J Electroanal Chem, 164 (1984) 89-119 Elsevier Sequoia S.A., Lausanne - Pnnted m The Netherlands 89 IRON(III)-TITANIUM(IV)-OXIDE ELECTRODES: THEIR STRUCTURAL, ELECTROCHEMICAL AND PHOTOELECTROCHEMICAL PROPERTIES BODO DANZFUSS and ULRICH STIMMING * Instttut fur Physlkahsche Chemic der Umoersttat Dusseldorf, Umversttatsstrasse 1, D-4000 Dusseldorf 1 (FRG) (Received 16th March 1983; m revised form 9th August 1983) ABSTRACT Iron(III)-txtanlum(IV)-oxades of the general composmon FexT11_xOy were prepared in the composition range x = 0-0.9 by thermal decomposition of the corresponding metal salt solutions. For a medium range of composition, 0.1 _<x _<0 7, amorphous orades were formed under the gwen conditions of preparation. Electrochemacal properUes such as reduction and re-oradatlon of the orade, electrode capacity behavlour, oxygen evolution and reduction and the redox reaction, FeE+/Fe3+, were lnvesUgated Photoelectrochemlcal properties were obtamed from photocurrent spectra and the dependence on potential of the photocurrent. Anneahng experiments showed that crystallization yields lower photocurrents and a shaft of the photocurrents spectra to shorter wavelengths Thus, amorphous semaconductors seem worth being investigated for a possible application in electrochermcal solar cells. An attempt is made to describe the impact of non-crystalhnlty on the photoelectrochemlcal behavlour of sermconductors. A model of the oxade is proposed to explain the electrochermcal and photoelectrochemical properties of the oxades FexTii_ xO v. (1) INTRODUCTION M e t a l oxides as electrode materials have gained increasing interest in recent years be cau s e of their specificity in electrochemical r e a c t i o n an d in m o s t cases, low cost of p r e p a r a t i o n . A l s o o f interest is the d e v e l o p m e n t of electrode materials w h i ch can be used in el ect ro che m i c a l solar cells. Since the w o r k of F u j i s h i m a an d H o n d a [1], T i O 2 is considered to be a possible m a t e r i a l because of its high q u a n t u m efficiency an d high resistance against p h o t o c o r r o s i o n . H o w e v e r , with a b a n d gap of ca. 3 eV, the theoretical efficiency with respect to sunlight of a solar cell with a T1 0 2 a n o d e is only a b o u t 4%. T h e r e f o r e a t t e m p t s have b e e n m a d e to m o d i f y the properties of T i O 2 by doping, in o rd er to achieve a higher a b s o r p t i o n at longer wavelengths [2,3]. Positive effects f r o m d o p i n g have been observed b u t the results were n o t satisfactory since * Present address. IBM, T.J Watson Research Center, Yorktown Heights, NY 10598, U.S A. New address (to where all correspondence should be addressed): Columbia Umverslty, Dept. of Chem Engmeering and Appl Chemistry, New York, NY 10027, U.S.A. 0022-0728/84/$03.00 © 1984 Elsevier Sequoia S A. 90 only a slight increase in absorption was found and the quantum efficiency suffered. Fe203, on the other hand, is an oxide with a higher electrochemical activity, but its corrosion stability is less than that of TiO 2. The onset energy for the photocurrent is reported to be at 2.2 eV [4,5], which is more in the Vaslble compared to TiO 2. A combination of the two should yield new properties in electrochemical, as well as photoelectrochemical behawour. Recently, investigations of mixed oxides concerning their electrochemical behaviour and possible use as materials in electrochemical solar cells were reported [6-9]. In this paper experiments are described where the electrode material was composed of a mixture of Fe(III)-Ti(IV) oxides in a wide range of composition. Their structure was investigated with respect to morphology, crystallinity and composiUon. The electrochemical properties are described by the electrode capacity, the oxidation-reduction behaviour of the oxide itself, and by reactions on it hke oxygen evolution and oxygen reduction and a pure redox reaction, F e 2 + / F e 3+. The photoelectrochemical behaviour is characterized by the wavelength and potential dependence of the photocurrent. The influence of a thermal post-treatment of the oxides and its special impact oh the photoelectrochemical properUes will be described. (2) EXPERIMENTAL The oxide electrodes were prepared by thermal hydrolysis of aqueous Fe(III)Ti(IV)-solutions on heated titanium metal sheets, followed by a thermal treatment (details given below). The circular electrodes with a diameter of 1 cm were mounted in a Teflon holder, allowing an area exposed to the electrolyte of ca. 0.5 cm 2. The ohmic backside contact was made by pressing a screw into the titanium metal. The counter-electrode was a platinum sheet with an area of 1 cm 2. The reference electrode was a m e r c u r y / m e r c u r o u s sulphate/0.5 M H2SO 4 electrode with all potentials given with respect to the standard hydrogen electrode (SHE). The electrolyte was normally 1 M HC104, except for the redox measurements with 0.05 M FeSO4/Fe2(SO4) 3 where the redox-free electrolyte contained 0.2 M H z S O 4. All solutions were deaerated by N 2 (99.995%), except for the measurements of the oxygen reduction where 02 was bubbled through the electrolyte. For all measurements the temperature was 25 _ 0.1°C. The electronic equipment consisted of a fast-rise potentiostat (band width 1 MHz), a potential sweep generator for potentiodynamic measurements and a differential amplifier for the current measurements. To determine the electrode capacity a small ac signal with a RMS amplitude of 0.316 mV and frequency of ca. 1 kHz was superimposed on the electrode potential and the resulting ac current was analysed by means of a lock-in amplifier. The photoelectrochemical results were obtained using monochromatic light from a 450 W Xe lamp and a grating monochromator in the wavelength range 200-800 nm. The intensity change of the hght with wavelength was followed by a thermopile and a pyroelectric detector. The photocurrents were measured directly or by using a chopper and lock-in amplifier. The structural investigations were carried out using the following set-ups. The 91 X-ray diffraction was measured with Co-K~-radiatlon since the usual Cu-K~-radiation yields a high fluorescence radiation by the iron. The composition, also in its spatial distribution, was obtained by microprobe measurements. The morphology was examined by light microscopy and scanning electron microscopy in an early stage of preparation to improve the preparation of the oxide. (3) PREPARATION AND SOLID STATEPROPERTIESOF THE OXIDES For the preparation of the oxides the method of thermal hydrolysis of aqueous solutions containing Fe(III)-Ti(IV) ions was chosen. This allowed a fairly easy variation of composition in the range of 100% Fe(III) oxide to 100% Ti(IV) oxide, though pure ion(III) oxide could not be made because of adhesion problems with the titanium substrate. The solutions were TffIV) chloride, prepared by oxidizing TiCI 3 with H202, and Fe(III) chloride, dissolved in 10% aqueous HC1 and 30% methanol. The solution was then sprayed in a oxygen stream onto the heated titanium substrate, previously cleaned with acetone, distilled water, HC1, H N O 3, H F and finally extensive rinsing with triply distilled water. The substrate temperature depended on the composition of the solution and was varied from 150 to 270°C with increasing iron content. This was necessary to improve the adhesion of the oxide to the substrate. The spraying procedure was as follows: after every 10 s spraying, there was a break of 5 s, with a total of 800 s spraying in one cycle. Seven cycles were performed; between each cycle the electrode was held at 400°C for 10 min in an oxygen atmosphere and, after the last cycle, for 1 h. Temperatures of 500 and 600°C after the last cycle gave a considerably lower activity of the formed oxides for the anodic oxygen evolution, and so were not chosen in the final preparation procedure. The effect of additional annealing on the capacity and photocurrent behaviour was studied with specimens which had been heated in an Ar atmosphere at 600°C for 24 h. For each composition, five samples were available for electrochemical and photoelectrochemical investigations. In this paper the oxides formed are denoted according to the composition of the solution using the general formula FexTi1-x0v, where x is given by eqn. (1): x = nFo/(nFe + nT,) (1) where rife and nT, are the numbers of iron and titanium ions in the spraying solution. While x changes from 0 to 1, depending on whether one has pure TiO 2 or pure Fe203, y changes between 2 and 1.5. The formed oxides had, according to their composition, different colors which are hsted in Table 1 together with the compositions which were investigated. An estimate of the oxide thickness was made by using the density of anatase as 3.9 and hematite as 5.3 g / c m 3 [10]. For the mixtures, a linear variation and 70% of the theoretical density were assumed. Thicknesses in the range of 1-10 /~m were obtained. The status of the surface during the preliminary stage of preparation was controlled using a scanning electron microscope (sere). The surface looked lumpy, 92 TABLE 1 Colour of Fe:,TlI _ ~Oyfor different compositionsx Composition x Colour 0.00 wtute 0.01 cream 0 027 hght yellow 0.10 ochre 0.25 hght brown 0.4-0.6 brown 0 7-0.9 red-brown with particles