International Journal of Hydrogen Energy 31 (2006) 1999 – 2017

www.elsevier.com/locate/ijhydene

Efficiency of solar water splitting using semiconductor electrodes

A.B. Murphya, c , P.R.F. Barnesa, c , L.K. Randeniyaa, c , I.C. Plumba, c,∗ , I.E. Greyb, c ,
M.D. Horneb, c , J.A. Glasscocka, c
a CSIRO Industrial Physics, P.O. Box 218, Lindfield NSW 2070, Australia
b CSIRO Minerals, Box 312, Clayton South VIC 3169, Australia
c CSIRO Energy Transformed Flagship, Australia

Available online 13 March 2006

Abstract
Reliable measurement of the photoconversion efficiency for semiconductor electrodes is essential to the assessment of electrode
performance. In this paper, the influence of the choice of light source on measured photoconversion efficiencies for semiconductor
photoelectrodes is examined. Measurements of efficiency performed under xenon lamp and solar illumination are compared with
efficiencies calculated by integrating the incident photon conversion efficiency (IPCE) over the lamp and solar spectra. It is shown
that use of a xenon lamp as the light source can lead to a large overestimate of the photoconversion efficiency, relative to that
obtained under standard AM1.5 solar illumination. The overestimate is greater when a water filter is fitted to the xenon lamp, and
when a wide-band gap semiconductor such as TiO2 is used as the photoelectrode. Achievable photoconversion efficiencies using
rutile TiO2 are calculated taking into account the losses due to imperfefct absorption, reflection and charge-carrier recombination;
these calculated efficiencies agree with the measurements to within experimental uncertainties. It is demonstrated that many
photoconversion efficiencies presented in the literature are overestimated. It is concluded that reliable estimation of efficiency
under standard conditions is best obtained by measuring the IPCE as a function of wavelength, and integrating over the AM1.5
solar spectrum, or by measuring under sunlight with a similar zenith angle to that of the AM1.5 spectrum.
䉷 2006 International Association for Hydrogen Energy. Published by Elsevier Ltd. All rights reserved.

Keywords: Hydrogen generation; Solar energy; Photoelectrochemistry; Photoelectrolysis; Photocatalysis; Spectral irradiance; Xenon lamp;
Solar spectrum; Semiconducting materials; Titanium dioxide

1. Introduction ratio of the chemical potential energy stored in the

Photoelectrochemical splitting of water to produce
gaseous hydrogen using solar energy has been re-
searched intensively since its initial demonstration in
1972 [1]. The most important figure of merit for a
semiconductor photoelectrode is the photoconversion
efficiency for water splitting, which is defined as the
∗ Corresponding author. CSIRO Industrial Physics, P.O. Box 218,
Lindfield NSW 2070, Australia. Tel.: +61 2 9413 7351;
fax: +61 2 9413 7631.
E-mail address: ian.plumb@csiro.au (I.C. Plumb).
0360-3199/$30.00 䉷 2006 International Association for Hydrogen Energy. Published by Elsevier Ltd. All rights reserved.
doi:10.1016/j.ijhydene.2006.01.014
form of hydrogen molecules to the incident radiative
energy. The benchmark efficiency is 10%, which is
generally considered to be required for commercial
implementation [2].
Most measurements of the photoconversion effi-
ciency are performed under illumination by artificial
light sources. This is convenient for many reasons—the
artificial sources are stationary, and their intensity is
essentially constant with time, while the spectrum and
intensity of solar radiation reaching the ground depends
on the time of day, atmospheric conditions such as
cloud cover, water vapour content and ozone column,
2000 A.B. Murphy et al. / International Journal of Hydrogen Energy 31 (2006) 1999 – 2017
albedo of the surrounding ground, etc. However, the
comparison of photoconversion efficiencies obtained for
different semiconductor materials requires that these ef-
ficiencies are presented for a standard solar spectrum,
usually the AM1.5 global solar spectrum, and artificial
light sources do not accurately replicate such spectra.
An examination of the literature reveals that in most
cases the photoconversion efficiency values presented
are obtained using illumination by a particular artificial
light source. For example, Nozik [3] used a UV source
with radiation between 300 and 400 nm to measure pho-
toconversion efficiencies of up to 4% for TiO2 photo-
electrodes. Giordano et al. [4] obtained an efficiency of
2.7% for a TiO2 electrode doped with platinum using
a mercury lamp. Khan and Akikusa [5] quoted an effi-
ciency for flame oxidised titanium of 2.0%; the publi-
cation indicates a xenon lamp was used, but Akikusa’s
PhD thesis [6] reveals that a mercury–xenon lamp was
used. Akikusa and Khan [7] reported an efficiency of
1.6% for a similar photoelectrode under xenon lamp
illumination. Khan et al. [8] claimed an efficiency of
8.35% for a flame-pyrolysed titanium photoelectrode
with a band-gap extended to 535 nm by carbon doping;
again xenon lamp illumination was used. Mishra et al.
[9] obtained photoconversion efficiencies as high as 3%
for TiO2 photolectrodes using a mercury–xenon lamp
source. Tang et al. [10] reported an efficiency of 2.5%
for a mesoporous anatase photoelectrode using a xenon
lamp. Radecka et al. [11] obtained an efficiency of 1.9%
using a xenon lamp for a mixed anatase and rutile thin
film deposited by radio-frequency sputtering.
As will be shown later, all of these efficiency figures
are close to or higher than those that are thermody-
namically possible for the standard AM1.5 global so-
lar spectrum, given the band gap of the semiconductor
material used and the bias voltage applied. It therefore
seems likely that the efficiencies are larger than those
that would be obtained for sunlight.
In this study, we investigate the influence of the light
spectrum on the photoconversion efficiency for cells
with semiconductor photoelectrodes. We consider in
particular the most widely used source, the xenon arc
lamp. We have had the spectrum of our lamp charac-
terised by an accredited standards laboratory. We exam-
ine the influence on efficiency of the spectrum of the
lamp, both with and without a water filter fitted. Water
filters are widely used to absorb the infrared output of
the lamp, thereby decreasing the heating of the irradi-
ated surfaces. We also investigate the influence of the
age of the lamp tube on the spectrum.
The photoconversion efficiency depends on many
factors in addition to the spectrum of the incident
radiation. These include the band gap of the semicon-
ductor, reflection of radiation before reaching the semi-
conductor, the level of absorption of the radiation in
the semiconductor, and the transport of charge carriers
through the semiconductor. We examine the influence of
each of these factors on the achievable efficiency, both
for solar illumination and xenon lamp illumination.
We have measured the photoconversion efficiency
for a cell with a rutile TiO2 photoelectrode under both
xenon lamp and solar illumination. We have also mea-
sured the incident photon conversion efficiency (IPCE),
which is the efficiency of conversion of photons inci-
dent on the photoelectrochemical cell to photocurrent
flowing between the working and counter electrodes, as
a function of wavelength of the incident radiation. This
allows the photoconversion efficiency to be calculated
for any radiation spectrum. The efficiencies obtained
from the measurements are compared with those calcu-
lated taking into account the spectrum of the radiation,
the band-gap, the absorption and reflection of the radi-
ation, and transport of the charge carriers.
The measurements reported here were obtained at
room temperature (T = 295 K). The effect of increased
temperatures on water splitting efficiencies has been
investigated by Licht [12,13].
In Section 2, we present the details of our mea-
surements, including the characterisation of the light
sources, and the results of these measurements. Our cal-
culations of the efficiency limits for the different light
sources are given in Section 3. The measurements are
compared with the calculations in Section 4; in addition,
the results of other workers are discussed in the light of
our results, and possible means of increasing the pho-
toconversion efficiency of semiconductor electrodes are
considered. Conclusions are presented in Section 5.
2. Experimental details and results
2.1. Sample preparation
The specimen used in this study was prepared by
oxidising a titanium substrate in a methane and oxy-
gen flame. Before oxidation, the titanium sheet (Sigma
Aldrich, 99.7%, 0.25 mm thick) was polished and
then etched in Kroll’s solution (one part 40% HF,
one part 70% HNO3 and three parts water) for 5 s. A
small (6 mm × 6 mm) piece of the sheet was held in a
methane–oxygen flame for five minutes, at a position
at which the flame temperature was measured to be
850 ◦ C using a K-type thermocouple. The oxygen to
methane volume ratio of the gas mixture was 1.15, so
 A.B. Murphy et al. / International Journal of Hydrogen Energy 31 (2006) 1999 – 2017
0.35
Diffuse reflectance
0.30
0.25
0.20
0.15
300
Specular reflectance
0.05
0.04
0.03
0.02
0.01
0.00
300
Fig. 1. Diffuse and specular reflectance of the rutile TiO2 film formed by flame-oxidising a titanium substrate. The reflectance is measured
relative to a matt Teflon standard. The best fit to the specular reflectance (assuming 15% by volume of the TiO2 film is air) is also shown. The
specular reflectance fit is performed only for wavelengths greater than 400 nm, since the strong absorption at shorter wavelengths precludes
the formation of an interference pattern.
the flame was lean with respect to stoichiometry. An
oxide layer was formed on both sides of the titanium
substrate. The side of the substrate exposed directly
to the flame is analysed here. The colour of the oxide
layer was mid-grey, with some red and green colours
indicating the formation of an interference pattern. The
preparation and characterisation of a range of similar
samples have been described in more detail elsewhere
[14].
2.2. Sample characterisation
X-ray diffraction analysis of the sample was per-
formed using a Philips diffractometer with a Cu K

