High-density asymmetric iron dual-atom sites for efficient and stable electrochemical water oxidation

Abstract

Double-atom catalysts (DACs) have opened distinctive paradigms in the field of rapidly developing atomic catalysis owing to their great potential for promoting catalytic performance in various reaction systems. However, increasing the loading and extending the service life of metal active centres represents a considerable challenge for the efficient utilization of DACs. Here, we rationally design asymmetric nitrogen, sulfur-coordinated diatomic iron centres on highly defective nitrogen-doped carbon nanosheets (denoted A-Fe₂S₁N₅/SNC, A: asymmetric), which possess the atomic configuration of the N₂S₁Fe-FeN₃ moiety. The abundant defects and low-electronegativity heteroatoms in the carbon-based framework endow A-Fe₂S₁N₅/SNC with a high loading of 6.72 wt%. Furthermore, A-Fe₂S₁N₅/SNC has a low overpotential of 193 mV for the oxygen evolution reaction (OER) at 10 mA cm⁻², outperforming commercial RuO₂ catalysts. In addition, A-Fe₂S₁N₅/SNC exhibits extraordinary stability, maintaining > 97% activity for over 2000 hours during the OER process. This work provides a practical scheme for simultaneously balancing the activity and stability of DACs towards electrocatalysis applications.

Introduction

Electrocatalytic water splitting and metal‒air batteries are considered promising options in sustainable energy conversion technologies. As the proton and electron provider on the anode electrode, the oxygen evolution reaction (OER) plays an important role in these devices. However, the four-electron OER process involves O–H bond rupture and O–O bond formation, the kinetics of which are sluggish; this is necessary for the development of highly efficient OER catalysts. To date, iridium- and ruthenium-based catalysts are still the benchmark commercial OER catalysts; however, their low crustal abundance, high cost, and inferior durability limit their long-term development^1,2,3. Consequently, developing earth-abundant and high-intrinsic-activity OER catalysts is important for improving current energy technology.

Recently, single-atom catalysts (SACs) with diverse well-defined coordination configurations have attracted increasing attention due to their maximized atomic utilization and high catalytic activity^4,5. Among the SACs developed in recent years, metal–N–C catalysts with symmetric metal–N₄ coordination structures have been extensively studied⁶. However, their OER performance is still unsatisfactory compared with that of benchmark iridium- and ruthenium-based catalysts due to their symmetric electron distribution. To overcome this challenge, it is necessary to adjust the local coordination environment around metal–N₄ SACs properly. The adsorption strength of OER intermediates can be optimized by breaking the inherent symmetric charge distribution⁷. In addition, SACs contain unitary active centers, resulting in the binding of only one oxygen-containing OER intermediate (e.g., O*, OH*, OOH*), which limits their high OER activity⁸. With the expansion of SACs, double-atom catalysts (DACs) have emerged as ideal candidates for highly efficient OER. The inherent advantages of SACs and the newly identified emerging binding sites/modes that occur within the catalytic centers for DACs are promising for overcoming the inherent activity limits⁹. The heterogeneous nature of the coordination configuration may lead to the formation of distinct catalytically active sites and reaction pathways, which in turn can influence the overall catalytic performance. Moreover, the atomic spatial configuration, the surrounding chelating atoms, and the coordination number are of paramount importance in optimizing the coordination environment of DACs¹⁰. Accordingly, dual-metal sites can be classified into three categories based on their coordination environment: binuclear homologous dual-metal sites, binuclear heterologous dual-metal sites, and isolated dual-metal sites¹¹. Among these, the symmetric distribution of M₂ sites on M₂–N–C DACs belongs to the binuclear homologous dual-metal site pairs¹². These catalysts have higher metal loadings than the isolated metal site catalysts due to the preferred diatomic pairs at high metal loadings. The OER involves multiple proton-coupled electron-transfer processes, and the symmetric distribution of M₂–N–C DACs possesses unitary active sites, which makes it difficult to achieve the optimal adsorption state of multiple reaction intermediates¹³. Consequently, the asymmetrical charge distribution on homonuclear dual-metal sites can maximize the potential of DACs for multistep reactions, which makes the optimization of activity and selectivity feasible.

Iron (Fe), which is a congener of Ru for the first transition series, is considered a promising OER catalyst because of its earth abundance, low cost, and biocompatibility^14,15. It has been reported that the monomeric amidate-ligated Fe(III)-aqua complex can be used as an electrocatalyst for the OER with a Faradaic yield of 45%¹⁶. However, the mononuclear Fe-based complex may collapse during the OER due to hydrogen atom transfer at the active Fe sites, which leads to stability problems¹⁷. Interestingly, the binuclear Fe complex [tpa(H₂O)FeOFe(H₂O)tpa](ClO₄)₄ was constructed as an efficient OER catalyst, which was six times greater than that of the mononuclear [Fe(OTf or Cl)₂(tpa)] complex¹⁸. Notably, the OER kinetics on Fe–N–C were sluggish because of the strong O* binding strength on the Fe sites¹⁹. Consequently, the rate-determining step for Fe–N–C towards the OER is the conversion of *O to *OOH. Inspired by the progress in OER research on Fe-based complexes and Fe–N–C, the construction of an asymmetric iron dual-atom catalyst is one of the most promising approaches for developing highly active and durable OER.

In this work, we design a sulfur, nitrogen-codoping carbonaceous support (SNC) with high defects to anchor Fe‒Fe dual-atom sites (A-Fe₂S₁N₅/SNC) through a multilayer stabilization strategy involving defect capture and coordination riveting. Comprehensive characterization verified that the asymmetric diatomic FeS₁N₂–FeN₃ moieties were densely dispersed on the support, with a high metal mass loading of 6.72 wt%. The A-Fe₂S₁N₅/SNC catalyst displays high OER activity. In alkaline electrolytes, A-Fe₂S₁N₅/SNC has a low overpotential of 193 mV at 10 mA cm⁻². Under constant current, A-Fe₂S₁N₅/SNC can catalyze the OER for more than 2000 h, which results in extraordinary stability compared with other reported atomically dispersed electrocatalysts. The theoretical calculations indicate that the enhanced OER activity derived from spontaneous electric polarization occurred at the iron dual-atom sites of the asymmetric A-Fe₂S₁N₅/SNC catalyst. Moreover, the alkaline AEMWE (anion exchange membrane water electrolyzer) catalyzed by A-Fe₂S₁N₅/SNC displays good performance, indicating its broad practical application potential. In particular, other asymmetrically coordinated A-M₂S₁N₅/SNC catalysts (M = Co, Ni, Cu, and Mn) with high metal loading were also prepared and investigated, demonstrating the universality of the developed strategy.

Results

Synthesis and characterization of A-Fe₂S₁N₅/SNC

A continuous two-step pyrolysis and carbonization approach, as shown in Fig. 1, was proven effective for gradually introducing numerous defects and high non-metal heteroatoms into A-Fe₂S₁N₅/SNC. Specifically, a certain mass ratio of 2-benzimidazolethiol and melamine was first ground into a homogeneous mixture, which underwent confined pyrolysis in a flowing inert atmosphere (Fig. 1a). The thiol group on the 2-benzimidazolethiol was easily extracted by hydrogen to form free radicals due to its high activity. Through free radical reactions, the active C atom on melamine is likely covalently bonded with the thiol group to form C–S–C bonds. At high temperatures, melamine transforms into a carbon nitride structure (NC) and accompanied by the polymerization of NC frameworks, numerous defects spontaneously form²⁰. During the thermal polymerization of 2-benzimidazolethiol, NC was encapsulated through a C–S–C bond linkage, and the constructed carbon framework was converted to S, N-codoped carbon nanosheets (SNC) after in situ restricted pyrolysis (Fig. 1b)²¹. In this matrix, sulfur was predominantly incorporated in the form of C–S–C. The presence of numerous sulfur-containing sites, coupled with the highly porous structure of the S–C, facilitated the preparation of diatomic iron catalysts with high loading through defect capture and sulfur tethering effects²². During the heating process, the edge S site formed by C–S bond breakage along with the adjacent N atom can effectively anchor individual Fe atoms through sulfur tethering and defect capture effects separately. Following initial anchoring, neighboring Fe atoms can migrate and interact with one another, resulting in the formation of a Fe‒Fe bond on the surface of the graphene substrate. Finally, diatomic A-Fe₂S₁N₅/SNC catalysts were obtained at selected temperatures through carbonization reduction under an N₂ atmosphere (Fig. 1c). The mass percentage of Fe in the A-Fe₂S₁N₅/SNC catalysts was calculated to be 6.72 wt% through an inductively coupled plasma‒optical emission (ICP‒OES) test.

