Comparative Study of Photocarrier Dynamics in CVD-deposited CuWO<subscript>4</subscript>, CuO, and WO<subscript>3</subscript> Thin Films for Photoelectrocatalysis.

The temporal evolution of photogenerated carriers in CuWO₄, CuO and WO₃ thin films deposited via a direct chemical vapor deposition approach was studied using time-resolved microwave conductivity and terahertz spectroscopy to obtain the photocarrier lifetime, mobility and diffusion length. The carrier transport properties of the films prepared by varying the copper-to-tungsten stoichiometry were compared and the results related to the performance of the compositions built into respective photoelectrochemical cells. Superior carrier mobility was observed for CuWO₄ under frontside illumination.

Keywords: carrier dynamics; CuWO4; CVD; Metal oxides; photoelectrocatalysis; thin films; TRMC

Low cost and non-toxic semiconducting metal oxides are attractive candidates for the application as light absorbers and catalysts in photoelectrochemical (PEC) processes important for the production of solar fuels such as water splitting and CO2 reduction [[1]], [[2]], [[3]], [[4]]. Much research efforts in this area have been focused on the use of n-type binary transition metal oxides such as TiO2, ZnO, Fe2O3, and WO3 as photoanodes [[1]], [[5]], [[6]] or p-type oxides such as CuO and Cu2O as photocathodes [[4]], [[7]], [[8]], [[9]]. However, when incorporated into PEC cells, the photoelectrodes from these compounds remain far below theoretical expectations. Many of these materials suffer from disadvantageous photoelectrochemical properties, such as too positive band edge positions, as well as electronic properties, such as low conductivity and high photocarrier (the term photocarrier refers to either a photo-excited electron or hole) recombination, which limits the efficiency in PEC applications [[5]], [[10]]. As a consequence, the focus in research has recently turned from binary oxides to ternary oxides as they offer a broader possibility for the variation of composition. In turn, more options to tune the material's optical, electronic and catalytic properties arise. For instance, BiVO4 is one of the most cited candidates showing reasonable stability and promising photoelectrocatalytic conversion efficiencies [[11]], [[12]]. Since good conductivity has often been reported for some binary metal oxides, in particular for WO3, CuO and Cu2O, this prompted the search for other ternary oxide based on these materials and especially CuWO4 with an even lower bandgap (ca. 2.3 eV) than BiVO4 (2.4–2.5 eV) and excellent stability in aqueous electrolytes [[13]], [[14]]. A recent report has shown that the catalytic activity of CuWO4 for water oxidation is very good and quantitative hole collection at larger bias potentials (>1.23 V vs. RHE) was possible [[15]]. However, a major factor limiting the efficiency is the poor charge separation efficiency and short charge carrier diffusion length leading to enhanced recombination especially at lower bias potentials [[12]], [[13]], [[14]]. Metal oxide photocatalysts are generally prone to rapid recombination of photoexcited charge carriers occurring on all time scales from picoseconds to milliseconds (ps-ms). The exploration of photocarrier dynamics is therefore an essential key to help identify carrier lifetime and mobility impediments which severely affect the diffusion length of carriers.

Recently we reported on a direct chemical vapor deposition (CVD) approach [[16]] which allowed to deposit stoichiometric CuWO4 films by controlled variation of the Cu/W ratio between CuO, and WO3. Variation of CVD process parameters enabled us to tune the stoichiometry to obtain stoichiometric, W-rich, and Cu-rich deposits. The addition of tungsten during the deposition process reduces the bandgap and enhances the absorption properties of the material in the visible range. CuWO4 films were found to exhibit higher incident photon to current efficiency (IPCE) than WO3 oxide under frontside illumination.

Building on the reproducibility and tuning possibility of this deposition technique, we herein investigate the photodynamics of near stoichiometric CuWO4 films along with their CuO, and WO3 counterparts using time-resolved laser spectroscopy. We examine film compositions prepared via carefully optimized CVD conditions that yielded the best photoelectrocatalytic activity. In particular, we monitor the temporal evolution of photogenerated carriers in these three materials via time-resolved microwave conductivity (TRMC) and terahertz spectroscopy (TRTS) to obtain the photocarrier lifetime, mobility and diffusion length.