source. The measurements indicated that the oxide layer
had a rutile structure.
The diffuse and total reflectance of the sample
were measured as functions of wavelength from 250
to 800 nm using the diffuse reflectance attachment of
a Cary 5 spectrophotometer. The reflectances were
measured relative to a Teflon standard. The specular
reflectance was obtained by subtracting the diffuse
reflectance from the total reflectance. The diffuse and
specular reflectances are shown in Fig. 1. The diffuse
reflectance curve shows a strong decrease in the re-
flectance at wavelengths below 400 nm, indicating an
absorption edge at around 3.1 eV, close to the band gap
of rutile.
2001
400 500 600 700 800
Measurement
Best fit, h = 280 nm
400 500 600 700 800
Wavelength (nm)
The thickness of the oxide layer was derived from the
interference pattern evident in the specular reflectance
spectrum. This was done using a least squares fit of the
specular reflectance from 400 to 800 nm to the expres-
sion [15] for the reflection coefficient of an absorbing
film of a given thickness on an absorbing substrate. The
real part of the refractive index of rutile was taken from
Cardona and Harbeke [16] and DeVore [17], as reported
by Ribarsky [18] and the imaginary part from the work
of Eagles [19]. Scanning electron microscope images
indicated that the sample had a porosity of 15 ± 10%.
The refractive index was accordingly reduced to allow
for 15 ± 10% by volume of air, using the formula given
by Mergel [20], based on the Bruggeman approxima-
tion. The best fit was obtained for TiO2 film thickness
h = 280 ± 40 nm, with the uncertainty allowing for the
estimated range of air volumes. The best fit curve for
15% air by volume is shown in Fig. 1. The feature of
the fitted curve that determines h is the separation of
the peaks of the interference pattern. The amplitude
of the fitted curve is normalised to have the same stan-
dard deviation from the mean specular reflectance as
the measured curve.
2.3. Light sources
The spectrum of a light source is usually described
by the spectral irradiance E
(
) (units W m−2 nm−1 ),
2002 A.B. Murphy et al. / International Journal of Hydrogen Energy 31 (2006) 1999 – 2017
defined by E
= dES /d
, where

∞
ES = E
(
) d
(1)
0 parameters. Values corresponding to the AM1.5 stan-
is the incident irradiance. The incident spectral photon
flux I
(units m−2 nm−1 s−1 ) is related to the spectral
irradiance by
E
(
)

I
(
) = , (2)
hc
where h is Planck’s constant and c is the speed of light in
a vacuum. The spectral photon flux is the more relevant
quantity in water splitting applications, since at most
one electron–hole pair is produced per absorbed photon,
and photon energy above the required energy to split a
water molecule is lost.
2.3.1. AM1.5 global solar spectrum
The standard reference solar spectra for calculating
the conversion efficiency of solar energy to electrical
or chemical energy are the AM1.5 spectra, defined by
the American Society for Testing and Materials [21].
The AM, or air mass, factor characterises the effect of
the Earth’s atmosphere on the solar radiation, and is
given by AM=1/ cos , where  is the solar zenith angle
(the angle between the overhead and actual position of
the sun).
Two AM1.5 spectra are defined. They describe re-
spectively the direct, and the total, or global, spectral ir-
radiance at air mass 1.5 under standard atmospheric and
surface conditions. AM1.5 corresponds to a solar zenith
angle of 48.2◦ , at which angle the sun’s radiation has
to pass through 1.5 times the thickness of atmosphere
relative to a solar zenith angle of 0◦ . The receiving sur-
face is inclined at 37◦ , so the incidence angle is 11.2◦ .
The global AM1.5 spectrum includes radiation incident
from all angles, and therefore includes scattered light.
The direct AM1.5 spectrum includes only the compo-
nent of the global spectrum that is directly incident from
the sun. The global spectrum is the relevant quantity
for solar water splitting experiments, since both direct
and scattered light will be incident on a water-splitting
reactor.
2.3.2. Solar spectrum for efficiency measurement
The spectrum of sunlight at the time and location of
photoconversion efficiency measurement was calculated
using the SMARTS code, version 2.9.2 [22,23]. (This is
the same code that has been used to derive the standard
AM1.5 spectra.) The measurement was made in Syd-
ney, Australia (latitude 33.87◦ S, longitude 151.22◦ E) at
midday on 28 May 2004. There was no cloud cover.
The zenith and azimuthal angles of the sun were, re-
spectively, 55.4◦ and 357.8◦ . The surface of the detec-
tor was tilted to face the sun. The calculation uses these
dard spectrum were used for all other parameters (e.g.,
atmospheric composition and surface albedo).
The calculated solar spectrum is similar to the AM1.5
global solar spectrum. The spectral irradiance of the
calculated spectrum is slightly lower at ultraviolet
and visible wavelengths, and similar at infrared wave-
lengths. The total irradiance of the calculated spectrum
was 962 W m−2 , consisting of 835 W m−2 direct and
127 W m−2 diffuse irradiance. The respective figures
for the AM1.5 global spectrum are 1000 W m−2 , con-
sisting of 900 W m−2 direct and 100 W m−2 diffuse
irradiance.
2.3.3. Xenon arc lamp
We used an Oriel 6271 ozone-free xenon lamp, with
an Oriel 61945 water filter fitted. The water filter was
filled with deionised water, and removed infrared ra-
diation above about 1000 nm and ultraviolet radiation
below about 250 nm. An Oriel digital exposure control
was used to maintain the total radiant power of the lamp
constant.
The spectral irradiance of the xenon lamp was mea-
sured by the Australian National Measurement Institute
[24]. Measurements were performed using a photomul-
tiplier detector for wavelengths from 240 to 400 nm, a
silicon detector from 300 to 1010 nm, an indium gallium
arsenide detector from 900 to 1680 nm, and a lead sul-
phide detector from 1300 to 2500 nm. The use of over-
lapping regions ensured that the measurements made
using the different detectors were comparable. The mea-
surement intervals used were 2 nm for wavelengths be-
low 400 nm, 5 nm for wavelengths between 400 and
1200 and 10 nm for longer wavelengths.
Spectra were measured both with and without the wa-
ter filter fitted to the xenon lamp. While the water filter
was used in our efficiency measurements, calculations
for both spectra are presented in order to broaden the
relevance of our results.
The spectrum of the lamp without the water filter ex-
tends to wavelengths longer than 2.5 m, which is the
limit for which the measurements could be performed.
However, by measuring the total radiant power flux with
a thermopile, and comparing this to the power obtained
by integrating the lamp spectrum to 2.5 m, it was de-
termined that less than 5% of the power was at wave-
lengths longer than 2.5 m.
Fig. 2 shows the spectral photon flux of the xenon
lamp with and without the water filter. Both spectra have
 A.B. Murphy et al. / International Journal of Hydrogen Energy 31 (2006) 1999 – 2017 2003

35 Xenon lamp, without water filter

Xenon lamp, with water filter

Spectral photon flux (1018 m-2 s-1 nm-1)

AM1.5 global sunlight
30
6

25 5

4
20
3

15 2

1
10
0
200 300 400 500 600 700
5

0
200 400 600 800 1000 1200 1400 1600 1800 2000 2200 2400
Wavelength (nm)

Fig. 2. Spectral photon flux for the xenon lamp, with and without the water filter, compared to the AM1.5 global solar spectrum. The lamp
had logged 1250 h of operation when the spectrum was recorded. The inset shows detail of short wavelengths. In all cases, the total irradiance
is normalised to 1000 W m−2 . Measurements were performed by the National Measurement Institute.

