Catalyst synthesis. Fe0.5NC catalyst was prepared from FeⅡ acetate (98%, Sigma-Aldrich), 1,10-phenanthroline (phen; ≥99%, Sigma-Aldrich), and ZnII zeolitic framework (ZIF-8, Basolite Z1200, BASF). One gram of the precursor mixture containing 0.5 wt% Fe and a mass ratio of phen/ZIF-8 of 20/80 were mixed using dry ball-milling for four cycles of 30 min at 400 rpm. A ZrO2 crucible with 100 ZrO2 balls of 5 mm diameter was used in this procedure. The resulting precursors were pyrolyzed at 1323 K under Ar flow for 1 h. Fe0.5NC contained Fe content of ca. 1.5 wt%, as confirmed by inductively coupled plasma-optical emission spectroscopy (ICP-OES). For the synthesis of Fe0.5NC-Pt, 0.5 wt% Pt was impregnated on Fe0.5NC by conventional wet impregnation and subsequent H2 reduction. 0.1 g Fe0.5NC was dissolved and dispersed in 100 mL of deionized water containing 1.3 mg of H2PtCl6·6H2O (≥ 37.5%, Pt basis, Merck), and then the solvent was evaporated at 353 K. The resultant powder sample was dried at 353 K under vacuum overnight and reduced at 523 K for 3 h under H2 flow (5%, 200 mL min− 1). Commercial Pt nanoparticle (HiSPEC 3000, 20 wt% Pt) was purchased from Thermo Fisher Scientific. For the synthesis of Fe2O3/NC, the NC catalyst was first prepared identically to Fe0.5NC, except for the addition of FeⅡ acetate during the ball-milling step. Note that owing to the presence of trace amounts of Fe impurities in the commercial ZIF-8 (> 100 ppm),35 Fe-free ZIF-8 was used for the synthesis of NC, which was prepared by mixing 2-methylimidazole (2-MeIm; 99%, Sigma-Aldrich) and Zn nitrate hexahydrate (Zn salt; 98%, Sigma-Aldrich) in aqueous solution (Zn salt/2-MeIm/water molar ratio of 1/60/2228).36 A, Fe2O3/NC was synthesized by a dry ball-milling of 98 mg NC and 2 mg commercial Fe2O3 (96%, Sigma-Aldrich) for four cycles of 30 min at 400 rpm without heat treatment. The Fe content in Fe2O3/NC was ca. 1.3 wt%, as confirmed by ICP-OES.
Physicochemical characterization. X-ray diffraction (XRD) patterns were obtained using a high-resolution X-ray diffractometer (X'Pert PRO MPD, PANalytical) equipped with a Cu Kα X-ray source. The XRD patterns were measured at an accelerating voltage of 60 kV and current of 55 mA, with scan rate of 10° min− 1 and step size of 0.02°. Raman spectra were obtained using a NRS-5000 series Raman spectrometer (JASCO) with 633 nm laser excitation. XPS measurements were performed using K-Alpha+ (Thermo Scientific) instrument equipped with a micro-focused monochromator X-ray source. The binding energy used for the peak deconvolution of the XPS-Pt spectra (for 4f7/2) was 72.4 and 73.8 eV for PtⅡ and PtⅣ, respectively, and the spin-orbit splitting for 4f5/2 and 4f7/2 peaks was 3.33 eV.37 ICP-OES analysis was performed using Optima 4300 DV (PerkinElmer Inc.) for determining Fe contents. Fe K-edge or Pt L3-edge X-ray absorption spectroscopy (XAS) signals were collected in the transmission mode at the Pohang Accelerator Laboratory (8C, Nano XAFS). To compensate any energy shift during data acquisition, the XAS energy scale was calibrated using each metal foil before the measurements. The EXAFS analysis was conducted using Athena and Artemis implemented in the Demeter program package. The EXAFS data in k space was Fourier-transformed with the Hanning window function (dk = 0.5 Å−1) after k3-weighting. Curve fitting of the FT-EXAFS of Fe0.5NC-Pt was carried out in the R-range of 1.1–2.3 Å with phase correction by including Pt–N and Pt–Cl scattering paths. The amplitude reduction factor (S02) for Pt was determined to be 0.83 from the curve fitting of the EXAFS of the Pt foil. The wavelet-transform of k3-weighted EXAFS data was analyzed with phase correction using the HAMA Fortran code. A Morlet function was used for the mother wavelet function (η = 10, σ = 1). The 57Fe Mössbauer spectrum was acquired using a 57Co source in Rh. The measurement was performed by keeping both the source and absorber at room temperature. The spectrometer was operated with a triangular velocity waveform, and a NaI scintillation detector was used to detect γ-rays. TEM, EDS, and HAADF-STEM analyses were performed using TECNAI G2 F30 S-Twin (FEI), TECNAI F20 UT (FEI), and Titan 80–300 (FEI), respectively.
