Efficient acidic hydrogen evolution in proton exchange membrane electrolyzers over a sulfur-doped marcasite-type electrocatalyst

Abstract

Large-scale deployment of proton exchange membrane (PEM) water electrolyzers has to overcome a cost barrier resulting from the exclusive adoption of platinum group metal (PGM) catalysts. Ideally, carbon-supported platinum used at cathode should be replaced with PGM-free catalysts, but they often undergo insufficient activity and stability subjecting to corrosive acidic conditions. Inspired by marcasite existed under acidic environments in nature, we report a sulfur doping–driven structural transformation from pyrite-type cobalt diselenide to pure marcasite counterpart. The resultant catalyst drives hydrogen evolution reaction with low overpotential of 67 millivolts at 10 milliamperes per square centimeter and exhibits no degradation after 1000 hours of testing in acid. Moreover, a PEM electrolyzer with this catalyst as cathode runs stably over 410 hours at 1 ampere per square centimeter and 60°C. The marked properties arise from sulfur doping that not only triggers formation of acid-resistant marcasite structure but also tailors electronic states (e.g., work function) for improved hydrogen diffusion and electrocatalysis.

A marcasite-type cobalt diselenide exhibits promise as an efficient cathode in proton exchange membrane water electrolyzers.

INTRODUCTION

Electrochemical splitting of water into green hydrogen (H₂) fuels, especially driven by renewable electricity, offers an elegant path toward carbon-neutral energy society (1, 2). Since proposed in 1789, alkaline water electrolysis has been progressively developed as a matured technology for industrial H₂ production, but its limited current density (high ohmic resistance), low partial load range, and low operating pressure are often drawbacks (3, 4). Proton exchange membrane (PEM) water electrolysis that relies on proton transfers can effectively surmount these issues in alkali, but the corrosive acidic environments require the use of expensive platinum group metal (PGM) catalysts, raising the stack cost (4–6). In present-day PEM electrolyzers, carbon-supported platinum (Pt/C) remains the catalyst of choice for the cathodic hydrogen evolution reaction (HER). Thanks to the fast HER kinetics in acid, the Pt loading at the cathode is typical 0.5 to 1.0 mg_Pt cm⁻² per PEM electrolyzer (5), whose cost is lower than that of IrO_x catalyst (~2 mg_Ir cm⁻²) used at the anode (4); thus, less previous research efforts have been made on the cost reduction at the cathode. While the Pt/C catalyst contributes small to the system cost nowadays, future large-scale H₂ production at terawatt range would demand a substantial amount of Pt, which is not sustainable and hampers commercialization of industrial-grade PEM electrolyzers (6, 7).

To enable widespread prevalence of PEM electrolyzers, the Pt loading at the cathode should be substantially reduced or ideally the Pt catalyst should be replaced with PGM-free materials (6, 7). Over the past decade, PGM-free catalysts for the acidic HER have been well explored (6, 8–21), exemplified by the molybdenum disulfide (MoS₂) catalyst inspired by nitrogenase enzyme (8–10). Subsequently, numerous transition metal dichalcogenides (11–15) and phosphides (16–20) were designed and synthesized to display decent activities in rotating disk electrode (RDE) measurements. Recently, King et al. (17) have demonstrated a very promising PEM electrolyzer using cobalt phosphide (CoP) as the HER catalyst, illustrating the practical relevance of PGM-free cathode (17). Despite big advances, under acidic environments, PGM-free materials generally show high propensity to chemical and structural changes, and even component dissolution, leading to decreased catalytic performances (4–6). To date, the design of PGM-free HER catalysts with exceptional activity and stability, particularly under realistic PEM electrolysis condition, remains a daunting challenge.

Besides MoS₂, transition metals and chalcogens can form many other important groups of minerals in nature, for example, pyrite (11, 12, 22, 23) and marcasite (24, 25). We (22) and others (11, 23) have reported pyrite-type cobalt diselenide (CoSe₂) to be promising catalyst for a variety of reactions with notable performances. Furthermore, we recently found that partial transformation of pyrite CoSe₂ into marcasite structure enables enhanced HER stability in acidic electrolyte (13). Unfortunately, this pyrite-marcasite mixed CoSe₂ catalyst still falls short of the requirement of practical PEM electrolysis. Yet, in nature, marcasite minerals often form under very acidic environments (26); moreover, the dissolution rate was experimentally observed to be more than 10-fold lower for marcasite than for pyrite (27). We therefore reason that marcasite-type CoSe₂ might show the potential as cathode for PEM electrolyzers, given its corrosion-resistant feature in acid.

Here, we report a complete structural transformation from pyrite- to marcasite-type CoSe₂ driven by sulfur doping. Not only creating marcasite structure with improved acid stability, but also the doping effect caused by partial substitution of Se with S also tailors electronic structures of the catalyst for more favorable hydrogen diffusion in the electrical double layer (EDL) and adsorption on the catalytic surface. The resultant sulfur-doped marcasite CoSe₂ (M-CoSe₂) as cathode was demonstrated to be active and stable in a practical PEM electrolyzer, highlighting potential of PGM-free catalysts for commercial-scale water electrolysis.

