Stabilizing ruthenium atom layer neighbors through molecular and metal support interactions for water electrolyzers

Thermodynamic stabilization through molecular and metal–support interactions

The alkaline HER is critical for both industrial applications and as a fundamental model for studying electrochemical reaction mechanisms. The alkaline HER requires two types of active sites in close proximity: one optimized for *H binding and the other for *OH binding⁴⁸. This dual-site requirement makes it an ideal platform for evaluating the design characteristics of noble metal ALNs in electrocatalysis (Fig. 1a). Among noble metals, Ru is particularly well-suited for AEMWEs due to its adequate *H binding energy and high cost efficiency (~40 % of Pt; Supplementary Table 1). Transition metal carbides serve as effective functional supports for efficient alkaline HER. First, their favorable binding of oxygenated species provides sufficient *OH, a key adsorbate in various catalytic mechanisms^13,49. Second, their strong affinity for noble metal species stabilizes the atomic structure of noble metal during both synthesis and catalytic processes^13,36,50. Furthermore, their high electrical conductivity and chemical stability enhance the efficiency of electrocatalysis⁴⁹.

Interaction mediator-assisted evaporation-induced self-assembly (EISA) using the BCP of poly(ethylene oxide)-block-poly(styrene) (PEO-b-PS) was employed for the synthesis of Ru ALNs on WC_x³². During the EISA, metal precursors were selectively incorporated into specific BCP microphases based on their hydrophilicity. Since a thorough understanding of MIs within EISA systems was essential to achieve atomic dispersion of noble metal precursors, molecular dynamics (MD) simulations were conducted for both hydrophobic Ru precursor (HPB Ru, Bis(2-methylallyl)(1,5-cyclooctadiene)ruthenium(II)) and hydrophilic Ru precursor (HPL Ru, Ruthenium(III) chloride hydrate) to predict the spatial positioning and dispersity of Ru precursors during EISA. The polymer solvation energy (ΔE_solv) of each precursor was calculated to determine its location within the BCP. The calculations revealed that HPL Ru preferentially localizes within PEO microphases due to its favorable ΔE_solv for PEO (Supplementary Fig. 2, Supplementary Table 2). Conversely, HPB Ru localizes within PS microphases due to its favorable ΔE_solv for PS (Fig. 2a, Supplementary Figs. 2, Supplementary Table 2), consistent with the like-dissolves-like principle. The dispersion propensity of Ru precursors (ΔE_disp) in the polymer matrix was calculated as the difference between ΔE_solv and the self-solvation energy (Supplementary Fig. 3, Supplementary Table 3). HPL Ru exhibited unfavorable ΔE_disp in the PEO matrix due to its strong intermolecular interactions, which can be attributed to its ionic nature. This unfavorable ΔE_disp predicts its aggregation within the PEO microphase. Conversely, HPB Ru showed favorable ΔE_disp in the PS matrix, where polymer-precursor interactions exceeded intermolecular interactions, enabling atomic dispersion of HPB Ru in the PS microphase (Fig. 2a). The predicted precursor-polymer interactions were further supported by thermogravimetry analysis with differential scanning calorimetry (TGA-DSC) measurements, which revealed distinct thermal behaviors for the precursor/polymer composites compared with the corresponding individual components (Supplementary Note 2).

To predict the atomic structure of Ru ensembles on WC (Ru_N/WC, N: the number of Ru) after high-temperature T (high-T) annealing at 1173 K, using the number of Ru atoms (N_Ru), MD simulations were performed at 1200 K. For Ru₁/WC, the high surface energy of Ru SAC caused vigorous thermal migration onto the WC support (Supplementary Fig. 6a, Supplementary Movie 1a). From Ru₂/WC to Ru₁₆₃/WC, two-dimensional Ru ALNs (one-atom-thick layer) were thermodynamically stabilized by SMSI with WC. In this regime, the proximity of neighboring Ru significantly reduced the mobility of Ru ALNs (Fig. 2b, Supplementary Fig. 6b-j, Supplementary Movies 1–4, 5a). However, when N_Ru exceeded a critical threshold (N_Ru ≥ 235), the WC support could no longer stabilize Ru agglomerates as ALNs, resulting in the formation of three-dimensional Ru NPs (Fig. 2b, c, Supplementary Fig. 6k-m, Supplementary Movie 5b-d). By plotting the logarithm of N_Ru against the Ru-Ru coordination number (CN), we observed a transition in atomic structure from ALNs to sub-monolayer configurations and ultimately to NPs as N_Ru increased (Fig. 2c, Supplementary Fig. 7, Supplementary Table 6). These results indicated that only Ru_N/WC below a critical size (N_Ru < 235) can form stable Ru ALNs on WC_x.

