of 30
www.sciencemag.org/content/344/6187/1005/suppl/DC1
Supplementary Materials for
Amorphous TiO
2
Coatings Stabilize Si, GaAs
, and GaP Photoanodes for
Efficient Water Oxidation
Shu
Hu
,
Matthew R.
Shaner
,
Joseph A.
Beardslee
,
Michael
Lichterman
,
Bruce S.
Brunschwig
,
Nathan S.
Lewis*
*Corresponding author. E-mail: nslewis@caltech.edu
Published 30 May 2014,
Science
344
, 1005 (2014)
DOI: 10.1126/science.1251428
This PDF file includes:
Materials and Methods
Supplementary Text
Figs. S1 to S17
References (
32, 33
)
Materials and Methods
Chemicals
All materials were used as received, except where otherwise noted. Sulfuric acid,
concentrated hydrochloric acid and hydrogen peroxide were purchased from EMD
Chemicals. H
2
O with a resistivity of 18.2 MΩ∙cm was obtained
from a Millipore de
-
ionized water system.
An aqueous solution that contained the Fe(CN)
6
3-/4-
redox couple was prepared by
making a solution of 50 mM K
3
Fe(CN)
6
(Fisher Scientific, 99.4%) and 350 mM
K
4
Fe(CN)
6
(Fisher Scientific, 99.4%) in 200 mL of H
2
O. A 1.0 M aqueous solution of
KOH (semiconductor grade, Sigma
-Aldrich, 99.99% trace metal basis) was also
prepared.
For non-
aqueous electrochemistry, acetonitrile (CH
3
CN, anhydrous, 99.8%, Sigma
Aldrich) was dried by flowing through a solvent column, and then was stored over 3Å,
activated, molecular sieves (Sigma
-Aldrich).
LiClO
4
(battery grade, 99.99%, Sigma
Aldrich) was dried by fusing the salt under a pressure <
1×10
-3
Torr at 300 °C. Prior to
use, the resulting material was stored under ultra
-high purity
(UHP) Ar(g) that contained
< 0.2 ppm of O
2
(g). Bis(cyclopentadienyl) iron(II) (ferrocene, FeCp
2
0
) was purchased
from Sigma Aldrich and was purified by sublimation under vacuum.
Bis(cyclopentadienyl) iron(III) terafluroborate (ferrocenium, FeCp
2
+
• BF
4
) was
purchased from Sigma Aldrich, recrystallized in a mixture of diethyl ether and
acetonitrile (EMD Chemicals), and dried under vacuum prior to use. These reagents
were used to prepare a solution that contained reagent concentrations of 1.0 M LiClO
4
,
0.5 mM ferrocenium, and 90 mM ferrocene.
Preparation of Si, GaAs, and GaP substrates
Three types of Si wafers were used as substrates for atomic layer deposition: p
+
-Si
(B-doped with a resistivity,
ρ
< 0.002 Ω∙cm, 3” diameter); n
-
Si (P
-doped with
ρ
= 2.06 –
2.18 Ω∙cm, 3” diameter); and n
+
-Si (As
-doped with
ρ
<0.002 Ω∙cm, 3” diameter). The Si
surfaces were first cleaned using an RCA SC
-1 procedure that involved soaking the Si
wafers in a 3:1 (by volume) solution of H
2
SO
4
(conc., ~ 18.4 M) and H
2
O
2
(conc., ~
1 M)
for 10 min, and then placing the samples for ~ 10 s in a 10% (by volume) aqueous
solution of hydrofluoric acid (conc. 31.8 M). The Si samples were then etched using the
RCA SC-
2 procedure of soaking in a 5:1:1 (by volume) solution of H
2
O, concentrat
ed
hydrochloric acid (11.1 M), and hydrogen peroxide (conc.~ 1 M) for 10 min at 75 °C.
A p
+
-GaAs wafer oriented to expose the (111) B surface (Zn-
doped, acceptor
concentration of 1×10
19
cm
-3
, 2 inch by AXT, Inc.) was also used as a substrate for
atomic
-layer deposition. The GaAs was placed for 15 s in a solution of 0.04% (by
volume) Br
2
(Acros Organics) in methanol (CH
3
OH, low water, J.T.Baker) and was then
immersed in 1.0 M KOH (aqueous solution of potassium hydroxide pellets,
semiconductor grade, 99.99% trace metals basis, Sigma
-Aldrich) for 15 s. The GaAs
samples were rinsed with copious amounts of deionized H
2
O, and were then dried using a
stream of N
2
(g).
Prior to ALD, n
-type GaP samples were etched, and ohmic contacts were formed,
using a published procedure.