RESULTS

S doping–driven pyrite-to-marcasite transformation in CoSe₂

Structurally, both pyrite and marcasite CoSe₂ (P-CoSe₂ and M-CoSe₂, respectively) have Co atoms occurring in octahedral coordination and contain characteristic Se─Se pairs (28). In comparison to pyrite whose metal octahedra do not share edges with each other, the Co coordination octahedra in marcasite share two edges (29). P-CoSe₂ in principle can be transformed into marcasite by rotating the Se─Se pairs through 90° (Fig. 1A) (29, 30). We thus sought to obtain pure marcasite structure by using our previously reported P-CoSe₂ nanobelts (31), whereas only thermal annealing cannot overcome the high-energy barrier of the structural transformation (fig. S1). Heteroatom doping (32, 33) has been demonstrated as an effective means of triggering structural transformation and is also expected to tune the electronic structure. Moreover, given that the substitution of more polarization Se with S in CoSe₂ has been described to result in greater covalency (34), we therefore chose S for doping to induce pyrite-to-marcasite transformation. Density functional theory (DFT) calculations predicted that the energy barrier for pyrite-to-marcasite transformation can be notably reduced when S doping was considered (Fig. 1B and see Materials and Methods).

Fig. 1. Pyrite-to-marcasite structural transformation.

(A) Atomic arrangements of P-CoSe₂ (001) (left) and M-CoSe₂ (101) (right). Arrows indicate the direction of inclination of Se─Se pairs. (B) Free energy diagram for the energetics of pyrite-to-marcasite transformation in CoSe₂. Insets present the models of initial states (IS; P-CoSe₂), transition states (TS), and final states (FS; M-CoSe₂) without S (top) and with S (down) dopant. (C and E) HRTEM images of P-CoSe₂ (C) along the [001] orientation and M-CoSe_1.28S_0.72 (E) along the [101] orientation. The inset in each image shows corresponding fast Fourier transform pattern. Scale bars 1 nm. (D and F) High-magnification HRTEM image of P-CoSe₂ (D) and M-CoSe₂ (F) from the marked area in (C) and (E), respectively. The inset in each image shows corresponding atomic model viewed along the same direction. (G and H) X-ray powder diffraction patterns (G) and Raman spectra (H) of P-CoSe₂ and M-CoSe_1.28S_0.72. a.u., arbitrary unit. JCPDS, Joint Committee on Powder Diffraction Standards.

The S doping–driven pyrite-to-marcasite transformation in CoSe₂ was realized by annealing freshly synthesized P-CoSe₂ nanobelts with sublimed sulfur powder under an argon atmosphere (see Materials and Methods and fig. S2). Scanning electron microscopy (SEM) image shows that the belt-like morphology was well preserved after reaction (fig. S3); moreover, energy-dispersive x-ray spectrum elemental mapping exhibits uniform doping of S across the whole structure (figs. S4 and S5). The atomic lattice patterns of P-CoSe₂ and M-CoSe₂ were imaged using high-resolution transmission electron microscopy (HRTEM) along the [001] and [101] orientations, revealing cubic (Fig. 1C) and orthorhombic (Fig. 1E) structures, respectively. Corresponding Fourier transforms shown as insets of Fig. 1 (C and E) exhibit clear differences between the symmetries of the P-CoSe₂ and M-CoSe₂. The magnified images at atomic-resolution show continuous lattice without distortion and defects after structural transformation (Fig. 1, D and F). X-ray diffraction data in Fig. 1G confirm the structural change from cubic P-CoSe₂ to orthorhombic M-CoSe₂. After transformation, the typical Raman peak of P-CoSe₂ at 189 cm⁻¹ from Se─Se stretching mode was shifted to 167 cm⁻¹, further verifying the formation of M-CoSe₂ (Fig. 1H) (35). We note that this method enables tunable doping level of S in CoSe₂ (figs. S6 to S12 and table S1), which potentially permits a controllable tailoring of the electronic structure. Furthermore, this S doping–driven structural transformation was observed to proceed very fast (within 1 min; fig. S11), indicative of low kinetic barriers of transformation caused by S doping.