Fig. 1: Two-step pyrolysis and carbonization synthesis process of A-Fe₂S₁N₅/SNC.

a The certain mass ratio of 2-benzimidazolethiol and melamine underwent confined pyrolysis in a flowing inert atmosphere. b The free radical reactions, polymerization processes, and thermally induced decomposition of 2-benzimidazolethiol and melamine during pyrolysis were performed to synthesize highly defective SNC nanosheets. c The SNC anchored dual Fe atoms through sulfur tethering and defect capture effects.

Scanning electron microscopy (SEM) revealed the curved and wrinkled nanosheet morphology of A-Fe₂S₁N₅/SNC (Fig. 2a and Supplementary Fig. 1), which was inherited from the SNC nanosheets (Supplementary Figs. 2, 3). The thickness of the nanosheets was calculated to be 4.5 nm (Supplementary Fig. 4). A-Fe₂S₁N₅/SNC consists of defect-rich thin nanosheets with many edge activity sites (Fig. 2b). The surface/edge defects can more easily anchor the metal, further changing the metal charge via atomic modulation strategies and affecting the electrocatalytic activity of the materials²³. The energy-dispersive X-ray spectroscopy (EDS) images and electron energy loss spectroscopy (EELS) mapping of A-Fe₂S₁N₅/SNC at high magnification confirmed the uniform distribution of Fe, S, N, and C over the entire A-Fe₂S₁N₅/SNC (Fig. 2c and Supplementary Fig. 5). Furthermore, no obvious metal particles/clusters are observed in the high-angle annular dark-field scanning transmission electron microscopy (HAADF-STEM) image (Supplementary Fig. 6), indicating the atomically dispersed Fe in the SNC matrix²⁴. Additionally, the morphologies of Fe₂N₆/NC, Fe₁S₁N₃/NC, and Fe₁N₄/NC are consistent with those of A-Fe₂S₁N₅/SNC (Supplementary Figs. 7–9). In Fig. 2d, aberration-corrected HAADF-STEM distinctly detects high-density and paired bright dots at the atomic level (circled by red dashed lines), corresponding to the atomically dispersed Fe‒Fe atom pair. The ratio of Fe diatomic pairs is determined by counting the pair distance for more than 140 sites on the AC‒STEM image of A-Fe₂S₁N₅/SNC (Supplementary Fig. 10), which reaches 80.7%. The intensity distribution along the X–Y direction in region A indicates that the spacing between the diatomic pair of Fe is ~0.25 nm (Fig. 2e), which closely matches the effective diameter of the Fe atoms in Fe₂N₆/NC (Supplementary Fig. 11). The EELS spectrum of small pixels of the annular dark field (red rectangle in Supplementary Fig. 5a) revealed visible signals of Fe, S, N, and C, confirming the densely populated FeS₁N₂–FeN₃ sites (Fig. 2f). Additionally, the relevant X-ray diffraction (XRD) and Raman characterization results are depicted in Supplementary Figs. 12, 13. N₂ adsorption and desorption isotherms (Supplementary Fig. 14 and Supplementary Table 1) revealed that the A-Fe₂S₁N₅/SNC, Fe₁S₁N₃/NC, Fe₂N₆/NC, Fe₁N₄/NC, and SNC catalysts had comparable Brunauer‒Emmett‒Teller (BET) specific surface areas. Pore size distribution analysis via the Barrett‒Joyner‒Halenda (BJH) model revealed a narrow peak at approximately 2.3 nm. The light element (i.e., C, N, S, H) contents of A-Fe₂S₁N₅/SNC were determined via CHNS analysis (Supplementary Table 2), which was consistent with the EDS analysis results. Furthermore, the ratio of Fe to S provides indirect evidence for the presence of Fe dual sites in A-Fe₂S₁N₅/SNC.

Fig. 2: Morphology and atomic local structure characterization of A-Fe₂S₁N₅/SNC.

a SEM; b HRTEM; and c EDS mapping images of Fe, S, N, and C for A-Fe₂S₁N₅/SNC (scale bar: 200 nm). d Aberration-corrected HAADF-STEM image of A-Fe₂S₁N₅/SNC. Inset: magnified region D, and the dual Fe sites are indicated by yellow ovals. e 3D model and the statistical Fe–Fe distance in the observed diatomic pairs A (top) and the intensity profile along the X–Y direction in regions B and C (bottom) of (d). f Electron energy loss spectroscopy (EELS) of A-Fe₂S₁N₅/SNC. g Fe K-edge XANES spectra (inset shows the magnified corresponding near-edge region), h FT k³-weighted Fe K-edge EXAFS spectra, and i WT-EXAFS plots of A-Fe₂S₁N₅/SNC, Fe foil, FePc, and FeS₂. Inset i: atomic configuration model of A-Fe₂S₁N₅/SNC.

The electronic and atomic structural information for the Fe species in the samples was further examined via X-ray absorption spectroscopy (XAS). Figure 2g and Supplementary Fig. 15 show the Fe K-edge X-ray absorption near-edge structure (XANES) spectra of A-Fe₂S₁N₅/SNC, Fe₁S₁N₃/NC, Fe₂N₆/NC, and Fe₁N₄/NC. The adsorption threshold positions of A-Fe₂S₁N₅/SNC, Fe₁S₁N₃/NC, Fe₂N₆/NC, and Fe₁N₄/NC were both located between those of FeS₂ and Fe₂O₃, which indicates that the average metal valence states of the Fe species in A-Fe₂S₁N₅/SNC, Fe₁S₁N₃/NC, Fe₂N₆/NC, and Fe₁N₄/NC were situated between them²⁵. From the XANES spectra shown in Supplementary Fig. 16, the average oxidation state of Fe in A-Fe₂S₁N₅/SNC obtained by fitting was 2.78. The Fourier transformed k³-weighted extended X-ray absorption fine structure (EXAFS) data of A-Fe₂S₁N₅/SNC, Fe₁S₁N₃/NC, Fe₂N₆/NC, and Fe₁N₄/NC are displayed in Fig. 2h and Supplementary Fig. 17. The intense peak at 1.47 Å confirms the existence of a Fe‒N scattering path in A-Fe₂S₁N₅/SNC, Fe₁S₁N₃/NC, Fe₂N₆/NC, and Fe₁N₄/NC. Concurrently, Fe‒Fe interactions of ~2.27 Å are found in both A-Fe₂S₁N₅/SNC and Fe₂N₆/NC, suggesting that dual iron atoms exist in the form of direct bonding^26,27. Notably, the peak appears at 1.84 Å in A-Fe₂S₁N₅/SNC, indicating the occurrence of Fe‒S scattering. The atomic configurations of A-Fe₂S₁N₅/SNC, Fe₂N₆/NC, and Fe₁N₄/NC were subsequently studied via the Fe K-edge wavelet transform (WT)-EXAFS (Supplementary Fig. 18), which demonstrated the isolated characteristics of the Fe species²⁸. The quantitative EXAFS fitting is studied to authenticate the local coordination configuration of Fe in A-Fe₂S₁N₅/SNC, Fe₁S₁N₃/NC, Fe₂N₆/NC, and Fe₁N₄/NC (Fig. 2i and Supplementary Figs. 19–22). The fitting results revealed that the coordination configurations of A-Fe₂S₁N₅/SNC, Fe₁S₁N₃/NC, Fe₂N₆/NC, and Fe₁N₄/NC are the FeS₁N₂-FeN₃, Fe₁S₁N₃, Fe₂N₆, and FeN₄ moieties, respectively. The first shell of A-Fe₂S₁N₅/SNC was fitted with Fe–N, Fe–S, and Fe–Fe back-scattering paths. Coordination numbers of approximately 2.5, 0.5, and 1.0 were determined using Fe–N, Fe–S, and Fe–Fe in A-Fe₂S₁N₅/SNC, with average bond lengths of 1.90, 2.30, and 2.56 Å, respectively (inset of Fig. 2i and Supplementary Table 3). Furthermore, the atomic structure of A-Fe₂S₁N₅/SNC was evaluated by comparing the experimental spectra with the simulated XANES spectra. XANES calculations were conducted on FeS₁N₂–FeN₃ moieties with different Fe–S or C–S positions (Supplementary Fig. 23). It appears that the FeS₁N₂–FeN₃ structural motif with direct Fe‒Fe bonding, one Fe bonded with one S atom and two N atoms, and the other Fe connected by three N atoms in Supplementary Fig. 23a, is in good agreement with the experimental spectra. In addition, the C, N, and S types in the carbon matrix were confirmed by soft XAS and XPS spectra (Supplementary Figs. 24-29). It appears that π* (C = C ring), π* (C–N–Fe or C–S–Fe), and σ* (C–C ring) exist in support of A-Fe₂S₁N₅/SNC, Fe₂N₆/NC, and Fe₁N₄/NC. Furthermore, three obvious peaks in the N K-edge soft XAS and N 1s XPS spectra for A-Fe₂S₁N₅/SNC, Fe₁S₁N₃/NC, Fe₂N₆/NC, Fe₁N₄/NC, and SNC represent the coexistence of pyridinic N, pyrrolic N, and graphitic N sites in the carbon matrix. Additionally, the C–S–N (C) species suggest the successful doping of the S atom into the NC fragment for A-Fe₂S₁N₅/SNC²⁹. The Fourier transform infrared (FTIR) spectrum of the A-Fe₂S₁N₅/SNC catalyst (Supplementary Fig. 30) revealed all the non-metallic bonds that constitute the basic framework of the catalyst³⁰. The NH₃-TPD technique was used to titrate the acidic sites. As shown in Supplementary Fig. 31, negligible acidic sites were found across the A-Fe₂S₁N₅/SNC catalyst. These results indicate that N and S are doped into the structure of graphene³¹.

