Interface engineering of Ta3N5 thin film photoanode for highly efficient photoelectrochemical water splitting

Interface engineering is a proven strategy to improve the efficiency of thin film semiconductor based solar energy conversion devices. Ta3N5 thin film photoanode is a promising candidate for photoelectrochemical (PEC) water splitting. Yet, a concerted effort to engineer both the bottom and top interfaces of Ta3N5 thin film photoanode is still lacking. Here, we employ n-type In:GaN and p-type Mg:GaN to modify the bottom and top interfaces of Ta3N5 thin film photoanode, respectively. The obtained In:GaN/Ta3N5/Mg:GaN heterojunction photoanode shows enhanced bulk carrier separation capability and better injection efficiency at photoanode/electrolyte interface, which lead to a record-high applied bias photon-to-current efficiency of 3.46% for Ta3N5-based photoanode. Furthermore, the roles of the In:GaN and Mg:GaN layers are distinguished through mechanistic studies. While the In:GaN layer contributes mainly to the enhanced bulk charge separation efficiency, the Mg:GaN layer improves the surface charge inject efficiency. This work demonstrates the crucial role of proper interface engineering for thin film-based photoanode in achieving efficient PEC water splitting. Solar-to-fuel energy conversion requires well-designed materials properties to ensure favorable charge carrier movement. Here, authors employ interface engineering of Ta3N5 thin film to enhance bulk carrier separation and interface carrier injection to improve the water-splitting efficiency.

A rtificial photosynthesis that converts solar energy into sustainable energy and fuels is a highly desired solution to meet the increasing global energy demand and address the environmental issues related to the consumption of fossil fuels [1][2][3][4][5][6][7] . Semiconductor materials that absorb sunlight to generate electron-hole pairs are at the core of artificial photosynthetic devices [8][9][10][11] . Suppressing the bulk and interfacial nonradiative charge recombination in the semiconductor light absorbers is essential to achieve a high solar-to-fuel conversion efficiency [12][13][14][15][16][17] . Especially, the semiconductor/electrode and semiconductor/electrolyte interfaces play a decisive role in device performance because photogenerated charges must cross these interfaces to participate in the photosynthetic reactions. Most defects at the interfaces are high-dimensional deep-level defects, which usually have higher capture cross-sections than those of the point defects in the bulk of semiconductor 18 . Therefore, the nonradiative charge recombination at interfacial defect sites is detrimental to the device performance. In addition to interfacial defects, energy level alignment at the interfaces is also a key factor that affects the final device performance. Nonideal energy level alignment not only reduces device efficiency because of inefficient interfacial charge transfer, but it also deteriorates device stability due to interfacial charge accumulation which could promote self-oxidation/reduction of the semiconductor. Therefore, effective carrier management through interface engineering is essential to improve the efficiency and stability of photoelectrode for artificial photosynthesis [17][18][19] .
Over the past decade, tremendous efforts have been devoted to interface engineering of various semiconductor photoelectrodes (e.g., Si 20,21 , α-Fe 2 O 3 22 , BiVO 4 23 , Cu 2 O 24 ) to improve their efficiency and stability through the passivation of interfacial defects or formation of hetero-/homojunctions. Ideally, the semiconductor thin film light absorbers should be sandwiched by a n-type electron transport layer (ETL) and a p-type hole transport layer (HTL) to achieve efficient charge separation, similar to the "n-i-p" device architecture commonly used in thin-film photovoltaics 25 . However, this type of "n-i-p" device architecture is rarely adopted in photoelectrochemical (PEC) devices 26 , especially for some of the emerging PEC materials. As one of the most promising photoanode materials for PEC water oxidation, Ta 3 N 5 has a high theoretical solar-to-hydrogen (STH) efficiency of 15.9% owing to its wide absorption range (up to 600 nm) and proper band positions for water splitting. However, the full thermodynamic potential of Ta 3 N 5 has yet to be unlocked. Although photocurrent approaching its theoretical value has been achieved 14 , the onset potential (usually >0.5 V vs. RHE) is still far higher than its theoretical limit (<0 V vs. RHE). Despite that various strategies have been employed to improve the bulk charge transfer efficiency, the applied bias photon-to-current efficiency (ABPE) of Ta 3 N 5 -based photoanode is still below 3.31% 27,28 . For pristine Ta 3 N 5 thin film based photoanode, the ABPE is still limited to 2.25% 29 . Promising results have been achieved by interface engineering of Ta 3 N 5 photoanodes with TiO x 14 , AlO x 30 , and GaN 15,31,32 to enhance the photocurrent, lower the onset potential, and improve the long-term stability. However, these previous studies only focus on either the Ta 3 N 5 /electrode or the Ta 3 N 5 /electrolyte interfaces. It is still lacking a concerted effort to engineer both interfaces to form a "n-i-p" device architecture that mimics the sandwich structure in perovskite solar cells. Such a synergistic interface engineering strategy is expected to further improve the efficiency of Ta 3 N 5 thin film based photoanode.
