The synergistic effect of cation mixing in mesoporous BixFe1−xVO4 heterojunction photoanodes for solar water splitting

Mixed metal vanadates are an interesting class of materials due to their favorable bandgap for visible light absorption and their catalytic activity.


Introduction
In the eld of photoelectrochemical (PEC) water splitting, bismuth vanadate (BiVO 4 ) has emerged as one of the highest performing metal oxide photoanodes. With recent progress and success of achieving 90% of its theoretical maximum photocurrent, it is clear that further improvement in the solar-tohydrogen (STH) efficiency of bismuth vanadate (BiVO 4 ) is limited by its bandgap (2.4-2.5 eV). 1,2 Numerous attempts have been reported on either identifying novel narrow bandgap photoanode materials or decreasing the bandgap of BiVO 4 . Some emerging materials, such as CuWO 4 (2.1 eV), 3 SnWO 4 (2.1 eV), 4,5 ZnFe 2 O 4 (2.0 eV), 6,7 Cu x V y O z (1.8-2.0 eV), 8,9 and FeVO 4 (2.06 eV), 10,11 have been reported to be promising photoanode candidates. However, these novel complex metal oxides still show much lower performance than expected, and further developments on a relatively longer timescale are necessary. Sulfurization or nitridation has been demonstrated to effectively reduce the bandgap of BiVO 4 , but its PEC stability may be a concern. 2,12 As an alternative, the bandgap of BiVO 4 can also be modied by incorporating another metal cation to form a metal vanadate mixture. The combination of a new metal cation-if chosen properly-could possibly raise the valence band of BiVO 4 without much effect on the conduction band or form a favorable heterojunction with extended light absorption. The effective bandgap, as a result, will be reduced.
Studies on mixed metal vanadates have been reported from a decade ago for photocatalytic applications (e.g., BiCu 2 VO 6 and BiZn 2 VO 6 ). [13][14][15][16][17][18] Combining two metal vanadates (Cu and Bi) into a solid solution to lower the bandgap of BiVO 4 has been experimentally demonstrated. 13,15 The absorption of BiCu 2 VO 6 extends to 600 nm, which is 80 nm longer than that of BiVO 4 . It was proposed that the bandgap reduction is due to band modulation through the incorporation of two metal cations. The additional advantage reported is that BiCu 2 VO 6 and BiZn 2 VO 6 possess higher photocatalytic activity than pristine BiVO 4 . 15,17 Similarly, a Bi 0.75 Fe 0.25 VO 4 mixture demonstrated higher photocatalytic efficiency in the degradation of methylene blue than FeVO 4 or BiVO 4 . 19 However, the detailed mechanism on why mixed vanadates can offer higher catalytic activity or improved photoelectrochemical performance is not well understood. In addition, a high performance mixed vanadate photoanode has not been reported in the literature.
In our previous report, a series of Bi x Fe 1Àx VO 4 (0 # x # 1) lms were fabricated by spray pyrolysis. 20 Initial results indicated the possibility of maintaining good mobility of BiVO 4 and extending visible light absorption by mixing FeVO 4 with BiVO 4 . The present study adopts electrospinning to deposit mixed vanadate lms and higher annealing temperature to improve the crystallinity. The combination of Fe and Bi vanadate shows higher catalytic efficiency and increased visible light absorption; in particular, the Fe 0.5 Bi 0.5 VO 4 composition demonstrates the highest PEC performance. Microscopic observations show that the Fe 0.5 Bi 0.5 VO 4 consists of adjacent FeVO 4 and BiVO 4 nanoparticles, which results in a favorable heterojunction to facilitate charge separation. Our study presents the synergistic effect of a Bi/Fe vanadate heterojunction and gives insights to further improve the performance of vanadate-based photoanodes.

