Revealing the relationship between photoelectrochemical performance and interface hole trapping in CuBi₂O₄ heterojunction photoelectrodes

Abstract

p-Type CuBi₂O₄ is considered a promising metal oxide semiconductor for large-scale, economic solar water splitting due to the optimal band structure and low-cost fabrication. The main challenge in utilizing CuBi₂O₄ as a photoelectrode for water splitting, is that it must be protected from photo-corrosion in aqueous solutions, an inherent problem for Cu-based metal oxide photoelectrodes. In this work, several buffer layers (CdS, BiVO₄, and Ga₂O₃) were tested between CuBi₂O₄ and conformal TiO₂ as the protection layer. RuO_x was used as the co-catalyst for hydrogen evolution. Factors that limit the photoelectrochemical performance of the CuBi₂O₄/TiO₂/RuO_x, CuBi₂O₄/CdS/TiO₂/RuO_x, CuBi₂O₄/BiVO₄/TiO₂/RuO_x and CuBi₂O₄/Ga₂O₃/TiO₂/RuO_x heterojunction photoelectrodes were revealed by comparing photocurrents, band offsets, and directed charge transfer measured by modulated surface photovoltage spectroscopy. For CuBi₂O₄/Ga₂O₃/TiO₂/RuO_x photoelectrodes, barriers for charge transfer strongly limited the performance. In CuBi₂O₄/CdS/TiO₂/RuO_x, the absence of hole traps resulted in a relatively high photocurrent density and faradaic efficiency for hydrogen evolution despite the presence of pronounced deep defect states at the CuBi₂O₄/CdS interface. Hole trapping limited the performance moderately in CuBi₂O₄/BiVO₄/TiO₂/RuO_x and strongly in CuBi₂O₄/TiO₂/RuO_x photoelectrodes. For the first time, our results show that hole trapping is a key factor that must be addressed to optimize the performance of CuBi₂O₄-based heterojunction photoelectrodes.

---

Introduction

Copper bismuth oxide (CuBi₂O₄) is a promising photoabsorber for photoelectrochemical (PEC) water splitting due to its optimal optical bandgap (1.5–1.8 eV), positive photocurrent onset potential (more positive than 1 V vs. RHE), and Earth-abundant chemical composition.^1–4 However, several limitations in CuBi₂O₄ must be overcome to improve its performance as a photocathode for the hydrogen evolution reaction. Perhaps the greatest limitation of CuBi₂O₄ is its susceptibility to photo-corrode under illumination in aqueous solutions, which is a common problem for Cu-based metal oxide photoelectrodes.^5–8 Surface modification via the formation of a heterojunction with a suitable buffer layer and/or an n-type protection layer could be highly effective at overcoming this limitation of CuBi₂O₄ based on previous reports on other Cu-based metal oxide photocathodes such as Cu₂O, CuO and CuFeO₄ as well as Si-based photoelectrodes.^9–13

To be effective, a suitable buffer layer and/or protective layer must cover CuBi₂O₄ conformally without any pinholes and have energy band positions that match favorably to allow for efficient transport of charge carriers across the solid-state and semiconductor–electrolyte interfaces. TiO₂ deposited by atomic layer deposition (ALD) has been reported as an excellent protective layer for unstable photoelectrodes while simultaneously allowing for efficient electron transfer to the electrolyte under the PEC conditions for hydrogen evolution.^11,14–16 Studies have shown that TiO₂-protected Cu₂O-based photocathodes exhibit a relatively high photocurrent density and significantly enhanced stability when using various buffer layers between Cu₂O and TiO₂ such as ZnO and Al doped ZnO (AZO).^15,17,18 However, relatively poor photocurrent onset potentials (0.45–0.55 V vs. RHE) were obtained in this structure due to the small photovoltage produced by the heterojunctions. It has been shown that the use of ZnS as a buffer layer between Cu₂O and TiO₂ can increase the photovoltage at the multilayer/electrolyte junction thereby shifting the onset potential cathodically to 0.7 V vs. RHE.¹⁹ The introduction of Ga₂O₃ as a buffer layer between Cu₂O and TiO₂ can improve the photovoltage even further (open-circuit voltage up to 1.2 V for Cu₂O solar cells²⁰ and photocurrent onset above 1.0 V vs. RHE for photocathodes).^21–23 In another work for CuO-based heterojunction photocathodes, ZnO showed rather poor performance as a buffer layer, in contrast to CdS.²⁴

The band positions of the various layers within TiO₂-protected heterojunction photocathodes are crucial in determining the overall performance. The photovoltage of the device is ultimately limited by the difference in the Fermi level of the photoabsorber and the conduction band of either the buffer layer or the TiO₂ protective layer, depending on which has a lower conduction band.⁵ Furthermore, numerous studies have emphasized the importance of band alignment between the photoabsorber and the buffer layer. In addition to band alignment, other key factors related to recombination and/or trapping at the layer interfaces can contribute to the overall performance. For example, it was recently shown by model calculations that Al-doped ZnO (AZO) buffer layers between Cu₂O and TiO₂ should enable a higher photovoltage than Ga₂O₃ buffer layers, but AZO may induce an interface recombination layer that hinders charge transfer and thus decreases the photovoltage.⁹ Therefore, in addition to optimal band energy alignment, heterojunction interface layers must be high-quality and possess low interfacial trap densities to maximize device performance. Incidentally, compared to band positions, defect states and recombination sites at interfaces are often more difficult to characterize. Modulated surface photovoltage (SPV) spectroscopy can be used to provide information about electronic transitions from which photogeneration can take place and about the direction of charge separation even over very short distances and with very high sensitivity.^25,26

In order to obtain information about dominant limiting factors in heterojunction photoelectrodes with CuBi₂O₄ as the absorber layer and ALD-deposited TiO₂ as the protective layer, we compared how different buffer layers (CdS, BiVO₄, and Ga₂O₃) affected the photocurrent density (with RuO_x as a co-catalyst layer for H⁺ reduction reaction) as well as the behavior of modulated SPV spectra. Fig. 1 shows the layer stacking of the various heterojunction photoelectrodes that were tested in this work. The energy positions of the valence band edges in the separate layers were measured by photoelectron spectroscopy. Our results show that, in addition to suitable band alignment for charge transfer, preferential trapping of holes at CuBi₂O₄/TiO₂, CuBi₂O₄/BiVO₄ and CuBi₂O₄/Ga₂O₃ interfaces drastically limits the photocurrent of the corresponding PEC systems. In contrast, preferential trapping of electrons at the CuBi₂O₄/CdS interface limits the photocurrent to a much lesser extent despite the appearance of pronounced deep defect states.

