Electronic Structure and Interface Energetics of CuBi2O4 Photoelectrodes.

CuBi2O4 exhibits significant potential for the photoelectrochemical (PEC) conversion of solar energy into chemical fuels, owing to its extended visible-light absorption and positive flat band potential vs the reversible hydrogen electrode. A detailed understanding of the fundamental electronic structure and its correlation with PEC activity is of significant importance to address limiting factors, such as poor charge carrier mobility and stability under PEC conditions. In this study, the electronic structure of CuBi2O4 has been studied by a combination of hard X-ray photoemission spectroscopy, resonant photoemission spectroscopy, and X-ray absorption spectroscopy (XAS) and compared with density functional theory (DFT) calculations. The photoemission study indicates that there is a strong Bi 6s-O 2p hybrid electronic state at 2.3 eV below the Fermi level, whereas the valence band maximum (VBM) has a predominant Cu 3d-O 2p hybrid character. XAS at the O K-edge supported by DFT calculations provides a good description of the conduction band, indicating that the conduction band minimum is composed of unoccupied Cu 3d-O 2p states. The combined experimental and theoretical results suggest that the low charge carrier mobility for CuBi2O4 derives from an intrinsic charge localization at the VBM. Also, the low-energy visible-light absorption in CuBi2O4 may result from a direct but forbidden Cu d-d electronic transition, leading to a low absorption coefficient. Additionally, the ionization potential of CuBi2O4 is higher than that of the related binary oxide CuO or that of NiO, which is commonly used as a hole transport/extraction layer in photoelectrodes. This work provides a solid electronic basis for topical materials science approaches to increase the charge transport and improve the photoelectrochemical properties of CuBi2O4-based photoelectrodes.

Introduction

The tetragonal copper bismuth oxide CuBi2O4 has a crystal structure that features a three-dimensional array of [CuO4]6– square-planar units, staggered along the c-axis and separated by Bi3+ ions, as shown in Figure S1 of the Supporting Information (SI). Such a crystal structure isolates the [CuO4]6– units from each other, in contrast to the crystalline structure of the isoelectric prototypical high-temperature superconducting (HTSC) cuprates CuR2O4 (where R is a trivalent rare-earth element), in which the [CuO4]6– square-planar units form infinite two-dimensional layers.[1] Since this structural feature is a fundamental characteristic of HTSC cuprates, CuBi2O4 with its characteristic isolated [CuO4]6– units is unlikely to show superconductivity. However, CuBi2O4 has been shown to have potential use as a material for the photoelectrochemical (PEC) conversion of solar energy into chemical fuels.[2−4] Photocathodes based on this material have been shown to generate photocurrent densities on the order of −1.0 mA/cm2 at 0.0 V vs the reversible hydrogen electrode (RHE) with a Faradaic efficiency for hydrogen evolution of ∼91%.[5] In the ternary oxide CuBi2O4, the hybridizations of Bi 6s states and Cu 3d states with O 2p within the valence and conduction bands are key factors that cause the desirable materials properties, such as visible-light absorption as well as p-type conductivity.[2] Therefore, knowledge of the elemental composition of the valence band maximum (VBM) can provide fundamental guidance to materials design that could lead to further improvement of their properties.

Similar to other copper-based oxides, the major drawbacks limiting the PEC performance of CuBi2O4 are the lack of stability against photocorrosion when in contact with aqueous electrolytes and the poor charge carrier transport properties.[2,3] These limitations mean that the development of CuBi2O4-based photoelectrodes will depend on the optimization of interfacial electric contacts with protection layers and charge transport layers that assist the separation of photogenerated charges. Since the charge transfer across interfaces plays a major role in the overall efficiency of multilayer photoelectrodes, the engineering of interface energetics is key to overcome efficiency losses due to unfavorable charge carrier dynamics and transport. The optimization of the interfaces is ultimately based on the electronic properties of CuBi2O4, and therefore, a detailed determination of the electronic structure is needed in order to guide the rational design of stable and high-performing photoelectrodes. Herein, we report a comprehensive description of the electronic structure and electron energetics in CuBi2O4 using a combination of X-ray photoemission spectroscopy (XPS), X-ray absorption spectroscopy (XAS), and density functional theory (DFT) calculations.

The electronic structure of CuBi2O4 has been the subject of a number of theoretical studies based on DFT calculations,[2,6,7] which generally agree that the Bi 6s and Cu 3d electrons contribute to the valence band through hybrid electronic states with O 2p. However, there are diverging suggestions regarding the energy levels of the contributions of the metal orbitals to the valence band. Whereas calculations by Sharma et al.[2] suggested that Bi 6s and Cu 3d have comparable contributions to the valence band maximum (VBM), the results presented by Janson et al.[6] and Di Sante et al.[7] suggest that the Bi contribution to the VBM is in fact negligible, so that an effective one-band model is appropriate for the description of the low-lying excitations in CuBi2O4. Several studies of the electronic structure of HTSC cuprates have established that calculations using [CuO4]6– clusters with D4 symmetry as models accurately describe electronic features and soft X-ray photoemission spectra of low-dimensional cuprates, because these models take into account configurations, interactions, and final states with more than one d hole (i.e., Cu 3d8).[8−10] For [CuO4]6– clusters, the excitation of low-lying energy levels (i.e., the VBM) is associated with the antiferromagnetic coupling of O 2p holes with Cu 3d9 that leads to a singlet state (3d9L, L = ligand hole), usually referred to as the Zhang-Rice single state.[9,10] Due to the hybridization with d8 and d10L2 configurations, this singlet state has a large stabilization energy and strongly affects the physicochemical properties of cuprates, e.g., the charge mobility and doping stability.

