Synthesis of the 3d-5d hybrid Fe,W-N-C catalyst and structural characterization
High-energy ball milling has enough power to break and reform chemical bonds, often used to construct defects on the support, which could anchor isolated metal atoms/clusters, while the surface wearing of the milling balls is often overlooked. Figure 1a illustrates the tungsten atoms falling off the tungsten carbide balls due to enormous shear and impact forces. When the raw material contains only carbon black, the fresh tungsten atoms with higher surface energy tend to combine with carbon atoms and aggregate into tungsten carbide nanoparticles during ball milling or subsequent pyrolysis steps (WC-C, Supplementary Figs. 1 and 2, Supplementary Table 1). In contrast, when sufficient phthalocyanine (Pc) complex is added, the tungsten atoms will stay isolated on the carbon black support (W-N-C, Supplementary Figs. 3 and 4). This could be attributed to the unique macrocyclic structure of Pc, with a central cavity occupied by two hydrogen atoms and surrounded by four nitrogen atoms in square planar geometry. This cavity is well-known for forming many stable metal complexes by replacing the two hydrogen atoms with metal atoms, such as Cu, Fe, Co, Ni, etc. When the two hydrogen atoms are knocked off during the ball milling process, the vacant cavity becomes an active trap for the fresh tungsten atoms, resulting in atomically dispersed W-N sites.
Inspired by the successful construction of the W-N site, we introduced one more component during the ball milling process: FePc, which possesses a similar structure to that of Pc but with two hydrogen atoms replaced by a Fe atom in the cavity, and constructed a 3d-5d Fe-W diatom catalytic site. As illustrated in Fig. 1b, in the highly energetic environment of ball milling (e.g., huge impact faces and high localized temperatures), the adsorption of both Pc and FePc molecules on the carbon black surface tends to reach thermodynamic equilibrium. Driven by the electron donor-acceptor interactions, the Pc and FePc tend to partially overlap and form N/FeN on the carbon surface at equiblium. When the N portion of the N/FeN site traps a highly active tungsten atom scratched off from the milling balls, the Fe-N/W-N diatomic site is successfully constructed. After that, a pyrolysis process at 900 °C for 2 h in Ar atmosphere stabilizes the sites without sacrificing the coordinating N atoms or causing the aggregation of metal atoms. It must be emphasized that 100% of the metals were utilized in the synthesis process. In the catalyst design, the Fe content is determined by the maximum amount of FePc allowed by atomic dispersion on carbon black, and the W content depends on the ball milling time to achieve the 1:1 atomic ratio. Extra milling time will introduce more W than that can be trapped and lead to the formation of tungsten carbide nanoparticles. Similarly, if the amount of Pc is insufficient, it will be difficult to introduce enough N sites on the carbon black surface to fix the W atoms that fall off the ball mill, resulting in some W atoms forming WC nanoparticles (Supplementary Fig. 5). In the obtained catalysts, named Fe,W-N-C, the contents of Fe and W were optimized at 1.25 wt% and 4.02 wt% (the atom ratio of Fe and W is close to 1:1) as determined by inductively coupled plasma-optical emission spectroscopy (ICP-OES, Supplementary Table 1). To demonstrate the versatility of this catalyst construction strategy, we extended it to other representative 3d-metals (such as Co and Ni) by simply replacing FePc with other metal phthalocyanine complex and successfully synthesized Co,W-N-C and Ni,W-N-C 3d-5d hybrid dual-atom catalysts (Supplementary Figs. 6 and 7, Supplementary Table 1).