Spectroscopic analysis

We studied the impact of S doping on the electronic and coordination structures of Co using x-ray absorption near-edge spectroscopy (XANES) at the Co K-edge. As Fig. 2A shows, both P-CoSe₂ and M-CoSe₂ with different levels of S doping exhibit a pre-edge peak at ~7709.2 eV originated from the Co intra-atomic 1s → e_g transition (36). The existence of pre-edge feature after structural transformation indicates that Co cations remain in octahedral environments (36). However, the intensity of this peak decreases greatly after incorporating S into CoSe₂ (inset in Fig. 2A), pointing to increased e_g filling of Co cations (36). Extended x-ray absorption fine structure (EXAFS) reveals an obvious decrease in Co─Se bond distance, from 2.04 to 1.91 Å, with increasing S doping levels (Fig. 2B), which was caused by the formation of Co─S bonds whose distance is shorter than that of Co─Se bonds. By fitting the Co K-edge EXAFS (fig. S13 and table S2), we disclose a decrease in the first shell coordination number (CN) for Co─Se(from 5.5 to 2.7) and a monotonic increase in the CN for Co─S (from 0 to 2.8), as the S doping amount in the structure increases (Fig. 2C), suggesting highly tailored local coordination environments. Wavelet-transformed EXAFS shows that, compared with that in P-CoSe₂, the Co─Se bonding in S-doped M-CoSe₂ is extended to high-k direction (from 8.7 to 9 Å⁻¹) (Fig. 2D), indicative of enhanced Co─Se covalency resulted from the doping of more electronegative S atoms (37). The wavelet transform analyses also show a gradually emerged Co─S scattering as the S doping level increases (Fig. 2D).

Fig. 2. Spectroscopic studies.

(A and B) Co K-edge XANES spectra (A) and corresponding Fourier transforms of k³-weighted EXAFS spectra (B) for P-CoSe₂ and different S-doped M-CoSe₂. FT, Fourier transform. (C) Coordination numbers in the first coordination shell of Co atoms for P-CoSe₂ and different S-doped M-CoSe₂ by fitting the Co K-edge EXAFS. (D) Wavelet transforms of k³-weighted EXAFS spectra of the Co K-edge for P-CoSe₂ and different S-doped M-CoSe₂. (E and F) Se K-edge XANES spectra (E) and corresponding Fourier transforms of k²-weighted EXAFS spectra (F) for P-CoSe₂ and different S-doped M-CoSe₂. (G) S K-edge XANES spectra of P-CoSe₂ and different S-doped M-CoSe₂.

The Se K-edge XANES spectra for P-CoSe₂ and different S-doped M-CoSe₂ all exhibit a sharp peak at ~12,660.1 eV (1s → e_g* transition), whose intensity gradually decreased as the doping amount increases (Fig. 2E), suggesting low spatial overlap (hybridization) of Se 4p and Co 3d states (38). Se K-edge EXAFS data in Fig. 2F reveals that the bond intensity of Se─Co reduces as the S doping level increases, indicating decreased Se─Co coordination number. For the S K-edge, the intensity of S intra-atomic 1s → e_g* transition peak (~2471.7 eV) was also observed to decrease as S doping increases (Fig. 2G), implying gradually lowered overlap of S 3p states and Co 3d orbitals (39). Overall, these spectroscopic studies reveal that S doping enables a controllable tailoring of the electronic structure and local coordination environment of CoSe₂.

HER performance in acidic media

We assessed the HER performances of P-CoSe₂ and different S-doped M-CoSe₂ in Ar-saturated 0.5 M H₂SO₄ solution (pH 0.32) in a three-electrode setup, with a reference measurement of commercial Pt/C (20 wt % Pt on Vulcan XC72R carbon) for comparison (see Materials and Methods). Our series of measurements unveil that M-CoSe_1.28S_0.72 obtained at 400°C for 5 min shows the optimal property (figs. S14 to S16). By optimization, the catalyst loading on glassy carbon RDE was fixed at 1.0 mg cm⁻² (fig. S17). Figure 3A shows that P-CoSe₂ demands an overpotential of 268 mV at 10 mA cm⁻² (normalized on the basis of geometrical surface area), which decreased to 67 mV for M-CoSe_1.28S_0.72, approaching to 28 mV over Pt/C catalyst. Tafel analysis in Fig. 3B yields a slope of 59, 50, and 32 mV dec⁻¹ for P-CoSe₂, M-CoSe_1.28S_0.72, and Pt/C catalysts, respectively. Moreover, electrochemical impedance spectroscopy (Fig. 3C and fig. S18) measured at 200-mV overpotential gives a charge transfer resistances (R_ct) of 1.5 ohms for M-CoSe_1.28S_0.72, which is greatly smaller than that of 145.6 ohms for P-CoSe₂, correlating to improved electron transfer from the M-CoSe_1.28S_0.72 surface to H⁺ ions (discharge process) (40). We further used gas chromatography to detect and quantify the H₂ gas evolved from the M-CoSe_1.28S_0.72–docerated carbon paper electrode at 10, 100, and 500 mA cm⁻². The measured H₂ product at the three current densities all matches well with the theoretical values, corresponding to Faradaic efficiency of ~100% (fig. S19).

Fig. 3. HER performance.