The CuWO4, CuO and WO3 thin films have been deposited on FTO coated 1.5 × 2.5 cm2 Pilkington TEC Glass (Xop glass, sheet resistance 12–14 Ω/sq.) or on quartz substrates using a custom built, horizontal cold-wall, low pressure MOCVD reactor; for details see Ref. [[16]]. Rutherford backscattering spectrometry (RBS) was used as a standard measurement for the determination of the elemental concentrations of the heavy elements Cu and W and in order to estimate the film thickness using the bulk density of the respective oxide. A 2.0 MeV 4He+ beam (intensity = 20–40 nA) was applied for films grown on Si(100) at RUBION, the Central Unit for Ions Beams and Radionuclides at the Ruhr-University Bochum. Nuclear reaction analysis (NRA) measurements were used for the concentration of the light elements nitrogen, carbon and oxygen via a deuteron beam (1 MeV). Using the SimNRA program the elemental concentrations were simulated combining values from RBS and NRA [[17]]. Further details and material characterization have been reported elsewhere [[16]]. Ultraviolet-visible (UV/vis) spectra were recorded with a Shimadzu UV-2600 spectrometer equipped with an integrating sphere IRS-2600. The spectra were corrected for scattering and reflection by placing the film at the entrance and at the exit of the integrating sphere to obtain the transmittance (Ft) and the reflectance (Fr), respectively. The absorptance was obtained as Fa = 1 − Fr − Ft. Tauc plots based on the UV/vis data were used to determine the optical bandgap [[18]]. The photocurrent measurements have been carried out using a SP-300 BioLogic potentiostat (Claix, France) and a three-electrode cell using a platinum counter electrode and an Ag/AgCl (3 M KCl) reference electrode. All potential values were recalculated and are reported with respect to the reversible hydrogen electrode (RHE). During the photocurrent measurements the photoelectrodes were pressed against an O-ring of the cell leaving an irradiated area of 0.5 cm2. The electrodes were irradiated from the backside (through the FTO glass) or from the frontside. For monochromatic wavelength-resolved measurements of incident photon-to-current efficiencies (IPCE) a tunable monochromatic light source (Instytut Fotonowy, www.fotonowy.pl) was used provided with a 150 W Xenon lamp and a grating monochromator with a bandwidth of ∼10 nm. Appropriate cut-off filters were used in order to eliminate second-order diffraction radiation. For the IPCE determination the average values out of measurements on three different electrodes were calculated for each curve. The value of photocurrent density was taken as the difference between current density under irradiation and in the dark. The IPCE value for each wavelength was calculated according to the equation (1),

Graph

$$\text{IPCE}(\%)=\left(\frac{i_{\text{ph}}}{\cdot }\right)\cdot 100$$ (1)

where iph is the photocurrent density, h is Planck's constant, c velocity of light, P the light power density, λ is the irradiation wavelength, and q is the elementary charge. The spectral dependence of lamp power density was measured by the NOVA II optical power meter equipped with a PD300-UV silicon photodiode (Ophir Optronics).

Time-resolved microwave conductivity (TRMC) measurements were performed using a gold-plated microwave cavity cell [[19]], [[20]]. A Q-switched frequency-tripled Nd:YAG laser operating at 355 nm with a 6 ns pulse width and repetition rate of 50 Hz illuminated the sample. The change in reflected microwave power ΔP/P from the cavity after optical excitation of the inserted sample was monitored as a function of time, and correlated to the photoinduced change in the conductance of the sample, ΔG, by

Graph

$$\frac{\mathrm{\Delta }}{P}\left(t\right)=-K\mathrm{\Delta }G\left(t\right)$$ (2)

where K represents a sensitivity factor derived from the dielectric properties of the medium and the resonance characteristics of the cavity. From the change in the measured photoconductance, the product of the charge carrier generation yield (ϕ) and the sum of electron and hole mobilities (Σμ) can be obtained according to [[21]], [[22]], [[23]]

Graph

$$\varphi \mathrm{\Sigma }\mu =\frac{\mathrm{\Delta }}{G}$$ (3)

where I0 is the incident laser intensity per pulse, e is the elementary charge, β is a geometrical factor related to the used microwave waveguide, and FA is the number of incident photons absorbed by the sample. The laser pulse intensities could be varied from ∼1013 to 1014 photons cm−2 pulse−1.

