High performance anion exchange membrane water electrolysis driven by atomic scale synergy of non-precious high entropy catalysts

Chemicals

Cobalt(II) acetate tetrahydrate (Co(OAc)₂·4H₂O, 98%), nickel(II) acetate tetrahydrate [Ni(OAc)₂·4H₂O, 98%], molybdenum(V) chloride (MoCl₅, > 99.6%), platinum (20 wt% on carbon black), iridium (IV) oxide, (IrO₂, 99.99%), and Nafion D-521 dispersion (5% w/w in water) were bought from Alfa Aesar. Iron(II) acetate [Fe(OAc)₂, 95%] was obtained from Acros Organics. Copper(II) acetate monohydrate [Cu(OAc)₂·H₂O, 98%] and ethylene glycol [(CH₂OH)₂, > 99%] were received from Showa Chemical. Potassium hydroxide (KOH, > 85.0%) was acquired from Honeywell. Carboxyl group-modified CNTs (COOH-CNTs, > 99.0%) were purchased from Golden Innovation. Nickel foam (NF, 95 PPI, 1.7 mm in thickness) was from MAY CHUN Co., Ltd, while Fumatech FAA-3-50 membranes were obtained from Fuel Cell Store. Commercial carbon paper (CP, GDL280) and nickel mesh (NM) were supplied by CeTech and Ing-Jing Precise Industrial Corp, respectively.

Syntheses of FeCoNiCuMo based high entropy catalysts

For preparation of FCNCuMO_X@CNT, 60 mg of COOH-CNT were ultrasonically dispersed in 10 mL ethylene glycol for 1 h. A 0.5 × 0.5 cm² NF was repeatedly soaked and methanol-lamp dried in the suspension until CNT loading reached 3 mg cm^-2 (CNT@NF). Precursors of Fe, Co, Ni, Cu, and Mo (0.0051 mmol each) were dissolved in 2.55 mL of ethylene glycol and drop-cast onto CNT@NF, followed by drying at 60 °C for 10 h. The sample was microwave-treated (800 W, 10 s, Ar) to yield FCNCuMO_X@CNT. To obtain FCNCuM@CNT, FCNCuMO_X@CNT@NF was calcined in a tube furnace under 5% H₂/Ar (250 sccm) at 350 °C for 2 h (10 °C min^-1). The final loadings of FCNCuMO_X and FCNCuM were ~1.4 and ~1.2 mg cm^-2, respectively. For comparison, ten quaternary catalysts (five oxides, five alloys) were synthesized similarly, excluding one element from the precursor mix. Oxides were used as OER catalysts (anode), and alloys for HER (cathode). For AEMWE assembly, FCNCuMO_X@CNT and FCNCuM@CNT were loaded on NM and CP, respectively, to minimize contact resistances.

Fabrication of IrO₂ and Pt/C benchmark electrodes

For alkaline electrolysis, 0.35 mg of IrO₂ (or 1.5 mg of Pt/C) was dispersed in 0.5 mL of mixed solvent (0.4 mL of ethanol, 0.1 mL of deionized (DI) water, and 10 μL of 5 wt% Nafion) and sonicated for 1 h to form a uniform ink. The ink was drop-cast on NF (0.5 × 0.5 cm) and dried at 70 °C for 10 h. The IrO₂ and Pt/C loadings were matched to FCNCuMO_X (1.4 mg cm^-2) and FCNCuM (1.2 mg cm^-2), respectively. For AEMWE, 1.4 mg of IrO₂ was dispersed in 7 mg of Sustainion XB-7 (5% in ethanol), 1 mL of isopropanol, and 1 mL of DI water (or 6 mg of Pt/C in 30 mg of XB-7), then sonicated for 1 h to obtain a uniform suspension. The IrO₂ (or Pt/C) suspension was applied onto a 1 cm x 1 cm NM (or CP) with spray coating to afford the precious metal based anode, IrO₂@NM (or cathode, Pt/C@CP). Similar to the case of the AWE, the mass loading of IrO₂ and Pt/C was controlled to be the same with that of FCNCuMO_X (1.4 mg cm^-2) and FCNCuM (1.2 mg cm^-2), respectively for fair comparison.

Pre-treatment of anion exchange membranes

The commercial anion exchange membrane (FAA-3-50) was cut into 3 × 3 cm² pieces, immersed in 1 M KOH at room temperature for 24 h for activation, and then soaked in deionized water for another 24 h to thoroughly rinse off residual KOH and remove the polyethylene terephthalate (PET) liner.

Material characterizations

Sample morphologies were observed using a scanning electron microscope (SEM, Hitachi SU8010). Crystalline structures were characterized by X-ray diffraction (XRD, D8 ADVANCE Eco) equipped with a Cu Kα source and high-resolution transmission electron microscopy (HRTEM, JEM-ARM200FTH). The valence states of constituent elements were determined by high-resolution X-ray photoelectron spectroscopy (HR-XPS, PHI Quantera SXM). Prior to XPS characterizations, the sample was treated with argon ion beam sputtering etching to reduce the surface oxide layer that inevitably formed for metallic samples under prolonged exposure to ambient atmosphere. Elemental compositions were measured via inductively coupled plasma-optical emission spectrometry (ICP-OES, Thermo iCAP 7000 SERIES). Raman spectra were obtained using a Raman spectrometer (MRID, ProTrusTech Co., Ltd) with a 532 nm excitation laser to detect OER-active intermediates. XAS of Fe, Co, Ni, Cu, and Mo K-edges was conducted in fluorescence mode at beamline TPS 44A1 of the National Synchrotron Radiation Research Center (NSRRC), Taiwan, using a solid-state detector. The Faradaic efficiencies of OER and HER were assessed by comparing gas yields measured via gas chromatography (GC-2014) with theoretical values calculated from the respective current densities.