To investigate the effects of BCP-precursor systems and SMSI with WC_x, aRuN/WC_x (samples prepared via self-assembly using HPB Ru; N denotes the relative amount of Ru precursor) and aRu5_p/WC_x (sample prepared via self-assembly using HPL Ru with a Ru content similar to that of aRu5/WC_x) were synthesized. For comparison, Ru on WO_x (aRu5/WO_x or aRu5_p/WO_x) was obtained by annealing at a lower T of 873 K to isolate the SMSI effect between Ru and WC_x support. Additional comparison samples were prepared using incipient wetness impregnation (IWI) with HPB Ru (Ru/WC_x) and HPL Ru (Ru_p/WC_x), followed by low-T reduction at 473 K under 10 v/v% H₂/Ar (_H₂) and high-T annealing at 1,173 K under Ar (_Ar). Particulate Ru samples on supports were also synthesized by chemical reduction (_R) (see Catalyst synthesis in Methods). Theoretical predictions indicated that the spatial positioning and dispersity of Ru precursors were governed by MIs within EISA systems. HPB Ru exhibited atomic dispersion in PS microphases, while HPL Ru formed aggregates in PEO microphases (Fig. 1b). During thermal treatment, HPB Ru reduced to sub-nanoclusters, which subsequently stabilized as ALNs via SMSI with WC_x due to their sufficiently small N_Ru (below the critical threshold). Furthermore, the limited mobility of Ru ensembles on WC_x allowed for systematic tuning of ensemble size by adjusting the Ru precursor amount (Supplementary Notes 1–3). In contrast, large agglomerates of HPL Ru formed large Ru NPs, unable to stabilize as ALNs due to their large N_Ru.

To clarify the effect of MIs, Ru content in comparison samples was set to ~10 wt%, matching the Ru content in aRu5/WC_x (Supplementary Table 7). Samples synthesized via self-assembly exhibited mesoporous structures with three-dimensional pore networks and large mesopores, as confirmed by N₂ physisorption, small-angle X-ray scattering (SAXS), and scanning electron microscopy (SEM) (Supplementary Figs. 8–20). The relative volume fractions of hydrophilic and hydrophobic microphases are critical determinants of mesoporous morphology, indicating that porosity depends not only on the precursor amount but also on its hydrophilicity²². Consequently, even with the same molar amounts of Ru precursor, samples prepared with HPB Ru exhibited distinct porosity compared to those prepared with HPL Ru (Supplementary Fig. 20). Consistent with previous reports, these results suggest that the chemical affinity of the Ru precursors governs their preferential interactions with specific BCP microphases during the EISA process, thereby influencing mesostructural correlations, structural ordering and porosity evolution (Supplementary Note 5)^51,52,53.

X-ray diffraction (XRD) analysis revealed the presence of crystalline WC_x in all samples, except for aRu5/WO_x and aRu5_p/WO_x (Fig. 2d, e, Supplementary Fig. 21, Supplementary Note 4). Notably, no crystalline Ru peaks were observed in either aRu5/WO_x or aRu5/WC_x, whereas aRu5_p/WO_x and aRu5_p/WC_x exhibited clear crystalline Ru peaks. This suggested that the MI of [PS-HPB Ru] effectively mitigated the MI of [HPB Ru], thereby facilitating the stabilization of Ru species as Ru ALNs (Supplementary Note 4). High-angle annular dark-field scanning transmission electron microscopy (HAADF-STEM) with energy dispersive spectroscopy (EDS) mapping of aRu5/WO_x showed uniform Ru distribution without large agglomerates (Fig. 2f, Supplementary Fig. 22), while aRu5_p/WO_x exhibited uneven Ru dispersion with large agglomerates (Fig. 2h, Supplementary Fig. 24). These results showed that selecting an appropriate BCP-precursor system was crucial. This choice was particularly important for achieving uniform atomic dispersion of Ru species, and it required a detailed understanding of MIs within EISA.

STEM-EDS of aRu5/WC_x revealed the uniform and selective dispersion of Ru ALNs on WC_x supports, showing that the SMSI with WC_x effectively prevented thermal sintering of Ru into NPs by maintaining Ru as ALNs (Fig. 2g, Supplementary Fig. 23). In comparison, aRu5_p/WC_x showed large Ru agglomerates, indicating that the SMSI with WC_x was insufficient to stabilize large Ru agglomerates (exceeding the critical size threshold) as ALNs (Fig. 2i, Supplementary Fig. 25). Likewise, aRuN/WC_x, synthesized via thermodynamic stabilization, exhibited the same features as aRu5/WC_x in STEM-EDS, demonstrating the versatility of thermodynamic stabilization strategy for a wide range of Ru contents (~10 wt%) (Supplementary Figs. 26–29). Furthermore, the Ru atomic ratio Ru/(Ru+W) obtained from EDS analysis was in close agreement with the X-ray fluorescence (XRF) results, indicating that Ru is uniformly distributed throughout the sample via MIs, leading to consistent values between bulk (XRF) and local (EDS) analyses (Supplementary Figs. 22–30, Supplementary Table 7). In contrast, XRD and STEM-EDS of IWI-prepared samples revealed that IWI alone was insufficient to stabilize Ru ALNs, as it lacked mechanisms to mitigate MI of [Ru precursor], resulting in the formation of large agglomerates before annealing (Supplementary Figs. 19, 29–32, Supplementary Notes 4, 6).

Z-contrast imaging using high-resolution HAADF-STEM is widely applied to elucidate the atomic structure of noble metal on supports, as the brightness scales with both the atomic number (Z) and the number of atoms overlapped along the electron-beam direction^54,55. However, in the case of Ru ALNs on WC_x, tungsten (Z = 74) possesses a much higher Z than ruthenium (Z = 44), resulting in stronger intensity and making it difficult to distinguish Ru from the underlying W in thick WC_x supports (>~2 nm). The simulated Z-contrast images for a single Ru atom and a single W defect on γ-WC ( ~ 0.6 nm thickness) exhibited similarly dark features (Supplementary Fig. 37a, b), indicating that such contrast cannot discriminate whether the dark spot originates from Ru or from a W vacancy. Furthermore, simulated Z-contrast images of Ru_N (N = 1-6) on γ-WC (~2 nm thickness) show nearly indistinguishable intensity profiles (Supplementary Fig. 38), suggesting that HAADF-STEM Z-contrast imaging is not suitable for resolving Ru ALN configurations on WC_x supports.