Terahertz spectroscopy was performed with a home-built THz spectrometer attached to a Coherent RegA femtosecond laser system consisting of a Ti:sapphire 80 MHz oscillator seeding a 150 kHz amplifier. The amplified pulses had a typical autocorrelation length of 70 fs (FWHM) and a maximum energy of 7 μJ at a wavelength of 800 nm. The beam provided by the system was split into three parts which serve for THz generation via optical rectification, THz detection via electro-optic sampling and optical excitation of the sample by a frequency-doubled pulse at 400 nm. The experimental setup covered a probe energy roughly from 0.5 to 2.5 THz [[24]]. The obtained signal with sub-ps time resolution is proportional to the THz photoconductivity.

In this study, the photocarrier dynamics and transport properties of CVD deposited CuWO4 thin films were analyzed by means of time-resolved conductivity measurements and compared to the binary counterparts CuO and WO3. As such experiments need to be performed on samples deposited on non-conductive substrates our CVD approach [[16]] that resulted in high quality CuWO4 photoabsorber films on F:SnO2 (FTO) was adapted to the deposition on quartz substrates. The process parameters could be tuned to yield stoichiometric, Cu-rich or W-rich films and deposits ranging from cupric oxide (CuO) via copper tungstate (CuWO4) to tungsten oxide (WO3) by adjusting the copper to tungsten ratio Cu/W. Figure 1a shows macroscopically smooth films of pure CuO, WO3 as well as of CuWO4 with a Cu/W ratio of 1.1 that exhibited the highest IPCE values when incorporated in a photoelectrochemical cell. Figure 1b directly visualizes a change in the absorption properties of the ternary compound in the visible range related to a decreased bandgap compared to WO3 and an increased bandgap compared to CuO. The film thickness of all three samples was in the range of 100 nm–150 nm, determined by RBS using the bulk density of the respective oxide. The results of the optical characterization via UV/vis spectroscopy including the derived optical band gaps are shown in Figure 2.

Graph: Fig. 1: (a) Photograph of CVD-grown thin films of cupric oxide (CuO), copper tungstate (CuWO4) with the ratio Cu/W = 1.1 and tungsten oxide (WO3) on FTO. (b) Bandgaps of the three material compositions CuO, CuWO4 with the ratio Cu/W = 1.1 and WO3 [[16]].

Graph: Fig. 2: UV/vis spectra (left) and Tauc plots (right) of CuO, WO3 and CuWO4 (Cu/W = 1.1) thin films (Abs. stands for absorptance) [[18]]. Bandgap energies of 1.4, 2.3 and 2.9 eV were determined for CuO, CuWO4 (Cu/W = 1.1) and WO3, respectively.

Figure 3 displays the photocarrier dynamics of the three films examined using TRMC with ns time resolution. The temporal development of the conductivity is derived from the absorption change of a fraction of the microwave electromagnetic field by free or loosely bound carriers [[21]]. The photoconductivity Δσ which is proportional to the product of the density of photoexcited electron-hole pairs, neh, and the sum Σμ of the combined electron and hole mobility (μe and μh) is written as:

Graph

$$\mathrm{\Delta }\sigma =e{n}_{\text{eh}}({\mu }_{\text{e}}+{\mu }_{\text{h}})=e{n}_{\text{eh}}\mathrm{\Sigma }\mu$$ (4)

Graph: Fig. 3: Time-resolved microwave conductance of CuO, WO3 and CuWO4 (Cu/W = 1.1) thin films in (a) 100 ns and (b) 10 μs time windows excited by a 355 nm 6 ns laser pulse of ∼4 × 1013 photons cm−2 pulse−1. The dashed lines represent biexponential fits to the respective data (see text). On the right side the data are shown in a normalized graph.

