Introduction

The excessive use of fossil fuels to meet global demands have increased carbon dioxide (CO2) emissions that are the primary cause of global warming1. Hydrogen (H2) is a clean energy carrier already used at an industrial scale, but that is almost wholly produced from fossil fuels today2. Solar driven (photocatalytic) water splitting has emerged as a particularly promising strategy that mimics natural photosynthesis by using semiconductor materials to absorb sunlight and split water into oxygen and renewable hydrogen.

Among n-type metal oxide semiconductors for photoelectrochemical (PEC) water oxidation–including titania (TiO2), hematite (α-Fe2O3), and tungsten trioxide (WO3)–bismuth vanadate (BiVO4) stands out due to its visible light active bandgap (~2.4 eV) and its resistance to photocorrosion in neutral and basic electrolytes3. Under AM1.5 G sunlight, BiVO4 has a theoretical maximum solar-to-hydrogen (STH) efficiency of 9.2% and photocurrent density of 7.5 mA cm−2.

Although BiVO4 has relatively low electron (~10 nm)4 and hole (~100 nm)5 diffusion lengths that can lead to high electron/hole recombination losses, these can be countered by materials design strategies, such as the formation of type-II staggered heterojunctions6. Spectroscopic studies show that bare BiVO4 requires 20–100 μs for electron extraction to occur, whereas the addition of a WO3 layer in a heterojunction enables sub-microsecond transfer into the WO3 conduction band7, improving overall PEC performance.

To improve the extraction of holes from BiVO4, catalytic materials that drive the oxygen evolution reaction (OER) such as cobalt oxides (CoOx) and cobalt phosphate (CoPi) have emerged as particularly promising non-precious metal candidates. In 2015, Ma et al. reported that surface modification of BiVO4 with CoPi resulted in a cathodic shift of the photocurrent onset potential by approximately 200–300 mV8. The performance enhancement was attributed to suppression of back electron/hole recombination, where CoPi accumulates long-lived holes at the BiVO4 surface8,9,10. During OER, redox cycling of cobalt between Co2+ to Co3+ enables a self-repair mechanism, improving stability during long-term operation11,12. In 2023, Varma et al. reported the fabrication of Fe-doped CoOx co-catalyst layers on BiVO4 photoanodes via photo-assisted electrodeposition (PED). Their optimised photoanode achieved a photocurrent density of 4.6 mA cm−2 at 1.23 VRHE13. Similarly, in 2022, Fang et al. used electrodeposition (ED) to deposit CoPi on BiVO4, resulting in a photocurrent density of 5.87 mA cm−2 and a marked cathodic shift in the onset potential from 0.45 VRHE to 0.20 VRHE 14.

Deposition via PED requires illumination and applied bias, which may be limited in scalability, and additionally requires that the electrolyte will not attenuate the illumination. Hence, we propose a more scalable alternative, where aerosol assisted chemical vapour deposition (AACVD) is used to deposit CoOx that is subsequently converted into CoPi through dark electrochemical treatment (ET). Moreover, by growing the BiVO4 layers using the AACVD method as well, this versatile technique may be used to produce all layers in the PEC system. In this study, we optimise the surface modified CoPi, and to benchmark this method, CoPi is fabricated using the conventional PED approach on AACVD-fabricated substrates.

Materials and methods

Experimental methods are briefly outlined here with further details presented in the Supplementary Information.

Chemical vapour deposition of photoanodes and cobalt oxide

The AACVD system utilised in this study is featured in several publications15,16,17. Briefly, 1.3 × 2.5 cm2 area and 3 mm thick FTO-coated glass substrates (TEC 15, Pilkington NSG) were first cleaned by successive sonication for 10 min each in deionised water with detergent, deionised water (18.2 MΩ cm, Sartorius), acetone (Technical, VWR Chemicals), and finally methanol (Reag. Ph.Eur., ACS, VWR Chemicals)16. In a typical BiVO4 deposition, vanadium (III) acetylacetonate (0.2 mmol, 0.0700 g, 97%, Sigma-Aldrich) and triphenyl bismuth (0.2 mmol, 0.0881 g, 98%, Santa Cruz Biotechnology Inc.) were dissolved in 40 mL of a 3:1 methanol:acetone mixture, aerosolised using an ultrasonic humidifier (2 MHz, Liquifog, Johnson Matthey), and carried using compressed air at a flow rate of 1.5 L min−1 over FTO substrates heated to 400 °C. The as-deposited BiVO4 films were finally annealed at 500 °C in air for 2 h. WO3/BiVO4 heterojunction fabrication was conducted by depositing BiVO4 onto pre-synthesised nanoneedle (NN) WO3 films, while the flat heterojunction system was formed by depositing BiVO4 on flat WO3 films, as reported elsewhere15. For a typical CoOx deposition on FTO, cobalt (II) acetylacetonate (Co(acac)2) (0.0257 g, 0.1 mmol, 5 mM, 98%, Fluorochem) was dissolved in 20 mL of ethanol (absolute >= 99%, Fisher Scientific) and delivered by compressed air carrier gas flow rate of 1.5 L min-1 over substrates heated to 400 °C.

