Formation and suppression of defects during heat treatment of BiVO4 photoanodes for solar water splitting

Metal oxide photoelectrodes typically suffer from poor carrier transport properties and extensive carrier recombination, which is caused by the presence of intrinsic or extrinsic defects in the material. Here, the influence of annealing temperature and atmosphere on the formation and suppression of defects in BiVO4—one of the best performing metal oxide photoanodes—is elucidated. Annealing in argon has little or no effect on the photoelectrochemical performance due to the competing effects of an increase in grain size (i.e., reduction of grain boundaries) and the unfavorable formation of oxygen vacancies. When annealing in air, the formation of oxygen vacancies is suppressed, resulting in up to 1.5-fold enhancement of the photocurrent and an order of magnitude increase of the charge carrier mobility. However, vanadium leaves the BiVO4 lattice above 500 C, which leads to a decrease in carrier lifetime and photocurrent. This vanadium loss can be avoided by supplying excess vanadium in the gas phase during annealing. This leads to enhanced charge carrier mobility and lifetime, resulting in improved photocurrents. Overall, this strategy offers a general approach to prevent unfavorable changes of cation stoichiometry during high-temperature treatment of complex metal oxide photoelectrodes.


Introduction
Photoelectrochemical (PEC) water splitting is a promising pathway for renewable energy production. Using a semiconductor immersed in an aqueous electrolyte, water is converted into hydrogen and oxygen by solar irradiation. Typically, this process can be divided into (i) light absorption in the semiconducting photoelectrode to generate charge carriers (electrons and holes), (ii) separation of the charge carriers, and (iii) water oxidation and/or reduction at the surface of the electrodes. 1 Metal oxides are particularly attractive as photoelectrode materials, mainly due to their generally good aqueous stability, non-toxicity, and low cost. [2][3][4] However, their PEC activities are typically limited by bulk recombination, which is partly due to the presence of defects. Metal oxide photoelectrodes are oen synthesized with low-or moderatetemperature processes, such as drop casting, spray pyrolysis, or electrodeposition. Although the simplicity of such processes offers important advantages, it tends to create relatively high defect densities (>10 18 cm À3 ) in the form of intrinsic point defects (e.g., cation and oxygen vacancies) as well as non-crystalline phases. 5 While some of these defects can promote the conductivity and therefore improve the PEC performance of the photoelectrode material, 6-9 many of these defects form electronic states deep in the bandgap of the oxide. Such states act as efficient charge carrier trapping or recombination centers, resulting in short carrier lifetimes. In addition, the electronic structure of many metal oxides results in large effective masses for the electrons and holes. In combination with the typical polaronic nature of electronic charge transport, this results in poor charge carrier mobilities. 10,11 It is the combination of short lifetimes and poor carrier mobilities that causes short carrier diffusion lengths and, therefore, recombination of the carriers before they are able to reach the interface.
The above-mentioned limitations also apply to bismuth vanadate (BiVO 4 ), which is one of the best performing metal oxide photoanodes for solar water splitting. The monoclinic scheelite phase, which is the most photoactive one, has a bandgap of 2.4-2.5 eV. 12 Carrier transport in the material is slow, with reported carrier mobilities in the range of 10 À2 -10 À1 cm 2 V À1 s À1 . 10,13,14 In addition to having a small polaron conduction mechanism, 10,11,[15][16][17] the presence of defects in BiVO 4 may lead to trap-mediated transport and/or bound polarons that reduce the mobility even further. 13,14,18 Nevertheless, through various developments in the last 5-10 years, $90% of the theoretical maximum photocurrent of BiVO 4 (7.5 mA cm À2 , assuming all AM1.5 photons with energy higher than the 2.4 eV bandgap contribute to the photocurrent) has already been achieved. 19 Pihosh et al. realized this record photocurrent by depositing a thin layer of BiVO 4 onto WO 3 nanorods to form a guest-host nanostructured photoelectrode. This orthogonalizes the direction of optical absorption and charge transport, 20,21 which ensures that enough photons can be absorbed while the photogenerated carriers only need to travel short distances.