Figure 3a and b depict the TRMC mobility decay in a 1000 ns and in a 10 μs time window after excitation with a 7 ns laser pulse of 355 nm with an intensity of ∼4 × 1013 photons cm−2 pulse−1. For an easier comparison of the decay times the same data are shown also in a normalized way. A lower limit for the mobility of the carriers can be obtained from the conductive transient peak assuming no major recombination losses at this early time [[25]], [[26]]. With the carrier mobility μ and the lifetime τ we are able to extract the carrier diffusion length L, given by L = (Dτ)1/2 with D = kTμ/e. To obtain the carrier lifetime we employed a double exponential model fit to the decay of the signal adapted from Savenije et al. [[23]]:

Graph

$$f\left(x\right)=a\cdot \exp \left(\frac{-}{x}\right)+b\cdot \exp \left(\frac{-}{x}\right)+c$$ (5)

The respective parameters for the examined samples are summarized in Table 1. The small offset parameter c represents a remnant small conductivity from poorly mobile extremely deeply trapped carriers.

The CuO sample exhibits the lowest mobility (see Table 1) and the fastest decay time which fits to numerous reports attesting photovoltaic and PEC performance of this material much lower than the theoretical maximum, due to fast carrier trapping and recombination [[27]], [[28]]. The TRMC signal drops by about 80% of its initial value within less than 50 ns crossing over into a small longer-lived tail that decays on the μs scale. Reported mobility values of CuO depend very much on the preparation parameters and the diverse measurement techniques and range between 10−4 and 102 cm2 V−1 s−1 [[29]], [[30]], [[31]], [[32]], [[33]], [[34]], [[35]]. For instance, Hall mobilities of CuO films are usually higher than field effect mobility obtained from thin-film transistors (TFT) measurements. High field electrode measured mobilities based on DC carrier injection methods, however, often emphasize band transport. TRMC probes free carriers shortly after photoexcitation and can give more insight into trapping processes that lead to an effective loss of mobility over time due to carrier localization. Keeping in mind that the conductivity is a product of the mobility and carrier density and assuming no severe carrier loss in the first few 10 ns, according to Figure 3 the mobility would have dropped already to 20% of its initial value. Such low mobility makes it hard for excited bulk carriers to reach the surface and makes them prone to recombination. Defects and impurities present in copper oxide are made responsible for a large number of hole traps associated with copper vacancies that are considered the origin of the p-type conductivity. It was recently proposed in a theory paper that CuO, being a magnetic oxide, also conducts holes through thermally activated spin polaron hopping due to spin flipping at Cu sites [[36]]. Carrier mobility is reduced not only because of strong electron–phonon coupling as in many oxides but also as a consequence of strong magnetic coupling. Othonos and Zervos investigated CuO nanorods with transient absorption as well as time-resolved photoluminescence (TRPL) [[37]] and observed an ultrafast initial drop in absorption (bleach) ascribed to the occupation of states in the valence bands or sub-bands with a recovery time of 10 ps. TRPL exhibited a decay signal up to tens of ns and was associated with carrier recombination via bandgap states. While sub-ns dynamics cannot be resolved with the TRMC setup the longer time component agrees well with the early decay of the photoconductivity of CuO shown in Figure 3. Born et al. [[38]] reported band-edge and trap-assisted recombination via midgap states with a short time constant of 50 ps in CuO nanocrystals measured using transient absorption. It can therefore be assumed that trapping and recombination sets in far earlier than the ns resolution of the TRMC experiment can time-resolve. Recently, Borgwardt et al. demonstrated for crystalline copper oxide (Cu2O) in a time-resolved photoemission experiment that copper vacancies can lead to ultrafast trapping of conduction band electrons within one picosecond preventing excited carriers to reach the surface at conduction band levels [[39]]. The electrons must then migrate to the surface via thermally activated hopping. Consequently, the persisting longer ns time-scale lasting CuO conductivity decay τ2 in Figure 3 is attributed to a sequence of even slower trapping/de-trapping processes leading to a remnant very low mobility of longer surviving carriers. A similar behavior as for Cu2O may also be possible for CuO. For the above reasons we chose to calculate an effective diffusion length L (Table 1) calculated only from τ1 that adopts transport of semi mobile carriers consecutive to ultrafast ps trapping and prior to deep trapping processes. The considerations above may to some extent be responsible for the known poor performance of CuO electrodes in photovoltaic and photoelectrochemical cells.

Tab. 1: Carrier lifetime, mobility and diffusion length derived from TRMC transients of Figure 2.