(
32
)
n
-GaP
samples were prepared from n
-GaP wafers that
had a sulfur dopant density of ~ 5 × 10
17
cm
-3
. Small (~ 0.3 cm
2
) pieces were cut from the
2
wafer using a scribe, and the pieces were then etched in 18.4 M H
2
SO
4
for 30 s, followed
by a rinse with copious amount
of de
-ionized H
2
O.
Fabrication of planar np
+
-Si and np
+
-GaAs samples
A 4” n
-Si (0.1-
0.3 ohm
-cm, <100> oriented, single
-side polished) wafer was cleaned
prior to formation of the p
+
emitter. The cleaning procedure consisted of a 30 s in
Buffered hydrofluoric acid (BHF) etch (Transene, Inc.); 20 min in an RCA I (5:1:1 of
H
2
O:H
2
O
2
:NH
4
OH) organic
-species cleaning solution at 70 °C; a BHF etch for 30 s; and
20 min RCA II (6:1:1 of H
2
O:HCl:H
2
O
2
) metals cleaning at 70 °C, with water rinses
between each proces
s step. Immediately following this cleaning procedure, the wafer
was cleaved into ~1” × 1” pieces, to fit into the doping furnace. Each piece was then
etched for 30 s in BHF, rinsed with H
2
O, dried with N
2
(g), and loaded into a boron-
doping furnace. Waf
er pieces were loaded back
-to-back on a quartz boat with the
polished side facing out, and were placed between Saint Gobain BN
-975 boron-
doping
wafers that over
-filled each n
-Si piece. For each doping run, six 1” × 1” n-
Si pieces were
loaded into 3 slots. Each doping run was performed in a 4” tube furnace at 950 °C under
a 10 L min
-1
flow of N
2
(g). In each run, the boat was loaded into the tube furnace during
a 1 min period, and was then soaking at 950 °C for 3 min and unloaded over another 1
min period. Following two doping runs (12 total samples), each sample was etched for
30 s in BHF and then was loaded, without a dopant source, into the doping furnace with a
5 L min
-1
flow of O
2
(g) for 30 min at 750 °C, to grow a low temperature oxide (LTO).
This L
TO allowed facile removal of residual boron on the surface through a 30 s BHF
etch.
Planar np
+
-GaAs junctions were grown epitaxially on an n
+
-GaAs wafer that had a
(100)
-oriented, polished surface (Si
-doped, donor concentration of 1×10
18
cm
-3
, 2 inch by
MTI, Inc.). A 2 μm thick n-
GaAs layer (Si
-doped, donor concentration of 2×10
17
cm
-3
)
and a 100 nm thick p
+
-GaAs layer (Si
-doped, accepter concentration of ~5×10
18
cm
-3
)
were sequentially grown on the n
+
-GaAs substrate. The edges of both the np
+
-Si and np
+
-
GaAs wafers were cleaved off and discarded, to eliminate shunts from the back side of
the sample to the front p
+
emitter.
Atomic layer deposition of TiO
2
TiO
2
films were deposited onto Si, GaAs, and GaP
substrates at 150 °C using a
Cambridge Nanotech S200 ALD system. The Si, GaAs, and GaP surfaces were prepared
as described above and loaded immediately thereafter into the ALD chamber. Each ALD
cycle consisted of a 0.015 s pulse of H
2
O, followed by a 0.10 s pulse of tetrakis
-
dimethylamidotitanium (TDMAT, Sigma
-Aldrich, 99.999%, used as received). A 15-
s
purge under a constant 0.02-
L min
-1
flow of research
-grade N
2
(g) was performed between
each precursor pulse. When idle, the ALD system was maintained under a continuous
N
2
(g) purge and had a background pressure of 2.7×10
-1
Torr.
Deposition of metal films and fabrication of island patterns
Both electron
-beam evaporation and radio-
frequency sputtering were used to deposit
Ni onto patterned and unpatter
ned TiO
2
surfaces. The electrochemical performance was
essentially identical for both sputtered and e
-beam evaporated Ni. The as
-deposited Ni
films were 100 nm thick for p
+
-Si and p
+
-GaAs in Fig. 1 and 4, were nominally 2 nm
3
thick for n-
GaP and np
+
-GaAs in Fig. 1, and were nominally 3 nm thick for the XPS
study in Fig. S17. The deposition rate was calibrated using profilometry. 100-
nm thick
Ni films were also deposited on flat quartz slides as a control sample in Fig. 1 and 4.
