High throughput discovery of enhanced visible photoactivity in Fe–Cr vanadate solar fuels photoanodes

Metal oxide solar absorbers are well suited for photoelectrochemical applications where requisite properties include stability in highly oxidizing environments, in addition to solar energy conversion. Metal vanadates are of particular interest due to their relatively low band gap energies compared to traditional, wide-gap photocatalysts. Concerted efforts on BiVO4-based photoanodes have revealed multiple avenues for improving the solar conversion efficiencies for photon energies above 2.5 eV but have not addressed the ultimate performance limitations from the undesirably high band gap energy. Fe and Cr vanadates have a lower band gap and thus a higher potential solar conversion efficiency, although to-date the absorbed 2–2.5 eV photons are not effectively converted to the desired anodic photocurrent. By using combinatorial synthesis and high throughput screening, we demonstrate that cation substitutions with the monoclinic MVO4 phase (M = Cr, Fe) improves the utilization of photons in this energy range. Given the portfolio of photoanode improvement techniques available, we suggest optimization of (Cr0.5Fe0.5)VO4-based photoanodes as a promising path for enable solar fuel technologies.


Introduction
Using solar energy for renewable generation of fuels requires a photoelectrocatalyst that efficiently utilizes the solar resource to power the oxygen evolution reaction, which liberates protons and electronics from water for use in fuel-forming reactions [1][2][3]. Metal vanadates have been the most intensely studied family of visible-gap metal oxide photoanodes, with the monoclinic BiVO 4 being the center of attention over the past decade [4][5][6][7][8]. While BiVO 4 has served as an exemplar material for advancing our understanding of complex photoanodes and establishing mechanisms to improve photoanode performance, utilization of solar energy is intrinsically limited by the ⩾2.4 eV band gap, motivating transfer of the lessons learned with BiVO 4 -based photoanodes to other metal vanadates with a lower band gap energy [9][10][11]. Perhaps the most valuable materials design tool emerging from the BiVO 4 work is the ability to dramatically improve conductivity and more generally carrier transport with substitutions on the cation sublattice. Among the dozens of metal vanadates with photoactivity below 2.4 eV, the most prominent systems are the Fe-and Cr-based vanadates wherein photoactive phases with band gaps below 2.4 eV include CrVO 4 -orth, CrVO 4 -mon, and Cr 2 V 4 O 13 [12] as well as FeV 2 O 4 , Fe 2 VO 4 [13], Fe 2 V 4 O 13 [14], FeV 3 O 8 [15], and FeVO 4 -tri [16][17][18][19][20][21][22][23][24][25][26]. Of these, FeVO 4 with band gap in the range 2.0-2.3 eV has received the most attention and is identified as a locally optimal composition via combinatorial exploration of the Fe-V oxide system [16,21]. Akin to α-Fe 2 O 3 , photoelectrochemical (PEC) performance of triclinic FeVO 4 is limited by its low carrier mobility and short hole diffusion length [22]. Initial work on heterojunction and coated FeVO 4 provide routes for photoelectrode optimization [25,26], which will be most effective upon optimization of the carrier transport and more generally the photoactivity of the FeVO 4 -based photoanodes.
The cation-substitution strategy for improving the photoactivity of FeVO 4 was initially demonstrated via doping with W [18] or Mo [22] and was more explored using combinatorial synthesis and screening methods by Nguyen et al [27]. Cation substituents explored in that work include Zn, Ni, Cr, Mo, and W with cation concentrations up to 10%. The best photoactivity in phosphate-buffered, near-neutral pH electrolyte with broadband illumination was observed with 7% Cr. Cr substitutions into FeVO 4 as well as the converse Fe substitutions into CrVO 4 have been shown to exhibit composition-tuned band edge positions [28]. The complex composition dependence on electronic structure is emblematic of the broader family of M +3 VO 4 orthovanadates [29], which exhibit complex composition-dependent crystal structure, cation ordering, and magnetic ordering with different transition metals and mixtures thereof on the M +3 sublattice [30][31][32][33]. Mixed phase iron bismuth vanadate compound has been demonstrated to maintain good mobility of BiVO 4 while extending visible light absorption by mixing FeVO 4 with BiVO 4 [22,34]. Unlike M = (Bi, Fe), complete miscibility on the M +3 sublattice has been observed for M = (Cr, Fe), with cation ordering at the 1:1 composition resulting in the quaternary oxide CrFe(VO) 4 [35]. Similar miscibility has been observed in the M 2 V 4 O 13 system [36], motivating our exploration of Cr-Fe-V oxide photoanodes with substantial amounts of both Cr and Fe, as opposed to the more dilute-substitutions that were explored previously [27]. Since the path for outperforming BiVO 4 -based materials commences with establishing substantial photoactivity between 2 and 2.4 eV, we investigate the spectrally-resolved compositional variation in photoactivity, demonstrating that compositions near (Cr 0.5 Fe 0.5 )VO 4 are optimal for photoactivity at both 2.1 and 2.4 eV, motiving future study of these quaternary oxides to efficiently utilize the solar resource and advance solar fuels technologies.