Consistent with these simulations, experimental Z-contrast images of both aRu0/WC_x (without Ru) and aRu3/WC_x exhibited similar distributions of bright (red circles) and dark spots (Supplementary Fig. 37c, d), confirming the intrinsic difficulty in distinguishing atomically dispersed Ru from local contrast variations within the WC_x lattice. In contrast, when Ru forms three-dimensional nanoclusters exceeding a critical size, the simulated image of Ru₂₃₅/WC reveals distinguishable Ru features on the γ-WC lattice (Supplementary Fig. 39a) due to the increased Ru thickness and cumulative scattering along the beam path⁵⁵. Similarly, experimental Z-contrast images of Ru/WC_x_R (Ru nanoclusters on WC_x supports, see Catalyst Synthesis in Methods and Supplementary Note 6) resolve sub-nanometer Ru aggregates (red circles, <1 nm) on WC_x supports (Supplementary Fig. 39b). These results collectively indicate that HAADF-STEM Z-contrast imaging can reliably identify three-dimensional Ru nanoclusters, but cannot distinguish two-dimensional Ru ALN configurations on heavy WC_x supports.

According to the Z-contrast image of aRu3/WC_x, no three-dimensional sub-nanometer Ru aggregates were verified (Supplementary Fig. 37c). Electron energy-loss spectroscopy (EELS) line scans using atomic-resolution STEM across the Ru L-edge revealed a spatially compact grouping of Ru signals, typically consisting of two to three adjacent atomic-scale peaks, together with a cumulative increase in the integrated areal density along the scan path (Supplementary Fig. 40). This spatially correlated Ru signal evolution supports the presence of Ru ensemble sites on the WC_x supports rather than isolated single atoms or three-dimensional aggregates.

Geometric and electronic structure of Ru atom-layer neighbors

To investigate the electronic and geometric structures of Ru ALNs, X-ray photoelectron spectroscopy (XPS) and X-ray absorption spectroscopy (XAS) were performed. During the transition from aRu5/WO_x to aRu5/WC_x, XPS showed a negative shift in Ru 3 d peaks (Supplementary Fig. 41a). Likewise, X-ray absorption near-edge structure (XANES) analysis at the Ru K-edge revealed a negative shift in the edge position for aRu5/WC_x compared to aRu5/WO_x (Fig. 3a). The edges of aRuN/WC_x aligned closely to that of metallic Ru (Supplementary Fig. 58a). These indicated that the SMSI with WC_x imparted metallic properties to Ru ALNs, which is distinct from those of comparison samples and previous findings of Ru SACs on various supports (Supplementary Figs. 41–58, Supplementary Note 7) ^39,56,57.

The Fourier transforms of extended X-ray absorption fine structure (FT-EXAFS) with corresponding fitting results reveal that aRuN/WC_x was coordinated by ~4 Ru-C bonds (~0.1 Å longer than Ru-O) and ~2 Ru-W bonds, indicating that Ru ALNs were selectively stabilized on WC_x supports (Fig. 3b, c, Supplementary Figs. 59, 67, Supplementary Tables 8, 9). In contrast, aRu5_p/WC_x and IWI-prepared samples exhibited ~2 Ru-O coordination in the first shell, which originated from surface oxygen bonds on Ru agglomerates⁵⁸. Furthermore, the CN of Ru-Ru in aRuN/WC_x gradually increased with higher Ru contents, suggesting that the Ru precursor amount systematically controlled the ensemble size of Ru ALNs (Fig. 3c, Supplementary Fig. 67, Supplementary Table 9). The Ru-Ru CN for aRuN/WC_x ranged between 1.5 and 2.5, with distinct bond lengths of Ru-Ru (~2.74 Å) and Ru-W (~2.84 Å), which differed from those for Ru metal and Ru agglomerates in comparison samples (Fig. 3c, Supplementary Tables 8–11). The low Ru-Ru CN suggests that each Ru ALN consisted of six or fewer Ru atoms, and the distinct bond lengths likely result from two-dimensional atomic structures of Ru ALNs via SMSI with WC_x supports (Supplementary Table 6, Supplementary Note 8) ⁵⁹. The FT-EXAFS with corresponding fitting for aRu5_p/WC_x and IWI-prepared samples revealed the presence of large Ru or Ru-W alloy agglomerates (Fig. 3b, c, Supplementary Tables 8–10, Supplementary Note 8).

Because HAADF-STEM analyses provide highly local structural information, they cannot alone verify whether the Ru ALN configuration represents the dominant structure across the catalyst. To address this, we compared the structural parameters predicted by theory with those obtained from bulk-averaged EXAFS fitting. Because the DFT models represent idealized local structures and do not fully capture the heterogeneity and dynamic electrochemical environment of the catalyst under operating conditions, the calculations are used mainly to support trends and guide the interpretation of the experimental results. Density functional theory (DFT) calculations suggested that two-dimensional Ru_N clusters are thermodynamically more favorable than their three-dimensional counterparts (Supplementary Figs. 68, 69). This trend is consistent with our MD simulations, in which three-dimensional Ru_N clusters were spontaneously stabilized into ALN structure below a critical size (N_Ru < 235) (Supplementary Fig. 6, Supplementary Movies 1–5). The theoretically predicted Ru-Ru bond lengths of Ru_N/WC_x (Supplementary Fig. 70) are consistent with those obtained from EXAFS fitting for aRuN/WC_x (~2.74 Å), but not with those of NP-based samples (~2.67 Å; Ru_p/WC_x_H₂, Ru/WC_x_H₂, Ru/WC_x_R, Ru/C_R). This characteristic elongation of the Ru-Ru distance is a distinctive signature of the ALN geometry. Furthermore, the experimentally observed Ru-Ru bond length elongation is inconsistent with both residual RuO_x species and incompletely dispersed Ru clusters and is more consistent with Ru ALN structures stabilized through simultaneous Ru-C and Ru-W coordination on WC_x (Supplementary Note 9).