Electrochemical measurements. The electrochemical properties were investigated using a VMP-300 potentiostat (Bio-Logic) in a three-electrode cell equipped with a graphite rod counter electrode and saturated Ag/AgCl reference electrode (RE-1A, EC-Frontier). A 0.1 M HClO4 electrolyte was prepared using ultrapure water (> 18 MΩ, Sartorious) and concentrated HClO4 (70%, Sigma-Aldrich). The reference electrode was calibrated against a Pt electrode in H2-saturated electrolyte to convert the potentials to the RHE scale correctly. The catalyst ink was prepared by dispersing 10 mg of the catalyst in an aqueous solution (868 µL deionized water, 7 µL isopropanol, and 80 µL 5 wt% Nafion solution). After ultra-sonication of the above suspension for 30 min, the working electrodes were prepared by dropping 15 µL of catalyst ink onto a glassy carbon disk (0.196 cm2) of the RDE (01169, ALS). The catalyst loadings were set to 800 µg cm− 2 for Fe0.5NC or Fe0.5NC-Pt, and 10 µgPt cm− 2 for Pt nanoparticles. The ORR polarization curves were recorded with a scan rate of 10 mV s− 1 and rotation speed of 900 rpm in an O2-saturated 0.1 M HClO4 electrolyte. ORR Faradaic currents were obtained by subtracting the polarization curves measured in an Ar-saturated electrolyte. The PRR polarization curves were measured with a scan rate of 1 mV s− 1 and a rotation speed of 900 rpm in an Ar-saturated 0.1 M HClO4 electrolyte containing 1.3 mM H2O2.
Online GDE/EFC/ICP-MS measurements. The GDE was fabricated by spraying catalyst ink onto a mesoporous layer (MPL) deposited on carbon paper. On a carbon paper with a 20 wt% polytetrafluoroethylene (PTFE) content (2 × 2 cm2, TGP-H-090, Toray), highly hydrophobic carbon MPL was first fabricated by spraying an ink emulsion containing 100 mg Ketjen black EC-300J, 400 mg PTFE (60 wt%, Sigma-Aldrich), and 20 mL isopropyl alcohol (IPA, 99.5%, Sigma-Aldrich), followed by heat-treatment at 513 and 613 K under Ar atmosphere for 30 min each. The resulting MPL had a Ketjen Black EC-300J loading of 2 mg cm− 2. Thereafter, catalyst ink (25 mg catalyst + 10 mL isopropanol + 250 µL 5 wt% Nafion solution) was sprayed onto the MPL to reach target catalyst loadings of 400 µg cm− 2. The active catalyst area of the GDE was 0.096 cm2. Fe dissolution was monitored using ICP-MS (7500ce, Agilent) coupled with a GDE-based EFC. The EFC is equipped with a U-shaped channel with a diameter of 1 mm and an opening for contact with the GDE. A graphite tube counter electrode (inner diameter = 1 mm) was placed at the inlet, and an Ag/AgCl reference electrode was connected to the outlet of EFC. A 0.1 M HClO4 electrolyte, de-aerated using a degasser (AG-32-01, FLOM Corp.), was flowed into the EFC at a flow rate of 200 µL min− 1. The electrolyte was mixed with 0.2 M HNO3 containing 5 ppb 187Re as an internal standard using a Y-connector (mixing ratio = 1:1). Ar or O2 gases flowed into the graphite serpentine gas channel at a rate of 5 mL min− 1. GDE was pre-treated with 200 fast CV cycles (200 mV s− 1) in the potential range of 1.0 − 0.0 VRHE at 298 K before the electrochemical measurement. After stabilization at the desired temperature for ca. 10 min, Fe dissolution was analyzed under potentiodynamic or potentiostatic conditions. Fe dissolution under potentiodynamic conditions was analyzed with three slow CV cycles (5 mV s− 1) in the potential range of 1.0 − 0.0 VRHE at various temperatures (298, 313, 333, and 353 K) under Ar or O2 flow. Fe dissolution under potentiostatic conditions was analyzed by holding a 0.6 VRHE at 353 K under Ar or O2 flow. ORR polarization curves were recorded with a scan rate of 5 mV s− 1 at 298 K. A manual iR compensation (MIR) program was used to compensate 85% of ohmic drop during the electrochemical measurements and 15% post-corrected. The uncompensated resistance was measured using electrochemical impedance spectroscopy (EIS) at open circuit potential (OCP).