Despite these encouraging results, the fabrication of a complete solar water splitting device with a nanostructured BiVO 4 photoanode is not straightforward. To generate the $1.5 V needed for water splitting, the BiVO 4 is usually placed in front of one or two smaller-bandgap semiconductors to form a tandem device. 22,23 Unfortunately, extensive optical scattering from nanostructured BiVO 4 photoelectrodes prevents the low energy photons from reaching the bottom absorber. To circumvent this, Pihosh et al. used Pt as a reective substrate and placed a doublejunction GaAs/InGaAsP tandem solar cell in front of their BiVO 4 at a 45 angle. The highly efficient III-V tandem cell absorbed enough of the diffusely back-reected photons from the WO 3 / BiVO 4 nanowires to reach an impressive 8.2% solar-to-hydrogen efficiency. 19 In another report a beam splitter was used to direct the short wavelength part of the solar spectrum to the WO 3 / BiVO 4 electrode and the long wavelength part to a perovskite solar cell, resulting in a device with a 7.7% solar-to-hydrogen efficiency. 24 For practical applications, however, a simple stacked tandem conguration with a non-scattering top absorber (and without expensive III-V semiconductors) is preferred. 25 To reduce the scattering of the WO 3 /BiVO 4 nanowires one could revert to photolithographic techniques to create highly regular WO 3 nanopillars followed by atomic layer deposition to deposit a BiVO 4 top layer with a homogeneous thickness. In this paper, we aim for a potentially easier approach, which is to improve the carrier transport properties of the BiVO 4 . If successful, this would avoid the need for nanostructuring altogether.
One possible way to alleviate carrier transport limitations is to apply a high temperature treatment that may reduce the concentration of point defects and improve the crystallinity. 26,27 While changes in the crystallinity can be readily observed by X-ray diffraction, changes in the nature and concentration of point defects are more difficult to track. To the best of our knowledge, no systematic studies have yet been carried out that correlate the inuence of such heat treatments on the defect properties and the resulting PEC performance. In this paper, we systematically elucidate for the rst time the formation of defects in BiVO 4 thin lms at high temperature in both oxidizing and reducing environments. We nd that high temperature treatments of BiVO 4 in air lead to an increase in grain size and improved carrier dynamics. At temperatures above 500 C, however, vanadium is lost from the lattice. The concomitant decrease in photocurrent is presumably caused by the formation of vanadium vacancies. By introducing excess vanadium (V) in the gas phase during the heat treatment the formation of these vacancies can be avoided, leading to improved photocurrents.

Results and discussion
We rst investigated the formation of defects in BiVO 4 at high temperature. Mass spectroscopy (MS) was performed on BiVO 4 powders to analyze possible losses of chemical species during heat treatment. BiVO 4 powders were heated to 700 C in air at a ramp rate of 7 C min À1 and then kept at this temperature. The MS signals represented as ion currents are shown in Fig. 1 as a function of time and temperature. Signals of V + (mass/ charge ratio, m/q, of 51, green curve), VO + (m/q ¼ 67, blue curve) and VO 2 + (m/q ¼ 83, purple curve) are clearly detected, indicating vanadium loss from the BiVO 4 powders at temperatures above 450 AE 20 C. At the same time, no Bi + (m/q ¼ 209, red curve) or BiO À (Àm/q ¼ 225, orange curve) signals are detected. The vanadium loss also occurs in thin lms of BiVO 4 ; XPS measurements reveal that the V : Bi ratio decreases from 0.76 for the lm heated at 450 C to 0.69 for the one heated at 700 C (Fig. S1 †). This is in agreement with a report by Wang et al., 28 in which BiVO 4 lms heated at temperatures higher than 500 C showed a decrease in the V : Bi ratio. Longer measurements of up to 8 hours show negligible change in the rate of vanadium loss (Fig. S2 †), and the rate follows the applied temperature. To further conrm the temperature at which the properties of BiVO 4 thin lms start to change, we performed an in situ UV-Vis measurement of a BiVO 4 lm heated in air. An increase of the absorption at around 500 nm (i.e., a darkening of the lm) was observed at temperatures above $470 C ( Fig. S3 †), which is in good agreement with the onset of vanadium loss found from the MS measurements.