Herein, we propose to synergistically engineer the Ta 3 N 5 / electrode and Ta 3 N 5 /electrolyte interfaces with an n-type indiumdoped gallium nitride (In:GaN) and a p-type magnesium-doped gallium nitride (Mg:GaN), respectively. The realization of the In:GaN/Ta 3 N 5 /Mg:GaN "n-i-p" heterostructure is possible because of the all-nitride compositions of the heterostructure that can be obtained by a single-step thermal nitridation process of their oxide precursors. The In:GaN and Mg:GaN interfacial layers not only facilitate selective charge extraction from Ta 3 N 5 because of the desired energy band alignment of the formed "n-i-p" heterojunctions, they also passivate interfacial traps due to the formation of lattice-matched interfaces between Ta 3 N 5 and the GaN layers. As a result, the ABPE is significantly improved from 2.29% for the pristine Ta 3 N 5 photoanode to a record-high value of 3.46% for the In:GaN/Ta 3 N 5 /Mg:GaN photoanode, when both are modified with a highly active oxygen evolution reaction (OER) co-catalyst. Detailed mechanistic study reveals that the In:GaN layer mainly contributes to the enhanced bulk charge separation efficiency by selectively extracting photogenerated electrons from Ta 3 N 5 through a mid-gap band generated by In doping in GaN. On the other hand, the Mg:GaN layer mainly contributes to improved surface charge injection efficiency by passivating surface traps in Ta 3 N 5 . These results demonstrate that interface engineering of semiconductor light absorbers with proper materials to construct band-aligned heterojunctions and passivate interface defects is an effective strategy to improve the solar-to-fuel conversion efficiency of artificial photosynthetic devices 33 .

Results and discussion
Structural characterizations of In:GaN/Ta 3 N 5 /Mg:GaN heterostructure. In:GaN/Ta 3 N 5 /Mg:GaN heterostructure films were fabricated by one-step thermal nitridation process of InO x -GaO x / TaO x /Mg:GaO x thin films deposited by electron-beam (EB) evaporation and atomic layer deposition (ALD). Figure 1 shows the schematic diagram for the preparation of In:GaN/Ta 3 N 5 /Mg:GaN thin films. The detailed procedures are given in Methods. The multilayer structure of the InO x -GaO x /TaO x /Mg:GaO x precursor film was resolved by using Auger electron spectroscopy (AES) in Supplementary Fig. 1. The scanning electron microscopy (SEM) images and X-ray diffraction (XRD) pattern of the prepared In:GaN/Ta 3 N 5 /Mg:GaN thin film on Nb substrate are shown in Supplementary Fig. 2. Polycrystalline film with grain size of several hundred nanometers and some small pores can been seen within the grains, which is characteristic of Ta 3 N 5 thin film converted from TaO x precursor film due to volume contraction in the nitridation process 29 .The layered structure was not directly observed from the cross-sectional SEM image because of the thin thicknesses of the In:GaN (~5 nm) and Mg:GaN (~20 nm) layers and the lack of contrast between the three layers formed in a single-step nitridation process. Scanning transmission electron microscopy (STEM) was employed to further characterize the structure of the sample. Figure 2a shows the STEM energy dispersive x-ray spectroscopy (EDS) elemental mapping of the cross section of In:GaN/Ta 3 N 5 /Mg:GaN film deposited on Nb substrate and Fig. 2b shows the corresponding annular dark-field (ADF) STEM image. The STEM image and the EDS mapping of the metallic elements were overlapped in Fig. 2c. The thickness of the Ta 3 N 5 layer is approximately 600 nm. A thin Ga-rich layer (~20 nm) on the top of the Ta 3 N 5 layer was clearly revealed by the EDS mapping. However, the In:GaN layer at the bottom interface of the Ta 3 N 5 layer was not clearly observed by EDS mapping. Therefore, AES cross sectional mapping and depth profile was employed to further characterize the sample (Supplementary Fig. 3). Similar to the EDS mapping results, AES mapping also only revealed the top Ga-rich layer, but not the bottom one. Nevertheless, AES depth profile clearly revealed the presence of a Ga-containing layer at the bottom interface of