been normalised to a total irradiance of 1000 W m−2 , of a replacement tube. These measurements were
to facilitate comparison with the AM1.5 global solar performed by measuring the lamp irradiance with a
spectrum. thermopile, after passing the light through a monochro-
The lamp spectra are almost identical for wavelengths mator with 12 nm pass band. The thermopile is
up to around 900 nm, apart from the influence of the described in Section 2.4. The measurements were per-
normalisation factor. The water filter absorbs a signif- formed with the water filter fitted to the lamp, and the
icant fraction of the light at wavelengths between 900 total irradiance was normalised to 1000 W m−2 in each
and 1100 nm, and all of the radiation at longer wave- case. To check the method of measurement, the spec-
lengths. In total, the filter absorbs about 50% of the lamp trum was also measured for the tube with 1250 h of op-
irradiance; hence when the spectra were normalised to eration; the results are almost identical to those shown
the same total irradiance, the spectral irradiance of the in Fig. 2 apart from the poorer wavelength resolution.
lamp without the filter is approximately 50% lower for The spectral photon flux for the three cases is shown
wavelengths less than 900 nm. in Fig. 3. The new tube has a greater proportion of the
Fig. 2 indicates that at short wavelengths, the spec- photon flux in the ultraviolet. Towards the end of tube
tral photon flux of the xenon arc lamp is greater than life, the proportion of the photon flux in the ultraviolet
that of AM1.5 global sunlight. The lamp emits a signifi- decreases rapidly. Unless otherwise noted, the results
cant photon flux for wavelengths from 250 nm, whereas presented in the remainder of the paper pertain to the
the solar photon flux is minimal at wavelengths below spectra measured for the tube with 1250 h of operation.
300 nm. The enhanced photon flux at short wavelengths Other lamps, such as non-ozone-free xenon arc
is greater for the lamp with the water filter, as a conse- lamps, mercury–xenon arc lamps, and mercury lamps,
quence of the normalisation. are also used for measurement of photoconversion ef-
The xenon lamp spectrum is similar to that presented ficiencies. Such lamps emit more strongly in the UV
by the manufacturer [25]. However, arc lamp spectra range than ozone-free xenon lamps [25].
vary from lamp to lamp, depending on the age of the
tube and other variables. The tube used here had logged 2.4. Measurement of efficiency
1250 h of operation before calibration. To investigate
the extent of the variability of the lamp spectrum, we Two light sources were used for the water-splitting
also measured the spectrum of the lamp near the end efficiency measurements, the xenon lamp with water
of the tube life (1337 h), and near the start of the life filter, and sunlight (at midday on 28 May 2004 in
2004 A.B. Murphy et al. / International Journal of Hydrogen Energy 31 (2006) 1999 – 2017

10

Spectral photon flux (1018 m-2 s-1 nm-1)

9

8 Tube 2, 5.7 h of operation

Tube 1, 1250 h of operation
7 Tube 1, 1337 h of operation
6

0
200 300 400 500 600 700 800
Wavelength (nm)

Fig. 3. Spectral photon flux for the xenon lamp with the water filter, for two different tubes and for three different tube ages. In all cases,
the total irradiance is normalised to 1000 W m−2 . Measurements were performed using a monochromator with a 12 nm bandpass.

potentiostat reference electrode. A Utah Electronics Model 0152 po-

A tentiostat was used to control the voltages. The elec-
trodes were immersed in 5M KOH electrolyte. A glass
-V W-R
frit separated the counter electrode from the working
and reference electrodes. Gaseous oxygen was purged
W R C from the counter electrode compartment using a con-
VW-C
tinuous flow of bubbled nitrogen gas. The presence of
N2(g)
the glass frit did not make a significant difference to
the photocurrent. However, allowing oxygen to dissolve
in the electrolyte surrounding the counter electrode led
to an increased photocurrent at a given cell bias volt-
quartz window
working age (VW.C in Fig. 4), by up to a factor of two at low
hv Pt
SCE
reference counter
bias voltages. This is attributed to the reduction of oxy-
porous glass frit gen molecules rather than the generation of hydrogen
molecules at the counter electrode. Measurements per-
Fig. 4. Schematic of the photoelectrochemical cell used for xenon formed in an oxygen-rich electrolyte can thus lead to
lamp measurements. A simplified equivalent circuit of the potentio- erroneously large efficiency values.
stat is included. VW.C and VW.R denote voltmeters measuring the The cell was constructed of glass, with a quartz win-
voltages between the working and counter-electrodes, and the work- dow to allow the incident radiation to enter with minimal
ing and reference electrodes, respectively. A denotes an ammeter
attenuation. The working electrode was placed 10 mm
measuring the photocurrent.
from the window. Swapping the positions of the work-
ing and reference electrodes had a negligible effect on
Sydney, Australia). Details of these sources were given the photocurrent, showing that attenuation by the elec-
in Section 2.3. trolyte and the resistance of the electrolyte were both
The xenon lamp measurements were performed using insignificant.
a standard three-electrode photoelectrochemical cell, The sunlight measurements were also performed us-
shown in Fig. 4. The three electrodes were the work- ing a three-electrode cell. In this case, the electrodes
ing electrode (the oxidised titanium sample), a plat- were immersed in 5M KOH electrolyte in a shallow
inum wire counter electrode, and a saturated calomel tray. Sunlight was directly incident on the electrolyte,
 A.B. Murphy et al. / International Journal of Hydrogen Energy 31 (2006) 1999 – 2017
and the working electrode was oriented so that it faced
the sun. The counter and reference electrodes were the
same as for the xenon lamp measurements, and a flow
of nitrogen gas was again used to purge oxygen from
the counter electrode.
The irradiance was measured using an Oriel 71751
sapphire window thermopile. The thermopile was cali-
brated against a standard thermopile by the Australian
National Measurement Institute, using a standard lamp
with filament temperature of approximately 2700 K as
the light source. The uncertainty in the calibration is
±2%; spatial variations in the incident light intensity
increased the uncertainty in irradiance measurement to
10%.
The cell efficiency  for the water splitting reaction
was determined using
 = jP (VWS − VB )/ES , (3)
where jP is the photocurrent produced per unit ir-
radiated area, VWS = 1.229 V is the potential corre-
sponding to the Gibbs free energy change per photon
required to split water, and VB is the bias voltage
applied between the working and counter electrodes
(VW.C in Fig. 4). The irradiance ES was measured
at a position outside the cell. The quartz window
used in the xenon lamp measurements reflects 4%
of this irradiance. No correction is made for this
loss, since a window is likely to be required in a
working water-splitting reactor. In the solar mea-
surements, no window is used; however, the water
is at an angle to the direct solar radiation equal to
the solar zenith angle of 55◦ . We calculate that 7%
of the direct solar radiation will be reflected by the
water. To ensure comparability with the lamp mea-
surements, the measured solar irradiance was reduced
by 3%.
Some workers [5,7,8,11] have used the voltage be-
tween the working and reference electrodes VW.R (rela-
tive to the open-circuit value of this voltage) in place of
the bias voltage VB . This is often the only voltage that
is reported. However, the bias voltage is applied across
the working and counter electrodes, and the photocur-
rent flows between these electrodes, so the electrical
power loss that has to be subtracted in calculating the
efficiency of the cell is the product of the photocurrent
and the voltage applied across the working and counter
electrodes. Typically VW.R < VB , so using VW.R instead
of VB leads to an overestimate of the cell efficiency.
The photocurrent was measured for a range of ap-
plied bias voltages, and the photoconversion efficiency
calculated using Eq. (3). The irradiance for the xenon
lamp illumination was 820 ± 80 W m−2 . The solar
2005
irradiance was measured to be 880 ± 90 W m−2 . This
agrees, within the uncertainty of measurement, with the
960 W m−2 calculated in Section 2.3.2.
Fig. 5 shows the measured photoconversion efficien-
cies for different bias voltages. The maximum efficiency
was found to be 0.54% at a bias voltage of 0.6 V for
xenon lamp illumination, and 0.28% at the same bias
voltage for solar illumination. The experimental uncer-
tainty for the measured photoconversion efficiencies is
estimated to be ±20%. The random repeatability er-
ror, which is a consequence of variation in factors such
as electrolyte purity, working electrode alignment, and
time for the photocurrent to reach equilibrium, was ap-
proximately 15%. The largest measurement error was
the 10% uncertainty in the irradiance measurements by
the thermopile.
2.5. Measurement of IPCE
The IPCE provides a measure of the efficiency of con-
version of photons incident on the photoelectrochem-
ical cell to photocurrent flowing between the working
and counter electrodes. An IPCE of 100% corresponds
to the generation of one photoelectron for each incident
photon. Losses associated with the reflection of inci-
dent photons, their imperfect absorption by the semi-
conductor, and recombination of charge carriers within
the semiconductor before they reach the electrolyte, all
contribute to the IPCE values falling below 100%.
The IPCE was measured by interposing a monochro-
mator between the xenon lamp and the photoelectro-
chemical cell. Similar results were obtained both with
and without the use of a water filter. The bandpass of
the monochromator was 12 nm (FWHM), which is suf-
ficiently narrow to have insignificant effect on the mea-
sured IPCE values. The irradiance within the pass band
was measured using the thermopile, averaging the mea-
surement over a 20 s period. The IPCE was then calcu-
lated as a function of wavelength using
IPCE(
) = jP (
)/[eI (
)], (4)
where I (
) is the incident photon flux passed by the
monochromator when set to wavelength
, and e is the
electronic charge. The uncertainty of measurement was
calculated by combining the standard error on the time
average of the irradiance with the errors noted for effi-
ciency measurements in Section 2.4.
Fig. 6 shows IPCE(
) for the TiO2 photoelectrode at
an applied bias voltage of 0.60 V, at which the maxi-
mum photoconversion efficiency is obtained. The IPCE
reaches a maximum of just under 60% for a wave-
length of 320 nm. The decrease at shorter wavelengths
2006 A.B. Murphy et al. / International Journal of Hydrogen Energy 31 (2006) 1999 – 2017

0.7
Xenon lamp with water filter
Sunlight, 28 May 2004, midday
0.6

Photoconversion efficiency (%)

0.5

0.4

0.3

0.2

0.1

0.0
0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2
Bias voltage (V)

Fig. 5. Photoconversion efficiency for the cell with TiO2 electrode as a function of bias voltage VB . Results are given for illumination by the
xenon lamp with water filter, and the sun.