Furthermore, the ensemble-size-dependent variation in the Ru-W CN obtained theoretically using a cutoff of <2.90 Å closely follows the experimental trend observed for aRuN/WC_x, whereas the trend calculated using a cutoff of <2.70 Å does not (Supplementary Fig. 80a, b). Consistently, the Ru-W bond length extracted from EXAFS for aRuN/WC_x (Supplementary Fig. 80e) differs from that of alloy-type Ru-W NP samples (~2.72 Å; aRu5_p/WC_x, Ru5_p/WC_x_Ar, Ru5/WC_x_Ar), reflecting the distinct coordination environment imposed by the two-dimensional Ru ALN structure (Supplementary Note 8). The consistency between the theoretical trends and experimentally obtained structural parameters supports the assignment of Ru ALNs as a predominant structural motif in aRuN/WC_x.

Taken together, the Ru ALN model is consistent with the experimentally observed structural fingerprints, including the characteristic Ru-Ru bond-length elongation, the distinctive Ru-W coordination environment, direct Ru-C anchoring interactions, and their ensemble size-dependent evolution. The consistency between theory and experiment across these independent structural descriptors supports that Ru species in aRuN/WC_x are predominantly stabilized as Ru atom-layer neighbors on WC_x supports (Supplementary Note 8).

In wavelet transforms of EXAFS (WT-EXAFS), aRuN/WC_x exhibited an intensity maximum at lower k compared to RuO₂ and Ru agglomerates in comparison samples, within the k range of 3 to 6 Å⁻¹ (Fig. 3d, Supplementary Figs. 84, 85). This lower k suggests that coordinating atoms in the first shell of aRuN/WC_x are lighter than oxygen, potentially indicating the presence of carbon (Supplementary Note 8)⁶⁰. To further distinguish the contributions of Ru-Ru and Ru-W in the second shell of aRuN/WC_x, a detailed WT-EXAFS analysis that offers higher resolution in k-space at the expense of R-space resolution was conducted (Fig. 3e, Supplementary Note 8). The lower k region (red square) corresponds to Ru-Ru, while the higher k region (blue square) corresponds to Ru-W. The relative intensity of Ru-Ru compared to Ru-W increased as Ru contents increased, which reflects the systematic tuning of ensemble size by adjusting the Ru contents. In conclusion, the thermodynamic stabilization strategy successfully stabilized Ru ALNs with distinct atomic structures via MIs and SMSI with WC_x, enabling systematic control over the ensemble size of Ru ALNs.

Origin of the enhanced kinetics for Ru atom-layer neighbors

To evaluate the Gibbs free energy of hydrogen binding (ΔG_H) based on the geometric structure of noble metal species, DFT calculations were performed. The ideal ΔG_H value for efficient HER is ~0 eV. Consistent with previous studies, hollow sites of Pt (111) and Ru (0001) provided the most stable *H adsorption, with ΔG_H for Ru (0001) calculated to be −0.269 eV (Fig. 4a, b, Supplementary Figs. 88, 90a, Supplementary Table 15). In contrast, isolated single-atom Ru on WC support (Ru₁/WC) exhibited only top-site adsorption with ΔG_H of 0.320 eV. Ru ALNs on WC support enabled bridge-site adsorption of *H when neighboring Ru atoms were present. Notably, despite identical bridge-site adsorption, the ΔG_H of Ru ALNs varied with the number of Ru atoms in the Ru ALN, which can be attributed to the ligand effect from neighboring Ru atoms. In particular, Ru₃/WC and Ru₄/WC exhibited near ideal ΔG_H values of −0.239 eV and −0.211 eV, respectively. Furthermore, WC_x supports showed favorable Gibbs free energy for oxygenated species, such as *OH and *H₂O, facilitating the enrichment of *OH species near *H binding sites (Supplementary Figs. 89, 90b, Supplementary Table 16).