The inuence of vanadium loss on the crystal structure of BiVO 4 thin lms was studied by X-ray diffraction. Fig. 2 shows the grazing incidence X-ray diffractograms of as-prepared and annealed BiVO 4 lms deposited on fused silica ('quartz') substrates. All lms show a pure monoclinic (scheelite-type) BiVO 4 phase (clinobisvanite; space group: I2/b JCPDS card No. 14-0688) without any phase segregation. However, at temperatures of 600 C and above, a re-orientation of the crystal lattice starts to occur and a strong preference for the (010) orientation is observed aer annealing at 700 C. This orientation is known to result in an enhanced photoelectrochemical performance. 29,30 We attribute this re-orientation to changes in the Temperature-time curve during the MS analysis (black) and ion currents for mass-to-charge ratios (m/q) of 51, 67, 83, 209, and 225, which correspond to V + (green), VO + (blue), VO 2 + (purple), Bi + (red) and BiO À (orange), respectively. amount of strain in the lm, as caused by the vanadium loss and/or possible change in particle size. 31 This is supported by the increased preference for the (010) orientation when uniaxially pressing BiVO 4 powder into a pellet (Fig. S4 †). The reorientation of the crystal structure is also accompanied by a shi of the symmetric V-O stretching mode peak ($826 cm À1 ) to a higher wavenumber, as shown in the Raman spectra ( Fig. S5 †). Since the wavenumber is inversely correlated with the interatomic distance, 32 the observed shi can be attributed to a decrease in V-O bond length, which is consistent with the loss of vanadium that occurs under these conditions (Fig. 1).
To understand the inuence of the annealing atmosphere, we performed the same high temperature MS measurement under argon ow. In contrast to the measurement in air (Fig. 1), neither V nor Bi loss was detected ( Fig. S6 †). Based on this, we conclude that the observed VO and VO 2 species in the airannealing experiment were formed by reaction of vanadium with oxygen from the environment and not oxygen from the lattice of BiVO 4 . The following defect-chemical reactions (in Kröger-Vink notation 33 ) are proposed to describe the loss of vanadium: It should be noted that a formal oxidation state of À5 for the vanadium vacancy ðV 00000 V Þ is energetically highly unfavorable so that it is likely to spontaneously ionize to the À4 or À3 state: Since no elemental V loss was observed in the Ar-annealing experiment, the presence of V (g) when annealing in air is believed to be the product of the following disproportionation reaction: Indeed, the change of Gibbs free energy, DG, for reaction (4) is negative (À61.6 kJ mol À1 at 500 C), suggesting that the reaction is spontaneous. The activation energy for the formation of V 000 V in BiVO 4 can be determined from an Arrhenius plot of the log of the ion current vs. the reciprocal temperature (Fig. S7 †). The activation energies for reactions (2) and (3) are found to be 0.18 AE 0.02 eV and 0.23 AE 0.01 eV, respectively.