of ca. 2 0 / t m diameter with gaps of ca. 1/~m between them. This was probably a result of the drying of the wet surface and was avoided in the final preparation process by using a low spraying rate and a cyclic exchange of the specimen. Light microscope pictures showed that by this procedure the particle size was increased to ca. 0.1 ram. The roughness factor of the electrodes, according to the sem pictures which show only the macroscopic roughness, was estimated to be < 3. Microprobe analysis of the oxide gave iron, titanium and oxygen as major components, with, in addition, traces of chlorine. The element distribution with the particles was nearly constant. However, between the particles the relative amount of titanium was much higher. This is due to gaps between the particles where the oxide layer is thinner and the titanium substrate contributes to the signal. X-ray photoelectron spectroscopy (XPS), which was done with the compositions x = 0, 0.1, 0.5, 0.9, gave Fe(III) and Ti(IV) as the main valence states of the metal. F r o m the data for the unsputtered surface the F e / T i ratio was obtained by integrating the peaks and taking into account the ionization cross sections [11]; the ratio values were higher than expected from the calculated x values. With Ar ÷ sputtering, the F e / T i ratio came close to the calculated x value. This indicates an iron enrichment at the surface with respect to the calculated x value of eqn. (1), but the results are not totally conclusive. X-ray diffraction measurements on the oxides should yield the following information: whether the oxide is crystalline, which modifications are formed, and if any mixed oxide of iron and titanium is formed. The hydrolysis of TiC14 solutions normally leads to the formation of anatase at low temperature, which is transformed into futile at higher temperatures [12]. In the presence of foreign ions, e.g. chloride ions, a stabilization of the anatase modification has been found [12]. The hydrolysis of Fe(III) salt solutions generally gives a-Fe203 or 3,-Fe203, where intermediate states can be a - F e O O H [13], "y-FeOOH [14] or Fe203 • 1.2 H 2 0 [15]. According to the phase diagram TiO2-F%O 3 [16] the formation of mixed oxides should be possible at x = 0.4, yielding pseudorutile, and at x = 0.67, yielding pseudobrookite. The solubihty limit for Fe203 is ca. 3% before the formation of a mixed oxide starts. The electrodes made with 400°C annealing as the final preparation step, gave reflections of anatase, hematite and titanium, depending on their composition. For x = 0-0.1, anatase was the only detectable phase but with decreasing intensity of the reflexes for higher x. The peaks in the diffractograms were broadened and the crystal size estimated to be < 0.5 /~m. In the range 0.1 < x < 0.7, no distinct reflections other than a contribution from the titanium substrate could be found. Only small 93 peaks, very slightly higher than the noise were present, which could be attributed neither to a pure nor a mixed oxide. In the range x > 0.7, hematite was identified, with an almost constant intensity of the reflections up to x = 0.9. At iron contents lower than 10% and greater than 70% crystalline compounds were formed, while in the entire intermediate range of composition the material was X-ray-amorphous. To improve crystalhnity and to study this effect in particular on the photoelectrochemical properties, several electrodes were additionally annealed in an inert Ar atmosphere for 24 h at 600°C. As an example, the oxide x = 0.5 was investigated by X-ray diffraction after this treatment. Distinct reflections were found which could be assigned to pseudobrookite and rutile. The formerly amorphous oxide was changed to a crystalline one by the thermal post-treatment. (4) ELECTROCHEMICAL PROPERTIES General behavmur The typical behaviour of the oxide electrodes can be seen from Fig. 1 which shows a potentiodynamic current vs. potential curve for an oxide with x = 0.7. In the intermediate potential range, c = 1.0-1.9 V, the current is on the order of 10 /~A/cm 2 which is near the value for the steady state corrosion of passive iron in acid solution [17] corresponding to the dissolution of Fe203. At potentials below 1 V a reduction process sets m, while in the reverse sweep the current shows an anodic peak. At high potentials, c > 1.9 V, oxygen evolution starts. The possible reactions in the potential range displayed in Fig. 1 are: 2 H 2 0 ~ 02 + 4 H ÷ + 4 e 2 H++ 2 e-~ H 2 (Oxygen evolution and reduction) (Hydrogen evolution) 2 Fe203 + 2 H + + 2 e - ~ 2 Fe304 + H 2 0 (2) (3) (Oxide reduction/oxidation) (4) Fe203 + 6 H + + 2 e - ~ 2 Fe 2+. aq + 3 H 2 0 (Reductive dissolution of the oxide) (5) A reduction of titanium oxide can be neglected because its equilibrium potential is very low, c < - 0 . 5 V. In the presence of the redox couple, Fe 2÷. aq ~ Fe 3+. aq + e - (6) which reacts by an outer sphere mechanism, an electronic current through the oxide is superimposed on the above-described reactions (dotted line). In the lower part of Fig. 1 the electrode capacity is shown as a function of the electrode potential in the same range. The potential dependence is similar to the behaviour of a highly doped n-type semiconductor. The hysteresis in the capacity curve is a function of how low the reverse potential in the sweep is and is probably due to stoichiometric changes in the oxide. 94 15 Fe.T 5.0y w,th x=07 i i - - H[IO~ IM/H2S Q 02M 10 - - - H[lO~ IM I FeSO~, Fe2(SOJ~005 M O5 % 0 25 -05 -20 -10 "5 -I 5 -20 .10 l!:' OS 1'0 1'5 2'0 E/V Fig 1 Typical electrochermcal behavlour of FexTh_xO ~ electrodes shown by potentlodynamlc current-potential curves, m a redox free electrolyte (sohd hne) or in electrolyte c o n t a m m g 0 05 M F e Z + / F e 3+ (dotted hne), and a capacity-potential curve C. Oxide reductton In the absence of oxygen in the soluUon, the cathodic and anodic currents in the potential range 0 - 1 V can be attributed to a reduction and re-oxidation of the oxide. As T102 is reduced only at much lower potentials, all reduction and oxidation currents in this potential range should be due to reactions of the ,ron in the oxide. For the onset of the cathodic current a Tafel-like current-potential relation, with a transfer coefficient a = 0.25, was found for all compositions x. For each oxide, the anodic charge was much smaller than the cathodic one. A possible explanation for this is that reaction (4) as well as reaction (5) occurs, which would thus reduce the current efficiency since parts of the iron are dissolved. The additional effect of a kinetic controlled reaction would decrease the current efficiency as well. Separation of the anodic and cathodic charge is rather difficult; superimposed small stationary currents would drastically change the relation between the two, as can be seen in Fig. 1 in the case of the presence of the redox system. However, to give an impression of this oxidation-reduction behaviour of the oxide, instead of the anodic charge, Qa, and the cathodic charge, Qc, their sum Q = Oa + Oc (7) was evaluated. This charge is simply the area described between the anodic and 95 60 ~ FexTIl_xOy 3o / 9 O2 05 x 08 Fig. 2. ReducUon and re-oxidation behawour of Fe~Tla_xOy as a function of x. The charge Q was evaluated from the area between the posmve and negative potenuodynanuc sweep below 1 V. cathodic part of the potentiodynamic sweep. The evaluation of the charge is made uncertain by the contribution of charging the electrode capacity, but this is almost constant for all compositions. In any case, Q represents the oxxdation-reduction behaviour of the oxides, in a qualitative manner, although it allows no chemical interpretation. A plot of Q as a function of the composition x is displayed in Fig. 2. With increasing x, Q also increases from 0.1 m C / c m 2 at x = 0.01 to 60 m C / c m 2 at x = 0.5, which is reasonable because the iron content of the oxide increases as well. At larger iron contents, however, the reducible iron oxide decreases sharply to ca. 5 m C / c m 2 at x = 0.9. The difference in reducibility of the iron, between medium x on one side and low and high x on the other side, is even more pronounced if Q is related to the actual amount of iron available in the oxide, i.e. if it is related to its mass. This shows that the iron is much more easily reduced in a medium range of composition, in which the oxide has been recognized to be amorphous (0.1 < x < 0.7). Oxygen evolutton and reductton Oxygen evolution is a redox reaction which, however, proceeds via an inner sphere mechanism. Thus reformation is not only obtained on the electronic proper- 96 0zevolution O.f FexTi).xOy wlfh chfferenf -3 x 0 -5 -G 20 2'5 3'0 3'5 r/V Fig. 3 Tafel plots of 0 2evolution for different composmons x ties of the oxide, but also on the properties of the material with respect to the adsorption step included in the reaction. Figure 3 shows current vs. potential curves for different compositions, in a semi-logarithmic plot. For pure TiO 2 electrodes the overvoltage is high and current densities of / ~ A / c m 2 can be reached only at potentmls greater than 3 V, but 10% to 25% iron already produces a strong decrease of the overpotential. However, higher iron contents do not lower the overvoltage to the same extent. The decrease is small in the range x = 0.5 to x = 0.9. This influence of composition can be seen in Fig. 4, where the electrode potential at constant current density (0.1 and 1 m A / c m 2) is plotted vs. the iron content x. The strongest influence of the iron is observed for low percentages, up to 25%, then it decreases and between 70% and 90% there is no longer any significant change. However, at the same time, a change of the transfer coefficient a occurs with changing composition. 