Electrocatalytic OER performance of A-Fe₂S₁N₅/SNC

To evaluate the alkaline OER electrocatalytic performance of A-Fe₂S₁N₅/SNC, we conducted electrochemical measurements in a 1.0 M KOH solution via a standard three-electrode configuration. The obtained A-Fe₂S₁N₅/SNC displayed more obvious advantages (Fig. 3a) in terms of overpotential (193 mV at 10 mA cm⁻²) than the Fe₁S₁N₃/NC (263 mV), Fe₂N₆/NC (280 mV), Fe₁N₄/NC (369 mV), SNC (409 mV), NC (437 mV), and even RuO₂ (255 mV). The optimal amount of Fe source was identified by the OER activities, and the results showed that the catalyst exhibited the optimized performance when the amount of iron (III) chloride was 40 mg (Supplementary Fig. 32). The corresponding LSV curves without iR compensation for the catalysts are shown in Supplementary Fig. 33. The A-Fe₂S₁N₅/SNC catalyst delivered a superior Tafel slope compared to Fe₁S₁N₃/NC, Fe₂N₆/NC, Fe₁N₄/NC, SNC, and RuO₂, suggesting faster OER kinetics (Supplementary Fig. 34a)³². Notably, A-Fe₂S₁N₅/SNC was the most active OER electrocatalyst among the reported SAC/DAC catalysts (Supplementary Fig. 34 and Supplementary Table 5), which might be attributed to the asymmetric iron dual-atom sites. The electrochemically active surface area (ECSA) represents the actual number of active sites of the electrocatalytic catalysts, which was evaluated through the double-layer capacitance (C_dl) of the catalysts³³. As derived from cyclic voltammograms at various sweep rates (Supplementary Fig. 35), the ECSA and C_dl of A-Fe₂S₁N₅/SNC were significantly greater than those of Fe₁S₁N₃/NC, Fe₂N₆/NC, Fe₁N₄/NC, and SNC (Supplementary Fig. 36 and Supplementary Table 4), confirming that the asymmetric configuration in A-Fe₂S₁N₅/SNC possessed more exposed reactive sites and hence enhanced the OER activity^34,35. To further determine whether the enhanced OER activity is solely contingent on the increased number of active sites, the ECSA-normalized OER polarization curves were calculated. From the ECSA-normalized LSV curves (Supplementary Fig. 37a), the A-Fe₂S₁N₅/SNC catalyst still exhibited the highest activity, suggesting intrinsically improved OER activity on the A-Fe₂S₁N₅/SNC catalyst due to the presence of asymmetric iron dual-atom sites³⁶. Electrochemical impedance spectroscopy (EIS) is a common indicator used to probe the interface resistance change between a catalyst electrode and an electrolyte³⁷. As illustrated in Supplementary Fig. 37b, the A-Fe₂S₁N₅/SNC catalyst exhibited a lower charge transfer resistance than the Fe₂N₆/NC catalyst. This suggests an accelerated charge-transfer mechanism following the addition of sulfur after S doping during the OER. Furthermore, the SNC support demonstrated a diminished charge transfer resistance relative to that of the NC support, indicating an increased electron-transfer rate within the SNC framework. As a result, the incorporation of the S dopant has been shown to increase the efficiency of charge transfer during the alkaline OER³⁸. The intrinsic catalytic activity of the catalysts was subsequently assessed by the calculated turnover frequency (TOF) shown in Fig. 3b, Supplementary Fig. 38a, and Supplementary Note 1. Notably, the A-Fe₂S₁N₅/SNC catalyst displayed an extremely high TOF throughout the entire overpotential window. In general, the TOF values for A-Fe₂S₁N₅/SNC were as high as 2875.0 O₂ h⁻¹ at an overpotential of 300 mV, 11 times and 183 times greater than those of Fe₂N₆/NC (273.0 O₂ h⁻¹) and Fe₁N₄/NC (15.7 O₂ h⁻¹), respectively. In addition, the mass activities (MA) of A-Fe₂S₁N₅/SNC were also greater than those of Fe₁S₁N₃/NC, Fe₂N₆/NC, and Fe₁N₄/NC (Supplementary Fig. 38b and Supplementary Note 2), demonstrating the high-intrinsic catalytic activity of A-Fe₂S₁N₅/SNC³⁹. The Faradaic efficiency (FE) was calculated in comparison to the amount of oxygen produced experimentally against the theoretical quantity. The FE of A-Fe₂S₁N₅/SNC was approximately 99.1% (Supplementary Fig. 39 and Supplementary Note 3), indicating that no side reactions occurred during the OER process. In addition, we evaluated the OER catalytic performance of the A-Fe₂S₁N₅/SNC under the industrial electrolytic water system (such as 80 °C, 30% KOH), and the results revealed that A-Fe₂S₁N₅/SNC has potential for industrial application (Supplementary Fig. 40).

Fig. 3: OER electrochemical activity and in situ XAFS characterization of A-Fe₂S₁N₅/SNC in 1.0 M KOH.

a LSV curves of the prepared catalysts. Note: the geometric surface area is 1.0 cm²; series resistance (Rs) are 2.7, 3.6, 4.5, 5.6, and 8.1 Ω for A-Fe₂S₁N₅/SNC, Fe₁S₁N₃/NC, Fe₂N₆/NC, Fe₁N₄/NC, and SNC, respectively. b TOF and mass activity at 300 mV, c normalized XANES spectra at the Fe K-edge for the OER (the insets are magnified images of the near-edge region for the OER), d k³-weighted FT-EXAFS at ex situ, 1.31, and 1.42 V vs. RHE (the shaded area highlights the shifts in the peak positions of Fe–N, Fe–S, and Fe–Fe in the first coordination shell), e chronopotentiometric measurement of A-Fe₂S₁N₅/SNC at 10 mA cm⁻² (no iR-correction).