Under the alkaline conditions (H₂-saturated 1 M KOH), the HER activity of aRuN/WC_x showed a volcano-type dependence on the ensemble size of Ru ALNs, with aRu3/WC_x achieving the lowest overpotential for −10 mA cm⁻² (η₁₀) at 22 mV (potentials referenced to the reversible hydrogen electrode, RHE) and Tafel slope of 41.9 mV dec⁻¹ (Fig. 4c, d, Supplementary Figs. 91, 92, Supplementary Tables 17, 18). Similarly, the trend in kinetic activity remained consistent at an identical Ru loading of 0.042 mg cm⁻² (Supplementary Fig. 93). Furthermore, aRu3/WC_x (−1202.0 mA mg_mass⁻¹), showed a higher current density at an overpotential of 50 mV per noble metal mass (j_50/mass) than commercial Pt/C, which displayed η₁₀ of 42 mV, a Tafel slope of 81.9 mV dec⁻¹, and j_50/mass of −312.0 mA mg_mass⁻¹ at 50 mV under identical noble metal loading (0.042 mg cm⁻²). DFT calculations (Fig. 4a) suggest that the enhanced activity of aRu3/WC_x within the aRuN/WC_x series is associated with the ensemble-size-dependent modulation of *H binding, consistent with Ru₃/WC_x or Ru₄/WC_x-like local structures being major contributors to the observed activity (Supplementary Note 10). By contrast, Ru NPs and agglomerates in comparison samples exhibited lower performance due to unfavorable ΔG_H values or lower noble metal utilization efficiency of particulate Ru (Fig. 4c, d, Supplementary Figs. 91, 92, Supplementary Table 17). During the long-term chronoamperometry at an overpotential of 50 mV, aRu3/WC_x showed a smaller decrease in activity than the comparison catalysts tested under the same conditions, showing only about half the current density loss of commercial Pt/C and other comparison catalysts (Fig. 4e, Supplementary Fig. 95). To minimize the mechanical and hydrodynamic effects induced by electrode rotation, the durability test was conducted using aRu3/WC_x coated onto carbon paper, which is a typical substrate for durability assessment^39,61. During the 100 h chronopotentiometry, aRu3/WC_x exhibited only minor voltage losses of ΔV ≈ 0.009 V at −200 mA cm⁻² and ΔV ≈ 0.050 V at −500 mA cm⁻² without any noticeable initial current decrease, owing to the SMSI between Ru ALNs and WC_x supports (Fig. 4f). The post-durability analyses confirmed that aRu3/WC_x preserved its mesoporous morphology and atomic-scale Ru dispersion after the long-term durability test, without forming the large Ru agglomerates (Supplementary Figs. 96, 97).

To examine the general applicability of the present ALN stabilization strategy, aPt3/WC_x and aIr3/WC_x were synthesized using hydrophobic Pt and Ir precursors, dimethyl(1,5-cyclooctadiene)platinum(II) (HPB Pt) and ((1,5-cyclooctadiene)-η⁵-indenyl)iridium(I) (HPB Ir), respectively, under identical noble metal molar loading conditions to those used for aRu3/WC_x (Supplementary Note 11). Structural analyses revealed that Ir could be stabilized as ALN-like structures on WC_x through the same stabilization mechanism, whereas Pt formed partially aggregated nanoclusters anchored on WC_x (Supplementary Figs. 98–105, Supplementary Table 19). These different stabilization behaviors suggest that thermodynamically favorable noble metal-WC_x interactions alone are insufficient for well-defined ALN formation, and that precursor-directed MIs during EISA are critical for suppressing premature aggregation and enabling surface confined ALN structures. The alkaline HER activities of the resulting catalysts were additionally evaluated at identical noble-metal loadings (42 μg cm⁻²). aPt3/WC_x exhibited an η₁₀ of 26 mV and a Tafel slope of 25.8 mV dec⁻¹, comparable to those of aRu3/WC_x, whereas aIr3/WC_x showed relatively lower HER activity with η₁₀ of 51 mV and a Tafel slope of 97.2 mV dec⁻¹ (Supplementary Fig. 106).

To reveal the origin of the enhanced kinetic activity of aRu3/WC_x, in situ XAS analyses were conducted under alkaline HER conditions (Fig. 5a-d). To monitor the potential-dependent structural evolution, in situ XANES and FT-EXAFS spectra were collected at both the W L₃-edge and Ru K-edge under a series of stepwise applied potentials, from open-circuit potential (OCP) to −50 mV (vs. RHE) with 10 mV intervals. At the W L₃-edge, the white line intensity of aRu3/WC_x increased when an overpotential was applied compared to that at the OCP (Fig. 5a), indicating an increase in the oxidation state of W. Likewise, the intensity of the first-shell peak in the corresponding FT-EXAFS spectra also increased under HER operation (Fig. 5b), suggesting a rise in the W-O/C CN. These results imply that the surface of WC_x supports becomes enriched with oxygenated species (e.g., *OH or *O) during alkaline HER, consistent with previous reports^62,63. In contrast, at the Ru K-edge, the edge position of aRu3/WC_x was negatively shifted relative to that at the OCP (Fig. 5c), indicating a partial reduction of Ru species due to *H adsorption on the Ru active sites⁶⁴. Moreover, no noticeable change was observed in the FT-EXAFS spectra at the Ru K-edge (Fig. 5d), suggesting that the Ru ALN structure remained intact during HER, maintaining its ALN configuration throughout the reaction.

To identify the surface species formed on the WC_x supports, in situ Raman spectroscopy was performed at OCP and under sequentially applied potentials ranging from 0.100 V to −0.300 V (vs. RHE) with 0.050 V intervals (Fig. 5e, f). The intensity of the Raman band centered at ~1060 cm⁻¹ increased as the potential decreased from 0.100 V to 0.000 V relative to that at the OCP, and then remained nearly constant upon further decrease to −0.300 V. According to previous reports, Raman bands observed around ~1060 cm⁻¹ are typically assigned to coupled O = W···W-OH vibrations in tungsten-based compounds⁶⁵. Combining the in situ Raman and in situ XAS results, it can be concluded that the surface of the WC_x supports becomes enriched with oxygenated species (e.g., *OH and *O) under alkaline HER conditions, consistent with the potential-dependent oxidation observed in the W L₃-edge XAS. Such surface oxygenation behavior is consistent with previous reports that transition metal carbides possess inherently high oxygen affinity, leading to spontaneous adsorption of oxygenated species in the presence of water during the catalytic reactions^49,66.