02 evoLuflonof 30 > FexTmt.xOy ~, 2O 01mA cm-z O0 O2 05 x 08 10 Fig 4. Overvoltage of 0 2 evolution as a function of x at constant current density; the bars show the variation of data for different samples of the same composition. 97 At potentials above 2.4 V, a is small and < 0.1 for all compositions encountered in that potential range. The curve for x = 0.25 itself exhibits a break, giving a higher a-value below 2.5 V. The transfer coefficient then increases continuously with the iron content up to a = 0.6 at x = 0.9. This change in the kinetics of the oxygen evolution can be associated with a change of the electronic states in the oxide at the corresponding energies, which will be discussed also in section 6. Oxygen reduction, formally the reverse reaction of oxygen evolution, proceeds only to the stage of H202, though the equilibrium potential of H 2 0 2 / H 2 0 is more positive than that of O 2 / / H 2 0 2 . The reduction to H 2 0 is observed in most cases at potentials lower than - 1 V [18]. Under our experimental conditions the lowest potential was - 0 . 3 V and therefore the reduction proceeds only to the stage of H 2 0 2 . The current density for oxygen reduction was obtained by measuring in both oxygen-saturated and oxygen-free solutions, and calculating the difference. A Tafel plot of the current vs. potential curve is given in Fig. 5, which was exactly the same for x = 0, 0.01 and 0.027. Oxides with a higher iron content did not yield an oxygen-reduction current at all. In the linear part of the Tafel plot the transfer coefficient is a = 0.5, but for high overvoltages a limiting current is observed, which is about 1 m A / c m 2 and depends on stirring, indicating a diffusion limitation of the oxygen to the electrode surface. It is interesting that oxygen reduction can be readily observed only at almost pure TiO 2. The fact that for x > 0.1 no current could be measured, can also be due to the loss of crystallinlty. The electronic conductivity of the oxide should play no decisive role under these conditions because the rate of other reactions increase with x and, in addition, the electrode potential is below the flatband potential. Under these conditions an accumulation layer is formed at the oxide surface and no limitation from the number or the transfer of electrons should be expected. This can be seen also from the fact that a = 0.5. Possibly, the reaction includes an adsorption step, which is connected to certain active sites on the -3 @ -5 at FexTil_xOy wtfh x<01 -6 -02 -0,1 0,0 ,0,1 ~/V Fig 5. Tafel plot of 0 2 reduction. 98 electrode surface which exists only on crystalhne TiO 2 but not on amorphous TiO2, as it is found for x > 0.1. The result is that oxygen reduction can proceed at TiO 2 as long as it is in the crystalline state. Iron oxide in the amorphous, as well as in the crystalline state, seems to have no activity for oxygen reduction. R e d o x reactton F e 2 + / F e 3 + Use of an outer sphere redox reaction gives clearer information on the electronic properties of a material and the electron transfer reactions therein. In our case the redox reaction chosen for investigation was Fe 2+- aq ~ Fe 3+. aq + e - (6) The presence of the redox couple in the solution changes the current, as can be seen in Fig. 1. The current t e of reaction (6) was evaluated by subtracting the current in the redox-free solution, i6E, from the current in the redox system-containing solution, /RE, le = /RE -- /GE (8) The current densities obtained are much lower than those on metal electrodes. Cathodic currents could be observed over the whole range of composition, while the anodic ones were much lower and only detectable for x > 0.5. Tafel plots are displayed in Fig. 6. Pure TiO2 shows the lowest current density but it is higher by two orders of magnitude than that found with passivated electrodes [19,20]. This is probably caused by the preparation process and is connected with a certain contamination of the material. Doping experiments with passive titanium have shown that even small quantities of a foreign substance can enhance the rate of reaction (6) considerably [20]. An amount of 10% of iron shafts the cathodic current to higher values by about one order of magnitude, but a further increase of the iron up to 90% has only a small effect. On the anodic side a measurable current starts with x = 0.5. In Fig. 6, curves are shown for x = 0.6 and 0.9. Up to potentials ~ = 1.5 V the current is some 1 0 / ~ A / c m 2 and the transfer coefficient is a < 0.1. At c = 1.5 V, a break is observed which is connected with a higher transfer coefficient a = 0.3. In principle, these reflect n-type semiconducting properties of the electrodes. The cathodic process needs much less overvoltage than the anodic one, which is almost blocked as the depletion layer built up at the oxide surface acts as a barrier for the electron transfer. The influence of an increasing iron content in the oxide on the cathodic process is probably due to a shift of the flatband potential Cfb with composition. As will be shown in the next section the flatband potential is changed from 0.1 V for pure TiO 2 to about 0.6 V for the highest iron content, x = 0.9. The barrier effective for the electron transfer is formed by the space charge layer gwen by eAqSsc= e ( , - ,fb) (9) with A,/,sc, the potential drop in the semiconductor, c fb, the flatband potential where 99 0906\ \x Fe2. ~ Fe3*+e- at Fe,Tq_,O x / w~thdifferent x \! \',/ °°l o1\ / \"1\ 09 06 /ill /,,," -S. o's & lO " " 2b E/V Fzg. 6. Tafel plots of the pure redox reactzon F e 2 + / F e 3+ for different compositions x. A@sc= 0, and is lowered at constant potential with increasing ¢ fb According to the shift of Cfb the increase in l e can be explained by a lowering of the barrier height and with that, a higher transfer probability for the electrons, since the concentration of free electrons determined by the doping level is of the same order of magnitude for the different compositions. For x >_ 0.6, the potential range where the cathodic current is measured is close to or even below the flatband potentzal. Therefore, the barrier former by the space charge layer according to eqn. (9) becomes unimportant, and curves for these compositions show only little variation compared to that for smaller x values. The anodic reaction, on the other hand, cannot be observed below x - - 0 . 5 . This is in accordance with the results obtained for passive titanium electrodes [19]. When c >> Cfb the thickness of the space charge layer allows little electron transfer and a process via the valence band is unlikely because of the large band gap. The small, almost potential-independent, current at higher x can be observed because more and more empty electronic states become available with the shift of c fb for the iron-richer oxide. This enables an electron exchange with occupied states in the electrolyte (Fe2+), resulting in anodic current which is barely potentialdependent. The break at c = 1.5 V indicates a contribution of lower lying terms above the valence band. This will be discussed in section 6 in connection with a model of the amorphous oxide. Capacity behavzour Capacity measurements with semiconductor electrodes usually give valuable information on the charge distribution at the electrode/electrolyte interface. At 100 potentials above the flatband potential, the capacity of an n-type semiconductor is determined by the capacity of the space charge layer built up at the semiconductor surface. Then, the Schottky-Mott equation (10) ~s applicable to describe the potential dependence of the electrode capacity, if certain limitation are considered [21] 1 C2 2 eND,Do (c - Ceb -- k T / e ) (10) From the extrapolation of a straight line in a Skottky-Mott plot, the flatband potential can be calculated. From the slope, the product N D ' of the donor concentration N and the dielectric constant D' is obtained. All electrodes except x = 0.9 show behaviour typical of an n-type semmonductor with a decreasing capacity at increasing electrode potential, as shown in Fig. 1. Schottky-Mott plots according to eqn. (10) of various electrodes are given m Fig. 7. The main influence of an higher iron content is a shift of the curve to higher potentials and to a smaller slope. For pure TiO 2 (x = 0), N D ' = 2 × 10 21 cm -3 is calculated from the slope which has a rather high value. The extrapolation of the linear part to 1 / C 2 = 0 yields a potential c o = 0.1 V. According to eqn. (10) only k T / e should be considered to obtain the flatband potential. However, for highly doped semiconductors an additional potential drop in the Helmholtz layer has to be taken into account [22,23], The extrapolated potential c ( 1 / C 2 = 0) = c o is then g~ven by eqn. (11) c o = ero + k T / e - eUD'Do/2CZn (11) Using C H = 20 /~F/cm 2, the flatband potential for x = 0 is cro--0.0 V. With