REPORTS

A Monolithic Photovoltaic-Photoelectrochemical Water is split directly upon illumination,

light being the only energy input.
Device for Hydrogen Production via Water The p-GaInP2/GaAs cells used in this
Splitting study were grown by atmospheric-pressure
organometallic vapor-phase epitaxy. Details
Oscar Khaselev and John A. Turner* of material growth and cell processing have
been previously reported (18). The cell
structure consists of a top layer of epitaxi-
Direct water electrolysis was achieved with a novel, integrated, monolithic photoelec- ally grown p-Ga0.52In0.48P (abbreviated as
trochemical-photovoltaic design. This photoelectrochemical cell, which is voltage biased p-GaInP2) 4.0 6 0.5 mm thick, connected
with an integrated photovoltaic device, splits water directly upon illumination; light is the in series via a low-resistivity, grown-in tun-
only energy input. The hydrogen production efficiency of this system, based on the nel junction (TJ) to a GaAs p/n bottom cell
short-circuit current and the lower heating value of hydrogen, is 12.4 percent. on a GaAs substrate. The photoelectro-
chemical characteristics of the present cell
were compared with those of a similarly
grown p-GaInP2 epilayer on a GaAs sub-
The direct photoelectrolysis of water at the move its band edges to some extent, their strate. We have studied a number of similar
semiconductor electrolyte interface is not energetic positions are still inadequate to devices with varying thicknesses of the top
only of scientific interest but also could effect water splitting (8). layer (19). The present configuration pro-
produce benefits such as cost and energy A number of approaches have been tried duced the best overall results and is report-
savings over coupled photovoltaic (PV)- to overcome some of the obstacles to the ed here.
electrolysis systems for H2 production (1). direct splitting of water. Cells with two Standard chemical and electrochemical

Downloaded from http://science.sciencemag.org/ on October 31, 2018

However, the lack of a stable, efficient simultaneously illuminated semiconductors procedures were used (20), with the surface
light-absorption system coupled with suit- have been used (9–11). Multiple-junction of the samples being coated with a platinum
able semiconductor-redox energetics has amorphous silicon devices have also been catalyst (21). Illumination for the photo-
kept photoelectrochemists from reaching reported by Sakai et al. (12) and Lin et al. electrolysis was produced by a fiber optic
this goal. (13). Augustynski et al. (14) have reported illuminator with a 150-W tungsten-halogen
Obstacles to direct photoelectrolysis of a water splitting system using a dye-sensi- lamp. To measure the light irradiance on
water are the lack of efficient light absorp- tized nanocrystalline solar cell coupled to a the surface of the sample, a calibrated PV
tion (for reasonable solar efficiencies, the tungsten trioxide electrode. However, each tandem cell was mounted into the electrode
band gap must be less than 2.0 eV), corro- of their systems operated separately and holder inside the cell in the manner similar
sion of the semiconductor (thermodynami- they were not integrated into a monolithic
cally, most useful semiconductors are pho- device. Although all of these systems had
tochemically unstable in water), and ener- some success, their efficiencies were rather
getics (the difficulty of matching the semi- low (,7%).
conductor band-edge energies with the H2 We now report a direct water electrolysis
and O2 evolution reactions) (2, 3). The system based on a novel, integrated, mono-
most photochemically stable semiconduc- lithic photoelectrochemical (PEC)/PV de-
tors in aqueous solution are oxides, but their vice (Fig. 1A). This device is patterned
band gaps are either too large for efficient after the GaInP2/GaAs p/n, p/n tandem cell
light absorption or their semiconductor device grown at the National Renewable
characteristics are poor (4). Semiconduc- Energy Laboratory (15). The solid-state
tors with better solid-state characteristics tandem cell consists of a GaAs bottom cell
are typically thermodynamically unstable connected to a GaInP2 top cell through a
with respect to oxidation (2). However, tunnel diode interconnect. The top p/n
p-type semiconductors generally offer some GaInP2 junction, with a band gap of 1.83
protection against photocorrosion, because eV, is designed to absorb the visible portion
under illumination the surface is cathodi- of the solar spectrum; and the bottom p/n
cally protected. Heller (5, 6) has shown GaAs junction, with a band gap of 1.42 eV,
that p-type indium phosphide is stable in absorbs the near-infrared portion of the
strong acid under illumination and H2 evo- spectrum transmitted through the top junc-
lution but requires an external bias for wa- tion. Although single-gap electrodes have a
ter splitting. In earlier work, we identified solar conversion efficiency limit of 32%,
p-type gallium indium phosphide (p- tandem-junction devices have an efficiency
GaInP2) as perhaps a perfect semiconductor limit of 42% (16). The maximum theoret-
for water splitting. Although its band gap is ical solar-to-electrical efficiency for the
in the ideal range (1.8 to 1.9 eV), the present combination of band gaps is about
energetics of its band edges are not correct 34% (17). Our device differs from the stan-
for water splitting, needing an additional dard solid-state tandem cell in that a PEC
300-mV bias (7). Although it is possible to Schottky-type junction has replaced the top
p/n junction. This device then is a PEC
National Renewable Energy Laboratory, 1617 Cole Bou- device that is voltage-biased with an inte-
levard, Golden, CO 80401–3393, USA. grated PV device. Under illumination, elec- Fig 1. (A) Schematic of the monolithic PEC/PV
* To whom correspondence should be addressed. E-mail: trons flow toward the illuminated surface device. (B) Idealized energy level diagram for the
[email protected] and holes flow toward the ohmic contact. monolithic PEC/PV photoelectrolysis device.