An in situ XAFS analysis was performed to elucidate the structural evolution and real-time change in the Fe chemical state during the OER^40,41. A homemade cell was used to perform the in situ XAS tests (Supplementary Fig. 41) through the commonly used Lytle detector (in fluorescence mode) to collect all the spectra. As shown in Fig. 3c, the Fe K-edge absorption edge gradually shifted to higher energy, accompanied by an increase in the applied potential from ex situ conditions to 1.42 V vs. RHE, indicating that the average valence state of the Fe sites increased as the OER proceeded^42,43. The formation of higher valence Fe could be further supported by the first derivative XANES curves in Supplementary Fig. 42. To further probe the evolution of the coordination configuration for the Fe atoms in A-Fe₂S₁N₅/SNC, FT–EXAFS spectra were recorded at the Fe K-edges (Fig. 3d), the Fe–N, Fe–S, and Fe–Fe shells displayed obvious dynamic changes during the OER. The coordination peaks shift to the low-R region, which might be derived from bridging adsorption of the oxygen-containing intermediates (i.e., O*, OH*, OOH*) on dual Fe sites⁴⁴. By considering the Fe–N, Fe–S, Fe–Fe, and Fe–O scattering paths, the first coordinated shell of the EXAFS curve-fitting analysis for A-Fe₂S₁N₅/SNC is displayed in Supplementary Figs. 43, 44 and Supplementary Table 7. Accordingly, slight shrinkage of the Fe‒N bond length could fix the dual Fe sites during the OER, thus avoiding possible dissolution and improving stability⁴⁵.

The A-Fe₂S₁N₅/SNC electrode exhibited electrochemical durability after 2000 h of continuous operation under constant current test conditions (10 mA cm⁻²) with >97% retention (Fig. 3e), which demonstrated its feasibility in industrial applications⁴⁶. The stability of A-Fe₂S₁N₅/SNC was also validated via LSV curves before and after 50000 CV cycles (with no loss) (Supplementary Fig. 45). A comparison of the long-term OER durability of A-Fe₂S₁N₅/SNC and other reported electrocatalysts is summarized in Supplementary Fig. 46 and Supplementary Table 5. Undoubtedly, A-Fe₂S₁N₅/SNC was one of the most stable reported OER catalysts. The morphology and structure of A-Fe₂S₁N₅/SNC after the durability tests were examined via TEM, XRD, Raman, and XPS analyses (Supplementary Figs. 47, 48), which revealed negligible changes. After cyclic testing, the ICP results revealed that the Fe content in A-Fe₂S₁N₅/SNC remained at 97.9%, confirming the stability of the catalyst (Supplementary Table 6 and Supplementary Note 5).

To determine the universality of the synthesis method, several other 3D transition-metal (for example, M = Co [6.43 wt%], Cu [7.15 wt%], Ni [6.98 wt%], and Mn [7.01 wt%]) diatomic A-M₂S₁N₅/SNC catalysts were produced (Supplementary Fig. 49). As demonstrated in Supplementary Figs. 50–53, A-M₂S₁N₅/SNC possesses a thin and curly nanosheet morphology with no obvious metal particles. In the AC HAADF–STEM image, many dense dimers are observed, indicating the presence of metal dual-atom pairs in A-M₂S₁N₅/SNC. The unsymmetrically arranged MS₁N₂–MN₃ coordination configuration was further confirmed by XAS spectra and quantitative EXAFS fittings (Supplementary Figs. 54–60 and Supplementary Table 8). The corresponding OER performance of A-M₂S₁N₅/SNC is also shown in Supplementary Fig. 61. The successful preparation of A-M₂S₁N₅/SNC demonstrates the feasibility and scalability of the strategy proposed in this paper for preparing asymmetric homonuclear diatomic catalysts, providing potential for the industrialization of diatomic catalysts.

Theoretical study of A-Fe₂S₁N₅/SNC for the OER

To elucidate the mechanism of the oxygen electrochemical reaction on the asymmetrical A-Fe₂S₁N₅/SNC catalyst in comparison with those on the Fe₂N₆/NC and Fe₁S₁N₃/NC catalysts, the first-principles method was adopted to study their proton-coupling electron-transfer (PCET) processes⁴⁷. Supplementary Figs. 62–65 show the possible configurations of the Fe₂N₆/NC and A-Fe₂S₁N₅/SNC catalysts. Given the spin-polarized Fe, the relative energies between the ferromagnetic (FM) and anti-ferromagnetic (AFM) states for each model and the formation energies were systematically evaluated to confirm the ground states of the stable models. The Fe₂N₆/NC and A-Fe₂S₁N₅/SNC models were all AFM ground states, and the two Fe atoms were named the Fe-1 and Fe-2 sites for ease of distinction. The Fe-2 site connects with one S atom and two N atoms, and the Fe-1 site coordinates with only N atoms in the asymmetrical A-Fe₂S₁N₅/SNC model. The Fe₁S₁N₃/NC model is depicted in Supplementary Fig. 66. Initially, the adsorption of intermediates in the OER/ORR was evaluated with these models, and the corresponding adsorbed configurations and adsorption-free energies are shown in Supplementary Figs. 67–69, Supplementary·Data 1, and Supplementary Tables 9–11, respectively. The O and OH species have strong binding energies of −0.64 and −1.49 eV on the bare A-Fe₂S₁N₅/SNC surface, −0.44 and −1.25 eV on the bare Fe₂N₆/NC surface, and 1.62 and 0.49 eV on the bare Fe₁S₁N₃/NC catalyst, respectively, all deviating from the ideal adsorption energies of the optimal OER/ORR catalyst, with values of 2.46 and 1.23 eV; this indicated that the O or OH intermediates were prone to occupy the Fe sites in the electrochemical surroundings⁴⁸. The Pourbaix diagram clearly shows the stable states of the catalyst surface under electrochemical conditions, which are related to the pH and electrode potential⁴⁹. Figure 4a and Supplementary Fig. 70 depict the free energies of a series of coverage surfaces following electrode potential U vs. RHE sampling for the A-Fe₂S₁N₅/SNC, Fe₂N₆/NC, and Fe₁S₁N₃/NC models. Under real electrochemical OER conditions, such as 1.50 V vs. RHE, their surfaces were covered by 3O with two Fe sites for diatomic catalysts and 2OH for the Fe₁S₁N₃/NC catalyst, termed 3O-Fe₂S₁N₅/SNC, 3O-Fe₂N₆/NC, and 2OH-Fe₁S₁N₃/NC in the OER surroundings, respectively. According to the OER mechanism of single-atom catalysts reported by Pacchioni et al.⁵⁰, at least one unconventional chemical intermediate, such as 2*OH, *O + *OH, 2*O, or *O₂ (“*” marks the surface site), was found to be more stable than the classical intermediate, including *O and *OOH species, in all cases. Therefore, we systematically evaluated the adsorption of these OER intermediates, which are composed of classical and unconventional intermediates, on the abovementioned surfaces, and the corresponding optimized adsorption configurations and adsorption energies are displayed in Supplementary Figs. 71–73, Tables 12–14, and Supplementary·Data 2. The Gibbs free energy changes of the elementary steps of the OER are depicted in Fig. 4b, c and Supplementary Figs. 74–77. The unconventional complex (*O + *OH) was more stable than the classical *OOH intermediate at the Fe−2 site of the 3O-Fe₂S₁N₅/SNC and 3O-Fe₂N₆/NC surfaces and had similar adsorption energies of *2OH and *O. Hence, their cascade OER processes on the diatomic catalyst could proceed through unconventional intermediates: H₂O (l)→*OH → *2OH → *O + *OH → *O₂ → O₂ (g). Considering the instability of the 2OH-Fe₁S₁N₃/NC catalyst induced by the unusual intermediates *2OH and *O + *OH, the classical steps were considered only for the OER process. Compared with the symmetric 3O-Fe₂N₆/NC catalyst with an overpotential of 0.41 V, the asymmetric diatomic catalyst 3O-Fe₂S₁N₅/SNC had the lowest overpotential (0.32 V) when an S source was used, and it also excelled with the single-atom catalyst 2OH-Fe₁S₁N₃/NC (0.35 V), which was consistent with the experimental measurements. Their potential-determined steps (PDSs) were the second dehydrogenation process [*OH → *2OH (O)]. Figure 4d depicts the OER mechanism on the 3O-Fe₂S₁N₅/SNC catalyst in light of unusual intermediates. The overpotentials of the classical intermediates for the OER [H₂O (l)→*OH → *O → *OOH → O₂ (g)] were 0.32, 0.44, and 0.35 V with the same PDSs of *OH → *O for 3O-Fe₂S₁N₅/SNC, 3O-Fe₂N₆/NC, and 2OH-Fe₁SN₃/NC, respectively. For both classical and unconventional intermediates, A-Fe₂S₁N₅/SNC exhibited optimal electrochemical OER performance. The above OER results were based on an implicit solvent model (VASPsol), which can generally capture electrostatic interactions with diverse intermediates and neglect the influence of hydrogen bonds on their stabilization and consequent OER mechanism via an explicit solvent model⁵¹.