Electrochemical treatment (ET) of CoOx to form CoPi

ET of CoOx to form CoPi was performed by immersing CoOx films into pH 7, 0.1 M aqueous potassium phosphate (KPi) (K2HPO4/KH2PO4, 99%, Thermo Scientific) buffer under an applied potential of 1.5 VRHE in the dark. An AgCl/Ag (saturated KCl) electrode served as the reference electrode and a platinum mesh as the counter electrode. Chronoamperometry lasting 10 minutes (resulting in a total charge transfer of ~0.8 mC cm-2) was applied (Supplementary Information Fig. S1) converting cobalt oxide into cobalt phosphate through cobalt ion dissolution–redeposition from the electrolyte 11.

Photoelectrochemical deposition (PED) of CoPi

PED of CoPi18 was carried out using white light illumination from a 75 W xenon lamp (Hamamatsu) in a quartz cuvette (5 cm × 3 cm × 13 cm). Cobalt nitrate (0.0146 g, 0.5 mM, 98%, Fluorochem) was dissolved in 100 mL of either pH 7 0.1 M KPi buffer electrolyte for CoPi deposition or borate buffer (1 M H3BO3, >=99.5%, Sigma Life Science, adjusted to pH 9) electrolyte for CoOx deposition for comparison. Electrodeposition was performed under three cycles of constant current conditions (20 μA cm−2). Between each cycle, the system was held at open circuit potential (OCP) of 0.61 VRHE for 260 s.

Physical characterisation methods

Thermogravimetric analysis (TGA) was performed using a Mettler Toledo TGA/DSC 1LF/UMX system, coupled with a Hiden Analytical HPR-20 QIC Evol mass spectrometer. SEM images & energy-dispersive X-ray spectroscopy (EDX) were acquired using a Zeiss Sigma microscope operating at 5 kV, equipped with an InLens detector. X-ray diffraction (XRD) measurements were carried out using a Bruker D2 Phaser diffractometer equipped with parallel beam optics and a Lynx-Eye detector. Raman spectra were collected using a DXR3® Raman Microscope (Thermo Fisher Scientific) with a laser wavelength of 532 nm. UV-Visible (UV–vis) transmittance and reflectance spectra were measured using a UV-2600 Shimadzu spectrophotometer equipped with an integrating sphere. X-ray photoemission spectroscopy (XPS) was measured with a Thermo Fisher K-Alpha+ with a monochromated Al K\(\alpha\) X-ray source. X-ray fluorescence (XRF) was measured using a Malvern Panalytical Epsilon 3 spectrometer.

Photoelectrochemical water splitting measurements

PEC linear sweep voltammetry and incident photon-to-electron conversion efficiency measurements were performed with an Autolab PGSTAT101 potentiostat and Nova software in a standard three-electrode configuration. An AgCl/Ag (saturated KCl) reference electrode was used, platinum mesh served as the counter electrode, and a phosphate-free 1 M, pH 9 borate aqueous buffer solution was the electrolyte. This buffer was selected to provide a distinct electrochemical environment from the phosphate-based electrolyte used during electrochemical treatment, which converts the surface of the (AACVD)-fabricated CoOx into CoPi. Illumination was provided by a 75 W xenon lamp attenuated with a KG3 filter. The intensity of this white light source was measured by UV–vis and NIR spectrophotometers to be 44% of the light intensity of the AM 1.5 G solar spectrum, and so measured photocurrents are plotted scaled to 1-sun illumination, given that photocurrents of BiVO4 have been shown to scale linearly at this regime 9.

The incident photon-to-current efficiency (IPCE) was determined by measuring the photocurrent response at discrete wavelengths ranging from 275 to 550 nm in 25 nm intervals using a monochromator (OBB-2001, Photon Technology International). During each measurement, a constant potential of 1.23 VRHE was applied. The incident light intensity at each wavelength was calibrated using a Thorlabs optical power metre (PM100D) equipped with an S120VC photodiode sensor.

Further photoelectrochemical impedance spectroscopy (PEIS), stability measurements, and charge separation and injection efficiency measurements were conducted under one sun illumination with a Class A Sun2000 Solar Simulator (Abet Technologies), equipped with a 550 W Xe lamp and AM 1.5 G filter. The light spectrum and intensity were characterised by StellarNet Black Comet UV/vis (280–900 nm) spectrophotometers, interfaced with SpectraWiz software. The measured light spectrum was compared with the NREL AM 1.5 G reference spectrum19 and the working distance was adjusted such that the cumulative light intensity over the measured region was equivalent to AM 1.5 G (one sun). A representative spectrum of the Sun2000 solar simulator is provided in Fig. S2. These photoelectrochemical tests were carried out using a three-electrode setup in a (quiescent) FC PEC H-Cell 2 × 15 mL - Front Contact Photo-Electrochemical H-Cell (Redox.me). The working electrode potential was controlled by an Autolab PGStat302N potentiostat (Metrohm). Electrode potentials were referenced with Redox.me refillable AgCl/Ag reference electrodes (sat. KCl, 0.197 VSHE at 25 °C) and a Redox.me Pt auxiliary electrode was used.

Photoelectrochemical impedance spectroscopy (PEIS) measurements were conducted to investigate the effect of the co-catalysts on bulk resistance. Frequencies of 105 to 10−1 (10 points per decade) were used with a constant applied electrode potential of 0.8 V ± 10 mV. PEIS data was processed using Nova 2.1.8 (Autolab). The equivalent circuit shown in Fig. S3 was selected for the analysis.