Fig. 1 Schematic of the layer stacking for the heterojunction photoelectrodes analyzed in this work. A TiO₂ protection layer was used in all cases, which was deposited either directly on the CuBi₂O₄ absorber, or on the Ga₂O₃, BiVO₄, or CdS buffer layer. RuO_x was used as a co-catalyst for the hydrogen evolution reaction. The substrate was SnO₂ : F, fluorine-doped tin oxide (FTO).

Results and discussion

Photoelectrochemical analysis

In this work we performed PEC measurements on bare CuBi₂O₄ photocathodes in addition to the CuBi₂O₄/CdS/TiO₂/RuO_x, CuBi₂O₄/Ga₂O₃/TiO₂/RuO_x, CuBi₂O₄/BiVO₄/TiO₂/RuO_x and CuBi₂O₄/TiO₂/RuO_x heterojunction photoelectrodes. They were measured under chopped AM 1.5 illumination in 0.3 M K₂SO₄ and 0.2 M phosphate buffer (pH 6.8) with Ar bubbling to purge dissolved oxygen from the electrolyte. As shown in Fig. S1a in the ESI,† the bare CuBi₂O₄ electrode exhibited a relatively large cathodic photocurrent density of −1 mA cm⁻² at 0.4 V vs. RHE under visible light illumination for the chopped linear sweep voltammetry (LSV) measurement. However, the photocurrent density decayed rapidly, as shown in the constant potential measurement under illumination at 0.6 V vs. RHE (see Fig. S1b†). Based on previous reports this is attributed to the reduction of Cu²⁺ to Cu¹⁺ and/or Cu in aqueous solution.^1,27–29 After 30 minutes the photocurrent density was only 6.4% of the initial value and by the end of the 5 hour measurement the illuminated area of the photoelectrode was transparent (see the inset of Fig. S1b†) presumably because the reduced copper dissolved into the electrolyte. Differential mass spectrometry (DEMS) measurements confirmed that the bare CuBi₂O₄ photocathodes did not produce a detectable amount of hydrogen (see Fig. S1c†).

The chopped LSV curves for the CuBi₂O₄/CdS/TiO₂/RuO_x, CuBi₂O₄/Ga₂O₃/TiO₂/RuO_x, CuBi₂O₄/BiVO₄/TiO₂/RuO_x and CuBi₂O₄/TiO₂/RuO_x heterojunction photoelectrodes are shown in Fig. 2a. Unlike the bare CuBi₂O₄ photocathode, the heterojunction photoelectrodes showed minimal dark currents at potentials more negative than 0.35 V vs. RHE, indicating that the ALD-deposited TiO₂ layer effectively inhibits the electrochemical corrosion of the underlying CuBi₂O₄. However, the photocurrent generated from the CuBi₂O₄/TiO₂/RuO_x and CuBi₂O₄/Ga₂O₃/TiO₂/RuO_x photocathodes (blue and black lines) were very small (ca. −0.04 mA cm⁻² and −0.01 mA cm⁻² at 0 V vs. RHE, respectively). This indicates that the CuBi₂O₄/TiO₂ and the CuBi₂O₄/Ga₂O₃ or Ga₂O₃/TiO₂ interfaces in these samples do not effectively charge transport to the RuO_x co-catalyst layer and into the electrolyte. The CuBi₂O₄/BiVO₄/TiO₂/RuO_x photoelectrode (green line) shows significantly higher photocurrent density (ca. −0.5 mA cm⁻² at 0 V vs. RHE) while the CuBi₂O₄/CdS/TiO₂/RuO_x photoelectrode (red line) shows by far the highest photoactivity with a plateau in photocurrent density at ca. −1 mA cm² at 0 V vs. RHE. It also has the most positive photocurrent onset at 0.8 V vs. RHE. For the CuBi₂O₄/CdS/TiO₂/RuO_x photoelectrode the bump in the dark current around 0.5–0.6 V vs. RHE is likely due to the reduction of dissolved oxygen that could not be completely purged from the PEC cell by Ar bubbling,⁵ or another reduction process on the electrode surface such as the pre-reduction of the RuO_x catalyst.¹⁶ The DEMS measurement (Fig. 2b) shows that when the photocurrent increases cathodically at potentials negative of 0.3 V vs. RHE (red line) there is a simultaneous increase in H₂ signal (black line) confirming that the CuBi₂O₄/CdS/TiO₂/RuO_x heterojunction photoelectrode can evolve hydrogen photoelectrochemically. The faradaic efficiency was estimated to be 92% based on calibration of the DEMS using a Pt metal electrode.

Fig. 2 (a) Chopped LSV scans for a CuBi₂O₄/CdS/TiO₂/RuO_x (red), CuBi₂O₄/BiVO₄/TiO₂/RuO_x (green), CuBi₂O₄/Ga₂O₃/TiO₂/RuO_x (black) and CuBi₂O₄/TiO₂/RuO_x photocathode (blue) under simulated AM1.5 illumination (these are representative measurements of at least 4 samples for each condition). (b) Differential electrochemical mass spectrometry (DEMS) LSV scans for a CuBi₂O₄/CdS/TiO₂/RuO_x photocathode with illumination by a xenon lamp, showing current (red) and H₂ signal (black). Measurements were performed in three-electrode configuration in 0.3 M K₂SO₄ and 0.2 M phosphate buffer electrolyte (pH 6.8) with Ar bubbling.