Electrochemical characterizations

Electrochemical characterizations were conducted on an electrochemical workstation (CH Instrument, CHi 760E) using a three-electrode system in 1 M KOH under room temperature. The exposed geometric surface area of the working electrodes was controlled to be 0.25 cm² (0.5 cm × 0.5 cm). Hg/HgO (1.0 M) and a graphite rod were used as the reference and counter electrodes, respectively. All potentials reported were converted to refer to the reversible hydrogen electrode (RHE) according to E (vs. RHE) = E (vs. Hg/HgO) + 0.059 pH + 0.105 V. Polarization curves were measured at a scan rate of 1 mV s^-1, and the resulting current densities were iR-corrected using a resistance of ~2-2.5 Ω.

Electrolyzer measurements

To assemble the AEMWE, the anion exchange membrane (FAA-3-50) was sandwiched between 1 cm × 1 cm anode and cathode to form the membrane electrode assembly (MEA), which was then integrated with spacers, flow channels, bipolar plates, and end plates, and fastened using a digital torque wrench at 1 N·m^-2. A 1 M KOH electrolyte was supplied to the anode at 30 mL·min^-1, with its temperature maintained at 50 °C using a controlled heating chamber. The polarization curves were recorded within an applied potential window of 1.2 to 2.0 V at a scan rate of 1 mV s^-1. And electrochemical impedance spectroscopy (EIS) was conducted at 1.8 V (vs. Hg/HgO) within a frequency window of 0.01 Hz to 100 kHz.

Computational methods

All calculations were performed using DFT with the Vienna Ab Initio Simulation Package (VASP)^[11,12]. The electron-ion interaction was described with the projector-augmented wave (PAW) method^[13,14]. The wavefunctions were expanded using the plane wave basis. The Perdew-Burke-Ernzerhof (PBE)^[15] exchange-correlation functional was employed to account for electron exchange and correlation energies. An energy cutoff of 560 eV was applied, and the convergence criterion for geometry optimization was set at 0.02 eV/Å. A FeCoNiCuMo HEA model was constructed using a 4-layer 3 × 3 (111) face-centered cubic (FCC) slab, followed by randomly assigning elements to each lattice point according to the atomic composition determined from ICP-OES. The lattice constant was obtained from experimental data and kept constant throughout the calculations. To mitigate errors arising from interactions between periodically repeated slabs, a vacuum layer of 15 Å was introduced. A k-point grid of 8 × 8 × 1 was sampled using the Monkhorst-Pack scheme for each computation. Spin-polarized calculations were performed to account for the magnetic properties of Fe, Co, and Ni. The adsorption energy (E_ads) was calculated using the formula E_ads = E_A/S - E_S - E_A, where E_A/S, E_S, and E_A were the calculated energies of the adsorption system, substrate, and adsorbate in the gas phase, respectively. ΔG_H* and ΔG_OH* were estimated using

The correction factors, 0.24 eV and 0.29 eV, are commonly applied and include contributions from both the zero-point energy and the entropy term^[16]. Hydrogen adsorption energies were calculated for 10 different FeCoNiCuMo HEA configurations [Supplementary Figure 1], with each configuration having 18 hollow sites exposed, totaling 180 different adsorption sites. The probability of each adsorption site was weighted based on the concentration of each metal element as determined by ICP-OES to obtain the average adsorption energy of the FeCoNiCuMo HEA model. For the corresponding quaternary samples, the probability of the adsorption sites involving the absent element was set to zero, and the probabilities of the adsorption sites involving the four present elements were re-weighted to obtain the average adsorption energy for the quaternary samples. Transition state structures were estimated using the climbing image nudged elastic band method (CI-NEB)^[17] with nine images. Nine images were positioned along the minimum energy path (MEP), and a spring force constant of -5.0 eV/Ų was applied between the images to relax them until the maximum force acting on each atom was less than 0.02 eV/Å.

Materials characterizations

The two high entropy catalysts, FCNCuMO_X@CNT and FCNCuM@CNT, were prepared with a simple and fast microwave synthesis method as illustrated in Scheme 1. Briefly, carboxyl group-modified CNTs were first deposited onto the skeleton surfaces of the porous substrate (NF, NM, or CP), followed by drop-casting a mixed Fe, Co, Ni, Cu, Mo precursor solution onto the substrate. The precursor loaded substrate was then treated in a microwave oven to form the HEO catalyst, FCNCuMO_X@CNT, used as the OER catalyst at the anode. As for the HER catalyst at the cathode, FCNCuM@CNT, it was obtained from thermal reduction of FCNCuMO_X@CNT in an atmosphere of 5% H₂/Ar. For comparison purposes, corresponding quaternary catalysts, both oxides and alloys, were also prepared.

Scheme 1. Schematic illustration for preparation of FCNCuMO_X@CNT and FCNCuM@CNT. CNTs: Carbon nanotubes.