Spectral features related to Zhang–Rice singlet states have been clearly identified in the valence band photoemission spectrum of CuO.[8,11,12] However, spectroscopic signatures of this singlet have not been observed in the photoemission spectrum of CuBi2O4,[13,14] which thus brings into question its stability within this material. In early experimental work, Goldoni et al. described CuBi2O4 as a charge-transfer insulator on the basis of their photoemission data, but no spectroscopic evidence of a Zhang–Rice singlet state was found.[14] Since the photoionization cross sections of the Bi valence states are too low compared to those of Cu 3d at the soft X-ray excitation energies that they used, the Bi contribution to the valence band could not be directly observed in the photoemission data. However, the Bi contribution to the valence band was addressed from the interpretation of electronic transitions observed in the electron energy loss spectroscopy data, which suggested that Bi 6s contributes to the top of the valence band.

In this paper, we report a detailed study of the electronic structure of CuBi2O4 based on an advanced and comprehensive X-ray spectroscopic approach, combining hard X-ray photoemission spectroscopy (HAXPES), resonant photoemission spectroscopy (ResPES), O K-edge XAS, and first-principles DFT calculations. Although photoionization cross sections of all electronic orbitals decrease as a function of the ionizing photon energy, the relative values are very different in the soft and hard X-ray regimes. As a result, soft X-ray photoemission spectroscopy preferentially probes Cu 3d valence states, whereas Bi 6s and Bi 6p make up most of the spectral features in HAXPES. This distinction allows identification of the Bi and Cu contributions to the valence band by taking valence band spectra at different ionization photon energies. We present an analysis of the optoelectronic properties of CuBi2O4 based on its electronic structure, concluding that the low visible-light absorption coefficient may come from a localized d–d forbidden electronic transition, leading to a low absorption coefficient in this spectral range and poor photochemical activity under low-energy visible-light irradiation. Additionally, a combined experimental and theoretical study of the electron energetics shows that the ionization potential (IP) of CuBi2O4 is higher than that of the related material CuO and that of NiO, which is the electronic basis for current materials science approaches to tailor charge carrier transport properties of CuBi2O4-based photoelectrodes.

Experimental Section

Sample Preparation

Samples were prepared by a spin-coating technique. Bismuth nitrate pentahydrate (Sigma-Aldrich, ≥99.99%) and copper nitrate trihydrate (Sigma-Aldrich, 99.999%) were dissolved in a solvent comprising a mix of acetic acid–water–2-methoxyethanol in a 2:1:1 volume ratio, to obtain a 2.6 M total metal concentration in solution. Citric acid was added to the solution at a 1:1 molar ratio with Cu just before using it for the deposition, which works as a gelling agent in the coating solution. The resulting solutions were used to spin-coat 2 × 2 cm2 sized fluorine-doped tin oxide (FTO) substrates at 3000 rpm. Prior to coating, the substrates were cleaned by ultrasound rinsing in acetone and then ethanol, followed by a UV/ozone cleaning treatment for 15 min. After coating, substrates were directly placed in a furnace at 450 °C and after 2 h the furnace was allowed to cool down naturally to room temperature, which led to the formation of a single-phase CuBi2O4 thin film as confirmed by XRD analysis (see Figure S1, SI).

The same procedure was used to prepare CuO. In this case, however, the solution concentration was 1.3 M copper nitrate trihydrate (Sigma-Aldrich, 99.999%).

For the nickel oxide hole transport layer, nickel nitrate hexahydrate (Sigma-Aldrich 99.999%) was dissolved in a 5:2 v:v solution of acetic acid–water to yield 5 mL of a 0.5 M solution. This solution was further diluted with 2.5 mL of 2-metoxyethanol and the resulting solution was used to spin-coat 2.5 × 2.5 cm2 sized FTO substrates at 3000 rpm. A mask made of sticky tape was used for selective area deposition of NiO, so that the same substrate had areas of bare FTO and areas covered by NiO. Prior to coating, the substrates were cleaned by ultrasound rinsing in acetone and then ethanol, followed by a UV/ozone cleaning treatment for 15 min. After coating, the substrates were calcined for 2 h at 500 °C to form single-phase NiO. After cooling naturally to room temperature, the samples were blown with nitrogen and cleaned with a UV/ozone treatment for 15 min, prior to the copper bismuthate layer addition, which was prepared as detailed before for the copper bismuthate samples on the FTO supports.