Note that the photocurrent density directly before the onset of dark current from electrochemical proton reduction (ca. −1 mA cm² at 0 V vs. RHE in Fig. 2a) is close to the maximum photocurrent density for bare CuBi₂O₄ before the onset of dark current from electrochemical corrosion (ca. −1 mA cm⁻² at 0.4 V vs. RHE in Fig. S1a†). This implies that most of the electrons that are photogenerated in CuBi₂O₄, which previously participated in the corrosion reaction (Cu²⁺ → Cu¹⁺ and/or Cu), are successfully injected into the CdS buffer and TiO₂ protection layers, and eventually into the electrolyte to drive the hydrogen evolution reaction.

In order to investigate the long-term stability of the CuBi₂O₄/CdS/TiO₂/RuO_x photocathodes, the photocurrent density was measured at a constant potential of 0.0 V vs. RHE for 5 hours with intermittent chopping of the front illumination (see Fig. S2 in the ESI†). The measurement shows a decay in photocurrent which was more than 100 times slower than for the bare CuBi₂O₄ photocathode (Fig. S1b†). A significant part of the decay was attributed to deterioration of the RuO_x co-catalyst as demonstrated by partial restoration of the photocurrent after re-deposition of RuO_x after 3 hours. At the end of the 5 hour measurement, and after the single re-deposition of the RuO_x co-catalyst, 30% of the initial photocurrent density was preserved. Visually the illuminated area of the CuBi₂O₄/CdS/TiO₂/RuO_x photoelectrode was almost indistinguishable from the non-illuminated area (see the inset of Fig. S2†). Under scanning electron microscope (SEM) there were no significant morphological differences between areas of the sample that were and were not PEC tested other than a slight smoothing of the electrode surface after PEC testing (see Fig. S3†). Therefore, the ALD-deposited TiO₂ layer effectively blocked the contact between CuBi₂O₄ and the aqueous electrolyte and hindered photo-corrosion. This was also the case for the other layer systems. For comparison, photos and SEM images of the CuBi₂O₄/CdS/TiO₂, CuBi₂O₄/Ga₂O₃/TiO₂ and CuBi₂O₄/BiVO₄/TiO₂ photoelectrodes are shown in Fig. S4–S6,† respectively.

Band alignment

In order to estimate the band positions of the different materials in the heterojunction photoelectrodes, the energy band offsets were measured through a combination of X-ray photoelectron spectroscopy (XPS), ultraviolet photoelectron spectroscopy (UPS), UV-visible spectrophotometry, and Mott–Schottky analysis (including work from previous studies). Fig. S7a† shows the XPS spectrum for Cu at the surface of bare CuBi₂O₄. There are two main peaks at 933.8 and 953.6 eV corresponding to the Cu 2p_3/2 and 2p_1/2 levels, both of which were attributed to the presence of the Cu²⁺ state on the surface.³⁰ The binding energies of Cd 3d, Ga 3d, V 2p and Ti 2p core levels of CdS, Ga₂O₃, BiVO₄ and TiO₂ are measured to be 406.3, 21.2, 516.7 and 459.3 eV, respectively (see Fig. S7b–e†). Fig. S8† shows the XPS survey results for the stack samples of the CuBi₂O₄/CdS/TiO₂, CuBi₂O₄/Ga₂O₃/TiO₂, CuBi₂O₄/BiVO₄/TiO₂ and CuBi₂O₄/TiO₂. The presence of strong Ti peaks and the absence of Cu, Cd, Ga and V peaks in the survey spectra indicates that the ALD-deposited TiO₂ layer fully covers the buffer layers.

For the band positions of bare CuBi₂O₄, we used our previously reported values from Mott–Schottky analysis and UPS measurements, to estimate a conduction band (E_C) of −0.3 V vs. RHE and a valence band (E_V) of 1.2 V vs. RHE.^4,5,31 The flat-band potential (φ_fb) of our spray-deposited BiVO₄ films is 0.37 V vs. RHE,^5,32 which can then be used to estimate the Fermi level (E_F) while taking into account the potential drop across the Helmholtz layer.^5,33 Fig. S9a–c† show the UPS spectra of CdS, Ga₂O₃ and TiO₂ thin films on FTO substrate measured with a 2 V bias. The work function (Φ), defined as the difference between the vacuum energy level and Fermi level (E_F), can be derived from the low kinetic energy cut-off in the secondary emission feature. The photon energy of the UV source (He I discharge) was 21.21 eV. Given that the Fermi level at the surface of these overlayers is considered independently of the spectrometer, the work function is determined to be 21.21 – 2 – secondary emission cut-off (SEC). The work function of CdS, Ga₂O₃ and TiO₂ thin films are calculated to be 4.12, 3.63 and 4.57 eV, respectively. Using 4.5 eV vs. vacuum as the reference value for the electrochemical proton reduction (0.0 V vs. RHE) we can estimate the Fermi energies of the CdS, Ga₂O₃ and TiO₂ layers at approximately −0.38, −0.87 and 0.07 V vs. RHE, respectively. These values are in agreement with the literature.^34–36

The valence band positions with respect to the position of the Fermi level, E_F − E_V, were determined by linear extrapolation of the UPS spectrum at the low binding energy side to the binding energy axis (see Fig. S10†). E_F − E_V for CdS, BiVO₄, Ga₂O₃ and TiO₂ thin films are calculated to be 2.5, 2.4, 4.6, and 3.3 eV, respectively, which is consistent with previously reported values.^37,38 Since CdS has a bandgap of 2.4–2.5 eV this would place the conduction band very close to the Fermi level at −0.38 V vs. RHE, which is within the wide range of previously reported E_C values for CdS.^5,39,40 The experimentally determined band energy values for all samples are summarized in Table S1 in the ESI† along with values from the literature.

