Addressing solar photochemistry durability with an amorphous nickel antimonate photoanode

Renewable generation of fuels using solar energy is a promising technology whose deployment hinges on the discovery of materials with a combination of durability and solar-to-chemical conversion efﬁciency that has yet to be demonstrated. Stable operation of photoanodes has been demonstrated with wide-gap semiconductors, as well as protected visible gap semiconductors. Visible photoresponse from electrochemically stable materials is quite rare. In this paper, we report the high-throughput discovery of an amorphous Ni-Sb (1:1) oxide photoanode that meets the requirements of operational stability, visible photoresponse, and appreciable photovoltage. X-ray absorption characterization of Ni and Sb establishes a structural connection to rutile NiSb 2 O 6 , guiding electronic structure characterization via X-ray photoelectron experiments and density functional theory. This amorphous photoanode opens avenues for photoelectrode development due to the lack of crystal anisotropy combined with its operational stability, which mitigates the formation of an interphase that disrupts the semiconductor-electrolyte junction. analyses


INTRODUCTION
Solar photoelectrochemical (PEC) generation of chemical fuels using sunlight, H 2 O, and CO 2 is a promising renewable energy technology whose technology readiness would be vastly improved by the identification of an efficient and robust photoanode for the oxygen evolution reaction (OER). 1,2 Multiphysics modeling has demonstrated that a variety of factors underlie the solar-to-chemical (STC) energy conversion efficiency, most centrally the band gaps of the 1, 2, or 3 absorbers. 3 The STC along with the cost of the source materials and the device durability are the central factors in determining the price of sustainably generated fuel, for which the US Department of Energy has established technology targets. 4 In this context, photoanode discovery research entails co-optimization concerning a breadth of performance criteria, including (1) operational PEC stability, (2) a band gap suitable for efficient utilization of the solar spectrum, (3) valence band energy (versus vacuum level) sufficiently negative to catalyze the OER, and (4) conduction band and Fermi energies sufficiently positive to perform the fuel-forming reaction or more commonly to couple to a photocathode in a tandem absorber configuration. Commensurate with these requirements are the energy conversion efficiency considerations of (5) maximizing external quantum efficiency (EQE), characterized as the fraction of incident photons that give rise to photoanodic current, and (6) maximizing photovoltage, which is typically characterized as minimizing the turn-on potential, i.e., the lowest potential at which anodic photocurrent is generated. [5][6][7][8] These efficiency considerations encompass the need for the photoanode to support the transport of photocarriers while mitigating carrier recombination. Regarding the chemical space of interest (7) earth-abundant elements are desirable for scalability of the resulting technology and (8) the metal elements (or more generally all elements other than O and H) and their equilibrium dissolved metals concentrations during operation must satisfy the requirements for device-level durability. We recently discussed this last requirement in the context of catalyst codesign. 9 For discovery research that is not constrained by a specific device architecture, requirements (1) and (8) correspond to the demonstration of a consistent photoactivity with low equilibrium dissolved metals concentrations, i.e., the dissolved metals concentrations at which the Pourbaix decomposition energy of the operational photoanode surface is zero. This last requirement alone limits the search space to metal oxides or semiconductors that form a metal oxide passivation layer. Many metal oxides also meet the band alignment and nonprecious metal requirements, motivating the community's concerted effort to identify metal oxide phases that meet additional requirements. The recent proliferation in photoanode discovery has produced a compendium of metal oxides that each meet several but not all of these criteria. [10][11][12] Combinatorial synthesis and screening has been an effective strategy for discovering these photoanodes. [13][14][15][16] Hence the search continues with an increased focus on the two criteria that have been least commonly demonstrated: low turn-on potential and stable operation with low dissolved metals concentrations.
The majority of discovered visible light active phases (sub-2.8 eV band-gap energy) are V-, W-, and Fe-based oxide photoanodes, such as WO 3 , a-Fe 2 O 3 , BiVO 4 , several copper vanadates, CuWO 4 , a-SnWO4, FeWO 4 , and BiFeO 3 . 5,6 Among these candidate photoanodes, performance limitations include insufficient utilization of the solar spectrum, poor charge carrier separation and transport, and poor operational stability. For example, WO 3 exhibits high operational stability in acidic electrolytes but suffers from a >2.7 eV band gap that substantially limits its solar conversion efficiency. Conversely, a-Fe 2 O 3 shows a relatively low band-gap energy near 2 eV and high stability in alkaline media but is limited by its low carrier mobility and short hole diffusion length. A group of copper vanadates were identified with a 2 eV band gap and excellent stability in pH 9-10 electrolytes due to surface self-passivation but suffer from poor carrier transport and high turn-on potential. [17][18][19][20][21][22] CuWO 4 has a band gap between 2.2 and 2.4 eV and is stable under acidic and oxidizing conditions but exhibits poor photoactivity due to fast recombination of charge carriers, low electronic conductivity, and the presence of midgap electronic states on its surface. 23 Recent investigations of a-SnWO 4 have demonstrated favorable charge carrier transport properties and a suitable band gap of 1.9 eV, but the PEC performance could be limited by rapid surface passivation by SnO 2 . 24,25 BiVO 4 is the most widely studied photoanode material, and the intensive effort to optimize its performance has resulted in nearly perfect carrier separation. 5 Its 2.4 eV band gap limits the STC efficiency 3 and its propensity for corrosion at all values of pH compromises device durability. 5 The thermodynamic driving force toward corrosion is characterized by the Pourbaix decomposition energy, which is typically calculated at the concentration of a dissolved metal at or below 10 À5 M. 26 A photoanode with an appreciable Pourbaix decomposition energy under operating conditions poses substantial challenges for attaining operational durability. 27 One strategy is to saturate the electrolyte with dissolved metals, as was recently demonstrated for BiVO 4 by using a high V 5+ concentration to suppress corrosion. 28 We note in the discussion that this strategy is often incompatible with device-level durability. Alternatively, the photoanode can ''self-passivate'' if a Pourbaix-stable phase forms on the surface to mitigate further corrosion. This interfacial layer plays a critical role in the photoanode efficiency, but its electronic properties cannot be rationally designed to optimize those properties because the chemistry and structure are dictated by corrosion processes. Identifying photoanodes with favorable Pourbaix decomposition energies is thus a promising avenue to overcome the hurdles in metal oxide photoanode performance, which we address in this work through combinatorial experiments akin to the theory-guided tiered screening workflow we reported previously. 11 Metal antimonates, especially transition metal rutile structures of the form MSb 2 O 6 (M = Mn, Co, Ni), have been studied for various applications, such as propane ammoxidation, 29 gas sensors, 30-32 lithium storage, 33 and visible light photocatalyst, for degradation of water pollutant. 34,35 More recently, using computational Pourbaix diagrams, these rutile antimonates were identified as Pourbaix-stable in potential ranges relevant for oxygen electrochemistry in strong acid electrolytes. 36,37 Several transition-metal antimonates have been investigated experimentally as electrocatalysts for OER, 38-41 the chlorine evolution reaction, 40,42 and oxygen reduction reaction. 43,44 This family of materials includes examples of visible light absorption, as well as activity and stability for OER, motivating our combinatorial investigation of Ni-Sb oxide library as visible light photoanodes, which, to the best of our knowledge, has not been previously reported. While we successfully synthesized one target phase, NiSb 2 O 6 , its underwhelming photoactivity hinders its further development as a photoanode, despite its favorable Pourbaix energetics. By virtue of the combinatorial nature of our experiments, we identified a more promising material that exhibits similarities to NiSb 2 O 6 but has a Ni:Sb ratio near 1 and is amorphous. This photoanode is referred to herein as am-NiSbO z and its appreciable EQE, despite its lack of crystallinity, is unique among known photoanodes. Our combined experiment and theory characterization of this material demonstrates that it performs quite well concerning the requirements for broad spectral response, operational stability, and turn-on potential.