Physical characterization and chemical compositions
We prepared a series of Bi x Fe 1Àx VO 4 (0 # x # 1) lms by electrospinning (Scheme S1 †) (the expression Bi x Fe 1Àx VO 4 here is determined by the composition of precursors and does not necessarily indicate the formation of a solid solution). For brevity, the different lms are denoted by the percentage amount of Bi in each lm, i.e., Bi0 corresponds to a FeVO 4 lm, Bi50 corresponds to Bi 0.5 Fe 0.5 VO 4 , and Bi100 corresponds to BiVO 4 . Prior to other measurements, EDX spectroscopy was used to conrm that the Fe and Bi ratio follows the initial amount added to the precursor (Fig. 1a). EDX mapping conrms the uniform distribution of Fe and Bi throughout the lm (Fig. S1 †). Fig. 1b shows the X-ray diffraction (XRD) patterns of the various Bi x Fe 1Àx VO 4 lms. At each extreme of the composition (i.e., Bi0 and Bi100), a pure phase of triclinic FeVO 4 and monoclinic BiVO 4 was obtained, respectively. For the mixed vanadate, both FeVO 4 and BiVO 4 peaks were detected from Bi5 to Bi50. From Bi65 and above, the FeVO 4 phase can no longer be detected by XRD, but only the BiVO 4 peak intensity changes with increasing Bi content (Fig. S2 †). This may be caused by the relatively small amount of FeVO 4 phase in the lms. Further investigations on the crystal structure by transmission electron microscopy are discussed later in the paper. Compared with the peak intensity of BiVO 4 , the FeVO 4 peak intensity is relatively lower. This is due to the fact that the lms were annealed at 550 C while annealing at 650 C is typically required to obtain good crystallinity of FeVO 4 . 10,20 This lower temperature was chosen as a compromise because V loss occurs at temperatures above 450 C. 21 Fig. S3 † indeed shows that all of our lms are slightly V-poor. The effect of V loss has been discussed elsewhere and is not the focus of this report. 21 The chemical composition and oxidation states in the lms were analyzed by XPS. All the spectra were calibrated using the adventitious carbon peak at 284.8 eV. Fe, Bi, V and O core level XPS spectra of FeVO 4 , Bi50 and BiVO 4 lms are shown in Fig. S4. † Peaks at 710-714 eV can be observed in FeVO 4 and Bi50 lms, which can be assigned to Fe 2p 3/2 multiplets with an oxidation state of +3. 22,23 The peak at 716 eV, which is present for all lms, corresponds to the Sn 3p 3/2 core level, which is caused by the exposed FTO. 24 XPS peaks at 158.7 eV and 164.1 eV conrm the existence of Bi 3+ in Bi50 and BiVO 4 . 25,26 The vanadium is present as V 5+ in all lms, as evident from the V 2p 3/2 peak at $516.9 eV. 27,28 The binding energy of O 1s at 530 eV matches with the oxygen lattice peak in the oxide. 27 The peak positions do not shi for all lms, suggesting that the chemical environment of the mixture remains unchanged.
A gradual color change from orange (FeVO 4 ) to bright yellow (BiVO 4 ) can be observed in the visual appearance of the lms (Fig. S5 †), similar to our previous report. 20 The porous appearance of the lms is due to the addition of PVP to the electrospinning precursor. The absorption spectra of the mixed vanadate lms obtained from UV-vis spectroscopy are shown in Fig. 2a and b. Other than the pure BiVO 4 lm, the absorption edge of the mixed vanadate lms extends beyond 520 nm (absorption edge of BiVO 4 ), validating the effect of Fe incorporation on increasing the visible light absorption range. For Bi10 to Bi50 lms, a higher Bi ratio increases the absorption at around 480 nm, which is a characteristic absorption shoulder of BiVO 4 (Fig. 2a). Beyond this Bi concentration (i.e., 50% and above), visible light absorption at the 520-600 nm range decreases with higher Bi content (Fig. 2b), possibly due to the difference of absorption coefficients of BiVO 4 and FeVO 4 . 20,29 Although the lms were detected to be mixtures of BiVO 4 and FeVO 4 , the 'apparent bandgap' of the mixture, instead of the actual bandgap, can still be determined. Thus, the direct and indirect bandgaps of all the lms can be calculated using Tauc analysis; the results are shown in Fig. 2c. The direct bandgap only shows minor differences for all compositions (2.45-2.5 eV), but the indirect bandgap increases with increasing Bi content.
Our results clearly show that a higher Fe content leads to a larger visible absorption range. However, the appearance of the absorption shoulder at 480 nm and the varying absorption spectra at $520-600 nm suggest that an optimal Bi/Fe ratio may exist in order to achieve the maximum total absorption. By integrating all the absorption spectra with the AM1.5 photons, the theoretical absorbed photocurrent density (J abs ) can be readily calculated for each composition and the results are shown in Fig. S6. † All the lms have a J abs of $3-4 mA cm À2 . With these absorption changes brought about by the Bi/Fe ratio, a maximum is observed for the Bi10 lm (4 mA cm À2 ).

