Carbon mesopore depth engineering boosts the performance of low-platinum fuel cells

Abstract

Ionomer adsorption-induced Pt poisoning and high local oxygen transport resistance constitute two critical challenges preventing proton exchange membrane fuel cells from achieving low-Pt loadings. Yet, strategies concurrently mitigating both issues remain limited, making it challenging to achieve satisfactory performance in both kinetically controlled low-current-density and mass transport-controlled high-current-density regions. Herein, we show that this limitation can be addressed with a mesoporous carbon support design whose key lies in constructing a small mesopore-dominated structure while rationally controlling pore depth to a medium level. This design creates an ideal local reaction environment for Pt nanoparticles inside the mesopores, one that protects most Pt from ionomer poisoning while achieving efficient oxygen diffusion. Using the developed carbon support, we synthesize an intermetallic PtCo nanocatalyst that delivers competitive membrane electrode assembly performance at low-Pt cathode loading (0.1 mg_Pt cm⁻²), with a H₂-air rated power density of 1.23 W cm⁻² at 0.67 V under heat-rejection-constrained conditions and a H₂-O₂ mass activity of 1.1 A mg_Pt⁻¹.

The heavy use of platinum catalysts limits the large-scale deployment of proton exchange membrane fuel cells. Here, the authors report a carbon support design with small mesopore-dominated structures and moderate pore depth that improves performance at low platinum loading.

Introduction

Proton exchange membrane fuel cells (PEMFCs) are zero-emission energy conversion devices with significant promise for decarbonizing and electrifying road transportation. However, their widespread adoption faces economic challenges¹. One major factor is the high platinum (Pt) loading required at the cathode to drive the sluggish oxygen reduction reaction (ORR). It has been estimated that achieving cost competitiveness for fuel cell vehicles necessitates a reduction in cathode Pt loading to below 0.1 mg_Pt cm⁻². While the exacerbation of kinetic-related voltage losses by reducing Pt loading is well-known, a series of reports published in recent years have demonstrated that it also creates voltage losses associated with mass transport^2,3. The reason is that low loading reduces the surface area of Pt, forcing each active site to handle a higher reactant flux, ultimately resulting in an increase in the local transport resistance, particularly for oxygen^4–6. Therefore, in the pursuit of reducing Pt usage, catalyst development is not only challenged by kinetic activity but also by local oxygen transport.

Generally, ORR catalysts are fabricated as carbon-supported Pt nanoparticles to improve Pt atomic utilization. In membrane electrode assemblies (MEAs), the cathode is composed of ionomers and catalysts. The ionomer forms a nanofilm on the catalyst surface, serving as a proton conductor that transports protons to the active sites⁷. While historically, Pt active structures were considered the determinant of catalyst kinetic activity, recent studies show a significant influence from the Pt/ionomer interface. At this interface, sulfonate groups from the ionomer strongly adsorb onto Pt, causing catalyst poisoning^8–10. Besides, molecular dynamics simulations suggest that the Pt/ionomer interface greatly elevates local oxygen transport resistance, due to ionomer densification caused by strong ionomer-Pt interaction^11,12. This theoretical prediction is supported by the following experimental findings: oxygen permeability is observed to decrease by up to an order of magnitude for ionomer nanofilms in the catalyst layer compared to the bulk counterpart, whereas similar oxygen permeability is observed for bulk and nanofilm ionomers in the absence of a Pt/ionomer interface^13,14. The above findings appear to indicate that reducing the Pt/ionomer interface benefits both kinetic activity and local transport; however, the practical reality presents a more complex picture. A well-known example is that catalysts made with solid carbon typically exhibit better high-current-density performance compared to those made with porous carbon, even though the porous carbon-supported catalyst has a significantly smaller Pt/ionomer interface area^15,16. Unlike solid carbon, porous carbon supports (e.g., Ketjenblack) normally possess internal mesopores and micropore openings, which are favorable to prevent ionomer infiltration into the carbon interior and thus reduce the contact of ionomer with Pt nanoparticles inside the carbon. A widely accepted explanation for the inferior high-current-density performance of porous carbon-supported catalysts is that the internal pores of porous carbon limit the reactant accessibility to Pt nanoparticles.

Nevertheless, choosing a porous carbon support presents an overwhelming advantage for activity enhancement. This is not only because it can mitigate ionomer poisoning of Pt, leading to a 2- to 4-fold increase in activity compared to solid carbon support¹⁷. Moreover, porous carbon is better suited for affording highly active Pt alloy catalysts, as its high surface area helps to minimize particle sintering during high-temperature synthesis¹⁸. This high surface area also endows porous carbon-supported catalysts with improved resistance against electrochemical particle coarsening under PEMFC operation¹⁶. As a result, porous carbon remains a highly attractive option, provided that the local transport issue can be resolved. In this regard, Anusorn Kongkanand et al. from General Motors reported a novel porous carbon support (HSC-f) that successfully reduced the local transport resistance without sacrificing activity¹⁶. HSC-f features micropore openings (1–2 nm) similar to traditional porous carbon supports like Ketjenblack, which effectively prevent ionomer infiltration into the carbon interior, thus ensuring high kinetic activity; however, HSC-f differs by having smaller primary carbon particles (~21 nm) and larger internal mesopores (7–14 nm), a combination believed to shorten and enlarge transport channels, thereby realizing good reactant transport¹⁹. Unfortunately, the preparation method for HSC-f remains undisclosed to date, and no engineering strategy has yet been reported in the literature that allows the realization of such finely controlled pore architectures.

Herein, we report a carbon mesopore depth engineering (CMDE) strategy to design effective carbon supports that can simultaneously optimize both the kinetic activity and local oxygen transport of the cathode catalyst, thereby enabling the fabrication of high-performance fuel cells at low-Pt loadings. Rather than involving complex pore morphology engineering, our strategy emphasizes two simple design elements: constructing a small mesopore-dominated carbonaceous structure and controlling the pore depth at a moderate level. The rationale for this pore configuration is that the ionomer has a limited thermodynamic penetration depth in small mesopores; thus, moderate pore depth control can reduce the formation of activity-detrimental Pt/ionomer interfaces, while maintaining the diffusion length at a reasonable level to ensure fast local oxygen transport. Utilizing this carbon support engineering, we synthesized an intermetallic PtCo nanocatalyst that delivered encouraging MEA performance at a low-Pt cathode loading of 0.1 mg_Ptcm^‒2, including a rated power density of 1.23 W cm^‒2 under heat-rejection-constrained conditions and a mass activity of 1.1 A mg_Pt^‒1.