the Ta 3 N 5 layer. It should be noted that Mg and In elements were also not detectable by either EDS or AES due to their low concentrations. The homogenous mapping results of Mg and In in Fig. 2a were likely resulted from noise spectra of the EDS instrument. Figure   Spectroscopic characterizations. XRD patterns of Ta 3 N 5 -based films with different layered structures deposited on quartz glass are exhibited in Fig. 3a. The main diffraction peaks can all be assigned to those of Ta 3 N 5 . No obvious change in the diffraction peaks is observed for films with different layered structures, indicating that the crystallinity of the Ta 3 N 5 layer is well maintained after adding the GaN layers. However, a close-up of the peak at~35°in the right panel of Fig. 3a reveals a distinct shoulder peak at~34.7°, corresponding to the diffraction peak of GaN (002), when Mg:GaN layer is formed on top of the Ta 3 N 5 film 15 . This indicates that these GaN-based layers have been successfully synthesized in a single-step nitridation process. To further confirm that the GaN layers can be formed by nitridation of GaO x precursor films, In:GaN and Mg:GaN thin films were deposited directly on quartz glass substrates by similar processes.
The XRD patterns of the In:GaN and Mg:GaN films in Supplementary Fig. 4 match that of hexagonal GaN (JPCDS no. 50-0792), while a weak peak belonging to GaInO 3 impurity phase at 35.1°is found in the In:GaN film. A slight peak shift to lower angle is observed the In:GaN film, indicating In is incorporated into GaN lattice because the ionic radius of In 3+ is large than that of Ga 3+ . The presence of Mg and In in the synthesized GaN films is further confirmed by X-ray photoelectron spectroscopy (XPS) results (Supplementary Figs. 5 and 6). Photoluminescence (PL) spectroscopy was employed to investigate the fluorescence properties of Mg:GaN and In:GaN films (Fig. 3b, c). The PL spectrum of Mg:GaN film can be deconvolved into two sub-peaks centered at~3.1 and~2.8 eV, which are typical fluorescence emission peaks related to the isolated Mg Ga centers and Mgrelated deep levels, respectively 35 . In contrast, the PL spectrum of In:GaN film shows a emission peak at~1.65 eV, which suggests In-doping induces mid-gap states in GaN. The In-induced midgap states could be beneficial to electron transport from the   Ta 3 N 5 light absorber through the In:GaN layer to the Nb electrode, as shown below. The band positions of the Mg:GaN and In:GaN films were determined using ultraviolet photoelectron spectroscopy (UPS) (Fig. 3d, e). By subtracting the cut-off energies (E Cutoff ) of the secondary electrons from the He I excitation energy (21.22 eV), the Fermi levels of Mg:GaN and In:GaN were obtained at 3.61 and 3.86 eV below the vacuum level, respectively. The lowbinding energy edges of the UPS spectra reveal that the valence bands of Mg:GaN and In:GaN are 1.99 and 2.02 eV below their Fermi levels, respectively. The Tauc plots of the absorption spectra revealed that the optical bandgaps of the In:GaN and Mg:GaN films are both~3.30 eV (Supplementary Fig. 7). Combining these data with the band positions of pristine Ta 3 N 5 obtained in our previous study 27 , the band alignment of the Nb/In:GaN/Ta 3 N 5 /Mg:GaN layered structure is obtained and shown in Fig. 3f. The band alignment shows that pristine GaN is not suitable for electron extraction from Ta 3 N 5 because its conduction band lies too high above that of Ta 3 N 5 . However, Indoping induces a mid-gap band at around −4.23 eV in In:GaN, which could serve as a conductive pathway for electrons. The ability of the In-induced inter-gap state to act as a channel for electrons was further verified by measuring cyclic voltammetry of a compact In:GaN film in the presence of a scavenger ( Supplementary Fig. 8). Such a conduction mechanism through inter-gap defect states has previously been demonstrated with amorphous TiO 2 coated on Si, GaAs, and GaP photoanodes 1 . Consequently, the photon-generated electrons and holes in the Ta 3 N 5 layer can be efficiently extracted through the In:GaN and Mg:GaN layers, respectively.
