acs.org/doi/full/10.1021/acs
nano.7b03148
6
microscopy allows us to examine the
two
heterostructure
devices in detail (Figure 5
(a)). We
observe
large
photocurrent
for
the
pn
hetero
junction geometry
(yellow region)
compared to the
Schottky junction geometry (light blue region). The decrease of the photocurrent in the
left
-
side
of the
ye
llow region
in Fig
ure
5(a)
is due to shadowing
from the electrical probes. A line cut of
the spa
tial photocurrent map
shown in
Figu
re 5
(b)
provides
a clearer distinction between the
two junctions, demonstrating
~
6x more photocurrent for the pn
junction
relative
to the Schottky
junction.
The
photocurrent
density
is directly related to the external quantum efficiency and
therefore
the
product of the
absorbance
and IQE.
I
n order to quantitatively compare the
electronic
differences between the two junctions
,
we need to normalize out the different optical
absorption in the two devices,
i.e.
compute the IQE of each device
퐼푄
퐸
퐸푥푝
(
휆
)
=
퐸
푄퐸
(
휆
)
퐴푏푠
(
휆
)
(
1
)
where
퐸푄퐸
(
휆
)
and
퐴푏푠
(
휆
)
are the e
xperimentally measured EQE and a
bsorbance of their
respective devices
(Figure 5 (c) (i) and Figure 5
(d) (i))
. A plot of the experimentally derived
IQE
(
i.e.
퐼푄
퐸
퐸푥푝
)
is shown in
purple in
Figure 5 (c) (ii) and Figure 5
(d) (ii).
T
his plot
also
confirms
that
a pn
junction geometry (with
퐼푄
퐸
퐸푥푝
~
40%
) formed of van der Waals materials is more efficient
for
carrier
collection than a Schottky junction
geometry (with
퐼푄
퐸
퐸푥푝
~
10%
).
Embedded in the above analysis is yet another convolution of the optical and electronic
prope
rties.
As per
Figure 4 (b
)
,
we found that
absorption in FLG and Au accounted for
~
20% of
the absorbance of the total heterostructure. Assuming very few photons abs
orbed in those layers
ultimately are extracted as
free carriers
(
i.e.
퐼푄
퐸
퐴푢
≈
퐼푄
퐸
퐹퐿퐺
≈
0
), the IQE
defined above
convolutes
the
parasitic
optical
loss with the electronic
lo
ss
in the
device
.
28
Thus
another useful
metric we
shall define is
퐼푄
퐸
퐴푐푡푖푣푒
, the active layer IQE:
퐼푄
퐸
퐴푐푡푖푣푒
(
휆
)
=
퐸푄퐸
(
휆
)
퐴푏푠
(
휆
)
−
퐴푏
푠
푃
(
휆
)
(
2
)
where the
additional term
퐴푏
푠
푃
(
휆
)
corresponds to the parasitic absorption in the other layers of
the device that do not contribute to current
(
i.e.
Au and FLG
in this device
)
.
Thus,
퐼푄
퐸
퐴푐푡푖푣푒
(
휆
)
is
a measure of the carrier generation and collection efficiency only in the
active
layer
(
i.e.
WSe
2
and MoS
2
)
of the device
and is
purely
an electronic efficiency as defined above
.
We shall use this
quantity to
accurately
compare electronic geometries.
Given the good agreement between
simulations and experiment
shown in Figure 4
, a simple method of estimating the parasitic
absorption described above is therefore through electromagnetic simulations.
퐼푄
퐸
퐴푐푡푖푣푒
of the
Schottky and
pn hetero
junction geometries
calculated with Eqn. 2
is
s
hown in Figure 5 (c) (i) and
Figure 5
(d) (ii)) with
dotted
green curves
.
Analysis of these plots reveals
several
important points.
First,
퐼푄
퐸
퐴푐푡푖푣푒
for
the
pn
junction geometry is
~
3
x higher than in the Schottky junction geometry when spectrally
This document is the unedited Author’s version of a Submitted Work that was subsequently accepte
d for
publication in
A
C
S
N
a
n
o
, copyright © American Chemical Society after peer review.
To access the final edited
and published work see
http://pubs.acs.org/doi/full/10.1021/acs
nano.7b03148
7
averaged.
