Multilayer oxide protection layer with multiple tunnelling paths for efficient and durable Si-based photocathode

Abstract

Constructing a low-resistance oxide protection layer is challenging but highly beneficial for realizing a practical photoelectrochemical device. The thickness of oxide layer strongly influences its behaviors of carrier transport and corrosion resistance, generally leading to a trade-off between efficiency and durability. Different from the previous methods, here we propose and demonstrate a universal approach to decouple the trade-off of oxide layer by multiple carrier-tunnelling paths. This approach with oxide/metal architecture ((O/M)_n, n is the number of nano-scale repeating unit) enables low-resistance carrier transport as required for high efficiency, while allowing the layer to be sufficiently thick, which reinforces durability. This approach can be applied to various oxide-based layers, such as (TiO₂/Fe)_n, (CeO₂/Fe)_n and (TiO₂/Pd)_n. In addition, a good correlation between carrier dynamics and oxide/metal architecture is established by employing systematic photoelectrochemical-electrical measurements and simulation models. Here we show important contributions for further developing the practical photoelectrodes in photoelectrochemical devices and controlling the carrier transport behaviors in complex multilayer structure.

Introduction

The photoelectrochemical (PEC) approach that directly converts simple small molecules (e.g., H₂O, NO₃⁻, CO₂) to valuable fuel molecules (e.g., H₂, NH₃, CH₃OH) using solar energy is an intriguing and sustainable technology^1,2,3. Tremendous efforts have been directed towards constructing efficient and stable photoelectrodes, which are the core component of the practical PEC system^4,5,6. In comparison to wide-gap semiconductors, narrow-gap semiconductors (e.g., Si) with band gaps less than 2.3 eV are more attractive as light absorbers of the photoelectrodes due to large optical absorption coefficients, broad-spectrum light collection, high theoretical photocurrent density, and good charge carrier mobilities^7,8,9. To maximize the solar-to-chemicals conversion, the semiconductor light absorbers with porous structure can simultaneously minimize the light reflection by trapping light in the pore channels and increase the surface active area for PEC reactions^10,11. Unfortunately, according to Pourbaix diagrams¹², these narrow-gap semiconductors with a very short window of stability are easy to be corroded/passivated quickly when immersed in aqueous electrolytes, especially in strong acidic and alkaline conditions. One way of combating this challenge is to introduce a protection layer into the photoelectrode as a means of separating the light absorber from the electrolytes.

Ideally, the protection layer should concurrently protect the underlying light absorber, facilitate the carrier (electron or hole) transport, and allow the high light transmittance¹³. Early attempts to extend photoelectrode lifetime have focused on depositing a thin noble metal layer (such as Pt) onto light absorber¹⁴. However, the parasitic light absorption/reflection and Fermi-level pinning effect of metal layer reduce the light harvesting and band bending of the photoelectrodes, leading to unsuccessful PEC performance at low photocurrents and photovoltages^15,16. Alternatively, transparent oxides, such as TiO₂¹⁷, SiO₂¹⁸, and NiO_x¹⁹, have been widely used as protection layers for various photoelectrodes due to high intrinsic chemical stability and minimal parasitic light absorption. Nevertheless, there is still a visible trade-off between photoelectrode efficiency and durability while oxides are employed as protection layers. High-quality thin oxide layers (less than 4 nm) generally show a high passage of current density (over 1 A cm⁻²) by carrier tunnelling^20,21, but is not stable in this situation. In the anti-corrosion field, over 30 nm-thickness oxide film can provide a typical protection layer for achieving long-term and durable operation^22,23. On the other hand, increasing the thickness of oxide layer leads to an exponential decay of carrier tunnelling probability, thus a nanoscale oxide layer is usually required. Two roles of carrier transport and corrosion resistance for the oxide protection layer need to be decoupled. Some pioneering work including our own study has demonstrated the defect formation and migration that can increase conductivity of thick TiO₂ layer (~20 nm) for accelerating carrier transport instead of tunnelling^24,25,26,27. However, this method is still difficult to completely resolve the trade-off of oxide layer. Here, in addition to the conductive channels, the defect sites can also serve as pinholes to induce the occurrence of pitting corrosion²⁸. Besides, this method is likely to be hard to generalize to other oxide layers or transparent compound layers (e.g., SiC).

Herein, to resolve the abovementioned issues, we demonstrate a general method for realizing a low-resistance oxide-based protection layer via combining carrier-tunnelling behaviors and multilayer complementarity. In this method, the architecture of the protection layer is oxide/metal multilayer structure ((O/M)_n, n is the number of nano-scale repeating unit), and its total thickness is constant (~36 nm). According to electric field simulations and experimental results, tuning the n value as well as the thickness of repeating unit can form the multiple carrier-tunnelling paths in the protection layer and optimize the interfacial resistance between oxide and metal for enhancing carrier transport. Using a classic silicon (Si) material as light absorber, a Si-based photocathode with optimal (O/M)₆ protection layer shows a faster transfer rate of charge carriers, lower loss of charge carriers and higher solar-to-ammonia conversion efficiency than that with other protection layers (n = 1, 2, 4, and 8). In addition, since the metal units are nanoscaled, it can also impede the parasitic light absorption/reflection for improving the light absorption of Si – thereby further increasing efficiency²⁹.

Results and discussion

Design, synthesis and characterization of oxide-based protection layer

To rationally design the multilayer structure, a series of electric field simulations for the intrinsic conductivity of various (O/M)_n protection layers with same total thickness (36 nm including 18-nm oxide units and 18-nm metal units) and different n values were implemented by finite element analysis (COMSOL Multiphysics). In these simulations, TiO₂ as one of the most common materials of oxide protection layer was selected as the oxide model of (O/M)_n, and cheap and abundant iron (Fe) served as the metal model to avoid the use of noble metal and increase the general applicability of our method. Unless otherwise stated, the oxide-based protection layer with (TiO₂/Fe)_n multilayer structure is simply labeled as (TF)_n, n = 1, 2, 4, 6, and 8 corresponding to unit thickness of 18/18, 9/9, 4.5/4.5, 3/3, and 2.25/2.25 nm, respectively. The parameters of (TF)_n materials and electric field are provided in Tables S1 and S2 of Supporting Information (SI). In this section, Fe is a conductor, and its resistance can be ignored. As a semiconductor, the conductive behavior of the TiO₂ bulk is simulated by a drift-diffusion equation. Figures 1a and S1 illustrate the electric potential distribution of the bulk for (TF)_n by finite element method in a constant-current situation. The simulation analysis indicated that the potential gradient of Fe layer was invariable in (TF)_n due to the quick redistribution of free electrons for eliminating the internal electric field, reflecting that Fe layer possesses a zero resistance. The bulk of the TiO₂ layer with an even-distributed resistance leads to a linear loss in potential. It can be seen that the increase of n in (TF)_n can effectively reduce the loss in potential (see the bottom of Figs. 1a and S1). As a result, the order of total TiO₂ bulk resistance is (TF)₈ < (TF)₆ < (TF)₄ < (TF)₂ < (TF)₁. In addition, the interface resistance of (TF)_n (n > 2) was simply used to sharply increase the loss of potential in this constant-current model, but the ohmic contact and Schottky contact at the interface of semiconductor/conductor were neglected.