High-throughput experimentation
Combinatorial exploration of the Ni-Sb oxide system commenced with the synthesis of a Ni x Sb 1-x oxide composition spread thin film on F-doped SnO 2 (FTO)/Tec7 glass substrate, which was subsequently annealed at 610 C in air. A series of compositions were characterized for photoactivity in a three-electrode scanning drop cell (SDC) ( Figure 1A) using toggled illumination chronoamperometry (CA) at 1.23 V versus RHE at 12 combinations of electrolyte and photon energy. A new series of as-synthesized compositions were used for each of the three electrolytes, which included pH 10 and pH 7 electrolytes with 0.01 M sodium sulfite (SLF) as well as pH 1 electrolyte with 0.1 M methanol (MET), as shown in Table S1. Each composition in each electrolyte was measured under 3.2, 2.7, 2.4, and 2.06 eV illumination ( Figure 1B). Sulfite and methanol served as sacrificial hole acceptors to characterize photoactivity without requiring the photoanode surfaces to support the OER. The results are summarized in Figure 1C using the EQE (Note S1) to adjust the measured photocurrent by the different illumination intensities (Tables S2 and S3). While all Ni x Sb 1-x oxide compositions exhibited photo activity in at least one of these 12 conditions, the compositions with x < 0.35 were only photoactive under 2.7 and 3.2 eV illumination. More Ni-rich compositions exhibited photoactivity at low photon energies with substantial sensitivity to composition, especially at pH 10, where increasing the nickel concentration from x = 0.35 to 0.5 resulted in a nearly 10-fold increase in EQE. With increasing x above 0.5, the EQE decreases slowly and then more dramatically, enabling the identification of the x = 0.5 composition as a primary target for further characterization as well as the x = 0.33 composition as a valuable point of comparison.
The selection of these compositions of interest is further supported by the phase behavior of the oxide composition spread. X-ray diffraction (XRD) measurements in Figure 2 revealed the presence of two crystalline phases. The rutile NiSb 2 O 6 structure exhibits appreciable intensity for x % 0.33, the formula unit value, with decreasing intensity as x increases such that the signal is near the detectability limit at x = 0.5. When x increases above 0.7 a weak NiO signal is observed with intensity increasing as x increases. Collectively the results indicate that the film contains an X-ray amorphous component for all compositions with x > 0.33, which corresponds to all compositions where photoactivity at 2.06 or 2.4 eV illumination was observed ( Figure 1C). This X-ray amorphous phase is in high phase purity from x = 0.5 to 0.6, which corresponds to the compositions where the highest photoactivity was observed for each photon energy. In our experience with metal oxide deposition, many compositions are X-ray amorphous as-deposited, but only the most refractory oxides resist crystallization upon annealing at 610 C. We have not found literature characterization of amorphous oxides in this composition region and herein refer to this phase as am-NiSbO z .
Computational and experimental assessment of durability While the lack of a structural model for am-NiSbO z limits the ability to computationally characterize its formation energy, we estimate the energetics of this phase by first calculating the Ni-Sb-O grand potential phase diagram ( Figure 3A). Using an oxygen chemical potential of À5.9 eV (½ O 2 from Materials Project [MP] plus À1 eV correction for experimental synthesis conditions relative to standard conditions) and considering all phases in the MP, the only stable ternary oxide phase occurs at a Ni composition of 0.33 (NiSb 2 O 6 , mp-505271). At the composition x = 0.5, a pyrochlore phase Ni 2 Sb 2 O 7 (mp-1190650) appears at 0.88 eV per metal atom above the solid-state free energy hull. This phase has been synthesized 45 but is not observed in this work. Since the exact formation energy of am-NiSbO z is unknown, we approximate that this phase lies on the free energy hull on the coexistence line between NiSb 2 O 6 and NiO ( Figure 3A), which is likely the lower limit of the true formation energy and serves as a non-arbitrary way to place this phase on the phase diagram. Under this approximation, the Pourbaix energetics of am-NiSbO z can be evaluated using the MP Pourbaix module, resulting in the Pourbaix diagram and corresponding map of Pourbaix decomposition energy (G pbx ) in Figure 3B. These bulk thermodynamics indicate that am-NiSbO z may self-passivate with a NiSb 2 O 6 layer in neutral to acidic OER conditions. In pH 10 electrolyte, am-NiSbO z is thermodynamically stable under OER conditions. The Pourbaix diagram of NiSb 2 O 6 is qualitatively similar ( Figure S1, Note S2), especially under mild alkaline conditions. We note that, if the true formation energy of am-NiSbO z corresponds to the above-hull energy in Figure 3A, the minimum G pbx of am-NiSbO z in Figure 3B would become this same above-hull energy due to the thermodynamic preference for decomposing into NiSb 2 O 6 and NiO, although for OER-relevant potentials in pH 10 the predicted lack of corrosion would remain unchanged.
At lower potentials, am-NiSbO z and NiSb 2 O 6 are predicted to undergo cathodic corrosion at potentials of 0.5, 0.35, and 0.3 V versus RHE for pH 1, 7, and 10, respectively ( Figure 3B). This prediction is consistent with the experimental observation of dark cathodic current for potentials below 0.4, 0.1, and 0.2 V versus RHE for pH 1, 7, and 10, respectively, as revealed by toggled illumination voltage sweeps under 3.2 eV illumination ( Figure S2). The final anodic sweep for each electrolyte is shown in Figure 4A. While a dark cathodic current obscures the observation of the turn-on potential for anodic photocurrent, the pH 7 electrolyte offers the largest potential window without substantial dark current, where the turn-on potential is near or below 0.1 V versus RHE ( Figure 4A). In pH 10, where the highest EQE was observed in Figure 1C, the photocurrent versus potential is mostly convex in the potential region with near-zero dark current ( Figure 4A), suggesting a fill factor in excess of 0.5, which is promising for a photoanode that has yet to undergo optimization. The substantial current transients upon illumination toggling illumination indicate the persistence of surface recombination despite the use of a sacrificial hole acceptor, suggesting that further improvements in photoactivity may be achieved with optimized surface treatments.
Additional characterization of photoanode stability in this electrolyte (pH 10) was performed in an electrochemical recirculation flow cell ( Figure 4B, insert) with a bipolar membrane separating the anolyte and catholyte to enable the photoanode to equilibrate with dissolved metals concentrations in the electrolyte. The corresponding results ( Figure 4B) demonstrate substantial photoanodic current at 1.23 V versus RHE throughout the 30-min measurement with near-zero dark current measured every 5 min. Inductively coupled plasma mass spectroscopy (ICP-MS) analysis of three-electrolyte aliquots from this experiment shows a very low Sb concentration on the order of 1 nmol L À1 , with a larger, but still low Ni concentration, which increases from approximately 9 to 20 nmol L À1 throughout the measurement. These results are consistent with the computed Pourbaix diagram ( Figure 3B), which indicates that critical dissolved metals concentrations for the onset of corrosion are below 1 mmol L À1 . After an initial transient in photocurrent, the slowly increasing photocurrent over the duration of the measurement coincides with a continually increasing dissolved Ni concentration. To ensure that photostability was not contingent on the presence of a sacrificial hole acceptor, this measurement was done without SLF. We note that the average Ni corrosion rate of 0.37 nmol L À1 min À1 corresponds to an anodic corrosion current density of n 3 1.4 3 10 À4 mA cm À2 assuming an anodic corrosion process with n e À per corroded Ni atom. This corrosion accounts for n 3 0.2% of the anodic charge passed during the experiment. Under OER conditions, Ni 2+ is the presumed aqueous species, which implicates a corrosion reaction with n = 2, e.g., NiO / Ni 2+ + 2 e À + 0.5 O 2 . We ignore the contribution of Sb corrosion to the current density since its dissolved metals concentration is more than 103 lower than that of Ni. Assuming that the only anodic process other than corrosion is the OER, the corresponding OER Faradaic efficiency is 99.6%.

