Synthesis and structural features of hetero-coordinated FeN4-C-NiP4. The synthetic route of FeN4-C-NiP4 is schematic illustrated in Fig. 1a (SEM images and color change of intermediates shown in Supplementary Fig. 1 and Fig. 2), which involves four steps. Step I: forming SiO2−@Fe3+ spheres. SiO2 nanospheres (150 nm) were pre-synthesized and underwent surface modification to be negatively charged. Then Fe3+ were readily adsorbed on SiO2− to form the SiO2−@Fe3+ spheres through electrostatic attractions, accompanied by color change from white to dark orange. Step II: adding excess adenosine monophosphate (AMP+) to wrap on the outer surface of SiO2−@Fe3+. AMP+ molecule was consisted of phosphate functional group, pentose nucleo-saccharide and nitrogenous base adenine, as structure shown in Supplementary Fig. 3). Selecting cheap and sustainable biomaterial of AMP+ as N, P, and C source was due to following three reasons: i) -NH2 from APM+ could provide lone pairs of electrons to strongly bind with Fe3+ via coordination interactions, as confirmed by color change and UV-vis in Supplementary Fig. 4. The resultant Fe3+-AMP+ complex tended to form atomic Fe sites after pyrolysis through spatial confinement effect; ii) AMP+ with large molecular weight could intertwine on the surface of SiO2−@Fe3+, forming a thick AMP+ layer to avoid the contact of inner Fe3+ with other metal precursors in subsequent steps. As confirmed by TEM and SEM images in Supplementary Fig. 5, it indicates the completely coverage of AMP+ layer with thickness of 25.0 nm on the outer surface of SiO2−@Fe3+; iii) the AMP+ was positively charged under isoelectric point, beneficial for further electrostatic adsorption. Step III: forming SiO2−@Fe3+@AMP+@Ni(CN)42− spheres. The electronegative [Ni(CN)4]2− could further adsorb on the outer surface of SiO2−@Fe3+@AMP+. Specifically, the cyano groups in Ni(CN)42− could disperse the central Ni2+ to avoid the aggregation after pyrolysis19. The precise element distribution of SiO2−@Fe3+@AMP+@Ni(CN)42− spheres was examined by EDX mapping (Supplementary Fig. 6), which verifies Janus-like structure consisted of inner Fe3+ and outer [Ni(CN)4]2−. Step IIII: pyrolysis and leaching to acquire FeN4-C-NiP4. The SiO2−@Fe3+@AMP+@Ni(CN)42− powder was carbothermally reduced to form SiO2@FeN4-C-NiP4 due to anisotropic coordination capacity of Fe3+ and Ni2+. After pyrolysis, AMP+ layer was transformed into N, P-doped carbon spheres, meanwhile the inner Fe3+-AMP+ and outer AMP+-Ni(CN)42− were transformed into atomic Fe-Nx sites and Ni-Px sites, respectively. Particularly, the post transition metal Ni2+, which contains much d-orbital electrons and exerts lower charge, preferred to form the Ni-P bonds because the unoccupied 3d orbitals of P could share the electrons pair of Ni2+ to form extra back donating bonds20. The increase of bonding number leads to preferential formation of Ni-P than Ni-N. Finally, after leaching to remove the SiO2, Janus-distributed and hetero-coordinated FeN4-C-NiP4 could be yielding.
The FeN4-C-NiP4 dual-atom sites were first investigated by X-ray diffraction (XRD) and Raman spectra. The XRD pattern exhibits two broad peaks at 25.7° and 44°, assigned to the (002) and (100) planes of graphitic carbon (Fig. 1b). No diffraction peak ascribed to Fe or Ni could be detected, demonstrating the absence of crystalline Fe and Ni-based species. The Raman spectra shows defects-related D band at 1325 cm− 1 and sp2-hybridized carbon-related G band at 1590 cm− 1 (Supplementary Fig. 7). The relative high ID/IG value (1.11) for FeN4-C-NiP4 reveals the defective structure of carbon after pyrolysis. The morphology of FeN4-C-NiP4 was investigated by large-scale HRTEM images (Supplementary Fig. 8) and high angle annular dark field scanning TEM (HAADF-STEM, Fig. 1c). Large number of interconnected and hollow nanospheres with diameter of 150 nm and wall thickness of 3.8 nm could be observed, manifesting the structural robustness. Meanwhile, no visible nanoparticles or clusters could be observed, implying the atomic dispersion of both Fe and Ni. The Brunauer-Emmett-Teller surface area of FeN4-C-NiP4 was calculated to be 290.71 m2 g− 1, with a hysteresis loop related to type-IV isotherm (Supplementary Fig. 9), indicative a high specific surface area with mesoporous characteristics. The mono-dispersion of both Fe and Ni atoms was directly monitored by HAADF-STEM (Fig. 1d). A number of bright dots with size of 2.0 Å are randomly dispersed, corroborating that both Fe and Ni atoms are isolated without the formation of Fe-Ni dual-atom pairs. The energy-dispersive X-ray (EDX) mapping reveals the co-existence of Fe, Ni, C, N, and P elements over the entire architecture (Fig. 1e), confirming the successful doping of P and well reservation of Fe and Ni sites. Inductively coupled plasma atomic emission spectroscopy (ICP-AES) indicates that the weight fraction of Fe and Ni in FeN4-C-NiP4 is 1.10 at% and 1.05 at%, respectively.