increasing x, Cfb becomes more and more positive up to more than 0.6 V for the iron-rich compositions. This change of ca, is also shown m Fig. 7. The N D ' values increase only slightly with x. A calculation of the donor density is rather uncertain because the dielectric constant D' of these mixed compounds is not known. For bulk crystalline TiO 2 and Fe203, D' is 80-120 and 12, respectively. However, since the particle size--according to X-ray diffraction--is small, especially where the oxide is amorphous, the dielectric constant can be considerably lower [24]. For a rough estimate, a c o m m o n value for all oxides of D ' = 10 can be assumed. The donor concentration in the whole range of composition is then very high, with only a small variation from 2 X 1020 to 8 X 1020 cm -3. For higher potentials, the curves in Fig. 7 show a break of the slope. This phenomenon is also observed with pure TiO 2 [25] and pure Fe203 [26]. In both cases this behaviour is explained by the existence of donors with different energy [21]. However, this explanation does not seem applicable here because it is restricted to crystalline material with defined donor states. Some of the electrodes, especmlly in the medium range of composition, display a pronounced curvature in their 1/C2(c) curves at higher potentials. In terms of eqn. (10) this would correspond to localized states in a certain energy range, a picture which is consistent with the amorphous state. Another explanation for a changing slope can be that the donor concentration is spatially different. This is possible, 101 [ O0 x=08 1 / / !'/ / 0005/4 f O~ ,--' ' -.4 ~l ()~"' 021_/ 0 0'8 1'2 ~/V 03 06 1'6 09 20 Fig 7 Schottky-Mott plots for different compositions, according to eqn. (10), the reset shows the change of ~fb, calculated from eqn. (11), as a funcUon of the composition assuming inhomogeneities within the particles or, if the particle size is small as suggested above, the thickness of the space charge layer could become comparable to the particle diameter, and at its surface additional terms become available, thus increasing ND'. However, it should be kept in mind that all these explanations are similar pictures, in a physical sense. In principle, a break in a Schottky-Mott plot could also be caused by a changing equivalent circuit of the electrode at low potentials, but then the flatband potentials obtained would be wrong by several hundred mV. Because the "as-prepared" and the "additionally annealed" samples show a reasonable agreement between c ro values obtained from capacity and photocurrent measurements, such an explanation does not seem very likely here. The conclusion drawn from this is that the oxide electrodes of different composition have a high number of donors which are connected to the rmcrocrystalhne or amorphous structure. Annealing of the electrodes in an inert Ar atmosphere lowered the donor density by orders of magnitude. This effect, together with the special impact of non-crystallinity on the capacity behaviour will he discussed in section 6. (5) PHOTOELECTROCHEMICAL BEHAVIOUR A N D I N F L U E N C E OF A N N E A L I N G The photoelectrochemistry of the pure materials, T i O 2 and F e 2 0 3 , is well known: the onset of the photocurrent on the low photon energy side is given by the absorption of the material and the band gap energies were found to be 3.0 eV for TiO 2 and 2.2 eV for F e 2 0 3. Mixtures of these two oxides, however, should show either an additive behaviour, or the characteristics of a newly formed compound (like an iron titanate), or a totally new behavlour. Values for the band gap of iron titanates were reported to he close to the value of Fe203 itself [6]. 102 Wavelength dependence of the photocurrent The materials described here are, depending on their compositton, amorphous or crystalline. This, and the composition itself, should have a large influence on the photoelectrochemlcal behawour. Photocurrent spectra were measured by the modulated beam technique for all compositions prepared between x = 0 and x = 0.9. The spectra shown in Fig. 8 are normalized to equal numbers of incident photons and, in addition, curves for different compositions are normalized to equal maxima. Pure TiO 2 (x = 0) exhibits the expected behaviour with an onset of the photocurrent at 3.0 eV. Admixtures of 1 and 2.7% Fe do not change the spectral distribution of the photocurrent. The actual magnitude, however, is gradually lowered by the iron content; this decrease is about one order of magnitude for x -- 0.027. The oxide of the composition x = 0.1 gives the same spectrum as for T~O2 with the ad&tion of a broad prepeak with a maximum at 2.7 eV. The spectrum of the composition x = 0.25 is already determined to a large extent by the low photon energy absorption (Fig. 8). The onset is at about 2.0 eV, which is 0.2 eV below the band gap of pure Fe203. At higher photon energies the features are very unlike the TiO 2 spectrum. A further increase in the iron content of the oxide results in a slight shift of the onset energy to longer wavelengths. As a typical example, the spectrum of x = 0.7 is displayed in Fig. 8. Its main features are a broad absorption between 2.0 and 3.5 eV and a steep edge at 1.9 eV. The actual photocurrent values do not depend very much on the composition. Details of the change of the onset energy of the photocurrent with )~/nm 700 600 500 l,O0 300 Fe Th Oy with different x x -x /-', ~o x:o7 \ I', -i / // ; / "io / / XX / / ~, ', IX! /\. "a / . 2'5 = "-.. //' 3'0 is ¢o ,.'5 photon energy hvleV Fig. 8. Pbotocurrent spectra of typical compositions, with the photon energy hv and the wavelength ~ on the abclssa; the photocurrent is normalized to equal number of incident photons and, m addition, curves of different composition x are normalized to equal m a m m a 103 composition are given below (see Fig. 10). In some of the spectra, especially for x = 0, the photocurrent passes a pronounced maximum after the absorption edge, and the photocurrent decreases at higher photon energies. This effect could be due to an increased surface recombination of the photoexcited carriers at high absorption coefficients. Such an effect is to be expected for a highly doped semiconductor with a small diffusion length of the minority carrier. The photocurrent spectra obtained for different compositions x, are different from those of the pure compounds and from an additive behavlour of them. No indication could be seen that this behaviour is due to the occurrence of a mixed oxide. Remarkable, is that the onset of the photocurrent is at photon energies lower than found for pure Fe203. Since there is obviously a connection between this behaviour and the X-ray amorphous structure of the oxades, annealing experiments in an inert atmosphere were c a m e d out. Influence o f anneahng on the photocurrent spectra The change of the photocurrent spectrum with annealing is shown in Fig. 9 for the composition x = 0.5. The originally broad peak between 2 and 3 eV in the as-prepared oxide is changed to a two peak spectrum with a much smaller peak width. In addition, a minor prepeak is observed at a longer wavelength. X-ray diffraction experiments carried out with these annealed samples show a change from the amorphous structure to one having discrete diffraction patterns. The identified substances for the medium range of composition are pseudobrookite and futile. This result helps to explain the change of the spectrum in Fig. 9. The two peak structure is probably due to the two compounds, identified by X-ray diffraction as pseudobrookite and rutile. The onset energy of the photocurrent for the two peaks is roughly 2.2 and 3.0 eV, close to the band energies of the two compounds. The values are clearly higher than that of the amorphous oxide with ca. 2 eV. In addition, the photocurrent values, which could be expected to be increased by the crystallization, dropped by orders of magnitude. A quantitative description of this effect will be given below. The two peak feature is a common characteristic of the annealed oxides, while the as-prepared oxides mostly have one broad absorption. The differences in the spectra with composition are seen in the changes of the hv o values, the onset energy of the photocurrent. In Fig. 9 the hv o values which are obtained by extrapolation, are indicated on the abscissa. A plot of the hv o values vs. the composition parameter x is displayed in Fig. 10 for both the as-prepared and the annealed oxide. As described above, even relatively small amounts of iron, e.g. 25%, decrease the absorption edge to ca. 2 eV. This value is not changed, up to a composition x = 0.6. For iron-richer oxides, x _> 0.7, the edge decreases shghtly to about 1.9 eV. In contrast to this is the result for the annealed oxides. For low and medium iron contents (x = 0.25-0.5) the shape of the TiO z spectrum is not changed and a peak at longer wavelengths, with an hv o value of 2.2 eV, grows with the iron content. The 2.2 eV value also stays constant for higher x but the shape of the spectrum changes: the increase of the 104 FeTti_ Oy w,fh x=O 5 / o c (24h 600°C,Ar ) as-prepored (lh ~O0°C,o,r) / /" / ////'/ hv.