A B
Given that corrosion is an insignificant contributor to the photocurrent in Figure 4B, the two most likely explanations for the rise in photocurrent over the 30-min measurement are a continued exchange of surface and dissolved Ni that increases OER activity compared with the initial surface, or improved carrier extraction due to the formation of a Sb-rich surface layer. The near-surface Ni composition was measured by X-ray photoelectron spectroscopy (XPS) (Figures S3 and S4) to be x = 0.46 after operation, which is commensurate with the Ni corrosion being higher than Sb and corresponds to a solid-state Sb enrichment by 4.3 nmol cm À2 . The corresponding thickness of a passivation layer with composition x = 0.46 compared with the bulk value of x = 0.5 is 8.8 nm. If the passivation layer was Ni-free, i.e., similar to Sb 2 O 5 , the thickness corresponding to the differential corrosion would be 3.7 nm. These nominal passivation layer thicknesses assume a planar thin film, and the increase in specific surface area from any non-planar structure would correspond to a proportional decrease in the passivation layer thickness. For example, with a roughness factor of 10, the passivation layer may be as thin as 0.37 nm. Figure 1 shows lower EQE at pH 1 and pH 7, coinciding with the indication from the Pourbaix diagram of Figure 3B that a greater extent of Ni corrosion will be required to enable passivation by NiSb 2 O 6 . The passivation layer thickness is likely thicker under those conditions, which may underlie the decrease in photoactivity.