Results and discussion

AACVD of cobalt oxide and subsequent electrochemical treatment

During AACVD, substrate temperature directly influences the thermal decomposition of the precursor and resulting film composition. The TGA profile of cobalt (II) acetylacetonate (Co(acac)2) in air (Fig. 1a) shows an initial weight loss of 6 wt% between 30 °C and 100 °C, corresponding to loss of surface adsorbed water. A gradual weight loss from 100 °C to 140 °C followed by a significant mass reduction between 140 °C and 230 °C is correlated to acac ligand expulsion. Combustion of residual carbonaceous species occurs up to 400 °C, after which the weight stabilises at a residual mass of approximately 10 wt%. Although this value is lower than would be expected from the molecular weights, melting and subsequent boiling of Co(acac)2 occurs at 165–170 °C and 215 °C respectively, concurrent with decomposition, explaining the non-stoichiometric mass loss. Accordingly, 400 °C was selected for all subsequent (AACVD)-CoOx depositions to ensure complete conversion of the cobalt precursors into the oxide.

Fig. 1: Analysis of a representative (AACVD + ET)-CoPi fi lm on FTO.
Fig. 1: Analysis of a representative (AACVD + ET)-CoPi fi lm on FTO.
Full size image

Analysis of a representative (AACVD + ET)-CoPi film on FTO deposited from a 5 mM Co(acac)2 precursor solution in 20 mL of ethanol at 400 °C and treated with 1.5 VRHE chronoamperometry for 10 min in KPi buffer solution. a TGA for the Co(acac)2 precursor, b Top-down and c Side-on SEM images, d XRD patterns, e Raman spectrum, and f EDX and XRF atomic % determination (n.d. not determinable).

Top-down and side-on SEM images for (AACVD)-CoOx on FTO are shown in the Supplementary Information Fig. S4a and S4b. The CoOx particles exhibit distinct angular facets, which are likely indicative of cubic CoO crystallites, such as those described by Melze et al20. Fig. 1b, c present top-down and side-on SEM images for the (AACVD + ET)-CoPi deposited on FTO from 5 mM precursor solution, showing an 85 nm thick layer on FTO and no visible changes from the 75 nm thick (AACVD)-CoOx samples, which is consistent with surface phosphate modification of the CoOx film. The resulting surface roughness and cross-sectional crystallite morphology may be a function of the FTO surface roughness.

Furthermore, XRD patterns, Raman spectroscopy, energy-dispersive X-ray spectroscopy (EDX) and X-ray fluorescence (XRF) of the (AACVD + ET)-CoPi samples were measured. In Fig. 1d, e, the XRD pattern shows peaks near 36 and 42 degrees two theta corresponding to cobalt oxide (CoO). The CoOx deposited herein by AACVD at a substrate temperature of 400 °C shows Raman peaks at 188, 467, 511, 602 and 672 cm−1 that are fully consistent with the Raman peak shifts for Co3O4 reported in the literature as having been prepared by the aerosol-assisted pyrolysis of cobalt acetate tetrahydrate between 250 and 350 °C21. This reported technique is similar to the AACVD investigated in this work, with nanopowders synthesised by pyrolysis in a tube furnace. The higher temperature required for AACVD may be a result of the surface heating orientation as the film is formed. The Raman peak shifts ranged from 184 to 191, 462 to 469, 508 to 511, 596 to 506, and 659 to 675 cm−1 respectively, indicating the AACVD material was closer in similarity to the Co3O4 prepared at 350 °C21. These peaks correspond to the Raman-active 3F2g, Eg, 2F2g, 1F2g, and A1g vibrational modes respectively, and are red shifted compared to Co3O4 standard22 which, along with bandwidth broadening, is indicative of decreased crystallite size21. The XRD data indicate the presence of CoO although the peaks are weak and do not exclude the presence of more amorphous Co₃O₄ phases. Raman spectroscopy more clearly shows features specific to Co₃O₄, but the Raman spectrum of CoO is characterised by a broad feature between 450 and 600 cm−122, which is possibly also present in the measurement. Ultimately, the catalyst may plausibly consist of a mixed phase with CoO and Co₃O₄. Catalytic performance reflects the state of the surface layer undergoing catalysis, but the specific stoichiometry would be expected to fluctuate under varying applied potentials. EDX (Fig. S5) and XRF (Fig. S6) spectra for (AACVD + ET)-CoPi (elemental abundance tabulated in Fig. 1f), show signals for cobalt and oxygen, indicating the successful synthesis of a cobalt oxide overlayer. The X-rays used in EDX were accelerated at 15 kV and up to 50 kV for XRF and hence the sample space analyzed is deeper (within the glass substrate) for the XRF measurements compared to for the EDX measurements. However, phosphate incorporation was undetectable by these methods. Figures S7 and S8 show XPS spectra and fitting tables of representative samples of (AACVD)-CoOx and (AACVD + ET)-CoPi on FTO. For (AACVD)-CoOx, XPS indicates cobalt is present as a 1:3 mixture of Co2+ and Co3+, identified with the XPS peak associated with Co2+ having an atomic ratio of 1:0.9 to oxygen associated with Co2+, and the peak associated with Co3+ having a ratio of 1:1.2 to oxygen associated with Co3+. After electrochemical treatment in phosphate buffer, the (AACVD + ET)-CoPi shows the presence of atomic 1.4 ± 0.4% phosphorous with a binding energy of 133 eV. The uncertainty is calculated from the residual standard error between the fitted curve and measurement data points, which is calculated to be 17 counts s-1 (Fig. S8). The standard error in the peak area is then 93 counts s-1, where the total area of the P 2p peak is 335 counts s-1 integrated over 5.5 eV of binding energy. As XPS is a surface sensitive technique (i.e. it probes up to ~10 nm), this result indicates that only the surface of the cocatalyst contains phosphate and explains why phosphorous is undeterminable by EDX or XRF analysis, as these techniques primarily probe the bulk (i.e. several microns deep). This limited conversion is consistent with the limited charge passed during chronoamperometry of ~0.8 mC cm−2 (Fig. S1). As detailed in the Supplementary Information, the charge passed is estimated to equal to a layer of CoPi 1.5 nm thick.