X-ray Diffraction and UV–Vis Spectroscopy

X-ray diffractograms were recorded with a Bruker D2 PHASER diffractometer using Cu Kα radiation. UV–vis spectra were taken using a Shimadzu UV–vis spectrophotometer with a deuterium light source (UV range) and a halogen light source (visible range). The thin film samples were measured in transmission mode using a piece of bare FTO substrate as reference and blank.

X-ray Spectroscopy

Lab-based X-ray photoemission spectra (XPS) were taken with a Thermo Scientific K-Alpha spectrometer using a 72 W monochromated Al Kα source (hν = 1486.6 eV). The X-rays are microfocused at the source to give a spot size on the sample of 400 μm in diameter. The analyzer is a double-focusing 180° hemisphere with a mean radius of 125 mm. It is run in constant analyzer energy (CAE) mode. The pass energy was set to 200 eV for survey scans and 50 eV for high-resolution region scans. The overall energy resolution gained was 0.5 eV, and all spectra were calibrated with respect to the Au Fermi level.

Hard X-ray and resonant photoemission spectra were measured at the I09 beamline of the Diamond Light Source in Didcot, UK.[15] For the ResPES experiment, the variable soft X-ray photon energy was tuned through a plane grating monochromator, whereas a Si(111) + Si(044) double-crystal monochromator was used for hard X-ray energies fixed at 8133 and 4068 eV. An EW-4000 photoelectron analyzer (VG Scienta) with a slit width set to 0.2 mm was used to record spectra at normal emission. The Cu L-edge X-ray absorption spectra associated with the ResPES experiment were measured in total electron yield (TEY) mode. The overall energy resolution of spectra taken at 8133 and 4068 eV photon energy was 0.32 and 0.27 eV, respectively. All spectra were calibrated with respect to the Au Fermi level.

The O K-edge X-ray absorption spectra were measured on the PGM beamline of the Laboratorio Nacional de Luz Sincrotron (LNLS), Campinas, Brazil. The variable soft X-ray photon energy was tuned through a plane grating monochromator, while the XAS spectra were taken in TEY mode with an effective resolution of 0.5 eV.

Samples studied by X-ray spectroscopy may be subject to beam damage. We addressed this issue by making sure that the chemistry of the material, as probed by coreline spectra (e.g., Cu 2p or Bi 4f), remained unchanged during the experiments. We took such coreline spectra before and after the VB recording.

Photoelectrochemical Characterization

The electrolyte was prepared by mixing equal parts of 0.1 M monobasic potassium phosphate (Sigma-Aldrich, ≥99.0%) and 0.1 M dibasic potassium phosphate dihydrate (Sigma-Aldrich, ≥99.0%). The resulting electrolyte solution had a measured pH of 6.85. The sample was placed in a PEC cell (Zahner model PECC-2) with a 3 mm diameter opening and 10 mL total electrolyte volume. A 50 μL portion of hydrogen peroxide (Sigma-Aldrich, 30 wt % in H2O) was added to the electrolyte in the cell and it was stirred thoroughly just before PEC measurements. Photoelectrochemical measurements were conducted using a Zahner Elektrik GmbH impedance measurement unit (model IM6), using a white-light-emitting diode (LED) with maximum intensity at 565 nm and 100 mW/cm2 intensity at the sample. For the three-electrode measurements of CuBi2O4 photocathodes, the counter electrode was a Pt wire with excess surface area, and the reference electrode was Ag/AgCl (3 M NaCl). Electrochemical potentials were converted to the RHE scale considering the calibrated potential of the Ag/AgCl (3 M NaCl) electrode, +0.241 V, and the measured pH of 6.85.

Computational Details

The spin-polarized density functional theory (DFT) calculations were performed using the Vienna Ab initio Simulation Package (VASP),[16] a periodic plane wave DFT code that includes the interactions between the core and valence electrons using the project augmented wave (PAW) method.[17] The Perdew–Burke–Ernzerhof (PBE) generalized gradient approximation (GGA) functional[18] was used for geometry optimizations. To overcome the limitation of the standard GGA-PBE functional in accurately predicting the electronic band gap of semiconducting materials, the screened hybrid functional (HSE06) with 25% Hartree–Fock exchange as proposed by Heyd–Scuseria–Ernzerhof was employed to determine the electronic structures of CuBi2O4, CuO, and NiO. A plane-wave basis set with a kinetic energy cutoff of 600 eV was tested to be enough to converge the total energies of CuBi2O4, CuO, and NiO to within 10–6 eV, and the residual Hellmann–Feynman forces on all relaxed atoms reached 10–3 eV Å. A Γ-centered Monkhorst–Pack k-mesh of (5 × 5 × 7), (7 × 7 × 5), and (9 × 9 × 9) was employed to sample the Brillouin zone of CuBi2O4, CuO, NiO, respectively. The bulk CuBi2O4 material was modeled in the tetragonal structure (P4/ncc [No. 130]), CuO in the monoclinic structure (C2/c [No. 15]), and NiO in the rock salt structure (Fm3̅m [No. 225]) with antiferromagnetic spin ordering (Figures S5–S7, SI). The optimized lattice parameters for CuBi2O4 (a = b = 8.498 Å and c = 5.903 Å) show very good agreement with the experimental lattice parameters (a = b = 8.499 Å and c = 5.803 Å).[3] Similarly, the optimized lattice parameters for CuO (a = 4.648 Å, b = 3.439 Å, c = 5.170 Å) and NiO (a = 4.209 Å) are all in close agreement (within 2% error) with reported experimental values[18,19] and previous theoretical values obtained with the PBE functional with and without the Hubbard (DFT+U) correction.[21,22]