Structural and electronic characterization
Continued characterization of the am-NiSbO z photoanode proceeded with transmission electron microscopy (TEM). Cross-section TEM images are shown in Figures 5A and 5B for am-NiSbO z samples, one in the as-synthesized condition as well as the sample that underwent the 30-min stability test in Figure 4B. The films exhibit an angled columnar structure that likely arises from the off-axis deposition in the combinatorial synthesis. A comparison of these images with the corresponding elemental maps ( Figure S5) reveals that the majority of each film has x near 0.5, while the columnar growth appears to have caused some lateral composition inhomogeneity due to shadowing of the depositing flux of Ni and Sb, whose deposition sources were located on opposing sides of the substrate to achieve the desired composition spread. The Sb-rich regions have a distinct appearance in the images of Figures 5A and 5B, motivating selected area electron diffraction (SAED) characterization of these regions as well as a region that represents the majority of the film. The results in Figures 5C and 5D demonstrate that the Sb-rich minority regions contain crystalline NiSb 2 O 6 , while the majority region has no discernable crystallinity ( Figure S6), in excellent agreement with the XRD analysis of Figure 2 where the persistence of a small NiSb 2 O 6 signal up to x = 0.5 is due to small crystallites on the sides of the pillars facing the Sb deposition source. Given that the NiSb 2 O 6 ll OPEN ACCESS sample at x = 0.33 has low photoactivity compared with the x = 0.5 region, we continue to conclude that the photoactivity arises primarily from the amorphous (am-NiSbO z ) material (ignoring any heterojunction effects). The elemental mapping by energy-dispersive X-ray spectroscopy (EDX) in TEM ( Figure S5) indicates that the oxygen stoichiometry z is approximately 3.4, which is similar to the value of 3.5 corresponding to the above modeling of this amorphous phase as being in equilibrium with a combination of NiSb 2 O 6 and NiO.
The higher-resolution image in Figure 5E reveals a feather-like nanostructure. We are not aware of a previous observation of such a morphology in an amorphous film. To confirm that the overall phase behavior and photoactivity are not produced by the lateral composition homogeneities, we synthesized and characterized an x = 0.5 film deposited and annealed using the same conditions but with a Pt/Ti/SiO 2 /Si substrate, which has much lower surface roughness than the F-doped SnO 2 /glass substrate. The results are shown in Figure S7 and support the existence of a nanostructured, amorphous am-NiSbO z photoanode with a broad spectral response.
To characterize the charge carrier dynamics of the photoanode, we proceeded with Hall effect measurements of the x = 0.33 and x = 0.5 samples. The 300-nm-thick films of these compositions were deposited on SiO 2 /Si wafers and post-annealed at 610 C, then cleaved into 1 3 1 cm samples with evaporated Au/Ti metal contacts. The resistivity of the x = 0.33 sample was measured to be 0.57 G 0.01 U-cm and from the negative Hall voltage we determined an electron mobility of approximately 0.09 G 0.01 cm 2 V À1 s À1 . The x = 0.5 sample exhibited a resistance above 100 MU, which prohibited reliable characterization of the resistivity or mobility. Low mobilities of only ca. 0.2 cm 2 V À1 s À1 have been measured in other high-performing oxide photoanodes, such as BiVO 4 , which is attributed to polaron formation. 46 Whether either the 0.33 or 0.5 compositions can meet or surpass such values with improvements in microstructure and density remains to be seen. In particular, the TEM analysis of Figures 5 and S5 indicates a high density of columnar voids that preclude lateral or in-plane conduction. While the deposition of x = 0.5 on a flatter substrate exhibits less nanostructure ( Figure S7), columnar voids are still present. As a result, this morphology is likely inflating the measured resistance compared with the intrinsic resistivity of the material, motivating future deposition of compact films to facilitate electronic characterization.
To characterize the relationship between am-NiSbO z and NiSb 2 O 6 , we conducted X-ray absorption near-edge spectroscopy (XANES), as shown in Figures 6A and  6B, where the signals from the Sb K-edge are nearly identical with edge positions matching that of Sb 2 O 5 . The signals from the Ni K-edge are similar for these two samples with edge positions between those of the NiO and NiOOH reference materials. The x = 0.33 (NiSb 2 O 6 ) sample exhibits a higher edge energy, suggesting that the Ni is slightly more oxidized in this material. The high-resolution core-level Ni 2p 3/2 XPS spectra of the same two samples in Figure S4 are also quite similar.