Results

Design principle of the CMDE strategy

The development of the CMDE strategy was initially inspired by our molecular dynamics (MD) simulations investigating the penetration behavior of ionomers into carbon nanopores of varying diameters, aimed at identifying an optimized pore size that minimizes direct Pt-ionomer contact. For this purpose, a molecular model comprising water, ionomers, carbon nanotubes (representing the pores), and hydronium ions was constructed for the penetration analysis (Fig. 1a), in which three pore diameters (d = 1, 3, and 5 nm) were considered. To evaluate ionomer penetrability, the corresponding free-energy profiles were calculated, with the origin of each profile normalized to G (0) = 0. As shown in Fig. 1b, ionomer penetration is thermodynamically unfavorable at 1 nm but becomes favorable at 3 and 5 nm. Notably, despite the complete ionomer exclusion in 1 nm pores, these micropores are physically too small to accommodate Pt nanoparticles, which typically require a size range of 3–5 nm to balance activity and stability^16,19,20. For the 3 nm mesopore, the free-energy change ($\mathrm{\Delta }G$, representing the difference in free energy between adjacent points along the pore depth direction) approaches zero at a penetration depth of ~2.6 nm, whereas for the 5 nm mesopore, $\mathrm{\Delta }G$ remains close to zero throughout the entire penetration process. These results indicate that ionomer penetration in small mesopores of ~3 nm is subject to an intrinsic depth limit, whereas this constraint is essentially absent in larger mesopores (≥5 nm). We further found that the limited thermodynamic penetrability in small mesopores of 3 nm appears to arise from a balance between two competing factors: the ionomer-carbon interaction and the effect of ionomer intrusion on the stable hydrogen-bonding network inside the mesopore. As shown in Fig. 1c, the average values of the radial distribution function between ionomer and carbon atoms in the carbon support ($\overline{g_{\mathrm{Ionomer}-\mathrm{C}}},r\le 0.5\mathrm{nm}$) increase with penetration depth and eventually converge, indicating the strengthened interactions between hydrophobic ionomer backbones and carbon surface, which stabilize the ionomer and promote penetration^21,22. Conversely, the number of hydrogen bonds between water molecules ($\mathrm{\Delta }{N}_{\mathrm{Hbond}}$) decreases with penetration depth, reflecting the disruption of the stable hydrogen-bonding network inside the mesopore, resulting in resistance to ionomer penetration^23,24. It should be noted that these MD simulations use idealized conditions (simplified pore geometry, fixed ionomer chain, no polarization); thus, results are qualitative rather than quantitative.

Fig. 1. Design principle of the CMDE strategy.

a Molecular model of ionomer penetration. The gray, red, white, yellow, and green balls represent carbon, oxygen, hydrogen, sulfur, and fluorine atoms, respectively. b Free-energy profiles for ionomer penetration into carbon mesopores with different diameters. c The average values of the radial distribution function between ionomer and carbon atoms (red line) and the number of hydrogen bonds between water molecules (blue line) at different ionomer penetration depths in a 3 nm mesopore. Insets show snapshots of the hydrogen-bonding network inside the mesopore without (right) and with (left) ionomer penetration. The green, white, yellow, pink, and red balls represent carbon, hydrogen, sulfur, fluorine, and oxygen atoms, respectively. The hydrogen bonds are shown as black dashed lines. d Schematic illustration of the CMDE strategy. Source data for this figure are provided as a Source Data file.

Although ionomer penetration into mesopores is inevitable, the limited penetration depth inherent to small mesopores inspired us to spatially isolate the majority of Pt particles from ionomer adsorption by extending the pore depth. On the other hand, it is conceivable that increasing the pore length elongates the diffusion path for reactant oxygen to Pt sites and product water to the external surface, potentially causing a local transport penalty. In particular, without efficient removal, product water is prone to capillary condensation, severely hindering oxygen access to Pt sites. Taking these opposing effects into account, we envision that a moderate pore depth configuration may exist, which can keep the majority of Pt nanoparticles free from ionomer adsorption, while controlling the diffusion distance within a threshold that still maintains efficient local transport (Fig. 1d).

Catalyst synthesis and characterization

To validate the CMDE strategy, we employed hollow mesoporous carbon spheres (HMCS) as a model system and synthesized three HMCS variants with comparable small mesopore-dominated architectures but distinct pore depths via a nanocasting method. The synthesis process involved three key steps^24–26: preparation of core–shell mesoporous silica spheres (SiO₂@mSiO₂) consisting of a solid SiO₂ core and a mesoporous SiO₂ shell, chemical vapor deposition (CVD) of ferrocene, and subsequent etching and annealing (Fig. 2a). The carbon shell thickness of HMCS was controlled by adjusting the thickness of the mesoporous SiO₂ shell. Statistical analysis of high-angle annular dark-field scanning transmission electron microscopy (HAADF-STEM) images confirmed that all three carbon supports exhibited similar hollow diameters of approximately 200 nm, with shell thicknesses of 18 nm, 86 nm, and 125 nm, respectively (Fig. 2b–e and Supplementary Fig. 1). The resulting hollow mesoporous carbon spheres are denoted as HMCS-X, where X represents the carbon shell thickness. Nitrogen physisorption measurements revealed that all resulting carbons exhibit type IV isotherms (Supplementary Fig. 2a), indicative of mesoporous structures. The specific surface areas of HMCS-18, HMCS-86, and HMCS-125 were determined to be 1915, 1803, and 1566 m² g^‒1, respectively. Pore size distributions derived from the quenched solid density functional theory method revealed similar profiles across all samples, dominated by small mesopores ranging from 3 to 5 nm (Supplementary Fig. 2b). According to the above MD simulations, we expected that such a mesopore distribution would allow for exerting a limited penetration depth for the ionomer.

Fig. 2. Catalyst synthesis and characterization.

a Schematic illustration of the synthetic route for the HMCS support. b–d HAADF-STEM images of Pt/HMCS-18, Pt/HMCS-86, and Pt/HMCS-125. e Shell thickness distributions of Pt/HMCS-18, Pt/HMCS-86, and Pt/HMCS-125. f–h Cross-sectional HAADF-STEM images of ultramicrotomed Pt/HMCS-18, Pt/HMCS-86, and Pt/HMCS-125. i Pt particle size distributions of Pt/HMCS-18, Pt/HMCS-86, and Pt/HMCS-125. Source data for this figure are provided as a Source Data file.