Fig. 4: Mechanistic exploration of A-Fe₂S₁N₅/SNC for the OER.

a Free energy of a series of coverage fraction surfaces sampled for the A-Fe₂S₁N₅/SNC catalysts; b The corresponding free energy change of the OER at the Fe−2 site; c comparison of Gibbs free energy changes of the OER intermediates in A-Fe₂S₁N₅/SNC and Fe₂N₆/NC at U = 0 V; d the proposed 4e^- mechanism of the OER on A-Fe₂S₁N₅/SNC.

The improved OER performance of 3O-Fe₂S₁N₅/SNC was attributed mainly to the enhanced oxygen capture ability induced by the involvement of an unsymmetric S source and dual-atom Fe sites. The Crystal Orbital Hamilton Population (COHP) results of the adsorbed O species between O and the active Fe site on the diatomic catalyst are shown in Supplementary Fig. 78. The 3O-Fe₂S₁N₅/SNC surface possessed stronger bond strength, with a more negative integrated COHP (ICOHP) value of −2.17 for the spin-up state and −3.44 for the spin-down state, which was principally due to Fe 3d_yz and O 2p_z, Fe 3d_xy, and O 2p_x, than that of 3O-Fe₂N₆/NC, with ICOHP values of −2.13 and −3.35, respectively. The charge and spin magnetic moment distributions of the active sites also play critical roles in the OER^52,53, as simulated by the Bader charge, differential charge, and spin-charge densities in Supplementary Figs. 79–80. For 3O-Fe₂S₁N₅/SNC, the Fe-1 site reduced much more electrons (+1.52 e⁻) to surrounding coordinate atoms than that of the Fe-2 site (+1.48 e⁻) to generate localized spontaneous electric polarization, yet the Fe-2 site possessed a larger magnetic moment (MM: 3.82 μ_B) than the Fe-1 site (3.60 μ_B), in contrast with the symmetrical 3O-Fe₂N₆/NC model for the Fe-1 and Fe-2 sites featuring the same charge (+1.53 e⁻) and magnetic moment (3.72 μ_B) distributions. Even though the obtained charge (+1.40 e⁻) and MM (2.85 μ_B) of the Fe site on the 2OH-Fe₁S₁N₃/NC catalyst were less than those of the symmetrical 3O-Fe₂N₆/NC, the derived electric polarization induced by the introduction of the S source into the FeN₄ moiety significantly increased the OER performance. Above all, the spontaneous electric polarization intensity endowed by heteroatom S doping and dual-atom sites can synergistically optimize the adsorption capability of OER intermediates, thus promoting the electrochemical oxygen evolution property.

Similarly, for the electrochemical ORR, such as 0.90 V vs. RHE, their surfaces were covered by 2OH species, named 2OH-Fe₁S₁N₃/NC, 2OH-Fe₂N₆/NC, and 2OH-Fe₂S₁N₅/SNC. Supplementary Figs. 73, 81, 82 display the optimized ORR intermediates, and the corresponding adsorption energies are shown in Supplementary Tables 14–16. Supplementary Fig. 83 shows the free energy changes of the ORR mechanism on these surfaces. The asymmetric diatomic 2OH-Fe₂S₁N₅/SNC catalyst had the largest onset potential (0.79 V) among those of 2OH-Fe₂N₆/NC (0.54 V) and 2OH-Fe₁S₁N₃/NC (0.65 V)⁵⁴, which matches well with the experimental results; their PDSs constitute the first hydrogenation step for the generation of *OOH species. The enhanced capture of OOH species was attributed to the large charge and spin magnetic moment population of the active Fe site, specifically for the asymmetric 2OH-Fe₂S₁N₅/SNC catalyst, as shown in Supplementary Fig. 84. Above all, the unsymmetrical geometric construction of the A-Fe₂S₁N₅/SNC catalyst realized spontaneous electric polarization via the coupling of dual-site and heteroatom S doping to achieve disparate OER/ORR performance.

AEMWE device performance of A-Fe₂S₁N₅/SNC

We further assembled an alkaline AEMWE (anion exchange membrane water electrolyzer) catalyzed by A-Fe₂S₁N₅/SNC, as shown in Fig. 5a. The A-Fe₂S₁N₅/SNC-based cell displayed better performance: a current density of 500 mA cm^–2 was reached at 1.94 V, whereas an even higher current density of 1.0 A cm^–2 was reached at 2.18 V. In contrast, the current density of the RuO₂ | |Pt/C cell at 2.22 V was 500 mA cm^–2 (Fig. 5b). Consequently, compared with those catalyzed by precious metal electrocatalysts, the performance of the AEM electrolyzer catalyzed by A-Fe₂S₁N₅/SNC was improved. To provide a comprehensive assessment of the alkaline electrolyzer enabled by A-Fe₂S₁N₅/SNC, as shown in Supplementary Table 17, the performance of the AEM electrolyzer using the A-Fe₂S₁N₅/SNC electrocatalyst exceeded that of most of the reported state-of-the-art catalysts. To characterize the stability of the electrolyzer, when assembled with A-Fe₂S₁N₅/SNC, the AEM electrolyzer could operate at 500 mA cm⁻² for at least 250 h with a negligible increase in voltage (Fig. 5c). Furthermore, the catalytic performances of the ORR (Supplementary Figs. 85–89) and the zinc-air battery (ZAB) were tested to validate the potential application of the bifunctional A-Fe₂S₁N₅/SNC catalyst. As depicted in Supplementary Fig. 90, the performance of the A-Fe₂S₁N₅/SNC-based battery is superior to that of the Pt/C + RuO₂ battery, which indicates its potential in commercial applications for A-Fe₂S₁N₅/SNC-based devices.

Fig. 5: Performance of AEMWE devices.

a Schematic illustration of the assembled AEM electrolyzer (GDL, gas diffusion layer); b polarization curves of the AEMWE electrolyzer using A-Fe₂S₁N₅/SNC and commercial Pt/C; c chronopotentiometry test of the AEMWE utilizing A-Fe₂S₁N₅/SNC held at 500 mA cm⁻². Inset c: Digital image of the AEMWE electrolyzer.