Computational Details. All DFT calculations were conducted based on the First-principle using the Vienna ab initio simulation program (VASP) to understand the structurally enhanced stability of Fe-N4 by impregnating Pt-based coordination compounds such as single-atom Pt, Pt-N4, Pt-N4Cl2, and Pt-N2Cl2 (including trans- and cis-structure).38 Projector Augmented Wave (PAW) pseudo-potential was implemented to efficiently describe by substituting the interaction of core electrons.39 The generalized gradient approximation (GGA) with the Revised Perdew-Burke-Ernzerhof (RPBE) functional was adopted to describe the electron exchange-correlation functional effects.40,41 In addition, we used the Methfessel-Paxton smearing method for better ionic and geometry optimization.42 To fix the number of plane waves, a plane-wave basis set was defined to effectively expand the Kohn-Sham wave functions of valence electrons with the kinetic energy cut-off of 520 eV. To investigate and compare the cohesive energies of Fe atom on the prepared models, at least 20 Å vacuum space along the z-direction was set to avoid undesirable interactions between the top and bottom of the unit cell box. The k-point mesh was sampled by (9 × 9 × 1) to structurally optimize pristine Fe-N4 models with/without Pt-coordinated compounds (single-atom Pt, Pt-N4, Pt-N4Cl2, and trans-/cis-Pt-N2Cl2) to integrate the Brillouin zone and calculate the total energies of the designed structures. The total energy was adopted within 1 x 10− 4 eV for the full ionic relaxation step and the maximum atomic forces were set below 0.05 eV Å−1.
Model design. Based on the experimental EXAFS analysis results (Fig. 3b, c, and Supplementary Table 4), we generated a variety of possible candidate models, such as Fe-N4, Pt-N4, Fe-N4/Pt-N4Cl2, and Fe-N4/Pt-N2Cl2. Initially, it was crucial to define the most stable pristine GNS model as the starting structure (more details can be found in our previous report43). Fe-N4 and Pt-N4 as reference models were designed by substituting Fe, Pt, and N to the carbon atom by removing the double vacancies of pristine GNS, as shown in Supplementary Fig. 16. In Supplementary Figs. 17–25, possible Fe-N4 sites were introduced into the pre-designed Pt-based coordination compounds to generate the possible configurations of Pt-impregnated Fe-N4. In particular, the configurations of Pt-N2Cl2 were defined as trans-/cis-structures on the edge sites or basal planes with and without carbon vacancies, as shown in Supplementary Fig. 20. Specifically, possible configurations of Fe-N4 with various Pt-based coordination compounds were generated as candidates by calculating the formation energies (\({E}_{f}\)) for the most stable structures with respect to different volumes or atomic ratios in specific volumes. \({E}_{f}\) was calculated using Eq. 1:44
\({E}_{f}={E}_{total}-{\sum }_{i}{\mu }_{i}{x}_{i}\) (Eq. 1)
where \({E}_{total}\) is the DFT-calculated total energy, and \({\mu }_{i}\) and \({x}_{i}\) are the chemical potential and quantity of element \(i\) in the designed model structures, respectively. After identifying the most stable structure in each model through formation energy calculations, the cohesive energies were calculated to compare the stabilities of the Fe atoms and select the representative Pt-impregnated FeN4 (Fig. 3a and Supplementary Table 3). The cohesive energies (\({E}_{coh}\)) are described by Eqs. 2:45
\({E}_{coh}={E}_{total}-{E}_{total w/o Fe atom}-{E}_{Fe atom}\) (Eq. 2)
where the cohesive energy (or binding energy) is described by the \({E}_{total}\), \({E}_{slab w/o Fe atom}\), and \({E}_{Fe atom}\) which are the total energy of the designed model, DFT energy without Fe atom in the designed model, and free single Fe atom, respectively, and are all derived from DFT calculation.
