Customized reaction route for ruthenium oxide towards stabilized water oxidation in high-performance PEM electrolyzers

Rational design of the intuitive descriptor

We initiate by recognizing the lack of pertinent physical parameters to describe the overall performance of the Ru-based catalysts, irrespective of the reaction mechanism. Previously, reactivity descriptors, including the e_g filling^47,48, d band center²⁸, p band center¹¹, and the adsorption energy difference between catalytic sites and dopants proposed by ref. ⁴⁹. were found feasible in describing the activity trend of the catalysts following either the AEM or LOM mechanism. However, a unified scale, which comprises information on activity and stability in combined reaction mechanisms (both AEM and LOM), is lacking, thereby obscuring the advancing direction of the catalysts.

We reasoned that, despite the complexity, the catalytic activity and stability are interrelated and bridged by the Ru-O bonding nature, i.e. bond covalency and level of difficulty in Ru-O bond breaking⁵⁰, as follows. Weakening the Ru-O bond covalency implies the burying of the O 2p orbital under the Fermi level as well as the Ru 3d front orbital, thereby prohibiting lattice oxygen from participating in OER and the formation of oxygen vacancy (ΔG_VO)^45,51. In thus formed catalysts dominated by AEM, too low degree of Ru-O hybridization indicates too weak oxygen binding strength, which causes decreased Ru vacancy formation energy (ΔG_VRu) and eases Ru loss⁵², resulting in a simultaneous decrease in activity and stability. Strengthening the Ru-O bonding will firstly lead to a moderate intermediate binding energy as well as an increased ΔG_VRu under AEM, thereby improving both the activity and stability^53,54. However, further increasing the bonding strength leads to arise of lattice oxygen redox, where O 2p becomes the front orbital and thereby causes lowered ΔG_VO value¹⁸. Such a condition, though greatly boosts the intrinsic activity, causes the formation of oxygen defects and easy detachment of Ru-Ox moieties from defects’ neck, leading to rapid catalyst degradation^28,37. To this end, an appropriate performance indicator might be a physical parameter that could precisely describe the local Ru-O bonding nature.

We postulate that the Ru charge might be able to serve as such an intuitive descriptor, because it could not only perfectly reflect the local Ru-O bonding structure, i.e., the ionic or covalent bonding, but also the bonding strength. Therefore, it is possible to customize both the reaction activity and stability, using Ru charge as a regulator (Fig. 1). We, therefore, begin our study by looking for a suitable model catalyst system to verify this idea, and found that the MRuO_x solid solution (M in +4 valence to avoid additional structure influence) with a clear Ru-O-M structure motif might be suitable for such a proposal. Specifically, the Ru charge can be regulated by adjusting the ionic electronegativity of M, where M⁴⁺ in higher ionic electronegativity than Ru⁴⁺ leads to reduced electron density on Ru, and vice versa. Consequently, we selected M = Ce⁴⁺, Sn⁴⁺, Ru⁴⁺, and Cr⁴⁺ with an ionic electronegativity of 1.608, 1.706, 1.848, and 1.861 to carry out our further study^55,56.

DFT calculational predictions

We firstly carried out DFT calculations with the simulation model of M_0.5Ru_0.5O₂ with Ru-O-M motifs constructed (Fig. 2a) based on the stable rutile RuO₂ (110)^18,57 (Supplementary Figs. 2–6 and Supplementary Note 1). The charge of Ru in Ru-O-M was first calculated. As the electronegativity of M⁴⁺ increases from 1.608 (Ce⁴⁺) to 1.861 (Cr⁴⁺), the charge of Ru increased from 1.49 (Ce_0.5Ru_0.5O₂) to 1.57 (Cr_0.5Ru_0.5O₂) (Fig. 2b). We then evaluated the dependence of theoretical OER activity on the Ru charge, in both AEM and LOM reaction pathways on Ru sites, as the M sites were calculated to possess extremely poor activities under both reaction pathways (M = Ce, Sn, Cr) (Supplementary Figs. 7–13 and Supplementary Note 2).