We synthesized Pt nanocatalysts from these mesoporous carbon supports by wet-impregnating them with H₂PtCl₆ and subsequently annealing them under a flowing reductive atmosphere, denoted as Pt/HMCS-X. The amount of H₂PtCl₆ during the wet-impregnation process was precisely managed to ensure a target Pt loading of 30 wt% in the resultant catalysts. As shown in X-ray powder diffraction (XRD) analysis (Supplementary Fig. 3a), the successful preparation of Pt nanocatalysts is evidenced by the characteristic diffraction peaks corresponding to monometallic Pt. HAADF-STEM imaging revealed a uniform dispersion of the Pt nanoparticles on the carbon supports for all prepared samples, with no discernible aggregation or excessive particle growth. Particle size distributions were obtained by measuring the diameters of over 200 nanoparticles (Fig. 2i and Supplementary Fig. 4), and the statistically derived average particle sizes were 3.30 nm for Pt/HMCS-18, 3.65 nm for Pt/HMCS-86, and 3.74 nm for Pt/HMCS-125, demonstrating good concordance with the crystalline sizes estimated from the XRD patterns (Supplementary Fig. 3b). We further used ultramicrotomy to prepare the ultrathin slices of these catalysts for investigating the spatial distribution of Pt nanoparticles on the carbon support. HAADF-STEM and energy-dispersive spectrometry (EDS) mapping images of the slices suggested that the majority of Pt nanoparticles were uniformly distributed within the carbon shell along its thickness direction in all cases (Fig. 2f–h and Supplementary Fig. 5). These results demonstrated that the three catalysts possess highly comparable features in terms of the carbon support’s pore architecture, Pt particle size, and their spatial distribution, thereby enabling a focused study on the effect of carbon shell thickness (and thus average pore depth) on fuel cell performance without interference from these parameters.

MEA performance and diagnostic analysis

We evaluated the MEA performance of these catalysts at a cathode loading of 0.1 mg_Pt cm^‒2, using differential hardware under conditions of 80 °C, 90% relative humidity (RH), and 150 kPa_abs. The H₂-air polarization curves showed that both Pt/HMCS-86 and Pt/HMCS-125 exhibited better performance at the kinetic-controlled low-current-density region compared to Pt/HMCS-18 (Fig. 3a). With increasing current density, the better performance of Pt/HMCS-86 over Pt/HMCS-18 became clearer, whereas that of Pt/HMCS-125 gradually diminished and was eventually overtaken by Pt/HMCS-18 at high current densities. A detailed comparison of voltage gaps across the current-density range, using Pt/HMCS-86 as the reference, revealed that Pt/HMCS-125 delivered the relatively high cell voltages below ~0.4 A cm^‒2, albeit with a marginal advantage of less than 7 mV (Supplementary Fig. 6). Beyond this point, the rapid voltage drop of Pt/HMCS-125 led to an intersection of its polarization curve with that of Pt/HMCS-18 at approximately 1.94 A cm^‒2. The H₂-air polarization curves corrected with the high-frequency resistance (HFR) indicated the same conclusion (Supplementary Fig. 7). These results demonstrated that the MEA performance was strongly governed by carbon shell thickness: thicker shells enhanced low-current-density performance, while thinner shells favored high-current-density performance. Consequently, Pt/HMCS-86, with its intermediate shell thickness, achieves the optimal balance across the entire current-density range.

Fig. 3. PEMFC performance and MEA diagnostic analysis of Pt/HMCS, Pt/V, and Pt/KJ300 catalysts.

a H₂-air polarization curves of Pt/HMCS-18, Pt/HMCS-86, and Pt/HMCS-125 measured at 80 °C, 90% RH, and 150 kPa_abs. Comparison of MA at 0.9 V_iR-free (b), and sulfonate coverages (c). d H₂-air polarization curves of Pt/HMCS-86, Pt/V, and Pt/KJ300 measured at 80 °C, 90% RH, and 150 kPa_abs. e Comparison of cell voltages at 0.4 A cm⁻², 1.0 A cm⁻², and 2.0 A cm^‒2. The error bars represent the standard deviation from two independently measured MEAs. Source data for this figure are provided as a Source Data file.

To gain further insight into the origin of the performance gaps, we conducted an array of MEA diagnostic measurements. The mass activities (MA) of the three catalysts, determined from H₂–O₂ Tafel plots (Supplementary Fig. 8), were 0.72 ± 0.04 A mg_Pt^‒1 for Pt/HMCS-125, 0.41 ± 0.05 A mg_Pt^‒1 for Pt/HMCS-86, and 0.15 ± 0.01 A mg_Pt^‒1 for Pt/HMCS-18 (Fig. 3b), generally reflecting the trend of their low-current-density performance observed in the H₂-air polarization below 0.4 A cm^‒2. CO stripping measurements revealed electrochemically active surface areas (ECSAs) of 61.9 ± 3.0, 61.8 ± 4.5, and 57.4 ± 9.8 m² g_Pt^‒1 for Pt/HMCS-125, Pt/HMCS-86, and Pt/HMCS-18, respectively (Supplementary Figs. 9 and 10). The comparable ECSA values make their specific activities (SA) follow the same trend observed in MA (Supplementary Fig. 11), Pt/HMCS-18 (0.25 mA cm_Pt^‒2) < Pt/HMCS-86 (0.61 mA cm_Pt^‒2) < Pt/HMCS-125 (1.17 mA cm_Pt^‒2). These results suggest that their activity differences are unlikely to arise from particle size effects. Therefore, it is reasonable to infer that the observed variations may be attributed to different degrees of Pt poisoning induced by ionomer adsorption. Supporting this hypothesis, sulfonate coverage determined by CO displacement and CO stripping measurements (Fig. 3c and Supplementary Fig. 12) showed an inverse trend relative to kinetic activity: Pt/HMCS-18 (14.1 ± 0.2%) > Pt/HMCS-86 (8.9 ± 0.5%) > Pt/HMCS-125 (2.5 ± 0.5%). Notably, the MA and sulfonate coverage values of Pt/HMCS-18 closely align with those previously reported for Pt nanocatalysts supported on solid carbon supports (MA: 0.15–0.23 A mg_Pt^‒1; sulfonate coverage: 10–20%)^27–29, suggesting that the ionomer may have fully penetrated its mesopores, thereby enabling direct contact with most of the Pt nanoparticles, akin to the behavior observed in solid carbon-supported catalysts. In contrast, the thicker carbon shells of Pt/HMCS-86 and Pt/HMCS-125 offer increased pore depth, which may hinder excessive ionomer penetration and thereby mitigate Pt poisoning. These electrode kinetic analyses provide evidence confirming the restricted penetration depth of ionomers within small mesopores, consistent with our MD simulation predictions.

To demonstrate the better performance of the optimized Pt/HMCS-86 catalyst over conventional carbon-supported catalysts, two additional Pt/C catalysts were synthesized using solid carbon (Vulcan XC-72) and porous carbon (Ketjenblack EC-300J) as supports, referred to as Pt/V and Pt/KJ300, respectively (Supplementary Fig. 13). As shown in Fig. S14, Pt/V (59.0 ± 1.4 m² g_Pt^‒1) and Pt/KJ300 (53.5 ± 2.1 m² g_Pt^‒1) displayed comparable ECSA with Pt/HMCS-86 (61.8 ± 4.5 m² g_Pt^‒1). The H₂-air polarization curves suggest that Pt/KJ300 outperformed Pt/V at low current densities, while the trend reversed at higher current densities (Fig. 3d), consistent with previously reported findings¹⁹. In contrast, Pt/HMCS-86 consistently exhibited the relatively high cell voltages across the entire current-density range, confirming its improved MEA performance. Consistent performance trends were observed in the HFR-free polarization curves (Supplementary Fig. 15). The cell voltages extracted at low (0.4 A cm^‒2), medium (1 A cm^‒2), and high current densities (2 A cm^‒2) reveal that Pt/HMCS-86 maintains the relatively high voltage levels with the least voltage decay among all samples (Fig. 3e and Supplementary Table 2), which stands as consistent evidence that it has achieved an optimal balance between activity and transport.