To investigate the inuence of vanadium vacancies on the photoelectrochemical (PEC) performance, we fabricated thin lm BiVO 4 samples on FTO-coated glass substrates annealed under different temperatures and atmospheres as described in the Experimental section. The temperatures were limited between 450 and 550 C, since scanning electron micrographs showed that agglomeration of the lms starts to occur at temperatures above 550 C (Fig. S8 †). Within this temperature range, the crystal structure and orientation of the lms do not change signicantly (see Fig. 2). Since we are interested in the bulk charge separation properties of the lms, the PEC measurements were performed in a 0.1 M phosphate buffer electrolyte (pH $ 7) with Na 2 SO 3 added as a hole scavenger to remove any surface catalytic limitations. Fig. 3 shows the AM1.5 photocurrents at 1.2 V vs. RHE for BiVO 4 lms annealed at different temperatures up to 550 C in air (black) and argon (purple) atmospheres. The heating atmosphere is found to have a profound inuence on the resulting PEC performance. Annealing in Ar does not result in any change in the photocurrent of BiVO 4 . In contrast, the photocurrent is strongly affected by heat treatment in air. The photocurrent increases with increasing temperature until it reaches a maximum ($1.5 fold increase) at about 500 C. Above this temperature, the photocurrent decreases with increasing temperature.
Since the PEC measurements were performed with Na 2 SO 3 as a hole scavenger, charge transfer limitations at the BiVO 4 / Fig. 2 Grazing incidence X-ray diffraction patterns of BiVO 4 films on quartz annealed in air at temperatures between 450 C (black) and 700 C (red). The labels indicate the diffraction peaks of the monoclinic scheelite crystal structure of BiVO 4 . electrolyte interface do not play a role. Thus, the observed temperature dependence is unlikely to be caused by any changes in the surface chemistry. We further rule out this possibility by comparing the oxidation states of Bi and V in BiVO 4 lms annealed at 450 C and 500 C in air and Ar with XPS (Fig. S9 †). The peak positions of all core levels remain the same, indicating no changes in the oxidation states of Bi and V despite the signicant differences in the PEC performance. The lack of a peak shi is not surprising, considering that even a large amount of vanadium vacancies (10 18 -10 19 cm À3 ) would still be much lower than the $1 at% detection limit of XPS. Furthermore, the photocurrent onset potential shows negligible variation with annealing temperature and atmosphere (see Fig. S10 †), which further conrms that the heat treatments do not signicantly affect the surface chemistry of the material.
Morphological changes with increasing temperature may also be the reason behind the observed PEC behavior. As shown in Fig. S8, † the grain size indeed increases with increasing annealing temperature. However, this is true for BiVO 4 lms annealed in air as well as those annealed in argon; thus, changes in morphology cannot explain the differences in photocurrent. Cross-section SEM images of the lms show that the thickness of the lms shows little change (Fig. S11 †), and AFM measurements (Table S1 †) reveal that the specic surface area of the lms is also not signicantly affected by the heat treatment.
We also rule out changes in optical absorption as a possible cause, since the monotonic increase of the long wavelength absorption with temperature (see Fig. S12 †) is inconsistent with the observed photocurrent maximum at 500 C in Fig. 3.
With changes in surface chemistry and lm morphology being ruled out, we attribute the observed differences in PEC performance to the formation and reduction of defects in BiVO 4 . For samples annealed in air, the increase in grain size reduces the amount of grain boundary scattering, leading to an increase in photocurrent with temperature up to 500 C (Fig. 3). Annealing in argon also results in larger grains, but, in contrast to air annealing, also leads to the formation of oxygen vacancies, V O . The formation of oxygen vacancies is evidenced by the loss of O 2 without the accompanying loss of Bi or V, as shown in Fig. S6. † While the presence of V O can improve the conductivity, too many oxygen vacancies may reduce the space charge width and adversely affect charge separation. Moreover, it has been reported that V O can form deep trap states and act as recombination centers. 34 This would cancel out the positive effect of the larger grain size and would explain the slight decrease in photocurrent for samples annealed in argon (Fig. 3). Anodic electrochemical treatment (i.e., applying positive bias) does not positively affect the photocurrent (Fig. S13 †); thus, it seems not possible to re-ll the oxygen vacancies in our Ar-annealed BiVO 4 . At temperatures above 500 the photocurrent for samples annealed in air rapidly decreases (Fig. 3, black curve). Under these conditions the MS experiments showed a rapid loss of vanadium in the form of VO + , VO 2 + , and V + (Fig. 1). This results in the formation of vanadium vacancies according to reactions (1) and (2). These have been reported to lie $0.3 eV above the valence band edge, 35 which would be deep enough in the bandgap to form trap states and act as recombination centers. In contrast, no vanadium loss is observed when annealing in argon (Fig. S6 †), nor does the photocurrent show any decrease under these conditions (Fig. 3, blue curve). Based on these observations, we conclude that the decrease in photocurrent for BiVO 4 lms annealed in air above 500 C is due to the formation of vanadium vacancies. We note that extending the annealing time of the BiVO 4 lms from 2 to 8 h at 500 C in air also results in a lower PEC performance (Fig. S14 †). We attribute this to the same V loss that causes the formation of V vacancies, which already starts at $450 C (see Fig. 1).