Composition library synthesis
The continuous composition spread of Cr-Fe-V was synthesized by reactive co-sputtering in a custom-designed combinatorial sputtering system [37] at room temperature and followed by post-position anneal in air at 650 • C for 1 h. The library was deposited in a mixed atmosphere of O 2 (0.6 mTorr) and Ar (5.4 mTorr) using Cr, Fe, and V sources placed 120 • apart with respect to the substrate plane of the 100 mm-diameter Pyrex glass with a conductive SnO 2 :F (FTO) coating, which served as the electrical back contact to the thin film photoanodes. The deposition proceeded for 10 h with the radio-frequency (RF) powers on Cr, Fe, and V sources set to 42, 62, and 150 W, respectively, to obtain the desired composition spread. Characterization data for this composition library have been aggregated in [38].

Composition and structure characterization
The bulk metal compositions were characterized by x-ray fluorescence (XRF) measurements using an EDAX Orbis Micro-XRF system with an x-ray beam approximately 2 mm in diameter. The Cr K, Fe K, and V K XRF peak intensities were extracted from the Orbis software and converted to normalized compositions using the sensitivity factor for each element calibrated by commercial XRF calibration standards (MicromatterTM).
The bulk crystal structure and phase distribution of composition library was determined by x-ray diffraction (XRD) measurements. XRD was acquired using a custom high throughput setup [39] incorporated into the bending-magnet beamline 1-5 of the Stanford Synchrotron Radiation Light Source at SLAC National Accelerator Laboratory. The characterization employed a monochromated 12.7 keV source in reflection scattering geometry with a 2D image detector. Diffraction images were processed into one-dimensional XRD patterns using WxDiff software with calibration from a LaB 6 powder standard, and further analyzed in the Bruker eva software.

Optical characterization
Optical properties of composition library were characterized using a custom-built, on-the-fly high-throughput scanning spectroscopy instrument [40]. The dual-sphere spectrometer recorded both transmittance (T) and reflectance (R) simultaneously at each sample, which were used to calculate the spectral absorption coefficient (α) up to a factor of film thickness (τ ): ατ = ln[(1 − R)/T]. The molar absorption coefficient was calculated by making τ the molar thickness from the XRF-measured molar concentration of Cr, Fe, and V. For select compositions, the traditional thickness-normalized absorption coefficient was calculated by adjusting the molar absorption coefficient by the molar density of the phase identified by XRD.