Discussion

In summary, a general and scalable approach for the synthesis of a series of A-DACs (A-M₂S₁N₅/SNC) with high metal loading is achieved. A-Fe₂S₁N₅/SNC exhibits outstanding OER catalytic performance due to the presence of asymmetric FeS₁N₂–FeN₃ moieties, indicating high AEMWE and ZAB potential. For the OER, A-Fe₂S₁N₅/SNC delivers a low overpotential of 193 mV at 10 mA/cm² under alkaline conditions, which is higher than those reported for diatomic catalysts and commercial RuO₂. Furthermore, A-Fe₂S₁N₅/SNC is stable for more than 2000 h at 10 mA cm⁻², which suggests that the durability and catalytic activity can be balanced for the A-Fe₂S₁N₅/SNC electrocatalyst. The AEMWE device with A-Fe₂S₁N₅/SNC catalysts also afforded durability for up to 250 h at 500 mA cm⁻² and exhibited an industrially required current density of 1.0 A cm⁻² at 2.18 V. For A-Fe₂S₁N₅/SNC, high activity comes from the cocatalysis of a high-density active Fe‒Fe site and high stability derives from the regulation of the metal atom microenvironment by S coordination. As corroborated by our operando characterizations and DFT calculations, unique configurations yield the optimal adsorption/desorption energy on Fe‒Fe sites for a radical intermediate with a reduced reaction barrier. We expect that this strategy will contribute to the promotion of the application of diatomic catalysts in the production of clean energy through water electrolysis.

Methods

Chemicals

Melamine (>99%, Aladdin), iron (III) chloride (>99%, Sigma-Aldrich), ruthenium (IV) oxide (99.9%, Aladdin), 2-benzimidazolethiol (>99%, Aladdin), trimesic acid (>98%, Aladdin), dicyandiamide (>99%, Sigma-Aldrich), ethanol (>99.5%, Aladdin), potassium hydroxide (95%, Aladdin), nafion117 solution (~5% in a mixture of lower aliphatic alcohols and water, Aladdin), commercial Pt/C (20 wt% metal, Aladdin), and carbon cloth (CC, type: WOS1009, thickness: 0.33 mm). All chemical reagents were utilized as received. The purity of oxygen and nitrogen gas is 99.999%.

Synthesis of SNC and NC

In a standard synthesis of sulfur and nitrogen co-doped carbon nanosheets, a mixture of 2-benzimidazolethiol and melamine in precise mass ratios (1:7) was initially milled to obtain a homogeneous precursor. The precursor was subsequently subjected to thermal polymerization and carbonization in a flowing nitrogen atmosphere within a horizontally aligned quartz tube, with a heating rate of 2 °C min^‒¹ until reaching 1000 °C, which was maintained for a duration of 2 h. The final synthesized product is designated as SNC, whereas the NC support was prepared using the same methodology as SNC, with the significant exception that the sulfur source (2-benzimidazolethiol) was excluded.

Synthesis of A-Fe₂S₁N₅/SNC

To prepare the A-Fe₂S₁N₅/SNC, the obtained SNC nanosheet support was ultrasonicated in 10 mL of ethanol, and then 40 mg of iron (III) chloride, 1.0 g of dicyandiamide, and 0.1 g of trimesic acid were added to the above suspension and stirred vigorously for 30 min. The suspension was then evaporated to dry the solvent in a rotary evaporator to obtain a gray powder. The gray powder was annealed at 900 °C (heating rate of 2 °C min⁻¹) and held for 3 h under high-purity nitrogen. After naturally cooling to room temperature, black A-Fe₂S₁N₅/SNC nanosheets were obtained.

Synthesis of Fe₂N₆/NC, Fe₁S₁N₃/NC, and Fe₁N₄/NC

To prepare the Fe₂N₆/NC, the obtained NC nanosheet support was ultrasound dispersion in 10 mL of ethanol, then 40 mg iron (III) chloride, 1.0 g dicyandiamide, and 0.1 g trimesic acid were added into the above suspension and stirred vigorously for 30 min. The suspension was then evaporated to dry the solvent in a rotary evaporator to obtain a gray powder. The gray powder was annealed at 900 °C with a heating rate of 2 °C min⁻¹ and maintained for 2 h under the N₂ atmosphere. After naturally cooling to room temperature, the black Fe₂N₆/NC nanosheets were obtained.

To obtain Fe₁N₄/NC, directly add 40 mg of iron (III) chloride, 1.0 g dicyandiamide, and 0.1 g trimesic acid to 15 mL of ethanol and vigorously stir for 30 min. The suspension was then evaporated to dry the solvent in a rotary evaporator to obtain a gray powder. The gray powder was annealed at 900 °C with a heating rate of 2  °C min⁻¹ and maintained for 2 h under the N₂ atmosphere. After naturally cooling to room temperature, the black Fe₁N₄/NC was obtained.

The synthesized NC nanosheet support was subjected to ultrasonic dispersion in 10 mL of ethanol. Subsequently, 40 mg of iron (III) chloride, 0.5 mL of thiophene, 1.0 g of dicyandiamide, and 0.1 g of trimesic acid were incorporated into the suspension, which was then stirred vigorously for 30 min. The solvent was subsequently removed through evaporation using a rotary evaporator, yielding a gray powder. This gray powder was then subjected to annealing at 900 °C, with a heating rate of 2 °C min⁻¹, and maintained at this temperature for 3 h in an atmosphere of high-purity nitrogen. Following natural cooling to ambient temperature, the resultant black Fe₁S₁N₃/NC was obtained.

General synthesis of A-M₂S₁N₅/SNC. (M = Co, Cu, Ni, and Mn)

For the A-M₂S₁N₅/SNC (M = Co, Cu, Ni, and Mn), the preparation process is the same as A-Fe₂S₁N₅/SNC catalyst, except that the use of metal salts is 0.25 mmol cobalt (II) chloride, copper (II) chloride, nickel (II) chloride, and manganese (II) chloride, respectively.

Characterizations

X-ray diffraction (XRD, D8 Bruker AXS, scan range of 10–80°) test was performed to identify the crystallinity and phase of samples. Raman (Lab RAM HR Evolution spectrometer, excitation line of 532 nm) was analyzed to record the stretching and bond vibration modes. The XPS data were acquired using an Escalab 250Xi instrument, employing magnesium and aluminium as the excitation sources from Kratos Analytical. The data were converted to VAMAS file format and subsequently imported into Avantage software for peak fitting analysis. Binding energies reported herein are referenced to the C (1 s) signal at 284.5 eV. The methodology for preparing XPS samples involves the following steps: First, a small quantity of the powder sample is positioned on a piece of aluminum foil. Subsequently, a section of double-sided Scotch tape, ~5 × 5 mm² in area, is cut and held with tweezers to adhere the sample. The tape, along with the sample, is then subjected to gentle pressure using a tablet press until the sample surface is rendered flat and free of contaminants. Finally, the prepared sample is placed on the sample stage for analysis. Scanning electron microscopy (SEM) data were obtained using an Ultra 55 Zeiss instrument to examine the morphology of the samples. Transmission electron microscopy (TEM) and high-resolution transmission electron microscopy (HRTEM) analyses were performed on an FEI Talos F200S at an accelerating voltage of 200 kV. Initially, the thoroughly ground sample was dispersed in an ethanol solution, homogenized through ultrasonic agitation, and subsequently deposited onto a carbon film-coated copper mesh using a pipette. Following the drying process at ambient temperature, specimens for TEM were subsequently prepared. Additionally, annular dark field scanning transmission electron microscopy (AC-HAADF-STEM) utilizing double spherical aberration correctors and electron energy loss spectroscopy (EELS) was carried out at 200 kV. Fourier transform infrared spectroscopy (FTIR) analyses were performed utilizing the Thermo Fisher-Nicolet 6700 instrument. The FTIR spectrum was obtained using a Bruker V70 spectrometer. The N₂ adsorption and desorption isotherm (collected from Quantachrone Autosorb iQ-MP-C) was obtained to derive the surface areas and the pore size distribution of the catalysts. ICP-OES (Thermo Fisher Scientific, IRIS Intrepid II XSP spectrometer) was used to determine the contents of Fe species in the samples.

Ex situ XAFS measurements

The X-ray absorption fine structure spectra data were collected at BL14W1 station in Shanghai Synchrotron Radiation Facility (SSRF, operated at 3.5 GeV with a maximum current of 250 mA), respectively. The data were collected at room temperature (Fe K-edge in fluorescence excitation mode using a Lytle detector). All samples were pelletized as disks of 13 mm diameter with 1 mm thickness using graphite powder as a binder.