Photoelectrodeposition (PED) of CoPi from cobalt-containing solutions

CoPi deposited by PED over three 100 s, 600 s, and 3000 s chronopotentiometric repetitions are calculated to be 7 nm, 43 nm, and 217 nm thick according to side-on SEM imaging, discussed below. These values and their corresponding current densities are tabulated in Table S1. In Fig. 2a, the UV-visible absorptance calculated from transmittance and total reflectance measurements for 43 nm thick (PED)-CoPi on FTO has the same trend as 75 nm thick (AACVD + ET)-CoPi on FTO, with both having broad absorption and being brown in colour. In Fig. 2b, c, the top-down and side-on SEM images of (PED)-CoPi deposited from three 3000-s chronoamperometry cycles show that CoPi forms very flat, uniform and dense plates on the nanoscale, but with deep cracks between these plates seen on the micron-scale that reach down to the FTO substrate. Furthermore, there are small spaces between the CoPi slabs and the FTO surface. These structural defects between CoPi regions and especially between CoPi and the FTO may cause easier delamination and hence explain the lower stability of the PED-CoPi film compared to the AACVD + ET-CoPi. Charge transfer is also adversely affected compared to the densely packed conformal AACVD + ET-CoPi as evidenced by the lower 1.53 mA cm−2 current density for (PEC)-CoPi on BiVO4 for sulphite oxidation compared to 1.83 mA cm−2 for (AACVD + ET)-CoPi on BiVO4. Furthermore, bubble formation in the case of PED-CoPi films may be expected to preferentially occur at the large gaps in the catalyst, letting larger bubbles to aggregate, whereas the (AACVD + ET)-CoPi film may encourage more disperse and ultimately smaller bubble formation which should encourage a more stable photocatalytic performance. In Fig. 2d, XRD measurement is unable to detect crystalline phases of CoOx, nor of crystalline cobalt phosphate (Figure S9). In Fig. 2e, the (PED)-CoPi/FTO Raman spectrum exhibited peaks at 183 cm−1, 462 cm−1, and 650 cm−1, with peaks between 509 and 590 cm−1 associated with Co–O stretching vibrations. The broader Co3O4 peaks observed in (PED)-CoPi/FTO, compared to (AACVD + ET)-CoPi, suggest a more amorphous structure. This interpretation aligns with their XRD and SEM analyses, which showed reduced crystallinity in PED films. Tabulated in Fig. 2f, EDX spectra (Fig. S10) and XRF spectra (Fig. S11) of PED-CoPi confirmed the presence of both cobalt and phosphorus, although the stoichiometry is uncertain because of variance in the sampling volume of each technique. Figures S12 and S13 show the XPS spectra of (PED)-CoOx deposited in non-phosphated pH 9 borate buffer solution and (PED)-CoPi respectively. For PED-CoOx, residual boron from the borate buffer solution is observed along with peaks associated with tin from the FTO. This observation may be through the large cracks that form on the surface of the PED films. (PED)-CoPi is observed to have 14% atomic mass of phosphorous although also 4% of potassium which must have been incorporated from the deposition in potassium phosphate buffer.

Fig. 2: Analysis of a 43 nm thick (PED)-CoPi film on FTO deposited from three 600 second chronoamperometric cycles.
Fig. 2: Analysis of a 43 nm thick (PED)-CoPi film on FTO deposited from three 600 second chronoamperometric cycles.
Full size image

a Calculated UV-visible absorptance spectra compared alongside the FTO substrate and a 75 nm thick (AACVD + ET)-CoPi film on FTO, b Top-down and c surface focused higher magnification top-down SEM images, d XRD patterns with reference patterns, e Raman spectrum, and f EDX and XRF atomic % determination (n.d. not determinable).

Optimising the co-catalyst thickness on BiVO4 photocatalysts

AACVD-fabricated BiVO4 on FTO were deposited with a thickness of approximately 200 nm from 20 mL of precursor solution following previously reported methods16,23. To determine the co-catalyst thickness on BiVO4 that achieves optimal PEC performance, a range of CoPi films were synthesised via AACVD on top of BiVO4 substrates by varying the Co(acac)2 precursor concentration. Representative current-voltage (J–V) curves and average photocurrents under 1.23 VRHE applied potential for (AACVD + ET)-CoPi in Fig. 3a, b show that precursor concentrations of between 0.03 and 0.125 mM yielded the highest average PEC performance, with 0.125 mM chosen for further investigation due to the likely stabilising effect of using a larger amount of co-catalyst. This optimal precursor deposition concentration is calculated to yield films of approximately 2 nm thick, given that a 5 mM precursor concentration was observed to yield an 85 nm-thick film by side-on SEM. Techniques such as high resolution transmission electron microscopy would be required to definitively confirm the thickness and morphology of such thin cocatalyst layers, but as shown in Fig. S1, integration of the current passed during chronoamperometry to convert CoOx to CoPi is calculated to result in a CoPi layer approximately 1.5 nm thick, corroborating the above estimation.