Scanning droplet cell (SDC) photoelectrochemistry (PEC) measurements
PEC characterization was carried out in a custom-designed fiber-coupled front-side illumination SDC instrumentation with a Gamry G 300 potentiostat controlled by custom automation software [41]. PEC photoanode screening protocol on the pseudo-ternary composition spread were reported previously [42].
Measurements were performed in an aqueous 1 atm O 2 -saturated borate electrolyte (pH 9) with 0.01 M sodium sulfite as sacrificial hole acceptor to increase the hole transfer kinetics at the film/electrolyte interface. A grid of 218 compositions was characterized with sequential chronoamperometries (CAs) under a series of light emitting diode (LED) illumination sources (3.2, 2.7, 2.4, and 2.1 eV) from which photocurrent and thus the spectral external quantum efficiency (EQE, also called incident photon-to-current efficiency) is calculated, as well as a toggled-illumination cyclic voltammetry (CV) under 3.2 eV illumination. At each sample, CAs were measured at 1.23 V versus reversible hydrogen electrode (RHE) with LEDs toggled with 0.5 s on/off illumination: 3.2 eV for 15 s, and 2.7, 2.4, and 2.1 eV all for 4 s each, and following CV from 1.23 V to 0.73 V and back to 1.53 V vs RHE (rate of 0.02 V s −1 ) under 3.2 eV illumination with light toggling 2 s on and 1 s off.

Composition interpolation
The XRF and optical measurements were performed on the same set of 1521 samples across the library on a 2 mm grid. The XRD and PEC measurements were carried out on a coarser grid of 302 and 218 samples, respectively, and those sample compositions were calculated using linear interpolation in the Cartesian library position space.

X-ray photoelectron spectroscopy (XPS)
The near-surface composition of select samples was measured by XPS using a Krato Axis Ultra Nova spectrometer with a base pressure <10 −9 Torr. The x-ray source is a monochromatic Al Kα source at 1486.6 eV and operated at 150 W. The binding energy was calibrated to the C 1s peak position at 285 eV. Elemental quantification was performed in the CasaXPS software using a Shirley background fitting of the Cr 2p, Fe 2p, and V 2p 3/2 , and O 1s peaks in the survey scans with relative sensitivity factors of 2.427, 2.957, 1.411, and 0.711, respectively.

Phase behavior of Cr-Fe-V oxide system
The utilization of combinatorial magnetron sputtering to explore (photo)electrocatalysts for solar fuel applications has greatly accelerated the materials discovery and development, especially for discovering cation off-stoichiometry compounds and complex mixed phase metal oxides [12,43,44]. To explore photoactivity of Cr-Fe-V oxides, the pseudo-ternary composition spread was prepared by reactive co-sputtering targeted at center composition of Cr:Fe:V = 1:1:2 followed by annealing at 650 • C in air. The non-confocal geometry of sputtering sources provided a continuous composition gradient spanning a 60-70 at.% range in the concentration of each cation element across the 100 mm-diameter FTO glass substrate. While oxygen is incorporated into the film during deposition, the oxygen stoichiometry of the photoanode samples is expected to be governed by the crystallization into metal oxide phases during the 650 • C anneal. For composition-based analyses, each photoanode sample is represented by its cation compositions measured by XRF.
