Semiconductor/Electrolyte Photoelectric Energy Conversion: The Use of a Molybdenum Oxide Coating to Avoid Corrosion

S. Gourgaud and D. Elliott
Department of Chemistry, University of Queensland, St. Lucia, Queensland 4067, Australia

Abstract

The electrolytic corrosion of n-type GaAs under illumination is avoided by a coating of hydrated molybdenum oxide (probably MoO₂). This has the properties of a solid redox system and acts as a hole acceptor to the substrate. It is in turn rereduced by some redox species in solution and thus effectively couples hole generation to the oxidation of the electrolyte. The molybdenum oxide can be coated electrolytically to a calculated thickness of at least several hundred angstroms and is capable of storing charge reversibly of many tens of millicoulombs per square centimeter.

Photoelectric energy conversion employing electrolytes rather than solid-state junctions has received some attention in recent years (1-4). A number of experiments have centered on the generation of hydrogen gas in an electrolytic cell containing a suitable semiconductor and counterelectrode [e.g., (5, 6)]. Others have considered the direct conversion to electric power with a similar arrangement.

Fig. 1. Diagrammatic representation of photoelectric conversion device: EC = conduction band; EV = valence band; EF = Fermi level; EO = standard potential of M(ox)/M(red).

The general design of a direct conversion device would be according to Fig. 1 for an n-type semiconductor. Light entering the surface of the semiconductor causes electron-hole pair generation if of sufficient energy to cross the bandgap. Electrons promoted to the conduction band drift toward the interior while holes, the minority carriers, drift to the surface. Here they encounter the reduced form, M(red), of the redox couple in solution. This component is oxidized by the holes, transported to the counterelectrode and, there, rereduced. This reduction is driven by the external connection from the semiconductor. The result is a passage of electrons externally in a circuit doing work with no net change in the electrolyte. Analogous arguments may be used for a p-type semiconductor where electrons (the minority carriers) drift to the surface and cause a reduction of some oxidized species in solution.

A photoelectric device based on this design has potential advantages over a conventional solid-state diode in ease of manufacture and scale-up to large sizes. This arises from the following factors: (i) use of polycrystalline substrates since no junctions need be grown; (ii) optimum light penetration and segregation of optical and electroconductive phenomena; (iii) the choice of a wide range of compound semiconductors (e.g., oxides, sulfides, etc.) of which only one carrier type can be made.

In a number of investigations of these types of devices, the principal disadvantages have been the electrolytic corrosion of the semiconductor and the inherent slowness of electrode reactions at such materials. Corrosion occurs for example with holes at an n-type semiconductor because of the preferential oxidation of the substrate compared with that of some ion in solution. Though there are some materials resistant to corrosion, e.g., TiO₂ and SnO₂, their bandgaps are too high for conversion of sunlight. The inherent slowness of electrode reactions involving oxidation or reduction of ions in solution is more fundamental and is related to the relatively small carrier density at the surface of a semiconductor. This slowness is responsible for the preferential oxidation of the electrode and when corrosion can be avoided it then causes unacceptably low currents. Some attempts have been reported where these disadvantages have been obviated by means of a surface coating, e.g., a metal film (7).

Fig. 2. Current-voltage curve for the reduction of molybdate at the HMDE: sweep rate = 0.5 V/min; area = 0.029 cm²; solution = 1M NaCI, 0.01M HCI (pH 1.8), 7 × 10^-4M Mo (as ammonium molybdate). The numbers on the curves represent successive sweeps.

Fig. 3. Reduction of molybdate at the HMDE as a function of pH. First current-voltage cycle. Data as for Fig. 2 except pH varied as shown by addition of HCI.

The reduction/oxidation linear sweep voltamogram is asymmetrical about the current axis; there is much more cathodic current passing than anodic. It is presumed that the hydrogen overvoltage is lowered compared with a mercury surface, so that appreciable hydrogen ion reduction takes place on the layer.