Previous research has identified that the depletion of water in the gap region, namely the gap water, causes sluggish alkaline HER kinetics by reducing the connectivity within the H-bond networks⁴⁶. To investigate variations in the interfacial water structure across support materials, in situ attenuated total reflection surface-enhanced infrared absorption spectroscopy (ATR-SEIRAS) was employed. Peaks at 3600, 3400, and 3180 cm⁻¹, assigned to the O-H stretching mode (ν_O-H) of water, can be attributed to weakly hydrogen-bonded, liquid-like, and ice-like water, respectively⁶⁷. According to the electric double layer-based model, the 3600 cm⁻¹ band corresponds to water molecules located in the interfacial gap region with partially disrupted hydrogen bonding, which act as bridging species linking the nearest interfacial water species and the bulk electrolyte^46,68. Therefore, an increased population of gap water facilitates proton transfer between bulk water and inner Helmholtz layer through an improved H-bond network. During alkaline HER, the proportion of gap water for aRu3/WC_x (17.4%) was significantly greater than that of Pt/C (Fig. 5g, h, Supplementary Fig. 107). This indicates that the *OH species on WC_x supports significantly enhance the H-bond network connectivity by increasing the gap water population, thereby accelerating HER kinetics compared to Pt/C with carbon supports (Supplementary Note 12).

The electrochemical characteristics of aRu3/WC_x are also distinct from those typically reported for hydrogen-spillover-dominated HER electrocatalysts⁶⁹. Previous studies on hydrogen-spillover-based HER electrocatalysts have generally reported characteristic Tafel slopes of ~20−25 mV dec⁻¹ associated with spillover-assisted HER kinetics^70,71, whereas aRu3/WC_x exhibited a larger Tafel slope of 41.9 mV dec^-1 (Supplementary Fig. 91). In addition, hydrogen-spillover-based electrocatalysts often exhibit pronounced electrochemical hydrogen desorption features originating from substantial populations of spilled-over hydrogen species on the support surface^71,72. In contrast, aRu3/WC_x did not exhibit any noticeable hydrogen desorption peak in CV measurements (Supplementary Fig. 108), suggesting that substantial accumulation of spilled-over hydrogen species on the WC_x support does not occur in our system. Taken together, these results support a mechanism in which the WC_x support primarily enhances interfacial proton-transfer kinetics through H-bond network modulation, while the Ru ALNs remain the principal HER-active sites for hydrogen adsorption and H₂ evolution.

To distinguish the effects of the active-site structure and WC_x support, additional DFT calculations were performed for Ru ensembles supported on graphene and γ-WC models, together with Bader charge and ΔG_H analyses (Supplementary Figs. 109, 110). Both the Bader charge and ΔG_H values varied with the Ru ensemble size on both supports, indicating that neighboring Ru-Ru interactions significantly influence the electronic structure and hydrogen adsorption energetics of Ru active sites. Furthermore, Ru ensembles on γ-WC consistently exhibited lower Bader charge values and ΔG_H values closer to the ideal hydrogen adsorption free energy than those on graphene, suggesting that additional electronic modulation can be induced through strong Ru-WC_x interactions. These results suggest that both the Ru ensemble structure and the WC_x support contribute to optimizing the hydrogen adsorption energetics for alkaline HER.

To experimentally distinguish the active-site structure effect and the WC_x support effect, atomically dispersed Ru clusters on N-functionalized carbon (aRu/NC) with a Ru-Ru CN of 1.92 comparable to that of aRu3/WC_x (CN = 1.72) were synthesized (Supplementary Figs. 111, 112). In addition, Ru NPs supported on WC_x and carbon (RuNP/WC_x and RuNP/C) were prepared to directly examine the support effect for similar NP active-site structures (Supplementary Figs. 113–115). During alkaline HER (Supplementary Fig. 116), aRu3/WC_x exhibited substantially higher HER activity (η₁₀: 22 mV, Tafel slope: 41.9 mV dec⁻¹) compared with aRu/NC (η₁₀: 128 mV, Tafel slope: 105.5 mV dec⁻¹), despite their similar Ru coordination environments. Likewise, RuNP/WC_x exhibited better HER activity (η₁₀: 20 mV, Tafel slope: 30.8 mV dec⁻¹) than RuNP/C (η₁₀: 36 mV, Tafel slope: 52.1 mV dec⁻¹), further confirming the beneficial role of the WC_x support. These results collectively indicate that enhanced HER activity of aRu3/WC_x originates from the cooperative effects of (i) optimized hydrogen adsorption energetics induced by both the Ru ensemble size and strong Ru-WC_x interactions and (ii) enhanced proton-transfer kinetics arising from the favorable interfacial water structure provided by the WC_x support compared with inert carbon supports (Fig. 5g, h, Supplementary Note 12).

At the same time, comparison between aRu/NC and RuNP/C further suggests that neighboring Ru-Ru interactions themselves also strongly influence HER activity under inert carbon supported conditions, as Ru NPs exhibited substantially higher activity than small Ru clusters supported on carbon. This trend is also consistent with the DFT results, which predict relatively non-ideal ΔG_H values for low-coordination Ru ensembles on inert carbon supports compared with Ru ensembles stabilized on γ-WC (Fig. 4a, b, Supplementary Figs. 109, 110). Although RuNP/WC_x and aRu3/WC_x possess different active-site geometries, both catalysts exhibited high HER activity owing to the beneficial Ru-WC_x interactions, together with the favorable interfacial water structure provided by the WC_x support. However, unlike Ru NPs, all Ru atoms in the Ru ALN structures are directly stabilized through chemical interactions with WC_x, enabling substantially stronger SMSI stabilization and consequently greater durability during long-term HER operation (Supplementary Fig. 116d).