The dual-atom configuration of Fe,W-N-C catalyst was investigated using scanning electron microscopy (SEM) and transmission electron microscopy (TEM). As shown in Fig. 2a, b the Fe,W-N-C catalyst is composed of a series of pearl-like spherical carbon particles with diameters of 50-80 nm, and there are no obvious metal/ metal Pc nanoparticles in the carbon matrix. These results are consistent with the X-ray diffraction (XRD) patterns, in which only two broad diffraction peaks belonging to the (002) and (101) crystal planes of carbon were detected (Fig. 2c). The atomically dispersed Fe,W atom pairs were directly imaged by HAADF-STEM at the atomic scale. As shown in Fig. 2d, a large number of isolated bright-faint dot pairs are uniformly dispersed on the carbon black surface, marked by green rectangles. The pronounced brightness difference in each pair of atoms is due to the sensitive Z-contrast of heavy elements. Since only two metal elements are possible in the system, the pair can be safely identified as 3d-Fe (faint) and 5d-W (bright) atomic pair without guesswork. Supplementary Fig. 8 exhibits the corresponding intensity profile for two typical Fe-W bimetallic pairs at site 1 and site 2 in Fig. 2d, and it was found that the distance between the two metal atoms is about 0.55 nm, which is consistent with the distance in the atomic structure model of the 3d-5d hybridized Fe-N/W-N diatomic site of the Fe,W-N-C catalyst pointed out in the density functional theory calculations part (Supplementary Fig. 9). In addition, the characteristic peaks of Fe and W atoms were both found in the electron energy loss spectroscopy (EELS) spectrum corresponding to the 1 nm × 1 nm small area HAADF-STEM image, providing another direct evidence of the existence of Fe and W diatomic sites (Fig. 2e-g). Since HAADF-STEM images describe two-dimensional projections, the projected distances between Fe and W atoms could differ (tagged by blue circles in Fig. 2d), depending on the angle between the W-Fe axis and the incident beam. The lower magnification HAADF-STEM image and corresponding energy-dispersive X-ray spectroscopy (EDX) mapping also revealed the uniform distribution of C, N, W, and Fe elements in the Fe,W-N-C catalyst (Fig. 2h and Supplementary Fig. 10). High-resolution X-ray photoelectron spectroscopy (XPS) results confirmed the existence of sufficient N (from Pc and FePc molecules) in the Fe,W-N-C catalyst, which not only promoted the anchoring of Fe and W atoms on the carbon black surface but also generated graphitic nitrogen to improve electron transfer in the carbon skeleton (Supplementary Fig. 11).
To gain more insight into the electronic structure and coordination environment of the 3d-Fe atom and 5d-W atom in Fe,W-N-C catalyst, X-ray absorption spectroscopy (XAS) was collected. As shown in Fig. 3a, a pre-edge peak at around 7114 eV was observed in the Fe K-edge X-ray absorption near edge structure (XANES), which is characteristic of the 1 s to 4p electric dipole transition, along with the charge transfer from ligand to the metal center, and could be recognized as the fingerprint of the porphyrin-like planar Fe-N. The absorption edge of the Fe,W-N-C catalyst is situated between FeO and FeO, indicating that the oxidation state of Fe is between +2 and +3. To elucidate the effect of the neighboring 5d-W site on the chemical state of the Fe center, a single atom Fe-N-C catalyst without W was prepared as a reference through a similar procedure (Supplementary Figs. 12 and 13). Notably, the Fe K-edge absorption of the Fe,W-N-C catalyst is significantly lower than that of the Fe-N-C catalyst, implying that the introduction of 5d-W site neighboring to the 3d-Fe site can effectively regulate the oxidation state of the Fe atom, which could prevent the electrochemical dissolution of the Fe center and enhance its electrocatalytic stability (vide infra, Supplementary Fig. 14). The Fourier transformed k-weighted Fe K-edge extended X-ray absorption fine structure (EXAFS) spectrum of the Fe,W-N-C catalyst exhibited a prominent peak at around 1.5 Å, which could be assigned to the Fe-N contributions in the first shell (Fig. 3b). However, the main peak of the Fe-N-C catalyst is located at 1.41 Å (similar to the position shown in the FePc complex, Supplementary Fig. 15). The difference in peak position of Fe,W-N-C catalyst and Fe-N-C catalyst is due to the introduction of W-N sites close to the Fe-N sites in the Fe,W-N-C catalyst, which reduces the electron transfer from the Fe atoms to the surrounding environment, resulting in a decrease in the oxidation state of the Fe atoms and an increase in the Fe-N distance in Fe,W-N-C. Compared with Fe Foil, there is no observable Fe-Fe scattering peak at 2.2 Å in both of the Fe,W-N-C and Fe-N-C catalysts. This confirms the absence of Fe aggregates and verifies that Fe atoms exist in an atomically dispersed form. Due to the high resolution in R-space and k-space, the wavelet transform (WT)-EXAFS analysis was carried out to further reveal the isolated state of the Fe atoms. The Fe,W-N-C catalyst demonstrated an intensity maximum at k ~ 4.7 Å responding to the Fe-N bonds, and no metallic Fe-Fe scattering signal can be detected (Fig. 3c, and Supplementary Fig. 16). As shown in Fig. 3d-f and Supplementary Fig. 17, the W atoms also exhibited an atomically dispersed nature with an oxidation state between 0 and +6.