The growth of the layer is sensitive to pH, occurring roughly between pH 1 and pH 4. Figure 3 shows a series of first cyclic scans as a function of pH. The first anodic peak is only detectable as a slight inflection in the first scan but the way it subsequently grows can be seen in Fig. 2. Taking the average of the cathodic and first anodic peaks for the second scans as indicating the equilibrium potential of the system, the variation with pH is about 0.110 V/pH unit, indicating that 2H⁺ ions are involved per electron. A second anodic peak (-0.18V at pH 1.8) is prominent at the beginning but ultimately disappears as the layer grows. It also shifts with pH (Fig. 3), but at about half the rate compared with the main system. It may represent the reoxidation of some of the layer back to the VIth state (presumably soluble) and is a further explanation for the asymmetry about the current axis; much more Mo(VI) is reduced and reoxidized than subsequently accumulates on the electrode as a stable insoluble layer.

Polarographic measurements have also been made. A two-stage reduction is observed, the first of which is very poorly defined and is approximately half that of the second part. This has been interpreted previously (10-14) as a VI → V step followed by a V → III step. It is therefore suggested that the V and III compounds are solid in this pH range and are responsible for the cyclic linear sweep pattern. The reduction/oxidation can be provisionally represented by the scheme

Mo (V) Ox + 4H⁺ + 2e ⇆ Mo (III) Ox + 2H₂O

Some confirmation of the two-electron transfer may be gained from the shape of the later curves (Fig. 2). As the layer grows a secondary anodic peak emerges, more negative than the main one by about 0.1V. A slight inflection on the cathodic curve in an analogous position seems to represent the same process taking place during reduction. As further growth proceeds, this feature eventually grows to dominate the whole linear sweep pattern. This causes a negative shift to the apparent equilibrium potential (judged from the mean point of the cathodic and anodic peaks). Compared with the graph in Fig. 3, the values for a fully grown layer may be 0.1V or more negative. This factor varies greatly with various parameters, e.g., sweep rate, Mo concentration, etc., which probably influence the physical form of the deposit if not its actual composition.

Physical Properties and Structure of Mo Ox

The material has been examined by a number of techniques. It was prepared in relative bulk for this purpose by electrolytic reduction on a mercury pool (≈ 20 cm²). A cyclic potential sweep was used, between –0.1 and –0.7V vs. SCE at the rate of 2 V/min. 10^-3M molybdenum was used (as ammonium molybdate) in a 1M NaCl solution adjusted to pH 1.5 with dilute HCl. The layer was generated for about 3 hr with constant stirring with nitrogen. The Mo Ox film was detached from the mercury by shaking, forming a suspension of small flakes that could be withdrawn from the vessel. The material was collected by centrifuging and was repeatedly washed with 100% alcohol to remove water. It was finally dried under vacuum.

The material appears to be amorphous down to the limit of resolution of the electron microscope. No electron diffraction pattern was detected and no x-ray powder pattern (24 hr Cu-Kα).

Spectroscopic examinations were made in both the infrared and the u.v./visible region. For the former, the material was incorporated into KBr disks and examined on a PE 225 spectrophotometer. The spectra, compared with those for some other molybdenum compounds, showed a large increase in the number of absorption bands (six bands were observed between 1400 and 1000 cm^-1, whereas hydrated Mo O₃ has only one and ammonium molybdate has none). This is attributed to a much lower site symmetry in the relatively amorphous compound. Bands appearing at 3595, 3450, 3150, and 1622 cm^-1 suggest the material is hydrated. However, the direct connection of hydroxyl groups to the metal atom is not indicated from the spectrum. According to Adams (15), evidence for this should occur in the range 500-576 cm^-1.