(A and B) HER polarization curves (A) and Tafel plots (B) of P-CoSe₂, M-CoSe_1.28S_0.72 and commercial Pt/C. Catalyst loading: ~1.00 mg cm⁻². Sweep rate: 2 mV s⁻¹. Rotation rate: 1600 rpm. (C) Electrochemical impedance spectra of P-CoSe₂ and M-CoSe_1.28S_0.72. (D) pH-dependent HER polarization curves at a rotation rate of 1600 rpm and a sweep rate of 2 mV s⁻¹ for M-CoSe_1.28S_0.72. (E) Free energy diagrams for hydrogen adsorption at different sites on the P-CoSe₂, M-CoSe_1.28S_0.72, and M-CoSe_1.03S_0.97. (F) Chronopotentiometry (E ~ t) recorded on P-CoSe₂ and M-CoSe_1.28S_0.72 catalysts at a constant current density of 10 mA cm⁻². (G) Comparison of overpotential at 10 mA cm⁻² and operating lifetime of M-CoSe_1.28S_0.72 with that of reported catalysts in acidic electrolytes. NDs, nanodots. NF, nanofoam. (H) Polarization curves of the PEM electrolysers with P-CoSe₂, M-CoSe_1.28S_0.72, and commercial Pt/C as cathode catalysts. (I) Chronopotentiometry curve of the PEM electrolyser with M-CoSe_1.28S_0.72 as cathode and commercial IrO₂ as anode operated at 1A cm⁻² and 60°C. Insets show photographs of the PEM electrolyser device.

We demonstrate that the HER activity of M-CoSe_1.28S_0.72 is largely dependent of the H⁺ ion concentration. The pH-dependent experiments reveals that HER rate decreases with increasing the pH (Fig. 3D). For pH 2.5 to 4, diffusion-limiting currents are seen, which indicate that the HER on M-CoSe_1.28S_0.72 surface is controlled by the mass transport of H⁺ ions, unveiling the critical role of efficient proton transfer in the interfacial EDL (41). To probe mechanistic insights into the observed HER performance, we calculated the hydrogen adsorption free energy (ΔG_H), showing that M-CoSe_1.28S_0.72 at Co sites has the smallest ΔG_H value of 0.05 eV, close to the thermoneutral ΔG_H ≈ 0 (Fig. 3E and figs. S20 to S22). We thus regard the enhanced HER over M-CoSe_1.28S_0.72 to the Co sites, whose coordination environment and electronic structure were optimized by S dopants.

We further examined the long-term stability—an even more formidable challenge for PGM-free catalysts in acidic electrolytes. We evaluated the electrochemical stability of M-CoSe_1.28S_0.72 on carbon paper at 10 mA cm⁻² in 0.5 M H₂SO₄ solution. As Fig. 3F shows, the M-CoSe_1.28S_0.72 catalyst performs stably, exhibiting no sign of improvement in overpotential after 1000 hours of continuous operation. By stark contrast, the P-CoSe₂ undergoes a rapid degradation owing to the dissolution of Co atoms into the acidic electrolyte (fig. S23). The composition and electronic states of M-CoSe_1.28S_0.72 after the stability test were well retained (figs. S24 to S26). We also compared the performance metrics of our M-CoSe_1.28S_0.72 with those previously reported in terms of overpotential at 10 mA cm⁻² and operating lifetime. Under similar testing conditions, the results obtained over M-CoSe_1.28S_0.72 greatly surpass prior PGM-free catalysts achieved by other approaches (Fig. 3G and table S3).

The high HER activity and stability motivated us to explore the performance of M-CoSe_1.28S_0.72 in realistic PEM electrolyzers. We thus incorporated the M-CoSe_1.28S_0.72 catalyst in the cathode and commercial IrO₂ in the anode of a 4-cm² PEM electrolyzer using a Nafion 115 PEM (see Materials and Methods for details). We fed pure water instead of aqueous H₂SO₄ from the viewpoint of practical use despite its larger ohmic loss. The polarization curves obtained at 60°C show that M-CoSe_1.28S_0.72 requires a cell voltage of 1.79 V at 1 A cm⁻², far exceeding that of P-CoSe₂ catalyst (Fig. 3H). Intriguingly, this voltage is mere 110 mV higher than that of using Pt/C cathode. By assuming that all metal atoms are HER active, we estimated the turnover frequency (TOF) of M-CoSe_1.28S_0.72 to be 0.95 H₂ s⁻¹ (fig. S27). If only the surface atoms (~2.5%) are considered, then the TOF of M-CoSe_1.28S_0.72 is estimated as 38.4 H₂ s⁻¹ (fig. S27). Despite these metrics still compare inferior to commercial Pt/C, our results indeed suggest great application potential of the M-CoSe_1.28S_0.72 cathode. The application potential of M-CoSe_1.28S_0.72 catalyst can be further demonstrated by the notable stability of the PEM electrolyzer, showing no appreciable voltage increase over 410 hours of continuous testing at 1 A cm⁻² (Fig. 3I). In addition, the catalyst loss after long-term operation is mere ~2% by the weight (fig. S28). To our best knowledge, this represents one of the most promising PGM-free cathode catalyst that has shown both high activity and stability in a realistic PEM electrolyzer (table S4).