Fig. 1: COMSOL multiphysics simulation results on the electric properties of (TF)_n.

a Electric potential distribution maps (top) and potential-layer depth curves (bottom) of the bulk for (TF)₁ (left), (TF)₆ (middle) and (TF)₈ (right) by finite element method in a constant-current situation. b The current density profiles of (TF)_n along the layer depth to simulate the electron conduction at the Schottky interfaces by WKB tunnelling model. c Simulation of mean current density vs. (TF)_n in the range of applied potential from 0 to 2.0 V by considering the bulk and interface resistance comprehensively.

Considering that the work function of Fe is higher than the electron affinity of TiO₂, electrons moving from Fe to TiO₂ and from TiO₂ to Fe usually occur in the Schottky contact and ohmic contact, respectively. The ohmic contact can facilitate the transfer of the electrons at the interface, which is regarded as zero resistance, as shown in the results of (TF)₁ (see Fig. 1a, b). On basis of the formation of depletion area and atomic-level properties of interface, a Wenzel–Kramers–Brillouin (WKB) tunnelling model³⁰ combined with approximation formula (see Eqs. 1 and 2) was employed to analyze the electrons passing through the Schottky interfaces.

$$\psi (x)\cong \frac{2D}{\sqrt{p(x)}}\sin \left[\frac{1}{h}\int _{x}^{{x}_{0}}p(x^{\prime} )dx^{\prime}+\frac{\pi }{4}\right]\,(x < {x}_{0})$$

(1)

$$\psi (x)\cong \frac{D}{\sqrt{p(x)}}\exp \left[-\frac{1}{h}\int _{{x}_{0}}^{x}\left|p(x^{\prime} )\right|dx^{\prime} \right]\,(x > {x}_{0})$$

(2)

where ψ(x), p(x), D, h, x and x₀ denote the wave function describing the state of a quantum mechanical microscopic system, kinetic energy function, normalization constant, reduced Planck constant, path traversed by the microscopic particle during tunnelling, and turning point of Classical and quantum physics behaviors (namely, the position of the particle’s total energy), respectively. In the tunnelling model, there is a strong electric field at the Schottky interface, where the electrons were accelerated and increased the possibility of tunnelling leading to a jump in current density at the Schottky interface (as shown in Fig. 1b). Due to the exponential dependence of tunnelling probability (Eq. 3)³⁰, the increase of n for (TF)_n increased the potential barriers, showing lower tunnelling current densities, viz.

$$T(E)\approx \exp \left(-\frac{2}{h}\int _{{x}_{1}}^{{x}_{2}}\sqrt{2{m}^{\ast }(V(x)-E)}dx\right)$$

(3)

where T(E), V(x), E and m^* are the tunnelling probability, potential energy function, total energy of the microscopic particle (kinetic and potential energy), and effective mass of the particle, respectively.

Accordingly, the order of tunnelling current density is (TF)₂ > (TF)₄ > (TF)₆ > (TF)₈. Integrating the bulk resistance and interface tunnelling effect, Fig. 1c shows the simulation of mean current density versus (vs.) (TF)_n in the range of applied potential from 0 to 2.0 V. Interestingly, (TF)₆ with unit thickness of 3 nm exhibited the highest mean current density, implying the lowest total resistance in (TF)₆. This optimized result is jointly determined by the decrease of bulk resistance and the increase of interface resistance with nanoscale multilayer structure.

In light of these calculations, we sought to fabricate the abovementioned (TF)_n as protection layer of Si-based photocathode for PEC NO₃⁻ reduction reaction (NO₃RR) via magnetron sputtering technology³¹. As shown in Fig. 2a, the assembly of Si-based photocathodes contained four main steps: the formation of Si with buried p-n junctions (n⁺p-Si) as light absorber¹; the deposition of carbon layer as electron transfer layer³²; the fabrication of (TF)_n with different n values as protection layer; and FeCu alloy layer as cocatalysts for NO₃RR³³. Fe- and Cu-based catalysts are known as promising candidates for NO₃RR, which can enhance NO₃⁻ adsorption, accelerate electron transfer, and suppress the HER³⁴. The FeCu cocatalysts/(TF)_n protection layer/carbon layer/n⁺p-Si photocathodes can be clearly denoted as C(TF)_nCSi (n = 1, 2, 4, 6, and 8). Preparation details of these photocathodes are described in the Experimental Section of SI.

Fig. 2: Synthesis and structural characterization of C(TF)_nCSi.

a Schematic of the chemical etching and physical vapor deposition to fabricate the C(TF)_nCSi. b Cross-sectional HRTEM images of C(TF)₆CSi. The inset is the magnified image of the specified area by HRTEM. c STEM image with corresponding elemental distribution maps for C(TF)₆CSi.

Field emission scanning electron microscopy (FESEM) was used to perform a reconnaissance study of the architecture of C(TF)_nCSi. As shown in Fig. S2, C(TF)_nCSi with different n values showing a similar surface morphology, implying similar surface properties (e.g., a good lyophilic surface with contact angles <35°, see Fig. S3) of FeCu cocatalysts for these photocathodes. Meanwhile, the cross-sectional FESEM images (Fig. S4) revealed that C(TF)_nCSi was composed of cocatalyst layer, (TF)_n protection layer with multilayer structure (n ≤ 4), and Si-based light absorber. However, when the n value was over 4, the multilayer structure of (TF)_n protection layer was hardly observed by FESEM due to the limitation of the resolution ratio. High-resolution transmission electron microscopy (HRTEM) was further applied to precisely determine the architecture and microstructure of C(TF)_nCSi. In comparison to C(TF)₁CSi with thick (TF) unit (18/18 nm, Fig. S5), C(TF)₆CSi (Fig. S6) and C(TF)₈CSi (Fig. S7) showed a composite protection layer with 6 and 8 thin (TF) units in the situation of same total protection layer thickness (36 nm), respectively, reflecting the successful fabrication of C(TF)_nCSi with nanoscale multilayer (TF)_n. In addition, the lattice spacing of ~0.205 nm was partially found in FeCu cocatalysts (Figs. 2b, S5 and S7) corresponding to (111) plane of Cu, in line with the presence of broad and low Cu diffraction peaks in X-ray diffraction patterns (XRD, Fig. S8). The (TF)_n protection layers were amorphous due to the absence of TiO₂ and Fe diffraction peaks. The cross sections of C(TF)₆CSi (Fig. 2c) and C(TF)₈CSi (Fig. S9) were also detected using scanning transmission electron microscopy (STEM) coupled with energy dispersive spectroscopy (EDS) mapping. Elemental mapping of C(TF)₆CSi (see Fig. 2c) definitely indicated that 6 (TF) units with ~3-nm Fe and ~3-nm TiO₂ were successively stacked as an oxide-based protection layer, and ~28-nm FeCu cocatalysts were on the protection layer. According to the data analysis of X-ray photoelectron spectroscopy (XPS, Fig. S10) and near-edge X-ray absorption fine structure spectroscopy (NEXAFS, Fig. S11), the surface of C(TF)_nCSi with different n values showed similar Fe/Cu ratio (~2.2), Fe³⁺/Fe²⁺ ratio (~2.6) and Cu²⁺/Cu¹⁺ ratio (~0.9), as exhibited in Fig. S12, identifying that the composition of FeCu cocatalysts is nearly same on these photocathodes. Meanwhile, XPS depth profiling measurement further revealed that the Fe unit layer of (TF)₆ protection layer was metallic with little oxidation (see Fig. S13), implying the successful deposition of nanoscale metal unit layer in the (O/M)_n architecture. Furthermore, via optical measurements and indirect allowed transition (Fig. S14), a band gap of ~1.1 eV assigning to Si and an almost identical visible light transmittance for different (TF)_n protection layers and C(TF)_n composite layers was found in all the C(TF)_nCSi. In summary, apart from the difference of the architecture of protection layer, C(TF)_nCSi approximately possessed the same Si-based light absorber, FeCu cocatalysts and optical properties, suggesting a similar light-harvesting ability and surface catalytic activity³⁵. Consequently, it can be easily imagined that the PEC NO₃RR performance of C(TF)_nCSi is controlled by changing the n values of (TF)_n protection layer due to the formation of multiple carrier-tunnelling paths.