CVD preparation of n-type WO3 semiconductor films according to the optimized deposition condition reported by Peeters et al. led to a nearly perfectly stoichiometric metal to oxygen ratio [[16]]. The TRMC decay of such a sample prepared on a quartz substrate following laser excitation is shown in Figure 3 a,b. Like for CuO above, we employ a double exponential model fit to the decay of the photoconductivity curve. The parameters shown in Table 1 indicate a much longer carrier lifetime and superior mobility for WO3 compared to CuO. The TRMC transient exhibits an initial decay and a longer component with time constants τ1 = 45 ns and τ2 = 1490 ns, respectively. The first decay time fits well to early TRMC measurements on spray pyrolyzed WO3 films by Patil et al. who reported a biexponential conductivity decay with 40 ns as first and ∼160 ns as second time constant [[40]], [[41]]. The authors claim that most of the measured conductivity can be attributed to conduction band electrons that represent the majority carriers with a mobility of 12 cm2 V−1 s−1 about four orders of magnitude higher than our value listed in Table 1. One explanation in the discrepancy between the reported mobilities could be sought in the nature of trap filling [[25]], [[42]], [[43]], [[44]] as reported for metal oxide semiconductors such as BiVO4, Cu2O or TiO2 in pulsed laser experiments. As a higher number of traps becomes populated with increasing pump laser power the measured mobility derived at the peak of a conductivity transient can approach band-like conductivity. This can be considerably higher than mobility of carriers localized at bandgap states such as defects or polarons, the latter referring to carriers that are self-trapped in a potential well they create by polarizing the surrounding atoms [[45]]. Carrier transport then proceeds thermally activated via strong electron-phonon coupling. Typically, reported mobilities for WO3 are in the range of 10 cm2 V−1 s−1 but even higher mobilities of about 35 cm2 V−1 s−1 were published using photo-Hall measurements [[46]]. Our TRMC experiments performed in a resonant microwave cavity [[26]] allow for measurements at laser pump powers of at least a factor of ∼100 lower and possibly better reflect the impact of native defects in WO3 with respect to trapping dynamics. The presence of traps has been reported by several groups [[41]], [[46]], [[47]], [[48]], [[49]], [[50]], [[51]], [[52]], [[53]], [[54]], [[55]]. But also the CVD deposited WO3 film studied here is stoichiometric [[16]] which possibly makes a big difference in carrier transport properties. Sachs et al. recently reported on the impact of oxygen-deficient WO3 in comparison to near-stoichiometric samples with respect to photocarrier dynamics and photocatalytic activity [[55]]. In a combined experimental and theory study the authors find that photoexcited holes are trapped on an ultrafast time scale at occupied oxygen vacancy states within the bandgap. Using transient absorption they derive that in highly oxygen-deficient material most of these holes are deeply trapped and almost unable to contribute to a photochemical reaction because of poor carrier transport. The localized carriers cannot be thermally activated and are considered the origin of the well-known photochromic effects of WO3 [[56]]. The report also explains why sub-stoichiometric samples, although giving rise to an enhanced absorption in the visible light region, show photoactivity mainly only in the UV region. A single oxygen vacancy donates two excess electrons to the lattice that localize at W sites and reduce W6+ to W5+ resulting in the formation of small polarons as has been reported for a vast number of metal oxides including WO3 [[45]], [[47]], [[56]], [[57]], [[58]], [[59]], [[60]], [[61]]. The small polaron hopping mechanism is known to exhibit low carrier mobility often leading to early recombination before photocarriers can reach the surface of the semiconductor. For BiVO4 for instance, one of the best performing oxides in photoelectrocatalysis, low mobility in the range of 0.01 cm2 V−1 s−1 was found [[25]]. The poor transport is, however, compensated by a long carrier lifetime which translates into a considerable diffusion length of around 40 nm. Lifetime and mobility derived for WO3 from our TRMC measurements listed in Table 1 are only slightly inferior to BiVO4 and the diffusion length yields ∼20 nm. Since hole trapping, as discussed above, is very effective in WO3, it can be concluded that the measured TRMC conductivity is mainly associated to mobile electrons. We prefer considering this an effective diffusion length calculated from the shorter TRMC time constant τ1 taking into account preceding faster carrier trapping processes that predetermine consecutive slower carrier transport via localized states. The longer μs time constant τ2 then reflects carriers that survive earlier recombination at the expense of extreme slow effective mobility due to slow de-trapping. Hall effect measurements have shown that the concentration of intrinsic mobile carriers in near-stoichiometric WO3 is larger by ∼5 orders of magnitude compared to highly oxygen-deficient counterparts [[55]]. This could possibly explain why other groups report extremely long lifetimes and mobilities resulting in long diffusion lengths using high-field injection measurement techniques that address only one carrier type [[46]], [[62]]. Ultrafast trapping in WO3 particles was also observed in TRTS experiments with conductivity decay times between 100 and 600 ps depending on the particle size [[63]]. The relaxation process was attributed to the filling of trap states. At early times ∼100 ps, the majority of trapping was related to bulk defects followed by later trapping at grain boundaries. An analogous behavior was found for BiVO4 where early trapping on a similar time scale precedes the growth of a polaron population that strongly impacts carrier transport [[42]]. Koide et al. [[64]] and Uemura et al. [[65]], [[66]] studied the early photoexcited state of BiVO4 and WO3 at atomic scale using transient X-ray absorption fine structure spectroscopy (XAFS) and transient X-ray absorption near-edge structure (XANES) spectroscopy. For both materials, photoabsorption of a short laser pulse led to a distortion of the W-O bonds and to the formation of a metastable state within 60 ps and 150 ps for BiVO4 and WO3, respectively. This state was associated with the photo-induced reduction of W6+ to W5+ and ascribed to the appearance of a small polaronic state. The metastable state decayed in about 2 ns and merged into a longer-lived component. The TRMC experiment shown in Figure 3 has a time resolution of a few ns and most probably probes the remaining polaron mobility at the initial stage of the recorded transient implying that our derived mobility (Table 1) may represent a lower bound. Such a phenomenon was also found for BiVO4 where ps-resolved TRTS measurements [[42]] produced peak mobilities by a factor of 10 higher than ns-resolved TRMC experiments.