Therefore, the DFT calculations, together with experimental control studies, support that the enhanced alkaline HER activity of Ru ALNs originates from the cooperative contributions of (i) electronic structure modulation induced by both neighboring Ru-Ru interactions and strong Ru-WC_x interactions, (ii) adsorption-configuration changes enabled by neighboring Ru sites, and (iii) interfacial kinetic enhancement provided by the WC_x support.

To sum up, aRu3/WC_x exhibited higher kinetic activity than commercial Pt/C through two synergistic effects of dual-type sites (Ru ALNs for *H and WC_x supports for *OH) revealed by theoretical calculations and in situ spectroscopic analyses. First, the OH-enriched WC_x supports increased the population of gap water at the catalyst-electrolyte interface, which bridges the active sites and the bulk electrolyte, thereby restoring the H-bond network continuity and facilitating proton transfer across the interface. Second, the controlled ensemble size of Ru ALNs regulates the H-adsorption configuration and strength through cooperative geometric coupling among adjacent Ru atoms and electronic interactions with the underlying WC_x supports, maintaining a stable and highly active Ru configuration during alkaline HER operation. These cooperative effects provide a plausible explanation for the enhanced reaction kinetics of aRu3/WC_x, as consistently demonstrated by the in situ XAS, Raman, and ATR-SEIRAS results (Fig. 5).

Single-cell performances of Ru atom-layer neighbors

aRu3/WC_x provides an effective catalyst architecture for efficient, low-noble metal AEMWEs. The ALN architecture of Ru ALNs combines the advantages of both SACs and NP catalysts to maximize Ru utilization while creating a favorable environment for the HER. To assess the practical applicability of aRu3/WC_x, single-cell performance was investigated under practical AEMWE operating conditions using 1 M KOH. Among the aRuN/WC_x-based AEMWEs, the aRu3/WC_x reached a current density of 9.82 A cm^-2 at 2.0 V_cell and 80 °C with a low noble metal loading of 0.10 mg_Ru cm^-2 (Fig. 6a, Supplementary Fig. 118, Supplementary Table 20). This result indicates that size-tuned Ru ALNs contribute to the increase in AEMWE performance under identical Ru loading (0.10 mg_Ru cm^-2). Furthermore, the aRu3/WC_x-based AEMWE with a lower noble metal loading (0.1 mg_Ru cm^-2) achieved a comparable current density of 9.82 A cm^-2 to the benchmark Pt/C-based AEMWE (9.77 A cm^-2) with a higher noble metal loading of 0.4 mg_Pt cm^-2 (Fig. 6b, Supplementary Figs. 117, 118, Supplementary Table 20). Owing to its lower noble metal loading and the comparatively lower cost of Ru, aRu3/WC_x achieved 4.0 times higher mass activity (98.2 A mg_mass⁻¹) and 9.2 times higher specific activity (activity per unit price of noble metal, 5880.9 A EUR⁻¹) compared with Pt/C (Fig. 6c, Supplementary Table 20). At 1.8 V_cell, aRu3/WC_x reached 4.52 A cm^-2 at 80 °C and 3.65 A cm^-2 at a milder operating T of 60 °C (Fig. 6a) despite its low noble metal loading of 0.10 mg_Ru cm^-2, exceeding US DOE’s 2026 performance target (3.0 A cm^-2 at 1.8 V_cell) and ultimate noble metal loading target (0.125 mg cm^-2) for PEMWEs⁷³. Furthermore, the aRu3/WC_x cathode reached 3.10 A cm^-2, exceeding the DOE 2026 target performance even with a commercial AEM and AEI (PiperION-A) at 60 °C (Supplementary Fig. 120). Among reported AEMWE catalysts, aRu3/WC_x demonstrated competitive specific activity and met both DOE targets even at low T of 60 °C (Supplementary Fig. 121 and Supplementary Table 21, DOE target of specific activity was calculated using Ru price).

To compare the activity and durability of aRu3/WC_x with those of commercial benchmark catalysts (Pt/C and Ru/C) at the single-cell level, AEMWE tests (1 M KOH, 60 °C) were performed under identical noble metal loading conditions (0.10 mg_mass cm^-2). Among the tested cathodes, aRu3/WC_x delivered a current density of 8.30 A cm⁻² at 2.0 V_cell, compared with 7.23 A cm^-2 for Pt/C and 6.58 A cm⁻² for Ru/C (Supplementary Fig. 122). Furthermore, both Pt/C and Ru/C cathodes exhibited more pronounced degradation after 100 h operation at 1.0 A cm⁻², as evidenced by larger differences between the polarization curves before (BOT) and after (EOT) testing (Fig. 6e) and greater voltage increases at each current density, compared to those of aRu3/WC_x (Fig. 6f). In particular, at 1 A cm⁻², the voltage increase of aRu3/WC_x was approximately one-quarter of those for both Pt/C and Ru/C. To evaluate the durability of aRu3/WC_x, the stability number (S-number) was determined by quantifying the generated H₂ and the dissolved noble metal content during the 100 h durability test⁷⁴. All samples exhibited nearly 100 % Faradaic efficiency for H₂ based on real-time hydrogen quantification using a mass flow meter and gas chromatography (GC) analysis of the cathode outlet stream (Fig. 6d and Supplementary Figs. 123−126), indicating that the measured current predominantly originated from hydrogen evolution. At 1.0 A cm⁻², aRu3/WC_x achieved an S-number of 1.136×10⁸, which is five times higher than that of Pt/C (2.221×10⁷) and nine times higher than that of Ru/C (1.311×10⁷) (Fig. 6g), indicating that the SMSI between the Ru ALNs and the WC_x supports contributes to the lower noble metal dissolution and voltage degradation of aRu3/WC_x under the tested AEMWE conditions.