The carrier transport properties of the air-annealed BiVO 4 lms are investigated using time-resolved microwave conductivity (TRMC). Typical TRMC curves (fSm vs. time) are shown in Fig. S15; † the carrier mobility can be obtained from the peak amplitude, and the carrier lifetime can be obtained from the decay of the transient curve. 36 Table 1 summarizes the TRMC results obtained at a laser pulse intensity of 3.5 Â 10 13 photons per pulse per cm 2 . Consistent with our previous report, 13 the carrier mobility (m) of BiVO 4 annealed in air at 450 C is rather low at $0.02 cm 2 V À1 s À1 , while its carrier lifetime (s) of $70 ns is relatively long. The carrier diffusion length (L D ) can be calculated using the following equation: here, D is the diffusion coefficient, k is the Boltzmann constant, T is the temperature, and e is the elementary charge. The carrier diffusion length of BiVO 4 annealed in air at 450 C is calculated to be $60 nm. The properties change signicantly when increasing the annealing temperature. When the BiVO 4 lm is annealed at 500 C, the carrier mobility increases by one order of magnitude to $0.21 cm 2 V À1 s À1 . In line with the discussion above, we attribute this to the observed increase in grain size; a smaller number of grain boundaries would reduce the extent of grain boundary scattering during intergrain charge transport. A similar explanation has been proposed for the behavior of TiO 2 lms aer annealing. 37 At the same time, the carrier lifetime decreases to $20 ns, which we attribute to the formation of V 000 V .
Although not yet dominating at this temperature, vanadium vacancies are likely to be formed as a result of the vanadium loss at temperatures above $450 C (see Fig. 1). As discussed above, these vacancies may act as deep defect states that cause recombination and therefore decrease the carrier lifetime. The net effect of the $10Â increase in mobility and the $3Â decrease in lifetime is an increase of the carrier diffusion length to $105 nm, which explains the observed increase of photocurrent when BiVO 4 is annealed in air up to 500 C (Fig. 3). Above 500 C, the formation of V 000 V dominates, which leads to increased recombination. Table 1 indeed shows decreased carrier mobility and lifetime of the BiVO 4 lm annealed in air at 550 C. The diffusion length decreases to $32 nm, consistent with the decrease in photocurrent observed in Fig. 3.
It should be noted that diffusion of impurities from the FTO substrate (e.g. Sn from the F:SnO 2 or Na from the glass) during high-temperature treatments may also affect the properties of the BiVO 4 . However, the transport properties measured by TRMC, which were obtained using Na-and Sn-free quartz substrates, show the same trends as the photocurrents that were measured using FTO substrates. Moreover, impurity diffusion cannot explain the pronounced inuence of the annealing atmosphere that is observed in Fig. 3 and 4. Therefore, impurity diffusion is unlikely to play an important role in our experiments.