In order to align the energies to the vacuum level, a slab-gap model was constructed and the corresponding electrostatic potential was averaged along the c-direction (see Figures S8–S10, SI), using the MacroDensity package.[23−25] The ionization potentials (IPs) were calculated when the slab vacuum level is aligned to the bulk eigenvalues, through core level eigenvalues in the center of the slab, using the O 1s orbital energy as a reference point. The electron affinity (EA) is calculated by subtracting the band gaps from the calculated ionization potentials. The work function (Φ), which is the minimum energy needed to remove an electron from the bulk of a material through a surface to a point outside the material, was calculated as Φ = Ev – EF, where Ev and EF are the energies of the vacuum and Fermi levels, respectively. A dipole correction perpendicular to all surfaces was applied, which ensured that there is no net dipole perpendicular to the surfaces that may affect the potential at the vacuum level. To allow for easy comparison with experimental data, the photoemission spectra have been simulated using the GALORE code,[26] which extracts and interpolates the cross-sectional weights from reference data and generates tabulated data for experimental comparison. The photoionization cross sections calculated by Scofield were applied.[27] It is assumed that the contributions of atomic s-, p-, d-orbital-projected states to the total photoemission spectrum will be proportional to the photoionization cross sections of corresponding orbitals in free atoms based on the Gelius model.[28,29]

Results and Discussion

The experimental study of the electronic structure of CuBi2O4 was carried out with thin film samples deposited on fluorine-doped SnO2 (FTO) by a reproducible sol–gel method described in the Experimental Section. It was ensured that no photoemission signal from the FTO substrate could be detected in the spectra of the samples, as shown in the survey spectra in Figure S2 (SI). Additionally, samples were conductive enough so that photoemission spectra were not subject to charging effects. The soft X-ray photoemission spectrum of as-prepared samples, shown in Figure S2 (SI), features intense peaks in the Bi 4f and Cu 2p regions with maxima at 158.4 and 934.1 eV, well within the expected binding energy range for Bi(III) and Cu(II) cations, respectively. The main Cu 2p peak in the spectrum of as-prepared samples, shown in detail in Figure S3 (SI), is highly symmetric, which is characteristic of CuBi2O4 and unlike CuO and other low-dimensional cuprates. Such symmetry is a consequence of the absence of nonlocal screening due to the isolation of square planar [CuO4]6– species in the unit cell of CuBi2O4.[30,31]

Cu 2p Resonance in the VB Photoemission: Cu 3d Contribution to the VB

Resonant photoemission spectroscopy is a well-established technique that provides a unique means to identify the character of the various states involved in the one-electron photoemission process. This technique is powerful for solids where electron-correlation effects are expected to play an important role in the electronic structure, owing to the presence of cations with open-shell valence states (e.g., Cu2+, a 3d9 cation) that allow intra-atomic electronic transitions.[32] In this sense, we explored the Cu 3d contribution to the valence band in detail, on the basis of the Cu 2p resonance in the valence band photoemission. Figure A shows valence band photoemission spectra of CuBi2O4 recorded with excitation photon energies below the onset (929.8 eV), at the onset (931.8 eV), and at the maximum (933.1 eV) of the Cu L3 absorption edge, as indicated in Figure B. A very strong resonant effect can be readily observed, in particular for the high binging energy spectral features. Tuning the excitation photon energy to the maximum of the Cu L3 edge (hν = 933.1 eV) induces the appearance of a very strong satellite with a maximum at 13.2 eV binding energy. The resonant photoemission spectrum observed here results from the interference of two photoemission channels leading to a Cu 3d8 final state: (i) a direct channel (Cu 3d9 + hν → Cu 3d8 + e–) and (ii) a deexcitation channel, which involves a photon absorption process followed by an Auger decay (Cu 3d9 + hν → Cu 2p53d10 → Cu 3d8 + e–).[32]

Figure 1

(A) Valence band photoemission spectra of CuBi2O4 taken using photon energies near the Cu L3. (B) Cu L3 X-ray absorption spectrum of CuBi2O4 indicating the selected photon energies for the resonant photoemission experiment.