By comparing the spectrum of the film with data given by Mitchell and Trifiro (16), the following assignments have been made:

970 cm^-1

stretching

M-O (terminal)

920 cm^-1

stretching

M-O (bridging)

735 cm^-1

stretching

M-O (bridging)

u.v./visible spectra were determined on a Cary 17 spectrophotometer. The material was ground and in the form of a suspension in absolute alcohol was evaporated onto a quartz plate. No peak was observed in the visible, but a band in the u.v. was observed at 41,500 cm^-1 (241 nm) which does not correspond to a band in any reported compound.

The tentative conclusion is that the material is a condensed molybdenum oxide molecule of average oxidation state IV involving bridging oxygen linkages and arranged amorphously in the film. This description is similar to that for a colloid and the material has a similar appearance to a gelatinous film when viewed in the visible microscope. It should be noted that in the pH range at which the material is generated condensed polyanions are the stable form for Mo(VI) in solution (e.g., Mo₆O₂₁^-6).

The Generation and Properties of Mo Ox on n-GaAs

A specimen of n-type GaAs (~ 10¹⁷ carriers/cm³) was cemented with epoxy resin into the end of a glass tube and mounted vertically in a rotating-electrode assembly. The surface of the electrode could be illuminated from underneath using a 100W quartz-iodine projector lamp.

Fig. 4. Current-voltage curves for the reduction of molybdate at a GaAs electrode (under illumination): sweep rate = 2 V/min; area = 0.27 cm²; rotation = 2000 rpm; solution = 1M NaCI, pH 1.85, 5.6 × 10^-4 Mo. The numbers on the curves represent particular stages in the sequence of 18 continuous cycles.

Figure 4 shows the current-voltage behavior for the rotating electrode under cyclic sweep conditions. The reduction peak for molybdenum (IV) is superimposed on the hydrogen evolution reaction and is displaced negatively with respect to the values previously reported for a mercury electrode. This large overvoltage illustrates the slowness of the reduction at the semiconductor surface. On scanning positively, no oxidation of the layer takes place in the absence of light. On illumination, however, the two-stage oxidation occurs, the first peak at about –1.0V, which is close to the Hatband potential. The holes responsible for the oxidation are released at the valence band edge about 1AV more positive than this. The layer grows satisfactorily only when the electrode is scanned in the presence of light between –0.7 and –1.1V. Initial chemical etching is also necessary.¹ If the potential is allowed to go more positive than about –0.6V, the electrode enters a region of anodic dissolution that disrupts the surface Mo Ox layer. Satisfactory growth of the layer does not occur at a surface so eroded and it must be reetched chemically.

¹ 1 part 30% H₂O₂, 1 part H₂O, 3 parts concentrated H₂SO₄.

The Oxidation of V²⁺ at a Mo Ox Covered GaAs Surface

Fig. 5. Oxidation of V²⁺ at a GaAs electrode (under illumination): sweep rate = 0.1 V/min; area = 0.27 cm²; rotation = 2000 rpm; solution = 1M NoCl, pH 1.35. ——— Oxidation of GaAs alone; – – – – oxidation of V²⁺ at bare GaAs electrode (10-1M V²⁺); ∙ ∙ ∙ ∙ oxidation of V²⁺ at Mo Ox-covered GaAs electrode (2 x 10-1M V²⁺).

The direct oxidation of V²⁺ at an illuminated, bare GaAs surface is sluggish and merges quickly into the anodic dissolution region (Fig. 5). This is a typical result for semiconductors and illustrates the poor coupling between holes accumulating at the surface and any reducing agent in solution. This is in spite of the fact that the holes appear at the valence band edge approximately 0.9V more positive than the V²⁺/V³⁺ potential. With a surface covering of Mo Ox, however, a large oxidation current for V²⁺ occurs. The V²⁺/ V³⁺ potential is more negative than that for Mo Ox, and presumably the latter is kept in the reduced form. Holes generated in the semiconductor cause the oxidation of the Mo Ox which is then rapidly reduced by the V²⁺ in solution. It is not known at this stage what is the rate-determining step. The net result is the oxidation of V²⁺ in solution at a velocity very much greater than without the Mo Ox layer.