(~) = 1'5 20 2q5 photon energy hv/eV 30 I 3.5 Fig. 9. Effect of anneahng on the photocurrent spectrum for the composition x = 0.5; the marks on the abossa, denoted by h v0 ( × , • or v), represent the onset energy of the corresponding peak lph(hP ) curve, which is rather steep for smaller x values, becomes much smaller so that the maximum photocurrent is reached at photon energies as high as 3.5 eV, compared to 2.7 eV for titanium-richer oxides. The spectrum of the annealed 3o- 4 ++1 2B as -prepared addl hona[ annent ,ng 26 *24 • . . . . . . ...~__ __ -...~ 22 20 i 01 i i 03 i i O5 x i i O7 i I O9 Fig. 10. The onset energy of the photocurrent, hp0, of the different peaks as a function of the composition x; hp o values are obtained from spectra as shown m Fig 9. 105 iron-rich oxides becomes very similar to that of pure Fe203. The difference between spectra of the as-prepared and the annealed oxides is obvious. Annealing yields the phases pseudobrookite and rutile in a medium range of compositton and the spectrum shows an additive behaviour of these two compounds. High iron contents change the spectrum towards that of Fe203 for the annealed oxide. Any contribution by "1"102 is probably hidden under the Fe203 spectrum. The magnitude of the photocurrent is also affected by the annealing process. In general it can be said that the current becomes smaller with annealing. This decrease is wavelength dependent and different for the various compositions. In order to give an impression of this effect, the ratio iap//laa(X) of the photocurrents for the "as-prepared", t,p, and the "additionally annealed" oxides, t,~, as a function of the composition x were calculated from the spectra. Figure 11 displays this plot for different wavelengths. The most important result is that high values of t,p/t~a of up to 400 are observed in the range x---0.25-0.7. These oxides are found to be amorphous. Annealing changes the photocurrents of the previously already crystalline oxides by a much lower extent. The maximum of the lap/laa(X ) c u r v e iS at about x = 0.4, but the values in the maxamum depend strongly on the wavelength. The high tap/Zaa values for the low energy light, reflect the shift of the spectrum to higher photon energies with annealing. Since the major effects are in the range where annealing changes the amorphous structure to a crystalline one, it can be concluded ¢.00~ Fe~Tq_xOy / \/55Onto i 350- 300- 250 -~ 200 150 100 50 -0 02 0~. x 06 OB 10 Fig. 11. Ratio of the photocurrents of the as-prepared, tap , and additionally-annealed, taa, oxades as a function of the composmon x, shown for vanous wavelengths X. 106 that the amorphous oxides gave a higher quantum yield than the crystalline ones. This is in contrast to expectation and a somewhat surprising result. Potential dependence and the effect of anneahng Like the wavelength dependence, the potential dependence of the photocurrent is also greatly influenced by the thermal post-treatment of the oxides. As an example, the current vs. potential curves of oxides with an composition x = 0.5 are shown in Fig. 12. The as-prepared oxide has the current onset at ca. 0.3 V, with a plateau of 0.1 # A / c m 2 up to 1.0 V. At higher potentials the current increases almost exponentially. A cathodic photocurrent could not be measured because the cathodic dark currents between 0.5 and 0 V were too large (see section 4). The annealed oxide, shown in the left part of Fig. 12, displays a quite different behaviour. At low potentials, the dark currents are small enough for the cathodic photocurrent to be measured. The current changes sign at ca. 0 V, a much lower potential than the current onset for the as-prepared oxide. The potential dependence of the anodic photocurrent is different as well; the curve becomes flatter at higher potentials. The photocurrents were measured using slowly chopped hght. The shape of the transients is different at different potentials. The as-prepared oxide shows a transient behaviour between 1.0 and 1.5 V, but not at lower or higher potentials, while the annealed oxide has no transient behaviour at all. The evaluation of the photocurrent from the transients is shown in the insets in Fig. 12. The change in properties of the oxides by annealing can be seen from the capacity 13 F.,T .O, W,Z.x=05 12 470nm II -o "-r o oC /65 -55 ADDITIONAL ANNEALING/ (LEFT SCALE)/,,-" 09 N os J ~ Z / AS PREPARED// (RIGHT SCALE)/ --t O 7 g__ o -I "45 ;~ = -40 -35 .~ 3o ~ 05. 25 ~ 04. 2O ~ 3:O6 ~ -50 03 15 02 I0 OI 05 0 0 05 (/V I0 15 Fig. 12 P h o t o c u r r e n t p o t e n t i a l c u r v e s , /ph(E), of the a s - p r e p a r e d a n d a d d i t i o n a l l y - a n n e a l e d oxide x = 0 5. The insets represent the t r a n s i e n t behavaour at the i n d i c a t e d p o t e n t i a l s a n d d e m o n s t r a t e , h o w t ph was evaluated. The two current scales are different. 107 FexTh_xOy wlfh x=05 15 t, -025 I / / 005 / add,honal an~eol~ / (2/*h 600"CAr) // -03 / 10 oo~ / / I 04 / No=2 1017cn~3 ~_ 003 6" / Mh~O'C air) 002 ii 05 10zoc,~3 8. / 10.001 / 1.10 / / 14 O0 05 10 15 20 ooo ¢/V Fig. 13. Schottky-Mott plots of the C(¢)-curves of the as-prepared and additionally-annealed oxide of the composiUon x = 0.5 behavlour. According to eqn. (10), carrier concentration and flatband potential can be determined from 1 / C 2 ( c ) (Schottky-Mott) plots. In Fig. 13 the effect of annealing on the capacity behaviour is shown for the composition where x = 0.5. Annealing reduces the excess carrier concentration by almost three orders of magnitude and the flatband potential is shifted to lower potentials by ca. 0.6 V. The ¢fb values for both oxides, 0.6 V and 0.0 V, are close to the onset potentials of the TABLE 2 The effect of annealmg m an Ar atmosphere for 24 h at 600°C on the photocurrent and capacity behavlour and the X-ray diffraction of an oxide with x = 0.5 Composmon ( x = 0 5) Photocurrent X-ray diffraction Capacity ([donor]/cm - 3 ) As-prepared Broad absorption, edge at 2 eV N o reflections of any Fe or Ti oxides 10 20 Annealed Distinct absorption by pseudobroolote and ruffle Reflections of pseudobrookite (F%TiO 5) and rutlle (T10 2) 2 x 1017 108 photocurrent in Fig. 12. In this way the results of the capacity and the photocurrent measurements agree reasonably well. The effect of the annealing of the oxide on various properties is summarized in Table 2. The anneahng process changes the oxides from an amorphous or microcrystalline into a crystalline structure which is associated with the occurrence of diffraction patterns. The photocurrent spectra are changed from a broad absorption to a distinct peak structure and the photocurrents drop, depending on the wavelength. The capacity behaviour reflects the crystallization, which decreases the donor density by three orders of magnitude. (6) MODEL OF THE OXIDE A description of the oxides in terms of a model has to include one of their basic characteristics, i.e. their non-crystalline response to X-ray diffraction. Therefore, an attempt to explain the observed electrochemical and photoelectrochemlcal behaviour has to deal with the amorphous structure of the material. Although amorphous semiconductors have been treated in solid state physics for over twenty years, there seems to be no application of them as electrode materials in electrochemistry (it has been assumed that passive films on metal electrodes are amorphous in some instances) and no investigation of their properties is known to the authors. Therefore, some characteristics of amorphous semiconductors will be discussed in the following, as far as they seem relevant to the experiments described in this paper. Amorphous semiconductor The amorphous state, also called non-crystalline or disordered, is basically characterized by the lack of a long range order in the material; only a short range order exists. The term short range order, however, covers a certain range of description depending on the degree of disorder. All electrochemical behaviour related to the solid state properties of the electrode material should show a dependence on whether the material is crystalline or not. However, very recent investigations with gold electrodes [27] show that there is no great difference between the amorphous and polycrystalline state in that case. For semiconductors more changes are expected, especially if electronic properties of the electrode material are involved. The density of the states at the band edges and in the band gap influences the electrochemical properties of the semiconductor to a great extent. Presumably, then, electrochemical behaviour, such as photoelectrochemical reactions, capacity behaviour and redox reactions, is affected by a change in the crystalline structure. As has been pointed out [28-30], the description of the electronic properties of amorphous semiconductors follows the same principles as known for crystalline ones. One of these principles is the description in terms of the density of states D(E), which is still applicable. However, the function D(E) is different within the bands and the band edges compared to a crystalline semiconductor. While in the 109 crystalline state the band edge is characterized by a sharp decrease in D ( E ) , non-crystalline sermconductors show extended tails into the band gap. Within the bands, D(E) show little structure compared to the crystalline state. The ideal crystal has no states in the band gap while the real one has, especially if intentionally doped. In Fig. 14(a)-(c) the