The AEM pathway activities of the materials are revealed to follow a volcanic-like trend as a function of Ru charge. The theoretical OER overpotentials of M_0.5Ru_0.5O₂ were calculated first, with the Sn_0.5Ru_0.5O₂ sample exhibiting moderate Ru charge (1.52) representing the lowest theoretical OER overpotential. For the samples with both lower (Ce_0.5Ru_0.5O₂) and higher (RuO₂ and Cr_0.5Ru_0.5O₂) Ru charge, increases in OER overpotentials are observed (Fig. 2c and Supplementary Figs. 14–17). While this phenomenon implies that an optimum Ru charge that is neither too high nor too low is beneficial for OER, the underlying reason is that Ru charge can be successfully used to reflect the Ru-O bonding interaction. Specifically, we calculated the overlap band center of Ru d orbital and O p orbital in the partial density of states (pDOS, Supplementary Fig. 18)⁵⁸ and found that the overlap band center (ε-Rud-Op) takes the same trending with the change in Ru charge, i.e., increases from −4.19 to −2.95 eV with an increase in Ru charge (1.49–1.57). To this end, the Ru charge can truly be utilized as the intuitive descriptor because it reflects the status of both Ru itself and the Ru-O orbital interaction. Indeed, as the Ru charge increases, the M_0.5Ru_0.5O₂ undergoes alteration in the potential determine step (PDS) from OH*/O* to O*/OOH* (Supplementary Figs. 13, 14), due to strengthening in Ru-O* binding, in line with bond shrinkage from 1.749 to 1.732 Å (Supplementary Fig. 19). Therefore, Ru sites charge can be successfully used to indicate the AEM reaction activity.

We then move to calculate the reactivity of lattice oxygen in different M_0.5Ru_0.5O₂ structures and the feasibility of the LOM route. First, upshifting in O 2p band center (Fig. 2d, e and Supplementary Fig. 20) and strengthening in Ru-O bond covalency are observed with increases in Ru charge, as indicated by increases in both ε-Op-Rud, O 2p band contribution near the E_F (from 2.2 to 25.2%)^40,59,60 and downshifted ICOHP between Ru and O (from −1.280 to −1.733 eV, Supplementary Figs. 21, 22). Second, we calculated the formation energies of oxygen vacancy(ΔG_VO) in different samples, to analyze the occurrence possibility of LOM¹⁸. The significantly downshifted ΔG_VO from 7.20 eV (Ce_0.5Ru_0.5O₂) and 8.25 eV (Sn_0.5Ru_0.5O₂) to 5.68 eV (RuO₂) and 5.29 eV (Cr_0.5Ru_0.5O₂) suggests the higher possibility of LOM involvement in RuO₂ and Cr_0.5Ru_0.5O₂ with high Ru charge (Supplementary Fig. 23). The lower ΔG_VO of Ce_0.5Ru_0.5O₂ compared to that of Sn_0.5Ru_0.5O₂ may originate from the distortion of the crystal structure⁶¹ (Supplementary Figs. 24–27 and Supplementary Note 3). Third, we turned to calculate the LOM energy variation for M_0.5Ru_0.5O₂ (Fig. 2f and Supplementary Figs. 28–31). Clearly, the theoretical overpotential of OER under the LOM path gradually decreases from 0.84 V for Ce_0.5Ru_0.5O₂ to 0.34 V for Cr_0.5Ru_0.5O₂. Therefore, the direct O-O coupling on the O p band above the E_F is more thermodynamically favored with a higher Ru charge (i.e., in RuO₂ and Cr_0.5Ru_0.5O₂). Based on the above calculations, the correlation between a charge of the Ru site and the reaction path (AEM and LOM) as well as reaction energy on M_0.5Ru_0.5O₂ (Fig. 1) is unveiled. While low charge (Ce_0.5Ru_0.5O₂ and Sn_0.5Ru_0.5O₂) at Ru sites leads to the domination of the AEM path, the LOM route is energetically more favored at elevated Ru charge. Moreover, by changing the Ru charge, the OER catalytic activity of catalysts following AEM can also be adjusted due to the regulation of Ru-O interaction in M_0.5Ru_0.5O₂.