To further evaluate the operational stability of the Ru ALNs under practical intermittent AEMWE conditions, intermittent start-stop cycling tests were additionally performed by alternating between 1.0 A cm⁻² and open-circuit voltage (OCV) every 1 h for a total operation time of 100 h (Supplementary Fig. 127). aRu3/WC_x exhibited substantially smaller voltage degradation than commercial Pt/C even under repeated start-stop operation, indicating the high operational stability of the Ru ALN/WC_x system. In addition, inductively coupled plasma mass spectrometry (ICP-MS) analysis of dissolved metal species after the intermittent operation revealed no detectable Ru dissolution from aRu3/WC_x, whereas measurable Pt dissolution was observed from commercial Pt/C (Supplementary Table 22). Although trace W dissolution was detected from aRu3/WC_x, the dissolved amount was limited, indicating that the WC_x support remained largely stable during intermittent operation. These results further suggest that the Ru ALNs remain stably anchored on the WC_x support without significant detachment under practical alkaline AEMWE operating conditions.

To examine whether the high-current-density AEMWE performance originates from catalyst-dependent bubble behavior or mass transport effects, we systematically investigated the influence of catalyst structural features, including Ru ALN morphology, Ru aggregation, and support properties, on wettability, bubble evolution, and mass transport characteristics using aRu3/WC_x, aRu5/WC_x, aRu5_p/WC_x, commercial Ru/C and Pt/C cathodes. Most catalysts exhibited comparable contact angles (105.3 − 121.4°), except for aRu5_p/WC_x (83.7°) (Supplementary Fig. 128). In addition, Pt/C and WC_x-supported catalysts exhibited similarly rapid microbubble detachment behavior with no noticeable differences in bubble size or surface coverage during real-time optical visualization of hydrogen bubble evolution, whereas Ru/C showed localized bubble accumulation (Supplementary Fig. 129). Furthermore, multistep chronopotentiometry combined with EIS revealed comparable residual overpotentials for Pt/C and WC_x-supported catalysts, while Ru/C exhibited a substantially larger residual contribution (Supplementary Fig. 130). Notably, aRu3/WC_x and aRu5/WC_x showed similar residual overpotentials, indicating that differences in Ru ALN morphology do not substantially affect mass-transport-related losses. These results demonstrate that the AEMWE performance of aRu3/WC_x does not primarily originate from bubble shielding or gas transport effects, but rather from the enhanced catalytic activity and favorable interfacial kinetics of the Ru ALNs on WC_x.

The post-durability analyses of AEMWE operation revealed that both commercial Pt/C and Ru/C cathodes underwent severe NP agglomeration along with noticeable Fe/Ni deposition, originating from the migration and reduction of dissolved Fe/Ni ions from the anode (Supplementary Figs. 132, 133). The aRu3/WC_x cathode also exhibited Fe/Ni deposition (Supplementary Fig. 134), indicating that Fe/Ni redeposition is a common system-level degradation factor in AEMWE operation rather than a catalyst-specific phenomenon⁷⁵. Although such Fe/Ni redeposition may partially contribute to active-site blocking and performance degradation, the markedly different durability behaviors among the investigated cathodes correlate more strongly with the degree of noble metal migration, aggregation, and dissolution. In this regard, aRu3/WC_x preserved the atomic structure of Ru species on the WC_x supports, showing only minor and localized Ru agglomeration even after prolonged operation (Supplementary Figs. 134, 135, Supplementary Note 13). These results indicate that the strong metal-support interaction between Ru ALNs and WC_x supports effectively suppresses Ru migration and aggregation, thereby contributing to the lower degradation of aRu3/WC_x compared with commercial nanoparticle catalysts. It should be noted that further mitigation of Fe/Ni redeposition requires advances in anode and membrane design beyond cathode catalyst optimization (Supplementary Note 13).

Even when compared with Pt/C cathodes with four times higher noble metal loading (0.40 mg_Pt cm⁻²), the aRu3/WC_x cathodes (0.10 mg_Ru cm⁻²) showed lower voltage degradation during 500 h operation at an industrially relevant current density of 1.0 A cm⁻² (Fig. 6h and Supplementary Fig. 131). After 100 h, aRu3/WC_x retained comparable current density to Pt/C, and after 500 h, it exhibited a higher current density than Pt/C in the same voltage region (Supplementary Fig. 131c, d). Moreover, the voltage increase of the Pt/C cathode was approximately twice that of aRu3/WC_x after 500 h, despite having four times higher noble metal loading (Fig. 6i). These results indicate that the SMSI between Ru ALNs and the WC_x support contributes to the lower voltage degradation of aRu3/WC_x compared with Pt/C cathodes with higher noble metal loading under AEMWE operating conditions. Such activity-durability balance suggests the potential of aRu3/WC_x electrocatalysts for practical water electrolysis.