Figure 1A shows the XRD pattern of FCNCuMO_X@CNT and FCNCuM@CNT. These high entropy materials were deposited on the surfaces of CNTs as nanoparticles, as evident from Figure 1B for FCNCuMOx@CNT and Figure 1C for FCNCuM@CNT. These nanoparticles have a size of around 6 nm [Figure 1D-F]. For the HEO catalyst, FCNCuM@CNT, two sets of diffraction peaks, (111) and (200) of the rock salt oxide located at 2θ of 37.1° and 43.5° and (311) and (440) of the spinel oxide located at 2θ of 34.4° and 60.6°, were observed from Figure 1A. The broad peak centering around 2θ of 35^o can be deconvoluted, as shown in Supplementary Figure 2, to determine the contributions from the spinel and rock salt oxides to be at a molar ratio of around 2:1. Evidently, FCNCuMO_X@CNT was composed of a mixture of two HEOs, rock salt and spinel. HRTEM was conducted to further support the conclusion. Figure 1G shows the HRTEM image with interlayer distances of 0.245 and 0.206 nm determined, in good agreement with the d-spacing of the (111) and (200) planes of the rock salt oxide, respectively. Spinel oxide was also detected, as evident from Figure 1H, with an interlayer distance of 0.257 nm observed, corresponding well to d-spacing of the (311) planes of the spinel oxide. As for the HEA catalyst, FCNCuM@CNT, characteristic diffraction peaks located at 2θ of 44.6°, 51.9°, and 76° were observed, as evident from Figure 1A, attributable to a single FCC phase of the HEA. Furthermore, interlayer distances of 0.207 and 0.181 nm were determined from the HRTEM image shown in Figure 1I, in good agreement with the d-spacing of the (111) and (200) planes of the FCC phase, respectively. Evidently, the mixed HEOs of FCNCuMO_X@CNT were completely reduced to an FCC HEA, FCNCuM@CNT. Note that the pronounced diffraction peaks located at 2θ of 26.1° for both FCNCuMO_X@CNT and FCNCuM@CNT [Figure 1A] were contributed by the presence of CNTs. In addition, the metallic elements of both FCNCuMO_X@CNT and FCNCuM@CNT are uniformly distributed in the sample as evident from the STEM-EDS elemental mapping presented in Supplementary Figures 3 and 4, respectively. Furthermore, the metallic compositions of the two catalysts were determined with ICP-OES, as summarized in Supplementary Table 1 for examination. Evidently, the atomic percentages of all constituent elements fall within the 5%-35% range, required for formation of high entropy materials.

Figure 1. (A) XRD patterns of FCNCuMO_X@CNT and FCNCuM@CNT. TEM images (B, D and E) and HRTEM images (G and H) of FCNCuMO_X@CNT. TEM images (C and F) and HRTEM images of (I) FCNCuM@CNT. XRD: X-ray diffraction; CNTs: carbon nanotubes; HRTEM: high-resolution transmission electron microscopy; TEM: transmission electron microscopy.

XPS analysis was performed to examine the surface oxidation states of FCNCuMO_X@CNT and FCNCuM@CNT. Figure 2 shows the HR-XPS spectra of Fe 2p, Co 2p, Ni 2p, Cu 2p, and Mo 3d for both catalysts. For FCNCuMO_X@CNT, the Fe 2p spectrum [Figure 2A] reveals six peaks attributed to Fe²⁺ (2p_3/2 709.1 eV, 2p_1/2 722.3 eV), Fe³⁺ (2p_3/2 712.1 eV, 2p_1/2 726.0 eV), and satellite peaks (2p_3/2 715.8 eV, 2p_1/2 732.2 eV)^[18,19] [Figure 2A]. The Co 2p spectrum [Figure 2B] displays peaks corresponding to Co²⁺ (2p_3/2 781.0 eV, 2p_1/2 797.4 eV), Co³⁺ (2p_3/2 779.7 eV, 2p_1/2 796.3 eV), and satellite peaks (2p_3/2 784.6 eV, 2p_1/2 802.5 eV)^[20,21]. Similarly, the Ni 2p spectrum [Figure 2C] shows features for Ni²⁺ (2p_3/2 854.0 eV, 2p_1/2 872.0 eV), Ni³⁺ (2p_3/2 856.0 eV, 2p_1/2 873.5 eV), and satellite peaks (2p_3/2 861.5 eV, 2p_1/2 879.5 eV)^[22]. The Cu 2p peaks [Figure 2D] include Cu²⁺ (2p_3/2 933.3 eV, 2p_1/2 952.9 eV) and satellite peaks (2p_3/2 941.4 eV, 2p_1/2 960.7 eV)^[23]. Finally, the Mo 3d spectrum [Figure 2E] consists of two peaks corresponding to Mo⁶⁺ (3d_5/2 232.1 eV, 3d_3/2 235.2 eV)^[24].

Figure 2. HR-XPS spectra of (A) Fe 2p, (B) Co 2p, (C) Ni 2p, (D) Cu 2p, and (E) Mo 3d of FCNCuMOX@CNT and FCNCuM@CNT. HR-XPS: High-resolution X-ray photoelectron spectroscopy; CNTs: carbon nanotubes.