Using the values given in Table S1,† the energies of conduction and valence bands were obtained and illustrated in Fig. 3 for the individual layers in the CuBi₂O₄/CdS/TiO₂, CuBi₂O₄/Ga₂O₃/TiO₂ and CuBi₂O₄/BiVO₄/TiO₂ heterojunction photoelectrodes in relation to the electrochemical redox potentials for proton reduction (H⁺/H₂) and water oxidation (H⁺, O₂/H₂O) at 0.0 and 1.23 V vs. RHE, respectively. Since the conduction band of CuBi₂O₄ is at −0.3 V vs. RHE and the Fermi level of TiO₂ is at approximately 0.0 V vs. RHE, all of the heterojunctions shown in Fig. 3 should be thermodynamically capable of reducing H⁺ as long as photogenerated electrons can be transported to the TiO₂ surface. However, there are differences in the conduction band offsets between CuBi₂O₄ and each buffer layer (ΔE_C,1) and between TiO₂ and each buffer layer (ΔE_C,2). For a wide range of heterojunctions, including CuO/TiO₂ junctions, it has been shown that a high conduction band offset can promote high interface recombination and therefore inhibit charge transport.⁴¹ Better band alignment of the CuBi₂O₄/buffer layer provides a larger driving force for charge transport in CuBi₂O₄, results in a smaller concentration of holes near the interface and reduces the interfacial recombination. For all of the systems shown in Fig. 3, ΔE_C,1 is by far the smallest between CuBi₂O₄ and CdS (less than 0.2 eV) while it is approximately 0.57 eV between CuBi₂O₄ and BiVO₄ and –0.67 eV between CuBi₂O₄ and Ga₂O₃ assuming that E_C is approximately 0.1 eV above E_F in the buffer layers. The significantly higher conduction band of Ga₂O₃ compared to CuBi₂O₄ creates an energy barrier, thus significantly limiting the charge separation efficiency and the injection of the photogenerated electrons to the TiO₂, which increases the recombination rate of photogenerated electrons and holes at the CuBi₂O₄/Ga₂O₃/TiO₂ interfaces, especially if they contain defect states. This alone is enough to explain the poor performance of the CuBi₂O₄/Ga₂O₃/TiO₂/RuO_x photoelectrode compared to the others. For all of the heterojunctions shown in Fig. 3, ΔE_C,2 is also smallest with CdS as the buffer layer. For the CuBi₂O₄/BiVO₄/TiO₂ system, transport of photogenerated electrons from CuBi₂O₄ to BiVO₄ is thermodynamically favorable, but the ΔE_C,2 value of −0.27 eV may hinder transport of these electrons into TiO₂. Overall, the relatively small conduction band offsets between CuBi₂O₄, CdS, and TiO₂ makes the CuBi₂O₄/CdS/TiO₂ heterojunction photoelectrode more favorable for obtaining a high photovoltage and transporting charges compared to the CuBi₂O₄/Ga₂O₃/TiO₂/RuO_x and CuBi₂O₄/BiVO₄/TiO₂/RuO_x photoelectrodes.

Fig. 3 Estimated band diagrams for systems containing (a) CuBi₂O₄/CdS/TiO₂, (b) CuBi₂O₄/BiVO₄/TiO₂ and (c) CuBi₂O₄/Ga₂O₃/TiO₂ in relation to electrochemical redox potentials for proton reduction (H⁺/H₂) and water oxidation (H⁺, O₂/H₂O). E_F is the Fermi level, E_C is the conduction band, E_V is the valence band, and ΔE_C,1 and ΔE_C,2 are the conduction band offsets.

Based on these energy band diagrams the CuBi₂O₄/TiO₂/RuO_x photoelectrode is also expected to contain a relatively small band offset of approximately 0.3 eV between CuBi₂O₄ and TiO₂. Therefore, assuming negligible formation of interface dipoles, the performance of the CuBi₂O₄/TiO₂/RuO_x photoelectrode is expected to be comparable to CuBi₂O₄/CdS/TiO₂/RuO_x and higher than the CuBi₂O₄/BiVO₄/TiO₂/RuO_x heterojunction system. As shown in Fig. 2a, this is clearly not the case, suggesting that additional factors aside the band offsets probably play an important role in the heterojunction systems. One of these factors, common in the case of heterojunctions, is the formation of intermixed interface layers, which may serve as hole or electron traps and hinder charge transfer.

Surface photovoltage analysis

To understand the relation between the charge transfer kinetics in the various heterojunctions and their PEC performance, modulated surface photovoltage (SPV) spectroscopy in the fixed capacitor arrangement was utilized.²⁶ Aside from its very high sensitivity, modulated SPV spectroscopy can provide information about fast and slow (or retarded) processes in relation to the modulation period. Modulated SPV spectra are measured with double-phase lock-in amplifiers. The in-phase (x) and phase-shifted by 90° (y) signals are related to the fast and slow responses, respectively. In general, the response times of the SPV are much shorter (or longer) than the modulation period after switching on and off illumination if y (or x) are equal to 0. Furthermore, positive (or negative) x-signals are related to preferential separation of photogenerated electrons towards the bulk (or surface) of the absorber. The sign of the y-signal in relation to the sign of the x-signal gives information about the preferential direction of trapped charge carriers. If the x- and y-signals are of opposite sign, the direction of separated trapped charge carriers is similar to that of fast charge separation and relaxation. In contrast, if the x- and y-signals have the same sign, the directions of separated charge carriers are opposite for the dominating fast and slow processes (see Fig. S11† for a schematic illustration, and for more detailed explanations also paragraph 2.3.4. in ref. 26 or the ESI in ref. 42).