To reveal the effect of pore depth on mass transport, we measured the total oxygen transport resistance (${\mathrm{R}}_{\mathrm{Total}}^{{\mathrm{O}}_2}$) using the limiting current method. A broad range of dry oxygen concentrations was employed to assess the variation in transport resistance from the low current density to the high-current-density region. The Pt/HMCS-125 catalyst exhibited the highest ${\mathrm{R}}_{\mathrm{Total}}^{{\mathrm{O}}_2}$ over the entire limiting current-density range, with Pt/HMCS-86 and Pt/HMCS-18 showing progressively lower values (Supplementary Fig. 16), indicating a positive correlation between oxygen transport resistance and carbon shell thickness. In the low-current-density region (below approximately 0.2–0.4 A cm^‒2), all catalysts showed nearly constant ${\mathrm{R}}_{\mathrm{Total}}^{{\mathrm{O}}_2}$ values. Beyond this range, the ${\mathrm{R}}_{\mathrm{Total}}^{{\mathrm{O}}_2}$ began to increase gradually. This phenomenon, consistent with previous reports, has been attributed to limited water condensation at low current densities, which creates a relatively dry condition favorable for oxygen transport³⁰. In contrast, increased water condensation at higher current densities leads to a relatively wet condition, hindering oxygen transport. Notably, during the transition from low to high current densities, Pt/HMCS-125 showed a clear increase in ${\mathrm{R}}_{\mathrm{Total}}^{{\mathrm{O}}_2}$, followed by Pt/HMCS-86 and Pt/HMCS-18, suggesting that thicker carbon shells promote water condensation in the cathode.

We further extracted the pressure-independent resistance (${\mathrm{R}}_{\mathrm{PI}}^{{\mathrm{O}}_2}$) from the total oxygen transport resistance by varying the outlet pressure (Figs. 4a and S17). Existing literature indicates that ${\mathrm{R}}_{\mathrm{PI}}^{{\mathrm{O}}_2}$ is primarily governed by local transport resistance, especially under low-Pt loading conditions^2,5,14. Unlike solid carbon catalysts, where local transport resistance is mainly associated with oxygen diffusion through the Pt/ionomer interface, porous carbon catalysts have an additional source of local transport resistance: the transport of oxygen through the carbon nanopores to reach the interior Pt nanoparticles³¹. As illustrated in Fig. 4a, ${\mathrm{R}}_{\mathrm{PI}}^{{\mathrm{O}}_2}$ follows the trend observed in the total resistance, including a nearly constant resistance under low current densities and an evident increase once the current density exceeds a certain threshold, particularly in the case of Pt/HMCS-125. Based on these results, it is reasonable to infer that the aggravated water condensation in the catalysts with thicker carbon shells likely originates from increased pore depth, which extends the diffusion path for water vapor from the carbon interior to the external surface, thereby hindering its removal. As a consequence, water condensation and accumulation occurred more readily in thicker mesoporous carbon shells at elevated current densities, leading to a significant increase in local oxygen transport resistance in the high-current-density region, as evidenced by the H₂-air polarization curve of Pt/HMCS-125 showing a large voltage loss from 1.5 to 2 A cm^‒2 (Fig. 3a). Furthermore, even under low current densities where water condensation is negligible, the ${\mathrm{R}}_{\mathrm{PI}}^{{\mathrm{O}}_2}$ still increased with carbon shell thickness (Fig. 4a). This means that greater pore depth can also impede local oxygen transport under dry conditions, though to a significantly lesser extent than under wet conditions. It is important to note that all catalysts were evaluated using the same ionomer-to-carbon weight ratio (I/C = 0.8) in the aforementioned tests. Since the I/C ratio is a well-recognized factor influencing oxygen transport properties, and the optimal value may differ among catalysts^32,33, we further examined the polarization behavior across a range of I/C ratios. Under the respective optimal I/C conditions for each catalyst, the relative performance trends remained consistent with those observed at the fixed ratio of 0.8 (Supplementary Fig. 18).

Fig. 4. Analysis of oxygen transport in the cathode catalyst layer.

a ${\mathrm{R}}_{\mathrm{PI}}^{{\mathrm{O}}_2}$ versus the limiting current density. b Oxygen diffusion coefficients in the mesopore with different water-filling rates. Insets show snapshots of oxygen diffusion in the mesopore, where gray, red, and white balls represent carbon, oxygen, and hydrogen atoms, respectively. Error bars represent standard deviation from four independent MD simulations. Source data for this figure are provided as a Source Data file.

We performed MD simulations to study the impact of water condensation on oxygen transport in mesopores. To evaluate oxygen transport efficiency under various water condensation levels, a series of models was constructed by introducing a constant number of oxygen molecules into a 5 nm mesopore containing different water-filling rates (α). As shown in Fig. 4b and Supplementary Table 1, the oxygen self-diffusion coefficient, $D_{\mathrm{O}2}$ (in 10^‒9 m² s^‒1), evolves through three stages with increasing α. Initially, for α in the range of 0.1–0.5, the $D_{O2}$ decreases mildly from 131.13 to 103.89. This is followed by an order-of-magnitude drop to 15.03 as α increases to 0.8, after which the coefficient stabilizes at a plateau of ~9 for α > 0.8. This trend reflects a shift in the transport mechanism, as indicated by the MD snapshots (Fig. 4b). Under low water-filling conditions, the pore retains abundant vacancy networks where the oxygen molecules predominantly localize, enabling Knudsen-like diffusion along connected pathways³⁴. As the pore becomes increasingly filled with water, the role of water-phase diffusion grows, which in turn drastically reduces the diffusion coefficient. Ultimately, near saturation, the vacancy network vanishes, and the transport is entirely dictated by diffusion within the water phase³⁵. The current-density-dependent rise in ${\mathrm{R}}_{\mathrm{PI}}^{{\mathrm{O}}_2}$ (Fig. 4a), together with the H₂-air polarization decay (Fig. 3a, e) and the sharp drop in oxygen diffusivity at elevated water-filling ratios in MD simulations (Fig. 4b), collectively support the conclusion that increased pore depth exacerbates water accumulation and impairs oxygen transport at high current densities.