To reveal the effect of the In:GaN underlayer and Mg:GaN overlayer on the carrier dynamics of Ta 3 N 5 , low-temperature and time-resolved PL (TRPL) spectroscopies were employed to study samples with different layered structures. Due to the strong optical anisotropy of its orthorhombic lattice, Ta 3 N 5 shows two intrinsic bandgaps of~2.1 eV along the a-axis and ∼2.6 eV along the b-and c-axes 29 . We found that at low temperature (10 K) under 375-nm laser excitation the PL of Ta 3 N 5 was dominated by an emission peak at~475 nm (~2.6 eV), corresponding to the near band edge radiative recombination along the b-and c-axes. Therefore, the intensities and lifetimes of this peak were measured to probe the carrier dynamics in Ta 3 N 5 -based samples with different layered structures. The steady-state PL spectra in Fig. 3g show that the PL intensity is quenched when Ta 3 N 5 film is modified with either In:GaN or Mg:GaN layers, suggesting their ability to extract photocarriers from Ta 3 N 5 . Moreover, the TRPL results reveal that the In:GaN and Mg:GaN layers also play a role in defect passivation besides carrier extraction. All the TRPL decay curves in Fig. 3h are well-fitted with a stretchedexponential decay (Supplementary Fig. 9). This decay law is typically observed in disordered systems and it is believed to be due to the dispersion diffusion of photoexcited carriers 36 . The degree of disorder depends on the trap density and is measured by the stretching component β (0 < β < 1). The higher the β value the less disorder (lower trap density) the semiconductor 37,38 . The details about the stretched-exponential decay fitting are given in Supplementary Note 1. The fitted results in Supplementary Table 1 show that both the average lifetime < τ> and the stretching component β increase significantly when the Ta 3 N 5 film is modified with In:GaN and Mg:GaN layers. This indicates the trap density of the Ta 3 N 5 layer is lowered by the interface and surface modification with the In:GaN and Mg:GaN layers. To further clarify the change of trap density in these samples, defectrelated PL spectra were measured under excitation of a 510-nm laser at 10 K (Fig. 3i). All the PL spectra could be deconvolved into two distinctive emission peaks centered at around 720 and 820 nm (Supplementary Fig. 10), which are ascribed to defect emissions of nitrogen vacancies and reduced Ta 3+ defects, respectively, according to our previous study 29 . The intensity of the defect-related PL emission significantly decreases with In:GaN and Mg:GaN modification, which indicates the trap density in the Ta 3 N 5 layer is reduced. In contrast to the abrupt lattice termination on the surface of bare Ta 3 N 5 layer, the In:GaN and Mg:GaN modification resulted in lattice-matched GaN/Ta 3 N 5 interfaces, thus reducing the defect density near the interfaces 15 .
PEC performance of In:GaN/Ta 3 N 5 /Mg:GaN photoanode. The PEC performance of Ta 3 N 5 films with or without interface engineering with In:GaN and Mg:GaN layers was tested after modifying their surfaces with a borate-intercalated nickel cobalt iron oxyhydroxide (NiCoFe-B i ) OER co-catalyst. The co-catalyst modification is essential to improve the activity and stability of the photoanodes (Supplementary Fig. 11). Figure 4a shows the photocurrent-potential (J-V) curves for Ta  The photocurrent onset potential, measured at a steady-state photocurrent density of~20 μA cm −2 (Fig. 4b), was also found to shift cathodically from 0.47 V vs. RHE for the Ta 3 N 5 photoanode to 0.38 V vs. RHE for the In:GaN/ Ta 3 N 5 /Mg:GaN photoanode. Consequently, a maximum ABPE of 3.46% was achieved for the In:GaN/Ta 3 N 5 /Mg:GaN photoanode (Fig. 4c), which is the highest ABPE reported so far for Ta 3 N 5based photoanode, to the best of our knowledge ( Supplementary  Fig. 12). The In:GaN/Ta 3 N 5 /Mg:GaN photoanodes samples also showed good reproducibility in PEC performance: the average ABPE for a batch of eight samples is 3.31% with a standard deviation of 0.11% (Supplementary Fig. 13).
