Particulate photocatalysis is a promising platform for sunlight-driven hydrogen production and CO2 reduction at scale.1–5 Although recently particulate photocatalysts have attained nearly 100% quantum efficiency,6 their state-of-the-art solar-to-hydrogen (STH) conversion efficiency, including those of oxides,5,7 oxynitrides,8,9 and oxysulphides,10 remains around 1%.11 An alternative direction is to utilise narrow bandgap semiconductive photoabsorbers, such as those II-V and III-V semiconductors, that are known for strong optical absorption, high quantum efficiency, and long carrier lifetime.12–14 For example, cadmium sulphide (CdS) and gallium-indium-phosphide (GaInP2) photoabsorbers have bandgaps of 2.42 eV and 1.87 eV, respectively, promising >10% STH efficiency.1,15 However, these semiconductors suffer from poor photo-stability in water, and their few-hour stability is much shorter than the thousands of hours operation by oxides.5 Hole scavengers leverage their fast and irreversible kinetics to compete against photooxidation.16 These scavengers provide the chemical bias to improve H2-evolution rates, but they cannot be regenerated to sustain continuous photocatalysis.
Existing coatings were shown to protect photoelectrodes for either reductive or oxidative reactions, but not for hosting both reactions simultaneously.17 Herein, we found the oxide coatings form a heterojunction, not a Schottky barrier, with the underlying protected semiconductors. Therefore, we developed a general approach to improve the stability and at the same time, to facilitate efficient charge separation and to transfer electrons and holes simultaneously during coevolving photocatalysis. Corroborated by both energetic measurements and numerical calculations, the charge separation was found driven by a semiconductor/coating/cocatalyst structure with locally varying barrier heights, rather than the specific morphology or crystal facets of the photocatalyst. As illustrated in Figure 1a, CdS particulate films and GaInP2 epitaxial films were coated by TiO2, the surfaces of which were subsequently loaded with Rh cocatalysts (see Methods). This simple two-step procedure transformed narrow-bandgap semiconductors into stable photocatalysts: the CdS/TiO2/Rh@CrOx panel continuously evolved H2 at 50.4 μmol h−1 cm−2, and showed an internal quantum yield (IQY) of 44.3% at 438 nm in a Na2S solution. The GaInP2/TiO2/Rh@CrOx panel achieved an H2-evolution rate of 144.7 μmol h−1 cm−2, equivalent to a photocurrent density of 7.8 mA cm−2. While the TiO2 surface oxidised the reversible sulphide/polysulphide (S2−/Sn2−) redox couples, the Rh-core and CrOx-shell, i.e. Rh@CrOx, cocatalysts evolved H2 and suppressed the re-reduction of redox mediators. The CrOx shell also prevented performance degradation and sulphur poisoning of the Rh cocatalysts.
Stability of the coating-stablised photocatalysts
A film of drop-casted CdS particles was encapsulated by TiO2 via atomic-layer deposition (ALD) (Fig. 1a, inset). Then, the ALD TiO2 was decorated with Rh@CrOx core-shell cocatalysts via photodeposition (see Methods). This cocatalyst-decorated coating not only suppresses photocorrosion of the CdS semiconductor but also allows coevolution of H2 and reversible redox mediators. In this case, photo-excited electrons and holes simultaneously inject to the active sites of Rh cocatalysts and bare TiO2 surfaces, respectively, to drive the hydrogen evolution reaction (HER) and sulphide oxidation reaction (SOR) along the photocatalyst-liquid interface. It is necessary to capture hole charges by the reversible redox mediators (A/A−) to sustain overall water splitting, but it is also challenging due to their reversible kinetics. In this work, we mainly focus on the reversible SOR based on the sulphide/polysulphide (S2−/Sn2−) redox mediators in alkaline pH, but it can be extended to other redox mediators, e.g., IO3−/I−, [Fe(CN)6]3−/4−, and Fe3+/2+ in various pH.18–20
CdS photocatalysts are known to rapidly degrade due to photooxidation of lattice sulphides, which forms soluble sulphates or insulating sulphur.21 Typical CdS photocatalysts were reported to evolve H2 that lasted for only a few hours.22,23 Therefore, we tested the stability of TiO2-stablised CdS panel by measuring the cumulative amount of H2 evolved in 50 mM Na2S solutions (pH 13.0) over time (see Supplementary Fig. S1, Note 1, and Video 1). As shown in Fig. 1b, the CdS/3-nm TiO2/Rh@CrOx panel produced 1-atm H2 for 150 hours continuously. To simplify the testing procedure, the redox solution was replenished by a new Na2S solution every 8 – 9 hours without disturbing the measurement. In comparison, the bare CdS panel produced H2 at a much lower rate while the activity dropped to almost zero after 10 hours. The lack of TiO2 coating made the CdS powder film delaminate even during the Rh photodeposition. The X-ray photoelectron spectroscopy (XPS) Cd 3d and S 3p core-level spectra showed little changes after 150 hours (Supplementary Fig. S2). The absence of SOx signature suggested that the TiO2 coating eliminated the primary failure mode of CdS photocorrosion. The H2-production activity decreased by 20% and 30% after the first 100 h, and 150 h, respectively, which was not due to the loss of CdS particles (Supplementary Fig. S3 and Note 1).10 In another compartment, (photo-)electrochemical redox-mediator reduction and water oxidation are paired with the H2 photocatalysis to complete overall water splitting (see the Photocatalytic reactor Section).