The time-resolved photoconductivity of stoichiometric CuWO4 with a Cu/W ratio of 1.1 deposited according to the optimized CVD conditions described by Peeters et al. [[16]] are shown in Figure 3. Compared to the binary compounds the mobility derived from the TRMC experiments performed on CuWO4 is the highest of the three deposited films. The lifetime of the photoexcited carriers more closely resembles the dynamics observed for WO3 with the decay of the transient also divided into two parts. The first decay parameter τ1 is almost identical whereas τ1 is slightly shorter for CuWO4 which can be seen in the normalized plot of Figure 3. In comparison to all other Cu/W ratios and CVD deposition conditions (not shown here) the Cu/W = 1.1 sample exhibited the longest decay time parameter τ1 while longer decay times up to a factor of ∼2.5 for τ2 were observed for other ratios. The resemblance with WO3 in the early TRMC time window lets us assume that the carrier relaxation mechanism is also similar and that a fast trapping mechanism as discussed above for CuO with potentially accountable Cu vacancies is not mainly involved. A diffusion length of ∼25 nm using the peak mobility and τ1 as carrier lifetime is derived for CuWO4 which is superior to the WO3 sample and close to BiVO4 [[25]]. Quantitative hole collection [[15]], [[67]] has been reported for films fabricated by atomic layer deposition (ALD) and electrochemical preparation under usage of appropriate scavengers to circumvent recombination via midgap surface traps [[68]] and ameliorate the slow oxidation kinetics [[69]] at the CuWO4 surface. A similar occupied midgap state corresponding to Cu 3d states was predicted in a combined photoelectrochemical and computational study [[70]]. According to calculations by Thang et al. [[71]], in oxygen-deficient CuWO4 two excess electrons can localize at Cu 3d as well as at W 5d states and reduce the transition metal ions from Cu2+ to Cu+ and W6+ to W5+, respectively. Their analysis claims that the energetically most convenient process is the transfer of excess electrons to Cu 3d empty states. For non W containing CuO trap-assisted recombination via midgap states was experimentally proven to be as fast as 50 ps [[38]]. These findings could possibly represent a major difference in the transport properties between CuWO4 and WO3. Although quantitative hole collection was demonstrated in a PEC for all holes that actually made it to the interface, bulk recombination could not be eliminated which is also the case for the stoichiometric materials as presented in our study. However, IPCE of nearly 100% was found for BiVO4 where TRTS experiments indicate that photoexcited carriers rapidly trap within 100 ps and form polarons that all make it to the surface [[42]]. The possibility of polaron formation in CuWO4 crystals was already reported in 1992 by Mathew et al. [[72]] applying temperature dependent electrical conductivity and recently picked up again in a DFT study by Hoang et al. [[58]]. Hole polarons are found to easily localize around oxygen atoms due to a an extremely low self-trapping energy of only 50 meV. This is consistent with the effective trapping of photoexcited holes in WO3 as discussed above. According to the calculations Cu, is found to be stable as Cu2+ in CuWO4 thus favoring electron transport in the form of polarons via W sites. This emphasizes our experimental findings that the TRMC carrier dynamics of CuWO4 better compare to WO3 than to CuO. We also performed TRTS measurements on CuWO4 to gain insight into the very early ps carrier dynamics. Figure 4 shows an ultrafast decay of the conductivity signal of ∼10 ps which is much faster than for BiVO4 [[42]]. In comparison, we have reason to believe that this decay represents rapid trapping and not recombination as is the case for BiVO4 that exhibits a loss of 80% of initial conductivity due an effective decrease of mobility connected to efficient trapping. The signal in Figure 4 does not completely drop to zero which is evidenced explicitly by the photoconductivity recorded at later times with TRMC that has ns time resolution. A tentative explanation for the fast dynamics could be that photoexcited electrons possibly first localize at shallow Cu sites leading to the prompt decay of conduction band carriers and subsequently populate W sites that take over responsibility for the main macroscopic transport like in WO3.