The currents are not very reproducible and are critically dependent on many factors, e.g., the precise composition and performance of the layer, its thickness, the pH and composition of the solution, etc. For the particular layer whose growth is shown in Fig. 4 (9.4 mC/cm²), the currents at constant voltage, after an initial transient, declined very slowly and the following values given in Table I were observed after a few minutes.

Table 1. Solution: 1M NaCI, 0.2M V²⁺, pH 1.5

---

V (vs.SCE)

mA/cm²

 

---

–0.9

0.41

 

–0.8

1.48

 

–0.7

2.0

 

–0.65

2.1

The current appears saturated at a potential more positive than about –0.7V suggesting that it is limited by the hole generation process and not by electrode kinetics. The saturation current observed is less than that for a silicon cell with the same light source (11 mA/cm²). However, since the lamp used was uncorrected and has its peak well in the infrared, the silicon cell (bandgap 1.12 eV, 1.1 μm) would be expected to be more sensitive. Another factor causing lower currents is the absorption of light in the liquid layer between the electrode surface and the glass window. This was not closely controlled for the rotating electrode, but in a working cell could be made very small by making the layer thin enough.

An Experimental Photoelectric Conversion Cell

Fig. 6. Diagram of experimental photoelectric conversion cell.

An experimental cell was constructed as shown in Fig. 6. The GaAs electrode was first plated with the Mo Ox layer before mounting. The counterelectrode was an annular pool of mercury chosen for its high H⁺/H₂ overvoltage. The solution containing approximately equal quantities of V²⁺ and V³⁺ was prepared externally by electrolytic reduction and pumped into the cell. Beside the two working electrodes, contact was made to the solution via a SCE. This enabled the potential of the GaAs electrode to be monitored to make sure it did not enter the anodic dissolution region. The open-circuit potential of the cell is the difference between that of the semiconductor (monitored at between –0.95 and –1.0V depending on the precise characteristics of the Mo Ox layer) and the counterelectrode which is determined by the V²⁺/V³⁺ potential (about –0.52V). A working value of about 0.45V has been obtained. The maximum current is governed by the overvoltage characteristic of each electrode. The properties of the semiconductor/ Mo Ox/ V²⁺ half of the system have been described above, and the observed currents are not much reduced because of the electrode being stationary, indicating that mass transfer is not a serious limiting effect for these currents.

At the counterelectrode, the intrinsic rate constant of the V³⁺/V²⁺ system is sufficiently great at a clean Hg surface (≈ 10^-3 cm/sec) that negligible activation overvoltage should result at ca. 20 mA/cm². In the actual cell constructed, a significant overvoltage was observed which considerably lowered the operating voltage of the cell. This may be due to surface contamination or concentration polarization, but could be largely eliminated by proper design.

Conclusion

A photoelectric conversion cell based on the preceding concepts seems a practical proposition. However, several factors have to be considered in justifying such a construction over conventional solar cells.

Efficiency.—The intrinsic efficiency of solid-state cells depends on the bandgap of the semiconductor and in sunlight reaches a theoretical maximum for materials with a gap of about 1.4 eV. However, in the present construction the operating voltage depends on electrochemical potentials unrelated to the band-gap, namely the difference between the flatband potential and the redox potential of the electrolyte. The optimum bandgap should be just a little larger than this for maximum conversion. The redox potential should be sufficiently negative to maintain a significant part of the Mo Ox layer in the reduced form, ca. < –0.4V. Consequently the only practicable variable is the flatband potential.

This potential at a given carrier concentration and electrolyte composition depends on the chemical nature of the material. With compound semiconductors it becomes more negative with lower molecular weight, e.g., CdS is more negative than CdSe and GaP is more negative than GaAs. This is also the direction of increasing ionic bonding between the elements. Since the bandgap also increases in the same sense (causing lower light absorption and therefore loss of efficiency), there is no advantage in looking for too negative values. There is not enough experimental information available on flatband potentials generally, particularly as a function of electrolyte composition to predict the material of highest efficiency. However, any material with a potential of –1.0V or more negative is worth pursuing experimentally. This includes GaAs, GaP, and probably CdS. An interesting possibility is that the potential may be shifted by surface-active materials in a manner similar to the shift in the ecm (or zero charge potential) of mercury and other metals.