A series of photolithogr
aphy, physical vapor deposition, and lift
-off steps was used
to fabricate Ni
-island patterns on TiO
2
overlayers that had been grown on n
-Si, np
+
-Si
and np
+
-GaAs substrates. The mask used for the island pattern was a square array of 3
μm
-diameter circles w
ith a 7
-μm pitch. Ni islands (100 nm thick) on TiO
2
were
fabricated by lifting off positive photoresist films, S1813 (Shipley), which were patterned
to mask deposited Ni layers. The photoresist was removed using PG remover
(MicroChem) followed by rinsing
the surface of the sample with copious amount of H
2
O.
Optical absorptance measurements
The optical absorptance of the n
-Si/TiO
2
/Ni
-island samples was determined by
integrating sphere transmission (
T
) and reflection measurements for both sides of the
samples (
R
1
and
R
2
) at normal incidence. For transmission measurements, the sample
was mounted at the front port of the integrating sphere, whereas for reflection
measurements the sample was mounted at the rear port of the sphere. The optical
reflection was normalized to a reflectance standard (99%, LabSphere, Inc) that was
placed at the same location as the sample. The absorptance (as a function of wavelength,
λ) was determined from the measured reflection and transmission, assuming sufficient
absorption of the sample, as follows:
)
(
1
)
(
)
(
1
)
(
2
1
λ
λ
λ
λ
R
T
R
A
=
.
Electrode preparation
Ohmic contacts to the semiconducting electrodes were formed by rubbing an In-
Ga
eutectic onto the unpolished back sides of the Si samples; by sequentially depositing
Ni/Cu films onto the unpolished back sides of the p
+
-GaAs samples; or by soldering In to
the back side of n
- p
+
-GaAs or n
-GaP samples. The np
+
-GaAs samples were then
annealed under forming gas at 450 °C for 30 s whereas the n-
GaP samples were annealed
under forming g
as at 400 °C for 10 min. For aqueous electrochemistry, Ag paste was
used to attach the samples to a piece of Cu tape as a current collector. The sample
assembly was then assembled into a custom
-made Teflon compression cell equipped with
an O
-ring seal (
0.0314 cm
2
in electrode area). For electrochemistry using glass
-rod
electrodes, Ag paste was used to attach the ohmic contact on the back side of the samples
to a coiled, tin
-plated Cu wire which was then threaded through a glass tube. The sample
was then encapsulated and sealed to the glass tube using grey epoxy (Hysol 9460F). An
optical scanner was used to image the exposed electrode surface areas and the areas were
measured using ImageJ software. All the electrodes in this study were typically 0.2 – 1.3
cm
2
in area.
Aqueous and non-
aqueous electrochemical measurements
A saturated calomel electrode (SCE, CH Instruments) was used as a reference
electrode for electrochemical measurements that were performed using aqueous
solutions, including photoele
ctrochemical, spectral response and faradaic efficiency
measurements. Pt gauze (100 mesh, 99.9% trace metal basis, Alfa
-Aesar), was used as
4
the counter electrode during electrochemical measurements with the Fe(CN)
6
-3/ -4
couple,
and a carbon rod placed wit
hin a fritted glass tube (Aceglass, Inc.) was used as the
counter electro
de for 1.0 M KOH(aq) solution. A carbon counter electrode in a fritted
glass tube eliminated the possibility of contamination of the electrolyte and thus the
working photoelectrode b
y trace Pt that would result from minute dissolution of the
material in a Pt counter electrode
. The formal potential for the oxidation of to O
2
(g),
0.19 V vs. SCE, was calculated from the measured pH of the solution (pH 13.7) and
using SCE = 0.24 V vs. NHE. Consistently, the open-
circuit potential of a Ni film
electrode in 1.0 M KOH was measured as 0.19 V vs. SCE. The
Nernstian potential of the
50 mM Fe(CN)
6
3-
and 350 mM Fe(CN)
6
4-
(aq) solution was 0.19 V vs. SCE, as obtained
by measuring the potential difference of a 2 mm
-diameter Pt disc vs. an SCE in the
solution, and the open-
circuit voltage was determined as the potential difference at open
-
circuit under illumination between the photoelectrode and a counterelectrode poised at
the Nernstian potential of the electrolyte solution.
A Pt wire (0.5 mm diameter, 99.99% trace metals basis, Alfa
-Aesar) was used as a
reference electrode and
a Pt gauze was used as the counter electrode for electrochemical
measurements made in the non
-aqueous solutions. A custom electrochemical cell with a
Pyrex flat bottom was used for non-
aqueous electrochemistry. During measurements, the
electrolyte was r
igorously agitated with a magnetic stir bar driven by a model
-train
motor. All of the data presented for electrochemical measurements in aqueous solutions
included compensation for the series resistance of the solution, as obtained from high
-
frequency ele
ctrical
-impedance measurements.