Based on the above findings, we can conclude that pore depth exerts opposing influences on kinetic activity and local oxygen transport. Due to the limited penetration depth of ionomer into the small mesopores (<5 nm), increasing pore depth enhances kinetic activity by mitigating ionomer-induced Pt poisoning, yet concurrently hinders local transport by extending the diffusion pathway. Nevertheless, a significant decline in high-current-density performance occurs only when the pore depth exceeds a certain extent, making the removal of ORR-produced water difficult. At this point, severe water condensation and accumulation lead to a transition of the oxygen transport mechanism from efficient gas-phase diffusion to sluggish liquid-phase diffusion. Below this extent, however, increasing pore depth primarily brings the benefit of improved activity without imposing a significant penalty on local oxygen transport. As a result, controlling the pore depth to a rational medium level enables the formation of a desirable local reaction environment for the majority of Pt active sites, achieving a good balance between kinetic activity and oxygen transport.

Developing high-performance PtCo catalysts

We selected the intermetallic PtCo phase, one of the most extensively studied Pt-based alloy catalysts for ORR³⁶, to validate whether HMCS-86 can serve as an effective support for enhancing the MEA performance of alloy catalysts. The synthesis of this catalyst was conducted using our previously developed sulfur-containing molecule-assisted method³⁷. The Pt content in the resulting catalyst (denoted as PtCo/HMCS-86) was controlled to be 30 wt%. Supplementary Fig. 19 presents the XRD pattern, in which two superlattice peaks at 24.0° and 33.3° were observed, which belong to the (001) and (110) planes of the face-centered tetragonal PtCo intermetallic compound, confirming the successful synthesis of the targeted phase. HAADF-STEM imaging revealed a uniform particle size distribution, with an average size of 5.3 nm (Supplementary Fig. 20). Before MEA testing, acid leaching was conducted to obtain an active and stable Pt shell/intermetallic PtCo core structure, as supported by aberration-corrected HAADF-STEM (Fig. S21). Energy-dispersive spectroscopy (EDS) mapping suggested a Pt/Co atomic ratio of ~2.4 for nanoparticles in the acid-leached catalyst (Supplementary Fig. 22 and Supplementary Table 3).

The MEA performance of PtCo/HMCS-86 was first examined under 80 °C and 150 kPa_abs and benchmarked against a commercial Pt/C catalyst (Tanaka Kikinzoku Kogyo TEC10E30E, denoted as TKK-Pt/C). The cathode Pt loading was set to 0.1 mg_Pt cm^‒2 for both catalysts. In H₂–O₂ testing (Supplementary Fig. 23), PtCo/HMCS-86 achieved a MA of 1.1 A mg_Pt^‒1, higher than that of TKK-Pt/C (0.41 A mg_Pt^‒1). As shown in Fig. 5a, its H₂‑air performance was competitive with that of the commercial benchmark, delivering current densities of 0.63 and 1.81 A cm^‒2 at 0.8 and 0.65 V, respectively, approximately 2.5 and 1.4 times greater than those of TKK-Pt/C (0.29 and 1.30 A cm^‒2). After 30k-cycle accelerated stress testing (AST) under the DOE-recommended square-wave protocol (0.6–0.95 V), PtCo/HMCS-86 retained a better performance over TKK-Pt/C, demonstrating favorable durability. Notably, its MA loss and voltage loss at 0.8 A cm^‒2 were 16% and 29 mV (Fig. 5b, c), both within the DOE durability targets for light-duty transportation (MA loss <40% and voltage loss <30 mV at 0.8 A cm^‒2). By contrast, TKK-Pt/C suffered a ~ 110 mV voltage drop at 0.8 A cm^‒2 and a 47% MA loss. We further tested a commercial PtCo/C catalyst (Tanaka Kikinzoku Kogyo TEC36E52, denoted as TKK-PtCo/C) under identical conditions and confirmed that PtCo/HMCS-86 again delivers favorable performance across the entire current-density range (Supplementary Fig. 24). In addition, to emphasize that the HMCS-86 support itself plays an independent and important role in improving MEA performance, we prepared PtCo nanoparticles supported on a commercial porous carbon (KJ600) as a reference catalyst (PtCo/KJ600; Supplementary Fig. 25), which exhibited lower performance than PtCo/HMCS-86.

Fig. 5. PEMFC performance of the HMCS-86-supported PtCo catalyst.

a H₂-air fuel cell polarization curves of PtCo/HMCS-86 and TKK-Pt/C before and after the 30k AST, measured at 80 °C, 90% RH, and 150 kPa_abs. b MA changes before and after the 30k AST. c Cell voltage variations at 0.8 A cm^‒2 before and after the 30k AST. d H₂-air polarization curves of PtCo/HMCS-86 and Pt/HMCS-86 measured at 94 °C, 65% RH, and 250 kPa_abs. Source data for this figure are provided as a Source Data file.

We further conducted H₂-air tests at 94 °C, 65% RH, and 250 kPa_abs to evaluate the MEA performance of the developed catalyst under the DOE-set heat-rejection-constrained (Q/ΔT < 1.45 kW °C^‒1). Analysis of the polarization curve revealed that PtCo/HMCS-86 achieved a rated power density of 1.23 W cm^‒2 at 0.67 V and a current density of 450 mA cm^‒2 at 0.8 V (Fig. 5d). As summarized in Table S4, the MA and rated power density of PtCo/HMCS-86 are competitive with those of most reported Pt alloy catalysts. Notably, under the same test conditions, the monometallic Pt catalyst (Pt/HMCS-86) delivered a power density of 1.0 W cm^‒2 at 0.67 V (Fig. 5d), which is comparable to that of previously reported Pt alloy catalysts (Supplementary Table 4).

Discussion

In summary, this work establishes a simple but effective carbon support design to enhance the overall MEA performance of cathode catalysts, which hinges on tailoring a small mesopore-dominated structure with moderate pore depth control. Through MEA diagnostic analyses and molecular dynamics simulations, we identified that the efficacy of this pore geometry stems from its ability to create a favorable local reaction environment, which both effectively protects most Pt active sites from ionomer adsorption and ensures the timely removal of ORR-produced water, thereby enabling oxygen transport to proceed via an efficient gas-phase diffusion mechanism. We believe these findings provide meaningful insights and valuable guidance for the development of next-generation high-performance cathode catalysts through carbon support engineering.

Methods

Chemicals and materials

Ammonium hydroxide (NH₃·H₂O; Sinopharm, 25 wt% in water), tetraethyl orthosilicate (TEOS; Macklin, 98% purity), octadecyltrimethoxysilane (OTMS; Macklin, 95% purity), ferrocene (Aladdin, 99% purity), hydrofluoric acid (HF; Sinopharm, 40 wt% in water), hydrochloric acid (HCl; Sinopharm, 30 wt% in water), hexahydrated chloroplatinic acid (H₂PtCl₆·6H₂O; Sinopharm, 99% purity), hexahydrated cobalt chloride (CoCl₂·6H₂O; Sinopharm, ≥99% purity), sodium thioglycolate (Macklin, ≥99% purity). All reagents were used without further purification.