Mechanistic investigation

To obtain mechanistic insights into the enhanced acidic HER performance of M-CoSe_1.28S_0.72, work function (WF) and potential of zero charge (PZC) of three selected catalysts (i.e., P-CoSe₂, M-CoSe_1.28S_0.72, and M-CoSe_1.03S_0.97) were measured and analyzed. We studied catalysts’ WFs because this electronic property has been reported to show a linear correlation with HER reactivity in acidic electrolytes (42, 43). Moreover, catalysts’ WFs were also found to be positively correlated to the PZC in acid (44), which could lead to variation in interfacial environment and consequently altered HER kinetics (45, 46). We measured the WFs using ultraviolet photoelectron spectroscopy (UPS) and observed an increase in WF from 4.07 eV for P-CoSe₂ to 4.94 eV for M-CoSe_1.28S_0.72 (Fig. 4A). Further increasing S doping level to M-CoSe_1.03S_0.97, however, results in a reduced WF value of 4.69 eV. We plotted the measured WFs as a function of exchange current densities for different catalysts and find that HER intrinsic activity increases linearly with WF, in agreement with the observation reported by Trasatti for metals (42) (Fig. 4B). The PZC, which was measured by Gouy-Chapman capacitance method (47), increases from −0.32 V for P-CoSe₂ to −0.17 V for M-CoSe_1.28S_0.72, which decreases to −0.22 V by further raising S doping amount to M-CoSe_1.03S_0.97 (Fig. 4C), showing similar trend observed in WF measurements. By comparison, PZC was measured to be 0.28 V for Pt/C catalyst, reasonably agreeing with the literature data (45).

Fig. 4. Electronic property.

(A) UPS spectra of the P-CoSe₂, M-CoSe_1.28S_0.72, and M-CoSe_1.03S_0.97. (B) Linear relationship between the WFs and the exchange current densities of different catalysts. The original data of different metals are taken from the report of Trasatti (42). (C) PZC values measured for P-CoSe, M-CoSe_1.28S_0.72, M-CoSe_1.03S_0.97, and the Pt/C catalyst. (D) In situ SEIRAS spectra recorded at potentials from 0 to −0.2 V on P-CoSe₂ (left), M-CoSe_1.28S_0.72 (middle), and M-CoSe_1.03S_0.97 (right) in Ar-saturated 0.5 M H₂SO₄. (E) Corresponding proportion of gap-H₂O molecules in the EDL of the P-CoSe₂, M-CoSe_1.28S_0.72, and M-CoSe_1.03S_0.97 catalysts.

Generally, PZC has a direct impact on the interfacial electrical field, which, in turn, affects the interfacial water configuration and consequently the proton diffusion in the EDL (45, 46). Very recently, by combining ab initio molecular dynamic simulation and in situ surface-enhanced infrared adsorption spectroscopy (SEIRAS), Li et al. (46) described that water distribution and the connectivity of H-bond networks in the EDL mostly determine the reactivity of hydrogen electrocatalysis. To characterize interfacial water configuration in the EDL, we performed in situ SEIRAS spectra on the studied catalysts at different biases in Ar-saturated 0.5 M H₂SO₄ electrolyte (see Materials and Methods and fig. S29). As shown in Fig. 4D, the SEIRAS spectra of all catalysts exhibit asymmetric peaks spanned a broad frequency range from ~2900 to ~3800 cm⁻¹, corresponding to the O─H stretching band of interfacial water (46, 48). Using Gaussian fitting, the O─H stretching peaks could be consistently deconvoluted into two components representing water molecules in gap region (gap-H₂O at ~3540 cm⁻¹) and other interfacial water molecules (non–gap-H₂O at ~3357 cm⁻¹), agreeing with prior report (46). Figure 4E reveals that, for the M-CoSe_1.28S_0.72 electrode, the proportion of gap-H₂O molecules exceeds that of P-CoSe₂ and M-CoSe_1.03S_0.97 electrodes at all potentials examined. This observation could be the result of its PZC closer to the equilibrium potential (Fig. 4C), which barely influenced the interfacial electrical field and resulted in negligible perturbation of the H-bond networks in the EDL, thus facilitating hydrogen diffusion for HER (45, 46).

To understand how S doping tunes the electronic structure (i.e., WF and consequent PZC) of CoSe₂, we further performed Co L-edge x-ray absorption spectroscopy (XAS). From Fig. 5A, we observed two main peaks that were attributed to the transitions of Co 2p_3/2 (L₃ edge) and 2p_1/2 core (L₂ edge) electrons to unoccupied e_g orbital (49), complementary to above Co K-edge results. In comparison to P-CoSe₂, the intensity of L₃ edge decreases gradually as the S doping level increases. Because no oxidation state change of Co was detected by x-ray photoelectron spectroscopy (XPS) (fig. S30), the intensity reduction should be ascribed to the spin-state transition of Co ions from low-spin (LS) to high-spin (HS) states after S incorporation (50). To investigate this, we carried out temperature-dependent magnetizations with a magnetic field of H = 1 kOe under field-cooling procedures, which enables us to probe the spin structures of Co ions. As Fig. 5B shows, the susceptibility, χ, derived from the magnetizations follows Curie-Weiss behavior above 250 K. By fitting the linear regions, we obtained effective magnetic moment μ_eff and the consequent unpaired electron for each sample (see Materials and Methods for details, fig. S31, and table S5). We find that the number of unpaired electron increases with increasing S doping amount (Fig. 5B, inset). For P-CoSe₂, the number was calculated to be 1.03, reasonably matching with its LS state (t_2g⁶e_g¹) reported in literature (51). The number increases to 2.03 for M-CoSe_1.28S_0.72 and 2.98 for M-CoSe_1.03S_0.97, which indicates a gradual transition from LS to HS (t_2g⁵e_g²), consistent with the XAS results. We further find that the HER activities of studied catalysts exhibit a volcano relationship as a function of the unpaired electron number, where the activities are compared in terms of the overpotential at 10 mA cm⁻² (Fig. 5C). Among these catalysts, M-CoSe_1.28S_0.72 with unpaired electron number of 2.03 sits at the top of the volcano.