To start, in order to compare between the different samples in terms of their general SPV response (not separated to slow and fast components), the x- and y-signals can be converted into amplitudes. The amplitude is defined as the square root of the sum of the squared x- and y-signals. The amplitude spectra of the CuBi₂O₄/CdS/TiO₂, CuBi₂O₄/BiVO₄/TiO₂, CuBi₂O₄/TiO₂ and CuBi₂O₄/Ga₂O₃/TiO₂ systems are presented on a semi-log scale in Fig. 4. The extracted band gap of CuBi₂O₄, which is determined close to where the SPV deviates from exponential dependance, amounts to 1.55–1.6 eV, in agreement with our previous report.⁵ When comparing the maximum measured SPV amplitudes, we find that there is no direct correlation between the maximum SPV signal and the PEC performance, i.e. for the CuBi₂O₄/TiO₂ sample, the maximum SPV amplitude is the highest, however for this sample very low photocurrent values were obtained compared to the samples containing CdS or BiVO₄ buffer layers (see Fig. 2a). Analysis of the amplitude signals below the bandgap can yield valuable information regarding tail states and other defect states in the bulk of the absorber as well as in the interfaces with the different protection layers. If assuming that charge separation is caused only by directed transport of mobile charge carriers in delocalized states, and the SPV measurements are performed in the low-signal case and under homogeneous absorption within the charge separation and/or diffusion lengths, and the x- and y-signals do not change their signs within the corresponding spectral range, an exponential increase of the amplitude near the band gap or the first derivative of a logistic growth function are directly related to the exponential distribution of states near the valence (or conduction) band edge or to the distribution of deep defect states, respectively, from which photogeneration into delocalized states takes place.

Fig. 4 Amplitude spectra of CuBi₂O₄/CdS/TiO₂, CuBi₂O₄/BiVO₄/TiO₂, CuBi₂O₄/TiO₂ and CuBi₂O₄/Ga₂O₃/TiO₂ (black squares, red circles, blue triangles and green stars, respectively). The solid green and red, blue, and black lines are fits below the band gap with one exponential term, the sum of two exponential terms, and the sum of a logarithmic growth and an exponential function for CuBi₂O₄/Ga₂O₃/TiO₂ and CuBi₂O₄/BiVO₄/TiO₂, CuBi₂O₄/TiO₂, and CuBi₂O₄/CdS/TiO₂, respectively. The two vertical dashed lines mark the approximate band gap range of CuBi₂O₄.

Thus, under the above assumptions, we find the following fitting functions were needed to obtain a good fit, depending on the sample:

For the CuBi₂O₄/BiVO₄/TiO₂ and CuBi₂O₄/Ga₂O₃/TiO₂ samples, a single exponential term (eqn (1)) was sufficient. For the CuBi₂O₄/TiO₂, a sum of two exponential terms was required (eqn (2)), and for the CuBi₂O₄/CdS/TiO₂ a sum of a logarithmic growth and an exponential function was required (eqn (3)). An identical band gap of 1.6 eV was used for all fits of the increasing part of the amplitude spectra. The main, common exponentials (eqn (1)) were characterized by the same tail energy (E_t,1) equal to 83 meV in all equations (we note that E_t is related, yet not equal to the so-called Urbach tail which is experimentally obtained from optical absorption). This gives evidence that E_t,1 is related to disorder-induced defect states near the band gap of the bulk of CuBi₂O₄ and that this disorder was not affected by the deposition of the different protection layers on top. Furthermore, the values of A₁ amounted to 20, 36, 130 and 15.5 μV for CuBi₂O₄/Ga₂O₃/TiO₂, CuBi₂O₄/BiVO₄/TiO₂, CuBi₂O₄/TiO₂ and CuBi₂O₄/CdS/TiO₂, respectively. Therefore, the strongest charge separation from defect states near the band gap of CuBi₂O₄ appeared in CuBi₂O₄/TiO₂.

The values of E_t,2 and A₂ were 35 meV and 600 μV, respectively. The fact that a second exponential term was required for fitting the increase of the SPV amplitude below the band gap of CuBi₂O₄ in CuBi₂O₄/TiO₂ gives evidence for the formation of an interface region with efficient absorption and modulated charge separation at reduced disorder near the band gap of CuBi₂O₄.

The logarithmic growth function in eqn (3) is related to the excitation and separation of mobile charge carriers due to absorption in relatively deep defect states (where E_d is denotes the maximum DOS of the defect distribution within the bandgap). The obtained E_d value amounts to 1.19 eV, about 400 meV within the bandgap of the CuBi₂O₄, in agreement with the deep defects observed in our previous work for this type of junction.⁵ These deep defect states were specific for the CuBi₂O₄/CdS interface since the defect related feature appeared only in the spectrum of CuBi₂O₄/CdS/TiO₂. It can therefore be concluded that partial inter-diffusion took place at the CuBi₂O₄/CdS interface leading to the formation of deep interfacial states. In spite of these additional deep defect states that were found only in the CuBi₂O₄/CdS/TiO₂ sample, this sample had the highest PEC performance. Thus, in order to a gain deeper understanding of how the modulated charge separation and recombination processes affect and correlate to the actual PEC performance, it is essential to understand exactly which type of free and/or trapped charge carriers are accumulated at the different interfaces, by performing a detailed analysis of the individual x and y components of the SPV signals, as shown next.

Fig. 5 shows the modulated SPV spectra of the in-phase and phase-shifted by 90° signals for the CuBi₂O₄/CdS/TiO₂, CuBi₂O₄/BiVO₄/TiO₂, CuBi₂O₄/TiO₂ and CuBi₂O₄/Ga₂O₃/TiO₂, samples. Starting with the in-phase signals, that give indication into which type of free charge carriers accumulate closer to the sample's surface, it is found that in the absorption range of CuBi₂O₄, the signs of the in-phase signals where negative for CuBi₂O₄/CdS/TiO₂, CuBi₂O₄/BiVO₄/TiO₂, CuBi₂O₄/TiO₂ and bare CuBi₂O₄ with an uncontrolled surface. This finding gives evidence to separation of electrons towards the surface (as expected for a p-type absorber with an electron-selective contact or a p-type semiconductor with a depletion region near the surface), which is favorable for a photoreduction reaction such as proton reduction to take place at the surface. Incidentally, the signals of the bare CuBi₂O₄ layer with an uncontrolled surface were extremely low (only on the order of a μV) in comparison to the CuBi₂O₄/CdS/TiO₂, CuBi₂O₄/BiVO₄/TiO₂, and CuBi₂O₄/TiO₂ samples. This suggests that without an appropriate charge extraction layer, the surface of the CuBi₂O₄ is not passivated, and/or the charge separation efficiency within the CuBi₂O₄ depletion region is rather low. In contrast, for the CuBi₂O₄/Ga₂O₃/TiO₂ sample, the sign of the in-phase signals was positive and only on the order of several μV. The positive sign of the in-phase signals indicates separation of holes towards the surface, however due to the very low signal, the charge separation efficiency in this case is quite low (as shown schematically in Fig. 7d), explaining the poor PEC performance of the CuBi₂O₄/Ga₂O₃/TiO₂/RuO_x heterojunction as a photocathode (Fig. 2a). This observed behavior is probably due to the unfavorable band alignment already discussed and shown in Fig. 3c.