www.sciencemag.org z SCIENCE z VOL. 280 z 17 APRIL 1998 425

to that used for the photoelectrodes. The For water-splitting reactions, however, H2 The photocurrent time profile for the
measured light irradiance was 1.19 6 0.05 production at the surface of the semicon- p-GaInP2(Pt)/TJ/GaAs electrode at short-
W/cm2: about 11 suns (22). ductor is preferred for two reasons: (i) the circuit condition is shown in Fig. 3. The
For the PEC/PV configuration to work H2 evolution reaction has the lowest over- current density decreased slightly for about
properly, the GaAs bottom cell must pro- voltage losses, so the requirement for a cat- 300 s but then stabilized at about 120 mA/
vide sufficient voltage to overcome any en- alyst is less, thus allowing an optimized cm2 and remained constant throughout the
ergetic mismatch between the band edges of counterelectrode to be used for the more first part of the experiment. The inset in
the GaInP2 and the water redox reactions, complex O2 evolution reaction; and (ii) Fig. 3 shows the long-term current stability
and must also provide any additional volt- under illumination, the semiconductor sur- of the p-GaInP2(Pt)/TJ/GaAs electrode.
age needed to overcome overvoltage losses face is cathodically protected. After 20 hours, the initial current density of
from the H2 and O2 evolution reactions. Figure 2 shows photocurrent-voltage 120 mA/cm2 decreased to 105 mA/cm2.
The total photovoltage output must include curves for platinum-catalyzed p-GaInP2(Pt) Gas bubbles continued to be evolved from
the thermodynamics for water splitting and p-GaInP2(Pt)/TJ/GaAs electrodes mea- the entire surface. Visual inspection of the
(1.23 V); polarization losses ha and hc for sured in a two-electrode configuration. The electrode surface after the experiment re-
the anodic and cathodic processes, respec- open circuit voltages in the dark were vealed some localized damage of the sample
tively; and the current-resistance potential ;–0.64 and –0.75 V for the p-GaInP2(Pt) along the upper edge of the epoxy coating,
drop in the bulk of the electrolyte, which and p-GaInP2(Pt)/TJ/GaAs electrodes, re- where gas bubbles collect and are held by
can be significant when gas evolution oc- spectively. The dark reduction current was the “lip” of the epoxy. The remaining por-
curs. An idealized energy-level diagram for small (in microampere range) for both elec- tion of the sample appeared undamaged.
the photolytic splitting of water with this trodes. Under illumination, the p-GaInP2 Because these experiments were done with
device (Fig. 1B) illustrates the process, (Pt) electrode started to generate hydrogen the electrode in a vertical position, a more
which includes two photons and one sepa- at a voltage 500 mV negative of 0 V bias, solar-like slanted position (not possible