Concurrently enhanced PEC performance on efficiency and durability

Next, we sought to verify whether the n value of (O/M)_n protection layer correlated with PEC performance, including solar-to-chemical conversion efficiency and carrier migration rate, as depicted in Fig. S15. In the conditions of the same Si-based light absorber and FeCu cocatalysts, PEC NO₃RR behaviors of the Si-based photocathodes in 1.0 M KOH-0.05 M KNO₃ aqueous solution (pH = 13.8) under 1 sun illumination were utilized to estimate the positive role of multilayer structure for (O/M)_n protection layer. Figure 3a displays the current density (J)-potential (V) curves of all the C(TF)_nCSi photocathodes for PEC NO₃RR. The dark currents of these photocathodes may be considered negligible. The highest performance was obtained on C(TF)₆CSi, realizing the highest saturation current density (J_sa, −19.7 mA/cm²) and lowest applied potentials (V_ap, 0.055 V vs. RHE corresponding to 10 mA/cm²). This is significantly better than −14.7 mA/cm² and −0.10 V measured on C(TF)₁CSi under identical testing conditions. With the increase of n value in C(TF)_nCSi, PEC performances of the photocathodes were gradually optimized, however, as n > 6, C(TF)₈CSi showed a weak PEC performance with a J_sa of −18.9 mA/cm² and 0.005 V vs. RHE corresponding to 10 mA/cm². The results perfectly accord with the simulated models (Fig. 1), indicating that the multilayer architecture with multiple carrier-tunnelling paths can form a low-resistance (TF)_n protection layer to accelerate the carrier transfer and reduce the carrier loss. Some similar variations in chopped current curves (Fig. S16) and light-assisted electrochemical surface areas (ECSAs, Fig. S17) are also observed on the C(TF)_nCSi photocathodes.

Fig. 3: PEC NO₃RR performance of Si-based photocathodes with various (O/M)_n protection layers.

a Current density (J)-potential (V) curves of all the C(TF)_nCSi photocathodes in 1.0 M KOH-0.05 M KNO₃ aqueous solution. The current density measured in the dark was a near horizontal line, namely 0 mA/cm². b Yield rate of NH₃ on the C(TF)_nCSi photocathodes at each given potential for 1-h PEC NO₃RR. c PEC EIS measurements of the C(TF)_nCSi photocathodes at 0 V vs. RHE. The inset is the equivalent electrical circuit image. d The IMPS plots of C(TF)₁CSi, C(TF)₆CSi and C(TF)₈CSi at 0 V vs. RHE. e Current density values of Si-based photocathodes with (CF)₁, (CF)₆, (CP)₁, (CP)₆, (TP)₁ and (TP)₆ protection layers at 0 V and −0.6 V vs. RHE. f Dependence of the lifetime of (O/M)_n protection layer on the thickness of oxide unit by combining the J-t measurements and corrosion rate calculations. The experiments were carried out under room temperature using a xenon lamp with the wavelength range of 320–780 nm and intensity of 100 mW/cm². FeCu alloy was utilized as the cocatalyst. Error bars represent standard deviations from three independent PEC measurements at least. The voltage is not iR corrected.

According to the chronoamperometry measurements (Fig. S18), the NH₄⁺/NO₂⁻ calibration curves (Fig. S19) and gas chromatography, the NH₃, NO₂⁻ and H₂ yield rate of C(TF)_nCSi photocathodes at each given potential in a 1.0-h period were shown in Figs. 3b and S20, respectively. For all the photocathodes, the NH₃ yield rate increased with the increase of V_ap from 0.1 V to −0.2 V vs. RHE. In comparison to other photocathodes, C(TF)₆CSi photocathode achieved a higher NH₃ yield rate of 0.42 mg/cm²/h with a maximum Faradaic efficiency of 77.3% (see Fig. S21) at 0 V vs. RHE (that is, no thermodynamic bias for hydrogen evolution reaction (HER)). This reflects that the introduction of low-resistance (TF)_n protection layer improves the solar-to-NH₃ conversion efficiency. Meanwhile, it can be easily observed that there were lower yield rates of NO₂⁻ and H₂ on the C(TF)₆CSi photocathode in the PEC NO₃RR process at 0 V vs. RHE than these on the other photocathodes (Fig. S20), in line with the higher NH₃ yield rate. Finally, the results of ¹H NMR spectra (see Fig. S22) were also consolidated for the high NH₃ yield rate on the C(TF)₆CSi photocathode via PEC NO₃RR.

Further, light-assisted electrochemical impedance spectroscopy (EIS) analysis was carried out to assess the PEC kinetics of C(TF)_nCSi photocathodes for NO₃RR from 100 kHz to 0.1 Hz at 0 V vs. RHE (Fig. 3c). The resistance of carrier transfer in the photocathodes exhibited the following order: C(TF)₆CSi < C(TF)₈CSi < C(TF)₄CSi < C(TF)₂CSi < C(TF)₁CSi, consistent with the PEC NO₃RR performance. The Bode plots of C(TF)_nCSi photocathodes also showed the similar results about their interface resistance (Fig. S23). An equivalent electrical circuit was further calculated to describe the PEC NO₃RR kinetic processes of C(TF)_nCSi photocathodes (the inset of Fig. 3c). Generally, the kinetic processes of a PEC reaction on the hierarchical photoelectrode are usually assessed by charge transfer resistance between solid/solid interface (R₁, like TiO₂ layer/Fe layer or TiO₂ layer/FeCu cocatalysts) and solid/liquid interface (R₂, like FeCu cocatalysts/electrolytes). The smaller the R₁ and R₂, the faster is the kinetics. The values of the R₁, R₂ and solution resistance (R_s) were obtained by fitting EIS data using the equivalent electrical circuit, as listed in Table S3. C(TF)₆CSi photocathode showed low R₁ and R₂, especially the lowest R₁, corresponding to the super PEC NO₃RR activity. Meanwhile, the data of intensity-modulated photocurrent spectra (IMPS) at different V_ap (see Figs. 3d and S24) demonstrated the carrier transfer time in the photocathodes via the calculations described in the SI. At 0 V vs. RHE, the carrier transfer times (τ_d) of C(TF)₁CSi, C(TF)₈CSi and C(TF)₆CSi were 30.3, 19.3, and 15.4 μs, respectively (Figs. 3d and S24c). Actually, compared to other photocathodes, C(TF)₆CSi measured at all the V_ap values presented a shorter carrier transfer time, meaning that photogenerated electrons reach the reaction surface more rapidly leading to a lower bulk-phase carrier recombination. These phenomena are in line with the results of steady-state photoluminescence spectra (Fig. S25) and open-circuit potential measurements (Fig. S26).