In situ synchrotron radiation XAFS measurements

A catalyst-modified carbon paper was used as the working electrode, graphite rod as the counter electrode, and Hg/HgO electrode as the reference electrode. A homemade electrochemical cell was used for in situ XAFS measurements. The experiments were performed at the BL14W1 station in SSRF.

Electrochemical measurements for the OER

The evaluation of the electrocatalytic OER performance of A-Fe₂S₁N₅/SNC was performed with a standard three-electrode system using an electrochemical workstation (CHI 760E) in Ar-saturated 1.0 M KOH. The electrolyte employed in this study is a 1.0 M KOH solution, maintained at a constant temperature of 25 ± 2 °C. This solution was prepared by dissolving 29.5 g of 95% pure KOH in 500 mL of deionized water, while maintaining continuous agitation to guarantee thorough dissolution. The determined practical electrolyte pH value is 13.86 ± 0.06 for a 1.0 M KOH solution. Prior to its application, the pH meter was calibrated with buffer solutions of varying pH levels, specifically potassium hydrogen phthalate solution (pH = 4.00), mixed phosphate solution (pH = 6.86), and sodium tetraborate solution (pH = 9.18). The ink was prepared using 2 mg of A-Fe₂S₁N₅/SNC, 10 μL of Nafion (5 wt%), 100 μL of ethanol, and 190 μL of deionized water to form a homogeneous catalyst ink (ultrasonicated for 2 h). The working electrode was 40 μL of ink deposited on a clean carbon cloth (area: 1 cm²). A graphite rod (counter electrode) and Hg/HgO (reference electrode) were used to perform the electrochemical tests. The catalyst loading amount on the carbon cloth was 0.26 mg cm⁻². Before measurement, the electrode was activated under CV conditions (scan rate 100 mV s⁻¹ for 100 cycles) until stable. Linear sweep voltammetry (LSV) curves of the catalysts were obtained at a constant scanning rate (5 mV s⁻¹) with 95% iR compensation. All potentials were referenced to the reversible hydrogen electrode (RHE), following the below equation: E_RHE = E_Hg/HgO + 0.098 + 0.0591 pH. The solution resistance (Rs) was evaluated using electrochemical impedance spectroscopy (EIS) across a frequency range of 100 kHz to 0.1 Hz, with an applied amplitude of 5 mV. The ECSA was evaluated by the electrochemical double-layer capacitance (C_dl), where the C_dl was obtained from the CV measurements at different scan rates (from 20 to 120 mV s⁻¹) in 1.0 M KOH. The stability of the electrode materials was obtained via the chronoamperometry method. The potentiostat software captured data at one-second intervals. All the measurements were performed once.

Electrochemical measurements for ORR

The oxygen reduction reaction measurements were performed in a three-electrode cell using an electrochemical workstation (CHI 760E) setup with a rotating disk electrode (RRDE-3A). A glassy carbon (GC: 4 mm inner diameter, 0.19625 cm²), platinum wire, and saturated calomel electrode (SCE) were used as the working, counter, and reference electrodes, respectively. The catalyst suspension was prepared by dispersing 2 mg catalyst into 250 μL ethanol/Nafion (5 wt%) mixed solution, followed by ultrasonication for 30 min to form a homogenous ink. Afterward, 10 μL ink was dropped onto the freshly polished GC electrode and naturally dried in air for all the catalysts. The catalyst loading amount on glassy carbon is 0.4 mg cm⁻². Before the ORR test, the electrolyte was purged with O₂ to form the O₂-saturated 0.1 M KOH solution (pH = 12.9 ± 0.12). Cyclic voltammograms were recorded at 50 mV s⁻¹ in O₂-saturated 0.1 M KOH. Linear sweep voltammetry polarization curves for ORR were obtained at a rotating rate of 1600 rpm, without iR compensation at 5 mV s⁻¹. All potentials were referenced to the reversible hydrogen electrode (RHE), according to the reference electrode calibration: E_RHE = E_SCE + 0.241 + 0.0591 pH. Durability tests were performed by cycling between 0.6 and 1.0 V versus RHE for 30,000 cycles at room temperature.

Zinc-air battery measurements

The A-Fe₂S₁N₅/SNC ink was uniformly dispersed onto Teflon-coated carbon fiber paper with the loading of 1.0 mg cm⁻², then dried at 60 °C. For comparison, a Pt/C electrode with the same catalyst loading was also prepared. The polished commercial Zn foil with a thickness of 0.2 mm was used as the anode. Both electrodes were constructed into a homemade Zn–air device in 6.0 M KOH and 0.2 M zinc acetate electrolyte saturated with O₂.

AEM water electrolyzer system assembly

The AEM water electrolyzer was composed of an anion-exchange membrane, electrodes, bipolar plates, current distributors, and end plates. The anode and cathode were assembled using A-Fe₂S₁N₅/SNC and commercial Pt/C, respectively. The anion exchange membrane is Sustainion X37-50 grade RT (50 μm × 2 cm × 2 cm), provided by the Sinero fuel cell store. The active area of the MEA was 4.0 cm² (2 cm × 2 cm). The AEM single cell was operated at room temperature with the injection of 1.0 M KOH into the MEA. The stability of the AEM single cell was tested at a constant current density of 500 mA cm^–2 for 250 h at room temperature.

Details of the DFT calculations

All the simulations were carried out via the spin-polarized density functional theory method executed via the Vienna Ab initio Simulation Package (VASP)⁵⁵ with a projector augmented wave pseudopotential (PAW) to handle the electron‒ion interactions. The PBE is functional with the scheme of the generalized gradient approximation (GGA) was applied to treat electronic exchange‒correlation interactions. DFT + U correction with an effective U value of 4.0 eV was adopted to depict the exchange interaction of the 3D orbitals for the Fe atom, referring to a previous report on diatomic catalysts supported on a nitrogen-doped carbon substrate⁵⁶. To describe the interactions among the intermediates and active sites accurately, the DFT-D3 dispersed corrections proposed by Grimme et al. were involved in evaluating their adsorbed free energies⁵⁷. Considering the practical electrochemical reaction surroundings, the solvation effect was drawn into the adsorbed free energy of intermediates via a polarized continuum model—VASPsol. The electronic kinetic cut-off with the plane wave basis set was 400 eV, and the corresponding convergence criteria of total energy and residual force per atom were less than 10⁻⁴ eV and 0.02 eV/Å, respectively⁵⁸. The k-point sampling mesh in the first Brillouin zone integration was 3 × 3 × 1 for geometric optimization and 5 × 5 × 1 for static simulations⁵⁹.

A symmetric Fe₂N₆/NC model constructed by embedding two Fe atoms into nitrogen-doped graphene with a 6 × 6 × 1 supercell was selected from four possible diatomic catalysts, named Fe₂N₃-6₁, Fe₂N₃-6₂, Fe₂N₄-6, and Fe₂N₄-8, according to a previous report^60,61, as shown in Supplementary Fig. 62. Given the spin-polarized characteristic of Fe, the relative energies between the ferromagnetic (FM) and anti-ferromagnetic (AFM) state for each model and the formation energies were systematically investigated via the PBE functional with and without DFT + U correction, as shown in Supplementary Fig. 63. Note that some AFM states with the initial setting were converted to a ferrimagnetic state after relaxation. Their ground states are AFM states, except for the Fe₂N₃-6₂ model with the FM ground state by the pure PBE method. The formation energies were calculated via the following formula: E_f = E_total - aE_C -b E_Fe - cE_N - dE_S, where E_total, E_C, E_Fe, E_N, and E_S are the total energies of the A-Fe₂S₁N₅/SNC and Fe₂N₆/NC models; the chemical potentials of C, Fe, N, and S refer to graphene, the metal iron bulk, the N₂ molecule, and the S₈ molecule, respectively, corresponding to the numbers marked by the letters a, b, c, and d. The Fe₂N₃-6₂ model has the lowest formation energy, with values of −7.85 and −11.79 eV according to the PBE functional with and without DFT + U correction. Therefore, the stable Fe₂N₃-6₂ model served as the as-fabricated Fe₂N₆/NC diatomic catalyst to investigate the following OER/ORR mechanism.