Though yet to be fully clarified, we
attribute
higher IQE in
pn
heterojunctions to
the
larger electric fields in a
pn
heterojunction that
may
lead to a higher exciton dissociation
efficiency and consequently IQE
.
Second,
compared to the IQE which included
the parasitic
absorption (purple dot
s
in Figure
5 (c) (ii) and Figure 5
(d) (ii)
),
the active layer IQE
curves
(green
dots)
are
spectrally flat within measurement
error
and calculations (
훿퐼푄퐸
/
퐼푄퐸
≈
0
.
07
). Thus,
the few broad peaks around the exciton
energies of WSe
2
(
~
770 nm) and MoS
2
(
~
610 nm and
~
67
0 nm) in
퐼푄
퐸
퐸푥푝
are not
attributed to
,
e.g.
,
resonant excitonic transport phenomena, but
rather
as
a simple
convolution
of the
optical and electronic effects when calculating the
electronic
IQE. In other words, consideration of parasitic absorption is critical when analyzing the
electronic
characteristics of
thin
optoelectronic device
s
.
However,
퐼푄
퐸
퐸푥푝
is still a useful metric,
as it effectively sets a lower bound on the true IQE.
G
ener
ally
,
we expect
퐼푄
퐸
퐸푥푝
≤
퐼푄
퐸
푇푟푢푒
≤
퐼푄
퐸
퐴푐푡푖푣푒
, as electromagnetic simulations tend to
slightly
overestimate the
absorption when
compared with
experimental
results
. Thus in this paper, we shall plot both expressions
when
comparing different
electronic device geometries
.
Finally
,
it is important to mention that an
active layer IQE of
~
70%
is achieved in van der Waals heterostructures without
complete
optimization of the electronic
configuration
of the device, such as
the
band
profiles
and
the
specific
choice of
contacts
.
With careful electronic design,
we suggest
it may be possible to
achieve active layer IQEs > 90%.
Optically transparent c
ontacts for carrier extraction
As a
nother
aspect
of analysis, we studied the role of
vertical ca
rrier collection compared
to lateral carrier collection in van der Waals heterostructures
. Graphene and its few
-
layer
counterpart can form a transparent conducting contact allowing for vertical carrier collection, in
contrast to in
-
plane collection
(see Fi
gure 1 (d))
.
T
he strong, in
-
plane covalent bonds of van der
Waals materials
suggest that i
n
-
plane conduction
may be
favorable
when contrasted
with the
weak
out
-
of
-
plane
van der Waals
interaction. However,
the length scale
for carrier
trans
por
t
in
-
plane (
~
휇
m)
is orders of magnitude
larger than
in the vertical direction
(
~
nm)
.
Therefore,
transport in a regime in
between the
se
two
limiting
cases is
not surprising
.
S
ilver exhibits
lower
absorption in the visible than gold, suggesting it
could
be an optimal
ba
ck
reflector for photovoltaic devices
,
as seen in Figure 4
. Thus,
we contrast the case of in
-
plan
e
and out
-
of
-
plane conduction concurrently with
the
presence of two different back
reflectors that
simultaneously fu
nction
as an electronic
back
contact (gold vs. silver) to a
pn
heterojunction, as
in Figure 6
(a) and (b).
Optical and photocurrent images of the devices are shown in
Supporting
Information S1.
Our results in Figure 6
(c) and (d)
show the
distinctions
between the various contacting
sc
hemes.
In the case of both silver and gold, a transparent top contact such as few
-
layer graphene
seems to enhance the carrier collection efficiency. This is particularly true in the case of silver,
where
퐼푄
퐸
퐴푐푡푖푣푒
enhancements of
~
5
x
is apparent.
In the case of
gold, the
IQE
is enhanced by
This document is the unedited Author’s version of a Submitted Work that was subsequently accepte
d for
publication in
A
C
S
N
a
n
o
, copyright © American Chemical Society after peer review.
To access the final edited
and published work see
http://pubs.acs.org/doi/full/10.1021/acs
nano.7b03148
8
about
~
1.5x when
parasitic absorption
is taken into account.
By analyzin
g the work
functions of
gold (
~
4.83
eV
29
)
and
silver (
~
4.26
eV
30
),
along with the electron affinity of MoS
2
(
~
4
.