To further confirm the universality and feasibility of (O/M)_n architecture, CeO₂ and Pd were also selected as the oxide and metal model to build the (CeO₂/Fe)_n protection layer (simplified as (CF)_n), (TiO₂/Pd)_n protection layer (simplified as (TP)_n), and (CeO₂/Pd)_n protection layer (simplified as (CP)_n) for replacing the (TF)_n in the Si-based photocathodes, respectively. Several composition and structure characterizations were used to primarily identify the successful preparation of these photocathodes (see Figs. S27 and S28). On basis of various PEC NO₃RR measurements (Fig. S29), the Si-based photocathodes with (CF)₆, (CP)₆ and (TP)₆ protection layer possessed higher photocurrent density at 0 V and −0.6 V vs. RHE than those with (CF)₁, (CP)₁ and (TP)₁ protection layer (see Fig. 3e). Among all the photocathodes, the photocathode with (CP)₆ protection layer showed highest photocurrent density (−16.0 mA/cm²) at 0 V vs. RHE, corresponding to the highest NH₃ yield rate (0.51 mg/cm²/h in Fig. S29b), implying that the carriers are transported in (CP)₆ architecture more rapidly. In addition, the almost identical J_sa of the photocathodes with (O/M)₆ protection layer reflects about the same carrier loss (Fig. 3a, e), irrespective of materials of oxide and metal unit. Meanwhile, we also investigated the effect of (TF)_n protection layer on PEC HER behaviors to further testify the availability of low-resistance strategy. As shown in Fig. S30, the HER performance of C(TF)_nCSi with the increase of n value showed a similar variation tendency with the NO₃RR performance (Fig. 3a). In summary, (O/M)_n architecture can fabricate low-resistance oxide-based protection layer to enhance the carrier transfer and improve the PEC performance.

The durability of (O/M)_n protection layer is another important indicator for pushing forward the process of practical Si-based photoelectrodes. Various long-term J-time (t) curves of C(TF)₆CSi photocathodes were measured under light irradiation at 0 V vs. RHE in 1.0 M KOH-0.05 M KNO₃ aqueous solution (Fig. S31). In a 3-h (Fig. S31a) or 12-h (Fig. S31b) PEC period, the C(TF)₆CSi photocathode exhibited little fluctuation in photocurrent density, suggesting the relatively stable operation for NO₃RR. As shown in Fig. S32, there was a nearly leaner increase in NH₃ production on C(TF)₆CSi in a 2-h PEC process, corresponding to the relatively smooth current region in the J-t curves. As the reaction time was extended over 20 h, a drastic fluctuation in photocurrent density was observed on the C(TF)₆CSi photocathode (see Fig. S31c). This phenomenon is attributed to the adsorption and desorption of the bubbles on the photocathode surface via HER, reflecting the change of surface composition based on the reaction transformation. To further determine the composition and structure evolution in the PEC process, the post-electrolysis C(TF)₆CSi photocathodes at different reaction times were characterized by FESEM and XPS. After 3-h PEC reaction, the C(TF)₆CSi photocathode retained a good coverage of C(TF)₆C composite layer with the thickness of ~70 nm (Fig. S33a), demonstrating little loss of FeCu alloy layer in this process. Meanwhile, XPS spectra of C(TF)₆CSi photocathode after 3-h photoelectrolysis showed the presence of Cu characteristic peaks (Fig. S33b) and the lack of Ti signal (Fig. S33c), testifying that the FeCu alloy layer is conformally covered on the (TF)₆ protection layer. Similar FESEM and XPS results (Fig. S34) were also found on the reacted photocathode as extending photoelectrolysis to 12 h, in line with the little fluctuation in photocurrent density (see Fig. S31b). This phenomenon was also confirmed by STEM image with corresponding elemental distribution maps, as shown in Fig. S35. However, in comparison to the initial thickness (~74 nm), the thickness of C(TF)₆C just was ~34 nm after 80-h PEC reaction (Fig. S36a), potentially including ~24-nm (TF)₆ protection layer and ~10-nm carbon layer (Fig. S5). In the corresponding XPS spectra, the absence of Cu characteristic peak (Fig. S36b) and the presence of Ti⁴⁺ peaks (Fig. S36c) further demonstrated the disappearance of FeCu alloy layer and the exposure of (TF)₆ protection layer. These results demonstrate that the C(TF)₆CSi photocathode is progressively corroded from the cocatalyst layer to protection layer during the PEC reaction in the strong alkaline electrolytes, as observed in the reported photocathodes². Being different from anisotropic corrosion of FeCu cocatalyst layer due to poor crystallinity, the TiO₂ unit of (TF)₆ protection layer with amorphous phase can show isotropous corrosion behaviors (see Fig. S37). However, further discussions on the degradation of CuFe cocatalyst layer in the PEC process are hardly completed in this work, which obviously deviate from the core theme of this work. Combining the fluctuation of J-t curves and the loss of (TF)₆ protection layer, it can be imaged that 36-nm (TF)₆ protection layer carry out a good level of protection for the Si-based photocathode over 100 h.

(O/M)_n protection layer/Si-based photocathodes (such as (TF)₆CSi) were fabricated and used to direct estimate the lifetime of protection layer and avoid the interference of FeCu alloy layer. To quantify the loss of (TF)₆ protection layer, the final Ti content in the electrolyte (1.0 M KOH-0.05 M KNO₃) after 12-h PEC reaction was evaluated by inductively coupled plasma optical emission spectrometer (ICP-OES). The results revealed the dissolution of 23 ng of Ti from the (TF)₆CSi photocathode with a geometric area of ~0.22 cm⁻² (Fig. 3f and Table S4). Assuming that the photocathode surface is flat, with a mass loss of ~49-ng Ti per cm² per day, 2.2-nm TiO₂ unit could be lost every day via operating continuously, meaning that the lifetime of (TF)₆ protection layer is expected to be 196 h (see Fig. 3f). Indeed, the actual surface area of (TF)₆CSi photocathode with pyramidal structure is considerably larger than the geometric area (Fig. S34a). Therefore, the actual loss of (TF)₆ is much lower than the calculated value, namely, the actual lifetime is much longer than the expected value. In addition, according to the J-t measurements and corrosion rate calculations, the estimated lifetimes of Si-based photocathodes with different (O/M)_n protection layers are over 100 h, as shown in Fig. 3f. Meanwhile, these Si-based photocathodes after 24-h PEC reaction still showed similar expected lifetimes to determine the feasibility of this calculation method. Durability data of Si-based photocathodes with various protection layers in various electrolytes were listed in Table S5, mainly including reported works in recent 5 years and this work. In view of these collected data, the durability measurements for most of Si-based photocathodes were implemented in the acid electrolytes, especially pH = 0, because most of pthe rotection layers relatively stabilize in such conditions. As mentioned in our previous work²⁶, it is critical to develop the efficient and durable Si-based photocathodes for PEC reactions under strong alkaline conditions. In comparison of the durability of the protection layers in these state-of-the-art Si-based photocathodes, the durability of (O/M)_n protection layers maintain at a high level (in Table S5). In short, the low-resistance (O/M)_n protection layer with efficient carrier transfer still maintains good durability with long lifetime, realizing the synchronized enhancement of conduction and protection functions.