70

60

50

40
IPCE (%)

30

20

10

0
300 320 340 360 380 400 420 440
Wavelength (nm)

Fig. 6. Measured incident photon conversion efficiency, expressed as a percentage, as a function of wavelength for the TiO2 photoelectrode.

is associated with an increase in reflection from the pho- assuming that the photocurrent is not dependent on the
toelectrode (see Section 3.3). At longer wavelengths, irradiance. The photoconversion efficiency can then be
weaker absorption of photons and the recombination of determined using
charge carriers lead to a decreased IPCE. These factors
∞
are considered in detail in Section 3.4.  = e(VWS − VB ) IPCE(
)I
(
) d
/ES . (5)
Using the IPCE measurements, it is possible to cal- 0
culate the photocurrent expected to be generated for The cell photoconversion efficiency for rutile TiO2 cal-
illumination by light with any spectral dependency, culated for different spectra is given in Table 1.
 A.B. Murphy et al. / International Journal of Hydrogen Energy 31 (2006) 1999 – 2017
Table 1
Photoconversion efficiencies for cells with rutile TiO2 photoanodes for different incident light sources
Spectrum Photoconversion efficiency
From IPCE (Eq. (5)) (%)
AM1.5 global sunlight 0.25 ± 0.05
Sunlight, midday, 28 May 2004, Sydney 0.21 ± 0.04
Xenon lamp without water filter (1250 h) 0.32 ± 0.07
Xenon lamp with water filter (1250 h) 0.51 ± 0.10
Xenon lamp with water filter (5.7 h) 0.66 ± 0.15
Xenon lamp with water filter (1337 h) 0.30 ± 0.06
Comparison of directly measured values with efficiencies calculated from the measured IPCEs integrated over the different source spectra.
The values given in Table 1 for sunlight and the xenon
lamp with the water filter can be compared to the pho-
toconversion efficiencies measured directly with those
sources. These efficiencies, given in Section 2.4, are re-
peated in Table 1 and are equal to the values derived
from the IPCE measurements within the experimental
uncertainty.
The age of the xenon lamp tube, and the presence of
the water filter, can strongly influence the measured cell
photoconversion efficiency for rutile TiO2 . In particular,
the efficiency determined using the lamp tube near the
end of its lifetime (1337 h of operation) is much less
than determined using a new tube or a tube that has
been operated for 1250 h. The presence of a water filter
increases the measured efficiency by around 50%.
3. Calculated photoconversion efficiency
There are fundamental thermodynamic limits to
the photoconversion efficiency, which depend on the
band-gap of the semiconductor being used, and on the
spectrum of the incident radiation. In practice, these
limiting efficiencies cannot be achieved, due to losses
such as the incomplete absorption of the incident ra-
diation in the semiconductor, reflection of the incident
radiation from the surface of the semiconductor and
other surfaces, and recombination of conduction band
electrons before reaching the electrolyte. In this sec-
tion we consider first the thermodynamic limits to
efficiencies, and then the relevant loss factors.
3.1. Thermodynamic limits to efficiency
We consider here an idealised semiconductor, in
which all photons of energy greater than the band-
gap energy Ug are absorbed. The resulting conduction
band electrons are transferred to the electrolyte, and a
2007
Directly measured (Eq. (3)) (%)
0.28 ± 0.06
0.54 ± 0.11
0.66 ± 0.12
hydrogen molecule is produced for each pair of such
electrons. The absorbed photon flux is then


g
JS = I
(
) d
(6)
0
where
g =hc/Ug is the band-gap wavelength. An upper
bound to the photoconversion efficiency is given by
⎧
⎨ JS G (1 − loss )
0
if Ug
G0 + Uloss ,
C = ES (7)
⎩
0 if Ug < G0 + Uloss ,
where G0 =eV WS =1.229 eV is the Gibbs free energy
change per electron for the water-splitting reaction. The
factor loss represents the radiative quantum yield; i.e.,
the ratio of re-radiated photons to absorbed photons.
Hence 1−loss gives the proportion of absorbed photons
converted into conduction band electrons. The reradia-
tion of photons is due to blackbody radiation from the
excited state. Bolton et al. [26] have calculated that the
value of loss that corresponds to a maximum value of
the efficiency is given by
loss P ≈ 1/ ln(JS /JBB ), (8)
where JBB is the blackbody photon flux at wavelengths
below the band-gap wavelength. loss P is small, less
than 0.02 for the range of band-gap wavelengths of in-
terest, and is practically independent of the spectrum of
the incident radiation.
The term Uloss is the energy lost per photon. This is
a significant loss term, and provides a lower limit to the
band-gap energy of a semiconductor that can be used
for water splitting. There are two major contributions
to Uloss . The first is thermodynamic, and can be shown
[26] to be equal to T Smix , where T is the temperature
and Smix is the increase in entropy of mixing associ-
ated with production of conduction band electrons by
photon absorption. This contribution is at least 0.4 eV.
2008 A.B. Murphy et al. / International Journal of Hydrogen Energy 31 (2006) 1999 – 2017

35

30

Xenon lamp, without water filter

Maximum efficiency (%)

25 Xenon lamp, with water filter
AM1.5 global sunlight
20

15

10

Maximum
5 band-gap
wavelength
0
200 250 300 350 400 450 500 550 600 650 700 750
Band-gap wavelength (nm)

Fig. 7. The maximum photoconversion efficiency possible as a function of semiconductor band-gap wavelength. Results are given for AM1.5
global solar illumination, and illumination by the xenon arc lamp with and without the water filter. The maximum band-gap wavelength of
610 nm is indicated; materials with greater band-gap wavelengths cannot be used to split water.

The second contribution is a kinetic loss due to the over-
potentials for oxygen and hydrogen production. This
has been estimated [27] to be at least 0.4 eV. There-
fore, we estimate Uloss
0.8 eV, which is in accordance
with Bolton et al.’s assessment [28]. This increases the
minimum band gap energy from G0 = 1.23 eV to
G0 +Uloss 2.03 eV, corresponding to a limiting band-
gap wavelength of 610 nm.
Note that there are two other loss factors implicit in
Eq. (7). First, although only photons with energy greater
than Ug can be utilised to excite electrons from the va-
lence band to the conduction band, the fraction of the
photon energy greater than Ug is lost through rapid ther-
mal equilibration with the semiconductor lattice. Sec-
ond, even though Ug has to be significantly greater than
the energy G0 that is ultimately stored as chemical
energy, the difference between Ug and G0 is lost.
Fig. 7 shows the maximum efficiency for water split-
ting, calculated using Eq. (7), with loss =loss P . A cut-
off band-gap wavelength corresponding to Uloss =0.8 eV
is indicated on the graph. The efficiency increases with
band-gap wavelength, since more photons with energies
greater than the band-gap energy are available as the
band-gap wavelength increases. The relative efficien-
cies of different radiation sources at a given band-gap
wavelength depend on the proportion of the photon flux
at wavelengths shorter than the band-gap wavelength.
For band-gap wavelengths below about 410 nm, the
maximum efficiency is greater for incident radiation
from the xenon lamp, both with and without a water
filter, than for AM1.5 solar radiation. At band-gap
wavelengths above 410 nm, the maximum efficiency
obtainable with the xenon lamp without a water filter
falls below that of AM1.5 solar radiation, while the
maximum efficiency for the xenon lamp with a water
filter remains larger for all band-gap wavelengths.
Table 2 gives the maximum photoconversion efficien-
cies obtainable for a range of semiconductors of inter-
est in water splitting, for the different radiation sources.
It can be seen that the maximum photoconversion effi-
ciencies for all materials are overestimated when using
a xenon lamp fitted with a water filter. The overesti-
mate is particularly severe for wide band gap materials
such as SrTiO3 and TiO2 . Use of a xenon lamp without
a water filter leads to overestimates of the efficiencies
of wide band-gap materials, and underestimates of the
efficiencies of narrow band-gap materials.
The efficiencies under xenon lamp illumination and
the efficiency ratios will vary depending on the tube
age and other characteristics of the xenon lamp. Using
the spectra shown in Fig. 3 we calculate that the max-
imum efficiency obtainable for cells using rutile TiO2
electrodes is 4.8% for the new tube (5.7 h operation),
and 2.3% for the tube at the end of its lifetime (1337 h
operation). These compare with 3.8% for the tube that
has been used for 1250 h. Hence, the ratio of the AM1.5
 A.B. Murphy et al. / International Journal of Hydrogen Energy 31 (2006) 1999 – 2017 2009