The stability of NiCoFe-B i modified Ta 3 N 5 and In:GaN/Ta 3 N 5 / Mg:GaN photoanodes was tested at 1.0 V vs. RHE in 1 M KOH under simulated sunlight. The photocurrent of the Ta 3 N 5 photoanode decreased continuously due to the self-oxidation of the Ta 3 N 5 surface (Supplementary Fig. 14). In contrast, the In:GaN/Ta 3 N 5 /Mg:GaN photoanode generated a stable photocurrent at~8.9 mA cm −2 for 160 min (Fig. 4d). The improved stability is ascribed to the passivation effect of the Mg:GaN layer, which prevents the self-oxidation of the Ta 3 N 5 surface, a key factor of performance degradation in Ta 3 N 5 photoanode 19 . For longer-term stability test, the In:GaN/Ta 3 N 5 /Mg:GaN photoanode maintained 80% of its initial photocurrent after 10 h and 70% after 15 h (Supplementary Fig. 15). Figure 4e plots the wavelength dependence of the incident photon-to-current conversion efficiency (IPCE) of the In:GaN/Ta 3 N 5 /Mg:GaN photoanode. The IPCE values were between 70% and 90% in a broad spectrum range of 400-550 nm, which indicates the impressive photon conversion efficiency of In:GaN/Ta 3 N 5 /Mg:GaN photoanode in the visible range. An integrated photocurrent density of 9.0 mA cm −2 was obtained by integrating the IPCE values over the standard AM 1.5 G solar spectrum, which matches well with the measured photocurrent density in Fig. 4a (~8.9 mA cm −2 ). Figure 4f shows the amount of O 2 evolved from the In:GaN/ Ta 3 N 5 /Mg:GaN photoanode quantified by the gas chromatography (GC). The dark line plots a quarter of the electron numbers (e − /4) calculated from the measured photocurrent (Supplementary Fig. 16), which gives the expected amount of O 2 evolution assuming 100% Faraday efficiency. The GC data points align well with the calculated e − /4 line, indicating nearly unity Faraday efficiency was achieved with the In:GaN/Ta 3 N 5 /Mg:GaN photoanode. This result demonstrates that the measured photocurrent is indeed contributed by OER and the ABPE calculated from J-V curve does represent the STH conversion efficiency of the In:GaN/Ta 3 N 5 /Mg:GaN photoanode.
To further clarify the roles of In and Mg doping in the GaN layers in the heterojunction structure, pristine GaN layers were employed for interface engineering for comparison. The GaN/ Ta 3 N 5 photoanode showed a degraded PEC performance compared with the Ta 3 N 5 photoanode ( Supplementary Fig. 17a), which is likely due to the mismatched band alignment that formed a large potential barrier on the back of the photoanode. In contrast, the In:GaN/Ta 3 N 5 photoanode showed obviously improved PEC performance compared with the Ta 3 N 5 photoanode, which is ascribed to the generation of a conductive midgap band in GaN by In doping, as revealed in Fig. 2f. The Ta 3 N 5 / GaN photoanode showed a slightly enhanced PEC performance compared with Ta 3 N 5 photoanode ( Supplementary Fig. 17b), consistent with the previous report 15 . Nevertheless, the improvement was not as great as that achieved with Ta 3 N 5 /Mg:GaN photoanode. These results demonstrate that proper doping of the interfacial GaN layers is crucial to improve the carrier separation efficiency and hole injection efficiency of Ta 3 N 5 photoanode.