Since the crystal structure of photoanode materials has a strong impact on the optical and PEC performance, a collection of 302 XRD patterns was used to identify the crystalline structure and distribution of each phase in the Cr-Fe-V pseudo-ternary composition spread. Unlike Bi-Cu-V oxide's complex phase diagram (containing 13 different crystal phases in 19 unique combinations), which requires employing deep reasoning networks to solve the underlying phase behavior [45], the phase behavior of the Cr-Fe-V oxide system is relatively simple due to the miscibility of Cr and Fe in metal vanadates [35,36]. The XRD patterns were analyzed by matching entries in the ICDD database, resulting in the identification of five primary phase structures summarized in table S1. Manual visualization of the XRD patterns revealed substantial peak shifting indicating the significant alloying or solid solution between Fe and Cr in a wide composition range. Thus the continuous solid solution phases are labeled as M 2 V 4 O 13 , MVO 4 -mon, and M 2 O 3 , where M = (Cr, Fe). Figure 1 shows the composition plots of each identified crystal phase with representative XRD patterns compared to ICDD library entries (see figure S1 for an alternate visualization) and resulting unique nine combinations of phase field labeled from A to I. The majority phase in the Cr-Fe-V oxide system is the monoclinic MVO 4 , which exists in 86% of measured samples across V concentration of 0.2-0.7 and Fe, Cr concentration of 0.1-0.7 (seven out of nine phase fields). The 100% pure phase appears in the narrow V composition region of 0.5-0.54 (phase field E). In the neighboring phase field at lower and higher V concentration, the monoclinic MVO 4 phase is mixed with M 2 O 3 (phase field F-H) and M 2 V 4 O 13 (phase field B-C), respectively. At very small V concentration below 0.3, M 2 O 3 becomes the majority phase, and mixes with both CrVO 4 -mon and CrVO 4 -orth phases at Cr-rich side (Cr > 0.55, phase field G), while mixing with FeVO 4 -tri at Fe-rich side (Fe > 0.6, phase field I).  To illustrate the continuous solid solution in the Cr-Fe-V oxide system across a wide range of compositions, we selected samples with V concentration of 0.54 and 0.24, respectively, to visualize in figures 2 and S2. At V concentration around 0.54 (phase field E) in figure 2(a), all samples exhibit monoclinic MVO 4 structure with Fe/Cr ranging from 0.28 to 5.1, coinciding with an increasing unit cell volume (decreasing 2-theta values) and a decreasing monoclinic distortion (decreases in the difference between, e.g. the 330 and 003 peaks). The systematic shifting of peak positions to higher d-spacing (lower 2-theta) with increasing Fe concentration also occurs in the neighboring phase fields: M 2 V 4 O 13 at higher V concentration (phase field A), and M 2 O 3 at lower V concentration (phase field F). Figure S2 illustrates the phase behavior at V concentration of 0.24, where MVO 4 -mon and M 2 O 3 are mixed with CrVO 4 -orth at Cr-rich side and with FeVO 4 -tri at Fe-rich side (across phase field G to F to I).