Table 2
Maximum possible photoconversion efficiencies for semiconductors of different band gaps, for AM1.5 global solar illumination, and illumination
by a xenon arc lamp with and without a water filter fitted

Material Band-gap Band-gap Maximum efficiency (%) Ratio of efficiencies

energy (eV) wave-length (nm) AM1.5 Xe lamp Xe lamp AM1.5/Xe lamp AM1.5/Xe lamp
without filter with filter without filter with filter

SrTiO3 3.70 335 0.22 0.69 1.0 0.32 0.22

Anatase TiO2 3.20 388 1.3 1.7 2.6 0.77 0.49
Rutile TiO2 3.00 413 2.2 2.3 3.8 0.98 0.60
WO3 2.70 459 4.8 3.7 6.2 1.32 0.78
CdS 2.40 517 9.1 6.0 10.4 1.52 0.87
TaON 2.38 520 9.3 6.1 10.7 1.53 0.88
C-doped TiO2 [8] 2.32 535 10.5 6.7 11.7 1.57 0.90
Fe2 O3 2.20 564 12.9 7.9 14.0 1.63 0.92
Ta3 N5 2.07 600 15.9 9.6 16.9 1.67 0.94
Hypothetical ideal material 2.03 610 16.8 10.0 17.7 1.68 0.95

efficiency to that for the xenon lamp with water filter
can vary from 0.95 to 0.45 for rutile TiO2 , depending
on the tube age. For most of the tube lifetime, the ratio
is between 0.45 and 0.60.
3.2. Effect of imperfect absorption on efficiency
The results presented in Section 3.1 were calculated
assuming the semiconductor absorbs all radiation at
wavelengths below the band-gap wavelength; i.e., it
was assumed that the absorbed photon flux is given by
Eq. (6). The absorption coefficients of real semiconduc-
tors in fact decrease as the wavelength approaches the
band-gap wavelength from below. To take this into ac-
count, the absorbed photon flux should be calculated
using


g
JS = (
)I
(
) d
,
0
where
(
) = 1 − exp[−k(
)h]
is the fraction of incident radiation absorbed in the
wavelength band from
to d
, k(
) is the absorption
coefficient and h is the thickness of the semiconductor.
It is assumed that absorption only takes place at wave-
lengths shorter than the band-gap wavelength
g of the
semiconductor.
In this section, we investigate the effect of tak-
ing into account real absorption coefficients, using
the data of Eagles [19] for the absorption coefficient
of single-crystal rutile. The absorption coefficient is
6.9 × 107 m−1 at 300 nm, but falls by almost four or-
ders of magnitude to 9.1 × 103 m−1 at the band-gap
wavelength of 413 nm.
Fig. 8 shows the absorbed spectral photon flux
(
)I
(
) for different thicknesses of rutile, calculated
using Eq. (9), for AM1.5 global solar radiation. It can
be seen that a thickness of about 1 mm is required
for essentially all photons with wavelengths below the
band-gap wavelength to be absorbed. Fig. 9 shows the
maximum photoconversion efficiency, calculated using
Eq. (7), for rutile semiconductors of different thick-
nesses, taking into account the imperfect absorption.
Results are given for AM1.5 global solar radiation and
for the xenon lamp with and without a water filter. The
band-gap energy is assumed to be 3.0 eV (correspond-
ing to wavelength 413 nm), for which C = 2.25% for
AM1.5 sunlight, if all radiation at wavelengths below
the band-gap wavelength is absorbed. Fig. 9, shows
(9) that this maximum efficiency is only obtained for thick-
nesses approaching 1 mm. For a thickness of 1 m,
the efficiency is 1.2%, and for a thickness of 0.1 m,
the efficiency is only 0.65%. For xenon lamp illumi-
(10) nation, particularly with the water filter, the efficiency
is greater, owing to the greater proportion of the more
strongly absorbed UV radiation.
The calculations presented here do not take into ac-
count reflection from the rear surface of the TiO2 layer.
The reflection coefficient from this surface is calcu-
lated to be about 30% for the titanium substrate used in
the experiments. It is therefore likely that the calcula-
tions presented here slightly underestimate the absorbed
power, particularly under conditions when a substantial
proportion of the incident radiation is transmitted by the
TiO2 layer.
2010 A.B. Murphy et al. / International Journal of Hydrogen Energy 31 (2006) 1999 – 2017

3.0
1 nm
10 nm

Spectral photon flux / 1018 (m-2 nm-1 s-1)

2.5 100 nm
1 µm
10 µm
2.0 100 µm
1 mm
AM1.5 global sunlight
1.5

1.0

0.5

0.0
300 320 340 360 380 400 420
Wavelength (nm)

Fig. 8. Wavelength dependence of the absorbed spectral photon flux for different thicknesses of rutile, for AM1.5 global solar illumination.

4.0
Xe lamp without water filter
3.5 Xe lamp with water filter
AM1.5 global sunlight
3.0

2.5
Efficiency (%)

2.0

1.5

1.0

0.5

0.0
10-9 10-8 10-7 10-6 10-5 10-4 10-3
Rutile thickness (m)

Fig. 9. Dependence of maximum photoconversion efficiency on thickness of rutile layer, taking into account absorption of photons in the
rutile. Results are given for AM1.5 global solar illumination, and illumination by the xenon arc lamp with and without the water filter.

3.3. Effect of reflection on efficiency For collimated radiation passing from medium 1 to
medium 2 at normal incidence to the boundary between
A proportion of the incident light is reflected at any the media, the reflection coefficient is given by
interface at which the refractive index changes. In a typ-
ical water-splitting reactor, a semiconductor electrode (n2 − n1 )2 + (2 − 1 )2
is immersed in an aqueous solution of a salt, an acid R12 = , (11)
(n2 + n1 )2 + (2 + 1 )2
or a base, and light enters through a quartz window.
Therefore reflection occurs at the interface between the where Ni = ni + ii is the complex refractive indices of
air and the quartz, the quartz and the solution, and the medium i. Here ni is the real part of the refractive index
solution and the semiconductor. (usually referred to as simply the refractive index) and
 A.B. Murphy et al. / International Journal of Hydrogen Energy 31 (2006) 1999 – 2017
0.40
0.35
0.30
Reflection coefficient 0.25
0.20
0.15
0.10
0.05
0.00
250
Fig. 10. Reflection coefficients at the interfaces between air and quartz glass, quartz glass and water, and water and rutile TiO2 . The coefficients
are for specular reflection at normal incidence.
i is the extinction coefficient, related to the absorption
coefficient k by k = 4i /
.
We consider here as an example the reflection losses
for a TiO2 electrode immersed in water, with a quartz
window between the water and the air. This was the
arrangement used for the xenon lamp measurements. It
was noted in Section 2.4 that the solar measurements
were performed in a vessel with no quartz window;
however, it is likely that reactors will use such a window.
Refractive index data for water and quartz were taken
from the SOPRA optical constant database [29], and for
rutile TiO2 as noted in Section 2.2. The TiO2 absorption
coefficient was assumed to be constant at wavelengths
below 300 nm, as suggested by the data of Eagles [19].
The relevant reflection coefficients, calculated using
Eq. (11), are shown in Fig. 10. Reflection losses at
the interface between the quartz glass and water are
very low, and those between the air and quartz glass
are about 4%. The reflection losses at the TiO2 surface
are substantial, reaching more than 30% at short wave-
lengths. It is interesting to note that while large absorp-
tion coefficients are required to minimise absorption
losses, they have the undesirable effect of increasing
reflection losses.
The influence of reflection on the efficiency of con-
version of solar energy to stored chemical energy can
be calculated by modifying the absorbed photon flux
JS :