Our ndings show that when BiVO 4 is annealed at high temperature in air, the resulting loss of vanadium is detrimental to the PEC performance. As a strategy to overcome the vanadium loss, we annealed our BiVO 4 lm in the presence of BiVO 4 powder (see Experimental section). By ensuring that the amount of BiVO 4 powder exceeds the amount of BiVO 4 in the lm by several orders of magnitude, we create a VO x -rich atmosphere during the annealing. The high concentration of VO x species is then expected to shi the defect equilibrium of reactions (1) and (2) to the le and thus suppress the vanadium loss from the BiVO 4 lm. Under these conditions, no reorientation of the crystal lattice is observed, even aer annealing at temperatures as high as 700 C (Fig. S16 †). This suggests that the re-orientation of crystallites in the lms is indeed induced by the loss of vanadium from the lms. The exact mechanism of this re-orientation is still unclear and beyond the scope of the current study. Fig. 4 shows that annealing in a VO x -rich atmosphere can partially compensate the loss of photocurrent observed in Fig. 3 when heating the lms at 550 C in air. This partial recovery is consistent with the improved carrier transport properties as measured by TRMC; annealing in VO x -rich atmosphere results in a 3.4-fold increase in the carrier diffusion length (Table 1). Fig. 5 summarizes our ndings on the inuence of different annealing atmospheres on the properties of BiVO 4 thin lms. In all cases, the grain size of BiVO 4 increases with increasing temperature, which results in a reduced number of grain boundaries and higher carrier mobilities. However, annealing in Ar introduces oxygen vacancies, which act as trap states in the lm. This reduces the carrier lifetime and therefore nullies the positive effect of larger grain sizes. The formation of oxygen vacancies can be prevented by annealing the samples in air. Unfortunately, the presence of air results in the loss of vanadium via volatile VO x species, especially above 500 C. This is accompanied by a decrease in the photocurrent, which is presumably caused by the formation of vanadium vacancies that act as recombination centers. This can be avoided by annealing in the presence of BiVO 4 powder in order to create a VO x -rich atmosphere. This prevents the formation of both oxygen vacancies and vanadium vacancies, and results in a BiVO 4 lm with higher carrier mobility and lifetime. This increases the carrier diffusion length and, therefore, the photocurrent of the lms.

Experimental section
Synthesis 100 nm-thick BiVO 4 lms were prepared by spray pyrolysis following a previously reported method. 14   (98%, Alfa Aesar) in 100 mL solvent consisting of acetic acid (98%, Sigma Aldrich) and ethanol (Sigma Aldrich) in a 1 : 9 ratio. The solution was sprayed on the heated substrate in 100 cycles with a spray rate of $1 nm per cycle. The as-deposited lms were then post-annealed in air at 450 C for 2 h. Further heat treatments of the lms are described in the Results and discussion section. In order to create a BiVO 4 -rich atmosphere, 0.6 g of BiVO 4 powder (pure DCCM-137, Dominion Colour Corporation) was placed in the oven during the high temperature treatment of the BiVO 4 lms. This amount was calculated to be $4 orders of magnitude higher than the amount of BiVO 4 in the 100 nm-thick lm.

Materials characterization
Mass spectrometry analysis was performed using a Netzsch thermobalance STA409C coupled with a mass spectrometer, which was operated under a continuous ow of Ar or synthetic air (80 mL min À1 ). The commercially-obtained BiVO 4 powder (see Synthesis section) was placed in an Al 2 O 3 crucible and xed to the sample holder of the thermobalance. X-ray diffraction (XRD) was conducted with a Bruker D8 diffractometer using a Cu Ka radiation (l ¼ 1.54184Å) at 40 kV and 40 mA. This measurement was done in a grazing incidence conguration with a 0.5 incident angle. A Horiba HR800 spectrometer with a HeNe laser was used for Raman spectroscopy. The HeNe laser is a 500 : 1 polarized 20 mW laser with a wavelength of 632.8 nm. Scanning electron microscopy (SEM) images were taken using a LEO GEMINI 1530 with an acceleration voltage of 3 kV. The BiVO 4 lms were coated with a thin carbon layer to prevent charging during the SEM measurements. UV-Vis spectroscopy was performed in an integrating sphere using a Lambda 950 spectrophotometer (Perkin Elmer) in a wavelength range from 320 to 800 nm. High temperature in situ UV-Vis spectroscopy measurements were performed by placing the BiVO 4 lms on a sample holder in a custom quartz tube equipped with a at quartz window, as described previously. 41,42 This tube was placed in an oven and annealed in air up to 700 C with a heating ramp of 7 C min À1 . The transmission spectra were measured with a high sensitivity CCD spectrometer (Maya2000 Pro, Ocean Optics) in a wavelength range from 300 to 850 nm, and a deuterium-halogen lamp (DH-BAL-2000, Ocean Optics) as the light source. The specic surface area was determined with an atomic force microscope (AFM, NT-MDT NovaPX).