Cost.—The semiconductor/electrolyte construction seems to offer advantages in reducing the cost per watt factor, particularly for large installations. There are, for example, a large range of materials (often one-carrier types, usually n-type) that would lend themselves to large-scale thin film polycrystalline construction. Also, the separation of the optical and electrical regions offers new design possibilities. A major limitation in the unit size of existing solid-state cells arises from the current-carrying capacity of the thin surface layer. If thin enough to allow adequate light penetration, its resistance limits the area to about 10-20 cm² for silicon cells (i.e., 0.1-0.2 W/unit for 10% efficiency).

Electrolytes should have a more favorable combination of optical/electrical properties and the layer can be made very thin providing that a cathode of sufficient surface area can be placed nearby. The necessary cathode area is also a flexible design parameter and it could take the form of a transparent conductive film or a metal grid. A grid provided with multiple feed-through contacts passing through holes in the semiconductor would enable the unit to be scaled up to meter sizes.

Stability.—The major problem of semiconductor corrosion is eliminated by operating in a potential range where it does not occur. The Mo Ox layer may also exert a moderate protective action. The long-term stability of this layer has not been measured in this particular configuration. However, previous results (8) suggest it is very stable. In any case it is renewable by a simple electrolytic process.

The particular electrolyte used, V²⁺/V³⁺, is of course prone to oxidation by air and by H⁺ (the latter oxidation is dependent on pH and the presence of catalytic metal surfaces). Persistence of V²⁺ for several days can be obtained with minimal precautions and it is to be expected that oxidation can be made almost negligible with good design. This electrolyte can also be regenerated electrolytically and since a moderate volume of solution could be used with a large area it could be periodically renewed at negligible energy cost. Even this would be eliminated by the use of more stable redox systems, e.g., Ti³⁺/ TiO²⁺, in various media. The latter is currently under investigation.

Manuscript submitted April 5, 1976; revised manuscript received Aug. 23, 1976.

REFERENCES

1. H. Gerischer, J. Electroanal. Chem. Interfacial Sci., 58, 263 (1975).
2. Y. G. Chaff and W. W. Anderson, Appl. Phys. Lett., 27,183 (1975).
3. H. Gerischer, Surf. Sci., 18, 97 (1969).
4. H. Gerischer, This Journal, 113, 1174 (1966).
5. A. Fujishima and K. Honda, Nature, 238, 37 (1972).
6. T. Ohnishi, Y. Nakato, and H. Tsubomura, Ber. Bunsenges. Phys. Chem., 79, 523 (1975).
7. Y. Nakato, T. Ohnishi, and H. Tsubomura, Chem. Lett., 883 (1975).
8. D. Elliott, D. L. Zellmer, and H. A. Laitinen, This Journal, 117, 1343 (1970).
9. P. Lagrange and J. P. Schwing, C. R. Acad. Sci. (Paris), 263, 848 (1966).
10. D. F. Boltz, T. de Vries, and M. G. Mellon, Anal. Chem., 21, 563 (1949).
11. E. P. Parry and M. G. Yakuleik, ibid., 26, 1294 (1954).
12. B. N. Popov and H. A. Laitinen, This Journal, 120, 1346 (1973).
13. L. Capanella and G. Pistoia, ibid., 120, 383 (1973).
14. G. P. Haight Jr., Anal. Chem., 23, 1505 (1951).
15. D. M. Adams, “Metal Ligand and Related Vibrations,” p. 248, Arnold Limited, London (1967).
16. P. C. H. Mitchell and F. Trifiro, J. Chem. Soc. A, 23, 3183 (1970).