Electrochemical characterizations of the heterostructure. Surface injection efficiency (η inj ) and bulk charge separation efficiency (η bulk ) were determined to further investigate the effect of In:GaN and Mg:GaN layers. The details about the calculation of η inj and η bulk are given in Supplementary Note 2. η bulk represents the fraction of photo-generated holes that arrive at the photoanode/electrolyte without recombination in the bulk, while η inj denotes the fraction of those holes that successfully inject into the electrolyte for water oxidation. The J-V curves of four different photoanodes were measured in 1 M KOH electrolyte with or without 0.5 M H 2 O 2 as a hole scavenger (Supplementary Fig. 18). The absorption photocurrent density (J abs ) of each photoanode was calculated based on the UV-visible absorption spectrum by assuming 100% absorbed photon-to-current conversion efficiency ( Supplementary Fig. 19). The potential dependent η bulk curves in Fig. 5a reveal that the improvement of η bulk is mainly attributed to the modification of In:GaN at the bottom interface of Ta 3 N 5 film. This is reasonable as In:GaN selectively extracts photogenerated electrons from the Ta 3 N 5 film, thus improving the bulk electron-hole separation efficiency. Meanwhile, the potential dependent η inj curves in Fig. 5b reveal that the improvement of η inj is mainly originated from the modification of Mg:GaN on the surface of Ta 3 N 5 film. These results have helped us distinguish the respective role of In:GaN and Mg:GaN in enhancing the PEC performance of the heterostructure photoanode.
Photoelectrochemical impedance spectroscopy (PEIS) was employed to compare the charge transfer processes between Ta 3 N 5 and In:GaN/Ta 3 N 5 /Mg:GaN photoanodes 14,22 . The PEIS Nyquist plots in Fig. 5c show that the radius of the semicircle curve for In:GaN/Ta 3 N 5 /Mg:GaN photoanode dramatically reduces compared with that of Ta 3 N 5 photoanode, which indicates the improvement of charge separation efficiency and injection efficiency. A typical two-RC-unit equivalent circuit model was used to fit the Nyquist plot of Ta 3 N 5 photoanode, while a more complex model with the addition of a heterojunction RC-unit was used to fit the Nyquist plot of In:GaN/Ta 3 N 5 / Mg:GaN photoanode (Supplementary Fig. 20) 22 . The PEIS Nyquist plots were well-fitted using the equivalent circuit models  Supplementary  Table 3. The trapping resistance (R trap ) and the charge transfer resistance (R ct ) of the In:GaN/Ta 3 N 5 /Mg:GaN photoanode are 6.0 and 23.0 Ω, respectively. In contrast, the Ta 3 N 5 photoanode shows a significantly larger R trap and R ct values of 449.3 and 503.7 Ω, respectively. Although additional resistance (R hete ) is generated in the In:GaN/Ta 3 N 5 /Mg:GaN photoanode due to the heterojunctions, it is relatively small (1.13 Ω). These results demonstrate that the separation and injection efficiencies are indeed improved by interface engineering with the In:GaN and Mg:GaN layers.
Mott-Schottky (M-S) analysis was employed to probe the change of surface energetics of the photoanodes with and without interface engineering. The M-S plot of Ta 3 N 5 photoanode in Fig. 5d shows that the flat-band potential is at 0.15 V vs RHE. However, the slope of the M-S plot changes at higher applied potential region (>0.77 V vs RHE), suggesting that the band bending develops to a different extent in different potential regions (Supplementary Fig. 21a). At low potential region (<0.77 V vs RHE), the Fermi level pinning due to the existence a distribution of surface states causes the band bending to develop in a lesser extent while part of the applied potential drops at the Helmholtz layer 39 . At high potential region (>0.77 V vs RHE), the decreased density of surface states allows for the band bending to develop in a higher extent, leading to a larger slope in the M-S plot. For the In:GaN/Ta 3 N 5 /Mg:GaN photoanode in Fig. 5e, the flat-band potential shifts cathodically to 0.05 V vs RHE while the transition of the two slopes happens at 0.37 V vs RHE. The lower applied potential to overcome the Fermi level pinning suggests there is a lower density of surface states due to the passivation effect of the Mg:GaN layer. The band bending can be developed in a larger extend at lower applied potential region, thus reducing the bias-potential requirement for the water oxidation process (Supplementary Fig. 21b). The open circuit potential (OCP) decay profile was employed to further demonstrate the passivation of interfacial traps by interface engineering [40][41][42] . Supplementary Fig. 22 shows the OCP decay profiles of Ta 3 N 5 and In:GaN/Ta 3 N 5 /Mg:GaN photoanodes. Compared with Ta 3 N 5 photoanode, the amplitude of the OCP decay is significantly larger in In:GaN/Ta 3 N 5 /Mg:GaN photoanode, which indicates larger band bending at photoanode/electrolyte interface under illumination 41 . The carrier lifetimes are derived from the OCP decay curves (Supplementary Note 3) and plotted against OCP in Fig. 5f. At the transient when illumination was switched off, the carrier lifetime in In:GaN/Ta 3 N 5 /Mg:GaN photoanode was reduced by a factor of two compared to that in Ta 3 N 5 photoanode. The faster decay kinetics upon switching off the illumination suggests the reduced charge trapping at the interfaces in In:GaN/Ta 3 N 5 /Mg:GaN photoanode. The larger band bending and reduced charge trapping in the depletion region near photoanode/electrolyte interface accelerate charge separation, leading to an improved PEC activity of the In:GaN/ Ta 3 N 5 /Mg:GaN photoanode.