PEMFC operation. A membrane electrode assembly (MEA) was prepared by a decal process via doctor blade coating to compare the single cell performance of Fe0.5NC and Fe0.5NC-Pt as cathode catalysts with all the Pt/C-based anodes. The cathode slurries were prepared by dispersing 0.26 g catalysts (Fe0.5NC and Fe0.5NC-Pt) in an aqueous solution containing 0.18 g deionized water, 0.14 g isopropanol, 0.14 g n-propanol, and 0.431 g 20 wt% Nafion solution (1000 EW, DuPont Fuel Cells). Moreover, the anode slurries were prepared by mixing 0.35 g catalysts (37.7 wt% Pt/C, TANAKA Precious metals) in an aqueous solution containing 1.98 g deionized water, 0.40 g isopropanol, and 2.30 g 5 wt% Nafion solution (1100 EW, DuPont Fuel Cells). The Nafion ionomer content in both the anode and cathode slurries was adjusted to 25 wt% of the total electrode solid amount. The slurries prepared in the N2-filled glove box were stirred in an ultrasonic water bath for 2 h, followed by a three-roll milling process (EXAKT 50I) to effectively break up the agglomerates in the slurries. The mixture was then stirred in an ultrasonic water bath for 1 h for better dispersion. Each cathode and anode slurry was coated onto a PTFE film as a decal substrate using a doctor-blade coater. The coated electrodes were dried at 353 K in an N2-purged vacuum oven, and the prepared cathode was hot-pressed onto a Nafion 211 membrane with an anode electrode at 100 bar and 383 K for 10 min. After cooling for 10 min via the cooling press, we successfully obtained MEA with an active area of 25 cm2. The catalyst contents of the cathode and anode electrodes were controlled at approximately 4.0 and 0.25 mg cm− 2, respectively. The cell performance of MEA was evaluated using an EIS46 potentiostat (Bio-Logic, HCP-803). The test cell includes MEA, gas diffusion layers (320 µm thickness, JNTG), a bipolar plate constituting two channels of graphite, and a current collector plate derived from gold-coated copper blocks to evaluate at single-cell scale. Before evaluating the cell performance of the MEA, we performed an activation process, in which it was necessary to remove impurities on the active sites of Pt and wet the Nafion membrane for effective proton transfer. This activation process was conducted in the following steps: (1) H2 and O2 as reactant gases with 500 and 1,500 mL min− 1 flow rates with the condition of 100% relative humidity (RH) at 353 K were provided for the anode and cathode, respectively. (2) A 0.1 V per step was applied for 30 s in the cycling range of 0.3–0.7 V until the performance of MEA at 0.6 V was saturated and stabilized. To evaluate the practical cell performance of MEA, polarization curves were obtained at ambient pressure, ranging from open circuit voltage (OCV) to 0.2 V at a scan rate of 20 mV s− 1 by linear sweep voltammetry under the same operating conditions as that of the activation process.