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