A Xe lamp
-based solar simulator, as well as ELH
-type and ENH
-type tungsten
-
halogen lamps, were used for photoelectrochemical experiments. The illumination
intensity was calibrated by placing a Si photodiode (Thor Labs) in t
he glass
electrochemical cell or in a Teflon compression cell, in the same location that was
occupied by the exposed area of a photoelectrode. The Si photodiode was previously
calibrated by measurement of the short
-circuit current
-density value under AM 1.5
simulated sunlight at 1
-Sun, i.e., at 100 mW cm
-2
of irradiance from a Xe arc lamp with
an AM 1.5 filter.
Oxygen
-evolution efficiency measurements
A Neofox fluorescence probe was used in an airtight cell with a side
-facing quartz
window. The oxygen
concentration was monitored throughout the measurement, and
was tabulated via measured values from a fluorescence detector. These values were
calibrated using the standard concentration of oxygen in water (7700 μg L
-1
or 2.4×10
-4
M) under a standard 20.9%
(by volume) oxygen atmosphere. A Ag/AgCl electrode was
the reference electrode, and the counter electrode was a carbon rod that was positioned in
a separate, fritted compartment. The cell volume was 52 mL, and the cell was purged
with a stream of UHP Ar
(g) for ~1.5 h prior to data collection. The experiment lasted for
30 min after a 10-
min waiting period at open circuit. The oxygen concentration was
monitored throughout the measurement.
The data were collected and modeled using a Matlab script. The o
xygen
concentration data were converted into micrograms of O
2
by first correcting for the O
2
leak rate measured during the first 10 min of the experiment, followed by calculating the
mass of O
2
in micrograms at each time point. The calculation process was
performed by
5
multiplying the reported percentage of oxygen by the number of micrograms of O
2
dissolved in water at room temperature under 1 atm, 7700 μg L
-1
(assuming the value the
same as in 1.0 M aqueous acidic solutions), and the cell volume (52 mL), divided by the
concentration of O
2
in air under 1 atm, which was in the value reported by the florescence
probe (20.9%). To compare the charge versus time data from the potentiostat with the
amount of oxygen generated versus time, 100% Faradaic efficiency was assumed for
comparison. Specifically, the amount of charge passed (in mA∙h) was multiplied by 3.6
to convert the data into coulombs; this value was then multiplied by 83 to convert the
value into micrograms of O
2
, because this is the conversion factor
of 1 coulomb of
electrons into 1 microgram of O
2
. Therefore, both the cumulative oxygen generated and
charge passed during the measurement can be shown in one plot with two comparable y
axes, where 0.33 mA∙h of charge passed corresponded to 100 μg O
2
gen
erated. The
electrode area for the measured Si, GaAs and GaP photoanodes are 0.26, 0.20 and 0.47
cm
2
, respectively.
Secondary
-ion mass spectrometry (SIMS)
Secondary
-ion mass spectrometry (SIMS) analysis was performed using a Cameca
SIMS
-7f GEO instrument. A Cs
+
primary ion beam was used to ionize and sputter
surface atoms from the sample. The sputtered, ionized atoms (secondary ions) were
collected and analyzed in a mass spectrometer. The samples were sputtered at normal
incidence with a 7 keV Cs
+
beam at substrate bias of 4 keV. The raster area was 125 μm
× 125 μm, with a centered 12500 μm
2
gated area. Count rates of
30
Si,
48
Ti,
12
C +
133
Cs
and
14
N +
133
Cs (complex ions) were recorded as a function of sputtering time.
Spectroscopic ellips
ometry
Complex refractive index (
n
,
k
) data for films of unannealed, as
-grown TiO
2
on n
+
-
Si were obtained using spectroscopic ellipsometry. The ellipsometric data were acquired
using a J.A. Woolam V
-VASE system. The non
-absorbing (
k
~0) portion of the dat
a were
fit using a Cauchey model that assumed a TiO
2
/SiO
2
bilayer structure on Si. The TiO
2
n
,
k
values were extracted from a point
-by-point fit, using fixed Cauchey and film
-thickness
parameters. Measurements were performed on TiO
2
films with thicknesse
s of 4 nm, 31
nm, 44 nm, 68 nm, and 143 nm, with the extracted
n
,
k
values being comparable for all
film thicknesses.