Graph: Fig. 4: TRTS photoconductivity transient of a CuWO4 (Cu/W = 1.1) thin film sample in a time window of 60 ps.

The peaks of the mobilities of the CuWO4 film obtained by TRMC at different laser intensities are displayed on the left side of Figure 5. The maximum values of the transients are a function of the number of incident photons per laser pulse and decrease towards higher intensity. The peak mobility at 4.1 × 1013 photons cm−2 pulse−1 noted in Table 1 drops by ∼60% when the fluence is increased to 2.3 × 1014 photons cm−2 pulse−1. A similar behavior was also found for the CuO and WO3 films, however, at lower initial maximum values as shown in Figure 5 on the right side. Such an intensity dependency was previously reported for undoped BiVO4 samples and was attributed to a combination of first and second order recombination [[25]]. The trend showing increasing mobility towards smaller laser intensity could suggest even higher mobility and longer diffusion length for photon densities comparable to AM 1.5 sunlight. However, lower fluence than ∼3 × 1012 photons cm−2 pulse−1 could not be used to probe the film due to low signal-to-noise ratio. Also, an uncritical extrapolation of the trend cannot be done as the phenomenon of trap-filling may mask lower trap-dominated effective mobility values at low photon fluences.

Graph: Fig. 5: Left: TRMC photoconductivity transient of a CuWO4 (Cu/W = 1.1) thin film for different excitation densities; right: Comparison of the peak mobilites of CuO, WO3 and CuWO4 films for different excitation densities.

Other stoichiometries with higher Cu/W ratios of 2.1 and 2.7 were also analyzed with TRMC but showed poorer mobility values at comparable excitation densities. Three different samples with an approximate ratio Cu/W = 1 from different batches prepared in intervals of several months exhibited a mobility spread of only 4–6 (10−3 cm2 V−1s−1) illustrating the good reproducibility of the CVD technique. Since SEM investigations [[16]] revealed better homogeneity for the latter stoichiometry the superior mobility may partially be related to a lower number of traps due to a lesser influence of grain boundaries. However, since carrier transport is expected to be polaronic small changes in local structure between different stoichiometries and possibly the concentration of oxygen vacancies may also affect polaron formation as well as activation energies. Introduction of foreign atoms as appropriate dopants may also help to engineer and potentially increase the diffusion length of these polarons.