Fig. 5. Enhancement mechanism.

(A) Co L-edge XAS of P-CoSe₂ and different S-doped M-CoSe₂. (B) Temperature-dependent susceptibilities for P-CoSe₂ and different S-doped M-CoSe₂. Inset shows the corresponding unpaired d electron number. (C) Volcano relationship between the HER activity and the unpaired d-electron number of different catalysts. (D and E) Crystal structures (D) and schematic energy band diagrams (E) of the P-CoSe₂ (left), M-CoSe_1.28S_0.72 (middle), and M-CoSe_1.03S_0.97 (right) catalysts. (F) Schematic illustration of the electronic coupling between H and Co in P-CoSe₂ (left), M-CoSe_1.28S_0.72 (middle), and M-CoSe_1.03S_0.97 (right) during Volmer reaction.

The above studies reveal that S doping enables fine modulation of the electron filling between e_g and t_2g orbitals, thus tailoring the electronic structure for enhanced HER. In P-CoSe₂ structure, the Co ions bind with six Se atoms in an octahedral CoSe₆ coordination via hybridization of Co 3d and Se 4p orbitals (Fig. 5D) (29, 30). As analyzed by Co K-edge EXAFS above, the addition of S monotonously increased the Co─S coordination number (CN_Co─S = 0 to 2.8) as the doping amount increased, yielding altered octahedral coordination environment (Fig. 5D). Because the bandwidth of S 3p is narrower than that of Se 4p states (52) (fig. S32) and partially replacing Se with S leads to increased Se─Se/S bond distance and consequently a decrease in the Se─Se/S bonding-antibonding splitting (53). This will caused an attenuated orbital overlap between Co d and Se/S p orbitals, which makes the e_g orbital moves closer to the Fermi energy E_F (Fig. 5E) (34). As a result, t_2g electron becomes readily to transfer into e_g orbital, therefore mediating the spin states.

Furthermore, the downshift of e_g orbital toward the Fermi level after S doping can also result in enhanced electron-nucleus interaction, which gives rise to increased WF (42, 54), consistent with our WF measurement in Fig. 4A. We emphasize that much higher S doping (e.g., M-CoSe_1.03S_0.97), however, leads to decreased WF because of forming HS state caused by further reduced ligand-field splitting energy (55), as schematically shown in Fig. 5E. Owing to the positive correlation between WF and PZC in acid (44), the above results together explain why M-CoSe_1.28S_0.72 has the largest WF and PZC as measured experimentally.

On the basis of above results, we interpret the optimal HER performance over M-CoSe_1.28S_0.72 as the improved WF and PZC. In acidic electrolytes, the superior WF of M-CoSe_1.28S_0.72 relative to P-CoSe₂ and M-CoSe_1.03S_0.97 reflects its strongly bound nature of surface electrons, which attracts a high local H⁺ ion concentration near the electrode and thus an enhanced HER kinetics. This finding is consistent with the work of Trasatti (42, 44), which exhibited that greater WF can lead to higher exchange current for HER on metals. Meanwhile, the large PZC of M-CoSe_1.28S_0.72 that approaches the equilibrium potential suggests less disturbed interfacial electrical field and thereby a better connectivity of H-bond networks, benefiting H⁺ ion diffusion toward the catalyst surface (45, 46). When arriving at the reaction surface, H⁺ first demands a d-electron pair to adhere. After that, it experiences a discharge process and then the resultant adatom requires an empty semi-d orbital to form σ-bonding *H (Volmer reaction) (56). Compared with P-CoSe₂, the M-CoSe_1.28S_0.72 catalyst contains more exposed d-electron pairs and empty semi-d–orbital sites, thereby giving superior HER rate (Fig. 5F). Regarding M-CoSe_1.03S_0.97, the substantial downshift of e_g orbital toward the Fermi level, however, causes too strong *H adsorption and therefore limited HER kinetics (Fig. 5F and fig. S33).