To clearly elucidate the coordination configurations of Fe and W atoms in the Fe,W-N-C catalyst, EXAFS fitting curves were simulated. As displayed in Fig. 3b, e and Supplementary Table 2, the fitting results indicated that both Fe and W atoms are coordinated with four N atoms in the first shell, with bond lengths of 1.90 Å and 2.09 Å, respectively. Since weak peaks at above 2.0 Å are observed in both the Fe K-edge EXAFS and W L-edge EXAFS spectra, combined with the weak WT-EXAFS intensity maximum in the k range of 5-7 Å, which is higher than that of the coordination with C and different from the Fe-Fe or W-W peak, suggesting that a nonnegligible long-range interaction between the Fe and W atoms may occurred (Fig. 3b, c, e, f and Supplementary Fig. 17). We included the outer shells of Fe and W in EXAFS fitting. Although the distance between Fe and W is too long (about 5.5 Å revealed by HAADF-STEM) to be accurately fitted in EXAFS, the simulated spectra fitted to the 4 shell of Fe (R = 3.33 Å) and the 5 shell of W (R = 4.29 Å) overlapped well in the experimental spectra, which further proved that the atomic structure model of the 3d-5d hybridized Fe-N/W-N diatomic site of the Fe,W-N-C catalyst pointed out in Fig. 3g is valid. Furthermore, in order to further verify the structure features, the XANES simulations for this representative structure (R = 7 Å cluster) using the FDMNES code were calculated and shown in Fig. 3h, i. It turned out that the theoretically calculated spectra showed similar features to the experimental spectra, particularly for the shape and the position of the peaks, demonstrating the well-defined structure of Fe,W-N-C catalyst.
The electrochemical ORR performance of the Fe,W-N-C catalyst was assessed using a rotating disk electrode (RDE) in oxygen-saturated 0.1 M KOH electrolytes. To verify the vital role of the Fe-N/W-N diatomic sites in oxygen electrocatalysis, Fe,WC-N-C catalyst (with isolated Fe atom and tungsten carbide nanoparticles coexisting) was synthesized using a similar method (Supplementary Figs. 18 and 19). As shown in Fig. 4a, the diatomic Fe,W-N-C catalyst exhibited a notable ORR catalytic activity with the most positive onset potential (1.03 V) and half-wave potential (E, 0.90 V) among other synthesized catalysts and the commercial Pt/C catalyst. Specifically, the E of the Fe,W-N-C catalyst is 60 mV and 30 mV higher than that of the single atom Fe-N-C catalyst and the Fe,WC-N-C catalyst, confirming the positive effect of the 5d-W species on the single atom 3d-Fe catalyst which can be maximized when the 5d-W species is also in single atom form. Fe,W-N-C also possessed the highest kinetic current density up to 17.14 mA cm at 0.82 V, nearly two times higher than the commercial Pt/C catalyst (Fig. 4b and Supplementary Table 3). Compared with the single atom Fe-N-C catalyst, the 3d-5d hybrid Fe-N/W-N diatomic site exhibited a lower Tafel slope (94 mV dec), revealing its lower oxygen binding energy and faster ORR kinetics (Supplementary Fig. 20, and Supplementary Table 4). The electrochemically active surface areas (ECSAs) of Fe,W-N-C and commercial Pt/C catalysts were estimated and compared by calculating the double-layer capacitance values (Cdl) via cyclic voltammetry curves (Supplementary Fig. 21). The higher Cdl value of the Fe,W-N-C catalyst compared to that of the commercial Pt/C catalyst, indicates its larger ECSA and more approachable active sites. Given its limited BET specific surface area of 75.6 m g (Supplementary Fig. 22), it can be concluded that most of the active sites exist on the carbon black surface.