Downloaded from http://science.sciencemag.org/ on October 31, 2018

rated electron-hole pair. Light incident on indicating that additional external voltage with our current configuration) would re-
the PEC/PV configuration first enters the was needed for this semiconductor to split duce or eliminate the accumulation of bub-
wide band gap p-GaInP2 layer, in which the water. Under illumination, the p-GaInP2 bles at this edge.
more energetic photons are absorbed, re- (Pt)/TJ/GaAs electrode showed an open-cir- The products of photoelectrolysis were
sulting in electron-hole pair excitation and cuit voltage of ;0.55 V, indicating the extra collected and analyzed with a mass spec-
producing photovoltage Vph1. Less energet- photovoltage generated by the GaAs cell. trometer. Within our experimental error,
ic photons penetrate through the GaInP2 Evolution of H2 started immediately, 400 the ratio was 2 :1 H2:O2 as expected (23).
and are absorbed by a GaAs bottom p/n mV positive of a short circuit. The photocur- The efficiency of H2 production was calcu-
junction generating photovoltage Vph2. rent density reached a limiting value of 120 lated with the following equation: efficien-
One set of holes and electrons are recom- mA/cm2 at ;0.15 V and remained almost cy 5 (power out)/(power in). The input
bined at the tunnel junction. If the result- constant with increasing bias. Copious power is the incident light intensity of 1190
ant photovoltage Vph5 Vph1 1 Vph2 is amounts of gas bubbles could be observed at mW/cm2. For the output power, assuming
greater than that required for photoelec- the semiconductor surface. Because these gas 100% photocurrent electrolysis efficiency,
trolyis for this particular cell configuration, bubbles can grow to sufficient size to act as the H2 production photocurrent of 120 mA/
it will drive the water reduction reaction at miniature lenses, causing pitting of the semi- cm2 is multiplied by 1.23 V, which is the
the semiconductor electrode and water ox- conductor electrode, 0.01 M of the surfac- ideal fuel cell limit at 25°C (the lower
idation at the counterelectrode. Two pho- tant Triton X-100 was added to the solution heating value of hydrogen). Calculated by
tons are required to produce one electron in to promote the formation of smaller bubbles this method, the H2 production efficiency
the external circuit, so four photons are that left the sample surface more rapidly. of our system is 12.4% (24).
required to produce one molecule of H2. The lower saturated photocurrent for the Gerischer (2) has calculated for a two-
This is an example of a D4 scheme (16). p-GaInP2 (Pt)/TJ/GaAs electrode as com- layer semiconductor system (a tandem cell)
This configuration can easily be de- pared to the p-GaInP2 electrode is taken to that the maximum efficiency of photoelec-
signed to do oxidation as well as reduction indicate that the p/n GaAs bottom cell is the trochemical water splitting would be 24%.
at the semiconductor-electrolyte interface. current-limiting junction. Bolton et al. (16) estimated 16% as a max-
imum realizable chemical conversion effi-
ciency for a tandem system of this type (a
D4 scheme in his nomenclature). Weber
and Dignam (25, 26) calculated that for
different tandem cell configurations, the
maximum water-splitting efficiency would
fall between 10 and 18%. Their efficiency
analysis took into account various solid-
state and electrochemical losses. Our results
are consistent with these calculations.
The key to making this system work
appears to be the requirement that the bot-
tom cell be the limiting electron provider.
Because these tandem systems operate by
Fig. 2 (left). Current-voltage characteristics for p-GaInP2(Pt)/ TJ/GaAs (curve 1) and p-GaInP2 (curve 2) requiring two photons (one per junction) to
electrodes in 3 M H2SO4 under white light illumination. Fig. 3 (right). Photocurrent time profile at produce one electron in the external cir-
short circuit for PEC/PV tandem cell in 3 M H2SO4 with 0.01 M Triton X-100 under tungsten-halogen cuit, great care is taken in the solid-state
white light illumination. Efficiency 5 120 mA /cm2 3 1.23 V 3 100/1190 mW/cm2 5 12.4%. CD, current systems to match the photon absorption
density. characteristics so that equal numbers of