As for FCNCuM@CNT, metallic states of the five constituent elements, absent from the HR-XPS spectra of FCNCuMO_X@CNT, are detected in addition to the oxidized states. The Fe 2p spectrum [Figure 2A] consists of eight peaks, assigned to Fe⁰ (2p_3/2 707.0 eV, 2p_1/2 720.0 eV), Fe²⁺ (2p_3/2 709.4 eV, 2p_1/2 722.5 eV), Fe³⁺ (2p_3/2 712.0 eV, 2p_1/2 725.0 eV), and satellite peaks (2p_3/2 714.9 eV, 2p_1/2 730.6 eV)^[25,26]. The Co 2p spectrum [Figure 2B] is composed of six constituent peaks, corresponding to Co⁰ (2p_3/2 778.0 eV, 2p_1/2 793.0 eV), Co²⁺ (2p_3/2 781.1 eV, 2p_1/2 796.2 eV), and satellite peaks (2p_3/2 786.2 eV, 2p_1/2 803.0 eV)^[20,21]. The Ni 2p spectrum [Figure 2C] features eight peaks, assigned to Ni⁰ (2p_3/2 852.8 eV, 2p_1/2 870.0 eV), Ni²⁺ (2p_3/2 854.0 eV, 2p_1/2 872.0 eV), Ni³⁺ (2p_3/2 856.0 eV, 2p_1/2 873.5 eV), and satellite peaks (2p_3/2 858.2 eV, 2p_1/2 876.9 eV)^[22,27]. The Cu 2p spectrum [Figure 2D] shows four peaks corresponding to Cu⁰ (2p_3/2 932.7 eV, 2p_1/2 952.5 eV) and Cu²⁺ (2p_3/2 934.8 eV, 2p_1/2 954.7 eV)^[23,28]. Lastly, the Mo 3d spectrum [Figure 2E] contains six peaks, attributed to Mo⁰ (3d_5/2 228.1 eV, 3d_3/2 231.0 eV), Mo⁴⁺ (3d_5/2 229.2 eV, 3d_3/2 232.9 eV), and Mo⁶⁺ (3d_5/2 232.0 eV, 3d_3/2 235.2 eV)^[24,29]. Note that the oxidized states of FCNCuM@CNT come from the surface oxidation that occurs when the sample is exposed to ambient atmosphere before XPS characterizations. The impact of surface oxidation is particularly intense for the present samples because of the small size of the catalyst nanoparticles, around only 6 nm. The surface oxidation states of the quaternary samples were also examined with XPS characterizations for comparison. Supplementary Figures 5-9 show the HR-XPS spectra of Fe 2p, Co 2p, Ni 2p, Cu 2p, and Mo 3d for all ten quaternary catalysts. Similar to the quinary samples, metallic states of all constituent elements are detected in CNCuM@CNT [Supplementary Figure 5], FNCuM@CNT [Supplementary Figure 6], FCCuM@CNT [Supplementary Figure 7], FCNM@CNT [Supplementary Figure 8], and FCNCu@CNT [Supplementary Figure 9], but not in CNCuMO_X@CNT [Supplementary Figure 5], FNCuMO_X@CNT [Supplementary Figure 6], FCCuMO_X@CNT [Supplementary Figure 7], FCNMO_X@CNT [Supplementary Figure 8], and FCNCuO_X@CNT [Supplementary Figure 9]. Moreover, the deconvolution results of the five constituent elements in these ten quaternary samples are nearly identical to those in the quinary samples [Supplementary Notes 1-5], indicating that the surface oxidation states of the quaternary samples are similar to those of the quinary samples.

XAS was conducted to gain further insights on the characteristics of the oxidation states of the constituent elements of the two samples. Different from XPS, a surface sensitive characterization, XAS offers information on the average oxidation states of the bulk of the samples. Supplementary Figure 10 shows the X-ray absorption near edge structure (XANES) spectra of FCNCuMO_X@CNT and FCNCuM@CNT. Evidently, the absorption edge positions, thus the valence states, of the constituent elements of FCNCuM@CNT downshift to lower absorption energies as compared to those of the corresponding elements of FCNCuMO_X@CNT. The absorption edge positions of the constituent elements of FCNCuM@CNT however are still significantly higher than those of the corresponding metal foils, again attributable to the great impact of surface oxidation on the average oxidation states of the ultra-small size nanoparticles. In conclusion, the outcome of XAS is consistent with that of HR-XPS.

Electrochemical characterizations

Electrochemical performances of the two high entropy (quinary) and ten quaternary samples, along with the two benchmark catalysts, Pt/C for HER and IrO₂ for OER, were investigated in 1 M KOH for AWE. Here, FCNCuMO_X@CNT, CNCuMO_X@CNT, FNCuMO_X@CNT, FCCuMO_X@CNT, FCNMO_X@CNT, FCNCuO_X@CNT, and commercial IrO₂ were used as the OER catalysts. Supplementary Figure 11 shows the OER linear sweep voltammetry (LSV) polarization curves and corresponding Tafel plots of the seven samples for comparison. Four points can be observed. First, all multi-element oxide samples exhibit better OER catalytic performances than the benchmark catalyst, IrO₂ (η₁₀ = 245 mV, η₅₀₀ = 486 mV), indicating the advantage of synergetic interactions realized in the multi-element oxides. Second, all quaternary samples exhibit similar electrochemical performances, including FNCuMO_X@CNT (η₁₀ = 212 mV, η₅₀₀ = 343 mV, Tafel slope = 49.5 mV dec^-1), FCCuMO_X@CNT (194 mV, 352 mV, 53.3 mV dec^-1), FCNMO_X@CNT (185 mV, 320 mV, 51.5 mV dec^-1), FCNCuO_X@CNT (190 mV, 318 mV, 69.2 mV dec^-1), except for CNCuMO_X@CNT (227 mV, 383 mV, 77.2 mV dec^-1). Evidently, the presence of Fe is critical on catalysis of OER. Third, the quinary sample of high configurational entropy, FCNCuMO_X@CNT, exhibits the highest OER activity (η₁₀ = 183 mV, η₅₀₀ = 294 mV), demonstrating the stronger cocktail effect of the five-element oxides over those of the four-element oxides. Fourth, the catalysts with co-existence of Fe, Co, and Ni, including FCNCuMO_X@CNT, FCNCuO_X@CNT, and FCNMO_X@CNT, show higher catalytic efficiency than those without the co-existence of Fe, Co, and Ni, implying that the synergy between Fe, Co, and Ni promotes catalysis of OER.