different situations of an ideal and real crystal, and a disordered system are compared. While in the doped crystal, localized states occur only at certain energies (Fig. 14b), in an amorphous semiconductor they are present at any energy depending on D(E) in the gap (localized states are represented in Fig. 14c as a shaded area). The borderline between localized and delocalized states is not a material constant as is the band gap in Fig. 14a. With increasing disorder of the structure, these border lines can move into the band till, at a certain critical degree of disorder, no delocalized states are left in the band. In the amorphous semiconductor, at the transition from localized to delocalized states, the mobility,/~(E), drops by orders of magnitude (to zero at 0 K); this is shown schematically in Fig. 14d. The energy difference E c - E v = E g is the mobility gap and has become a usual characterization of amorphous semiconductors and their electrical properties. It should be kept in mind that the definition of Ec, E v and Eg are different for crystalline and amorphous semiconductors, for the latter it can, for example, depend on the preparation mode. The mobility gap can be determined from the temperature dependence of the dc conductivity. The total conductivity x is given by the integral = efD(E)#(E)f(E)dE x (12) where f(E) denotes the Boltzmann distribution function. Experimentally most materials obey the relation = A exp(-EJkT) (13) CBAND ONDUCTION IE Ec L~VALENCE ~ j Ev D(E) D(E) (a) (b) PERFECT DOPED CRYSTAL D(E) (c) ,u.(E) (d) DISORDERED SYSTEM Fig. 14 Schematic representation of the density of states, D ( E ) , for a perfect crystal (a), a doped crystal (b) and a disordered system (c), where the shaded area denotes locahzed states, (d) displays the mobdlty, # ( E ) , at the moblhty edge m a disordered system. 110 with Ea, the activitation energy of the actual process and A, a constant. Depending on the temperature, different mechanisms with different E a determine the conductivity. At higher temperatures, thermal excitation of carriers from delocahzed states in the valence band to delocalized states in the conducUon band contribute most to the conductivity. At lower temperatures other mechanisms become more probable, such as excitations into localized states, with a subsequent hopping for carrier transport, and hopping of carriers between localized states at the Fermi level. The latter process, which occurs only at very low temperatures, has, beside its low activation energy Ea, a low value of the constant A in eqn. (13). A distinction between the different mechanisms can be made, based upon their different activation energy E a obtained from In x ( 1 / T ) plots according to eqn. (13). The optical absorption a ' of crystalline semiconductors at the edge is described by eqn. (14), ignoring any electron-hole interaction (e.g. exciton formation) ot'h~,ec ( h u - Eg)" (14) with n = 2 for indirect transitions, n = 1 / 2 for direct allowed and n = 3 / 2 for direct forbidden transitions [31]. Equation (14) allows one to determine the optical band gap and the type of transition whtch ~s involved upon the different power law of the a b s o r p t i o n - p h o t o n energy relation. Very few crystalline semiconductors show, at low absorption coefficients, a so-called Urbach tail [32] which is described by the e m p m c a l eqn. (15): a ' = a o' exp[ 3"(hv - Eg)/kT] (15) with the constants a" and 3'- Up to now, no conclusive theoretical interpretation for this behaviour exists. While it is very rare for crystalline semiconductors, it is observed quite frequently and is obviously a general behaviour for amorphous semiconductors. A determination of the band gap according to eqn. (15) is difficult but is frequently done at the point where In a ceases to be linear with hp, a more or less arbitrary point. While transitions in crystalline semiconductors are determined by selection rules, some of these, especially the k-conservation rule, are not valid for amorphous materials [29]. Transitions then, are non-vertical in an E-k diagram without requiring phonon participaUon and are called non-direct. At higher absorption coefficients, above the Urbach tail, some amorphous semiconductors show a behaviour described by the relation a'hv = const.(hv - Eg) 2 (16) which ~s similar to that of indirect transitions in crystalline semiconductors (see eqn. (14)), but is related to a change of the density of states D(E) o~ v ~ [33]. However, power laws different from n = 2 in eqn. (16) are also found, with n = 1 and n = 3 in other cases. Values observed for the optical band gap are generally not very far from the gap in crystals. Both higher and lower values are found, but it should be kept in mind that the evaluation of the band gap involves some inaccuracy. Measurements of photoconductivity or photocurrents include both the absorption 111 and conductivity characteristics of the material. The photocurrent, normalized to the electric field E', or the photoconductivity are given, under steady state conditions, by [29] t p h / E ' = e/~r~/Io(1 -- R){1 d exp(- a'd) }/d (17) with ~', the carrier life time for monomolecular recombination; 77, the quantum efficiency, which is 1 if each absorbed photon contributes to the current; Io, the incident photon flux; R, the reflectivity and d, the thickness of the sample. In photoelectrochemlstry with crystalline semiconductors, the changing electric field in the space layer of the thickness dsc must be considered; then by expressing/x and ~-in terms of the diffusion length L, eqn. (17) changes to lph = e,/Io(1 -- R) 1 exp[ - a'dsc(' - 'fb)l/2] } 1 + a'L (18) A similar form of this was derived by Butler [34] following concepts developed by Gaertner [35]. Equation (18) can describe the potential dependence of the photocurrent. The underlying model, however, does not include a treatment of surface states and recombination process in the space charge layer. Therefore, the actual potential dependence of the photocurrent may be different from eqn. (18). Since no description of the photoelectrochemistry of amorphous semiconductors exists, photoconductivity measurements with amorphous semiconductors in solid state systems are briefly reviewed, which might demonstrate the differences in comparison to crystalline materials. In general, most amorphous semiconductors are good photoconductors. The spectral dependence of the photoconducltivity and the absorption show a similar behaviour near the absorption edge, as was shown in the case of chalcogenide films [36]. At higher photon energies, photocurrents stay nearly constant, while crystalline semiconductors often show distinct peaks which are due to surface recombination at high absorption levels, thus lowering the photocurrent. This indicates a recombination process in amorphous semiconductors being more uniform throughout the whole material. Another effect of non-crystallinity on the photocurrent is described by Davis and Mott [37,38]. It results in a shift of the quantum efficiency to higher photon energies and in an exponential field-dependence of the photocurrent. After excitation by a photon, hv, the carriers share the excess energy hv - Eg which is lost by thermalization in a very short time because of high scattering in the disordered structure. Therefore, electron and hole still have their mutual electrostatic attraction which reduces their separation and leads to recombination. This lowers the quantum efficiency, which is shifted to higher photon energies as has been observed with selenium [39]. On the other hand, the electrostatic attraction between electron and hole can be overcome by a sufficiently high electric field which lowers the escape barrier. The photocurrent then becomes field dependent according to a Poole-Frenkel mechanism. 112 Between the amorphous and the crystalline state of semiconductors, there certainly exast considerably differences. However some of these, for example the conduction mechanism of the solid, can be studied only under conditions which include temperatures much lower than those normally used in electrochemistry. This limits the information which can be obtained on the solids state properties of the electrode material in electrochermcal systems, although systematic work on the electrochemistry of amorphous semicondutors is still not being done. There is, however, a possible solution to these temperature limitations: that is, to extend the temperature range of electrochemical measurements to low values, including the freezing of the electrolyte. Very recently it was shown that electrochemical reactions can be studied, even in frozen electrolytes, down to 120 K [40]. Thus, with the knowledge of electrochemical reactions at low temperatures, the investigation of solid state properties also becomes possible in electrochemical systems using a wide range of temperatures. Photoelectrochemtcal behavzour Spectral dependence. The absorption characteristics of the oxides described here could not be determined separately. However, under the assumption that the carrier life time "r and mobility g are not a function of the photon energy, which is not always valid for an amorphous semiconductor, the measured photocurrent t ph in the region of the edge should be proportional to the absorption coefficient iph(hv) cc a ' ( h v ) (19) Then, according to eqn. (15) a straight line in a ln(hv.tph)(hv) plot indicates a so-called Urbach tail. In Fig. 15, corresponding plots are shown for the compositions x = 0.4, 0.5 and 0.7. In fact, straight lines are obtained in the region of the tail (see Fig. 8). The constant -/in eqn. (15), calculated from the slope, is 0.34 (x = 0.4 and 0.5) and 0.26 (x = 0.7), which is of the same order of magnitude as other amorphous semiconductors [29]. The upper limit of the linear behaviour which can be identified with the band gap is indicated with arrows in Fig. 15. At higher photon energies, a spectral dependence according to eqn. (20), which follows from eqns. (16) and (19), would be expected Iphhi) const.