OPEN ACCESS
even though the X-ray absorption spectroscopy (XAS) data suggest that the bulk Ni is more oxidized in the NiSb 2 O 6 sample.
To characterize the environments around Ni for am-NiSbO z and NiSb 2 O 6 , we obtained Ni-K extended X-ray absorption fine structure (EXAFS) signals. The fitting of the first shell in the EXAFS signal ( Figures S8 and S9) was performed using the Ni-O scattering path templated by the computational NiSb 2 O 6 structure (mp-505271). The fitted parameters (Table S4) 34 While the am-NiSbO z shows substantially higher EQE than the NiSb 2 O 6 photoanode, it is worth noting that the optical absorptivity of am-NiSbO z is approximately 1.53 lower over the entire visible range (1.5-3.0 eV). The increasing photoactivity with increasing Ni is not explained by optical absorption, which is contrary to the general behavior that the high defect density of amorphous materials contributes to substantial optical absorption.
To further characterize the electronic structure of am-NiSbO z , we consider the connection to Ni 2 Sb 2 O 7 , which has approximately the same cation composition as the samples with the highest EQE in Figure 1C, and the connection to NiSb 2 O 6 established by the XAS and band energetics experiments (Figures 6 and 7). For each of these crystalline structures, we performed density functional theory (DFT) calculations using the Vienna Ab initio Simulation Package (VASP) 47,48 with Hubbard U corrections 49,50 applied to the Ni d-states, revealing sensitivities with respect to the band-gap and formation energies (Figures S12 and S13). Both ferromagnetic (FM) and antiferromagnetic (AFM) configurations of spins centered on the Ni sites were considered. In particular, AFM orderings were found to be more stable than the FM ordering by 213 meV per metal atom for NiSb 2 O 6 and by 636 meV per metal cation for Ni 2 Sb 2 O 7 . For NiSb 2 O 6 , both FM and AFM orderings had very similar lattice parameters and adopted a tetragonal crystal structure. For Ni 2 Sb 2 O 7 , however, the FM ordering corresponded to a cubic structure, while the AFM ordering produces a small tetragonal distortion with two of the lattice parameters expanding by 0.36% and the third contracting by 0.33%. Given the default MP FM ordering used for Figure 3, this strong magnetic-ordering dependence of Ni 2 Sb 2 O 7 formation energy indicates that its above-hull energy in Figure 3A may be over-estimated. Furthermore, Ni 2 Sb 2 O 7 with FM ordering was found to be metallic in our DFT+U calculations, underscoring the importance of carefully considering magnetism in these materials (Table S5). The element projected density of states of NiSb 2 O 6 calculated with DFT-PBE+U ( Figure 8A) leads to a band gap of 1.03 eV, a valence band of predominantly oxygen character, and a conduction band with a very low density of states of predominantly antimony character.
To improve the treatment of short-range exchange and correlation effects and ascertain the associated sensitivity toward the electronic structure, NiSb 2 O 6 in the AFM state was also evaluated using the hybrid functional of Heyd, Scuseria, and Ernzerhof ll OPEN ACCESS (HSE06), 51,52 which can provide an improved description of band gaps and electronic structure. The resulting density of states ( Figure 8B) has a larger band gap of 1.96 eV that well matches the experimental results of Figure 7. The conduction band character and density are consistent with the HSE06 calculation, while the valence band has approximately equal contributions from nickel and oxygen, as opposed to the oxygen-rich character observed in the PBE+U calculation.
The PBE+U band structure of NiSb 2 O 6 ( Figure 8D) reveals that the conduction band edge is highly dispersive. Furthermore, the density of states of Ni 2 Sb 2 O 7 appears to have a very similar band edge character to that of NiSb 2 O 6 ( Figure 8C). Given the similarity of these electronic structures irrespective of Ni:Sb stoichiometry and level of first principles theory, which is corroborated by the similarity of band energetics of am-NiS-bO z and NiSb 2 O 6 in Figure 7, we hypothesize that Ni-Sb oxides may generally host highly dispersive bands near the conduction band edge, raising the possibility of facile band transport of electrons upon optimization of materials synthesis.