Design for efficient charge separation and product coevolution
The introduction of coatings and cocatalysts improved the photocatalytic activity of CdS and GaInP2. The commercial CdS powder film showed a 1-atm H2-evolution rate of only 2.2 μmol h−1 cm−2 in a 100 mM Na2S solution. The rate increased by a factor of 1.3 and 5.8, respectively, after coating 3 nm of TiO2 only and after loading 1 wt% Rh on the TiO2 (Fig. 2a). Upon growth of CrOx shells on individual Rh nanoparticles, a significant boost for the activity to 50.4 μmol h−1 cm−2 was achieved. Similar improvement was observed with the GaInP2 panel. With 3-nm TiO2 and Rh@CrOx deposition, the n-type GaInP2 film evolved 1-atm H2 at 144.7 μmol h−1 cm−2 in the Na2S solution, 11 times of the as-grown GaInP2 (Fig. 2b). This H2-evolution rate was stable, while the bare GaInP2 dropped to 28.8% of the initial rate within the first 1.5 hours. To our knowledge, this is the first demonstration of coevolving GaInP2 photocatalysts. The equivalent photocurrent density of 2.7 and 7.8 mA cm−2 was achieved by the TiO2-stablised CdS and GaInP2 panels, respectively. Besides S2−/Sn2−, coevolution of H2 and [Fe(CN)6]3−/4− redox couples by the CdS/TiO2/Rh@CrOx panel was also observed in a 50 mM K4Fe(CN)6 aqueous solution at pH 7.0 (Supplementary Fig. S4), showing the versatility of this coating strategy. The rate can be further improved by tuning the coating energetics and making facile charge transfer to redox mediators in future studies.24,25
Efficient charge separation, leading to the simultaneous injection of electrons and holes to the liquid, can be realised through the varying barrier heights along the same photocatalyst-liquid interface. Cocatalyst deposition introduces two types of sites, i.e., semiconductor/TiO2 and semiconductor/TiO2/Rh@CrOx, when in contact with the redox-mediator solution. We investigated their barrier-height energetics through light-intensity dependent OCP measurements by using CdS particle-assembled electrodes and GaInP2 planar electrodes (see Methods).26 The ohmic back contact for the CdS and GaInP2 electrodes allows for probing of the averaged electrochemical potentials of photo-excited electrons. Fig. S5a shows that as the light intensity increased, the OCPs of the CdS/TiO2 electrode measured in a 10 mM Na2S solution (pH 12) became increasingly negative and reached a steady-state value of 0.34 V vs. RHE, denoted as OCPlight. The OCPlight shifted to −0.09 V vs. RHE after loading the CdS/TiO2 electrode with Rh@CrOx (Fig. 3a), indicating a reducing potential for the Rh@CrOx sites to drive H+ reduction. The OCPlight of the CdS/TiO2/RhCrOx electrode was consistent with the conduction band (CB) edge position of the CdS/TiO2 sites, which was determined from the Mott-Schottky analysis independently (Supplementary Fig. S5b and Note 2). Therefore, both sites are able to coevolve HER and SOR, according to their band edge positions (Supplementary Fig. S6).