g
JS = [1 − R(
)](
)I
(
) d
,
0 air–quartz, quartz–water and water–rutile interfaces,
2011
Air-quartz
Quartz-water
Water-TiO2
Total
300 350 400 450 500 550 600
Wavelength (nm)
where R(
) is the total reflection coefficient (i.e., the
sum of the individual reflection coefficients). Fig. 11
shows the influence of reflection on the photoconver-
sion efficiency for different thicknesses of TiO2 . Reflec-
tion leads to a decrease in efficiency of between 24%
and 33%, depending on the thickness of the TiO2 . The
decrease is essentially independent of the light source.
Reflection reduces the maximum efficiency for AM1.5
sunlight to 1.7% for a thick rutile layer, and to 0.45%
for a 0.1 m layer.
There are several sources of uncertainty in the reflec-
tion coefficient calculations. First, the refractive index
of water has been used instead of the refractive index of
a basic solution. However, this makes only a small dif-
ference to the calculated reflection coefficients. The re-
fractive index of water at 589 nm is 1.333, while that of
5M KOH is 1.376 [30]. The total reflection coefficient
at 589 nm is reduced from 15.9% to 14.8% when the re-
fractive index of water is replaced by that of 5M KOH.
Second, the reflection coefficients have been cal-
culated for normal incidence. Reflection coefficients,
and therefore losses, increase as the angle of incidence
(which is zero at normal incidence) increases. For
global solar irradiation, a proportion of the radiation
is scattered from the atmosphere, and therefore has a
non-zero angle of incidence even when the electrode is
facing the sun. Hence, reflection losses will be larger
than those presented in Fig. 11. We estimated the
influence of this by calculating diffuse reflection coef-
ficients (averaged over all angles of incidence) for the
(12)
2012 A.B. Murphy et al. / International Journal of Hydrogen Energy 31 (2006) 1999 – 2017
4.0 Xe lamp without filter
Xe lamp with filter
3.5 AM1.5 global solar
3.0
2.5
Efficiency (%)
2.0
1.5
1.0
0.5
0.0
10-9 10-8 10-7 10-6
Rutile thickness (m)
Fig. 11. Dependence of photoconversion efficiency on thickness of rutile layer, taking into account reflection from the quartz, water and rutile,
and absorption of photons in the rutile. Results are given for AM1.5 global solar illumination, and illumination by the xenon arc lamp with
and without the water filter. Also shown is the ratio of photoconversion efficiency achievable including and excluding reflection.
taking into account refraction of the rays at the first two
interfaces. The reflection at the first interface increases
significantly, resulting in an increase in the total reflec-
tion coefficient by a factor of about 30%. We applied
these reflection coefficients to the diffuse component of
the global AM1.5 spectrum, and the normal incidence
reflection coefficients to the remainder of the spectrum.
The photoconversion efficiency decreased by a factor
of between 9% (for a 1 nm thick rutile layer) and 6%
(for a 1 mm thick rutile layer). Again, these changes
are relatively small.
Finally, the reflection coefficients have been calcu-
lated assuming that the TiO2 layer has a smooth surface,
while the surface is rough, which may reduce the to-
tal reflectance, integrated over a hemisphere. Scanning
electron microscope images show that the rms rough-
ness of the surface is much less than 100 nm. Calcula-
tions using a physical optics model [31] indicate that
the total reflectance from such a surface is at least 97%
of that from a smooth surface.
3.4. Calculation of diffusion length of charge carriers
from IPCE measurements
The measured incident photon conversion efficiency
for the TiO2 electrode, shown in Fig. 6 was used in
Section 2.5 to calculate the photoconversion efficiency
under illumination by different light sources. In this
section, we use the measured IPCE to estimate the dif-
Efficiency with reflection / efficiency without reflection
1.0
0.8
0.6
0.4
0.2
0.0
10-5 10-4 10-3
fusion length of charge carriers in the TiO2 . We use
the simple Schottky-barrier model presented by Ghosh
and Maruska [32,33], in which the semiconductor layer
of thickness h is assumed to behave like a conven-
tional Schottky barrier. There is a barrier region (carrier
depletion, or space charge, region) of width lb at the
semiconductor–electrolyte interface. All carriers pro-
duced in the barrier region, or that reach the barrier re-
gion by diffusion, are assumed to reach the electrolyte.
Carriers produced outside the barrier region (i.e., a dis-
tance greater than lb from the semiconductor–electrolyte
interface) have to diffuse through the bulk of the semi-
conductor, with diffusion length L, to reach the barrier
region. A similar model was used by Butler [34].
The number of carriers produced per unit time within
the interval dx at a distance x from the surface is pro-
portional to GI
k exp(−kx) dx, where G is the quan-
tum conversion efficiency. Assuming that G is constant
with respect to x and writing  = 1/L gives

k
IPCE = G [1 − exp(−kl b )] + exp(lb )
k+

× {exp[−(k + )]lb − exp[−(k + )]h} .
(13)
A least-squares fit of the IPCE, calculated using (13),
to the measured data shown in Fig. 6 was performed.
The three terms that were fitted were L, lb and G. The
measured IPCEs were divided by (1−R) to compensate
 A.B. Murphy et al. / International Journal of Hydrogen Energy 31 (2006) 1999 – 2017
100
90
80
70
60
IPCE (%)
50
40
30
20
10
0
320 340
Wavelength (nm)
Fig. 12. Measured IPCE at 0.6 V bias voltage, modified to compensate for reflection losses, for the flame-oxidised TiO2 electrode, with the
best fit of the Schottky barrier diode model to the measurements. The contribution of the barrier region and the bulk region to the IPCE are
shown for the best fit. Also shown is , the percentage of the radiation absorbed in the TiO2 layer.
for reflection losses. The thickness h = 280 ± 40 nm
measured by reflectometry was used. The best-fit values
of lb , L and G were found to be 8 ± 7 nm, 180 ±
30 nm and 0.96±0.05, respectively, for h=280 nm. The
uncertainties reflect the values for which the fit becomes
noticeably less accurate. The values are only weakly
dependent on h; for example, varying h by ±40 nm
altered L by only ±10 nm.
The calculated dependence of IPCE on wavelength
is shown in Fig. 12 together with the components due
to photons absorbed in the barrier region (x lb ) and
the bulk region (lb < x L). These components corre-
spond respectively to the first and second terms on the
right-hand side of Eq. (13). The agreement between
the calculation and the measured data is reasonable for
wavelengths up to 380 nm; it is not clear whether the
discrepancy at longer wavelengths is a consequence of
deviations of the absorption coefficient from the single-
crystal rutile data that was used, or deficiencies in the
simple Schottky barrier model. Also shown in Fig. 12 is
(
), calculated using Eq. (10) with h = 280 nm, which
corresponds to the IPCE assuming that all photons ab-
sorbed within the thickness of the TiO2 layer partici-
pate in electrochemical reactions. The measured IPCE
is significantly smaller than this curve for wavelengths
above 330 nm, indicating that recombination of charge
carriers is significant. For shorter wavelengths, where
absorption occurs close to the surface of the semicon-
ductor, the quantum yield is consistent with the radiative
limit of (1 − loss P ) ≈ 0.98 calculated in Section 3.1.
2013
measured
model, best fit
(L = 180 nm, lb = 8 nm, G = 0.96)
barrier contribution
bulk contribution
α(λ)
360 380 400 420
The value of L = 180 ± 30 nm indicates that only
photons absorbed in about the top 200 nm of the TiO2
can produce charge carriers that are used in electrolysis.
This suggests that, for the present sample, there is little
benefit in having a layer of thickness much greater than
200 nm, and that diffusion of carriers through the bulk
is the main contribution to photocurrent generation.
4. Discussion
4.1. Comparison of measured and calculated
efficiencies
The photoconversion efficiencies calculated for dif-
ferent radiation sources are compared with measured
efficiencies in Table 3. For each source, six calculated
efficiency figures are given. The first is the theoretical
maximum value, calculated using Eq. (7). The second
takes into account losses due to imperfect absorption,
and the third due to reflection as well as imperfect ab-
sorption and charge transfer. The influence of the bias
voltage of 0.6 V that is applied in the measurements is
taken into account in the fourth figure using
 = (VWS − VB )/1.23 V, (14)
where  and  are, respectively, the efficiencies cal-
culated excluding and including the effect of the bias
voltage.
2014 A.B. Murphy et al. / International Journal of Hydrogen Energy 31 (2006) 1999 – 2017

Table 3
Comparison of measured photoconversion efficiencies for rutile TiO2 with those calculated for various effective rutile thicknesses h , reflection
coefficients R and bias voltages VB

Efficiency AM1.5 global Sunlight, 28 May 2004, Xenon lamp with Xenon lamp
sunlight midday, Sydney water filter without water filter

Calculated
h = ∞ 2.25% 1.94% 3.75% 2.29%
R=0
VB = 0

h = 280 ± 40 nm 0.92 ± 0.04% 0.76 ± 0.04% 2.09 ± 0.06% 1.35 ± 0.04%

R=0
VB = 0

h = 280 ± 40 nm 0.66 ± 0.03% 0.54 ± 0.03% 1.47 ± 0.05% 0.94 ± 0.03%

R from Fig. 10
VB = 0

h = 280 ± 40 nm 0.34 ± 0.02% 0.28 ± 0.02% 0.75 ± 0.02% 0.48 ± 0.01%

R from Fig. 10
VB = 0.6 V

h = 180 ± 30 nm 0.57 ± 0.03% 0.47 ± 0.03% 1.34 ± 0.06% 0.86 ± 0.03%

R from Fig. 10
VB = 0

h = 180 ± 30 nm 0.29 ± 0.02% 0.24 ± 0.02% 0.69 ± 0.03% 0.44 ± 0.02%

R from Fig. 10
VB = 0.6 V

Measured
From IPCE 0.25 ± 0.05% 0.21 ± 0.04% 0.51 ± 0.10% 0.32 ± 0.07%
Direct 0.28 ± 0.06% 0.54 ± 0.11%