The electron transfer numbers at various potentials were calculated using the linear sweep voltammetry curves collected at different RDE rotating speeds. As shown in Fig. 4c, d, the limiting current density increases with the rotation speed, and the electron transfer numbers were calculated to be ~ 4 in the potential range of 0.2-0.8 V. Also, a nearly complete 4-electrons transfer pathway and less than 2.55 % HO yield could be observed in a wide potential range from 0.2 to 1.0 V by rotating ring disk electrode (RRDE) measurements, further certifying the high selective to the 4-electrons transfer pathway (Fig. 4e). This is highly desirable since the competing 2-electrons transfer pathway not only reduces the energy efficiency, but also poisons the Fe active site through the Fenton reaction between the generated HO and the Fe sites. To confirm the effect of the W-N site in the Fe,W-N-C catalyst after the Fe-N site being poisoned, the nitrite stripping tests were conducted. As shown in Supplementary Fig. 23, the reduced activity after introducing nitrite indicates that the Fe-N site in the Fe, W-N-C catalyst is the adsorption site of O/ ORR intermediates. It is worth noting that the ORR catalytic activity did not decrease completely to the metal-free level, indicating that the W-N site can also drive the ORR process at a higher overpotential. After that, the gravimetric site density (MSD) of Fe-N site in Fe,W-N-C catalyst was roughly estimated as at least 10.18 µmol g (due to the high hydrogen evolution catalytic activity of the W site at high overpotentials, which will mask the dissolution peak) by analyzing the current density difference between the unpoisoned and poisoned curves. Since the ORR kinetic current density of the Fe,W-N-C catalyst at 0.95 V (vs RHE) in 0.1 M KOH electrolyte is 0.83 mA cm, the turnover frequency (TOF) of the Fe,W-N-C catalyst is estimated to be 4.2 s.
Besides catalytic activity and selectivity, methanol tolerance and durability are also essential indices for evaluating the ORR catalytic performance. As displayed in Supplementary Fig. 24, the Fe,W-N-C catalyst exhibited high methanol tolerance ability with negligible current density decay. In contrast, the commercial Pt/C catalyst suffered from a sharp drop in the current. Additionally, the Fe,W-N-C catalyst demonstrated a satisfying ORR long-term catalytic stability with a relative current density retention of 97.94% after a 10-h chronoamperometric test (Fig. 4f). Due to the unique 3d-5d hybrid Fe-N/W-N dual-atom site, the ORR catalytic performance of Fe,W-N-C is comparable to other reported high-activity noble/non-noble metal catalysts (Supplementary Table 5). The stability of Fe,W-N-C catalyst has been further tested by potential cycling from 0.2 V to 1.1 V. As shown in Supplementary Fig. 25, the 1500th CV curve is largely consistent with the initial one, and the LSV curves indicate that despite the catalyst undergoing 1500 cycles of CV tests, there is no obvious decay in E and limited current density, suggesting the outstanding ORR stability of the Fe,W-N-C catalyst. By comparing the CV and ICP results before and after the accelerated durability test (It was found that the Fe content in the catalyst before and after the accelerated durability test were accounted for 1.25 wt% and 1.24 wt%, respectively), negligible metal leaching was observed, which further demonstrated the enhanced catalytic stability of the Fe,W-N-C catalyst.
Inspired by the good ORR catalytic performance, we further examined its catalytic performance in oxygen evolution reaction (OER) to evaluate its potential application as a cathode for rechargeable ZABs. As illustrated in Supplementary Fig. 26 and Supplementary Table 6, the Fe,W-N-C catalyst exhibited the lowest potential as 1.56 V to deliver a current density of 10 mA cm, among Fe,WC-N-C (1.68 V), Fe-N-C (1.81 V), W-N-C (1.64 V), WC-N-C(1.72 V), and the benchmark IrO (1.59 V) catalysts. Additionally, the smallest Tafel slope of Fe,W-N-C catalyst confirmed its more favorable OER kinetics (87 mV dec, Supplementary Fig. 27). Most importantly, the Fe,W-N-C catalyst also demonstrated high OER catalytic stability for over a 15-h chronopotentiometry test, far exceeding the commercial IrO catalyst (Supplementary Fig. 28).