Electron microscopy
Transmission electron microscopy (TEM) was performed using a JOEL JEM
-2100F
operating at an accelerating voltage o
f 200 kV. The bright
-field images and selected
-area
electron diffraction patterns were acquired under TEM mode, while the elemental
-
contrast images (Fig. 3A) and energy
-dispersive x
-ray spectroscopy data (Fig. 3B) were
acquired under scanning TEM mode. T
he cross
-sectional TEM sample was polished to 5
μm in thickness using a series of diamond lapping films, and the samples were then
further polished to 50 – 100 nm thin foils by a Gatan precision ion polishing system.
X-ray ph
otoelectron spectroscopy
X-ray photoelectron spectroscopy (XPS) was performed using a Kratos Axis Ultra
system with a base pressure of < 1×10
-9
Torr. A monochromatic Al Kα source was used
6
to illuminate the sample with 1486.7 eV photons at a power of 150 W. A hemispherical
analyzer oriented for detection along the sample surface normal was used for maximum
depth sensitivity. Survey scans were obtained at a resolution of 1 eV with a pass energy
of 80 eV. Detailed scans were acquired at a resolution of 25 meV with a pas
s energy of
10 eV, using variable acquisition times. When cleaning of adventitious species such as C
was required, an octopole ion gun was used to sputter
-clean the sample surface with 500
eV Ar ions for 30 – 60 s. ~10 min of air exposure occurred between sample transfers
from the ALD or sputtering deposition tool to the XPS load lock.
Measurement of charged
-carrier density (Hall measurements)
TiO
2
films were grown on glass slides by the aforementioned ALD procedure. The
glass slides were cleaned by s
equential rinsing with acetone, isopropanol and ethanol for
1 min each. The samples used for Hall measurements were made by cutting TiO
2
-coated
glass into 1 cm × 1 cm square pieces, and In was directly soldered onto the four corners
of the TiO
2
film. The
current
-supply leads and voltage probes were connected to the Hall
measurement circuit board through In
-soldered Cu wires. Ohmic conduction was
confirmed by current
-voltage (
I
-
V
) measurements of any two of the four In-
contacted
corners, with the supplied current in the range of 50 – 100 nA. The magnetic field was
fixed at 3000 gauss. TiO
2
films having thicknesses of 4 nm, 31 nm, 44 nm, 68 nm, and
143 nm were measured, and all showed n-
type conduction with an average electron
concentration of 4×10
16
cm
-3
.
Hg droplet measurements
A Hg droplet was placed on ALD TiO
2
-coated Si samples, with the contact area
confined by a 2 mm diameter O
-ring. A Pt wire was inserted into the Hg droplet to make
the top contact, whereas the ohmic bottom contact was formed by rubbing In-
Ga eutectic
into the unpolished back side of the Si. The bias polarity was kept the same as in the
electrochemical measurements of Fig. 1 and 4, i.e. positive voltage corresponded to a
forward bias of the p
+
-Si substrate with respect to a Hg droplet, and positive current
corresponded to anodic currents in which holes flowed from the Si to the Hg droplet.
Mott
-Schottky analysis
The Mott
-Schottky analysis (Fig. S15) of the area
-corrected differential
capacitance–potential data was obtained using electrochemical impedance
spectroscopy.
(
31
) The solution was quiescent and kept in dark during the measurements.
The electrochemical impedance data were fit to a model that consisted of a parallel
resistor and capacitor with a fixed constant phase element arranged electrically in series
with a se
parate resistor.
Supplementary
Text
Load
-line
analysis
The
J -E
behavior of a serially
connected Si PV cell and an electrocatalytic anode
was calculated by negatively shifting the
J
-E
characteristic of the electrolytic anode by
the photovoltage of the PV ce
ll at each photocurrent density
. The photocurrent density
(
J
) – voltage (
V
) performance of the ideal Si PV cell was calculated using the ideal diode
equation:
=
−퐽
0
(
푞푉
푘푇
1)
, where
is light
-limited current density (
=27.7
7
mA
∙cm
-2
),
0
is dark current density,
q
is the unsigned charge on the electron,
k
is
Boltzmann’s constant, and
T
is the absolute temperature (
T
= 300 K). Obtaining a
V
oc
of
the ideal Si cell that reproduced the observed 0.32% thermodynamic water
-oxidation
efficiency required a value of
0
= 1.55 × 10
−6
mA
∙cm
-2
for the PV cell component of
the series
-connected system used in the model. An analogous procedure for GaAs
-based
systems required a PV cell having a
0
= 1.25 × 10
−12
mA
∙cm
-2,
which produced a
V
oc
=
0.77 V, in combination with a
J
L
= 14.3 mA∙cm
-2
and a fill factor of 0.86 and a PV
energy
-conversion
efficiency of 9.5%, in series with a p
+
-GaAs
/ TiO
2
/Ni
-film dark
electrode
(Fig. S2B), to produce the
J
-
E
behavior observed for the n-
p
+
-GaAs
/ TiO
2
/Ni
-
film integrated photoelectrode system (Fig. 1D). If a 25% absorption loss of Ni catalysts
is considered, a 12.7% GaAs PV cell would be required, and the water
-oxidation
efficienc
y would be 6.6% at a light
-limited photocurrent density of 19.1
mA∙cm
-2
when
there’s negligible optical absorption loss.