Improved charge carrier efficiencies
The AM1.5 photocurrent density of all lms was measured in 0.1 M phosphate buffer (KPi, pH ¼ 7). Most lms show higher photocurrent under back-side illumination as compared to front illumination (Fig. S7 †). 30 min deposition time (electrospinning) is found to be optimal to ensure sufficient lm thickness and avoid delamination of lms during annealing. As the PEC performance is similar for 30 min and 45 min  Table S1. † Among all the compositions, the Bi50 lm shows the highest photocurrent (0.47 mA cm À2 at 1.23 V vs. RHE). The peak photocurrent trend at a Bi 50% ratio is even more obvious at higher potential, while other Bi rich or Fe rich composition lms only generate moderate photocurrents. From Bi0 to Bi50, the photocurrent increases almost linearly with increasing Bi content; for Bi50 to Bi100, the photocurrent instead decreases. To exclude any surface limitation (due to slow charge transfer or surface recombination), the photocurrent was also measured in KPi with added H 2 O 2 as a hole scavenger. The hole scavenger photocurrent trend is similar to that in KPi (Fig. 3b), with a Bi50 lm showing the highest photocurrent of up to 1.82 mA cm À2 at 1.23 V vs. RHE.
The charge transfer and separation efficiencies can be calculated from the photocurrent density with and without a hole scavenger (denition and calculation steps are described in the ESI †). The transfer efficiency (h trans ) of all compositions is shown in Fig. S10 † and the overall trend at 1.23 V vs. RHE is shown in Fig. 3c. It is interesting to note that the charge transfer efficiencies of all Bi/Fe mixed vanadate lms are more or less constant and higher than those of pure FeVO 4 and BiVO 4 over the whole potential range. This is in good agreement with the observation that mixed vanadates have higher catalytic activity than a single metal vanadate. 15,17,19 In addition, a closer look at the trend shows that the h trans of the lms with higher Fe content (Bi5 to Bi50) is slightly higher than those of the Bi rich lms (Bi65 to Bi100). We attribute this to the higher catalytic activity of Fe sites for oxygen evolution as compared to Bi. Indeed, the dark water oxidation current of FeVO 4 and Bi50 lms is higher than that of BiVO 4 (Fig. S11 †).
Since the J abs values of all lms are relatively close to each other ( Fig. S6 †), the charge carrier separation efficiency (h sep ) ( Fig. 3c and S12 †) follows a similar trend to the hole scavenger photocurrent. The incorporation of Bi/Fe at around a 1 : 1 ratio (i.e., Bi50) results in an optimal charge separation. The h sep of Bi50 at 1.23 V vs. RHE is 10-fold and 1.7-fold higher as compared to that of FeVO 4 and BiVO 4 , respectively. It is therefore clear that the photocurrent improvement in the mixed Bi-Fe-vanadate lms can be mainly attributed to the more efficient charge carrier separation.
The much higher h sep can be explained by the formation of FeVO 4 and BiVO 4 heterojunctions with favorable band alignment to facilitate efficient charge carrier separation. 23 As the two vanadates are homogeneously distributed, indeed the 1 : 1 ratio is expected to be optimal to ensure enhanced charge separation at all parts of the lm. A BiVO 4 /FeVO 4 heterojunction in a 1 : 1 ratio has also been reported to obtain the highest photocatalytic activity for pollutant degradation, 30 which agrees very well with our results and interpretation.
Electrical impedance measurements were performed to shed light on the charge transfer processes. The Nyquist plots for all compositions derived from EIS measurements under illumination are shown in Fig. S13. † The different radii of semicircles can reect the overall resistance. It is clear that Bi50 has the smallest semicircle; either increasing or decreasing the Fe content only results in a larger radius of the semicircle. An equivalent circuit is shown in Fig. S14, † which is typically used for the reported heterojunction structure and porous electrodes (the electrospun lms show some extent of porosity as evidenced by SEM images). Here, R s is the serial resistance, C sc is the space charge layer capacitance, R ct is the charge transfer resistance and CPE is the constant phase element. [31][32][33] From FeVO 4 to Bi50, R ct reduces with increasing Bi ratio; however R ct increases again from Bi50 to BiVO 4 . Such a trend similar to the changes of the semicircle radius in the Nyquist plot implies that Bi50 has the lowest charge transfer resistance, which can be properly explained with the presence of a BiVO 4 /FeVO 4 heterojunction.