(hv - E s ) " = (20) However, all exponents, n = 1, 2 and 3, observed up to now in eqn. (16) do not fit our results when applying eqn. (20), which is analogous to eqn. (16). Only n = 0.5, normally assigned to direct transitions in crystalline material, following eqn. (14). yields a straight line (Fig. 16). There is obviously no previous case of an amorphous semiconductor with such a steep increase of tph(hv), and any explanation must be considered tentative. However, a very narrow valence band with a density of stat~ function much steeper than D(E) cc v~, winch would be assumed for the conduction band, could explain such a dependence [41]. Still, hv(tph = 0) values can b~ calculated from Fig. 16 and compared to those in Fig. 15, which shows that 113 i 6 i i i i i i i i FexTll xOy AMORPHOUS o A. zx o~o~° j. d ° , , ~ " . . o -'° ~o/ do,.o" p i o" ,op -- ' / / t!+ /? b )V /o o'° ~" ~d /o~ I 0 18 o x=o7 ,~=os ox=o4 II d / ~o,° z~ I 210 I I 22 I 214 I I 26 PHOTON ENERGY (h~/eV) Fig. 15 Plot o f ln(tr, hh~,)(h~), according to eqns. (15) and (19), for the amorphous oxides x = 0.4, 0.5 and 0.7; values are taken f r o m photocurrent spectra as shown in Fig. 8. reasonable agreement for the optical band gap is reached. The values are the same within +0.025 eV and are 2.0 eV (x = 0.7), 2.12 eV (x = 0.5) and 2.16 eV (x = 0.4). As described above and shown for x = 0.5 in Fig. 9, annealing has a considerable influence on the photocurrent spectra. The generally broad wavelength response is changed to a distinct peak structure, the edge is shifted to higher photon energies (Fig. 10), and the photocurrents are lowered by orders of magnitude depending on I I I I t l l FexTIi.xOy AMORPHOUS ox=07 ~Xx=O5 ! o,=o, g' ° d 0 I t8 d / ~ o°, z~.~,r~o,, ao ° ~ .'J ~ ?_.O I I I ?-2 / o, o2 d ~ ?-4 ?-.6 PHOTON ENERGY (hzs/eV) Fzg 16. Plot o f (lphhp)2(hp), according to eqn (20), for the amorphous oxides x = 0.4, 0.5 and 0.7; same v a l u e s as m Fig. 15; the m a r k s o n the a b c l s s a r e p r e s e n t the a r r o w s in Fig. 15, f o r c o m p a r i s o n . 114 I I I I I FexT=l_xOy 0 AMORPHOUS 0 CRYSTALLINE I [ I x=O 5 °/° . ,/J 8 N .¢: c. I ./." 0 I 18 i • o s~ ,,~'~o o ~ ; 20 I J 6~ • 4-~... I 22 I 24 I I 26 0 PHOTON ENERGY (hz,,/eV) Fig 17. Absorption edge for the compositionx = 0 5 of the amorphous oxide, accordingto eqn (20) with n = 0 5, and the crystalhneoxade,accordingto eqn. (14) with n = 2, valuesare taken from Fig 9. the wavelength. In addition, annealing also changes the type of absorption edge. While the amorphous material follows eqn. (20) with n = 1/2, after annealing a straight line is found only for an (hr. iph)l/2(hv) plot. This corresponds to indirect transitions in a crystalline semiconductor with n = 2 m eqn. (14). Both plots are displayed in Fig. 17 for x = 0.5 using the data of Fig. 9. Unfortunately, the absorption edge of the annealed sample overlaps with a prepeak, so that it cannot be traced to lower photon energies. Although Fig. 9 shows a shift to hagher photon energies with annealing, the optical gap obtained from extrapolation in Fig. 17 is, at ca. 2.0 eV, lower than the direct-transition like behaviour of the amorphous oxide. However, the photocurrents are much larger and increase more steeply with hv for the amorphous material (see Fig. 11), so that the lower band gap of the annealed sample is of no consequence. The observation of an indirect transition for the annealed oxide, which according to X-ray analysis consists mainly of Fe2T105, is in accordance with results of Ginley and Butler [6]. They found indirect transitions for various iron titanates, including Fe2TiO5. Their value for the band gap is higher, at 2.2 eV, but their sample preparation was different, especially due, to the much higher temperatures used. Potentzal dependence of the photocurrent. As described above and shown representively for x = 0.5 in Fig. 12, the potential dependence of the photocurrent depends strongly on whether the oxide is amorphous or crystalline. For the crystalline material eqn. (18) should be applicable for describing the potential dependence of the photocurrent. Using capacity data of the annealed oxide from Fig. 13, the following assumptions should hold for low band bending and not too high absorption coefficients a'L << 1 and a'd~(c - Cfb)]/2 << 1 115 Equation (18) can then be extended to (21) iph OCa'd~¢ (c - ceo) t/2 According to eqn. (21) a/p2h(E ) plot should yield a straight line with the interception C(tph = 0 ) = cfb. Such a plot is displayed in Fig. 18a. As expected, it is linear but deviates at higher potentials. The value for the flatband potential ~vo is m agreement with the capacity measurements within +0.1 V, and yields 0.0 V. The amorphous oxide shows a quite contrary behaviour. The photocurrent increases exponentially with c after passing a small plateau which sets in at ca. 0.4 V. A possible explanatton for the exponentml increase is that there exists a field dependence of the photocurrent as was found for various amorphous semiconductors. This was explained by Davis [38] with a Poole-Frenkel mechanism. In this case, the number of free carriers depends on the field and the logarithm of the photocurrent should be linear with the square root of the field according to eqn. (22) log 'ph = A'( E ' ) '/2 (22) A test of the validity of eqn. (22) by a log lph(c ~/2) plot lS shown in Fig. 18b. The electrode potential is chosen since although the actual field strength E ' is not known, the potential and the field in the oxide are assumed to be proportional to each other. Such behaviour, according to eqn. (22), is expected for fields larger than 104-105 V / c m . This would mean that the potential has to drop within the semiconductor in a surface layer of at most 100 nm. This condition is certainly fulfilled m our case, as the amorphous oxides behave rather like a highly doped n-type material (ca. 1020 cm -3) which results in a Debye length d~ < 10 nm. The plot in Fig. 18b is hnear over almost 0.5 V but shows some deviation at higher potentials. Therefore, it appears that eqn. (22) describes satisfactorily the potential dependence of the photocurrent. I [ i r ~ i i I I i i -50 FexTIl_xOy x=05 I0 O8 -55 (o} CRYSTALLINE (b)AMORPHOUS -60 ~E o -65 ~ oo 5 oe o. 04 -70 02 O0 i O0 02 04 ~/v 6 J 09 I10 i II 112 ~112/VII2 113 - 7 5 Fig 18 Potential dependence of the photocurrent for the composition x = 0 5 of the crystalhne oxide (a), according to eqn (21), and the amorphous oxade (b), according to eqn (22); values are taken from Fig. 12. 116 Description m terms of energy states The combination of electrochemical and photoelectrochemlcal data allows the description of the oxades in terms of energy states. The photoelectrochemical measurements give the optical band gap, while the kinetics of electron transfer reactions give information on the bands involved and their energy position. Capacity data yields the flatband potential and donor density and from this, the position of the Fermi level relative to the band edges can be estimated. For the crystalline semiconductors n-TiO 2 and n-Fe203 the band edges are shown in Fig. 19. For a description of the amorphous oxide it is assumed that the states related to iron and titanium have basically the same energy as the Fe203 and TiO2, respectively. However, the states are localized and their density D ( E ) depends on composition. The " b a n d gap" depends on the density of the states as well, which can be seen from Fig. 16 for the composition range x = 0.4-0.7. In the middle part of Fig. 19 an amorphous oxide FexTll_xOy for a medium iron content is illustrated. The dashed horizontal lines mark the position of the optical band gap, and the diagonal lines represent states related either to iron or to titanium, the fibrous structure of the lines in the middle represents the extension of states into the gap, similar to the picture in Fig. 14c. A change of composition x mainly influences the density of the states attributed to iron and titanium. Because in Fe203, both the conduction and the valence band edges exceed those of TiO 2 towards the band gap, major changes in the electrochemical behaviour are already expected for small values of x. The measurements of redox reactions and photocurrents confirm this. The capacity behaviour, however, does not appear to support this picture in the CONDUCTION BAND -0 A , - \ _ _ _ /H2U2/U 2 : ~ Fe2+/ge3+ "~H20/O 2 2 W -7 ~> ~ -3 -8 VALENCE BAND • r AMORPHOUS n-T~O 2 n-Fe203 FexT=l-xOy Fig. 19. SchemaUc r e p r e s e n t a t i o n of the b a n d edge of crystalline n-TiO 2 a n d n-Fe203 and a m o r p h o u s Fe~TI 1 _xOy of a m e d i u m c o m p o s i t i o n The abscissa denotes the distance r, the left o r d i n a t e the energy vs the v a c u u m a n d the right o r d i n a t e the electrode p o t e n t i a l vs S H E , S H E was a s s u m e d to be at a n energy of E = - 4 . 