DISCUSSION

In summary, we have demonstrated a complete pyrite-to-marcasite structural transformation in CoSe₂ induced by sulfur doping. The obtained M-CoSe_1.28S_0.72 catalyst shows outstanding HER activity and stability in acidic electrolyte. Moreover, a PEM electrolyzer assembled with this catalyst as cathode can be stably operated at 1 A cm⁻² for over 410 hours, suggesting potential practical use. The high operating stability of our catalyst can be explained by a greater covalency of Co─Se(S) bonds in marcasite caused by sulfur doping. In addition, this doping also suitably tailors electronic structures (i.e., WF and consequent PZC) of the catalyst, leading to an improved proton transfer in EDL, and a more favorable hydrogen adsorption for enhanced HER. While our PEM performance still compares inferior to the platinum-based counterpart, this work opens up the possibility of developing low-cost PEM electrolyzers that make use of PGM-free materials as electrode catalysts. Future effort should focus on further improving the activity and stability of the PGM-free catalysts, aiming to make future terawatt-scale PEM electrolysis a reality.

MATERIALS AND METHODS

Material synthesis

All chemicals were used as received without further purification. The M-CoSe_1.28S_0.72 was prepared through a two-step method. First, P-CoSe₂ nanobelts were synthesized as described in our previous work (31). Then, 50 mg of fresh P-CoSe₂ nanobelts and 0.5 g of sublimated sulfur powder were placed at two ceramic boats, respectively, while sublimated sulfur powder is at the upstream side. The samples were heated at 400°C for 5 min with a heating rate of 2°C min⁻¹ in Ar atmosphere. The synthetic temperature, time and the mass of sublimated sulfur powder, as well as the obtained products were summarized in table S1. All the obtained samples were carefully washed and dried before use.

Material characterizations

X-ray powder diffraction (XRD) was obtained from a Japan Rigaku DMax-γA rotation anode x-ray diffractometer equipped with graphite monochromatized Cu-K radiation. The morphology of the samples was achieved by SEM (Zersss Supra 40) and TEM (Hitachi H7700). HRTEM measurements and energy-dispersive spectroscopy (EDS) mappings were carried out using FEI Talos F200X, equipped with Super X-EDS system (four systematically arranged windowless silicon drift detectors) at 200 kV. Raman spectra analysis carried out with a LABRAM-HR confocal laser micro-Raman spectrometer with a wavelength of 532 nm. XPS data were gathered using an ESCALAB-MKII x-ray photoelectron spectrometer with Mg Kα radiation as an exciting source (Mg Kα = 1253.6 eV). UPS was obtained at the BL11U beamline of National Synchrotron Radiation Laboratory in Hefei (China). The x-ray absorption spectra of Co L-edges were taken on the BL10B beamline of National Synchrotron Radiation Laboratory in Hefei (China). The x-ray absorption spectra of Co and Se K-edges were carried out at the beamline 1W1B station of Beijing Synchrotron Radiation Facility (China), and the S K-edges were performed at the beamline 4B7A station of Beijing Synchrotron Radiation Facility (China). Inductively coupled plasma atomic emission spectroscopy data were investigated by an Optima 7300 DV instrument.

DFT calculations

The DFT calculations were performed by Vienna ab initio simulation package (57) program with projector augmented wave method, and the kinetic energy cutoff was set to be 500 eV. The convergence criterion for the electronic self-consistent iteration was set to be 10⁻⁷ eV. The atomic positions were fully relaxed until the force on each atom is less than 0.02 eV Å⁻¹. The Perdew-Burke-Ernzerhof (58)–generalized gradient approximation exchange-correlation functional was used throughout. The slab model of P-CoSe₂ (001) surface and M-CoSe₂ (101) surface were constructed on the basis of the optimized crystal structure, and the S atoms were located at M-CoSe₂ (101) surface to consider the effect of the doped S. The vacuum layer was set to be 15 Å to ensure the separation between slabs. To study the rotation of the Se─Se bond, its energy barriers were determined by using the climbing image nudged elastic band method. The Gibbs free energy is calculated by

$$G=H-T\text{Δ}S=E_{\mathrm{DFT}}+E_{\mathrm{ZPE}}-TS$$

where E_DFT is the total energy from the DFT calculation. E_ZPE is the zero-point energy, S is the entropy, and T is the temperature.

Electrochemical measurements

All the electrochemical measurements were performed in a standard three-electrode cell at ambient temperature connected to a VSP-300 Potentiostat (BioLogic, France). Ag/AgCl (3.5 M KCl) electrode and graphite rod were used as the reference and counter electrodes, respectively. The potentials reported in this work were normalized versus the RHE through a standard RHE calibration (E = E_Ag/AgCl + 0.20 V). An RDE with glassy carbon (Pine Research Instrumentation, 5.00 mm in diameter and disk area of 0.196 cm²) was used as the working electrode.