The deexcitation channel only probes photoionization from Cu 3d orbitals leaving a d8 final state and it is only available for the on-resonance spectrum. On the other hand, the direct channel also probes direct photoionization from O 2p and Bi 6s orbitals and is available at any photon energy. In view of the drastic increase of the VB spectral intensity at the Cu 2p resonance, it is clear that the deexcitation channel dominates the resonant photoemission, and therefore, the resonance reflects the contribution from Cu 3d orbitals that leave a d8 final state upon excitation. The data show that such a contribution resides mostly at the high binding energy side of the main valence band, which is consistent with the description of CuO and HTSC cuprates as charge transfer insulators. A very similar spectroscopic structure was found in the Cu 2p resonant photoemission spectrum of CuO.[11,33] Tjeng et al. could satisfactorily reproduce this spectroscopic structure in their 2p resonant photoemission calculations of the Cu 3d spectral weight for a square planar [CuO4]6– cluster model.[33] Calculations by Tjeng et al. confirmed that the Cu 2p resonance in the VB photoemission primarily derives from a singlet Cu 3d8 final state, with peaks at 16.9 and 13.2 eV binding energies containing atomic 1S and 1G character, respectively (see Figure A). This further confirms the charge transfer nature of CuO that can be translated to CuBi2O4.

A more detailed picture of the pure Cu 3d contribution to the valence band can be identified by the intensity difference between on- and off-resonance spectra, as shown in Figure . Although the main Cu 3d spectral feature resides at energy levels around 13 eV below the Fermi level, the resonance enhancement of the upper valence band indicates a hybridization between Cu 3d and O 2p states, so that partial Cu 3d character is present across the whole spectrum up to the VBM. The spectral structure indicated at the top of the valence band can be associated with a Zhang–Rice (Z–R) singlet final state (3d9L, L = O 2p hole), which is the first ionization state in the [CuO4]6– cluster model.[8−10] The Cu 2p resonant enhancement of the top of the valence band spectrum depicts its partial d8 final state character, which is consistent with the hybridization of a primer 3d9L final state with a d8 final state, required for the stabilization of the Zhang–Rice final state.[13,33]

Figure 2

Valence band photoemission of CuBi2O4 near the Fermi level showing the spectral intensity difference between on-resonance (at hν = 933.1 eV) and off-resonance (at hν = 929.8 eV) spectra.

Hard X-ray Valence Band Photoemission: Probing the Contribution of Bi 6s and Bi 6p to the VB

The photoemission spectrum of CuBi2O4 in the valence band region measured with an Al Kα1 lab X-ray source (1486.6 eV), as well as those measured with 4068 and 8133 eV at a synchrotron, is shown in Figure A. The main valence band has a spectral onset at about 0.7 eV and expands up to 8 eV in binding energy. In addition, a broad peak is found on the high binding energy side of the main valence band, centered at 12 eV binding energy. Four spectral features, labeled I, II, III, and IV, can be identified in the main valence band region. The intensity of II and IV relative to I and III continuously increases while moving from soft to hard X-ray ionization photon energy (hν), resulting in a marked change in the shape of the VB spectrum. At the same time, switching from soft to hard X-ray hν leads to a marked increase in the relative intensity of the broad satellite feature at the high binding energy side.

Figure 3

(A) Experimental VB photoemission spectra of CuBi2O4 measured with 1486 eV (Al Kα1), 4068 eV, and 8133 eV ionizing photon energy. (B) Photoionization cross section dependence on the ionizing photon energy for valence orbitals in CuBi2O4.[27] (C) Calculated VB photoemission spectra of CuBi2O4 at 8133 eV ionizing photon energy.

It is important to note that despite differences in probing depth inherent to soft (surface) and hard (bulk) X-ray photoemission spectroscopies, all features can be clearly observed in all spectra. Therefore, we argue that these features are intrinsic to the material rather than due to surface defects. The observed changes in the VB spectral feature originate from the hν-dependence of the photoionization cross section of the atomic orbitals contributing to the valence levels. For CuBi2O4, the valence band is mainly O 2p, with contributions from Cu 3d, Bi 6s, and Bi 6p due to a certain degree of covalency in the bonds, as detailed in our DFT modeling of the electronic structure shown in Figure S5 (SI). However, due to a large photoionization cross section, Cu 3d photoelectrons dominate the spectral features of the soft X-ray spectrum (see Figure B). In fact, calculations of the Cu 3d spectral weight, using a [CuO4]6– cluster with D4 symmetry, have been shown to reproduce fairly well the spectral features observed in the Al Kα VB photoemission spectrum of CuBi2O4.[13,14] From those calculations and in accordance with our previous discussion, feature I at the top of the VB spectrum can be assigned to the Cu 3d–O 2p hybrid with the Zhang–Rice final state, which has not been observed previously in the photoemission spectrum of CuBi2O4, probably due to insufficient spectral sensitivity and resolution. Furthermore, as discussed above, the high binding energy broad feature can be partially assigned to Cu 3d with d8 final states; however, such a satellite feature is also observed in valence band photoemission spectra of Bi-based oxides and is assigned to the main contribution from Bi 6s.[34,35]

As shown in Figure B, the photoionization cross section values for O 2p and Cu 3d decrease much faster than those of Bi 6s, Bi 6p, and Bi 5d, and as a result, calculations of the Cu 3d spectral weight using a [CuO4]6– cluster cannot reproduce the hard X-ray photoemission spectrum because Bi 6s, Bi 6p, and Bi 5d states dominate most spectral features. The evolution of the valence band spectrum as the ionization photon energy is shifted from the soft to the hard X-ray regime allows us to conclude that the states associated with features II and IV at 2.3 and 6.5 eV, respectively, have pronounced Bi 6s and Bi 6p character hybridized with O 2p. In addition, the sharp increase in relative spectral intensity for the high binding energy broad feature is consistent with its partial assignment to Bi 6s as the main contribution.