Purging the redox solution with H2 resembles the local environment during H2-evolving photocatalysis. When multiple charge transfer pathways coexist at the photocatalyst-liquid interface, e.g., a two-redox liquid junction, the OCP is the potential at the detailed balance of the kinetic rates for all the forward and backward charge-transfer pathways.26 Unlike the almost unaltered OCPs of the CdS/TiO2 electrode after H2 purging (Supplementary Fig. S5a), the OCPdark of CdS/TiO2/Rh@CrOx approached 0 V vs. RHE, while the OCPlight remained at − 0.10 V vs. RHE (Fig. 3a). In other words, the potential of the Rh cocatalysts (inside CrOx shells) shifted upwards to align with E0(H+/H2). Meanwhile, the CdS CB position stayed at − 0.10 V vs. RHE. Instead of a fixed value, the barrier height at those CdS/TiO2/Rh@CrOx sites was reduced to 0.11 V, much smaller than the CdS/TiO2 barrier height of 0.60 V (Supplementary Note 2). The variable barrier height of Rh nanoparticles is not specific but rather general for many nanoparticulate cocatalysts: they were typically photodeposited over a hydrous or electrolyte-wetted TiO2 surface, thus applicable to our coevolution coating strategy.26,27 Following the same procedure, we obtained electron barrier heights of 0.68 V and 0.14 V for the GaInP2/TiO2 and GaInP2/TiO2/Rh@CrOx liquid interface, respectively (Supplementary Fig. S5c,d and Note 2).
Our strategy achieves local charge separation without leveraging the barrier-height difference specific to crystal facets.6,28 Both photocatalysts feature a barrier-height difference, or asymmetric barrier energetics.29 As illustrated in Fig. 3b and 3c, the barrier-height difference induced an electric field between the HER and SOR sites. This field made photogenerated electrons transport to the Rh@CrOx sites for H2 evolution, whereas holes transport to and accumulated at the bare TiO2 surface to oxidise the redox mediator. The detailed calculation for the band diagram is explained in the Supplementary Note 2 and Fig. S7, and the various energetic parameters are summarised in Supplementary Table S1.
However, only decorating Rh cocatalysts but without regulating the re-reduction of redox mediators cannot achieve sufficient barrier-height difference. The OCPlight of CdS/TiO2/Rh electrodes, i.e., the Rh potential, was not raised to more negative potentials (Supplementary Fig. S8). As a two-redox liquid junction,26 the Rh site could also transfer electrons to Sn2−/S2−: this process was fast enough to compete with the desirable pathway of H+ reduction. The resulting 0.45 V barrier height of the Rh sites was comparable to that of the CdS/TiO2 sites. Therefore, this lack of energetic asymmetry adversely affected the charge separation efficiency for semiconductor/TiO2/Rh panels without the CrOx selectivity shell layer (Fig. 2). Besides, the selective Rh@CrOx cocatalysts ensure the high activities regardless in 1-atm Ar, air, or H2 atmospheres (Supplementary Fig. S9). Such high selectivity against O2 reduction or Sn2− reduction is consistent with the unchanged OCPlight at ca. − 0.10 V vs. RHE for the semiconductor/TiO2/Rh@CrOx panels. The continuous H2 production eventually replaces the headspace with 1-atm H2. Essentially, the observed activity trends highlighted the dual functionality of the Rh@CrOx cocatalysts: i) improving kinetics and ii) creating barrier-height asymmetry to facilitate charge separation.
Modelling of steady-state charge separation
So far, the HER and SOR sites have been proven to bear different barrier heights. Therefore, we choose the CdS photocatalyst to build a semiconductor model and to simulate a steady-state distribution of photo-excited electrons and holes in two dimensions, i.e., inside the CdS particle and along the TiO2-liquid interface. The materials and junction properties used are discussed in Supplementary Note 3 and listed in Table S2. The CdS/TiO2 solid-solid interface is modelled as a heterojunction. The barrier-height energetics vary along the liquid junction dynamically during H2 photocatalysis: for simplicity, fixed Schottky barriers of 0.11-V and 0.60-V barrier height were set to simulate those HER and SOR sites, respectively. Their geometric parameters used in the simulation were estimated from Fig. S10 – S11.