0
eV
31
),
the Schottky
-
Mott rule suggests in both cases that a Schottky b
arrier should form equal to
휙
퐵
=
휙
푀
−
휒
,
32
where
휙
퐵
is the Schottky barrier height,
휙
푀
is the work funct
ion of the metal, and
휒
is the electron affinity of the semiconductor
. However,
several reports
33
–
35
have indicated that
gold appears to form an
electrically
Ohmic contact to MoS
2
, which we observe here.
Conversely,
the above data suggests that silver and MoS
2
follow the
traditional Schottky
-
Mott rule, leading
to the formation of a small Schottky barrier of
~
0
.
26
eV. Given that the energy barrier is
about
10
푘
푏
푇
, very few electrons can
be extracted out of the
pn
heterojunction
when
silver
is used
as a
back contact,
leading to very low IQEs.
By taking into account just the active layer (dashed lines),
we see that gold is
~
2
x better as an electronic contact than silver.
Finally, we examine the role of vertical carrier collection on the I
-
V characterist
ics of the
two de
vices (Figure 6
(e) and (f)).
In the case of gold
, we see purely an
enhance
ment of
the short
circuit current
with vertical carrier collection
. On the other hand, vertical carrier collection for
silver drastically increases both the short circuit current
de
nsity
and the open circuit voltage.
This
p
henomenon is consistent with
the
previously described nature of gold (Ohmic) and silver
(
S
chottky)
contacts
. Namely,
on silver
in the absence of a transparent top contact, due to
both
the Schottky barrier
and the
large in
-
plane propagation distance,
carriers
are
collected with poor
efficiency leading to a
small
퐼
푠푐
. C
onsequently
,
a
high recombination rate of the generated
carriers
which are inefficiently
extracted
lead
s
to
small
푉
표푐
values
. On the other hand,
even in the
absence of a top transparent electrode, gold
enables efficient
extraction of electrons
from t
he
pn
heterojunction
as an Ohmic contact
.
Thus, the
short circuit current and
open
-
circuit voltage
in gold
are
high
er
compared to the silver back contact
,
even in the absence of a transparent
electrode
.
When introducing
few
-
layer graphene as a transparent top contact,
the propagation
distance is significantly reduced in the
silver device
and carriers can be extracted wit
h
much
higher efficiency
, leading to a large enhancement of both the current and voltage. Whereas for
gold,
the few
-
layer graphene enhances the already high carrier collection
(
yielding
larger
퐼
푠푐
) but
only has a
negligibly small
enhancement
effect
on t
he open
-
circuit voltage.
Overall,
these results
demonstrate
that vertical carrier collection plays a crucial role in high photovoltaic device
performance
in van der Waals heterostructures
.
Thickness dependence
on charge collection efficiency
As a final
point of analysis,
we briefly examined the effect of
thickness on
퐼푄
퐸
퐴푐푡푖푣푒
under
vertical carrier collection.
We
compared the optoelectronic cha
racteristics of a thicker
pn
hetero
junction (11 nm hBN/1.5 nm FLG/4 nm WSe
2
/9 nm MoS
2
/Au) with a
thinn
er
pn
hetero
junction (1.5 nm FLG/4 nm WSe
2
/5 nm MoS
2
/Au).
The experimentally measured
absorbance and EQE are plotted in
Figure S4
for
reference
.
B
y normalizing out the
differences in
absorption
between the
pn
junction
s
, we see a somewhat surprising result
when we analyze
the
This document is the unedited Author’s version of a Submitted Work that was subsequently accepte
d for
publication in
A
C
S
N
a
n
o
, copyright © American Chemical Society after peer review.
To access the final edited
and published work see
http://pubs.acs.org/doi/full/10.1021/acs
nano.7b03148
9
active layer IQE (
dashed lines,
Figure 7
). In particular, despite
the
roughly
50% more length
in
active layer thicknesses (13 nm vs. 9 nm)
and qualitatively different absorban
ce and EQE spectra,
the thick
pn
junction exhibits nearly the
same active l
ayer IQE compared to the thin
pn
junction.
In fact, it appears to be slightly more efficient, but this is within the error bar of the measurement
and simulation
s
(
훿퐼푄퐸
/
퐼푄퐸
≈
0
.
07
, see Methods
for details of errors
).