In order to support the assignment of the contribution of Bi states to the VB, we have simulated the hard X-ray photoemission spectrum at hν 8133 eV using the GALORE code.[26] Cross section weighted partial density of states (PDOS) extracted from band structure calculations, shown in Figure S5 (SI), were summed to generate the simulated spectra shown in Figure C. In general terms, the simulation compares well with the measured spectrum and allows us to conclude that feature II (centered at around 2.3 eV) has a pronounced Bi 6s–O 2p character, whereas the spectral feature labeled as IV results from an overlap of VB contributions for Cu 3d and Bi 6p.

An early experimental study of the electronic structure of CuBi2O4 by Goldoni et al.[14] was based on soft X-ray photoemission and electron energy loss (EEL) spectroscopies. A very low relative value of the Bi 6s and Bi 6p cross sections at soft X-ray energy hindered the direct observation of the contribution of Bi electronic states to the valence band. However, the authors used the interpretation of EEL spectra to conclude that there was a contribution of Bi 6s at the top of the valence band. On the other hand, as shown in our work, hard X-ray excitation energies provide direct observation of the contribution of Bi states within the valence band of CuBi2O4. Our results reveal that this contribution is deeper in energy into the valence band, with maximum Bi 6s–O 2p hybrid states at 2.3 eV below the Fermi level, leaving the valence band maximum with a predominant Cu 3d–O 2p hybrid character. Therefore, taken together, our results suggest that the electronic structure of CuBi2O4 can be described within the charge-transfer semiconductor picture, similar to its parent binary oxide CuO.[33,36]

O K-Edge X-ray Absorption

Due to dipole selection rules, the O K-edge XAS probes the transition from O 1s to unoccupied states with partial O 2p character hybridized with Cu 3d, Bi 6s, and Bi 6p states. Theoretical and experimental descriptions of the XAS process have shown that the interaction of valence electrons with the core hole created upon excitation leads to a slight modification of the electronic structure of the final state, which complicates a straightforward comparison of XAS spectra with the ground-state density of states.[37,38] This effect is particularly pronounced for transition-metal L-edge XAS, although detailed calculations of the effect of the O 1s core hole in transition-metal oxides demonstrate that its effect on the band structure is weak. Thus, the O K-edge spectra can be qualitatively related to unoccupied density of states of primarily transition-metal character, provided there is enough hybridization with O 2p to generate measurable oscillator strength.[39] The O K-edge XAS of CuBi2O4, shown in Figure A, has an onset at 529.6 eV and the main absorption expands up to 537.5 eV. Four features can be seen in the main absorption spectrum, as indicated in Figure A with letters a–d. By comparing with the O K-edge XAS of CuO, we tentatively assign the onset feature a to be Cu 3d–O 2p hybrid states at the bottom of the conduction band.

Figure 4

(A) O K-edge XAS of CuBi2O4 along with that of CuO for comparison. (B) Empty PDOS from DFT calculations for CuBi2O4; the O K-edge XAS is included in the same scale for comparison. (C) UV–vis absorption spectrum of CuBi2O4.

In order to assign the contribution of Bi states to the conduction band of CuBi2O4, we directly compare our experimental O K-edge XAS spectrum to the corresponding hybrid DFT-calculated PDOS in Figure B. To make the comparison illustrative, we shifted the spectrum onset to match that of empty states in the DFT calculations. The comparison confirms the Cu 3d character of spectral feature a at the conduction band minimum (CBM), whereas features b–d, located at higher energy levels, can be mostly related to Bi 6p–O 2p hybrid states.

Together with the photoemission study of the valence band, showing that the VBM is predominantly a Cu 3d–O 2p hybrid electronic state, our results provide more insights into the optical absorption spectrum of CuBi2O4, shown in Figure C. Previous reports have argued that the weak optical absorption observed at 1.8 eV arises from the band gap absorption.[2−4] However, since the lowest-energy optical absorption arises from the transition from the VBM of mixed O 2p/Cu 3d character to the empty Cu 3d eg state, our results suggest that the weak visible absorption onset is most probably due to a low cross section parity-forbidden crystal field d–d transition. Apart from this weak transition, charge-transfer (CT) transitions previously identified for CuBi2O4 can been seen in the UV–vis spectrum of our sample at 2.08 eV (p–d intracenter CT transition), 2.55 eV (d–d intercenter CT transitions), and >3.2 eV (p–d intracenter CT transition).[40] Charge-transfer transitions can be associated with the band gap excitation in CuBi2O4, which is an indirect transition, according to our band structure diagram along high-symmetry points of the Brillouin zone (see Figure S5, SI). Although one major asset of CuBi2O4 for photoelectrochemical applications is its low onset of visible-light absorption, it shows negligible photochemical activity under low-energy visible-light irradiation. For instance, it has been shown that the incident photon to current efficiency (IPCE) is negligible for hν ≤ 2.25 eV,[3] and we have recently shown that the surface photoreduction of CuBi2O4 does not proceed when irradiated with hν ≤ 2.30 eV.[41] These observations are consistent with the localized nature of the low-energy optical transitions. Low-energy visible-light absorption associated with a parity-forbidden crystal field d–d transition has also been observed in large band gap Cu(II)-based oxides, such as CuWO4 and CaCu3Ti4O12.[42,43]