The heterojunction explanation is also supported by the trend in the onset potential as a function of Bi ratio. The addition of BiVO 4 to FeVO 4 leads to a dramatic onset potential drop from 0.94 V vs. RHE (Bi0, FeVO 4 ) to $0.4 V vs. RHE (Bi80-Bi95) (Fig. 3d). The cathodic onset potential shi is a typical effect of the heterojunction, which has been reported in TiO 2 ,    show no apparent trend of the electrochemical active surface area of all lms (Fig. S16 †). Cross-sectional view images also conrm similar particle sizes across the depth prole ($150-170 nm) for different compositions (Fig. S17 †).
To further conrm the formation of a heterojunction, TEM and high resolution TEM (HRTEM) images of the Bi50 sample were taken as shown in Fig. 4. A clear phase contrast can be observed between the bright and dark regions (Fig. 4c and d), where the dark regions are surrounded by the bright regions. Clear lattice fringes can be observed in the bright regions, with d-spacings of 0.85 nm and 0.34 nm corresponding to the (001) and (021) planes of FeVO 4 , respectively ( Fig. 4a and b). The dark regions, however, do not show any clear fringes. To conrm the composition in the dark regions, Energy Dispersive X-ray Analysis (EDAX) was carried out at the sample locations shown in Fig. 4c and d, and the elemental mapping is shown in Fig. S18. † The dark regions are veried to be BiVO 4 particles while FeVO 4 particles (bright regions) are more dispersed. From the mapping images, the microscopic structure of BiVO 4 particles intimately surrounded by FeVO 4 particles can be proposed. This structure also explains why only the FeVO 4 phase can be clearly seen in the HRTEM images (bright regions), while the dark regions do not show lattice fringes due to the convolution between the BiVO 4 and FeVO 4 structure. This is further supported by the higher surface Fe/Bi ratio of the Bi50 lm (2.2 as revealed from XPS, see Table S2 †) as compared to the bulk Fe/Bi ratio of 1 (see Fig. 1a). Such an intricate structure shows clear evidence for the heterojunction, which enhances the charge carrier separation efficiency in the mixed vanadate lms, particularly in the case of Bi50. A core-shell BiVO 4 /FeVO 4 structure has indeed been reported to enhance charge separation. 23 Mott-Schottky measurements were carried out to obtain the at-band position of our lms, which can be further used to estimate the Fermi level position. The at-band potential of BiVO 4 and FeVO 4 is extrapolated to be $0.13 V and $0.72 V vs. RHE, respectively (Fig. S19 †), which agree very well with reported values. 20,39 For a conclusive conrmation of band alignments, UPS was conducted and the spectra are presented in Fig. 5. The work function can be determined using the secondary electron cut-off. The position of the valence band maximum (E V ) with respect to the Fermi level (E F ) can be determined from the low kinetic energy cut-off of the spectrum. 39 As shown in Fig. 5, the work function for BiVO 4 and FeVO 4 can be determined to be 4.5 eV and 4.9 eV, respectively, while the E V with respect to the E F is determined to be 2.36 eV and 2.06 eV, respectively. Although the measured work function of BiVO 4 and FeVO 4 is slightly higher than the reported values (BiVO 4 : 4.56 eV 39    We also note that the optimal charge carrier separation efficiency in the Bi50 lm can be caused by the improved charge carrier transport property (i.e., carrier mobility). This is, however, unlikely to be the case. While we have no direct evidence of the carrier mobility in our electrospun lms-timeresolved microwave conductivity (TRMC) measurements of our lms are not possible since electrospinning the lms on quartz (which is required for TRMC) results in mechanically unstable lms-our previous study on sprayed mixed Bi-Fe-vanadates have shown that the carrier mobility continuously decreases with decreasing Bi content. 20 Inuence of the vanadate mixture on visible light photoactivity Finally, we investigated the photoactivity of the lms under different illumination wavelengths. The incident photon-tocurrent conversion efficiency (IPCE) curves of all lms are shown in Fig. S20. † Although the absorption edge of FeVO 4 is $600 nm, which is considerably longer than that of BiVO 4 (520 nm), the IPCE of FeVO 4 at wavelengths above 525 nm is negligibly small ($0) (Fig. 6a). The mismatch of the optical absorption onset and photoelectrochemical onset wavelength of FeVO 4 can be attributed to the low carrier mobility and indirect absorption energy loss in a long wavelength range. 20,40 Similar to the observation in UV-vis spectra, the IPCE at all wavelengths was enhanced for the mixed vanadates as compared with both FeVO 4 and BiVO 4 (Fig. S20 †), with Bi50 showing the highest IPCE (Fig. 6a). Further IPCE measurements in H 2 O 2 were carried out to better illustrate the onset wavelength for photoactivity (Fig. 6b). The Bi50 lm clearly exhibits an extended PEC response to its absorption edge at 565 nm (corresponds to its indirect bandgap of 2.2 eV in Fig. 2c), while FeVO 4 and BiVO 4 displayed the same onset wavelength as in KPi ($520 nm, Fig. 6b). More direct evidence of the photo response of Bi50 at 550 nm is shown in Fig. S21. † It is clear that Bi50 generated measurable photocurrent under monochromatic illumination at 550 nm.