Fig. 3: PEC performance under (front) illumination and dark conditions in borate buffer for (AACVD + ET)-CoPi/BiVO₄ deposited from cobalt oxide precursor solutions with concentration indicated.
Fig. 3: PEC performance under (front) illumination and dark conditions in borate buffer for (AACVD + ET)-CoPi/BiVO₄ deposited from cobalt oxide precursor solutions with concentration indicated.
Full size image

a Selected J–V curves in borate buffer and b mean ± 80% confidence interval (CI) current densities measured at 1.23 VRHE from measurements on 3–5 independent samples. 95% confidence intervals are shown in Fig. S14. PEC performance of (PED)-CoPi/BiVO₄ with (c) J–V curves at deposition durations shown (total of 3 PED cycles per sample). d Current density extracted from 1.23 VRHE in borate buffer for (PED)-CoPi/BiVO₄ photoanodes.

For PED-CoPi, three thicknesses were prepared on BiVO4 substrates using chronopotentiometry. Thickness was controlled by changing the deposition duration between 100 s, 600 s, and 3000 s per cycle with a total of three cycles conducted in each trial. These conditions yielded CoPi film thicknesses of ~7 nm, ~43 nm, and ~217 nm, estimated by side-on SEM (Fig. 2c for the 3000 s cycle sample). The sample deposited for 600 s exhibited the highest photocurrent density at moderate applied potentials (Fig. 3c, d). The difference in optimal thickness between the (AACVD + ET)-CoPi and (PED)-CoPi on BiVO4 is likely related to morphology of these samples. (PED)-CoPi is grown with wide, deep cracks (Fig. 2c) and therefore water can easily penetrate down to the BiVO4/CoPi interface for larger film thicknesses, which is reported to be integral to its function8. The (AACVD + ET)-CoPi is grown as a densely packed layer and is therefore less likely to be as porous for water to penetrate to the BiVO4/CoPi interface, leading to a thinner optimal layer thickness. Specifically, thicker depositions of (AACVD + ET)-CoPi may ultimately be formed of BiVO4/CoOx/CoPi, where the additional interfaces could hinder the charge transfer dynamics.

PEC performance of AACVD and PED cocatalysts on photoanodes

Linear scan voltammetry (LSV) curves

The LSV curves in Fig. 4a show that films of (AACVD + ET)-CoPi on BiVO4 yielded a more than 6-fold enhancement in photocurrent compared to bare BiVO4 under front illumination in phosphate-free, borate buffered electrolyte, reaching 0.74 mA cm−2 at 1.23 VRHE. In contrast, the (PED)-CoPi on BiVO4 fabricated with a 600 s per cycle deposition exhibited a near 5-fold increase, generating a photocurrent density of 0.60 mA cm−2.

Fig. 4: J–V plots measured in borate buff er under front illumination.
Fig. 4: J–V plots measured in borate buff er under front illumination.
Full size image

J–V plots measured in borate buffer under (front) illumination for comparing (a) cocatalysts on bare BiVO₄ and b co-catalysts on NN BiVO₄/WO3 Heterojunctions, c compilation of onset potential and current density at 1.23 VRHE for the samples in a, b.

In Fig. 4b, the use of nanostructured BiVO4-coated WO3 photoanode substrates coated with (AACVD + ET)-CoPi increased photocurrent density by 20% from 1.15 to 1.41 mA cm−2 at 1.23 VRHE. Similarly, 1.23 mA cm−2 was achieved with (PED)-CoPi on a nanostructured BiVO4-coated WO3 photoanode. However, the photocurrent onset potential remained relatively unchanged for both methods.

All photocurrent onset potentials and current densities at 1.23 VRHE are compiled in Fig. 4c for comparison and shown in the Supplementary Information Fig. S14. The onset potential for the (AACVD + ET)-CoPi on BiVO4 shifted cathodically by approximately 0.2 V compared to the bare BiVO4. This shift is likely due to a reduction in the water oxidation overpotential, introduced by CoPi, which facilitates improved charge separation and mitigates surface recombination losses. Conversely, the PED-fabricated sample showed approximately the same photocurrent onset potential as BiVO4. At modest applied potentials of 1.0 VRHE, the photocurrents of (AACVD + ET)-CoPi on BiVO4 exceeds that of (PED)-CoPi on BiVO4 and both far exceed the photocurrent of bare BiVO4. There is minimal variation in onset potentials observed for the heterojunction systems, compared to that of BiVO4. The staggered band alignment at the WO3/BiVO4 interface facilitates efficient carrier separation24, diminishing the additional impact of surface hole-collecting catalysts like CoPi. The catalytic contribution of surface CoPi is less significant when efficient charge separation is already provided by the WO3/BiVO4 interface.