Fig. 5 Modulated SPV spectra of the in-phase and phase-shifted by 90° signals (black and red solid lines, respectively) for the CuBi₂O₄/CdS/TiO₂, CuBi₂O₄/BiVO₄/TiO₂, CuBi₂O₄/TiO₂ and CuBi₂O₄/Ga₂O₃/TiO₂, samples ((a)–(d), respectively). For comparison, the thin dashed black and red lines in (d) represent the measured in-phase and phase-shifted by 90° signals for a bare CuBi₂O₄ layer with uncontrolled surface.

With regards to the phase-shifted by 90° signals, positive phase-shifted by 90° signals were found for CuBi₂O₄/CdS/TiO₂ and CuBi₂O₄/Ga₂O₃/TiO₂, suggesting that for these two samples, electrons that were separated towards the external surface were trapped in the range of the CuBi₂O₄/CdS or the CuBi₂O₄/Ga₂O₃ interfaces. The electron traps in the CuBi₂O₄/CdS interface can probably be attributed to the deep defect states seen in the amplitude spectrum of the CuBi₂O₄/CdS/TiO₂ sample in the energy region of 1–1.5 eV. In contrast, the signs of the phase-shifted by 90° signals were negative for CuBi₂O₄/BiVO₄/TiO₂ and CuBi₂O₄/TiO₂, i.e. holes were separated towards the external surface and trapped. This finding suggests that for these two heterojunction types, predominant trap states for holes were formed at CuBi₂O₄/BiVO₄ and CuBi₂O₄/TiO₂ interfaces, as shown schematically in Fig. 5b and c. Such an accumulation of trapped holes at the interface with CuBi₂O₄, compared to trapped electrons in the case of CuBi₂O₄/CdS, is much more detrimental for the PEC performance since it will lead to an increase in the total number of photogenerated electrons that will (non-radiatively) recombine. This explains the lower observed photocurrents for CuBi₂O₄/TiO₂ heterojunction in Fig. 2a compared to CuBi₂O₄/CdS/TiO₂.

Although in both CuBi₂O₄/BiVO₄/TiO₂ and CuBi₂O₄/TiO₂ a combination of negative in-phase and phase-shifted by 90° signals was observed, the PEC performance of the CuBi₂O₄/TiO₂ sample was extremely poor compared to the sample with the BiVO₄ buffer layer, which requires explanation. This behavior can be attributed to an additional, pronounced difference that is observed with and without the BiVO₄ buffer layer: for the CuBi₂O₄/TiO₂ sample, in the region between 2.5–3.5 eV, the phase-shifted by 90° signal becomes larger than the in-phase signal, and the in-phase signal changes sign. Such pronounced qualitative differences can be analyzed quantitively by considering the so-called phase angle of the signal. The in-phase (x) and quadrature (y) components of the signal can be converted into the phase angle, which is defined as the arctan of the ratio between the y- and x-signals. Since the behavior of the x- and y-signals depends sensitively on transport and charge transfer processes, changes in the phase angle can provide information about changes in dominating processes in modulated charge separation. Spectra of the phase angles of CuBi₂O₄/CdS/TiO₂, CuBi₂O₄/Ga₂O₃/TiO₂, CuBi₂O₄/BiVO₄/TiO₂ and CuBi₂O₄/TiO₂ are given in Fig. 6.

Fig. 6 Spectra of the phase angles of CuBi₂O₄/CdS/TiO₂ and CuBi₂O₄/Ga₂O₃/TiO₂ ((a), black squares and green stars, respectively) and of CuBi₂O₄/BiVO₄/TiO₂ and CuBi₂O₄/TiO₂ ((b), red circles and blue triangles, respectively).

For CuBi₂O₄/CdS/TiO₂, the phase angles changed from about 140° at 1.4 eV to 119° at 2.6 eV and to 124° at photon energies above 2.85 eV. Between 2.2 and 2.6 eV, the phase angle peaked slightly from 124° to 125° at 2.4 eV (band gap of CdS), decreased to the minimum of 119° at 2.6 eV, and then increased to its saturation value after that. Therefore, since those changes were small, absorption of light by defect states near the band gap of CdS and fundamental absorption in CdS led only to little modification in trapping.

For CuBi₂O₄/Ga₂O₃/TiO₂ and CuBi₂O₄/BiVO₄/TiO₂, the phase angles changed from about 75° to 86° and from about 189° to 196°, respectively. No well-defined signatures could be found within those weak changes. Therefore, deposition of Ga₂O₃ or BiVO₄ onto CuBi₂O₄ does not result in the formation of additional transitions that influence the trapping dynamics.

In contrast, drastic changes were observed in the spectrum of the phase angles for CuBi₂O₄/TiO₂. At photon energies between 1.45 and 2.15 eV, the phase angle increased from about 182° to 192°. The fact that the phase angles were so close to 180° means that the forward and backward electron transfer was the fastest at the CuBi₂O₄/TiO₂ interface. Between 2.15 and 2.7 eV, the phase angle increased strongly to 224° which means that the electron transfer became very slow in relation to the modulation period, i.e. strong electron trapping set in. The very strong change of the phase angles up to values exceeding 270° (296° at 3.05–3.10 eV) shows that a large change in the modulated charge separation occurs at these photon energies. Specifically, a phase angle around 270° indicates that the photo-generated holes move towards the external surface and/or that the photo-generated electrons move towards the CuBi₂O₄/TiO₂ interface. This explains the very poor performance of the CuBi₂O₄/TiO₂ photocathode sample without a buffer layer. At photon energies 3.1 eV, the phase angle changed towards lower values (244° at 3.6 eV) giving evidence for an increased influence of the separation of electrons photo-generated in TiO₂ towards the external surface. The very strong change of the phase angles for CuBi₂O₄/TiO₂ can only be interpreted by the formation of defect states within the band gap of TiO₂ near the CuBi₂O₄/TiO₂ interface, probably caused by inter-diffusion and/or partial reduction of TiO₂ by transferred and trapped holes. This notable difference in the dynamics and distribution of the charge carriers with and without the BiVO₄ buffer layer, illustrated in Fig. 7b and c, demonstrates the need to prevent direct contact between the CuBi₂O₄ and the TiO₂.