Photoaction spectra (IPCE vs. wavelength plots) of CuO, WO3 and CuWO4 (Cu/W = 1.1) thin films prepared on FTO under the same CVD conditions [[16]] as for deposition on quartz are shown in Figure 6. The three different compositions were measured in borate electrolyte (pH 7) at 1.12 V vs. RHE under irradiation from the frontside and from the backside. To a great extent the spectra mirror the trend found in the TRMC measurements. The highest overall mobility observed for CuWO4 (Cu/W = 1.1) coincides with the highest IPCE results obtained under frontside illumination with UV light. Yourey et al. also observed slightly higher photocurrents for electrodeposited CuWO4 films compared to WO3. Upon backside illumination, the WO3 CVD-grown photoanodes exhibit the best photoconversion efficiencies.

Graph: Fig. 6: IPCE of CuO, WO3 and CuWO4 (Cu/W = 1.1) films for frontside (left) and backside (right) illumination.

CuO that displayed the fastest decay of photoconductivity in TRMC experiments produced no photocurrent at all in a PEC which is in line with the p-type conductivity of this material and the somewhat negative valence band position rendering the oxidizing power of holes very low. The higher IPCE of CuWO4 under frontside illumination indicates a longer diffusion length for electrons (majority carriers) than for holes (minority carriers) in line with fast hole trapping as discussed above. For UV light, most of the charges are generated near the semiconductor/electrolyte interface under frontside illumination, and holes can more easily reach the electrolyte. Under the backside illumination the holes must travel the largest distance to contribute to the photocurrent. Neglecting space charge layer effects, the carrier diffusion length of ca. ∼25 nm calculated from TRMC does not completely match the layer thickness (100–150 nm) so that not all photocarriers make it to the surface. Previous SEM measurements indicate the fewest grain boundaries for CuWO4 (Cu/W = 1.1) which would imply less carrier scattering and higher effective carrier mobility [[16]]. Our observations give promise for CuWO4 as a future high quality thin top-absorber in a photocatalytic tandem or multilayer device. For thicker layers other preparation techniques may be favorable, such as for instance bulk-solution concepts where CuWO4 grains are electrically wired with carbon nanotubes in a CuWO4 carbon nanotube nanocomposite to enhance inter-grain transport [[73]]. In comparison to CVD grown CuWO4, the long-lived higher TRMC conductivity tail of WO3 that prevails beyond the 10 μs range, as shown in Figure 3 (normalized curves), could possibly point to a small fraction of extended carrier diffusion via trapped/de-trapped holes that contribute to the photocurrent to some extent. This could be the reason for higher IPCE values of WO3 under backside illumination. From the above discussions concerning the latter material we propose that mixing WO3 with Cu to form CuWO4 represents a superior way of increasing visible light absorption compared to deliberate incorporation of oxygen vacancies.

Photogenerated charge carrier dynamics in CuWO4, CuO and WO3 thin films grown by CVD was studied using time-resolved microwave conductivity and terahertz spectroscopy. On the nanosecond timescale the combined carrier mobility and diffusion length in CuWO4 with Cu/W = 1.1 exceed the values for the binary compounds CuO and WO3. This is in line with the analysis of wavelength-resolved photocurrent (IPCE) spectra recorded under illumination from the frontside vs. backside which points to hole transport as the efficiency-limiting step. Taking into account the excellent surface catalytic properties and good stability of CuWO4, our results suggest that one way CuWO4 could be employed beneficially would be in form of a thin layer covering another photoanode with optimum light harvesting and charge separation properties. Alternatively, as the efficiency of CuWO4 is chiefly limited by short hole diffusion length of ca. 25 nm, a viable approach to improvements in photoconversion efficiency might include fabrication of carefully designed, thicker porous photoelectrodes in which the primary crystallite size is in the range of ca. 50 nm (assuming that the concentration of electrons in the porous and compact photoelectrode is similar, i.e. at small applied positive bias and/or if the thickness of the space charge layer is significantly smaller than the crystallite size).

This work has been funded by the German research foundation (DFG) within the Priority Program SPP 1613 (Solar H2; DE 790/12-1, BE 5102/4-1, EI 673/3-1). S.M. is grateful for financial support from the BMBF through the MeOx4H2 project, funding Nr. 03SF0478A. Detlef Rogalla from the RUBION facility at the Ruhr University is thanked for the RBS/NRA measurements and evaluation.