Based on the exploration of NH₃ energy and perniciousness of NO₃⁻, developing green PEC technology to convert excess NO₃⁻ species into NH₃ has drawn the attention of many researchers, which can facilitate the artificial nitrogen cycle and keep global nitrogen balance^36,37. Figure S38 summarizes the photocurrent density and NH₃ yield rate of various photocathodes for PEC NO₃RR at 0 V vs. RHE without heating and reducing sacrificial reagents (additional references listed in Table S6). Additionally, the electrocatalytic (EC) NO₃RR is also regarded as a green, viable approach to NH₃ production under ambient conditions^38,39. Therefore, this comparison of performance against current density and yield rate was extended to some work of EC NO₃RR at 0 V vs. RHE (Fig. S38). In view of these collected data, most of the studies on the (photo)-cathodes showed poor NO₃RR behaviors with low current density (<5 mA/cm²) and NH₃ yield rate (<0.2 mg/cm²/h) at 0 V vs. RHE, especially PEC reaction, because most of the photocathodes with protection layer have poor carrier transfer. This is despite a comparison with the good EC NO₃RR performance with iR correction^40,41, the C(TF)₆CSi and C(CP)₆CSi photocathodes in this work showed a competitive current density and NH₃ yield rate for solar-driven NO₃⁻-to-NH₃ conversion (see Fig. S38). In short, the (O/M)_n protection layer with multiple carrier-tunnelling paths and thick layer thickness supports the Si-based or other narrow-gap semiconductor-based photoelectrode to achieve a good PEC performance and a long durability simultaneously.

Deciphering the formation and role of multiple carrier-tunnelling paths in the (O/M)_n protection layer

In order to explore the intrinsic electron transport, the electrical properties of the (O/M)_n protection layers with different thicknesses were conducted by temperature-dependent, metal contact current-voltage (I-V) measurements¹³. In these experiments, a two-electrode cell in micro electrical platform⁴² (Fig. 4ai) was used to collect the electronic signals of protection layers in the dark. Figure 4b presents the electrical resistances of the unit of (Fe)_n, (TiO₂)_n and (TF)_n on FTO substrates calculated form their I-V characteristics at different n values (Fig. S39). Similar to the relationship of unit thickness and n value for (TF)_n in the Simulation section, the unit thickness of (Fe)_n and (TiO₂)_n with n = 1, 2, 4, 6, and 8 was associated with 18, 9, 4.5, 3, and 2.25 nm, respectively. Firstly, irrespective of materials, the resistance decreases with the increase of n value as well as the decrease of unit thickness, which conforms to our design and theoretical simulations. In comparison to the resistance of (TiO₂)_n and (TF)_n unit, (Fe)_n unit had lower resistance (0.52–0.07 kΩ) in the range of unit thickness from 18 to 2.25 nm, and exhibited a near-linear change with the dependence of unit thickness (see Fig. 4b), corresponding to the typical electrical properties of a metal layer. Due to the combination of TiO₂ and Fe and the presence of interface resistance, the unit of (TF)₁ showed the highest resistance (23.9 kΩ) as the thickness was 18/18 nm. A similar exponential decrease was observed in the unit of (TiO₂)_n and (TF)_n via increasing the n value from 1 to 6. In this stage, the resistance variation in the unit of (TF)_n can be mainly attributed to the decrease of TiO₂ thickness, following the electrical behavior of semiconductor materials. When the n value was 6 (namely, the thickness of TiO₂ in (TF)₆ = 3 nm), the (TF)₆ unit showed a low resistance (1.68 kΩ, see Fig. 4b) close to the resistance of metal-like materials, implying the formation of carrier-tunnelling path in TiO₂. The resistance of (TF)₈ unit reduced slightly to 1.30 kΩ with a further decrease of unit thickness to 2.25/2.25 nm. At the same time, the resistance of (TF)₈ unit was far higher than that of (TiO₂)₈ unit (0.57 kΩ), and the resistance difference of 0.38 kΩ between the unit of (TF)₆ and (TF)₈ was approximate to that between the unit of (TiO₂)₆ and (TiO₂)₈. As a result, for n ≥ 6, the effect of the interface resistance between TiO₂ and Fe on the resistance of (TF)_n unit can become predominant. Additionally, estimates the total resistance of (TF)_n by using the unit resistance multiplied by the n value showed an order of (TF)₁ > (TF)₂ > (TF)₄ > (TF)₈ > (TF)₆. The significant dependence on the thickness of the measured resistance across the (O/M)_n protection layer indicates the importance of using a vapor deposition technology such as magnetron sputtering, which displays good uniformity and control of thickness at the nanoscale.

Fig. 4: The analysis of intrinsic carrier-transport properties for (O/M)_n protection layer.

a Schematics of the two-electrode measurements in micro electrical platform at 15 °C: (i) the unit of protection layer on FTO substrate; (ii) C(TF)_nCSi photocathodes under illumination. b The electrical resistances of the unit of (Fe)_n, (TiO₂)_n and (TF)_n on FTO substrates with different n values. c Experimental I-V characteristics of C(TF)_nCSi photocathodes showing strong light-dependent current behaviors. d Schematic of photo-assisted KPFM experiments performed on the photocathodes with and without illumination under open-circuit potential conditions. e The surface potential as a function of testing time on C(TF)₁CSi and C(TF)₆CSi photocathodes by chopping the light. f The internal electric field of C(TF)₁CSi and C(TF)₆CSi photocathodes in the PEC NO₃RR process.

On the other hand, Fig. S40a shows the temperature-dependent electrical resistance curves through the unit of (Fe)₆, (TiO₂)₆ and (TF)₆ on FTO substrates from the analysis of I-V measurements (Fig. S40b–d). It is noticeable that as temperature increased, the resistance of the unit for (TiO₂)₆ and (TF)₆ decreased, as explained for semiconductor properties⁴³, different from the resistance change of the unit for (Fe)₆. The resistance of the unit for (TF)₆ was discovered to have very little dependence on a wide temperature range from 15 °C to 200 °C (see Fig. S40a), a strong indication of electron transport by tunnelling⁴⁴. Further, the temperature-dependent electrical properties of the (TF)₁ and (TF)₆ protection layers were also probed to corroborate the electron tunnelling behaviors of the (O/M)_n architecture (see Fig. S41). The current density values of (TF)₆ protection layer with 3-nm TiO₂ unit layer showed little dependence on temperature (Fig. S41c), suggesting the presence of electron tunnelling transport. From the I-V measurements (Fig. S41b), the resistance of the (TF)₆ protection layer was estimated to be ranged from ~3.2 kΩ to ~1.9 kΩ in a wide temperature range from 25 °C to 200 °C. (TF)₁ protection layer with 18-nm TiO₂ unit layer resulted in a decreasingly temperature-dependent resistance from ~131.0 kΩ to ~56.2 kΩ (Fig. S41a), implying a more thermally activated and bulk-limited conduction mechanism such as Frenkel-Poole conduction. The significant dependence on the thickness of the measured electronic conduction across the (TF)₆ protection layer indicates the importance of the (O/M)_n architecture by using a magnetron sputtering deposition method. As a result, the (TF)₆ protection layer can have the multiple carrier-tunnelling paths, leading to a lower resistance.