To investigate the mechanism of OER catalyzed by FCNCuMO_X@CNT, in-situ Raman spectroscopy was conducted. Supplementary Figure 12 shows the in-situ Raman spectra of FCNCuMO_X@CNT in 1 M KOH recorded at increasing applied potentials. Evidently, a multi-peak broad feature band appears within the wavenumber range of 460-720 cm^-1 at open circuit potential (OCP). This broadband can be attributed to one Raman active mode (A_1g, ~550 cm^-1) of the rock salt structure^[30] and two Raman active modes (T_2g, ~507 cm^-1 and A_1g, ~670 cm^-1) of the spinel structure^[31]. The contribution of the spinel structure can be further divided into two sub-peaks, one located in the wavenumber range of 460 cm^-1 to 660 cm^-1 for the octahedral group and the other located between 660 and 720 cm^-1 for the tetrahedral group of the spinel oxide^[32]. The combined showing of the feature modes of both rock salt and spinel structures is in good agreement with the finding that FCNCuMO_X@CNT contains both spinel oxide and rock salt oxide [Figure 1A]. Upon raising the applied potential to 1.416 V, the two feature Raman bands centering at 480 and 556 cm^-1 become more pronounced, with enhanced peak intensity by 1.11 and 1.17 folds, respectively, as compared to those at OCP. This phenomenon signifies emergence of OER-active intermediates, NiOOH and FeOOH. Additionally, at an applied potential of 1.456 V, the intensity of the feature Raman band centering at 508 cm^-1 increases by 1.42 times as compared to that at OCP, attributable to the emergence of another OER-active intermediate, CoOOH^[33]. Emergence of the OER-active intermediates, including NiOOH, CoOOH, and FeOOH, indicates further oxidation of Ni, Co, and Fe on the surface of FCNCuMO_X@CNT under elevated anodic potentials to trigger the OER. Consequently, Ni, Co, and Fe serve as the active sites of FCNCuMO_X@CNT for catalysis of OER. Fe is particularly important in this regard, as it significantly impacts the OER activity of NiOOH. Lin et al.^[34] found that Fe-doping alters the rate determining step (RDS) of the OER on NiOOH, greatly decreasing the Tafel slope of NiOOH. In addition, Fe-doping also reduces the kinetic energy barrier of the RDS. Our experimental results are in line with the above finding. For the case of CNCuMO_X@CNT, without the presence of Fe and thus lack of generation of FeOOH and the enhancing effect toward the OER activity of NiOOH, its OER activity is the lowest among all multi-element oxide catalysts.

To examine the roles played by the five constituent elements of FCNCuM@CNT toward catalysis of HER, the six multi-element alloy catalysts samples, including FCNCuM@CNT, CNCuM@CNT, FNCuM@CNT, FCCuM@CNT, FCNM@CNT, FCNCu@CNT, together with commercial Pt/C, were characterized for catalytic performances toward HER. Supplementary Figure 13 shows the HER LSV polarization curves and Tafel plots of the samples, respectively. Several points can be made. First, the overpotentials achieved by FCNCu@CNT and FCCuM@CNT (η₁₀ = 68, 57 mV > 50 mV; η₅₀₀ = 320, 303 mV > 300 mV), without the simultaneous presence of Ni and Mo, are appreciably higher than those of the other four multi-element alloy catalysts, implying the active roles played by Ni and Mo toward catalysis of HER^[35]. The synergy between them effectively promotes catalysis of HER. Second, among all quaternary alloy catalysts, FCNM@CNT shows the best performance (η₁₀ = 43 mV; η₅₀₀ = 228 mV), as outstanding as the HEA catalyst, FCNCuM@CNT (η₁₀ = 38 mV < 50 mV; η₅₀₀ = 230 mV < 250 mV), implying the relative inertness of Cu toward catalysis of HER. Third, the mechanism of HER involves three fundamental steps, including Volmer step (electrochemical hydrogen adsorption), Heyrovsky step (electrochemical hydrogen desorption), and Tafel step (chemical hydrogen desorption). When treating the Volmer, Heyrovsky, and Tafel step as the RDS, the theoretical Tafel slopes are derived to be 120, 40, and 30 mV dec^-1, respectively. From Supplementary Figure 13B, the Tafel slope of FCNCuM@CNT is 104 mV dec^-1, falling between 120 and 40 mV dec^-1, indicating that the Volmer and Heyrovsky steps are comparable^[36]. Fourth, FCNCuM@CNT exhibits the highest HER activity among all multi-element alloy catalysts, again demonstrating the stronger cocktail effect of the five-element alloy over those of the four-element alloys. The HER catalytic performances of FCNCuM@CNT, although not as good as Pt/C (η₁₀ = 12 mV, η₅₀₀ = 174 mV), are considered outstanding for non-precious metal based HER catalysts. Figure 3A and B summarizes η₁₀, η₅₀₀, and Tafel slopes of all sample catalysts toward OER and HER, respectively. Evidently, FCNCuMO_X@CNT and FCNCuM@CNT exhibit outstanding catalytic performances toward OER and HER, respectively, making them promising catalysts for anode and cathode, respectively, of AEMWEs. The coupling of FCNCuMO_X@CNT and FCNCuM@CNT is expected to outperform the IrO₂//Pt/C couple for AEMWEs since the combined overpotentials for OER and HER of AWE of the FCNCuMO_X@CNT//FCNCuM@CNT couple, 221 mV for 10 mA cm^-2 and 524 mV for 500 mA cm^-2, are appreciably lower than those of the IrO₂//Pt/C couple, 257 mV for 10 mA cm^-2 and 660 mV for 500 mA cm^-2.

Figure 3. Electrochemical characterizations of FCNCuMO_X@CNT, FCNCuM@CNT, and benchmark electrodes, Pt/C and IrO₂, in 1 M KOH. Summary of Tafel slope and overpotentials, η₁₀ and η₅₀₀, for (A) OER and (B) HER achieved by all sample catalysts. CNTs: Carbon nanotubes; OER: oxygen evolution reaction; HER: hydrogen evolution reaction.