Given the enhanced ORR and OER catalytic activity and stability, rechargeable ZABs with Fe,W-N-C catalyst loaded on the cathode were assembled to demonstrate their practicability. As shown in Supplementary Figs. 29 and 30, the ZAB with Fe,W-N-C cathode delivered a high specific capacity of 781 mAh g and a corresponding energy density up to 953 Wh kg, outperforming the ZAB with commercial Pt/C + IrO catalyst (specific capacity: 678 mAh g; energy density: 780 Wh kg). Notably, the cell exhibited a stable and repeatable discharge/charge cycling curve for over 10,000 h (over 30,000 cycles) at the current density of 5 mA cm with negligible decay on the discharge/charge voltage (the voltage gap is nearly unchanged and maintained at around 0.72 V) and areal energy density (based on the area of the air cathode), which is comparable to other reported ZABs (Fig. 5a, Supplementary Figs. 31-33). Even when operated at a high current density of 50 mA cm, the Fe,W-N-C based ZAB still can deliver stable performance for over 2000 h (12,000 cycles), far exceeding the ZAB with commercial Pt/C + IrO as the cathode and other reported ZABs (Supplementary Figs. 34 and 35). Surprisingly, the Fe,W-N-C based ZAB exhibited a relatively stable discharge/charge cycling curve of more than 550 h at the current density of 100 mA cm, revealing its possibility of stable operation in high-current power-consuming facilities (Supplementary Figs. 36 and 37). Such high ZAB stability has seldom been achieved to date (Supplementary Table 7). In addition, the Fe,W-N-C cathode ZAB presented a high discharged voltage plateau with a maximum peak power density of 252 mW cm, in contrast to only 140 mW cm for the Pt/C + IrO based ZAB (Fig. 5b), demonstrating its practical application potential as an alternative catalyst to Pt-based catalysts in ZABs. To meet the demand for flexible energy devices, we also assembled a flexible solid-state ZAB with Fe,W-N-C as the cathode. As shown in Fig. 5c and Supplementary Figs. 38 and 39, the solid-state ZAB remains stable even after the iterative bending test. It can also light up a series of LED lights on a luminous wristband, promising its practical application in flexible electronics (Supplementary Fig. 40).
To trace the origin of the notable stability, solid-state ZAB with Fe,W-N-C cathode was assembled and in-situ XAS analyses were carried out as illustrated in Fig. 5d. The ZAB was first discharged and charged at 5 mA cm and then cycled at a higher current density of 10 mA cm after a 2-min rest. As shown in Fig. 5e, the Fe adsorption edge slightly shifts to lower energy compared to the blank state during the first discharge process, and then moves back to higher energy during the charge process. This low-high energy shift repeated in the second discharge/charge cycle, indicating that the discharge/charge processes of ZAB indeed have certain fluctuations in the valence state of the Fe center, which could be attributed to the adsorption of reactants/reaction intermediates on Fe site. The intensity of the pre-edge peak in Fe K-edge XANES is slightly lower than that in ex-situ measurement, suggesting the adsorption of reactants on Fe. Based on the fluctuation of the valance state and the decrease in pre-edge intensity, Fe can be safely identified as the central metal of the active site in both ORR and OER processes. Despite fluctuations in the valence state of Fe during the discharge/charge cycles, the oxidation state remained between +2 and +3 without over-oxidization or over-reduction to cause Fe aggregation or dissolution (Supplementary Fig. 41). Furthermore, nine XANES curves were recorded at the resting state of the ZAB after each discharge/charge cycle, and the oxidation state of the Fe remained stable, which further demonstrates its catalytic stability (Fig. 5f).