DISCUSSION
It is worth noting that the vast majority of metal oxide photoanodes are based on crystalline phases, while the most effective passivation layers are amorphous due to their lack of crystal anisotropies and defects such as grain boundaries. 53,54 Amorphous semiconductors are normally considered to have poor charge carrier transport due to chemical disorder and low optoelectronic efficiency compared with crystalline semiconductors. Since the discovery of enhanced photocatalysis of socalled black TiO 2 , 55 there is a growing interest in exploring amorphous semiconductors for solar cell, photoelectrocatalytic, and photocatalytic applications. 56 For example, varying composition in amorphous metal oxide films to tune band gap and light absorption; 57 exploring amorphous mixed-metal oxides as photocatalysts to enhance water oxidation kinetics; 58 forming crystal-amorphous junction semiconductors to improve charge separation and transportation; 59,60 nano-structuring amorphous materials to increase specific surface and expose more active sites; 61 developing amorphous metal oxide on crystalline semiconductor as a protection layer to prevent corrosion. 62,63 However, there are few reports on amorphous semiconductors used as photoanodes for solar fuel generation.
Amorphous TiO 2 has been shown to exhibit some photoactivity 64 but has been more effectively utilized as a capping layer atop traditional n-TiO 2 absorbers. A sol-gelderived CuSnO 3 amorphous thin film with band gap 2.3 eV had a low turn-on/onset potential of 0.3 V versus RHE for OER in pH 6.5, but the photocurrent is unstable with a 50% drop within 25 min. 65 Photoanodes made from amorphous (Zn, Si) 2 GeO 4 nanowire arrays exhibited a better efficiency and stability in PEC water splitting compared with the crystalline Zn 2 GeO 4 . 66 In this work, the combinatorial study of photoactivity ( Figure 1) reveals that the amorphous material has a higher EQE at and above 2.8 eV than the neighboring composition regions, and perhaps even more striking is that the amorphous photoanode exhibits photoresponse down to a photon energy of approximately 2 eV. Given the chemical and electronic similarities established between am-NiSbO z and NiSb 2 O 6 , the visible photoresponse of am-NiSbO z is commensurate with the electronic structure characterization in Figures 7 and 8, where the band diagrams also suggest that the electronic structure of nickel antimonates may generally be conductive to band transport. However, these observations do not reveal why am-NiSbO z outperforms NiSb 2 O 6 , which can be addressed in future work by disambiguating whether the increased concentration of Ni +2 or the lack of crystallinity are most central to photoactivity. Subsequent investigation of the corresponding change in the electronic structure and/or the formation energy and electronic contribution of defects will facilitate understanding of amorphous photoanodes as well as directions for further optimization of am-NiSbO z .
Among amorphous semiconductors, am-NiSbO z is especially promising due to its thermodynamic Pourbaix stability, as evidenced by the low dissolved metals concentrations under operation, approximately 1 and 20 nmol L À1 for Sb and Ni, respectively. Table S6 compares dissolved metals concentrations of am-NiSbO z with other photoanodes, e.g., WO 3 , a-SnWO 4 , and BiVO 4 . While dissolved metals concentrations are rarely reported alongside photocatalytic activity and the variation in PEC protocols hinders comparisons among literature reports, we established the comparisons of Table S6 by considering reports from PEC operation at 1.23 V versus RHE for 20-240 min. The PEC stability of photoanodes is strongly pH and electrolyte dependent per the Pourbaix thermodynamics of the respective photocatalyst material. WO 3 is Pourbaix-stable under OER conditions in acid and has enabled operational stability at low pH both as a photoanode and as a self-passivation layer for a-SnWO 4 . 25 WO 3 in 0.5 M H 2 SO 4 exhibits the lowest dissolved metals concentration observed for these photoanodes (0.87 mmol L À1 ), 67 which is 403 higher than our measurement of am-NiSbO z in pH 10 (0.021 mmol L À1 ). Efforts to mitigate BiVO 4 corrosion revealed relatively low dissolution in low-concentration phosphate electrolytes (pH 6.8, 0.1 M KPi). 68 After 60 min operation, the dissolved metals concentrations were negligible for Bi and 0.08 mmol L À1 for V, which is 43 higher than that demonstrated by am-NiSbO z in this work. The low dissolved metals concentration observed in Figure 4B is not only an important demonstration of stability but is also imperative for long-term durability.
Operational stability at any dissolved metals concentration can only be maintained if the dissolved metals remain in the anolyte (the electrolyte in electrochemical communication with the photoanode). For long-term operation, a finite rate of metal precipitation can be expected within the electrolyte, particularly a polymer electrolyte membrane, and/or on the opposing electrode, which has been documented as a prominent device degradation pathway. 69 The flux of dissolved metals from the anolyte will also instigate further corrosion of the photoanode. These considerations highlight the importance of developing photoanodes with favorable Pourbaix energetics, as opposed to dosing the electrolyte with large dissolved metals concentrations. The development of photoanodes that operate with minimal dissolved metals concentration can address the durability challenges of solar fuel generators via codesigned mutual compatibility with other device components. 9 The combination of computational Pourbaix screening and high-throughput experimentation is particularly well suited for deploying this codesign research strategy, where the former identifies composition systems of interest and the latter identifies the performant materials. In the present case, the performant material is a previously unknown amorphous NiSbO z photoanode, where z is estimated to be 3.5 corresponding to formal oxidation states of Ni +2 and Sb +5 . Electrochemical analyses demonstrate that am-NiS-bO z exhibits a promising combination of stability, broad spectral response, and a turn-on potential below 0.4 V versus RHE. The computational Pourbaix stability is mirrored by a very low dissolved metals concentration during operation in pH 10 electrolyte, which is pertinent not only to device-level durability but also to the maintenance of the semiconductor-liquid junction given that operational stability is not reliant on a thick protection layer. The additional inherent lack of crystal anisotropies in this amorphous photoanode opens new opportunities for scalable synthesis. Collectively these results highlight the opportunity for optimizing the photoactivity of am-NiSbO z and more generally for developing amorphous photoelectrodes to address the activity-durability-scalability challenges in solar fuel generation.