To exclude the effect of the space charge layer of solid/liquid interface, photo-assisted I-V measurements were conducted on C(TF)_nCSi photocathodes in micro electrical platform at 15 °C with illumination (Fig. 4a(ii)). Figure 4c exhibits the I-V characteristics of the C(TF)_nCSi photocathodes with and without illumination. As stated in the process of PEC NO₃RR, the dark current of the C(TF)_nCSi photocathodes can be ignored, especially at zero applied bias (0 V). Under illumination, all the C(TF)_nCSi photocathodes showed a fast increase in forward bias current with a weak dependence of leakage current on applied reverse bias, reflecting the presence of a rectifying interface. It was found in Fig. 4c that there was a visible photocurrent in the C(TF)_nCSi photocathodes driven by the photovoltage (V_ph) before 0 V. Among the photocathodes studied, the highest performance was obtained on C(TF)₆CSi photocathode, achieving the maximal V_ph (0.4 V), the highest current (0.33 mA) at 0 V, and the highest saturation current (~0.50 mA). These results directly verify that the design of nanoscale multilayer structure for (O/M)_n protection layer can facilitate carrier transport and effectively reduce the loss of photogenerated carriers.

Photo-assisted Kelvin probe force microscopy (KPFM) was also used to detect the surface potential (φ) of C(TF)₁CSi and C(TF)₆CSi photocathodes and understand the surface physical response by compensating the electrostatic force between photocathode surface and microscopic tip under illumination (Fig. 4d). It should be pointed out that, in these measurements, the n⁺p-Si wafers with flat surface were employed to replace those with pyramidal structure to avoid the interference from an uneven morphology. The topography images of C(TF)₁CSi and C(TF)₆CSi photocathodes with and without light irradiance (Fig. S42) showed a flat and dense surface, in line with HRTEM images (Figs. 2b and S5–S7). Under open-circuit potential conditions, the potential distribution on the surface of C(TF)₁CSi and C(TF)₆CSi photocathodes was relatively uniform in the dark, and there was a visible negative shift on their surface potentials (see Fig. S43) as the photocathodes were illuminated with 532-nm light. This result means that many photogenerated electrons from n⁺p-Si can pass through the oxide-based protection layer and inject into the FeCu cocatalysts to form the plentiful reduction sites⁴⁵.

To accurately uncover the photoresponse of surface potential, the surface potentials of C(TF)₁CSi and C(TF)₆CSi photocathodes were recorded as a function of testing time while turning on/off the light source (see Fig. 4e). Although the intensity of light source is an important affect factor, the difference of surface potential (Δφ) before and after illumination can still indirectly estimate the surface photovoltage (V_sph) and surface work function of the photocathodes, in connection with the surface band bending⁴⁶. The absolute value of Δφ (15 mV) for C(TF)₆CSi was higher than that for C(TF)₁CSi (8 mV, in Fig. 4e), indicating a higher surface photovoltage (−15 mV) and more significant reduction of surface work function (15 mV). According to our previous work⁴⁵, the higher surface photovoltage could accompany with the stronger built-in electric field existed in the C(TF)₆CSi. The intensity of the built-in electric field is quantitatively evaluated the carrier transfer efficiency. For a given photoelectrode, the built-in electric field is usually related with the surface photovoltage and surface charge density by following equation⁴⁷,

$${E}_{{{\rm{b}}}}={(-2{V}_{{{\rm{sph}}}}{\rho }_{{{\rm{s}}}}/{\varepsilon }_{{{\rm{r}}}}{\varepsilon }_{0})}^{0.5}$$

(4)

In the expression, E_b, V_sph, ρ_s, ε_r and ε₀ denote the built-in electric field intensity, surface photovoltage, surface charge density, relative dielectric permittivity of silicon (a fixed value of 11.7) and vacuum permittivity (a constant of 8.85 × 10⁻¹⁴F cm⁻¹), respectively⁴⁵. In accordance with the research findings reported by Le Formal et al. ⁴⁸, the quantity of carriers accumulated on the photoelectrode surface exhibits a linear correlation with the integral value derived from the difference between transient photocurrent density and steady-state photocurrent density, as illustrated in Fig. S44. The surface charge density of C(TF)₆CSi in the PEC NO₃RR process was ~1.6 times higher than that of C(TF)₁CSi. Combining the obtained data from KPFM measurements and transient PEC reactions, the built-in electric field of C(TF)₆CSi for PEC NO₃RR is calculated to be 4.17 kV cm⁻¹ (Fig. 4f), which is ~1.9 times higher than that of C(TF)₁CSi. This enhancement is originated from more carriers through the (TF)₆ protection layer and less depletion on the photovoltage of Si-based photocathode. In fact, the built-in electric field of C(TF)₆CSi is close to that of conventional n⁺p-Si in the photovoltaic devices⁴⁹. Thus, the architecture of (O/M)_n protection layer can accelerate the carrier-dynamic behaviors and drastically improve the PEC performance, including high photocurrent density and good durability.

In summary, we have proposed a universal approach to construct a low-resistance oxide-based protection layer for a practical Si-based hierarchical photocathode by employing multiple carrier-tunnelling paths. This approach allows the protection layer to realize facile and low-loss photogenerated carrier transport, while providing a sufficient thickness for maintaining the required durability. We have demonstrated thoroughly this approach is common and feasible in various oxide/metal materials. Accordingly, this approach is expected to speed up the pace of development of practical PEC devices. Beyond photoelectrochemistry, the design of low-resistance oxide-based layers with nanoscale multilayer structure can be extended to other chemical and electric fields, such as photocatalysis, electrocatalysis, photovoltaic cells, transistors, and supercapacitors.

Methods

Materials and chemicals

KNO₃ (>99%) and KOH (85%) were received from Sinopharm Chemical Reagent Co., Ltd. P₂O₅ (99.9995%) was received from Meryer (Shanghai) Biochemical Technology Co., Ltd. Silicon wafer was purchased from Hefei KJ Materials Technology Co., Ltd. Argon (Ar, >99.999%) was purchased from Changsha Rizhen Gas Co., Ltd. Iron target (99.95%), copper target (99.99%), titanium target (99.995%), platinum target (99.99%) and cerium target (99.99%) were received from ZhongNuo Advanced Material (Beijing) Technology Co., Ltd.

The preparation of electrolyte

Dissolve 66 g of KOH and 5.06 g of KNO₃ in 100 mL of pure water, cool the mixed solution to ambient temperature and then transfer it to a 1L-volumetric flask. After adjusting the volume to 1 L, the pH value of the mixed solution is 13.8.

Preparation of the samples

To prepare the heterogenous junction of silicon (n⁺p-Si), p-type Si wafer was put into 2% (mass/volume) potassium hydroxide (KOH) solution and etched at a temperature of 80 °C for 1 h. The etched Si was cleaned using 5% HF solution to remove the SiO₂ layer. The phosphorus pentoxide (P₂O₅) powders were ultrasonically dissolved in 10 mL anhydrous ethanol to prepare 0.1 M P₂O₅ solution. 1–2 drops of the P₂O₅ solution were applied onto the surface of the treated Si wafer, and the wafer was quickly annealed at 800 °C for 8 min to form the n⁺p-Si.