ALD growth rate
ALD samples with 250, 750, 1000 and 1500 and 3000 cycles of pulsed precursors
were grown on p
+
and n
+
Si substrates. These growths produced film thicknesses of 4.3
nm, 30.7 nm, 43.5 nm, 68.3 nm and 142.5 nm, corresponding to a growth rate of ~ 0.5 Å /
cycle. The ellipsometrically measured and fitted film thicknesses agreed with the
thicknesses obtained from TEM micrographs (Fig. S7).
Energy
-conversion properties
of photoelectrodes
A bare n
-p
+
-Si electrode in contact with a non
-aqueous electrolyte that contained 0.5
mM ferrocenium and 90 mM ferrocene (Fc) with 1.0 M LiClO
4
dissolved in dry CH
3
CN
exhibi
ted
V
oc
= 0.50 – 0.55 V, a short
-circuit current density,
J
sc
, of 33.6 ±
5.0 mA
cm
-2
,
and a fill factor,
ff
, of 0.29, with a photoelectrode energy
-conversion efficiency of 5.42%
under simulated 1-
Sun illumination (Fig. S3). The
V
oc
of the n-
p
+
-Si junction
was
therefore within experimental error of the photovoltages exhibited by the n-
p
+
-
Si/TiO
2
/Ni
-island electrodes. The presence of the thick TiO
2
film thus resulted in < 50
mV voltage losses for both minority
- and majority
-carrier hole conduction.
In cont
rast to
the behavior shown in Fig. 1
B, a photovoltage of < 5 mV was observed when Ni islands
were directly deposited and patterned onto n-
Si electrodes (Fig. S4), presumably due to
the formation of nickel silicide at the Si/Ni interface. In addition to p
roducing negligible
photovoltages under alkaline anodic operating conditions, the Ni on the n-
Si/Ni samples
was undercut and the underlying Si was etched consequently, in accord with prior
observations (
14
)
and with expectations based on the chemical and photoanodic
instability of unprotected or poorly protected Si electrodes in alkaline media.
Description of XPS data
Fig. S17 shows the X
-ray photoelectron spectra for the Ti 2p and Ni 2p core
-levels
for a 3
-nm Ni film deposited on ALD
-grown TiO
2
. The chemical shift of the Ti 2p1/2
and Ti 2p3/2 peaks indicates that only the Ti
4+
chemical states were present in the as
-
grown TiO
2
. The interfacial Ni may be present in various oxidation states, with Ni
2
O
3
and nickel oxide hydroxide (when present) contributing to the peak at 855.9 eV binding
energy. At the Ni/TiO
2
interface, the oxidation state of Ti remained Ti
4+
as compared
8
with ALD
-grown TiO
2
, because the binding energy of the Ti 2p XPS core
-level spectra
remained at 459.0 eV after Ni deposition.
The valence-
band offset of TiO
2
relative to the Si substrate was determined by XPS
using the Kraut method, as described by Wallrapp
e
t al
. (
33
)
High
-resolution scans were
obtained from the Fermi edge to the Si 2p and Ti 3p peaks, for Si that was cleaned by the
RCA SC-
2 procedure as well as for a bulk film of TiO
2
deposited via ALD. The valence
-
band maximum (VBM) for the RCA SC
-2 cleaned Si and for the 43.5 nm TiO
2
film on Si
was determined by a linear extrapolation of the Fermi edge to the binding energy axis
(Fig. S13A and S13C). The differences between the intensity maxima of the Si 2p and Ti
3p core
-level peaks and their corresponding valence
-band maxima, respectively, were
taken as the energy of the core level orbitals, which were found to be 98.86±0.005 eV
and 34.36±0.02 eV, respectively. A high-
resolution scan of a Si substrate with a thin TiO
2
layer (190 cycles of ALD on Si) was acquired over an energy range that spanned the Si
2p peak and the Ti 3p peak (Fig. S13B). The core level of
fset (of 62.30 ± 0.06 eV)
between the Si 2p and Ti 3p signals was determined by the difference in energy values
between the intensity maxima of the two peaks. The overall valence
-band offset across
the Si/TiO
2
interface was then calculated as the differen
ce between the Si 2p to VBM
value and the Ti 3p to VBM value, subtracted from the offset between the Si 2p and Ti 3p
peaks. The VB offset value was determined to be 2.22 ± 0.08 eV for ALD
-grown TiO
2
films on Si with interfacial layers prepared by the RCA SC
-2 etching procedure.