Fig. 7 Schematic model illustrating the distribution of charge carriers in the different heterojunctions upon illumination with white light: (a) CuBi₂O₄/CdS/TiO₂; (b) CuBi₂O₄/BiVO₄/TiO₂; (c) CuBi₂O₄/TiO₂ and (d) CuBi₂O₄/Ga₂O₃/TiO₂.

Based on the analysis of the SPV results and the correlation found with the PEC performance of the different heterojunctions, a schematic model illustrating the distribution of charge carriers in the different heterojunctions upon illumination with white light is shown in Fig. 7. The CuBi₂O₄/CdS/TiO₂ sample (Fig. 7a) shows the highest PEC performance, which we attribute to the uniform separation of free and trapped electrons toward the TiO₂ surface. When the CdS layer is replaced with BiVO₄ (Fig. 7b), interfacial hole traps are formed at the CuBi₂O₄/BiVO₄ interface, enhancing non-radiative recombination of electron–hole pairs and resulting in lower photocurrents. In the case of direct contact between the CuBi₂O₄ and TiO₂ (Fig. 7c), large defect-related SPV signals are observed, and a large majority of the photogenerated electrons and holes are trapped at the CuBi₂O₄/TiO₂ interface, resulting in poor PEC performance. For the case of CuBi₂O₄/Ga₂O₃/TiO₂ (Fig. 7d), the charge separation efficiency is quite low compared to the other heterojunctions, probably due to the large energetic barriers due to the higher conduction band of Ga₂O₃ shown in Fig. 3. The charge separation direction is not favorable for photocathodic reactions, explaining the very poor PEC performance observed. Overall, based on the modulated SPV results, it can be concluded that one of the most notable differences between the different heterojunctions is the lack of trapped holes in the heterojunction containing the CdS buffer layer. Hence, we conclude that trapped holes at the interface with the CuBi₂O₄ are the main cause for interfacial recombination and poor photocurrents.

Conclusions

In this work, we showed that photocurrents generated from bare, unprotected CuBi₂O₄ photocathodes were mainly due to photo-corrosion of CuBi₂O₄ and not from H₂ production, as claimed by many reports on Cu-based photocathodes. We investigated the influence of different buffer layers between the CuBi₂O₄ absorber and an n-type TiO₂ film that serves as a protection layer. We found that a CdS buffer layer in combination with the ALD-deposited TiO₂ protection layer and RuO_x co-catalyst layer yielded a stable photoelectrode with the highest photocurrent density and faradaic efficiency for the hydrogen evolution reaction. In contrast, neither high photocurrent nor efficient hydrogen evolution was obtained for CuBi₂O₄/Ga₂O₃/TiO₂, CuBi₂O₄/BiVO₄/TiO₂ and CuBi₂O₄/TiO₂ heterojunctions. However, band alignment considerations alone cannot explain the observed trends in photoelectrochemical performance. Therefore, modulated surface photovoltage measurements were used to investigate the mechanism governing the charge transport in these heterojunctions. The modulated SPV results strongly correlated with the observed trend of the photoelectrochemical performance and revealed the formation of different interfacial states, depending on the CuBi₂O₄/buffer layer junction. Most importantly, it revealed which type of charge carrier was trapped at the interface with the CuBi₂O₄ suggesting that trapping of holes near the interface strongly limits the photo-electrochemical performance of the heterojunctions.

Author contributions

The manuscript was written through contributions of all authors. All authors have given approval to the final version of the manuscript.

Conflicts of interest

There are no conflicts to declare.

Acknowledgements

Angang Song thanks the China Scholarship Council (file no. 201607040078) for financial support. We thank Christian Höhn for assistance with XPS and UPS measurements. We thank Lars Steinkopf for preparing CdS buffer layer and Matthias J. Müller for depositing the TiO₂ protection layer by ALD.