426 SCIENCE z VOL. 280 z 17 APRIL 1998 z www.sciencemag.org

 REPORTS
photocarriers are generated in the top and tics were measured with an EG&G 263A poten- system design issues, a more realistic limit as to the
tiostat. The electrolyte, 3 M H2SO4, was freshly amount of light concentration that can be used for
bottom cells. However, when the photocur- prepared from deionized water having resistivity of photoelectrolysis is in the range of 10 to 20 suns.
rent-matched configuration is placed in an 18 megohm/cm. All solutions were made of analyti- 23. The gas products of the photoelectrolysis were an-
aqueous environment, the system either cal-grade reagents. Before electrochemical mea- alyzed with a UTI-100C mass spectrometer. A
surements, the samples were etched in 1: 20 :1 HCl: sealed cell was used, connected directly to the high-
produces no current in the external circuit CH3COOH:H2O2 solution. A single-compartment vacuum chamber of the mass spectrometer via a
or decomposes (27). A thicker, top p-type cell was used. Although the electrodes were spatially gas inlet system. Before being sealed, the cell was
layer and the resultant mismatch of elec- separated, no attempt was made to separate the purged for several minutes with nitrogen to remove
products of the anode and cathode reactions. any oxygen. The cell was then sealed, and the pho-
tron-hole formation in the two-junction
21. To reduce the overvoltage losses associated with toelectrolysis reaction was run for several hours. The
region appear to be the key for proper the noncatalytic surface of the semiconductor, a thin external current was monitored with an ammeter.
operation. layer of platinum catalyst was electrochemically de- Periodically, the gas mixture from the upper part of
posited on the surface of the semiconductor elec- the cell was sampled by the mass spectrometer sys-
trodes from a 20 mM H2PtCl6 solution. Photoas- tem. The same cell was used and the same experi-
REFERENCES AND NOTES mental technique was repeated with the use of two
___________________________ sisted galvanostatic deposition was performed at a
cathodic current density of 1 mA/cm2, with a plati- platinum electrodes at the same external current.
1. A. J. Bard and M. A. Fox, Acc. Chem. Res. 28, 141
num quantity corresponding to a charge of 10 mC/ Within experimental error (;620%), the results were
(1995).
cm2. A recent publication (28) has shown that the the same, indicating stoichiometric water splitting.
2. H. Gerischer, in Topics in Applied Physics, vol. 31,
platinum not only acts as a catalyst for hydrogen 24. For 100% electrolysis efficiency, this calculation is
Solar Energy Conversion, B. O. Seraphin, Ed.
evolution but also drastically reduces the corrosion identical to Bolton’s calculation of efficiency (29). For
(Springer-Verlag, Berlin, 1979), pp. 115 –172.
reaction of a similar III-V compound, indium phos- the size of these samples and the amount of hydro-
3. The ability of a semiconductor electrode to drive the gen and oxygen being generated, a calculation using
phide. We would expect that platinum would offer a
electrochemical reaction of interest is determined by external current flow is far more accurate.
comparable corrosion-inhibiting mechanism here.
its band gap (the energy separation between the
22. We used concentrated light for two reasons: (i) The 25. M. F. Weber and M. J. Dignam, Int. J. Hydrogen
valence and conduction band edges) and the posi-
iiii
photocurrent and the amount of gases produced are Energy 11, 225 (1986).
tion of the valence and conduction band edges rela-
higher and therefore easier to quantify, and (ii) be- 26. , J. Electrochem. Soc. 131, 1258 (1984).
tive to the vacuum level (or other reference elec-
cause of their cost, the major terrestrial application of 27. S. Kocha, D. Montgomery, M. Peterson, J. A. Turn-
trode). In contrast to metal electrodes, semiconduc-