Furthermore, Faradaic efficiency of FCNCuMO_X@CNT and FCNCuM@CNT for OER and HER was determined at current densities of 100 and -100 mA cm^-2, respectively, for a duration of 60 min. As shown in Supplementary Figure 14, the measured quantities of O₂ and H₂ generated in the experiment closely match the theoretical values calculated based on the respective current densities, giving Faradaic efficiency close to 100% and confirming the absence of side reactions during OER and HER.

In-situ XAS

In-situ XAS was conducted to gain deeper mechanistic insights on the OER catalyzed by FCNCuMO_X@CNT. Figure 4A-E shows the XANES spectra at K-edge of Fe, Co, Ni, Cu, and Mo during OER, with the applied potentials increasing from OCP to 0.65 V (vs. Hg/HgO) and back to OCP in 1 M KOH. The absorption edges of Co and Ni upshift to higher energies upon raising the applied potential from OCP to 0.65 V and downshift back to lower energies when the applied potential returns to OCP, implying further oxidation of Co and Ni to higher valence states under anodic potentials for formation of active species of higher valences for catalysis of OER. On the contrary, the absorption edge of Mo experiences a slight downshift at 0.65 V, implying its relative inertness toward catalysis of OER. Finally, the absorption edges of Fe and Cu remain almost unchanged under variations of the applied potentials, indicating the stable oxidation states of Fe and Cu under anodic potentials. From XANES spectra, one can conclude that Co and Ni are active sites toward catalysis of OER. The roles played by Fe, Cu, and Mo are to be further examined and confirmed with Fourier-transformed extended X-ray absorption fine structure (FT-EXAFS) spectra.

Figure 4. In-situ XAS characterizations of FCNCuMO_X@CNT under variations of applied potentials. XANES spectra of K-edge of (A) Fe, (B) Co, (C) Ni, (D) Cu, and (E) Mo; Inset figure of panel (C) shows locally enlarged absorption spectra; FT-EXAFS spectra of K-edge of (F) Fe, (G) Co, (H) Ni, (I) Cu, and (J) Mo; (K) Locally enlarged spectra of panel (H); (L) Fitted first-shells of Ni at OCP and 0.65 V. CNTs: Carbon nanotubes; XAS: X-ray absorption spectroscopy; XANES: X-ray absorption near edge structure; FT-EXAFS: fourier-transformed extended X-ray absorption fine structure; OCP: open circuit potential.

Corresponding FT-EXAFS spectra of Fe, Co, Ni, Cu, and Mo are shown in Figure 4F-J, from which variations in local coordination environments of the metal centers under anodic potentials can be examined for their involvement in catalysis of OER. The analysis is focused on the first shells, the metal-O coordination, of the spectra for data reliability and accuracy. First, the FT-EXAFS spectra of Cu and Mo remain invariant upon variations of the applied potentials, indicating constant local coordination environments of the two elements during OER and thus the inertness of Cu and Mo toward catalysis of OER. On the contrary, the oscillation amplitudes of the FT-EXAFS spectra of Fe and Co increase when raising the applied potential from OCP to 0.65 V (vs. Hg/HgO) [Figure 4F and G], indicating enhancements in coordination extent with neighboring species, which can be attributed to the coordination of Fe and Co with OH^- under anodic potentials for catalysis of OER. Note that, for OER in alkaline media, the active site is under continuing attacks of OH^- to generate hydroxides, oxides, and finally oxyhydroxides for release of oxygen and recovery of the active site. One thus concludes that Fe and Co are actively involved in catalysis of OER^[37]. As for the FT-EXAFS spectra of Ni, the situation is more complicated. The oscillation amplitude drops, accompanied by a slight downshift in coordination distances [Figure 4H]. Here, two different motifs can be identified for the Ni-O coordination shell, Ni-O_short and Ni-O_long, as shown in Figure 4K. Ni-O_short accounts for nickel of a higher valance state, such as NiOOH, whereas Ni-O_long is contributed by nickel of a lower valence state, such as Ni(OH)₂ and NiO. The first shells of the FT- EXAFS spectra of Ni recorded at OCP and 0.65 V can be fitted to reveal the detailed coordination environment of Ni [Figure 4L]. For the case of Ni at OCP, it is purely Ni-O_long, whereas for the case of Ni at 0.65 V, it is a combination of 20% Ni-O_long and 80% Ni-O_short, as summarized in Supplementary Table 2. Upon raising the applied potential from OCP to 0.65 V, the oscillation amplitude of Ni-O_long decreases while the oscillation amplitude of Ni-O_short emerges and dominates, leading to a slight shrinkage of the average Ni-O bond caused by elevation of the valence state of Ni^[38,39]. In conclusion, Ni is also an active site toward catalysis of OER. With the above analyses, it can be concluded that Fe, Co, and Ni are active sites of FCNCuMO_X@CNT for catalysis of OER, consistent with the findings observed from in-situ Raman study.