One sun (AM 1.5 G) measurements with sodium sulfite sacrificial reagent

To investigate the effect of the co-catalysts on charge transfer, LSV measurements were conducted for each electrode under one sun (AM 1.5 G) illumination for sulfite oxidation to assess the intrinsic photoactivity without being limited by surface reaction kinetics. Figure 5 compares water and sulfite oxidation for BiVO4 electrodes with CoOx and CoPi co-catalysts. Figure S15 further compares the water oxidation performance for BiVO4 and WO3/BiVO4 heterojunctions with and without cocatalysts. Both the CoPi co-catalysts (synthesised by AACVD and PED) proved effective in enhancing charge separation, \({\phi }_{{\rm{sep}}}\) and charge injection, \({\phi }_{{\rm{inj}}}\) efficiencies compared to bare BiVO4 when conducting sulfite oxidation, with a markedly stronger effect when (AACVD + ET)-CoPi is used. The charge transfer efficiencies, along with the other parameters from Eqs. 5, 6, are summarised in Table S2 for one sun measurements under back illumination.

Fig. 5: Chopped LSV curves for water and sulfi te oxidation on BiVO4 photoanodes.
Fig. 5: Chopped LSV curves for water and sulfi te oxidation on BiVO4 photoanodes.
Full size image

a without a co-catalyst, b with (AACVD)-CoOx, c with (AACVD + ET)-CoPi (C), and with (PED)-CoPi. All measurements were taken under 1 sun (100 mW cm−2, AM 1.5 G) simulated sunlight under back illumination, with a chopping frequency of 1 Hz and scan rate of 20 mV/s.

Figure S16 shows the comparison of water and sulfite oxidation for WO3/BiVO4 electrodes with CoOx and CoPi co-catalysts. These electrodes exhibited high charge transfer efficiencies, in line with values reported previously for heterojunction electrodes fabricated by AACVD25. The electrodes with CoOx and CoPi co-catalysts also saw cathodically shifted onset potentials for water oxidation compared to baseline WO3/BiVO4 electrodes. The incorporation of CoOx and CoPi co-catalysts on the heterojunction photoanodes also enhanced charge injection (transfer) efficiencies (from the co-catalyst surface to the electrolyte) from 32 % in the heterojunction to 44%with the addition of CoOx and 86% with the addition of CoPi, with bulk charge separation efficiencies being observed to drop slightly because the photocurrents for sulfite oxidation were unexpectedly lower with the presence of co-catalyst. This decrease in bulk charge separation efficiency was attributed to increased bulk charge transfer resistance arising from additional solid | solid interfaces in the photoanode structure with the incorporation of the co-catalyst. This effect was less detrimental with the incorporation of the PED-synthesised CoPi, likely due to the more conformal coating of the co-catalyst layer afforded by the AACVD method. PEIS measurements (Fig. S17) showed that charge transfer resistance (Table S3) was approximately 16 times lower for the WO3/BiVO4 heterojunction (at 1.78 kΩ) than for bare BiVO4 (at 28.5 kΩ), providing an already lower baseline for further improvements prior to co-catalyst incorporation. Nevertheless, while charge transfer resistance was further reduced with co-catalysts, the capacitance increased for the (AACVD)-CoOx and (AACVD + ET)-CoPi co-catalysts relative to the WO3/BiVO4 baseline value, consistent with the explanation above. Nonetheless, the charge transfer efficiencies of the CoPi-coated WO3/BiVO4 electrodes are comparable to the 35.0 and 84.6% separation and injection efficiencies of the NiFeOOH-coated WO3/BiVO4 also fabricated by AACVD by the authors in previous work 26.

Photoelectrochemical impedance spectroscopy

Figure 6 compares Nyquist plots for BiVO4 electrodes with CoOx and CoPi co-catalysts. Compared to bare BiVO4 electrodes, PEIS analysis showed that the incorporation of the co-catalysts significantly reduced charge transfer resistance (represented by Rct in the equivalent circuit model and evidenced by the increased curvature in Fig. 6). A summary of the equivalent circuit parameters for different BiVO4 based electrodes is provided in Table S3. The charge transfer resistance was lowest with (AACVD + ET)-CoPi and (PED)-CoPi, at 502 Ω and 472 Ω respectively. The addition of co-catalysts also reduced the CPE capacitance compared to bare BiVO4. The capacitance values were lowest when the CoPi co-catalysts were incorporated. With the (AACVD + ET)-CoPi co-catalyst, the capacitance decreased tenfold, to 5.8 × 10−6 F and with the (PED)-CoPi co-catalyst, there was a further decrease, to 1.9 × 10−6 F. This significant drop in capacitance suggests the incorporation of CoPi resulted in the passivation of surface states, widening the depletion region and boosting charge separation.

Fig. 6: Nyquist plots.
Fig. 6: Nyquist plots.
Full size image

Nyquist plots from PEIS measurements (0.1 to 1 × 105 Hz, under AM 1.5 G 100 mW cm−2 simulated sunlight, at 0.8 VRHE) comparing BiVO4 with co-catalysts deposited from AACVD & AACVD + ET, inset zoomed in for co-catalysed samples.