Since FeVO 4 and BiVO 4 show zero IPCE signal at wavelengths higher than 525 nm, the extended IPCE response for Bi50 can be attributed to the absorption in the FeVO 4 phase assisted by the presence of the BiVO 4 /FeVO 4 heterojunction to improve charge separation as mentioned above. The equivalence of the optical onset and PEC onset wavelength for the Bi50 lm therefore serves as evidence for a successful extension of the photoactivity in the mixed BiVO 4 /FeVO 4 lms.
Finally, Bi50 was measured in KPi at 1.5 V vs. RHE under illumination to investigate its stability. The photocurrent of the lm continuously decreases in the 4000 s chronoamperometry measurement (Fig. S22a †). Therefore, a thin layer of CoPi was deposited by a well-established method for surface protection and the thickness of CoPi is estimated to be $2 nm according to a report. 41 Aer CoPi deposition, other than the initial transient, the photocurrent stability is improved. The lm is able to retain a photocurrent of $0.6 mA cm À2 up to 4000 s (Fig. S22a †), implying that a thin layer CoPi can help protect the surface and improve the stability. This CoPi deposited Bi50 lm was later used for oxygen evolution measurements due to its improved stability. The amount of oxygen generated was quantitatively recorded by gas chromatography. The faradaic efficiency for oxygen evolution was calculated to be $88% (Fig. S23 †).

Conclusions
A series of Bi x Fe 1Àx VO 4 mixture mesoporous lms was successfully fabricated by electrospinning. Instead of forming a solid solution, a mixed phase of BiVO 4 and FeVO 4 was detected from the X-ray diffraction patterns, and the indirect bandgap decreases with decreasing Bi/Fe atomic ratio. The mixed phase with an atomic ratio of Bi/Fe ¼ 1, i.e., Bi50, was found to have the optimum charge carrier transfer and separation efficiencies as compared with BiVO 4 and FeVO 4 . Improved charge transfer efficiency was achieved due to a higher catalytic activity of the Fe and Bi cation mixture, while the enhanced charge separation was attributed to the homogeneously distributed BiVO 4 /FeVO 4 heterojunction as observed from TEM. In addition, IPCE measurements revealed that the Bi50 composition demonstrates photoactivity up to $560 nm, which is larger than that of pure BiVO 4 and consistent with the indirect bandgap of $2.2 eV. Our work illustrates a simple yet successful strategy of forming a mixed phase heterojunction to expand light utilization as well as charge carrier transfer and separation efficiencies. Further improvement of the mixed phase lms may be obtained through strategies well-reported in the literature, such as doping and nanostructuring.
The resulting solution was then used for electrospinning deposition. 4 mL prepared solution was sucked into a 16 cm diameter syringe. The syringe and the electrospinning needle (20G, 13 mm length) were connected with chemical resistant tubing (Tygon 2375, 1/16 Â 3/16 00 ). The setup is shown in Scheme S1. † Fluorine doped tin oxide (FTO) substrates (14 U cm À2 , Nippon Sheet Glass) were subjected to a standard cleaning procedure (sonicated 20 min each in soap water, deionized water, acetone, deionized water and ethanol). Aer drying with nitrogen ow, the FTO substrates were placed at the center of the electrospinning stage (collection plate) partially wrapped with aluminum foil to allow contact between the FTO conductive side and the stage. During the electrospinning process, a voltage of 25 kV was applied between the needle and the collection plate at a distance of 12 cm. The deposition rate was controlled at 1.5 mL h À1 using the syringe pump. The deposition was carried out at room temperature at different time intervals, i.e., 15, 30 and 45 min. All the lms were nally annealed at 550 C in air for 3 h. The samples were denoted by the Bi content, for example Bi 0.3 -Fe 0.7 VO 4 can be denoted as Bi30.