In summary, an "n-i-p" heterostructure In:GaN/Ta 3 N 5 / Mg:GaN photoanode was prepared by one-step thermal nitridation process of InO x -GaO x /TaO x /Mg:GaO x precursor film deposited by EB evaporation and ALD. Mechanistic studies revealed that the n-type In:GaN layer mainly improved the bulk charge separation efficiency by selectively extracting photogenerated electrons from Ta 3 N 5 through a conductive mid-gap band induced by In-doping. Meanwhile, the p-type Mg:GaN layer mainly contributed to the enhanced surface charge injection efficiency of the photoanode by passivating the traps on the surface of Ta 3 N 5 . Benefiting from the improved bulk charge separation efficiency and surface injection efficiency, the In:GaN/ Ta 3 N 5 /Mg:GaN photoanode yielded a maximum ABPE of 3.46%, which was significantly improved from 2.29% of the Ta 3 N 5 photoanode. Owning to the passivation effect of Mg:GaN layer, the stability of the In:GaN/Ta 3 N 5 /Mg:GaN photoanode was also drastically improved. The application of In:GaN and Mg:GaN as interfacial modification layers could be extended to other oxynitride semiconductors (e.g., TaON, BaTaO 2 N, LaTiO 2 N) to further improve their conversion efficiency for PEC water splitting. Finally, it is proposed that effective carrier management through interface engineering should be generally considered to improve the efficiency and stability of photoelectrode for artificial photosynthesis. The deposition process was carried out at a constant current density of 30 μA cm −2 for 10 min under AM 1.5 G illumination (SAN-EI Electric, XES-40S3-TT). After deposition, the samples were rinsed with deionized water.

Methods
PEC measurement. All the PEC measurements were conducted using a potentiostat (Bio-Logic SP-200) in three-electrode configuration with a Pt counter electrode and a Hg/HgO reference electrode. All measured potentials versus Hg/HgO were converted to RHE scale according to the Nernst equation. The Pt cathode chamber was separated from the photoanode chamber using a Nafion 117 membrane. Simulated sunlight from a solar simulator (SAN-EI ELECTRIC, XES-40S3-TT) with intensity calibrated to 100 mW cm −2 (AM1.5 G) using a certified reference cell (Konica-Minolta AK-200) was used as the light source. J-V curves were recorded under a cathodic scan at a rate of 10 mV s −1 at 283 K in 1 M KOH (pH 13.6) solution under magnetic stirring and Ar bubbling. ABPE was calculated from the J-V curves with the equation: ABPE = [(1.23 -V app ) × J light /P light ] × 100%, where V app is the applied potential (versus RHE), J light is the photocurrent density under AM 1.5 G light, P light equals to 100 mW cm −2 for the simulated sunlight. IPCE spectrum was measured at 1 V vs RHE from 350 to 650 nm with an interval of 10 nm using a monochromatic light source (Zolix Sirius 300 P). The IPCE value at each wavelength (λ) was calculated by: IPCE = [(1240/λ) × (J light − J dark )/P light ] × 100%, where J light was the measured photocurrent density, J dark was the current density in the dark, P light was the irradiance of the monochromatic light. The amount of oxygen evolved from the photoanode was quantified by a gas chromatography (Shimadzu GC-2014). PEIS was performed at 1.0 V vs RHE under AM 1.5 G simulated sunlight in frequency range of 1 Hz to 1 MHz. Fitting of the Nyquist plots was performed by ZSimDemo software. M-S plots were measured in the potential range of −0.6-1.3 V vs RHE with an AC amplitude of 10 mV in the frequency range of 1.0-3.0 kHz under dark conditions. The OCP decays at the light on/off transient were recorded after illuminating the sample for 10 min in 1 M KOH.