EXPERIMENTAL PROCEDURES
Resource availability Lead contact Further information and requests for resources should be directed to the lead contact, John M. Gregoire (gregoire@caltech.edu).

Materials availability
This study did not generate new unique reagents.  1 The deposition proceeded at room temperature without intentional heating in a mixed O 2 (0.9 mTorr) and Ar (5.1 mTorr) atmosphere with 10 À8 Torr base pressure, followed by a post-deposition anneal in a box oven at 610 C in air for 1 h. The non-confocal geometry of sputter sources provided a continuous composition gradient spanning a 60-70 at % range in the concentration of each cation element across the substrate. A duplicate Ni x Sb 1-x O z composition library was fabricated on a 100-mm-diameter Pt/Ti/SiO 2 /Si wafer under the same deposition and annealing process.

X-ray fluorescence
The metal oxide compositions were characterized by X-ray fluorescence (XRF) using an EDAX Orbis Micro-XRF system to obtain Ni, Sb metal contents and values of x = Ni/(Ni + Sb) with ca. 1 at % relative uncertainty. The sensitivity factor for Ni and Sb was calibrated by commercial XRF calibration standards (MICROMATTER). The oxygen signal and thus stoichiometry was not detectable by the XRF experiment.
Assuming the bulk density of oxides of each element, the XRF measurements also provided an estimate of film thickness.

XRD
The bulk crystal structure and phase distribution of the composition library were determined by XRD using a Bruker DISCOVER D8 diffractometer with Cu Ka radiation from a Bruker Ims source. The X-ray spot size was limited to a 2 mm length scale, over which the composition was constant within approximately 1-2 at %. Diffraction images were collected using a two-dimensional VÅ NTEC-500 detector and integrated into one-dimensional patterns using DIFFRAC.SUITE EVA software. The crystal structures present in each sample were identified by matching the XRD patterns with entries in the International Crystallography Diffraction Database in the EVA.

XPS
XPS spectra were measured to determine the near-surface chemistry using a Kratos Axis Ultra Nova instrument with a base pressure <10 À9 Torr. A monochromatic Al Ka (1,486.6 eV) source with a power of 150 W was used for all wide, valence band, and core-level detail scans, whereas 10 W was used for work function measurements. The collected spectra were calibrated to the carbon 1s peak of 284.8 eV. Data were analyzed using CasaXPS. The valence band edge is the region close to zero binding energy (maximum kinetic energy). The VBM was determined by the intersection of a linear fit to the background and the onset of the valence band. The work function of the sample was determined by measuring the secondary electron cutoff of the sample while biasing the sample with a stabilized voltage source (+40 V) to overcome the work function of the analyzer. The spectra were then corrected by the applied bias, and the position of the secondary electron cutoff edge was determined by the interception of a linear fit to the edge. In an intensity versus kinetic energy plot, this value at the intersection will directly yield the value of the work function. For the binding energy scale, the value must be converted by the photon energy. A Shirley background model was applied to quantify the core-level intensities, which were corrected by the analyzer transmission function and relative sensitivity factors to obtain corrected peak intensities and atomic ratios.