AEMWE characterizations

FCNCuMO_X@CNT and FCNCuM@CNT perform well as catalysts for OER and HER, respectively, in AWE, and were further applied as the catalysts for anode and cathode, respectively, in AEMWE. For AEMWEs, the contact between the electrodes, anode and cathode, and the anion exchange membrane is critical, concerning the contact resistance of the MEA. In this regard, suitable ubstrates for support of the catalysts are used to minimize the contact resistance to improve the electrochemical performances of the assembled AEMWE. Here, FCNCuMO_X@CNT and FCNCuM@CNT were loaded on NM and CP to serve as the anode and cathode, respectively for AEMWEs. For comparison purposes, corresponding quaternary catalysts and noble metal-based AEMWEs were also fabricated for characterizations. The polarization curves of the seven AEMWEs are shown in Supplementary Figure 15, with the current densities achieved at 1.8 and 2.0 V displayed in Figure 5A for comparison. In addition, the current densities achieved at 1.8 V (I_1.8V) from AEMWE measurements are correlated with the sums of η₁₀ (η_{10_sum}) determined from AWE measurements as presented in Supplementary Table 3 and the attached figure. Evidently, I_1.8V correlates well with η_{10_sum}, with lower η_{10_sum} giving higher I_1.8V. Several points are further observed from Figure 5A, Supplementary Figure 15, and Supplementary Table 3. First, the quinary catalysts, FCNCuMO_X@CNT and FCNCuM@CNT, based AEMWE and AWE achieve the highest current densities and lowest combined overpotentials, confirming again the enhancement induced from the positive synergy of Fe, Co, Ni, Cu, and Mo. Second, the Fe-absent catalysts, CNCuMO_X@CNT and CNCuM@CNT, based AEMWE and AWE exhibit the lowest current densities and highest combined overpotentials. This can be explained as follows. The OER occurring at the anode is the bottleneck of the water electrolysis process, because of the high overpotential required to trigger the sluggish four-electron transfer reactions involved. Consequently, the OER activity of the anode catalyst plays a dominant role in determining the electrochemical performance of the AEMWE and AWE. And, as discussed in an early section, Fe is a critical component to OER-activities. The Fe-absent catalysts, CNCuMO_X@CNT and CNCuM@CNT, based AEMWE and AWE thus exhibit the worst performance among all characterized samples.

Figure 5. (A) Summary of current densities achieved at 1.8 and 2.0 V by sample AEMWEs; (B) Operation stability of FCNCuMO_X@CNT and FCNCuM@CNT based AEMWE under current density of 500 mA cm^-2 for 100 h. CNTs: Carbon nanotubes; AEMWE: anion exchange membrane water electrolysis.

In addition to polarization curves, EIS was conducted to examine the charge transport and charge transfer resistances involved in the six sample AEMWEs. Supplementary Figure 16 shows the Nyquist plots for the six sample AEMWEs recorded at 1.8 V and 50 °C. These Nyquist plots were fitted with an equivalent circuit model (inset of Supplementary Figure 16), composed of five electrical components, including a system resistor (R₀), two charge transfer resistors (R₁ for the cathode and R₂ for the anode), and two constant phase element capacitors (CPE₁ for the cathode and CPE₂ for the anode). The fitting results for R₀, R₁, and R₂ are summarized in Supplementary Table 4 for comparison, from which three key points can be made. First, R₀, obtained as the intercept on the horizontal axis at high frequencies, corresponds to the high frequency resistance (HFR) of the AEMWE and is mainly contributed by the contact resistances between the two electrodes and the ion exchange membrane within the MEA^[1]. The R₀ values remain almost constant across the six sample AEMWEs since the same kinds of substrates, NM for the anode and CP for the cathode, were used for assembly of the six sample AEMWEs. Second, the values of R₂, the charge transfer resistance at the anode (OER), are significantly higher than those of corresponding R₁, the charge transfer resistance at the cathode (HER)^[40], since OER, a four-electron transfer reaction, is much more sluggish than HER, a two-electron transfer reaction. Third, the current density delivered by the AEMWE correlates well with the value of R₂ - the higher R₂, the lower current density. This phenomenon reflects the fact that OER is the bottleneck process in AEMWE operations.

The best performing quinary catalysts based AEMWE was further characterized for its operation stability. Figure 5B (in terms of ∆V/V₀) and Supplementary Figure 17 (in terms of V) show the remarkable operation stability of the FCNCuMO_X@CNT@NM//FCNCuM@CNT@CP couple based AEMWE operated under a high current density of 500 mA cm^-2 for 100 h, with negligible decay of 0.014% in applied cell voltages. After the stability test, physical characterizations of the catalysts, including scanning electron microscopy (SEM, Supplementary Figure 18) and transmission electron microscopy (TEM, Supplementary Figure 19), were conducted to confirm that the catalysts maintained their morphological and crystalline structures, demonstrating outstanding electrochemical and mechanical stability. The present developed high entropy catalysts, FCNCuMO_X@CNT and FCNCuM@CNT, are thus proven excellent electrode catalysts for AEMWEs. Supplementary Table 5 compares the electrochemical performances of the present high entropy catalysts with those of the state-of-the-art non-precious metal based catalysts reported in recent years for AEMWEs. Evidently, the FCNCuMO_X@CNT and FCNCuM@CNT-based AEMWE stands out among them. Particularly, in terms of the specific activity, defined as the current density achieved at 2 V per unit of combined mass loading of the anode and cathode catalysts, the FCNCuMO_X@CNT and FCNCuM@CNT-based AEMWE largely outperforms all other samples with an ultrahigh specific activity of 293 mA mg^-1, making them highly efficient catalysts for AEMWEs.