Interface Energetics in CuBi2O4 and Choice of Contact Materials

As discussed above, although both Bi 6s and Cu 3d contribute to the valence band via hybridization with O 2p, the VBM of CuBi2O4 can be described as O 2p with significant contribution from Cu 3d and negligible contribution from Bi 6s. Likewise, the CBM is predominantly empty Cu 3d hybridized with O 2p, providing this material with a chemistry similar to that of the parent compound CuO. In fact, the material properties of CuBi2O4 are closely related to those of CuO; e.g., both materials exhibit p-type semiconductivity[2,44] and behave as cathodes in photoelectrochemical cells.[3,45] Both materials present a characteristic low carrier (hole) mobility, <10–2 cm2 V–1 s–1,[3,46,47] which is detrimental to PEC applications of these materials. However, as described in detail below, the ionization potential (IP) of CuBi2O4 is larger than that of CuO.

We have explored the electron energy levels in CuBi2O4 relative to other more common p-type semiconductor oxides, such as CuO and NiO, by determining their IPs via photoemission spectroscopy. To this end, we measured the work function (Φw) by determining the difference between the width of the photoemission spectrum and the source energy (Figure S4, SI). Then, the IP was estimated from the onset of the valence band spectrum, considering that IP = Φw + VBM. An estimation of the conduction band minimum (or electron affinity) can be obtained from the reported optical band gaps of these materials, i.e., 1.44 eV for CuO,[20] 3.7 eV for NiO,[19,48] and 1.8 eV for CuBi2O4.[2−4,41]

Using this information, we constructed schematics of the band energy levels relative to the vacuum level for CuBi2O4, CuO, and NiO (see Figure A). In order to support the experimentally determined relative band energy positions, we have also carried out two separate computational analyses on kusachiite CuBi2O4, cubic CuO, and rock salt NiO: (i) the electronic structures of the bulk crystals and (ii) the absolute vacuum alignment from a well-converged slab-gap model. The band gap of CuBi2O4 is predicted at 1.98 eV, in close agreement with the current and previous experimentally estimated value of 1.80 eV and the previous DFT+U predicted value of 1.90 eV.[2−4] The standard GGA-PBE functional underestimated the band gap of CuBi2O4 at 1.05 eV. The band gaps of CuO (1.76 eV) and NiO (3.69 eV) calculated using the HSE06 hybrid functional are consistent with known optical band gaps for CuO (1.3–2.1 eV eV)[49,50] and NiO (3.4–4.3 eV),[51−53] depending on how the location of the band edge is defined: the location of the first absorption feature, the midpoint of the first rise, or where the maximum slope of absorption extrapolates to zero. Previous LDA+U calculations give a band gap of 3.4 eV for NiO.[21] For CuO, previous DFT+U calculations predict the band gap at 1.25 eV, whereas the hybrid functional predicts a band gap of 1.42 eV (for mixing ratio α = 0.15).[21]

Figure 5

(A) Measured and (B) calculated electron affinity (CBM) and ionization potential (VBM) of CuBi2O4, CuO, and NiO with respect to the vacuum level.

CuBi2O4(101), CuO(111), and NiO(111) were chosen for the slab calculations, as they do not contain dangling bonds and resulted in low-energy, nonpolar terminations (Figures S7–S10, SI). The slabs are constructed of thicknesses larger than 20 Å [24.362 Å for CuBi2O4(101), 23.805 Å for CuO(111), and 24.304 Å for NiO(111)], and in every simulation cell, a vacuum region of 15 Å perpendicular to the surface was tested to be sufficient to avoid interactions between periodic slabs. The electrostatic potential was averaged along the c-direction in order to determine the external vacuum level (see Figures S8–S10, SI), using the MacroDensity package.[23−25] Displayed in Figure B are the calculated IP and electron affinity (EA) values of the CuBi2O4(101), CuO(111), and NiO(111) surfaces. Consistent with the experimental values, the IPs of CuBi2O4(101), CuO(111), and NiO(111) surfaces were predicted as 6.01, 4.75, and 5.02 eV, whereas the EA values were predicted as 4.04, 2.99, and 0.92 eV, respectively.