We then move to our next step, i.e., probing the applicability of the Ru charge in describing the stability of the catalysts. Two indicators, i.e., the formation energies of Ru (ΔG_VRu) and O (ΔG_VO) vacancies, are used jointly to describe the stability of the catalysts. On one hand, it is expected that increasing the Ru charge results in strengthened Ru-O bonding and thereby increased ΔG_VRu in perfectly crystallized M_0.5Ru_0.5O₂ samples, with ΔG_VRu increasing gradually from 9.44 eV in Ce_0.5Ru_0.5O₂ to13.92, 14.60, and 15.13 eV for Sn_0.5Ru_0.5O₂, RuO₂, and Cr_0.5Ru_0.5O₂, respectively (Fig. 2g). On the other hand, however, increase in Ru charge causes ease formation of O_V (Supplementary Fig. 23) and thereby decreases in ΔG_VRu (Supplementary Fig. 32) due to lower overall Ru-O chelation number. Specifically, with the formation of one neighboring O_V, the ΔG_VRu value for RuO₂ and Cr_0.5Ru_0.5O₂ decreases from 14.60 and 15.13 eV to 8.51 and 7.95 eV, respectively, resulting in poor stability (Fig. 2g). To this end, the Sn_0.5Ru_0.5O₂ near the apex of the volcano curve under the AEM path demonstrates the best stability. The rationale behind this is that the low Ru charge (Ce_0.5Ru_0.5O₂) makes the Ru species prone to dissolution in acid, too high Ru charge leads to the readily generation of O_V defects and smaller ΔG_VRu, thus resulting in the overall destabilization. Since metal dissolution may occur during acidic OER, defective structures with Ru and M vacancies were also constructed for calculations (Supplementary Figs. 33–43 and Supplementary Note 4). Although the inclusion of metal center vacancies leads to the trend of crossover between AEM and LOM pathways (Supplementary Fig. 44), in line with recent work from Alexandrov et al.^62,63, Ru charge is still valid in describing the reaction path and energy as the vacancies influence the Ru charge and OER reaction in the same way. Specifically, the inclusion of vacancies leads to the evolution of the Ru-O bonding feature and thereby Ru charge (Supplementary Fig. 45 and Supplementary Note 5), which thus determines the theoretical activity under different paths. As a result, the theoretical performance of defective M_0.5Ru_0.5O₂ on the scale of Ru charge perfectly matches that of the non-defective structure. To here, it is clear that the Ru charge can be used to describe not only the major reaction path and OER activity, but also the structural stability, due to its successful reflection in both electronic features of Ru and overall Ru-O bonding interactions (Fig. 1).

Synthesis and characterization

To experimentally validate the DFT prediction, MRuO_x oxide solid solution with a M/Ru ratio close to 1 was synthesized according to the previous reports (see Methods for details)^16,35. Powder X-ray diffraction (XRD) patterns show that all prepared MRuO_x samples possess a rutile-type structure without any distinct diffraction peaks corresponding to MO_x and RuO₂ (Supplementary Fig. 46), indicating the formation of MRuO_x oxide solid solution, in line with the uniform elements distribution revealed by energy-dispersive X-ray spectroscopy (EDX) (Supplementary Figs. 47–50). Rietveld refinement analysis (Supplementary Fig. 51 and Supplementary Table 1) suggests the random displacement of Ru by M in MRuO_x solid solution, giving the formation of Ru-O-M structure motif (Supplementary Fig. 52). The atomic ratios of M and Ru were close to 1, as analyzed by inductively coupled plasma optical emission spectrometer (ICP-OES) and EDX (Supplementary Fig. 53 and Supplementary Table 2).