Thus, the fabricated A-Fe₂S₁N₅/SNC diatomic catalyst was constructed based on the Fe₂N₃-6₂ model considering eight kinds of Fe₂S₁N₅/SNC configurations in Supplementary Fig. 64, named “1–8”. In the Fe₂N₃-6₂ model, the “1–3” models replace one nitrogen site using one sulfur atom with different nitrogen sites. One sulfur atom substitutes one carbon and one nitrogen site with different carbon‒nitrogen sites to generate models “4–6”. The “7–8” models were similar to the “4–6” configurations, with fewer nitrogen atoms. The relative energies and formation energies of the PBE functional with and without DFT + U correction are displayed in Supplementary Fig. 65. Using the DFT + U method, the AFM states were their ground states, and the opposite was true for the pure PBE method, except for model “8” with the FM ground state, which had a negligible energy difference of 0.02 eV. Model “1” has the lowest formation energy, with values of −6.01 and −10.17 eV obtained via the PBE functional with and without DFT + U correction, respectively. The bond lengths of the Fe‒Fe, Fe‒S, and Fe‒N pairs are 2.36, 2.29, and 1.96 Å, respectively, using the DFT + U method, which matches well with the EXAFS fitting results. Therefore, model “1” acted as an A-Fe₂S₁N₅/SNC diatomic catalyst for the following OER/ORR research, unlike the Fe₂N₆/NC model.

The density of state, charge, and spin population was further calculated via the VASPKIT⁶² tool to treat these postprocessing results. In the electrochemical reaction, the energy of H⁺ + e⁻ was referred to as the energy of 1/2 H₂ according to the computational hydrogen model (CHE) proposed by Nørskov et al. ⁶³. To accurately depict the real configuration of the active site in the electrochemical surroundings, the Pourbaix diagram was calculated via the following equation:

$$ \ast+{{\rm{m}}}{{{\rm{H}}}}_{2}{{\rm{O}}}\to*{{{\rm{O}}}}_{{{\rm{m}}}}{{{\rm{H}}}}_{{{\rm{n}}}}+\left(2{{\rm{m}}}-{{\rm{n}}}\right)\left({{{\rm{H}}}}^{+}+{{{\rm{e}}}}^{-}\right)$$

$$\Delta {{\rm{G}}}\left({{\rm{U}}},{{\rm{pH}}}\right)={{{\rm{G}}} * }_{{{\rm{OmHn}}}}+\left(2{{\rm{m}}}-{{\rm{n}}}\right){{{\rm{G}}}}_{{{\rm{H}}}++ {{\rm{e}}}-}-{{\rm{G}}} * -{{\rm{m}}}{{{\rm{G}}}}_{{{\rm{H}}}2{{\rm{O}}}}$$

$$={{{\rm{G}}} * }_{{{\rm{OmHn}}}}+\left(2{{\rm{m}}}-{{\rm{n}}}\right)\left(1/2{{{\rm{G}}}}_{{{\rm{H}}}2}{{\rm{\hbox{-}}}}{{\rm{eU}}}+{{\rm{k}}}{{\rm{B}}}{{\rm{T}}}{\mathrm{ln}}[{{\rm{c}}}{{\rm{H}}}+]\right)-{{\rm{G}}} * -{{\rm{m}}}{{{\rm{G}}}}_{{{\rm{H}}}2{{\rm{O}}}}$$

$$={{{\rm{G}}} * }_{{{\rm{OmHn}}}}+\left(2{{\rm{m}}}-{{\rm{n}}}\right)\left(1/2{{{\rm{G}}}}_{{{\rm{H}}}2}-{{\rm{eU}}}-0.059{{\rm{pH}}}\right)-{{\rm{G}}} * -{{\rm{m}}}{{{\rm{G}}}}_{{{\rm{H}}}2{{\rm{O}}}}$$

where G = E + ZPE + ΔH(0→298.15 K) − TS. E is the DFT energy, which is the free energy correction, and for the adsorbates, it can be calculated via the vibrational frequency based on the harmonic approximation; however, for free gas molecules, the standard free energy correction was taken from the computational chemistry database (https://cccbdb.nist.gov/)⁶⁴. The letters m and n are the numbers of oxygen and hydrogen atoms in the adsorbate, respectively. The “*” refers to an adsorption site. The electrochemical OER and ORR mechanisms were comprehensively researched via all the possible intermediates in the proton-coupling electron-transfer (PCET) processes according to our previous work^65,66.

For the electrochemical OER process, the possible elementary steps are as follows:

$${{{\rm{H}}}}_{2}{{\rm{O}}}+\ast \to*{{\rm{OH}}}+{{{\rm{H}}}}^{+}+{{{\rm{e}}}}^{-}$$

(1)

$$*{{\rm{OH}}}\to*{{\rm{O}}}+{{{\rm{H}}}}^{+}+{{{\rm{e}}}}^{-}$$

(2)

$$*{{\rm{OH}}}+{{{\rm{H}}}}_{2}{{\rm{O}}}+\ast \to 2*{{\rm{OH}}}+{{{\rm{H}}}}^{+}+{{{\rm{e}}}}^{-}$$

(3)

$$*{{\rm{O}}}+{{{\rm{H}}}}_{2}{{\rm{O}}}+\ast \to*{{\rm{O}}}{{\rm{OH}}}+{{{\rm{H}}}}^{+}+{{{\rm{e}}}}^{-}$$

(4)

$$2*{{\rm{OH}}}+\ast {{\rm{O}}}+\ast {{\rm{OH}}}+{{{\rm{H}}}}^{+}+{{{\rm{e}}}}^{-}$$

(5)

$$*{{\rm{O}}}+{{{\rm{H}}}}_{2}{{\rm{O}}}+\ast \to*{{\rm{O}}}+\ast {{\rm{OH}}}+{{{\rm{H}}}}^{+}+{{{\rm{e}}}}^{-}$$

(6)

$$*{{\rm{OOH}}}\to*{{{\rm{O}}}}_{2}+{{{\rm{H}}}}^{+}+{{{\rm{e}}}}^{-}$$

(7)

$$*{{\rm{O}}}+\ast {{\rm{OH}}}\to 2*{{\rm{O}}}+{{{\rm{H}}}}^{+}+{{{\rm{e}}}}^{-}$$

(8)

$$*{{{\rm{O}}}}_{2}\to \ast+{{{\rm{O}}}}_{2}({{\rm{g}}})$$

(9)

$$2*{{\rm{O}}}\to \ast+{{{\rm{O}}}}_{2}({{\rm{g}}})$$

(10)

For the electrochemical ORR process, the possible elementary steps are as follows:

$$ \ast+{{{\rm{O}}}}_{2}+{{{\rm{H}}}}^{+}+{{{\rm{e}}}}^{-}\to*{{\rm{OOH}}}$$

(11)

$$*{{\rm{OOH}}}+{{{\rm{H}}}}^{+}+{{{\rm{e}}}}^{-}\to*{{\rm{O}}}+{{{\rm{H}}}}_{2}{{\rm{O}}}$$

(12)

$$*{{\rm{O}}}+{{{\rm{H}}}}^{+}+{{{\rm{e}}}}^{-}\to*{{\rm{OH}}}$$

(13)

$$*{{\rm{OH}}}+{{{\rm{H}}}}^{+}+{{{\rm{e}}}}^{-}\to \ast+{{{\rm{H}}}}_{2}{{\rm{O}}}$$

(14)

For the OER process, the potential-determined step (PDS) is the primary step with the maximum reaction energy value, where G_PDS = max {G_i}, I = 1, 2, 3, 4. The overpotential of the OER (η_OER) is (G_PDS/e – 1.23) V. For the ORR process, the PDS is the elementary step with the maximum reaction energy value, where G_PDS = max {G_i}, i = 1, 2, 3, 4. η_OER = [1.23 – (–G_PDS/e)]/e = 1.23 + G_PDS/e.