The gradient of electron potential energy within the CdS indicated the steady-state electron-hole separation. A uniform carrier generation rate of ca. cm−3 s−1 was calculated under 1-sun illumination (see Methods). Fig. 3d shows that the electron potential energy reached the maximum at the SOR sites and the minimum at the HER sites. Driven by this potential gradient, the simulated electron and hole transport directions, indicated by the red and black arrows, respectively, were consistent with the 1D schematic in Fig. 3b. From the HER and SOR site at the liquid interface into the CdS bulk, ca. 200 nm or 20% of the absorber volume was under an electric field (Supplementary Fig. S12a − 12c). Within this volume, either the electron or hole current dominated the total currents (Supplementary Fig. S12d). Away from the surface, electron diffusion current became comparable to and eventually higher than the electron drift current (Supplementary Fig. S12e). This spatial distribution indicated a steady-state electron accumulation at the Rh@CrOx site (Supplementary Fig. S12f). Essentially, this diffusion-controlled charge separation is distinctive to the drift-dominated regime typically observed in solid-state energy conversion:30 even a barrier-height difference much less than the bandgap enables efficient charge separation during coevolution.
Too thick TiO2 coatings may reduce the hole injection rates, and thus, affect charge separation efficiency and photocatalytic activity. Experimentally, we observed that H2 rates of the CdS photocatalyst decreased dramatically when the TiO2 thickness reached 10 nm (Supplementary Fig. S13). This thickness-dependent behaviour should not always occur but was attributed to the energetic mismatch during hole transport from CdS to the redox mediator. Through > 3-nm thick TiO2, the hole hopping transport was slow as compared to the direct tunnelling transport,31 because the Ti3+-defect band does not match with the energy level of CdS valence bands. In the modelling, different charge injection currents can be simulated by modulating the hole surface recombination velocity (SRV) at the 10-nm TiO2. The modelling outcome is consistent with the measured activity: reducing the hole SRV by 10 times decreased the photocurrent, or the H2-evolution rate, to a third of the 3-nm TiO2 coated photocatalyst (Supplementary Fig. S13). The calculated effective resistivity of ~ 108 Ω for the 10-nm TiO2 layer also agrees with the literature report.31 The low SRV led to significant hole accumulation up to 1021 cm−3 at the 10-nm TiO2 surface (Supplementary Fig. S14). Therefore, more carrier recombination occurred inside the CdS than with 3-nm TiO2 coatings, reducing the photocurrents adversely. Insufficient hole transport that causes activity degradation is irrespective to CdS carrier lifetime or radiative efficiency.
Improving quantum yields of CdS photocatalysts
With 3-nm TiO2, 1 wt% of Rh, and 1 wt% Cr loading, one CdS photocatalytic panel produced H2 at a rate of 50.4 μmol h−1 cm−2 in a 100 mM Na2S solution (pH 13.5) (see Methods). We measured an apparent quantum yield (AQY) of 24.6% at 438 nm for the same panel in the 100 mM Na2S solution. Considering 55.5% of the light absorbed at that wavelength (Supplementary Fig. S15), the internal quantum yield (IQY) of the panel was calculated to be 44.3%. Stacking multiple panel devices can take full advantage of the incident light. With a three-panel stacking, the H2 evolution rate increased to 90.6 μmol h−1 cm−1, and the AQY reached 44.2%, approaching the IQY (Supplementary Fig. S16 and Video 2). This H2 rate promises STH efficiency of 5.9% in a redox-mediated solar H2 generator.