We then move on to probe the local structure and chemical states of Ru and O in MRuO_x. X-ray photoelectron spectra (XPS) show that both Ru 3d and Ru 3p_3/2 (Fig. 3a) signals could be deconvoluted into two doublets, representing Ru⁴⁺ (fill in green) and the satellite peaks (fill in yellow)^54,64. Notably, the Ru 3d and Ru 3p binding energies for CeRuO_x (280.1 and 461.9 eV), SnRuO_x (280.4 and 462.1 eV), RuO_x (280.6 and 462.5 eV), and CrRuO_x (280.8 and 462.8 eV) consecutively shifted to more positive values, implying a successive increase in Ru valence states, in line with the DFT calculation results. We further collected the Ru K edge spectra through X-ray absorption near-edge spectroscopy (XANES) to quantitatively describe the oxidation state of Ru. As presented in Fig. 3b, taking Ru foil and commercial RuO₂ as references, the oxidation states of Ru in MRuO_x were estimated to be +3.3, +3.6, +3.9, and +4.2 for CeRuO_x, SnRuO_x, RuO_x, and CrRuO_x, respectively. The variation in Ru valence and its consistency with DFT calculation results verifies the feasibility of tailoring the Ru electron density via controlling the element M in the Ru-O-M motif, making it promising for reaction path and O* adsorption energy regulation.

We next turned to validate experimentally if the OER path of MRuO_x truly follows the DFT calculation results as a function of Ru charge, initiated by monitoring the Ru-O bond covalency via Ru XANES spectrum at L₃ edge (Fig. 3c)⁶⁵. As reported previously, the Ru XANES spectrum at the L₃ edge can be fitted by two peaks ascribable to e_g and t_2g, and the energy splitting ΔE (ΔE = E_eg – E_t2g) is positively correlated with the Ru-O bond covalency. Notably, we find that ΔE increases from 1.51 to 1.95 eV (Fig. 3c, right) with the increase in valence state (+3.3 to +4.2), thus testifying to the greatly enhanced Ru-O bond covalency with increased Ru charge, in line with the increase in Rud–Op overlap in the calculation (Fig. 2d, e). O 1 s XPS spectra (Supplementary Fig. 54 and Supplementary Table 3) further suggest significant increases in O_V contents in RuO_x (36%) and CrRuO_x (41%) compared with that of CeRuO_x (25%) and SnRuO_x (15%), in good compliance with the increase in bond covalency¹¹. The occurrence of LOM is further verified by in situ ¹⁸O isotope-related DEMS measurements on all MRuO_x samples (Fig. 3d, Supplementary Figs. 55–63, and Supplementary Note 6) according to the previous reports^66,67,68. By extrapolating the integration of the mass spectra signal (¹⁶O/¹⁸O) per second to the OER onset, the content of the LOM path during OER is determined (Supplementary Figs. 55–63e). Specifically, the AEM path is revealed to be dominant on CeRuO_x and SnRuO_x, as the ¹⁶O/¹⁸O ratio of O₂ catalyzed by CeRu¹⁶O_x and SnRu¹⁶O_x in ¹⁸O-labeled 0.5 M H₂SO₄ is 6.88 and 6.85% (Supplementary Figs. 56, 57e), respectively, which is consistent with that of the commercial IrO₂ (6.56%, Supplementary Fig. 55), attributable to the isotopic abundance of ¹⁶O in H₂¹⁸O. On Ru¹⁸O_x and CrRu¹⁸O_x, however, the ¹⁶O/¹⁸O ratio in products significantly increased to 28.63 and 41.96%, leading to an increment in the LOM ratio of RuO_x and CrRuO_x to 12.6 and 36.7%, respectively (Supplementary Figs. 58–59). Besides, the DEMS measurements in ¹⁸O-labeled 0.1 M PBS (pH = 6) also demonstrate similar results (Supplementary Figs. 60–63). It is noted that the LOM ratio differs significantly in literature as reported by refs. ^67,68,69, etc., mainly owing to the difference in structural properties of the catalysts, while our result is in high accordance with that reported by ref. ³⁸, due to the similar solid solution structure (Supplementary Note 7). To here, it is clear that the Ru charge is a useful parameter to customize the OER path on MRuO_x.