7 eV vs. vac. [42,43]. 117 same way. Assuming the Schottky-Mott equation (10) is still applicable for all compositions, a high donor density is observed. According to eqn. (23) then, E F = E ~ + k T In n c / n v (23) with E~, the Fermi level for the undoped semiconductor which is close to the mid gap position, and n c and n v, the carrier concentration in the conduction and valence band, respectively, the Fermi level is expected to be close to the conduction band for high donor concentrations. Amorphous semiconductors have a large density of localized states in the gap (see Fig. 14). This should pin the Ferrm level at a close-to-mid-gap position and make this rather insensitive to doping because of the large number of localized states already present. If, however, the density of the localized states below the conduction band and above the valence band vary considerably the Fermi level is shifted towards the higher density of the localized states and the amorphous semiconductor behaves more like a doped one. It is expected that annealing, i.e. a change from the amorphous to the crystalline state, will shift the Fermi level towards the conduction band if donor states are present which now become effective. Figure 13, showing the capacity curves for the amorphous and crystalline oxide, illustrates the correct tendency, namely, that the flatband potential is shifted to lower potentials with annealing. However, the position of the band edges would then have to be at higher energies by ca. 0.5 eV compared to the picture in Fig. 19. Therefore, it seems that further investigations are necessary to understand more clearly the capacity behaviour of amorphous semiconductors and the effect of crystallization on the electrode capacity. In Fig. 19, at the right axis, the equilibrium potentials of the different reactions are indicated. For the F e 2 + / F e 3+ redox couple, the changes of the reaction rate with the composiUon x can be explained as follows. The cathodic reaction can be observed with pure TiO2 because at potentials below 0.74 V the band bending at the surface is some 0.1 eV, allowing a measurable reduction of Fe 3+. At higher x more states are available below the conduction band. Also at these energies the electrolyte has more unoccupied states due to Fe 3+. aq ions and thus the current should be higher. In addition, the shift of the flatband potential to higher potentials with x, lowers the barrier for the electron transfer and increases the cathodic current at a given electrode potential. However, this effect of x on the cathodic current IS smaller than on the anodic one. For pure TiO2, the barrier for an anodic electron transfer from an occupied state in the electrolyte, Fe 2+. aq, to an unoccupied state in the conduction band is too high at potentials c > 0.74 V. Additional states below the conduction band lower this barrier considerably and allow either resonance tunneling into the conduction band or a hopping conduction in the localized states. It is understandable that only for x > 0.5 can anodic currents be observed. The low anodic transfer coefficient, a < 0.1, indicates that it is only a weakly potentialdependent process. The electron exchange happens above the Fermi level where a potential change does not change the occupation of either state. The break in the curve with a higher transfer coefficient at c > 1.5 V may perhaps be explained as follows. At 1.5 V, the Fermi level is at a mid gap position with respect to the iron 118 states. Therefore, at higher potentials an increasing number of these states above the valence band become unoccupied and can contribute to the ano&c electron transfer. On this basis, the explanation of the ano&c oxygen evolution is straightforward. With increasing iron content, more states above the valence band are available to facilitate the anodic electron transfer at higher energies, i.e. lower electrode potentials. This description, in terms of a band structure or, at least, energy states, has to stay a prehminary one as long as detailed stu&es of the effect of crystallinity of semiconductors on their electrochemical behaviour are lacking. (7) CONCLUSIONS The oxides described here seem to represent an interesting class of materials in several ways: because of an amorphous structure in a wide range of composition it was possible to study this with respect to the electrochemical and photoelectrochemical behaviour; the electrochemical properties of the electrode material can be changed by changing the composition in a rmxed oxide structure, especially if they are amorphous; the photoelectrochermcal behaviour shows a better photocurrent performance for the amorphous than for the crystalline oxides. The latter point is interesting as a possible apphcatlon of semiconductors in electrochemical solar cells. In contrast to expectation, the onset of the photocurrent of the amorphous oxides was at lower photon energies compared to the crystalhne oxides, which is basically due to a steeper increase of the current at the absorption edge. Moreover, the crystalline oxides showed a photocurrent lower by orders of magnitude. Although the performance quahties are far from being good for such an application, it seems worthwhile to consider amorphous materials as possible candidates for photoelectrodes in electrochemical solar cells. The experiments show the great impact the structure of the electrode material has, in a chemical as well as a physical sense, on the electrochemical behaviour. However, to give a definite description of amorphous semaconductors with respect to electrochemistry, more experimental and theoretical work must be done. ACKNOWLEDGEMENTS We thank Ms. K. Schubert who carried out preliminary experiments on oxide preparation, Mr. M. Hirsch who did the XPS measurements, Prof. J.W. Schultze for helpful discussions and Drs. D. Rath and W. Schmickler for critical reading of the manuscript; and special thanks are due to Prof. Sir N.F. Mott for stimulating discussions on amorphous semiconductors with one of us (U.S.). This work was in part supported by funds from the "Bundesminlster fuer Forschung und Technologle" of the German Federal Government. 119 REFERENCES 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 A Fujlsbama and K Honda, Bull. Chem. Soc. Jpn., 44 (1971) 1148 D. Haneman and F. Steenbeeke, J. Electrochem. Soc., 124 (1977) 861. A. Monnler and J. Augustynska, J. Electrochem. Soc., 127 (1980) 1576. K.L. Hardee and A.J Bard, J Electrochem Soc., 123 (1976) 1024 R K Qmnn, R.D. Nasby and R.J. Baughman, Mater Res. Bull, 11 (1976) 1011 D S Gmley and M.A. Butler, J Appl. Phys., 48 (1977) 2019 J H. Kennedy and K.W Frese, J. Electrochem Soc., 123 (1976) 1683. T Watanabe, A Fujlshlma and K Honda, Bull Chem. Soc Jpn., 49 (1976) 355. N. Hatanaka, T Kobayasht, H Yoneyama and H. Tamura, Electrochtm. Acta, 27 (1982) 1129. R C Weast, Handbook of Chemistry and Physics, CRC Press, Cleveland, 57th edn, 1976 J H. Scofieli:l, J Electron Spectrosc., 8 (1976) 129. Gmehn's Handbuch der Anorgamschen Chemae, Titan, 8th ed, 1951. G. Brown, m X-Ray Ident. and Cryst. Struct. Clay Minerals, London, 1961, p. 386. L Rooksby, m ref. 13, p 264 A.A van der Glessen, J Inorg Nucl. Chem., 28 (1966) 2155 L.A Bursdl, J Sohd State Chem., 10 (1974) 72. K J. Vetter, Electroctum Acta, 16 (1971) 1923 K.J Vetter, Electrochenucal Kinetics, Acadermc Press, New York, 1967 U. Stlmrmng, Dechema-Monographaen, Vol 90, Verlag Chemle, Weinhelm, New York, 1981, p. 259. W Schmackler and U. Stimmmg, Tlun Sohd Films, 75 (1981) 331. V A Myamhn and V. Pleskov, Electrochenustry of Semiconductors, Plenum Press, New York, 1967 B. Pettlnger, H R Schoeppel, T. Yokoyama and H Genscher, Ber. Bunsenges Phys. Chem., 78 (1974) 1024. R. De Gryse, W.P Gomes, F Cardon and J Venmk, J Electrochem. Soc., 122 (1975) 711 M. Takeuclu, T Itoh and H. Nagasaka, Tlun Sohd Fdms, 51 (1978) 83. F. Moellers, H.J Tolle and R. Memmmg, J Electrochem Soc., 121 (1974) 1160 J H. Kennedy and K W. Frese, J Electrochem. Soc., 125 (1978) 723. J Lipkowskl R.M. Reeves and M.R. Krlshnan, J. Electroanal Chem., 140 (1982) 195. M.H Cohen, H, Fntzsche and S.R. Ovslunsky, Phys Rev. Lett, 22 (1969) 1065. N.F. Mott and E.A. Davis, Electromc Processes in Non-crystalhne Materials, Clarendon Press, Oxford, 1971. J Tauc m S. Nudelman and S.S. Mltra (Eds), Optical Properties of Sohds, Plenum Press, New York, 1969. R H. Bube, PhotoconductLvity of Sohds, Wdey, New York, 1967 F Urbach, Phys Rev., 92 (1953) 1324. J Tauc in F Abeles (Ed), Optical Properties of Sohds, Elsevier, Amsterdam, London, 1972 M A. Butler, J Appl Phys., 48 (1977) 1914. W.W. Gaertner, Phys Rev., 116 (1959) 84 H.K Rockstad, J. Non-cryst. Sohds, 2 (1970) 192 E.A. Davas and N F Mott, Phi1 Mag., 22 (1970) 903. E.A Davis, J. Non-cryst. Sohds, 4 (1970) 107. J L Harke and P.J. Regensburger, Phys. Rev., 139 (1965) A970. U. Stimn~ng and W Schnuckler, J. Electroanal Chem, 150 (1983) 125 N.F. Mott, personal commumcatlon. R. Gomer and G TUcson, J. Chem. Phys., 66 (1977) 4413 W.N. Hansen and D.M. Kolb, J. Electroanal. Chem., 100 (1979) 493