Preparation of catalysts

Synthesis of SiO₂@mSiO₂ Templates. The synthesis of SiO₂@mSiO₂ templates was performed using a modified Stöber process²⁴. For the template used in HMCS-18 synthesis, a solution containing 175 mL of ethanol, 32.5 mL of deionized water, and 3.8 mL of NH₃·H₂O was first mixed under stirring for 5 min. Then, 6.95 mL of TEOS was added under vigorous stirring. After 12 h, a mixture of 4.95 mL of TEOS and 1.95 mL of OTMS was introduced into the white suspension. The reaction was continued overnight under stirring. The resulting product was collected by centrifugation, washed, and dried at 65 °C overnight. The dried powder was calcined in air at 550 °C for 3 h with a heating rate of 1 °C min^‒1. The calcined powder was ground with a mortar and pestle to obtain the final template. Templates for HMCS-86 and HMCS-125 were synthesized following the same procedure, except that the amounts of TEOS/OTMS were adjusted to 9.9/3.9 mL and 14.85/5.85 mL, respectively.

Synthesis of HMCS carbon supports. HMCS carbon supports were prepared by CVD of ferrocene on the SiO₂@mSiO₂ template^25,26. Typically, 2 g of the SiO₂@mSiO₂ template and 10 g of ferrocene were placed in separate quartz crucibles and loaded into a quartz tube installed in a tubular furnace. The template crucible was positioned at the center of the furnace heating zone, while the ferrocene crucible was placed near the tube inlet and heated using an external jacket. The tubular furnace was heated to 500 °C at a rate of 5 °C min^‒1 under an argon flow of 100 mL·min^‒1, after which the heating jacket was rapidly heated to 150 °C to sublimate the ferrocene. The tubular furnace was then further heated to 550 °C at a rate of 5 °C min^‒1 and maintained for 2 h. After the reaction, both heating systems were turned off, and the product was allowed to cool naturally to 25 °C under continuous argon flow. The resulting material was passivated in air for over 2 h before further handling and then annealed at 800 °C for 4 h under an argon flow of 200 mL min^‒1 (heating rate: 5 °C min^‒1). To remove the silica template, 2 g of the composite was treated with 20 mL of HF for 12 h, followed by repeated washing with deionized water. Residual iron species were further removed by acid leaching with excess HCl (7 mL per gram of the carbon product) under overnight stirring. The acid-leached product was then repeatedly washed with deionized water, dried, and subsequently thermally treated at 1000 °C for 1 h under an argon flow (200 mL min^‒1) to obtain the HMCS. The reproducibility of the HMCS synthesis was validated by preparing two independent batches, which exhibited comparable shell thicknesses and pore size distributions (Supplementary Fig. 26).

Synthesis of the Pt/HMCS, PtCo/HMCS, and PtCo/KJ600 catalysts. For the Pt/HMCS catalysts, 30 mg of HMCS was dispersed in 50 mL of deionized water and sonicated for 2 h. Subsequently, 9.5 mL of H₂PtCl₆·6H₂O aqueous solution (1.5 mg_Pt mL^‒1) was added, and the mixture was sonicated for another 2 h. Afterward, the mixture was stirred overnight to allow sufficient time for capillary action to draw the precursor solution deep into the mesopores, ensuring homogeneous infiltration. Then, the solvent was removed by slow rotary evaporation to prevent rapid precipitation at the pore mouths and promote uniform deposition throughout the shell thickness. The resulting powder was annealed at high temperature under a flow of 200 mL min^‒1, 5 vol% H₂/Ar mixed gas. The temperature was ramped at 5 °C min^‒1 to 900 °C, held for 2 h, and then naturally cooled to 25 °C. For the PtCo/HMCS-86 catalyst, 100 mg HMCS-86, 87 mg H₂PtCl₆·6H₂O, 51.9 mg CoCl₂·6H₂O, and 29 mg sodium thioglycolate were added to 50 mL of deionized water. The mixture was sonicated for 2 h and stirred overnight to obtain a uniform dispersion. The resulting mixture was concentrated by rotary evaporation, annealed at 1100 °C for 1 h under a 95% Ar/5% H₂ atmosphere with a heating rate of 5 °C min^‒1, and then allowed to cool naturally to 25 °C. Before the electrochemical test, the as-prepared PtCo/HMCS catalyst was initially oxidized in air at 250 °C for 6 h, then etched with 0.1 M HClO₄ in an oil bath at 60 °C, followed by annealing at 400 °C for 2 h under Ar/H₂ to form the stable Pt shell/PtCo core structure. The PtCo/KJ600 catalyst was prepared using the same synthetic protocol as that employed for PtCo/HMCS‑86.

Characterization

XRD patterns were recorded on a Rigaku DMax-γA rotating anode diffractometer with Cu Kα radiation (λ = 1.54056 Å) operating at 40 kV and 150 mA. Atomic-resolution high-angle annular dark-field scanning transmission electron microscopy (HAADF-STEM) images were obtained on a JEM-ARM 200 F microscope operating at 200 kV. Low-magnification HAADF-STEM imaging and energy-dispersive X-ray spectroscopy mapping were conducted on an FEI Talos F200X microscope equipped with a Super X-EDS system, operated at 200 kV. Particle and shell sizes of HMCS were statistically analyzed from HAADF-STEM images by sampling over 200 particles. SEM images were obtained using a HITACHI SU8220 Cold Field Emission Scanning Electron Microscope at 10.0 kV and 0.8 nm. Nitrogen sorption isotherms were measured at 77 K on a Quantachrome Autosorb-iQ analyzer.

MEAs fabrication

Cathode catalyst inks were made by mixing the in-house catalysts with an ionomer dispersion in a mixed solvent of deionized water and isopropanol in a 1:1 v/v ratio. The total solvent volume was approximately 20 mL per 40 mg of catalyst. The I/C ratio was screened from 0.6 to 1.0 for all HMCS catalysts, while fixed values of 0.6 (for Pt/V) and 0.8 (for TKK-Pt/C, TKK-PtCo/C, and PtCo/KJ600) were used for the catalysts. The mixture was sonicated for 2 h in an ice-water bath using an ultrasonic homogenizer operating at 48 kHz, with temperature control to prevent solvent evaporation and ionomer degradation. The anode catalyst ink was prepared using TKK-Pt/C, following the same procedure as for the cathode ink. The catalyst-coated membranes (CCMs) with an active area of 5 cm² were fabricated on a Nafion membrane (GORE, ~12 μm, used as received without further treatment) using an ultrasonic spray coater operating at 48 kHz and 95 °C (SonoTek). The Nafion membrane used for CCM preparation was cut to ~7 cm² and sandwiched between two sub-gaskets (PEN adhesive film, ~40 μm each) with a central square cut-out of 5 cm², thereby defining an effective active area of 5 cm² and limiting the non-active area to less than 2 cm². The Pt loadings of all CCMs were controlled at 0.05 and 0.1 mg_Pt cm^‒2 for the anode and cathode, respectively. The prepared CCMs were assembled into a single-cell fixture together with two gas diffusion layers (Sigracet 22BB, ~215 μm) and two gaskets (PTFE, 150 μm). Depending on the catalyst layer thickness, the compression of the gas diffusion layers ranged from 23% to 26%. The hardware bolts were torqued to 10 Nm to ensure proper sealing of the single cell. Both the anode and cathode employed a seven-channel serpentine flow field design, featuring negligible pressure drops between the inlet and outlet³⁸.