OER catalytic performance in a three-electrode configuration

The OER performance of the prepared MRuO_x was firstly evaluated in a conventional three-electrode setup at a catalyst loading of 41.65 μg_cat cm⁻² (Supplementary Figs. 64, 65 and Supplementary Note 8) with 0.5 M H₂SO₄ electrolyte according to the guideline recently set^70,71. The linear sweep voltammetry (LSV) curves normalized by the geometric area (Fig. 4a), electrochemical active surface area (ECSA) (Supplementary Fig. 66a, b), Ru mass loading on the electrode (Supplementary Fig. 66c) and corresponding Tafel slopes (Supplementary Fig. 66d–f) revealed the variation in OER catalytic activity of MRuO_x. As summarized in Fig. 4b and Supplementary Fig. 67, the activity trend of MRuO_x is in good agreement with DFT prediction. Specifically, on samples dominated by AEM path (CeRuO_x and SnRuO_x), the OER activity is significantly enhanced with an increase in Ru charge (strengthened O* adsorption), accompanied by the variation in reaction determine step as indicated by the difference in Tafel slope (60.2 and 38.2 mV dec⁻¹) (Supplementary Fig. 66d and Supplementary Note 9). Further increase in the Ru charge leads to a performance decline in AEM but the occurrence of LOM. Therefore, an activity drop in RuO_x and a slight lift in CrRuO_x are observed, in line with the DFT calculation. Notably, the SnRuO_x customized near the apex of AEM exhibited remarkable OER catalytic activity. Satisfactorily, a small catalytic overpotential (η) of 194 mV is needed for SnRuO_x to deliver a current density of 10 mA cm⁻², which is 176 mV lower than that of commercial RuO₂ (Fig. 4a). Furthermore, the mass activity and turnover frequency (TOF) of SnRuO_x at 1.48 V vs. RHE is 36.4 times that of commercial RuO₂, reaching 2360 A g_Ru⁻¹ and 0.63 s⁻¹ (Supplementary Fig. 68). These results unambiguously suggest SnRuO_x as one of the most active Ru-based catalysts towards acidic OER (Fig. 4e and Supplementary Table 4).

We then tested the stability of MRuO_x during operation to probe into the effectiveness of Ru charge in describing the stability. To start with, chronoamperometry (CA) tests (5 h) suggest SnRuO_x as the most stable catalyst, with high current retention of 91.7, 77.15, and 66.48% after operating at 1.5, 1.7, and 1.9 V vs. RHE, respectively (Supplementary Fig. 69). By contrast, the CeRuO_x, CrRuO_x, and RuO_x only demonstrate current retention of 54.75, 83.09, and 53.84% at 1.5 V vs. RHE, respectively, implying their poor stability. Furthermore, to eliminate the interference of other factors beyond the catalyst, i.e., thin film quality^72,73, active sites blockage⁷⁴, and back electrode passivation^75,76, the stability of MRuO_x was evaluated by monitoring the dissolution behavior of M and Ru during CA tests. By calculating the Ru stability number (S-number)⁷⁷ during CA tests, the stability of MRuO_x is found to be in line with the ΔG_VRu from the DFT calculations (Supplementary Fig. 70). Given the perfect match in catalytic stability between experiments and theoretical computations, we further run a CP test at 100 mA cm⁻² for the SnRuO_x sample to verify its stability (Fig. 4c). Excitingly, during a 250 h test, the overpotential only increased by 26.8 mV, giving a degradation rate of merely 107.2 μV h⁻¹, three orders of magnitude smaller than that of commercial RuO₂ (Supplementary Fig. 71) and comparable to that of the advanced Ir-based catalysts (Fig. 4e and Supplementary Tables 5, 6)^{12,14,18,19,38,52,53,54,78,79,80,81,82,83,84,85,86,87,88,89,90}. Besides, by carrying out CA tests in a weak acidic environment (0.1 M PBS, pH = 6), the contribution of M (M = Ce, Sn, and Cr) dissolution on the performance degradation of MRuO_x is eliminated (Supplementary Figs. 72, 73 and Supplementary Note 10), further demonstrating the stability of MRuO_x is dependent on the Ru stability. These results corroborate the effectiveness of the Ru charge not only in describing the trend in catalytical activity, but more importantly, in screening out the most durable catalysts ever reported.