Photoelectrochemical characterizations were performed in a three-electrode conguration. The measurement consisted of a Pt counter electrode, an Ag/AgCl reference electrode (XR300, saturated KCl and AgCl solution, Radiometer Analytical) and the BiVO 4 lms as the working electrode. Electrical contact to the lms was provided by a copper wire as well as a tin-plated copper foil (CCK-18-101, Farnell) connected to the exposed FTO substrate. The electrolyte was 0.1 M potassium phosphate (KPi) buffer (pH $ 7) with added 0.5 M sodium sulte (Na 2 SO 3 ) as a hole scavenger. Potentials with respect to the reference electrode (V Ag/AgCl ) were applied by a potentiostat (EG&G PAR 273A) and converted to the reversible hydrogen electrode scale (V RHE ) by using the Nernst equation: V RHE (V) ¼ V Ag/AgCl (V) + 0.0591 Â pH + V Ag/AgCl 0 (7) where V Ag/AgCl 0 is the standard potential of the Ag/AgCl reference electrode (199 mV at 25 C). Photocurrents were measured under back-side (i.e., light arrives from the substrate-side) AM1.5 solar illumination (100 mW cm À2 ) using a solar simulator (WACOM, type WXS-50S-5H, class AAA) with a scan rate of 20 mV s À1 . Time-resolved microwave conductivity (TRMC) measurements were performed using a setup that has been described in detail elsewhere. 14,37 In short, the BiVO 4 lms deposited on quartz were placed in a microwave cavity cell. A frequencytripled Q-switched Nd:YAG laser at a wavelength of 355 nm was used as the excitation source with a 7 ns pulse. Microwaves in the X-band region (here: 8.4-8.7 GHz) were generated by a voltage controlled oscillator (SiversIMA VO3262X). For the carrier transport properties calculation, the dielectric constant of BiVO 4 was taken as 68. 43

Conclusion
In summary, we elucidated the formation of defects in BiVO 4 upon high temperature anneal treatments and the inuence of these treatments on the carrier dynamics and photoactivity of the material. We showed that annealing BiVO 4 thin lms in air up to 500 C increases the grain size and enhances the carrier mobility and diffusion length. This results in an increase of the AM1.5 photocurrent by a factor of $1.5. At temperatures above 500 C, the carrier mobility, lifetime, and photocurrent decrease due to the loss of vanadium. This presumably leads to the formation of vanadium vacancies in the lm, which act as deep defect states that cause recombination. The activation energy for the formation of V 000 V in BiVO 4 was found to be $0.2 eV. Finally, we showed that it is possible to suppress the loss of vanadium and the concomitant decrease in photocurrent at temperatures above 500 C by annealing the photoanode lms in the presence of excess BiVO 4 powder. Further optimization of this vanadium-rich atmosphere may result in complete recovery or even increase of the photocurrent in BiVO 4 . Overall, we believe this strategy offers a general approach to prevent unfavorable changes of cation stoichiometry during high temperature treatment of complex metal oxides.

Conflicts of interest
There are no conicts to declare.