Description of optical absorption data
Tauc plots (Fig. S14) were generated by plotting (
α
h
ν
n)
1/2
as a function photon
energy, where
α
is the absorption coefficient (
α
= 4 π
k
/ λ), and
k
is Planck’s constant.
The two small peaks observed in the Tauc plot were consistent with the presence of
defect bands in the unannealed, as
-grown TiO
2
films. The optical band gap of the as
-
grown TiO
2
, 3.34 ± 0.01 eV, was determined by a linear extrapola
tion of the optical
absorption edge in the Tauc plot to the y axis.
Ir overlayers on TiO
2
Ir was selected as an alternative metal to evaluate the generality of the
TiO
2
/electrocatalyst structure relative to TiO
2
/Ni. Ir has a higher work function (5.3 – 5.7
V) than Ni (4.8 – 5.0 V) as measured in vacuum. RuO
x
has limited stability under acidic
and basic conditions. IrO
x
is considered as a stable OER catalyst in base, and only
IrRuO
x
catalysts, used in dimensionally stable anodes, are long
-term stable
in acid. In
base, although RuO
x
is more active than IrO
x
, RuO
x
is not significantly more active than
Ni-based OER catalysts: for example, McCrory
et al.
(29)
recently showed that IrO
x
has
an overpotential of 330 mV as compared to NiFeO
x
of 360 mV when both produce
anodic current densities of 10 mA∙cm
-2
for the OER.
Long
-term stability of photoelectrodes
An n
-Si/TiO
2
/Ni photoelectrode has run for >200 hours cumulatively, and the film
did not prevent
hole conduction, however longer term studies would be required to assess
the very long term thermal stability of the film towards oxidation. Additionally, separate
studies are needed to establish the efficacy of the unannealed film over very large
9
electrode areas to understand the defect den
sity and defect tolerance of the electrodes in a
manufacturing environment.
10
11
Fig. S1.
Secondary-ion mass spectrometry data of TiO
2
films on Si with Ni overlayers (a) and
without Ni overlayers (b) of nominally 100 nm Ni thickness deposited by sputtering
.
The
thickness of the TiO
2
films in (a) and (b) was 143 nm. If Ni had percolated through the
TiO
2
films, the Ni signal in (a) would be above the background Ni concentration, as
shown in (b). (c) C and N impurity profiles in ALD-TiO
2
films of 4 nm, 31 nm, 44 nm,
and 143 nm in thickness. The C and N impurity levels were constant with depth
throughout the films of various thicknesses, and were comparable to each other for
various thicknesses.
A
B
10
0
10
1
10
2
10
3
10
4
10
5
10
6
10
7
2500
2000
1500
1000
500
0
10
0
10
1
10
2
10
3
10
4
10
5
10
6
10
7
1500
1000
500
Sputtering cycles
Sputtering cycles
Counts per second
Sputtering cycles
Counts per second
58
Ni
48
Ti
28
Si
58
Ni
48
Ti
28
Si
143 nm
68 nm
31 nm
4 nm
30
Si
48
Ti
12
C+
133
Cs
14
N+
133
Cs
10
1
10
3
10
2
10
5
10
4
10
7
10
6
10
8
Counts per second
C
143 nm
143 nm
12
Fig. S2.
Dark
J-E
behavior of TiO
2
-coated p
+
-Si (a) and p
+
-GaAs (b) electrodes, covered with 100
nm Ni films, in contact with 1.0 KOH (aq). The cyclic-voltammetry
data in both (a) and
(b) were measured in the dark, and have been compensated for the ohmic series
resistance of the electrolyte. The formal potential for water oxidation is labeled at 0.19
V
vs. SCE. (a) 4
143 nm thick TiO
2
, as well as a 100 nm thick, continuous Ni film on
quartz, as a control; and (b) is the same structure on p
+
-GaAs but with a TiO
2
thickness of
68 nm. The black curve in (b)
shows increasing corrosion current for a p
+
-GaAs
electrode without the TiO
2
overlayer present. The electrode area was typically 0.2
1.3
cm
2
.