The origin of the high stability

To further probe into the stabilization mechanism of SnRuO_x during OER operation, we adopted in situ XAS and in situ Raman spectroscopy to investigate the dynamic variation in local Ru structure and chemical environment. RuO_x was also tested as the counterpart catalyst for comparison. According to the literature, the over-oxidation of Ru originating from the structure damage under high oxidation potential is one of the main culprits leading to severe Ru dissolution⁹¹. Thus, Ru K edge XANES spectra were collected to monitor the change in Ru oxidation states of SnRuO_x and RuO_x during operation. As presented in Fig. 5a, b and summarized in Fig. 5c, SnRuO_x and RuO_x show big differences in Ru valence states during operation (Supplementary Fig. 74), especially at higher applied potentials. Specifically, switching on the electrode potential from open circuit potential (OCP) to 1.4 V vs. RHE leads to a rapid rise in Ru valence from +3.6 to +4.2 in SnRuO_x, followed by a gentle and steady change from +4.2 to +4.4 with potential further lifted to 2.0 V vs. RHE. On the contrary, the Ru oxidation states in RuO_x increase at a constant rate from +3.9 to +4.9 with the applied potential ranging from OCP to 2.0 V vs. RHE, showing no sign of stabilization. This result indicates that even at high oxidation potentials, the stable structure of SnRuO_x can be maintained under the customized AEM path, whereas RuO_x undergoes significant structural evolution, probably ascribable to the involvement of LOM.

We then carried out in situ Raman spectroscopy to clarify the structure evolution during OER. For both SnRuO_x and RuO_x, two major peaks at ~435 and 595 cm⁻¹ assignable to E_g and A_1g vibration modes were observed (Supplementary Fig. 75)^92,93. With the potential increases from OCP to 1.8 V vs. RHE, the SnRuO_x sample maintains a constant Raman shift, suggesting the consistency in Ru-O bonding structure during OER. However, a ca. 9 cm⁻¹ positive shift in RuO_x is noticed, implying the shrinkage in Ru-O bonding length during OER. To intuitively uncover the local structure evolution of SnRuO_x and RuO_x during OER, Fourier transforms (FTs) analysis of extended X-ray absorption fine structure (EXAFS) spectra was carried out to provide R-space information. A prominent peak of ~1.45 Å (phase uncorrected) corresponds to the first Ru-O coordination shell can be observed (Fig. 5d, e and Supplementary Figs. 76, 77) for both samples. Interestingly, while only small fluctuations in Ru-O bond length (r_Ru-O) are observed for SnRuO_x (from 1.97 to 1.96 Å) with potential increases from 1.0 to 2.0 V vs. RHE, a significant r_Ru-O shrinkage from 1.96 to 1.86 Å is observed in RuO_x (Fig. 5f), in well accordance with the Raman results. Meanwhile, while quantitative Ru K edge fitting results (Supplementary Table 7) show no significant change in Ru-O coordination number (CN) for SnRuO_x, a significant decrease in CN from 6.4 to 4.7 is evidenced in RuO_x, ascribable to the generation of O_V due to the occurrence of LOM.