UV-vis spectroscopy
The optical properties of the composition libraries were characterized using a custom-built on-the-fly scanning UV-vis dual-sphere spectrometer to record transmittance (T) and total reflectance (R) simultaneously. 2 The fractional T and R spectra were used to calculate the spectral absorption coefficient (a) as a = -ln [T*(1ÀR) À1 ]/t, where t is film thickness. The optical spectra were acquired on a 2-mm-grid of 1,521 positions on the composition library and automatically processed for Tauc analysis. 3

TEM and EDX
TEM experiments were carried out in an FEI Tecnai Osiris FEG/TEM operated at 200 kV in bright-field and high-resolution TEM mode. SAED patterns were taken using the same machine. EDX elemental mapping was acquired using Bruker Quantax. This characterization was performed using a Eurofins EAG Precision TEM in Santa Clara, CA.

Mobility measurements
Electronic transport measurements were performed in a van der Pauw geometry. Linear regression fitting of current-voltage sweeps collected between all adjacent contacts resulted in a minimum R 2 value of 0.9999, confirming ohmic contacts. The final excitation current of 10 mA chosen for Hall analysis was within this ohmic range. Hall effect measurements employed the FastHall approach using a LakeShore cryotronics probe station equipped with a 1 Tesla permanent magnet. This method is more amenable to low mobility films than conventional DC Hall effect measurements. In the 100 measurements performed, the Hall voltage was found to be negative 90 times. To further verify the measurement's reliability, the field was intentionally reversed and, in this case, we measured an equal-but-opposite signal (similar mobility, but majority positive voltages).

Photoelectrochemistry
Combinatorial PEC measurements were conducted in our previously reported fibercoupled SDC instrumentation with a Gamry G 300 potentiostat controlled by custom automation software. 4 Experiments were carried out in three 1 atm O 2 -saturated aqueous electrolytes (see Table S1) with a sacrificial hole acceptor to increase the hole transfer kinetics at the film/electrolyte interface.

ICP-MS
ICP-MS using a Thermo Fisher Scientific iCAP RQ instrument was used to determine the concentration of dissolved Ni and Sb metals in the electrolyte used for electrochemistry at different durations throughout the PEC measurement.

Pourbaix calculations
Energetics of the Ni-Sb-O phases were obtained from the Materials Project Database. 71 The Pourbaix diagram was generated using the Pourbaix module in Pymatgen. 26,72 DFT calculations First principles DFT calculations were performed with a plane-wave basis and projector-augmented wave potentials 73 using the VASP. 47,48 These calculations were performed in the generalized gradient approximation as implemented by Perdew, Burke, and Ernzerhof (PBE) 49 using additional Hubbard U correction terms in the formalism developed by Dudarev (PBE+U) 50 to account approximately for on-site correlation. A U value of 6.2 eV was used for the Ni ions, consistent with the standard Hubbard U values used in the MP. The plane wave energy cutoff was 600 eV. A 6 3 6 3 2 k grid was used for NiSb 2 O 6 and a 6 3 6 3 6 k grid was used for Ni 2 Sb 2 O 7 . Three collinear magnetic orderings for the Ni ions were considered for both materials, FM ordering and two AFM orderings each. For NiSb 2 O 6 the Ni octahedra are not connected. Planes of disconnected Ni octahedra in the ab plane are layered with each plane offset by half a lattice parameter, making triangles with an in-plane separation of $4.7 Å and out-of-plane separations $5.7 Å apart. The most stable AFM ordering had FM ordering in-plane and AFM ordering between neighboring planes. For Ni 2 Sb 2 O 7 the magnetic ordering is substantially more complex, as previous experimental evidence indicates that this material is a frustrated antiferromagnet. 53 Investigation of various non-collinear magnetic orderings would be more appropriate for identifying the ground-state magnetic ordering of Ni 2 Sb 2 O 7 , but that is beyond the scope of this work. Ni sites are clustered in groups ll OPEN ACCESS of four with edge-sharing polyhedra and the lowest energy collinear AFM magnetic ordering investigated consisted of two of the polyhedra in each cluster with parallel magnetic moments and the other two with their magnetic moments in the opposite direction. Structural relaxations were performed for both materials for both FM and AFM collinear magnetic orderings with forces converged to <0.001 eV/Å . The density of states and band structure calculations were performed for both NiSb 2 O 6 and Ni 2 Sb 2 O 7 using PBE+U with 10 3 10 3 2 and 10 3 10 3 10 k-point grids, respectively. The density of states of NiSb 2 O 6 was also calculated using HSE06 51,52 with an energy cutoff of 500 eV, a 6 3 6 3 2 k-point grid, and the structure obtained from relaxation with PBE+U.