The first four figures are calculated assuming that
the quantum efficiency reaches its maximum possible
value, so that all charge carriers produced in the full
thickness of the TiO2 layer participate in water-splitting
reactions. The fifth and sixth figures take into account
charge transfer between the semiconductor and the elec-
trolyte, using an effective TiO2 layer thickness equal to
the diffusion length of 180 ± 30 nm derived in Section
3.4. The sixth figure, which takes into account the bias
voltage as well, is the best estimate of efficiency from
our calculations. It is compared in Table 3 to the effi-
ciencies measured both directly and derived from the
measured IPCE. It can be seen that the calculated effi-
ciency for sunlight agrees well with the measurements,
while that for the xenon lamp is slightly higher.
There are many uncertainties in the calculated effi-
ciencies, such as the influence of roughness of the TiO2
surface on reflection coefficients, the use of absorption
and reflection coefficients for single crystal rutile, the
neglect of absorption of radiation reflected from the
TiO2 –Ti interface, and the use of a simple model to
calculated the effective thickness of the film. These un-
certainties account for the small discrepancies between
the measured and calculated efficiencies. Similar cal-
culations for different semiconductors would provide a
useful guide to the achievable efficiencies.
Of particular interest is the large difference between
the theoretical maximum efficiency, and that achievable
once loss mechanisms are taken into account. The losses
are of the order of a factor of 7 to 8 for sunlight. This
is a major concern, given that the theoretical maximum
efficiency for a semiconductor with an ideal band-gap
energy of 2.0 eV is only 17%. Clearly to obtain the
target efficiency of 10% will require substantial effort
to minimise losses. If possible, the flat-band potential
of the conduction band should be more negative than
the hydrogen redox potential to remove the need for a
bias voltage. A thick nanostructured or mesostructured
semiconductor photoelectrode could be used to ensure
all the incident light is absorbed, while allowing the
electrolyte to penetrate the semiconductor to minimise
recombination losses [35].
4.2. Influence of the light source on efficiency
Our results show that the spectrum of the light source
has a large impact on the measured photoconversion
 A.B. Murphy et al. / International Journal of Hydrogen Energy 31 (2006) 1999 – 2017
efficiency. We have investigated the most widely-used
light source, an ozone-free xenon lamp. The mea-
sured efficiency is overestimated, relative to that for
the AM1.5 global solar spectrum, for wide band-
gap semiconductors and when a water filter is fitted.
Other commonly-used lamps, such as mercury lamps,
mercury–xenon lamps and non ozone-free xenon lamps
have spectra more heavily weighted to the ultraviolet
than ozone-free xenon lamps, and will lead to more
severe overestimates of efficiency.
Even when the spectrum of the light source is known,
it is not possible to derive a universal conversion factor
to convert measured photoconversion efficiencies to ef-
ficiencies under AM1.5 solar illumination. The conver-
sion factor would depend on the band gap, and on the
wavelength dependence of the absorption and reflection
characteristics of the semiconductor, and hence would
vary from material to material. A further difficulty is
that the spectrum of a given type of lamp depends on
its age. We have shown that the conversion factor can
vary by a factor of around two, depending on the age
of the tube.
Our results indicate that measurement of efficiency
under AM1.5 solar illumination has to be performed ei-
ther directly, or by measuring the IPCE as a function of
wavelength and integrating over the AM1.5 spectrum.
There are significant uncertainties for solar measure-
ment, since the spectrum can vary depending on the
zenith angle of the sun, and atmospheric conditions, par-
ticularly the ozone column, and ground conditions. We
have found, for example, differences of 15% to 20% in
the efficiencies calculated for AM1.5 global spectrum
and for the solar spectrum relevant to our measurement
performed under solar illumination. Probably the most
reliable means of relating measured efficiencies to the
AM1.5 global solar spectrum, or any other standard
spectrum, is to measure the IPCE and integrate over the
chosen spectrum using Eq. (5)
4.3. Relevance to other work
4.3.1. Measurements under solar illumination
Few previous publications have compared photocon-
version efficiencies obtained under solar and artificial il-
lumination. Houlihan et al. [36] presented data for TiO2
and doped TiO2 electrodes, finding larger efficiencies
under sunlight than under xenon lamp illumination (for
example, 0.70% under sunlight and 0.26% under xenon
lamp for a TiO2 electrode). However, the solar irradi-
ance was assumed to be 400 W m−2 at around midday
in June in Pennsylvania, USA [37]; this is too low by
a factor of around 2.5. Fujishima et al. [38] similarly
2015
underestimated the solar irradiance as 290 W m−2 , ob-
taining an efficiency of 0.4% for TiO2 produced by
flame oxidation; using a more realistic irradiance would
reduce the efficiency to well below 0.2%.
The most reliable published solar measurement is
probably that of Ghosh and Maruska [32] who obtained
an efficiency of 0.6% for aluminium-doped TiO2 un-
der solar illumination for an irradiance of 1.05 kW m−2 .
The efficiency for undoped TiO2 was about 35% of the
Al-doped TiO2 , and is hence about 0.21%, which is
consistent with our result.
4.3.2. Measurements under artificial illumination
It was noted in Section 1 that many workers have
quoted efficiencies obtained under artificial illumina-
tion, and that some of these efficiencies are close to or
higher than the theoretical thermodynamic limits for the
semiconductors used. Here we examine some specific
cases in the light of our results.
Khan et al. [8] reported obtaining an efficiency of
8.35% with carbon-doped TiO2 with an extended band-
gap wavelength of 535 nm. Table 2 shows that the max-
imum efficiency under AM1.5 global solar illumina-
tion is 10.5% for this band-gap wavelength. The stated
bias voltage was 0.3 V; taking this into account using
Eq. (14) gives a maximum efficiency of 8.0%. A simi-
lar maximum efficiency figure of 8.1% was calculated
by Hägglund et al. [39]. Losses due to imperfect ab-
sorption, reflection and recombination will almost cer-
tainly reduce the efficiency to below 2%. It should also
be noted that Khan et al. measured the bias voltage
between the working electrode and the reference elec-
trode, and defined the cell bias voltage as the differ-
ence between this in the open and closed circuit state.
The actual bias voltage was likely to be closer to 0.5 V
than the stated 0.3 V. This will decrease the efficiency
by a further factor of around 20%. Additionally, no pre-
cautions were taken to eliminate oxygen from the elec-
trolyte surrounding the counter electrode, which may
cause a further overestimate of efficiency.
Khan et al. used a xenon arc lamp with an infrared
filter as their light source. Table 2 indicates that this will
contribute strongly to the discrepancy between the re-
ported efficiency of 8.35% and the maximum efficien-
cies estimated here. It is also possible that the spectral
response of the photometer used by Khan et al. to mea-
sure the irradiance was weighted to short wavelengths,
which would lead to an underestimate of the incident
power and therefore an overestimate of the photocon-
version efficiency. We note that Khan et al. [8] obtained
an efficiency of 1.08% for oven-oxidised TiO2 , with a
standard rutile band gap of 3.0 eV. This is substantially
2016 A.B. Murphy et al. / International Journal of Hydrogen Energy 31 (2006) 1999 – 2017
above the values we presented in Table 3 even for xenon
arc lamp illumination. The photoconversion efficiencies
we have obtained for oven-oxidised TiO2 have gener-
ally been lower than those we have obtained for flame-
oxidised TiO2 .
Other workers have also presented photoconversion
efficiencies above those possible under AM1.5 global
solar illumination. For example, the 2.5% reported by
Tang et al. [10] for an anatase photoelectrode under
xenon lamp illumination is clearly greater than the 1.3%
possible for AM1.5 illumination (and the 1.7% possible
for a xenon lamp without a water filter). The discrepancy
here is due to an underestimate of the total irradiance
[40]. Mishra et al. [9] reported efficiencies up to 3%
for TiO2 photoelectrodes using a mercury–xenon lamp
source; again this is much higher than is possible for
AM1.5 illumination. The 1.9% efficiency obtained by
Radecka et al. [11] using a xenon lamp for a mixed
anatase-rutile photoelectrode is greater than is possible