120
100
80
60
40
20
0
Current density
J
(mA cm
-2
)
0.8
0.6
0.4
0.2
0.0
E
(V vs. SCE)
120
100
80
60
40
20
0
Current density
J
(mA cm
-2
)
0.8
0.4
0.0
-0.4
E
(V vs. SCE)
A
B
4 nm TiO
2
31 nm TiO
2
44 nm TiO
2
68 nm TiO
2
143 nm TiO
2
Ni only, no TiO
2
p
+
-Si/TiO
2
/Ni film
p
+
-GaAs/TiO
2
/Ni film
p
+
-GaAs
/TiO
2
/Ni
p
+
-GaAs
/Ni
O
2
/H
2
O
O
2
/H
2
O
13
Fig. S3.
J-E
behavior
of an np
+
-Si buried junction (photoactive) in contact with 0.5 mM
ferrocenium and 90 mM ferrocene (Fc) and 1.0 M LiClO
4
in CH
3
CN, under simulated
ELH-type 1-Sun illumination
(blue curve) and in dark (black curve). The top p
+
-Si layer
formed an ohmic contact to the Fc
+/0
redox couple.
14
Fig. S4.
J-E
behavior of Ni islands (100 nm thick, 3 μm circles in 7 μm pitch) on n-Si in 1.0 M
KOH(aq), in the dark (A) and under illumination (B).
120
100
80
60
40
20
0
Current density
J
(mA cm
-2
)
0.6
0.4
0.2
0.0
-0.2
-0.4
E
(V vs. SCE)
A
B
120
100
80
60
40
20
0
Photocurrent density
J
(mA cm
-2
)
0.6
0.4
0.2
0.0
-0.2
-0.4
E
(V vs. SCE)
15
Fig. S5.
J-E
data of an np
+
-Si photoelectrode coated with TiO
2
and Ni islands at various stages of
chronoamperometry shown in Figure 3. The fill factors at each stage barely decreased,
and the light-limited photocurrent density varied linearly with the illumination intensity.
Beginning
End
40
30
20
10
0
Photocurrent density
J
(mA cm
-2
)
1.5
1.0
0.5
0.0
E
(V vs. SCE)
40
30
20
10
0
Photocurrent density
J
(mA cm
-2
)
1.5
1.0
0.5
0.0
E
(V vs. SCE)
16
Fig. S6.
Photocurrent vs. time data, under simulated 1-Sun illumination at the formal potential for
water oxidation (0.19
V vs. SCE), of a
2-nm Ni/118 nm TiO
2
-coated np
+
-GaAs
photoanode (A) and a 2-nm Ni/118 nm TiO
2
-coated n-GaP (B) photoanode in 1.0 M
KOH (aq) under constant simulated AM 1.5 1-Sun illumination. Stable photocurrents for
> 25 hours are shown in the inset of (A). The GaAs and GaP electrode area was 0.82
cm
2
and 0.40
cm
2
, respectively.
15
10
5
0
Photocurrent density
J
(mA cm
-2
)
30
25
20
15
10
5
0
Time (hours)
12
10
8
6
4
2
0
Photocurrent
J
(mA)
6
5
4
3
2
1
0
Time (hours)
A
B
1.6
1.4
1.2
1.0
0.8
0.6
0.4
0.2
0.0
Photocurrent
J
(mA)
5
4
3
2
1
0
Time (hours)
Fig. S7.
Cross
-section TEM characterization of the Ni/ TiO
2
/p
+
-Si interface. (A) Bright
-field
micrograph showing that the TiO
2
film thickness was measured to be 68.4 ± 0.8 nm. (B)
Selective-
area diffraction pattern at the Ni/TiO
2
interface.
A
B
Ni
TiO
2
Si
68.4 nm
Ni 111
Ni 200
Ni 220
Ni 311, Ni 222
Ni 400
NiO
220
NiO
222
Ni 331
3.383A
NiO
111
17
18
Fig. S8.
Dark
J-E
behavior of a nominally 100 nm Ni film on a 68 nm thick ALD-TiO
2
film on
p
+
-GaAs (black curve) and bare 68 nm ALD TiO
2
/p
+
-GaAs
(red curve) in contact with 50
mM of K
3
Fe(CN)
6
and 350 mM of K
4
Fe(CN)
6
in H
2
O.
200
150
100
50
0
Current density
J
(mA cm
-2
)
1.5
1.0
0.5
0.0
-0.5
E
(V vs. SCE)
200
150
100
50
0
Current density
J
(mA cm
-2
)
1.5
1.0
0.5
0.0
-0.5
E
(V vs. SCE)