Stability testing

To evaluate the stability of the CoPi-coated BiVO4 electrodes, chronoamperometry measurements were conducted at a constant potential of 1.23 VRHE under continuous one sun illumination (Fig. 7a). Over four hours, the current density produced by the bare BiVO4 electrodes decreased steadily in line with previously reported BiVO4 photo-corrosion effects27,28,29,30. After testing, the bare BiVO4 electrodes saw a reduction in photocurrent during linear sweep voltammetry of more than 90% at 1.23 VRHE. BiVO4 degradation is typically driven by two mechanisms: the photo-oxidation of Bi3+ ions (which occurs at the commonly used photoanode characterisation potential of 1.23 VRHE) and V5+ leaching, which occurs as soon as BiVO4 comes into contact with the electrolyte28. Previous studies have demonstrated that co-catalysts can mitigate these effects26,31, which address Bi3+ oxidation by increasing the favourability of water oxidation kinetics over self-oxidation, and by providing a physical barrier at the BiVO4/electrolyte interface that can suppress V5+ leaching. The CoPi co-catalyst fabricated by AACVD appears effective in suppressing BiVO4 degradation, with the photocurrent at 1.23 VRHE retaining near to 90% of its initial value after four hours of chronoamperometry, as shown in Fig. 7b. The drop in photocurrent of the CoPi-coated BiVO4 electrodes during chronoamperometry is attributed to polarisation of the electrode surface. This behaviour is consistent with reports that redox charging and ion accumulation within the CoPi layer under sustained anodic bias temporarily suppress hole injection from BiVO4 into the co-catalyst layer10,32, which can be reversed by a brief potential sweep or open circuit relaxation. Moderately longer-term stability measurements to the tens of hours would be expected to yield similar stability results, with minor losses in performance possibly arising from gradual CoPi photocorrosion and fragmentation off of the electrode33, culminating in more rapid performance degradation after a certain proportion of the BiVO4 is exposed. Long-term stability commensurate with commercial applications would need to be demonstrated over hundreds of hours on prototypical devices, as scaled up photoelectrodes would be expected to show different failure modes than laboratory-scale samples.

Fig. 7: Chronoamperometry (CA) stability testing.
Fig. 7: Chronoamperometry (CA) stability testing.
Full size image

a CA comparing BiVO4 with 0.125 mM precursor solution (AACVD + ET)-CoPi and 600-s cycles (PED)-CoPi coatings at an applied potential of 1.23 VRHE, b) J–V curves comparing (AACVD + ET)-CoPi before and after stability testing, and c before and after photographs of the samples tested in a) over 4 h.

The (AACVD + ET)-CoPi-coated BiVO4 electrodes were able to recover to near their original performance after chronoamperometry, although the (PED)-CoPi-coated BiVO4 electrodes degraded irreversibly. As with the bare BiVO4 electrode, the photocurrent did not recover after de-polarisation of the electrode and visible changes to the photoactive area are observed after the stability test, as shown in Fig. 7c. After 2.5 h, it was observed that the (PED)-CoPi-coated BiVO4 electrodes showed no photo-activity. This difference between the two deposition methods is attributed to the AACVD procedure resulting in a more homogenous and surface-adhered coating of CoOx than the cocatalysts fabricated by PED. This manifested in cracks in the (PED)-CoPi layer and spaces between the (PED)-CoPi and FTO (Fig. 2), which may result in a less effective barrier at the BiVO4/electrolyte interface.

Incident Photon-to-Current Efficiency (IPCE) and Half-Cell Solar-to-Hydrogen (HC-STH) Efficiency

Figure 8 presents the IPCE profiles under front-illumination for CoPi-modified BiVO4 photoanodes fabricated using AACVD + ET and PED. Figure 8a compares CoPi-modified samples on BiVO4 substrates. The (AACVD + ET)-CoPi/BiVO4 photoanode exhibited an average incident photon-to-current efficiency (IPCE) of 19%, representing a near fivefold improvement compared to the bare electrode. Meanwhile, the PED-CoPi/BiVO4 configuration achieved a slightly lower IPCE of 16%, corresponding to a fourfold enhancement. The peak IPCE values for the bare, (AACVD + ET)-modified, and PED-modified BiVO4 electrodes were seen at 400–425 nm, with peak efficiencies of 4%, 20%, and 16%, respectively.

Fig. 8: Incident photon-to-current effi ciency (IPCE) profi les measured in borate buff er under front illumination.
Fig. 8: Incident photon-to-current effi ciency (IPCE) profi les measured in borate buff er under front illumination.
Full size image

a flat BiVO4 with CoPi-decoration by AACVD + ET or by PED b a NN heterojunction of BiVO4/WO3 with CoPi-decoration by AACVD + ET or by PED.

Figure 8b shows the corresponding heterojunction nanostructured (NN) systems compared to (AACVD + ET)-CoPi WO3/BiVO4, and (PED)-CoPi WO3/BiVO4, which show higher IPCE compared to the single-layer BiVO4. For illumination at wavelengths between 350 and 425 nm, the (AACVD + ET)-CoPi heterojunction maintained an average IPCE of 29%, reflecting a 1.2-fold enhancement over the NN heterojunction within the same spectral region. The (PED)-CoPi heterojunction showed an even higher average IPCE; approximately 31%. All three heterojunction configurations exhibited peak IPCE values at 400 nm: 25% for the bare, 29% for (AACVD + ET)-fabricated cocatalyst, and 31% for PED-deposited cocatalyst samples.