To enhance electron transport capability, the carbon (C) layer was deposited on the surface of n⁺p-Si (labeled as CSi) by methane (CH₄) plasma treatment at 800 °C for 210 s. The prepared CSi wafers (4 × 4 cm²), flat Si wafers (4 × 4 cm²), FTO wafers (1 × 2 cm²) and BK7 glass slides (2 × 2 cm²) as substrates were placed in the magnetron sputtering system (Chuangshiweina Co. Ltd, MSP-3200) to deposit the oxide-based protection layer and cocatalyst layer. Firstly, the argon (Ar) plasma treatment was carried out at 8.6 Pa for 20 min to clean the substrate surface. Subsequently, oxide-based protection layer with oxide/metal multilayer structure ((O/M)_n, where n is the number of nano-scale repeating unit) was deposited on the substrates by orderly sputtering different metal targets in pure Ar and mixed Ar/O₂ atmosphere. The deposition pressure, substrate temperature and bias voltage are 2.0 Pa, 50 °C, and 0 V, respectively. The iron (Fe), titanium (Ti), cerium (Ce), and palladium (Pd) targets were sputtered to fabricate the Fe unit, TiO₂ unit, CeO₂ unit, and Pd unit for assembling the (O/M)_n protection layer, respectively. Detailed deposition parameters of (TiO₂/Fe)_n (labeled as (TF)_n), (TiO₂/Pd)_n (labeled as (TP)_n), (CeO₂/Pd)_n (labeled as (CP)_n) and (CeO₂/Fe)_n (labeled as (CF)_n) are listed in Tables S7–S10, respectively. The n value of (O/M)_n protection layer was controlled by changing the deposition time, as shown in Tables S7–S10. Finally, Fe-copper (Cu) alloy (FeCu) cocatalysts were deposited by co-sputtering Fe and Cu target with 200-W radio-frequency (RF) power and 4-W direct-current (DC) power, respectively.

The rear surface of the Si-based photocathode modified with the (O/M)_n protective layer and FeCu cocatalyst was first subjected to polishing treatment. Subsequently, a Ti film with a thickness of approximately 500 nm was deposited onto the polished rear surface, and a Cu metal wire was attached to form an ohmic back contact. Silver paste was then applied to secure the Cu wire in place. After drying, the entire rear surface and a portion of the front surface of the Si-based photocathode was encapsulated with epoxy resin, which defined an exposed active area of ~0.2 cm². The actual geometric area of the exposed surface was accurately measured via calibrated digital imaging and ImageJ³². Considering that the thickness of FeCu cocatalyst layer is ~28 nm, the loaded alloy mass is approximately 4.6 μg.

Composition and structure characterization

Aiming to determine the microstructure and composition of the film, field emission scanning electron microscopy (FESEM, Hitachi SU8220) was used to observe the surface and cross section of the samples. The nanoscale multilayer structure of Si-based photocathodes was further examined by high-resolution transmission electron microscopy (HRTEM, Tecnai G2 F20 S-Twin). Scanning transmission electron microscopy (STEM, Tecnai G2 F20 S-Twin) with a high angle annular dark field (HAADF) detector and energy dispersive x-ray spectroscopy (EDS) were employed in the analysis of compositional distributions in the photocathodes. The elements and electron structure of the photocathodes were conducted by a Shimadzu AXIS SUPRA X-ray photoelectron spectroscopy (XPS) instrument. The binding energies of the elements were referenced to the C1s peak (284.8 eV) arising from adventitious carbon. The near-edge x-ray absorption fine structure spectroscopy (NEXAFS) of Fe and Cu L-edges for Si-based photocathode were detected at the Materials Science Beamline (MSB) at the Elettra synchrotron light source in Trieste (Italy) using a Specs Phoibos 150 multichannel electron energy analyzer, photon energies that had been calibrated against the Fermi level using an Ar ion etched and clean Cu foil standard, and KolXPD obtained from https://www.kolibrik.net/en/solutions-products/kolxpd was used in the curve fitting of species peaks. To verify the crystal structure of Si-based photocathodes, X-ray diffraction (XRD) spectra were measured using a Bruker D8 Advance instrument with the radiation of Cu Kα in the 2θ range from 30° to 80° and a grazing angle of 1° at a scan rate of 2°/min. Photoluminescence (PL) spectra were measured at room temperature on a fluorescence spectrophotometer (Fls-980, Edinburgh) with an excitation wavelength of 405 nm. Photo-assisted scanning Kelvin probe force microscope (KPFM) tests were performed on an Atomic Force Microscopy system (Asylum Research Cypher S, Oxford Instruments) to obtain the surface potentials and morphologies (experiment setup shown in Fig. S45). Inductively coupled plasma optical emission spectrometry (ICP-OES) was carried out on the Agilent 720ES to quantify the concentration of Ti and Ce ions in the electrolytes after 12-h reactions. To investigate the wettability characteristics, the contact angles between the reaction electrolytes (1 M KOH + 50 mM KNO₃) and photocathode surface were measured using a Dataphysics OCA 20 contact angle goniometer.

The optical diffuse reflectance and transmittance were monitored on a UV-visible-near-IR spectrophotometer (Hitachi, UH-4150) from 300 to 2600 nm. The Kubelka-Munk theory was generally used for the analysis of diffuse reflectance (R) spectra to obtain the absorption coefficient (α) of the Si-based photocathodes as followed.

$$F(R)={(1-R)}^{2}/2R=\alpha$$

(5)

where F(R) was Kubelka–Munk function. The optical energy bandgap of the samples was calculated by using the classical relation of optical absorption, where B, E_g and hυ denoted as the band tailing parameter, the optical band gap and the photon energy, respectively. The value of m should be taken as 0.5 or 2, corresponding to the direct or indirect allowed transition which dominated over the optical absorption, respectively.

$$ah\nu=B{(h\nu -{E}_{{{\rm{g}}}})}^{m}$$

(6)

Photoelectrochemical measurements

The PEC performance of the Si-based photocathodes was performed with a CHI-760e Instruments workstation in an H-type cell with three electrodes under simulated 1 sun illumination (100 mW/cm², AM 1.5 G). Prior to testing, the light intensity was calibrated using an optical power meter. The pretreated Nafion 117 membranes with 1-h treatments with 5% H₂O₂ and DI water at 80 °C were used as the proton conductive cation exchange membranes. The electrolytes in both chambers were 8 mL mixed solution containing 1 M KOH and 50 mM KNO₃. To increase the mass transfer, the electrolytes were stirred by magnetic stirring apparatus with magnetons in the PEC measurements. The working electrode, counter electrode and reference electrode were the Si-based photocathode, platinum foil and Ag/AgCl (filled with saturated KCl solution), respectively. The voltage was not iR corrected. Readings for the Ag/AgCl electrode were recalculated for the reversible hydrogen electrode (RHE) as follows.

$${E}_{{{\rm{RHE}}}}={E}_{{{\rm{Ag}}}/{{\rm{AgCl}}}}+0.059\times {{\rm{pH}}}+0.197$$

(7)

During PEC reactions, the Ar flow was constantly purged into the electrolyte from one side of the cathode chamber to maintain an inert gas atmosphere. Chronopotentiometry tests were conducted at the potential of 0.1 V vs. RHE, 0 V vs. RHE, −0.1 V vs. RHE and −0.2 V vs. RHE, each PEC NO₃⁻ reduction reaction (NO₃RR) test was repeated three times at least. Linear sweep voltammetry (LSV) tests were conducted at a scan rate of 50 mV/s with chopped light alternating between dark and light every 3 s. Error bars represent standard deviations from three independent PEC measurements at least.

The intensity-modulated photocurrent spectroscopy (IMPS) of the photocathode was obtained using a Zahner electrochemical workstation at applied potentials of 0 V, –0.1 V and −0.2 V vs. RHE. The light-assisted electrochemical active surface area (ECSA), open-circuit voltage (OCP), and light-assisted EIS of the photocathodes were measured using the CHI-760e Instruments workstation. The scan rate for ECSA test was 20 mV/s, 40 mV/s, 60 mV/s, 80 mV/s, and 100 mV/s and the double layer capacitance (C_dl) was obtained by calculating the slope.