Fuel cell testing

All PEMFC tests were conducted using a Scribner 850e fuel cell test stand linked with a Scribner 885 potentiostat. Prior to performance measurement, an activation protocol was conducted, consisting of a break-in procedure followed by a voltage recovery (VR) step. The break-in procedure entailed repeated voltage sweeps between the open-circuit voltage and 0.30 V in H₂–O₂ and H₂-air environments at 80 °C, 100% RH, and 150 kPa_abs, until the cell performance reached a steady state. VR was subsequently performed under 40 °C, 150% RH, and 150 kPa_abs, following a sequence of 5 min at 0.85 V, 10 min at 0.6 V, and 1 h at 0.1V¹⁷. The VR procedure was repeated until no further increase in current density was observed at 0.85 V and 0.6 V.

The H₂-air polarization curve measurement was performed in the anodic direction using the constant current mode under the conditions of 80 °C, 100% RH, and 150 kPa_abs. Each data point was collected over a 3-min interval and subsequently averaged for the final 1-min period. The flow rates of H₂ and air were maintained at 0.5 and 2 L min^‒1, respectively.

The testing procedure for MA included recording H₂–O₂ polarization curves from 0.775 V to 0.9 V at the conditions of 80 °C, 150 kPa_abs, and 100% RH, wherein the gas flow rates of H₂ and O₂ were both 0.2 L min^‒1. Each measurement point was maintained for 4 min, with the average of the final minute being used, and the measurements were taken in the anodic direction. Corrections for high-frequency resistance and hydrogen crossover were applied to the Tafel plots, which were then utilized to calculate the MA at 0.9 V³⁹.

ECSA was measured by the CO stripping method at the conditions of 80 °C, 150 kPa_abs, and 100% RH. A gas mixture containing 1% CO in N₂ was first introduced to the cathode at a flow rate of 0.2 L min^‒1 for at least 20 min, while maintaining the cathode potential at 0.1 V. Subsequently, the cathode was purged with pure N₂ at 0.5 L min^‒1 for at least 30 min to remove CO. Two cyclic voltammetry (CV) curves were then recorded sequentially using a Scribner 885 potentiostat by sweeping the potential from open-circuit voltage to 1.2 V at a scan rate of 100 mV s^‒1, with 5% H₂ in N₂ (0.2 L min^‒1) and N₂ (0.005 L min^‒1) flow at anode and cathode, respectively. The area between the first and the second anodic scans was used to determine the CO stripping charge, which was used with an area-specific charge of 420 μC cm^‒2 to calculate the ECSA.

Sulfonate coverage was obtained at 40 °C under ambient pressure and 100% RH via a two‑step procedure. First, the cathode was cleaned by cyclic voltammetry scanning to 0.9 V at 100 mV s^‒1, with 5% H₂ in N₂ (0.2 L min^‒1) and N₂ (0.2 L min^‒1) flow at anode and cathode, respectively. Second, the cathode was held at 0.4 V until a steady current was reached, then the cathode gas was switched to 1% CO/N₂ (0.2 L min^‒1) to record the CO displacement curve. CO stripping was subsequently performed as described earlier for the stripping measurement. The sulfonate coverage was calculated by comparing the displacement charge with the stripping charge^29,40.

The total oxygen transport resistance(${\mathrm{R}}_{\mathrm{Total}}^{{\mathrm{O}}_2}$) was determined by the limiting current method. Linear sweep voltammetry (LSV) was performed with a Gamry 5000E potentiostat at 80 °C, 150 kPa_abs, and 70% RH, with flow rates of 2.0 L min^‒1 (H₂) at the anode and 5.0 L min^‒1 (O₂/N₂ mixture) at the cathode. By varying the dry O₂ mole fraction (0.5%, 0.7%, 1%, 1.5%, 2%, 2.5%, 5.0%, 7.5%, 10.0% and 12.5% O₂ in N₂), a series of LSV curves was obtained. The limiting currents extracted from these curves were linearly regressed against the dry O₂ mole fractions, and the resulting slope corresponds to the ${\mathrm{R}}_{\mathrm{Total}}^{{\mathrm{O}}_2}$. To extract the pressure-independent component (${\mathrm{R}}_{\mathrm{PI}}^{{\mathrm{O}}_2}$) from the ${\mathrm{R}}_{\mathrm{Total}}^{{\mathrm{O}}_2}$, the above LSV measurements were performed at different outlet pressures (100, 150, 200, and 250 kPa_abs). A linear fit of ${\mathrm{R}}_{\mathrm{Total}}^{{\mathrm{O}}_2}$ versus outlet pressure was performed, and the intercept, after correction according to the method reported by Baker et al.⁴¹, was taken as ${\mathrm{R}}_{\mathrm{PI}}^{{\mathrm{O}}_2}$.

Catalyst durability was evaluated following the DOE-recommended AST protocol, consisting of 30,000 square-wave potential cycles between 0.6 and 0.95 V with a 3 s hold at each step (80 °C, 100% RH, ambient pressure, H₂–N₂ flows of 0.2/0.075 L min^‒1).

Molecular dynamics simulation

Ionomer penetration. To simulate the ionomer penetration process, the simulation model consists of the pore model (d = 1, 3, 5 nm), a 25-mer ionomer chain, water molecules, and 25 hydronium ions. The mesopore model is inspired by Kikkawa’s work²¹, and the mesopore depth is 5 nm at d = 1/3 nm and 10 nm at d = 5 nm, respectively. The general AMBER force field (GAFF)⁴² was employed for the mesopore and ionomer model, while MSPC/E⁴³ and Ding’s model⁴⁴ were used for water and hydronium ions, respectively. The charges of ionomer atoms were determined using the AM1-BCC method^45,46. The Lennard-Jones (LJ) parameters for sp² aromatic carbon were used to describe the van der Waals interactions among carbon atoms in the mesopore model. The hydrogen bonds were constrained using the LINCS method⁴⁷. The Coulomb interactions were calculated using the PME method⁴⁸ with 1 nm cutoff. All MD simulations were performed by using the GROMACS package (timestep: 0.2 fs; temperature: 300 K)⁴⁹. Other settings were the standard settings of GROMACS. The initial and final configurations of the MD simulations are provided in PDB format in Supplementary Data 1.