Notes and references

1. S. P. Berglund, F. F. Abdi, P. Bogdanoff, A. Chemseddine, D. Friedrich and R. van de Krol, Chem. Mater., 2016, 28, 4231–4242.
2. D. Kang, J. C. Hill, Y. Park and K.-S. Choi, Chem. Mater., 2016, 28, 4331–4340.
3. R. Patil, S. Kelkar, R. Naphade and S. Ogale, J. Mater. Chem. A, 2014, 2, 3661–3668.
4. A. Song, P. Plate, A. Chemseddine, F. Wang, F. F. Abdi, M. Wollgarten, R. van de Krol and S. P. Berglund, J. Mater. Chem. A, 2019, 7, 9183–9194.
5. A. Song, P. Bogdanoff, A. Esau, I. Y. Ahmet, I. Levine, T. Dittrich, T. Unold, R. van de Krol and S. P. Berglund, ACS Appl. Mater. Interfaces, 2020, 12, 13959–13970.
6. H. Gerischer, J. Electroanal. Chem. Interfacial Electrochem., 1977, 82, 133–143.
7. S. Chen and L.-W. Wang, Chem. Mater., 2012, 24, 3659–3666.
8. Y. Xu, J. Jian, F. Li, W. Liu, L. Jia and H. Wang, J. Mater. Chem. A, 2019, 7, 21997–22004.
9. P. Cendula, M. T. Mayer, J. Luo and M. Grätzel, Sustainable Energy Fuels, 2019, 3, 2633–2641.
10. Y. W. Chen, J. D. Prange, S. Dühnen, Y. Park, M. Gunji, C. E. D. Chidsey and P. C. McIntyre, Nat. Mater., 2011, 10, 539–544.
11. B. Seger, T. Pedersen, A. B. Laursen, P. C. K. Vesborg, O. Hansen and I. Chorkendorff, J. Am. Chem. Soc., 2013, 135, 1057–1064.
12. R. E. Brandt, M. Young, H. H. Park, A. Dameron, D. Chua, Y. S. Lee, G. Teeter, R. G. Gordon and T. Buonassisi, Appl. Phys. Lett., 2014, 105, 263901.
13. S. Hu, M. R. Shaner, J. A. Beardslee, M. Lichterman, B. S. Brunschwig and N. S. Lewis, Science, 2014, 344, 1005.
14. A. Paracchino, N. Mathews, T. Hisatomi, M. Stefik, S. D. Tilley and M. Grätzel, Energy Environ. Sci., 2012, 5, 8673–8681.
15. A. Paracchino, V. Laporte, K. Sivula, M. Grätzel and E. Thimsen, Nat. Mater., 2011, 10, 456–461.
16. S. D. Tilley, M. Schreier, J. Azevedo, M. Stefik and M. Graetzel, Adv. Funct. Mater., 2014, 24, 303–311.
17. C. G. Morales-Guio, S. D. Tilley, H. Vrubel, M. Grätzel and X. Hu, Nat. Commun., 2014, 5, 3059.
18. J. Luo, L. Steier, M.-K. Son, M. Schreier, M. T. Mayer and M. Grätzel, Nano Lett., 2016, 16, 1848–1857.
19. P. Dai, W. Li, J. Xie, Y. He, J. Thorne, G. McMahon, J. Zhan and D. Wang, Angew. Chem., 2014, 53, 13493–13497.
20. W. Cui, W. Niu, R. Wick-Joliat, T. Moehl and S. D. Tilley, Chem. Sci., 2018, 9, 6062–6067.
21. L. Pan, J. H. Kim, M. T. Mayer, M.-K. Son, A. Ummadisingu, J. S. Lee, A. Hagfeldt, J. Luo and M. Grätzel, Nat. Catal., 2018, 1, 412–420.
22. C. Li, T. Hisatomi, O. Watanabe, M. Nakabayashi, N. Shibata, K. Domen and J. J. Delaunay, Energy Environ. Sci., 2015, 8, 1493–1500.
23. W. Niu, T. Moehl, W. Cui, R. Wick-Joliat, L. Zhu and S. D. Tilley, Adv. Energy Mater., 2018, 8, 1702323.
24. W. Septina, R. R. Prabhakar, R. Wick, T. Moehl and S. D. Tilley, Chem. Mater., 2017, 29, 1735–1743.
25. V. Duzhko, V. Y. Timoshenko, F. Koch and T. Dittrich, Phys. Rev. B, 2001, 64, 075204.
26. T. Dittrich and S. Fengler, Surface Photovoltage Analysis of Photoactive Materials, 2020.
27. Y. H. Choi, K. D. Yang, D.-H. Kim, K. T. Nam and S. H. Hong, Mater. Lett., 2017, 188, 192–196.
28. J. Lee, H. Yoon, S. Kim, S. Seo, J. Song, B. U. Choi, S. Y. Choi, H. Park, S. Ryu, J. Oh and S. Lee, Chem. Commun., 2019, 55, 12447–12450.
29. G. Sharma, Z. Zhao, P. Sarker, B. A. Nail, J. Wang, M. N. Huda and F. E. Osterloh, J. Mater. Chem. A, 2016, 4, 2936–2942.
30. A. Song, A. Chemseddine, I. Y. Ahmet, P. Bogdanoff, D. Friedrich, F. F. Abdi, S. P. Berglund and R. van de Krol, Chem. Mater., 2020, 32, 2408–2419.
31. R. Gottesman, A. Song, I. Levine, M. Krause, A. T. M. N. Islam, D. Abou Ras, T. Dittrich, R. van de Krol and A. Chemseddine, Adv. Funct. Mater., 2020, 1910832.
32. F. F. Abdi, L. Han, A. H. M. Smets, M. Zeman, B. Dam and R. van de Krol, Nat. Commun., 2013, 4, 2195.
33. M. Butler and D. Ginley, J. Electrochem. Soc., 1978, 125, 228–232.
34. A. Orendorz, J. Wüsten, C. Ziegler and H. Gnaser, Appl. Surf. Sci., 2005, 252, 85–88.
35. J. M. Bolts and M. S. Wrighton, J. Phys. Chem., 1976, 80, 2641–2645.
36. M. Mohamed, K. Irmscher, C. Janowitz, Z. Galazka, R. Manzke and R. Fornari, Appl. Phys. Lett., 2012, 101, 132106.
37. J. Resasco, H. Zhang, N. Kornienko, N. Becknell, H. Lee, J. Guo, A. L. Briseno and P. Yang, ACS Cent. Sci., 2016, 2, 80–88.
38. K. H. Ye, H. Li, D. Huang, S. Xiao, W. Qiu, M. Li, Y. Hu, W. Mai, H. Ji and S. Yang, Nat. Commun., 2019, 10, 3687.
39. V. Stevanović, S. Lany, D. S. Ginley, W. Tumas and A. Zunger, Phys. Chem. Chem. Phys., 2014, 16, 3706–3714.
40. T. Feurer, P. Reinhard, E. Avancini, B. Bissig, J. Löckinger, P. Fuchs, R. Carron, T. P. Weiss, J. Perrenoud, S. Stutterheim, S. Buecheler and A. N. Tiwari, Prog. Photovoltaics, 2017, 25, 645–667.
41. J. Morasch, H. F. Wardenga, W. Jaegermann and A. Klein, Phys. Status Solidi A, 2016, 213, 1615–1624.
42. P. Prajongtat and T. Dittrich, J. Phys. Chem. C, 2015, 119, 9926–9933.

---

Footnote

† Electronic supplementary information (ESI) available. See DOI: 10.1039/d0sc03030a

---