DFT study

To explore the roles and significances of constituent elements of the catalysts during HER and to understand the key factors influencing the catalytic performances of the samples, DFT calculations were conducted to investigate the synergistic effects of multiple metal active sites in alkaline HER. Hydrogen adsorption energy, ΔG_H*, has been a popular descriptor for HER activities. Theoretically, a value of ΔG_H* close to 0 indicates that the active site binds with H neither too strongly nor too weakly, facilitating the HER^[41,42]. Nevertheless, this concept may be applicable only to HER in acidic media where hydrogen ions are abundant^[43]. For HER in alkaline media, the hydrogen ions in the electrolyte need to be provided by cleavage of water molecules. Consequently, ΔG_H* alone is insufficient to correlate with activities in alkaline HER. Figure 6A summarizes experimentally determined η₁₀’s achieved by the FCNCuM HEA catalyst and corresponding five quaternary alloy catalysts. Evidently, FCNCuM exhibits the highest HER activity, with the lowest overpotential of 38 mV. Nevertheless, the ΔG_H* of FCNCuM (yellow line) is not the one closest to zero [Figure 6B], inconsistent with the experimental finding if ΔG_H* is taken as the sole HER descriptor. Interestingly, the FCNCu quaternary alloy (Mo-absent, purple line) possesses a ΔG_H* closest to zero among the six alloy samples, but exhibits the highest overpotential [Figure 6A]. With the above observations, we speculate that in alkaline HER, in addition to ΔG_H*, consideration of the activation energy of the water dissociation reaction is also necessary. This speculation also aligns with the experimental finding for Tafel slopes determined for the six alloy catalysts. As evident from Supplementary Figure 13B, the Tafel slopes of the six alloy catalysts are either between 40 and 120 mV dec^-1 (FCNCuM and FCNM) or at around 120 mV dec^-1 (CNCuM, FNCuM, FCCuM, and FCNCu), indicating comparable Volmer and Heyrovsky steps or a single RDS of Volmer, respectively. In alkaline media, both Volmer (H₂O + e^- → H* + OH^-) and Heyrovsky (H₂O + H* + e^- → H₂ + OH^-) steps involve water dissociation, emphasizing the importance of water dissociation and also supporting the speculation.

Figure 6. (A) Summary of η₁₀’s of six alloy samples; (B) ΔG_H* for six alloy samples and Pt; (C) Activation energies of three representative adsorption models; (D) Illustration of transition along reaction path for FCNCuM during water dissociation.

Calculations for activation energies of water dissociation were performed for three representative models, including coupling of a strong H adsorption site with a strong OH adsorption site (H_strong+OH_strong), a strong H adsorption site with a weak OH adsorption site (H_strong+OH_weak), and a weak H adsorption site with a weak OH adsorption site (H_weak+OH_weak). ΔG_H* and ΔG_OH* on all possible sites, including on-top, bridge, and hollow, were first calculated for choices of the H_strong, H_weak, OH_strong, and OH_weak sites. Supplementary Table 6 summarizes some representative strong and weak sites for H and OH. The strongest and weakest adsorption sites, including FeFeFe hollow site as the H_strong site, FeCuCu hollow site as the H_weak site, Mo on-top site as the OH_strong site, and FeCoNi hollow site as the OH_weak site, were chosen to construct the three models. As evident from Figure 6C, the H_strong+OH_strong case gives the lowest activation energy (0.71 eV) required for water dissociation, which is most favorable for breaking the bond between H and OH in water molecules. The pairing of adsorption sites of H and OH and resulting activation energies for water dissociation of the three models are summarized in Supplementary Table 7 for comparison. Figure 6D and Supplementary Figure 20 illustrate the transition along the reaction path for the model of FCNCuM HEA during the water dissociation process. A water molecule is first adsorbed at the on-top site of Mo, a strong OH adsorption site. The dissociation of the water molecule proceeds, with the liberated H settling to a neighboring FeFeFe hollow site, a strong H adsorption site. It is worth mentioning that the case of H_weak+OH_strong is absent from the discussion because of the unique nature of HEAs. HEAs consist of five or more randomly distributed metal elements in the lattice, meaning that different adsorption sites are randomly dispersed on the catalyst surface. During the simulation, it was observed that H tends to be attracted to strong hydrogen adsorption sites located in the neighborhood of the strong OH adsorption site in the dissociation process, making it unlikely to have H converge to weak hydrogen adsorption sites. Consequently, this scenario is unlikely to occur in our HEA systems.

Supplementary Figure 21 summarizes the relative adsorption ability of the five constituent elements of FCNCuM toward H and OH, from which some conclusions can be drawn. First, Fe is an element with a strong hydrogen adsorption. It can assist in water dissociation but has difficulty desorbing hydrogen, which is unfavorable for HER. Second, Cu is an element with a very weak hydrogen adsorption ability. Although Cu can easily desorb hydrogen, it may cause an excessively high activation energy for water dissociation, making water cleavage and thus hydrogen ion production difficult, thus hindering HER. Third, Mo has a strong hydroxide adsorption ability, which can help lower the activation energy for water dissociation, making water cleavage and thus hydrogen ion production easier, thereby driving HER. Fourth, Co and Ni are elements with moderate hydrogen adsorption abilities. As evident from Figure 6B, the ΔG_H* of the Ni-absent sample, FCCuM, is more positive as compared to that of the Co-absent sample, FNCuM, indicating that FCCuM (w/o Ni) has a weaker hydrogen adsorption ability. This suggests that Ni has a stronger hydrogen adsorption ability as compared to that of Co. Therefore, Ni is more effective than Co in assisting cleavage of the O-H bond. Furthermore, the hydrogen adsorption ability of Ni is weaker than Fe, making dehydrogenation relatively easier, which is favorable for HER. In summary, the results indicate that Mo can stabilize OH^-, aiding bond cleavage, and Ni has a moderate hydrogen adsorption ability, which helps in bond cleavage without making desorption too difficult. Therefore, Mo and Ni are the two critical elements for catalysis of HER in alkaline media. The synergy between them ensures high HER activities. This is consistent with the experimental finding obtained in this work for alkaline HER, where FCNCu (w/o Mo) and FCCuM (w/o Ni) exhibit the poorest HER performance. It also explains why NiMo alloys are in general excellent HER catalysts^[44].