The density functional theory (DFT) calculations were conducted to clarify the regulation of 5d W-N sites on neighboring Fe-N sites. Based on the HAADF-STEM-EELS, XPS, and XAS analysis, the atomic configurations of the main active sites in Fe-N-C and Fe,W-N-C catalysts are illustrated in Figs. 6a and 6b (Supplementary data 1). By analyzing the charge density differences of the Fe-N and 3d-5d hybrid Fe-N/W-N sites, it could be found that after modification with the neighboring 5d W-N site, the electron transfer from Fe atoms to the surrounding decreased, indicating the reduced oxidation state of the Fe atom, consistent with the XAS results. Given the oxygen-rich environment in ORR, we analyzed the adsorption and transition pathways of O on the W-N site. As shown in Supplementary Fig. 42, the adsorption of O molecules on W will spontaneously convert from the end-on adsorption to the side-on adsorption, and the O-O bond will break and form a stable W-(O) configuration, which remains the same during the catalytic process. Therefore, the actual active site in Fe,W-N-C catalyst should be denoted as Fe-N/W-NO. To further clarify the regulation of 5d W-NO sites on Fe-N sites and the effects on ORR catalytic activity, we investigated the ORR catalytic process at pure Fe-N sites and 3d-5d hybrid Fe-N/W-NO sites. As illustrated in Fig. 6c, due to the strong adsorption of *OH on the Fe center of the Fe-N-C catalyst, the potential-determining step is the desorption of *OH. Also, the positive ∆G for O activation (*O → *OOH) step reveals its inertness. In contrast, after introducing the neighboring 5d W-NO site, the intramolecular hydrogen bond forms between the H atom in *OOH and the O atom in the W-NO site, accelerating the activation of O (Fig. 6d). Most importantly, the energy required for the desorption of *OH (rate-determining step) was reduced by 0.14 eV, suggesting that the formation of 3d-5d hybrid Fe-N/W-NO is beneficial for optimizing the adsorption energies of ORR intermediates. This maybe due to the electron-withdrawing effect of the adjacent W-NO site on Fe site, which leads to a decrease in the electron density of Fe (the positive charge on Fe in the Fe-N-C catalyst and Fe,W-N-C catalyst are 1.085 and 1.103, respectively, Supplementary Fig. 43), thereby affecting the amount of charge that can transferred to *OH, which is beneficial to the desorption of *OH. The Bader charge transfer results revealed that the *OH accepted less charge from Fe-N/W-NO site than pure Fe-N site, which also confirmed the weak adsorption and stronger desorption ability of *OH on Fe-N/W-NO site. (Fig. 6e, f). The Projected density of state (PDOS) analysis also corroborated that the introduction of neighboring 5d W-NO site reduces the overlap between the Fe-3d orbitals and O-2p orbitals (in *OH intermediates), especially the overlap between Fe-3dz orbital and O-2pz and 2py orbitals, which leads to the weak adsorption of *OH (Fig. 6g, h, Supplementary Fig. 44 and 45).
As for the OER process, since the initial system is not an oxygen-saturated environment, the active site of Fe,W-N-C catalyst is first assumed to be the original Fe-N/W-N site of the catalyst. However, when the W-N site serves as the adsorption site of the HO/OER intermediates, the desorption of *O from the W-N site requires higher energy, which reveals the inertness of OER on the W-N site in the Fe, W-N-C catalyst (Supplementary Fig. 46). This also proves that in the OER reaction, the W-N site will eventually form the configuration of W-NO. Also, under the modification of the W-NO site, lower energy is required for Fe-N site to drive the OER processes. It is further proved that during the OER reaction, the Fe-N site is also the real adsorption site of the HO/OER intermediates, while the W-NO site is used to optimize its electronic structure to promote the occurrence of OER.
To reveal the catalytic stability of the Fe-N/W-NO site, the oxidation state transition of Fe atoms and the demetallation energy of the Fe site during the ORR process were calculated. As shown in Fig. 6i, the oxidation state of Fe atoms tends to be stable in each step under the regulation of the neighboring 5d W-NO site, indicating that Fe atoms will not be over-oxidized or over-reduced by the adsorbed oxygen-containing intermediates. Notably, the positive demetallation energy differences between the Fe-N/W-NO and Fe-N sites in all reaction stages demonstrate the strong binding energy of Fe-N bonds in Fe-N/W-NO site, which well explains the enhanced catalytic stability of the Fe,W-N-C catalyst.