Production analysis

The ammonia (NH₃) production was quantified by the indophenol blue method, supplemented with ¹H nuclear magnetic resonance (NMR) measurements. The indophenol blue method was implemented with three chromogenic reagents, whose compositions were defined as follows: (1) 1 M sodium hydroxide (NaOH) solution incorporated with 5 wt% sodium citrate (Na₃C₆H₅O₇) and 5 wt% salicylic acid (C₇H₆O₃); (2) 0.05 M sodium hypochlorite (NaClO) solution; (3) 1 wt% sodium nitroferricyanide (Na₂[Fe(CN)₅NO]·2H₂O) solution³². In the specific experimental protocol, 0.9 mL of the post-photoelectrolysis electrolyte was mixed with 1 mL of reagent (1), 0.5 mL of reagent (2) and 0.1 mL of reagent (3) sequentially. Subsequently, the mixed solution was kept in a dark environment for 2 h. The UV-visible spectrophotometer was employed to determine the absorbance at a characteristic wavelength of 656 nm, and the obtained absorbance values exhibited a good linear dependence on the NH₃ concentration within the tested range. The standard curves for quantitative determination of NH₃, which were constructed respectively via the indophenol blue method and ¹H NMR spectroscopy, were depicted in Supplementary Figs. S19 and S22. The calculation equations for NH₃ yield rate and FE were provided as follows³². All electrolytes collected after PEC reactions were measured at least three times to ensure data reliability. In Eq. 8, the parameters were defined as follows: C represented the NH₃ concentration (mg/L); V referred to the volume of the electrolyte involved in the reaction (L); t denoted the total reaction duration (h); and A was the area of the photocathode (cm²). In Eq. 9, n was the number of electrons transferred during the reaction process, which was assigned a value of 8 corresponding to the reduction of NO₃⁻ to NH₃. F was Faraday constant, with a fixed value of 96,500 C/mol. Additionally, M represented the molar mass of NH₃ (g/mol), while Q denoted the total electric charge consumed throughout the PEC reaction (C).

$${{\rm{Yield}}}\,{{\rm{rate}}}\,({{\rm{mg}}}/{{\rm{cm}}}^{2}/{{\rm{h}}})=\frac{C\times V}{t\times A}$$

(8)

$${{\rm{FE}}}\,(\%)=\frac{C\times V\times n\times F}{M\times Q}\times 100\%$$

(9)

The Griess test was employed to quantify the nitrite (NO₂⁻) concentration in post-photoelectrolysis electrolytes³². The chromogenic reagent for this test was prepared by dissolving 0.04 g of N-(1-naphthyl) ethylenediamine dihydrochloride and 0.8 g of sulfonamide in a mixed solution consisting of 2 mL of phosphoric acid and 10 mL of deionized water. In the detection process, 50 μL of the electrolyte was first subjected to 100-fold dilution with deionized water. Subsequently, 100 μL of the aforementioned chromogenic reagent was added to the diluted sample, and the mixture solution was kept in the dark for 20 min. Following this treatment, the absorbance of the mixture was measured at a wavelength of 540 nm using a UV-visible spectrophotometer. A set of sodium nitrite (NaNO₂) solutions with gradient concentrations was utilized to establish the corresponding standard curve, as shown in Fig. S19. The calculation methods for NO₂⁻ yield rate and FE were consistent with those for NH₃, as detailed in Eqs. 8 and 9. Notably, no other reaction products were detected in the electrolytes below the detection limit.

¹H NMR measurements

The NMR spectroscopy was applied for the determination of NH₃ yield rate. After a 1-h PEC reaction, the resultant electrolyte was subjected to a dilution as the pH value was adjusted to approximately 3 using 1 M hydrochloric acid (HCl). Subsequently, 0.5 mL of the diluted electrolyte (with pH calibrated to 3) was pipetted into NMR sample tube. Then, 100 μL of deuterated dimethyl sulfoxide (d6-DMSO) was added as the deuterated solvent, along with 50 μL, 5.0 mM sodium 3-(trimethylsilyl)-1-propanesulfonate (DSS) solution as an internal standard solution. All NMR spectra were acquired on a 600 MHz NMR spectrometer (Bruker) fitted with a low-temperature probe, with the final data representing the cumulative outcome of 256 scans. Additionally, the standard curves for ¹⁴NH₃ were established by measurements performed on ¹⁴NH₄Cl reference samples.

Electrical measurements

The layer thickness and temperature-dependent electrical experiments were performed in a two-electrode cell in micro electrical platform as shown in Fig. S46. During the material preparation, one end of FTO wafer was masked, and one unit layer of (Fe)_n, (TiO₂)_n, or (TF)_n (n = 1, 2, 4, 6, 8) was deposited on the remaining part. The thickness of unit layers for (Fe)_n and (TiO₂)_n was 18, 9, 4.5, 3 and 2.25 nm corresponding to n values of 1, 2, 4, 6, and 8, respectively. In addition, the thickness of unit layers for (TF)_n was 18/18, 9/9, 4.5/4.5, 3/3, and 2.25/2.25 nm corresponding to n values of 1, 2, 4, 6, and 8, respectively. The samples were fabricated with four Au contacts on the surfaces via thermal evaporation, in order to further enhance electron transport. In the photoexcitation electrical measurements, Si-based photocathodes were investigated and the detailed fabrication process has been described above.

The current versus voltage (I-V) curves were carried out using the combination of a Keithley 2450 Source Meter and a probe station. For the samples, two probes from the probe station were connected to the source meter and subsequently pressed onto the deposited region and masked region, respectively. The substrate bias was scanned from −10 to 10 mV while monitoring the current density and the resistance was calculated from the I-V curves. The units of (Fe)_n, (TiO₂)_n, and (TF)_n were used to analyze the effect of film thickness on resistance at a constant temperature. The units of (Fe)₆, (TiO₂)₆, and (TF)₆ were also used to analyze the effect of temperature on resistance from 15 to 200 °C.

In the photoexcitation electrical measurements, a laser light with a wavelength of 520 nm was used to illuminate the Si-based photocathodes. Two probes were pressed onto the photocathode surface and Cu belt (see Fig. 4a(ii)), and the electrons conducted along the vertical direction. The substrate bias was scanned from −1 to 1 V. Before illumination, all samples were placed in darkness and tested for the I-V curves.

Finite element analysis

The finite element analysis was performed using COMSOL Multiphysics (version 6.3). Tables S1 and S2 showed the material properties of TiO₂ and Fe and the constructed electrical filed. In this simulation, the effect of the material’s cross-sectional area was neglected, and with nanoscale film thickness, a two-dimensional (2D) model was built. We selected the EC physical field and the semi-physical field for simulating the Fe and TiO₂, respectively. Under the influence of these two physical fields, the simulation showed that the current within the bulk material reached a steady-state value, while abrupt changes were allowed to occur at the Schottky junction contact. The Schottky junction equation could be described as followed.

$$J=A * {T}^{2}\exp \left(-\frac{q{\phi }_{{{\rm{B}}}}}{kT}\right)\left[\exp \left(\frac{qV}{kT}\right)-1\right]$$

(10)

Among them, J was the current density, A^* the Richardson constant related to the material, T the temperature, q the charge of an electron, ϕ_B the height of the Schottky barrier, k the Boltzmann constant, and V the applied voltage. An endogenous electric field can be generated by applying an external voltage at the model boundaries. Driven by this electric field, electrons migrated along the thickness direction of the film, resulting in distinct dielectric loss characteristics.