for anatase or rutile under AM1.5 illumination when the
stated 0.2 V bias voltage is taken into account. An earlier
paper by Khan and Akikusa [5] quoted an efficiency of
2.0% for flame-oxidised titanium with a bias voltage of
0.7 V; the maximum achievable under AM1.5 sunlight
is 0.9% in this case.
A number of workers have presented IPCE measure-
ments, but have not taken the next step of integrating
these data over the AM1.5 solar spectrum to obtain an
efficiency that is easily comparable with the results of
others.
5. Conclusions
We have demonstrated that the use of a xenon lamp
as the light source can lead to significant overestimates
of the photoconversion efficiencies, compared to effi-
ciencies under the standard AM1.5 solar illumination.
The overestimate is particularly large when a water fil-
ter is used on the xenon lamp, and for wide-band gap
semiconductors such as TiO2 . We conclude that reliable
estimation of efficiency under standard solar conditions
is best performed by measuring the IPCE at the cell bias
voltage of maximum efficiency as a function of wave-
length, and integrating over the AM1.5 solar spectrum.
Measurements under sunlight should also provide rea-
sonable estimates of efficiency, provided that the solar
zenith angle is close to the 48.2◦ used for the AM1.5
spectrum.
Many of the photoconversion efficiencies presented
in the literature are of limited usefulness, since they
have been measured with xenon lamps and other light
sources, without taking into account the differences
between the spectra of the light source and the AM1.5
solar spectrum. Other problems are also evident in
some efficiency measurements, such as the use of the
voltage between the working and reference electrodes,
rather than the working and counter electrodes, as the
bias voltage, and incorrect measurement of the total
irradiance.
We have calculated the thermodynamically achiev-
able photoconversion efficiency for semiconductors of
different band gaps, and investigated the absorption,
reflection and recombination losses for rutile. These
losses, and the loss due to the requirement for a bias
voltage, are found to account for the discrepancy be-
tween the theoretically achievable efficiency and the
measured efficiency. The losses lead to a major reduc-
tion in the efficiency, and will have to be minimised if
photoconversion efficiencies approaching 10% are to
be obtained.
Acknowledgments
We thank Mr. Paul Gwan of CSIRO Industrial Physics
for assistance with preparing the rutile electrode, Ms.
Christina Li of CSIRO Minerals for assistance with the
IPCE measurements, and Mr. Errol Atkinson and Dr.
Frank Wilkinson of the Australian National Measure-
ment Institute for calibrating the xenon lamp and the
thermopile.
References
[1] Fujishima A, Honda K. Electrochemical photolysis of water at
a semiconductor electrode. Nature 1972;238:37–8.
[2] Bard AJ, Fox MA. Artificial photosynthesis: solar splitting of
water to hydrogen and oxygen. Acc Chem Res 1995;28:141–5.
[3] Nozik AJ. Photoelectrolysis of water using semiconducting
TiO2 crystals. Nature 1975;257:383–6.
[4] Giordano N, Antonucci V, Cavallaro S, Lembi R, Bart JCJ.
Photoassisted decomposition of water over modified rutile
electrodes. Int J Hydrogen Energy 1982;7:867–72.
[5] Khan SUM, Akikusa J. Stability and photoresponse of
nanocrystalline n-TiO2 and n-TiO2 /Mn2 O3 thin film electrodes
during water splitting reactions. J Electrochem Soc 1998;145:
89–93.
[6] Akikusa J. Investigation of thin film n-TiO2 , n-Fe2 O3 and single
crystal p-Si, p-SiC semiconductors for photosplitting of water,
PhD thesis, Duquesne University, Pittsburgh, USA, 1996.
[7] Akikusa J, Khan SUM. Photoresponse and AC impedance
characterization of n-TiO2 films during hydrogen and oxygen
evolution reactions in an electrochemical cell. Int J Hydrogen
Energy 1997;22:875–82.
[8] Khan SUM, Al-Shahry M, Ingler Jr WB. Efficient
photochemical water splitting by a chemically modified n-TiO2 .
Science 2002;297:2243–5.
 A.B. Murphy et al. / International Journal of Hydrogen Energy 31 (2006) 1999 – 2017
[9] Mishra PR, Shukla PK, Singh AK, Srivastava ON. Investigation
and optimization of nanostructured TiO2 photoelectrode in
regard to hydrogen production through photoelectrochemical
process. Int J Hydrogen Energy 2003;28:1089–94.
[10] Tang J, Wu Y, McFarland EW, Stuckey GD. Synthesis and
photocatalytic properties of highly crystalline and ordered
mesoporous TiO2 films. Chem Commun 2004;2004:1670–1.
[11] Radecka M, Wierzbicka M, Komornicki S, Rekas M. Influence
of Cr on photoelectrochemical properties of TiO2 thin films.
Physica B 2004;348:160–8.
[12] Licht S. Solar water splitting to generate hydrogen fuel:
photothermal electrochemical analysis. J Phys Chem B
2003;107:4253–60.
[13] Licht S. Solar water splitting to generate hydrogen fuel—a
photothermal electrochemical analysis. Int J Hydrogen Energy
2005;30:459–70.
[14] Barnes PRF, Randeniya LK, Murphy AB, Gwan PB, Plumb IC,
Glasscock JA. et al. TiO2 photoelectrodes for water splitting:
carbon doping by flame pyrolysis?. Dev Chem Eng Mineral
Process 2006;14:51–70.
[15] Heavens OS. Optical properties of thin films. London:
Butterworths, 1955. p. 76–7.
[16] Cardona M, Harbeke G. Optical properties and band structure
of wurtzite-type crystals and rutile. Phys Rev 1965;137A:
1467–76.
[17] DeVore JR. Refractive indices of rutile and sphalerite. J Opt
Soc Am 1951;41:416–9.
[18] Ribarsky MW. Titanium dioxide (TiO2 ) (rutile). In: Palik ED,
editor. Handbook of optical constants, Orlando, FL: Academic,
1985. p. 795–800.
[19] Eagles DM. Polar modes of lattice vibration and polaron
coupling constants in rutile (TiO2 ). J Phys Chem Solids
1964;25:1243–51.
[20] Mergel D. Modeling thin TiO2 films of various densities as an
effective optical medium. Thin Solid Films 2001;397:216–22.
[21] ASTM. Standard tables for reference solar spectral irradiance at
air mass 1.5: Direct normal and hemispherical for a 37◦ tilted
surface. Standard G159-98. West Conshohocken, PA: American
Society for Testing and Materials, 1998.
[22] Gueymard CA. Parameterized transmittance model for direct
beam and circumsolar spectral irradiance. Solar Energy
2001;71:325–46.
[23] Gueymard C. SMARTS2, a simple model of the atmospheric
radiative transfer of sunshine: Algorithms and performance
assessment. Professional Paper FSEC-PF-270-95. Cocoa, FL:
Florida Solar Energy Center, 1995.
[24] National Measurement Institute. Australian Government
Department of Industry, Tourism and Resources, P.O. Box 264,
Lindfield NSW 2070, Australia. http://measurement.gov.au
2017
[25] Oriel Instruments. The book of photon tools. CT: Stratford,
1999.
[26] Bolton JR, Haught AF, Ross RT. Photochemical energy storage:
an analysis of limits. In: Connolly JS, editor. Photochemical
conversion and storage of solar energy (Proceedings of the
third international conference, Boulder, CO, 1980). New York:
Academic Press; 1981. p. 297–339.
[27] Weber MF, Dignam MJ. Efficiency of splitting water
with semiconducting photoelectrodes. J Electrochem Soc
1984;131:1258–65.
[28] Bolton JR, Strickler SJ, Connolly JS. Limiting and realizable
efficiencies of solar photolysis of water. Nature 1985;316:
495–500.
[29] SOPRA Optical Constant Database, http://www.sopra-sa.com/
more/database.asp, accessed 27 May 2004.
[30] Lide DR, editor. CRC handbook of chemistry and physics, 84th
ed. Boca Raton, FL: CRC Press, 2003. p. 8.59–85.
[31] He XD, Torrance KE, Sillion FX, Greenberg DP. A
comprehensive physical model for light reflection. Comput
Graph 1991;25:175–86.
[32] Ghosh AK, Maruska HP. Photoelectrolysis of water in sunlight
with sensitized semiconductor electrodes. J Electrochem Soc
1977;124:1516–22.
[33] Ghosh AK, Morel DL, Feng T, Shaw RF, Rowe Jr
CA. Photovoltaic and rectification properties of Al/Mg
phthalocyanine/Ag Schottky-barrier cells. J Appl Phys
1974;45:230–6.
[34] Butler MA. Photoelectrolysis and physical properties of the
semiconducting electrode WO3 . J Appl Phys 1977;48:1914–21.
[35] Santato C, Ulmann M, Augustynski J. Photoelectrochemical
properties of nanostructured tungsten trioxide films. J Phys
Chem B 2001;105:936–40.
[36] Houlihan JF, Hamilton JR, Madacsi DP. Improved solar
efficiencies for doped polycrystalline TiO2 photoanodes. Mat
Res Bull 1979;14:915–20.
[37] Houlihan JF, Madacsi DP, Walsh EJ, Mulay LN. Improved solar
energy conversion efficiencies for the photocatalytic production
of hydrogen via TiO2 semiconductor electrodes. Mat Res Bull
1976;11:1191–8.
[38] Fujishima A, Kohayakawa K, Honda K. Hydrogen production
under sunlight with an electrochemical photocell. J Electrochem
Soc 1975;122:1487–9.
[39] Hägglund C, Grätzel M, Kasemo B. Comment on “Efficient
photochemical water splitting by a chemically modified n-TiO2 ”
(II). Science 2003;301:1673b.
[40] Tang J, Private communication, 2005.