The numerical simulation also helped us discover the property that limited the measured IQY to be 44.3%. Experimentally, we obtained a CdS carrier lifetime of 0.1 ns (Supplementary Fig. S17a), which is short. Using the experimental lifetime, we introduced Shockley-Read-Hall recombination in the model (see Methods) to show that 44.0% of photo-excited electrons are collected as photocurrents. Varying the modelling parameter further showed the possibility of reaching 90% IQY, as long as the carrier lifetime of commercial CdS powders could be longer than 0.5 ns (Supplementary Fig. S17b). Such sensitivity to the carrier lifetime is consistent with the diffusion-controlled charge separation driven by a small barrier-height asymmetry. Therefore, increasing the carrier diffusion length or the depletion region volume is desirable (Supplementary Note 4),29 and can follow various reported strategies.32
We constructed a solar-fuel generator, which allows for H2 and O2 evolution at two separated reactor chambers continuously, or at two separated times. As illustrated in Fig. 4a, a photocatalysis (PC) reactor was integrated with a photovoltaic-driven electrolysis (PVE). The reversible S2−/Sn2− redox mediators shuttle photogenerated hole charges between the PC reactor and PVE device. A triple-junction (3-J) amorphous Si (a-Si) solar cell of 1.7 − 2.0 eV bandgaps absorbs the light transmitted through the CdS panel, to power the redox regeneration at a matching current.33 This proof-of-concept redox-mediated water-splitting reactor was constructed for simplicity.34,35 For the redox-mediator regeneration, the PVE component can be essentially replaced by a suspension or a panel of particulate photocatalysts:36 e.g., the iron-, cobalt-, and iodine-mediated O2 evolution by BViO4 and WO3 photocatalysts.20,37,38 More generally, the GaInP2/TiO2/Rh@CrOx panel can also operate inside the H2-evolving chamber when a GaInP2 epi-layer is lifted-off from reusable GaAs substrates and mounted on a glass panel.39 The light transmitted through the GaInP2 layer can drive redox regeneration.
Fig. 4b shows a separated production of stoichiometric H2 and O2 by alternating the PC and the PVE processes. The amount of H2 evolved was directly measured by an online gas chromatograph (GC), while the amount of O2 was quantified based on the charge passed and the 100% Faradaic efficiency of IrOx catalysts (see Methods). The membrane-electrode electrolysis device consists of a Pt/C catalyst-loaded cathode and an IrOx/C catalyst-loaded anode. The cathode reduces Sn2− ions to regenerate S2−, whereas the anode produces O2 and protons by oxidising pure water. A cation exchange membrane (CEM) prevents the O2 crossover, while replenishing the redox solution with the protons needed for H2 evolution. In the PC step, 115.3 μmol of H2 was evolved under 1-sun illumination at 1 atm, equivalent to 22.2 C of charges received by the reversible S2−/Sn2− redox mediators. In the second PVE step, the same amount of charge was passed to the electrolysis device. This step generated 57.7 μmol of O2, i.e., a 2:1 molar ratio of H2 and O2. For this demonstration, the 25.6 μmol h–1 cm–2 rate of H2 evolution under 1-sun illumination indicated STH conversion efficiency of 1.7%, equivalent to a photocurrent of 1.4 mA cm−2. The H2 rate of 25.6 μmol h−1 cm−2 after two consecutive cycles was kept the same as in the first cycle, and the potential of the redox mediator solution also returned to the original open-circuit potential after each PVE regeneration (Supplementary Note 5 and Fig. S18). Both indicated complete regeneration of the redox mediators.
The rates of oxygen evolution and redox regeneration by the electrolysis device are further scalable. Thus, it is also reasonable to assume that all remaining sunlight transmitted through the CdS panel can be utilised to power the S2−/Sn2− regeneration at a rate matching to the H2 production. In this case, the reactor should achieve STH conversion efficiency as high as 5.9% by stacking three CdS panels to maximise the light capture. Besides, the redox mediators buffer the photo-excited charges, such that the reactor operation can be flexibly tailored by using short-term energy storage, even with intermittent solar irradiation or diurnal solar cycles. A similar water-splitting reactor based on the TiO2-stabilised GaInP2 panel can pair with efficient O2-evolving photocatalysts, such as stabilised GaAs or Si,29 to utilise the transmitted light and regenerate redox mediators. The regeneration rate can match to that of the H2 evolution by the GaInP2 panel shown in Fig. 2b, achieving STH efficiency of 9.4%. Equivalently, 1-sun H2 production rates of 4.8 10–7 and 7.7 10–7 kg s–1 m–2 are achievable from coating-stablised CdS and GaInP2, respectively, which meets the US Department of Energy MYRD&D 2020 target for dual bed photocatalytic systems with product separation.40