We further performed quasi in situ soft-XAS measurements to collect the O K edge information, to directly evidence the formation of O_V by recording the total electron yield (TEY) intensity before and after operating under different potentials. Two pre-edge peaks at ~532 and 534.5 eV assignable to the unoccupied orbital of O 2p hybridized with Ru 4d t_2g and e_g orbitals are observed for both SnRuO_x and RuO_x (Fig. 5g–i)⁴⁵. According to previous reports, the reduction in t_2g–p/e_g–p peak intensity (I_t2g-p/I_eg-p) implies the formation of the O_V due to the electron occupation of the lowest energy state (t_2g) of octahedral symmetry^94,95. For RuO_x, the ratio of I_t2g-p/I_eg-p decreases monotonously with potential increases from 1.2 to 2.0 V vs. RHE, implying the generation of a high amount of O_V during OER. On the contrary, no obvious variation in I_t2g-p/I_eg-p is observed for SnRuO_x, suggesting that there is no increase in O_V content and thus ruling out the occurrence of LOM, which endows it with excellent operational stability. These results are in good agreement with the catalyst design idea as SnRuO_x was customized to follow the AEM path and exhibited high stability during OER due to moderate Ru charge.

Performance of PEMWE devices

Encouraged by the overall high activity and stability of SnRuO_x towards OER, we finally assembled a single cell using the catalyst as the anode to evaluate its performance in a real PEMWE device (Supplementary Fig. 78, see Methods for details). The steady-state polarization curve of membrane electrode assemblies (MEAs) operated at 50 °C (Fig. 6a) shows that the performance of PEMWE can be greatly improved with the utilization of SnRuO_x. Specifically, cell voltages of only 1.565, 1.655, and 1.735 V are needed to reach current densities of 1, 2, and 3 A cm⁻², respectively, far superior to that obtained with commercial RuO₂ (1.733 V@1 A cm⁻² and 1.883 V@2 A cm⁻²) (Supplementary Fig. 79) and also well surpassing the DOE 2025 target (3 A cm⁻²@1.9 V)⁹⁶. Taking noble metal cost and cell performance into account, the overall advantages of SnRuO_x are self-evident (Fig. 6b). With anode noble metal cost of US$ 0.0194 cm⁻², high energy efficiencies at 78.7 and 74.3% were achieved at 1 and 2 A cm⁻², respectively, corresponding to energy consumptions at 41.9 and 44.4 kWh kg⁻¹H₂, outperforming the prior reported most cost-efficient OER catalysts (Fig. 6b, Supplementary Table 8, and Supplementary Note 11)^{20,97,98,99,100,101,102}. According to the calculation from US DOE, only US$ 0.838 is required for this PEMWE device to produce 1 kg H₂, which is much lower than the DOE target (H2⁻¹), further proving the efficiency of the SnRuO_x at the anode.

The long-term operational stability is even more exciting, as shown in Fig. 6c, the SnRuO_x-based cell demonstrates well-maintained PEMWE performance after running under 1 A cm⁻² for 1300 h, with apparent degradation rate only at 53 μV h⁻¹. Such stability not only represents a record for Ru-based catalysts, but even superior to that of the advanced Ir-based OER catalysts recently reported (Fig. 6d and Supplementary Table 9)^{14,20,38,52,97,99,103,104,105,106,107}. To more accurately determine the lifetime of SnRuO_x in PEMWE, a modified MEA setup without metallic parts in the anode and cathode water cycle was employed according to the previous reports (Supplementary Fig. 80–84, Supplementary Table 10, and Supplementary Note 12)¹⁰⁸. With all the Ru dissolved in the anode/cathode water and membrane monitored during 10 h operation at 1 A cm⁻², the S-number of SnRuO_x is calculated to be 1.04 × 10⁶ (Supplementary Note 13)⁷⁷. These results reveal the inherently robust nature of SnRuO_x as an OER catalyst with highlighted practical application prospects.