This
observation
is
c
orroborated with the experimentally derived IQE
(dotted curves, Figure 7)
, which has nearly the
same spectr
um between the two thicknesses, but differ
in magnitude
due to differences in
parasitic absorption
.
This
result
suggests
that in the ultrathin limit
(
~
10 nm)
of van der Waals
heterostructures with vertical carrier collection,
the IQE
has
a
weak dependence on active layer
thickness
. This
weak
dependence may be due to
a combination o
f increased scattering competing
wit
h
charge transfer
,
36,37
tunneling
,
4,38,39
and
exciton
quenching
40,41
effects
as the vdW
heterostructure becomes
thicker. The exact role of each of these effect
s
, as well as possibly other
effects
,
will require a
new
theoretical fr
amework
and
experimental measurements
to analyze
their
relative
contributions
to charge collection efficiency
.
CONCLUSIONS
O
ur results suggest important
challenges
that must be
addressed
to enable high
photovoltaic effi
ci
ency. For example
, despite
the usefulness of gold as an electrical back contact,
we found
from electromagnetic simulations
that it
accounts for nearly 20% of the parasitic loss
in the heterostructure
s
reported
here
. Schemes using
optically transparent carrier
selective
contact
s
could
be used to avoid this
parasitic
optical loss
.
Another
open question
is the role and
importance of exciton dissociation and transport
. Indeed
, the
large
exciton
binding energies in
t
ransition metal dichalcogenides (
~
50
−
100
meV
in the bulk)
25,26
suggests
that
a significant
exci
ton
population
is
generated immediately after illumination.
However, it is
not
yet
clear
whether
such
an
excit
on
population
fundamentally limit
s
the internal quantum efficiency
of the
device, posing an upper limit
on the maximum achievable EQE
in
van der Waals material
s
based
photovoltaic devices.
Finally, the problem of open
-
circuit voltage must als
o be addressed.
For
example,
the type
-
II band alignment between
ultrathin
MoS
2
and WSe
2
suggests a
renormalized
bandgap
of
~
400
−
500
m
e
V
,
42
given by the minimum conduction band energy and maximum
valence band energy
of the two materials
. In accordance with the S
hockley
-
Queisser limit, this
would severely reduce
the
maximum
power conversion efficiency
attain
able
by a factor of
~
3.
T
herefore, t
o achieve higher
open circuit voltages,
a monolithic
device structure
may be required
to
avoid
low energy
interlayer recombination
states
.
However, our results described here also suggest a
different approach
in
addressing the
optical and electronic considerations
for ultrathin van der Waals heterostructures
when
compared with conventional photovoltaic structures. For example,
our observation
that ultrathin
van der Waals heterostructures can be optically treated a
s a single effectiv
e medium is
a regime
of optics that is uncommon for the visible to near
-
infrared wavelengths analyzed in photovoltaic
This document is the unedited Author’s version of a Submitted Work that was subsequently accepte
d for
publication in
A
C
S
N
a
n
o
, copyright © American Chemical Society after peer review.
To access the final edited
and published work see
http://pubs.acs.org/doi/full/10.1021/acs
nano.7b03148
10
devices.
Likewise, our observation of
weak thickness dependence o
f the
charge collection
efficiency represents a realm
of electronic transport that is also quite unconventional
and
unexplored
when compared to traditional photovoltaic structures
.
Thus, the combination of
the
above observations
may enable entirely different photovoltaic device physics and architectures
movin
g forward
.
To summarize, we have shown that external quantum efficiencies > 50% and active layer
internal quantum efficiencies
> 70%
are possible in
vertical
van der Waals heterostructures
.
We
experimentally demonstrated absorbance > 90% in van der Waals
heterostructure
s
with good
agreement to electromagnetic simulations.
We
further
used the active layer internal quantum
efficiency to quantitatively compare
the
electronic
charge collection
efficiencies of
different
device geometries
made with van der Waals materials
.
By further reducing parasitic optical losses
and performing a careful study on exciton dissociation and charge transport while simultaneously
engineering the band profiles and contacts
, van der Waals photovoltaic devices
may be able to
achieve external quantum efficiencies > 90%.
Our results presented here show a promising
and
exciting
route to
designing and
achieving efficient
ultrathin
photovoltaics
composed of van der
Waals heterostructures
.