To evaluate the overall energy conversion potential, the solar predicted photocurrents (SPP) (Equation 3) were used to calculate the HC-STH conversion efficiencies (Equation 4). For bare BiVO4, shown in Fig. 9 below, the SPP was calculated as 0.17 mA cm−2, yielding a HC-STH efficiency of 0.21%. Significant enhancements were observed following CoPi deposition: (AACVD + ET)-CoPi and (PED)-CoPi increased the SPP to 0.94 and 0.49 mA cm−2, corresponding to HC-STH efficiencies of 1.16% and 0.60%, respectively. When integrated into the NN heterojunction structure, further improvements were noted. The (PED)-CoPi/NN heterojunction achieved the highest SPP of 1.34 mA cm−2, followed by 1.19 mA cm−2 for the (AACVD + ET)-CoPi/NN heterojunction, and 1.11 mA cm-2 for the bare NN heterojunction. The corresponding HC-STH efficiencies for the bare, (AACVD + ET)-, and (PED)-modified heterojunctions were 1.36%, 1.46%, and 1.65%, respectively.

Fig. 9: Incident photon-to-current efficiency (IPCE) and solar predicted photocurrent (SPP) measurements in borate buffer under front illumination.
Fig. 9: Incident photon-to-current efficiency (IPCE) and solar predicted photocurrent (SPP) measurements in borate buffer under front illumination.
Full size image

a bare BiVO₄, (AACVD + ET)-CoPi/BiVO₄, and PED-CoPi/BiVO₄ photoanodes, and b for cocatalysts on BiVO4/WO3 heterojunctions, illustrating in the shaded regions their solar flux harvested.

In comparison, (AACVD + ET)-CoPi deposited on flat WO3/BiVO4 heterojunction yielded an SPP of 1.12 mA cm−2 and a corresponding HC-STH efficiency of 1.38%, (Fig. S18) which is an improvement over the bare flat heterojunction that exhibited an SPP of 0.71 mA cm−2 and a corresponding HC-STH efficiency of 0.87% but is less effective than the nanostructured photoanodes. A summary of all photoanode performances in 1 M borate buffer in this work is presented in Table 1.

Table 1 Summary of photoanode systems tested in 1 M borate buffer in this work.
Full size table

Comparison to reported cobalt cocatalyst-based BiVO4 photoanode systems

In 2023, Chen et al34. fabricated a CoOx film on Fe-doped BiVO4 by drop-casting Co(acac)2 precursor solution, followed by annealing, resulting in a photoanode demonstrating a photocurrent density of approximately 4.0 mA cm−2 at 1.23 VRHE. They achieved an IPCE of 54% at 400 nm, which was a threefold enhancement compared to their bare BiVO434. Also in 2023, Varma et al13. reported the fabrication of Fe-doped CoOx co-catalyst layers on BiVO4 photoanodes via photo-assisted electrodeposition. Their optimised photoanode achieved a photocurrent density of 4.6 mA cm−2 at 1.23 VRHE 13.

Although the AACVD-fabricated CoPi-decorated BiVO4 system presented in this study exhibits a lower photocurrent density of 0.74 mA cm−2 under front-side illumination and HC-STH efficiency of up to 1.5%, it still achieves a 6x enhancement factor over bare BiVO4 samples, and superior stability, demonstrating the potential of this technique for commercial devices. Moreover, the AACVD fabrication method employed ensures scalability, reproducibility, and cost-effectiveness over conventional methods used for CoPi fabrication (i.e., photo-assisted electrodeposition or drop-casting) and BiVO4 preparation (i.e., spin-coating, spray pyrolysis, or electrodeposition).

Conclusions

An aerosol-assisted chemical vapour deposition and electrochemical treatment (AACVD + ET) technique was developed to grow surface-modified CoPi films to serve as co-catalyst layers on BiVO4 and BiVO4/WO3 heterojunction photoanodes for application in photoelectrochemical (PEC) water splitting. The physical properties and performance of these CoPi films were compared with those made by a traditional photoelectrodeposition (PED) method. XRD and Raman characterisation suggest that the films first grown by AACVD form a crystalline mixed CoO and Co2O3 structure. After electrochemical treatment (ET), linear sweep voltammetry confirmed the catalytic activity of the photoanodes coated with co-catalysts, and chopped light scans in the presence of sulfite oxidation showed that (AACVD + ET)-CoPi on BiVO4 demonstrated the highest charge separation efficiency of 84%, higher than (PED)-CoPi on BiVO4 at 70%. EDX and XRF analysis confirmed the existence of phosphate in (PED)-CoPi films, whereas XPS analysis was required to detect minute amounts of phosphate in (AACVD + ET)-CoPi films, suggesting only surface modification took place. For flat BiVO4 photoanodes coated with CoPi, those grown using AACVD + ET exhibited higher, average IPCE (~19% vs ~16% across the 350–430 nm wavelength range), SPP (0.94 mA cm−2 vs 0.49 mA cm−2) and calculated HC-STH efficiencies (1.16% vs 0.60%) compared to (PED)-CoPi on BiVO4. Nyquist plots derived from photoelectrochemical impedance spectroscopy suggest that CoPi deposited by either method provide similarly effective reductions in surface resistance and capacitance. However, only (AACVD + ET)-CoPi was significantly stable over medium-term continuous operation, retaining more than 90% of its original photocurrent after four hours chronoamperometry, compared with (PED)-CoPi which was totally degraded. This demonstrates the viability of this technique to produce scalable photoanodes.