Characterization
The morphology of the lms was observed by eld emission scanning electron microscopy (FESEM) using a JEOL-7600F microscope. Energy dispersive X-ray spectroscopy (EDX) analysis was done with an Oxford INCA detector. For each composition, measurements were done in at least ve different locations on the lms to obtain an average value. The crystal structure of the lms was measured by X-ray diffraction (XRD) using a Shimadzu thin lm X-ray diffractometer (Cu Ka source). Optical absorption spectroscopy was measured on a UV-vis-NIR spectrophotometer (PerkinElmer, Lambda 950). For structural characterization, Transmission Electron Microscopy (TEM) analysis was performed using a JEOL 2100F transmission electron microscope. Powders carefully scratched from the lm deposited on an FTO substrate were used for the TEM analysis. The powders were loaded onto the copper grid aer sonicating in ethanol for 20 min to allow better dispersion. Energy Dispersive X-ray Analysis (EDAX) was performed for elemental mapping.
Photoelectrochemical (PEC) measurements were carried out on a CHI 660D workstation (CH Instruments Inc.). The mixed vanadate photoanodes were irradiated using a 150 W xenon solar simulator (67005, Newport Corp.) with a solar lter (KG 3) to achieve a measured light intensity of 1 sun (100 mW cm À2 ). PEC measurements were performed in a PEC cell in a 3-electrode conguration, where the vanadate lms were used as the anode, a Pt coil (3.6 cm 2 ) as the counter electrode, and an Ag/ AgCl electrode (KCl 3 M solution) as the reference electrode. The PEC performance of the photoanodes were measured in a potassium phosphate buffer electrolyte (0.1 M KPi, pH ¼ 7). The illuminated area on the photoanodes is xed at 0.125 cm 2 . Unless stated, all the PEC and IPCE measurements were performed under back-side illumination. Potential conversion to the reversible hydrogen electrode (RHE) scale was calculated using the equations below.
where E 0 Ag/AgCl ¼ 0.21 V vs. NHE at 298 K. For PEC measurements with a hole scavenger, 0.5 M hydrogen peroxide (H 2 O 2 , 30% w/w in H 2 O, Sigma Aldrich) was added to the KPi buffer solution. Incident photon-to-current conversion efficiency (IPCE) measurements were carried out using a combination of a xenon lamp (MAX-302, Asahi Spectra Co. Ltd.) and a monochromator (CMS-100, Asahi Spectra Co. Ltd.) from 350 to 620 nm wavelength. The IPCE values can be calculated using eqn (2).

¼
1240ðV nmÞ Â photocurrent densityðmA cm À2 Þ incident light power densityðmW cm À2 Þ Â wavelengthðnmÞ (2) Electrochemical active surface area (ECSA) measurements were performed using the same conguration as the PEC measurements, but without any irradiation (i.e., in the dark). The potential was swept between 0 V to 0.6 V vs. Ag/AgCl at different scan rates (20,50,100,150,200, and 300 mV s À1 ). The lms were masked with tape to leave a xed area (0.283 cm 2 ) for electrical contact. More details of the ECSA measurements can be found in other reports. 42,43 Electrical impedance spectroscopy (EIS) and Mott-Schottky measurements were conducted on an Autolab potentiostat (Metrohm-Autolab, AUT 83285) in a three electrode conguration using 0.1 M KPi solution. All the lms were masked with tape to leave the same contact area (0.283 cm 2 ).
X-ray photoelectron spectroscopy (XPS) and ultraviolet photoelectron spectroscopy (UPS) measurements were conducted on the same VG ESCALAB 2201-XL instrument (operation condition <5 Â 10 À10 mbar) equipped with a monochromatic Al Ka (1486.7 eV) X-ray source and a He I discharge lamp (21.2 eV) UV source.
A stability test was carried out in KPi at an applied potential of 1.5 V vs. RHE for both untreated and CoPi deposited Bi50 lms. CoPi was deposited by electrodeposition at 1.5 V vs. RHE for 60 s using 0.1 M KPi electrolyte containing 0.5 mM Co(NO 3 ) 2 . A two-compartment electrochemical cell divided by an ion exchange membrane was used for the gas evolution test. The CoPi-Bi50 lm was subject to 1.5 V vs. RHE in 0.1 M KPi under AM1.5 illumination. The amount of O 2 evolved was detected using a gas chromatograph (GC) (Agilent 7890A).

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