PEC characterization of Cr-Fe-V oxide system
Spectral EQE calculated using the photocurrent from CAs and incident LED light power was used to investigate the relationship between the PEC activity and the incident photon energy. The EQE at 1.23 V vs RHE for each of the four photon energies mapped in both library position space and composition space is shown in figure 3. While almost all Cr-Fe-V oxide compositions exhibited photoactivity under 3.2, 2.7, and 2.4 eV illumination, with decreasing photon energy the local EQE maximum moves from Fe-rich compositions at the edge of the composition library toward the center of the composition space. With 2.4 eV and 2.1 eV illumination, the top several performers all reside in phase field E from figure 1(b) where the monoclinic MVO 4 structure is the only phase observed by XRD. The global maximum in EQE at these photon energies is near Cr 0.23 Fe 0.24 V 0.53 , and the toggled-illumination CA and CV data for this sample are shown in figure S3. The XRF-derived ratio of V/M in the optimal MVO 4 samples is approximately 1.13, whose difference from the formula unit value of 1.0 is comparable to the uncertainty of the XRF measurement. The optimal films may indeed be V-rich, which is analogous to similar V excess in BiVO 4 resulting in high-performance, phase-pure photoanodes [46].

Optical property of Cr-Fe-V oxide system
Optical characterization of the Cr-Fe-V oxide composition library proceed with characterization of the composition-dependence of the spectral absorption coefficient, α. For each composition, absorption coefficient spectrum was obtained from measured transmittance (T) and reflectance (R), then molar absorption coefficients were averaged over three different ranges of incident photon energy and normalized with XRF-determined molar concentration of metals. Figure 4 shows the composition mapping of averaged molar absorption coefficients of 1521 compositions, which reveals a variety of compositional trends (library position mapping is shown in figure S4). Particularly distinctive compositions include the Cr-Fe composition line with V concentration between 0.5 and 0.55 where the monoclinic MVO 4 (phase field E)  was observed in figures 1 and 2, which exhibits relatively low absorption. Other compositions containing a mixture of MVO 4 with other phases, especially M 2 O 3 , exhibit much higher absorption, for example Cr 0.2 Fe 0.6 V 0.2 in phase field G that exhibits high absorption over the full spectral range.
The optical spectroscopy analysis is combined with the EQE data for the primary composition of interest, Cr 0.23 Fe 0.24 V 0.53 , in figure 5. The spectral transmittance, total reflectance, and absorption coefficient of this sample are shown in figure 5(a). Tauc analysis was used to characterize the band gap, resulting in approximate indirect-allowed and direct-allowed band gap energies of 2 and 2.6 eV, respectively. These values are similar to the band gaps we reported previously for FeVO 4 , Cr 2 V 4 O 13 , CrVO 4 -orth, and CrVO 4 -mon photoanodes [12]. The EQE is also plotted in figure 5(b) revealing photoactivity at 2.1 eV, just above the approximate indirect band gap energy. Given the uncertainty in the band gap energy and the relatively low EQE at 2.1 eV, these results are consistent with photoactivity onset from bandgap excitation, although the absorption and photoactivity may also involve mid-gap states, for example from point defects, whose importance is noted in the Discussion. Automated Tauc analysis to observe compositional trends in the approximate direct and indirect band gap energies are shown in figure S5.

Property variations in the FeVO 4 -CrVO 4 pseudo-binary space
X-ray photoelectron spectroscopy (XPS) was employed for characterization of the near-surface composition of eight MVO 4 photoanode samples, spanning Fe/Cr values from 0.34 to 4.1, after their PEC characterization ( figure S6). The positions of the Fe 2p, Cr 2p, V 2p, and O 1s peaks do not noticeably vary with Fe/Cr composition. Core level spectra of the transition metals are consistent with the nominal oxidation states Fe +3 , Cr +3 , and V +5 . The O 1s signal contains a high-binding energy peak near 532.4 eV that is presumed to Table 1. Composition characterization of MVO4 photoanodes. For a series of eight photoanode samples, the composition measured by XRF before photoelectrochemistry and by XPS after photoelectrochemistry are shown using a the molar ratio X/(Fe + Cr + V) with X = Fe, Cr, V for both XAS and XPS, and additionally X = O for XPS. The formula unit value of this ratio for X = O is 2.  The compositions with Fe/Cr near 1, e.g. near Cr 0.23 Fe 0.24 V 0.53, provide the global maximum for EQE at 2.1 eV, with the Fe/Cr = 1 ± 0.04 compositions exhibiting >2.4× higher EQE that compositions with Fe/Cr > 4, as shown in figure 6(b). The average absorption coefficient shows a decrease with Fe/Cr between 0.3 and 1 and further increasing Fe has no effect on the coefficient.
Quantitative analysis of the internal quantum efficiency (IQE) is best done with in situ measurement of the fraction of absorbed incident light during PEC, which was not performed in the present study. Figures 3,5, and 6 reveal that the maximal EQE samples are near the minimum in α, which indicates that the IQE of the compositions near Cr 0.23 Fe 0.24 V 0.53 will be substantially higher than the other materials in this composition library. This finding suggests that the equimolar Cr-Fe occupation in the MVO 4 structure enhances the charge separation and/or carrier transport efficiency.