The mesopore and bulk hydrated ionomer (λ = 14) were equilibrated for 1-ns NVT simulations. The ionomer was then positioned 0.5 nm apart from the mesopore model, with water molecules randomly added in the simulation box to reach 1 g cm^‒3 density for modeling a dilute ionomer system. The system was equilibrated for 100-ps NVT and 1-ns NPT simulations. In all NPT simulations, the Parrinello−Rahman barostat⁵⁰ with a 10 ps relaxation parameter and 4.5 × 10⁻⁵ bar^‒1 compressibility was used, controlling only the z-axis in the barostat at 1 bar target pressure.

The penetration depth (Z) was calculated by applying Eq. (1)

$$Z=Z_0-Z_i$$ 1

where $Z_0$ is the z-coordinate of the center of mass (COM) of the graphene layer (excluding CNT atoms), and $Z_i$ is the z-coordinate of the COM of the $i$-th monomer. Given the slow structural relaxation of ionomers (diffusion coefficient≈10⁻¹¹ m² s^‒1)⁵¹, the replica-exchange umbrella sampling (REUS) method⁵² with a harmonic bias potential (Eq. (2)) was employed to avoid local-minimum free-energy traps and enhance conformational space sampling within limited MD simulation time.

$$U^{\mathrm{bias}}=\frac{k\left(t\right)}{2}\left(Z-Z_j^0\right)$$ 2

where

$$k\left(t\right)=\left\{\begin{array}{c}\hfill \frac{500t}{1\mathrm{ns}}\mathrm{kJ}/{\mathrm{nm}}^2,\mathrm{NVT\; simulation}\hfill \\ \hfill 500\mathrm{kJ}/{\mathrm{nm}}^2,\mathrm{NPT\; simulation}\hfill \end{array}\right)$$

and

$$Z_j^0=\left(0.1j-1\right)\mathrm{nm}$$

for the (j + 1)-th replica (j = 0, 1, 2, …47 for d = 1, 3 nm; j = 0, 1, 2, …80 for d = 5 nm).

First, 48 or 81 replicas of each initial structure were duplicated and then relaxed for 1-ns NVT and 4-ns NPT simulations under the biased potentials. Further, 10-ns NPT simulations were performed for sampling. The PLUMED package⁵³ was used for the REUS calculations, and free-energy profiles were calculated using the weighted histogram analysis method (WHAM)⁵⁴. The radial distribution function (RDF) between type A and B particles was calculated by applying Eq. (3)

$$g_{AB}\left(r\right)=\left(\frac{n_B}{4\pi r^{2}dr}\right)\left./\right)\left(\frac{N_B}{V}\right)$$ 3

where $n_B$ is the number of B particles located at a distance $r$ from particle A in a shell of thickness $dr$, $N_B$ is the number of B particles in the system, and $V$ is the total volume of the system. The average values of the $g_{\mathrm{Ionomer}-\mathrm{C}}\left(r\right)$ at $r$ ≤ 0.5 nm ($\overline{g_{\mathrm{Ionomer}-\mathrm{C}}}\left(r\le 0.5\mathrm{nm}\right)$) across replicas were calculated to evaluate the interaction strength between hydrophobic ionomer backbones and carbon support surface at different penetration depths. The decrease of hydrogen bonds within the pore ($△N_{\mathrm{Hbond}}$) was calculated by applying Eq. (4)

$${△N}_{Hbond}=N_{Hbond,Z=0}-N_{Hbond,Z}$$ 4

where $N_{\mathrm{Hbond},Z=0}$ and $N_{\mathrm{Hbond},Z}$ are the number of hydrogen bonds within the pore at Z = 0 and varying Z, respectively. Four independent calculations were performed for each pore diameter using different initial structures and random seeds, and the average value was reported.

Oxygen diffusion. To simulate oxygen diffusion in a mesopore, the simulation model consists of a three-walled armchair carbon nanotube (d = 5 nm, h = 5 nm) as the pore model, water molecules, and 20 oxygen molecules. The water-filling rate (α, α = 0.1, 0.3, 0.5, 0.8, 1.0, 1.1) was determined by applying Eq. (5)

$$\alpha =N_{{\mathrm{H}}_2\mathrm{O}}/N_{{\mathrm{H}}_2\mathrm{O}}$$ 5

where $N_{{\mathrm{H}}_2\mathrm{O}}$ and $N_{{\mathrm{H}}_2\mathrm{O}}$ are the number of water molecules in the pore at varying $\alpha$ and at 1 g cm^‒3, respectively.

All simulation settings remained identical to those in the preceding section except for the temperature (353 K). The water molecules and oxygen molecules were randomly positioned in the pore, and the whole model was relaxed for a 5-ns NVT simulation. Further, a 100-ns NVT simulation was performed to obtain the Z axial diffusion coefficient of oxygen molecules by applying Eq. (6)

$$D={\lim }_{t\to \infty }\frac{1}{6t}⟨{\left(r\left(t\right)-r\left(0\right)\right)}^2⟩$$ 6

where r(t) and r(0) are the positions of the target molecules at times t > 0 and t = 0, respectively. The MSD calculation was performed in the range 0 ≤ t ≤ 5 ns, much smaller than the total simulation time, to guarantee the sampling accuracy. Four independent calculations were performed for each $\alpha$ using different initial structures and random seeds, and the average value was reported.

Author contributions

H.-W.L., L.T. and C.-S.M. conceived and designed the experiments. L.Z. and K.C. designed the theoretical simulation. C.-S.M. and Z.-R.L. conducted the catalyst synthesis, characterization, and PEMFC tests. K.C. conducted the molecular dynamics simulation. S.L., C.X., K.-Z.X., X.-C.X., R.-Y.S. and T.-W.S. provided support for the PEMFC tests. Y.-W.D. carried out the nitrogen physisorption measurements. L.T., C.-S.M., K.C., L.Z., J.L. and H.-W.L. cowrote the manuscript. All authors discussed the results and contributed to the manuscript revision.

Peer review

Peer review information

Nature Communications thanks Ziyun Wang and the other anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Funding

H.-W.L. discloses support for the research of this work from the National Natural Science Foundation of China (Grant Nos. 22325903, 22221003) and the Anhui Province Ecological Environment Science and Technology Project (Grant No. 2024hb014). L.T. discloses support for the research of this work from the National Natural Science Foundation of China (Grant No. 22509195) and the Fundamental Research Funds for the Central Universities (Grant No. WK9990250181). L.Z. discloses support for the research of this work from the National Natural Science Foundation of China (Grant No. 22373055, 22521202) and the Program of Beijing Huairou Laboratory (Grant No. ZD2022006A).

Data availability

All data supporting the findings of this study are available within the article and its Supplementary Information. Source data are provided with this paper.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-73971-8.