There is agreement between measured and simulated values of the electron energetics for each material. In particular, the simulated values reproduce very well the relative energy levels of the CBMs and VBMs of CuBi2O4, CuO, and NiO, which is the most important parameter that determines the most likely electron transfer in a stacked photoelectrode design. For instance, there has been an increasing interest in using NiO thin films as a hole transport layer in solar energy devices, mainly due to their successful incorporation in hybrid perovskite photovoltaic cells,[54,55] as well as all-oxide transparent diodes.[56,57] However, it is important to consider that NiO must have a smaller ionization potential than the semiconductor photoabsorber coupled to it, in order to form a suitable hole selective contact and behave as a hole transport layer. In this sense, a NiO thin film may act as a hole transport layer for CuBi2O4-based photoelectrodes, as Lee et al.[58] and Song et al.[59] have shown recently. However, despite the fact that CuO is a well-known photocathode material, coupling a NiO layer to CuO would bring about a detrimental effect to the PEC efficiency of the resulting system, owing to incompatible interface energetics. Therefore, the relative high IP of CuBi2O4 provides the basis for materials engineering approaches to enhance the hole mobility and transport, which is very important for Cu-based photoelectrodes because of their poor charge carrier transport properties, as described above.

We have carried out a proof-of-concept experiment in which we compared the effect of NiO as a contact layer in CuO- and CuBi2O4-based photoelectrodes. To achieve this setup, we have developed a multilayer thin film deposition method in order to make photoelectrodes for a comparative photoelectrochemical study. In a first step, NiO thin films were deposited by a sol–gel method on FTO substrates with a pattern defined by a mask. The resulting NiO-patterned FTO substrates were used for an overall spin-coating deposition of CuO and CuBi2O4, as described in the Experimental Section, so that part of the photoabsorber film is deposited on bare FTO and the other part over NiO/FTO, featuring similar film thicknesses. Figure A shows the linear sweep voltammetry scan under chopped illumination of a CuBi2O4 photocathode on bare FTO (black line) and that of the same material on NiO/FTO (red line). Similarly, Figure B shows the photoresponse of CuO deposited on bare FTO and on NiO/FTO. A degassed solution of 200 mM sodium sulfate and 100 mM sodium phosphate buffer at pH 6.8 was used as the electrolyte, with the addition of 50 mM of hydrogen peroxide as an electron scavenger.

Figure 6

(A) Chopped-light linear sweep voltammetry scans for CuBi2O4 (black) and NiO/CuBi2O4 (red) photoelectrodes. (B) Chopped-light linear sweep voltammetry scans for CuO (black) and NiO/CuO (red) photoelectrodes. PEC measurements were done in 0.2 M K2SO4 + 0.1 M phosphate buffer solution (pH 6.8) with 0.3% w/w H2O2.

A substantial increase in the photoresponse is observed for CuBi2O4 photocathodes upon introducing NiO as a hole transport layer to the electrode, reaching 2.1 mA cm–2 at 0.6 V vs RHE, which is an effect that has been previously observed in CuBi2O4-based photoelectrodes.[58,59] In contrast, the incorporation of NiO in CuO photoelectrodes causes a detrimental effect that lowers the observed photocurrent by about 50%. The very fast kinetics for H2O2 reduction minimizes any catalytic limitation at the surface of the electrode, so the observed increased photoresponse reflects the better charge carrier transport properties of CuBi2O4–NiO photoelectrodes, in contrast to the detrimental effect in using NiO as the contact layer in CuO photoelectrodes.

Conclusion

In this paper, we have reported a detailed study of the electronic structure of CuBi2O4 based on advanced X-ray spectroscopic techniques: HAXPES, ResPES, and O K-edge X-ray absorption spectroscopy. Owing to the dependence of the relative photoionization cross sections for different electronic orbitals on the ionizing photon energy, soft X-ray photoemission spectroscopy preferentially probes Cu 3d valence states, whereas Bi 6s and Bi 6p make up most of the spectral features in HAXPES. This difference allowed us to identify the Bi and Cu contributions to the valence band by taking VB spectra at different ionization photon energies. On the basis of this strategy, we found that (i) the main Cu 3d spectral feature resides at energy levels around 13 eV below the Fermi level. However, due to the strong hybridization between Cu 3d and O 2p states, Cu 3d character is present across the whole spectrum up to the valence band maximum (VBM). We identified a spectroscopic feature at the top of the VB associated with a Zhang–Rice singlet final state (Cu 3d9L), which has been predicted for CuBi2O4 but not observed in previous experimental studies. (ii) There is a strong Bi 6s–O 2p hybrid electronic state with a maximum at 2.3 eV below the Fermi level, although the contribution is negligible at higher binding energies, leaving the VBM with a predominant Cu 3d–O 2p hybrid character. (iii) The CBM has strong Cu 3d character, whereas Bi 6p–O 2p hybrid electronic states make up most of the rest of the conduction band. On the basis of these electronic features, we suggest that the low visible-light absorption at the onset of the absorption feature at hν ∼ 1.8 eV originates from a localized d–d forbidden electronic transition, leading to a low absorption coefficient for visible light and consequent poor photochemical activity under low-energy visible-light irradiation. In addition, a combined experimental and theoretical study of the interface energetics shows that the IP of CuBi2O4 is higher than that of the related material CuO and that of NiO, which provides a basis for materials engineering approaches to enhance the hole transport, which is critically important for Cu-based photoelectrodes, owing to their poor charge carrier transport properties.