Discussion
The EQE measurement at 1.23 V vs RHE with 3.2 eV photon illumination is the condition where most binary metal vanadate phases, e.g. FeVO 4 -tri, Fe 2 V 4 O 13 , CrVO 4 -mon, CrVO 4 -orth, and Cr 2 V 4 O 13 , have been demonstrated to be photoactive in our previous paper [12]. In the Cr-Fe-V oxide system, the presence of multiple photoactive phases with substitutional alloying-based tuning of composition result in detectable photoactivity for most compositions. The only compositions found to be relatively inactive are compositions with V concentration below 0.3 and Cr, Fe concentration between 0.3 and 0.6, where Cr 2 O 3 is the dominate phase. The local maxima in performance with 3.2 and 2.7 eV illumination appear in two regions: (a) near the Fe-V binary line with Cr concentration below 0.1, and Fe concentration between 0.4 (Fe-rich MVO 4 -mon, phase field E) and 0.6 (the mixture of FeVO 4 -tri and CeFeO 3 , phase field I), and (b) at a composition with minimum Fe (<0.1) and V concentration of 0.6 (phase mixture of CrVO 4 -mon and CrVO 4 -orth, phase field D), towards more V composition (phase mixture of CrVO 4 -m and M 2 V 4 O 13, phase field B). These observations are largely consistent with the known Cr and Fe metal vanadate photoanode materials from experiments using broad-band illumination [12][13][14][18][19][20][21][22].
While the present high throughput-based discovery cannot uniquely identify the mechanism underlying the improved performance provided by the equal population of Fe and Cr on the M site of the MVO 4 structure, the combinatorial data can provide guidance for future mechanistic studies. In the monoclinic BiVO 4 system, one demonstrated benefit of substitutional alloying is lowering of the monoclinic distortion to facilitate carrier transport [47]. While figure 6(a) show considerable lattice constant modulation with Fe-Cr composition, there is no apparent structural parameter with maximal value corresponding to the maxima in the EQE data of figure 6(b). The analysis of carrier lifetime and mobility by Zhang et al indicated that FeVO 4 carrier transport is in the carrier-tunneling regime wherein a high density of recombination centers enables tunneling from a trapped carrier site to a nearby recombination site [22]. Figure S7 provides circumstantial evidence for a high concentration of recombination centers in the Fe-rich MVO 4 samples with the observation of substantial current transients upon illumination. The time resolution of the present data does not afford detailed modelling of the current transients, but their intensity decreases with increasing Cr concentration. Such current transients are typically attributed to population of surface states, and their persistence in the presence of a sacrificial hole acceptor is commensurate with the extension of these states into the bulk of the semiconductor. This observation points toward the need to understand the extent by which Cr substitutions mitigate mid-gap states and the associated density of recombination centers. One mechanism by which this may occur is by lowering the concentration of oxygen vacancies, which have been demonstrated to be central to carrier transport and surface chemistry of BiVO 4 and other metal oxides [48,49]. Notably the XPS characterization of O concentration has neither the precision nor the penetration depth to ascertain defect-level changes in the oxygen stoichiometry of the thin film photoanodes.
The combinatorial synthesis of a broad range of composition coupled with high throughput synchrotron diffraction and spectrally-resolved EQE measurements reveals the composition region that performs well over a broader spectral range, the MVO 4 structure with composition near (Cr 0.5 Fe 0.5 )VO 4 -mon. The compositions near the local maximum in EQE in phase field D from 3.2 and 2.7 eV illumination are inactive at lower photon energies, demonstrating that non-spectrally-resolved measurements at the materials discovery stage can miss the most promising materials for further development, especially toward the development of a photoanode that utilizes lower-energy photons than state-of-the art photoanode BiVO 4 .

Conclusion
A series of M-V-O and M-O phases (M = Cr,Fe), and mixtures thereof, were fabricated by combinatorial synthesis, where all phases of interest exhibit Fe-Cr substitutional alloying on the M sublattice. High throughput XRD, photoelectrochemistry, and optical spectroscopy were combined to study the composition and structure-based variation in properties pertinent to solar fuels photoanodes. FeVO 4 and CrVO 4 -based photoanodes have been investigated for improving upon the state-of-the-art BiVO 4 by utilizing the 2-2.4 eV portion of the solar spectrum, and the present work indicates that the substitutional alloy phase (Cr 0.5 Fe 0.5 )VO 4 provides the greatest opportunity for developing next-generation photoanodes.

Data availability statement
The data that support the findings of this study are openly available at the following URL/DOI: https://data.caltech.edu/records/20061.