To make the working electrodes, 5 mg of catalyst powder was dispersed in 1 ml of 1:3 (v/v) isopropanol/deionzed water (DIW) mixture with 20 μl of Nafion solution (5 wt %), which was ultrasonicated to yield a homogeneous ink. Then, 40 μl of catalyst ink was pipetted onto the glassy carbon disk to ensure the catalyst loading of ~1.0 mg cm⁻². The fresh electrolytes (0.5 M H₂SO₄) were bubbled with pure argon for 30 min before measurements. The HER polarization curves were recorded at a sweep rate of 2 mV s⁻¹ and 1600 rpm (to remove the H₂ bubbles formed in situ) at ambient temperature. The electrochemical impedance spectroscopy measurement was performed in the same configuration at 200-mV overpotential over a frequency range from 100 KHz to 100 mHz at the amplitude of the sinusoidal voltage of 5 mV. The polarization curves were replotted as overpotential (η) versus log current (log j) to get Tafel plots to assess the HER kinetics of investigated catalysts. The Tafel slope (b) can be obtained by fitting the linear portion of the Tafel plots to the Tafel equation [η = b log (j) + a]. The exchange current density (j₀) were calculated from Tafel curves using extrapolation method.

The M-CoSe_1.28S_0.72 modified carbon paper (catalyst loading: ~1.0 mg cm⁻²) was used as working electrode to conduct chronopotentiometry experiments at a constant current density of 10 mA cm⁻². To estimate the double-layer capacitance, cyclic voltammograms were measured at different sweep rates in the potential region of 0 to 0.1 V versus RHE at ambient temperature. All the polarization curves were corrected with iR compensation that resulted from the solution resistance.

PEM water electrolyzer measurements

To prepare the catalyst ink, the catalyst and the Nafion solution (5 wt %) was dispersed in ethanol, followed by ultrasonicated to yield a homogeneous ink. To prepare the catalyst-coated membrane (CCM), the anode (IrO₂) and cathode (M-CoSe_1.28S_0.72) catalysts are sprayed onto sheets of polytetrafluoroethylene (PTFE), respectively. After that, the PTFE-supported cathode catalysts, Nafion 115, and PTFE-supported anode catalysts are hot pressed together at 135°C for 3 min under a pressure of 2 T. After cooling, the PTFE on the surface were carefully peeled off to get the CCM with an electrode area of 4 cm⁻². The final catalyst loading was 3.0 mg cm⁻² for the anode and 1.0 mg cm⁻² on the cathode. The CCM prepared was preserved in distilled water for further measurements.

To construct the PEM electrolyzer for performance evaluation, a titanium felt with Pt coating was used as the porous transport layers (PTLs) in the anode. For the cathode, a titanium felt without Pt coating is used as the PTL. The PEM electrolyzers were operated at 60°C. Deionized water was used as reactant for the constructed PEM electrolyzer, which was supplied through peristaltic pump circulation.

In situ SEIRAS experiments

The Au film was firstly prepared for SEIRAS. Briefly, the silicon prism was washed by aqua regia solution. Then, the silicon prism was polished with a 0.05-μm Al₂O₃ slurry, followed by ultrasound in acetone and DIW, respectively. After that, the silicon prism was dried, soaking 120 s in the NH₄F solution. Last, the silicon prism was immersed in a mixture of gold-plated solution and 2 wt % HF aqueous solution under 55°C water bath for 7 min to chemically deposit gold film. After deposition, the gold film on silicon prism was washed by DIW. Next, the working electrodes were prepared via dropping ink onto the obtained gold film. Briefly, 10 mg of catalyst powder was dispersed in 1 ml of isopropanol, which was ultrasonicated to yield a homogeneous ink. Then, 200 μl of catalyst ink was uniformly pipetted onto the Au film. Last, the obtained working electrode was dried naturally for use.

In situ SEIRAS tests were carried in a spectroelectrochemical cell with a three-electrode configuration. A graphite rod was used as the counter electrode and an Ag/AgCl (3.5 M KCl) electrode as the reference electrode. The cell was integrated into a NICOLET iS50 Fourier transform infrared spectrometer equipped with a liquid nitrogen–cooled mercury cadmium telluride (MCT) detector. During the in situ SEIRAS tests, Ar was kept bubbling into the electrolyte. In addition, each spectrum consisted of 32 single beams with a resolution of 4 cm⁻¹.

Temperature-dependent magnetization measurements

The temperature-dependent magnetization (M) measurements were carried out with a Magnetic Property Measurement System Superconducting Quantum Interference Device (MPMS SQUID) magnetometer and under magnetic field strength (H) of 1 kOe for all the samples. The effective magnetic moments (μ_eff) for all the samples were obtained via μ_eff = (8C)^1/2 × μ_B, wherein μ_B is Bohr magneton and C is Curie constant and is obtained by fitting the susceptibility (χ⁻¹ = M/H) above 250 K via Curie-Weiss law. The fractions of Co ions in HS and LS states are calculated from the relationship: μ_eff = g × μ_B × [S_HS × (S_HS + 1) × V_HS + S_LS × (S_LS + 1) × V_LS]^1/2, where S_HS = 3 and S_LS = 1. V_HS and V_LS (= 1 − V_HS) represent to the fractions for Co²⁺ with HS and LS states, respectively. Consequently, the number of unpaired electron (x) is calculated via x = S_HS × V_HS + S_LS × V_LS.

Supplementary Materials

This PDF file includes:

Figs. S1 to S33

Tables S1 to S5

References