Downloaded from http://science.sciencemag.org/ on October 31, 2018

the solid-state analog of these tandem cells is in con- er, Sol. Energy Mater. Sol. Cells, in press.
tor electrodes in contact with liquid electrolytes have
centrator systems (.100 suns). Therefore, any real 28. P. Bogdanoff, P. Friebe, N. Alonso-Vante, J. Electro-
fixed energies where the charge carriers enter the
application using the present solid-state design for chem. Soc. 145, 576 (1998).
solution. This fixed energy is given by the energetic
position of the semiconductor’s valence and con- photoelectrolysis would also use concentrated light. 29. J. R. Bolton, Solar Energy 57, 37 (1996).
duction bands at the surface (where these bands However, the PEC systems will probably be limited to 30. We thank S. Kurtz and J. Olson for the tandem cell
terminate at the semiconductor/electrolyte inter- less than 100 suns, because that would give rise to a samples and A. Dillon for help with the mass spec-
face). The energetic position of these band edges is current of about 1A /cm2, a practical limit for electrol- trometer measurements. This work was supported
determined by the chemistry of the semiconductor/ ysis. The evolution of gas bubbles on the illuminated by the Hydrogen Program of the U.S. Department of
electrolyte interface, which is controlled by the com- surface is also an issue, as these bubbles will scatter Energy.
position of the semiconductor, the nature of the sur- light, reducing the efficiency. Given this limitation as
face, and the electrolyte composition. So even well as problems regarding the amount of catalyst and 18 November 1997; accepted 23 February 1998
though a semiconductor electrode may generate
sufficient energy to effect an electrochemical reac-
tion, the energetic position of the band edges may
prevent it from doing so. For spontaneous water Gastrointestinal Tract as a Major Site of
CD41 T Cell Depletion and Viral Replication
splitting, the oxygen and hydrogen reactions must lie
between the valence and conduction band edges,
and this is almost never the case.
4. G. E. Shakhnazaryan et al., Russ. J. Electrochem.
30, 610 (1994).
in SIV Infection
5. E. Aharon-Shalom and A. Heller, J. Electrochem.
Soc. 129, 2865 (1982). Ronald S. Veazey, MaryAnn DeMaria, Laura V. Chalifoux,
6. A. Heller, Science 223, 1141 (1984). Daniel E. Shvetz, Douglas R. Pauley, Heather L. Knight,
7. S. Kocha, J. Turner, A. J. Nozik, J. Electroanal.
Chem. 367, 27 (1991). Michael Rosenzweig, R. Paul Johnson, Ronald C. Desrosiers,
8. S. Kocha and J. A. Turner, J. Electrochem. Soc. 142,
2625 (1995).
Andrew A. Lackner*
9. H. Yoneyama, H. Sakamoto, H. Tumura, Electro-
chem. Acta 20, 341 (1975). Human and simian immunodeficiency virus (HIV and SIV ) replicate optimally in activated
10. A. J. Nozik, Appl. Phys. Lett. 29, 150 (1976). memory CD41 T cells, a cell type that is abundant in the intestine. SIV infection of rhesus
11. R. C. Kainthla, B. Zelenay, J. O’M. Bockris, J. Elec- monkeys resulted in profound and selective depletion of CD41 T cells in the intestine
trochem. Soc. 134, 841 (1987 ).
12. Y. Sakai et al., Can. J. Chem. 66, 1853 (1988). within days of infection, before any such changes in peripheral lymphoid tissues. The loss
13. G. Lin et al., Appl. Phys. Lett. 55, 386 (1989). of CD41 T cells in the intestine occurred coincident with productive infection of large
14. J. Augustynski, G. Calzaferri, J. Courvoisier, M. numbers of mononuclear cells at this site. The intestine appears to be a major target for
Graetzel, in Proceedings of the 11th World Hydrogen
Energy Conference, T. N. Veziroglu, C.-J. Winter,
SIV replication and the major site of CD41 T cell loss in early SIV infection.
J. P. Baselt, G. Kreysa, Eds. (DECHEMA, Frankfurt,
Germany, 1996), pp. 2378 –2387.
15. K. A. Bertness et al., App. Phys. Lett. 65, 989 (1994).
16. J. R. Bolton et al., Nature 316, 495 (1985).
17. S. R. Kurtz et al., J. Appl. Phys. 68, 1890 (1990). It is now thought that ongoing HIV rep- fact that the gastrointestinal tract con-
18. K. A. Bertness et al., in AIP Conference Proceedings lication results in a continual loss of tains most of the lymphoid tissue in the
No. 306, R. Noufi and H. Ullal, Eds. (American Insti- CD41 T lymphocytes that is nearly bal- body (2, 3). Furthermore, it is likely that
tute of Physics, New York, 1994), p. 100.
19. O. Khaselev and J. A. Turner, in preparation.
anced by the production of new CD41 T the behavior of HIV in the unique immu-
20. For electrochemical measurements, the wafers were lymphocytes (1). This model explains nologic environment of the intestinal
cleaved into samples ;0.2 cm2 in area. The samples some of the puzzles of HIV infection, but mucosa differs from that observed in the
with gold ohmic contacts were mounted on a Teflon-
covered screw electrode with silver epoxy and heat-
the events that occur in the initial stage periphery.
ed to 80°C for 1 hour. The electrical contact was of infection remain largely unexplored. The gut-associated lymphoid tissue
insulated from the electrolyte by an epoxy coating Although it is clear that HIV targets lym- (GALT) consists of organized lymphoid tis-
that also covered the samples’ edges. A convention- phoid tissue, nearly all studies in this area sue (Peyer’s patches and solitary lymphoid
al two-electrode configuration was used with a plat-
inum gauze or foil counter-electrode coated with ru- have focused on peripheral blood and follicles) as well as large numbers of acti-
thenium metal. Photoelectrochemical characteris- lymph nodes. These studies overlook the vated memory T lymphocytes diffusely dis-

www.sciencemag.org z SCIENCE z VOL. 280 z 17 APRIL 1998 427

A Monolithic Photovoltaic-Photoelectrochemical Device for Hydrogen Production via Water
Splitting
Oscar Khaselev and John A. Turner

Science 280 (5362), 425-427.

DOI: 10.1126/science.280.5362.425

Downloaded from http://science.sciencemag.org/ on October 31, 2018

ARTICLE TOOLS http://science.sciencemag.org/content/280/5362/425

REFERENCES This article cites 8 articles, 1 of which you can access for free
http://science.sciencemag.org/content/280/5362/425#BIBL

PERMISSIONS http://www.sciencemag.org/help/reprints-and-permissions

Use of this article is subject to the Terms of Service

Science (print ISSN 0036-8075; online ISSN 1095-9203) is published by the American Association for the Advancement of
Science, 1200 New York Avenue NW, Washington, DC 20005. 2017 © The Authors, some rights reserved; exclusive licensee
American Association for the Advancement of Science. No claim to original U.S. Government Works. The title Science is a
registered trademark of AAAS.