The valence states and chemical environment of FeN4-C-NiP4 were elucidated by XPS (Supplementary Fig. 10). The wide-scan spectrum displays obvious signals of Fe, Ni, C, N, and P (content: 4.64 at%) in FeN4-C-NiP4. The high-resolution Fe 2p and Ni 2p spectra show main peaks for Fe2+ (709.8 and 723.1 eV), Fe3+ (715.2 and 728.9 eV) and Ni2+ (854.9 and 872.6 eV), respectively, manifesting the high-valence state of Fe and Ni in FeN4-C-NiP4. Theoretically, when electronic effect is existed between Fe and Ni, the Ni atoms could donate partial electrons to Fe atoms, leading to the positive shift of Ni region and correspondingly negative shift of Fe21. Importantly, the FeN4-C-NiP4 demonstrate standard location of Fe2+/Fe3+ and negative shift of Ni2+, suggesting a non-existent interaction between Fe and Ni. This is possibly due to a relative far distance between Fe and Ni sites, providing evidence for the spatially separated location of Fe and Ni sites in FeN4-C-NiP4. The N 1s spectrum reveals the coexistence of pyridinic N (398.1 eV, 36.8%), pyrrolic N (399.4 eV, 32.4%), and graphitic N (400.6 eV, 30.8%). The P 2p spectra could be fitted into three peaks at 131.9 eV (P 2p3/2 for M-P), 132.9 eV (P-C), and 133.8 eV (P-O), respectively. The obvious M-P signal implies the direct bonding between metal and P atoms.
In order to further confirm the Ni-P bond in FeN4-C-NiP4, the Ni2P-doped C nanosheets and mono-component NiN4 single-atoms were fabricated. Importantly, through directly calcining the mixture of K2[Ni(CN)4] and AMP powders, keeping the same ratio used for FeN4-C-NiP4, Ni2P-doped C nanosheets with high purity could be obtained, indicating the strong binding ability between Ni and P (structural features confirmed by HRTEM, XRD, and XPS in Supplementary Fig. 11). Particularly, both the XPS Ni 2p and P 2p spectra for Ni2P-doped C nanosheets show similar location with those for FeN4-C-NiP4, providing evidence for the possible Ni-P bonding in FeN4-C-NiP4. Furthermore, mono-component NiN4 with pure N coordination were synthesized according to previous work22, using the histidine to eliminate the P source. The formation of NiN4 was confirmed by XRD, HRTEM, HAADF-STEM, EDX mappings, and XPS, as displayed in Supplementary Fig. 12. Supplementary Fig. 13 compares the XPS Ni 2p3/2 spectra for FeN4-C-NiP4, Ni2P-doped C nanosheets, and NiN4. Obviously, the Ni 2p3/2 peak of FeN4-C-NiP4 and Ni2P-doped C nanosheets show a negative shift of 0.8 eV than that of the NiN4, implying non-formation of Ni-N bond in FeN4-C-NiP4. Due to the weaker electronegativity of P (2.01) than N (3.04), the charges of Ni adjacent to P atom are less positive, resulting in more negative location of Ni-P than Ni-N.
Atomic structure analysis. XAFS analysis were conducted to decipher the coordination configuration of FeN4-C-NiP4 at atomic scale (Supplementary Fig. 14). The pre-edge peak for Fe K edge is situated near the Fe2O3 and shifts to higher energy region than Fe foil and Fe2P, indicating that the Fe valence in FeN4-C-NiP4 is near + 3 (Fig. 2a). The pre-edge peak for Ni K-edge is located between Ni2P and NiO, demonstrating the average oxidation state of Ni in FeN4-C-NiP4 is between + 1 and + 2 (Fig. 2b). It is worth noting that the wide-range line-type of Fe K-edge for FeN4-C-NiP4 is similar to that of the Fe2O3 with a salient peak at 7132 eV, while the line-type of Ni K-edge for FeN4-C-NiP4 is similar to that of the Ni2P with mild ridge type. This suggests that the Fe and Ni for FeN4-C-NiP4 might possess similar coordination with Fe2O3 and Ni2P, respectively. The Fourier transform (FT) EXAFS spectra of Fe K-edge and Ni K-edge for FeN4-C-NiP4 are illustrated in Fig. 2c (fitting details shown in Supplementary Figs. 15 and 16). For Fe moiety, the predominant peak located at 1.41 Å is assigned to the backscattering of Fe-N in the first shell, meanwhile no related peak corresponding to Fe-Fe, Fe-P, or Fe-Ni could be detected. For Ni moiety, the sole peak located at 1.81 Å is attributed to the scattering of Ni-P first-shell coordination, which is consisting with the location of Ni-P bonds for Ni2P. Meanwhile, no Ni-Ni, Ni-N, or Ni-Fe contribution could be observed, confirming the entirely Ni-P bonding in FeN4-C-NiP4. EXAFS wavelet transform (WT) analysis present single intensity maximum at 3.5 Å−1 for Fe K-edge and 5.4 Å−1 for Ni K-edge, respectively (Supplementary Fig. 17). These could be assigned to the Fe-N and Ni-P path, further verifying the isolated feature of FeN4 and NiP4 in FeN4-C-NiP4. Quantitatively, the coordination numbers of Fe-N and Ni-P were extracted by EXAFS fitting. Due to the relative large difference in bond length, the coordination parameters of M-N and M-P could be well distinguished. As listed in Supplementary Tables 1 and 2, the coordination numbers for Fe-N were calculated to be 4.3 with bond length of 2.02 ± 0.01 Å. The coordination numbers for Ni-P were calculated to be 3.8 with a much longer bond length of 2.25 ± 0.02 Å. Thus the hetero-coordinated Fe1-N4 and Ni1-P4 configuration are determined to be the dominating structural motifs.
In order to investigate the role of ligands on tuning the electronic structure, the charge and spin density distribution, and projected density of states (PDOS) of FeN4-C-NiP4 were analyzed by DFT calculations, in comparison with N-coordinated FeN4-C-NiN4. Since the XPS and EXAFS verify negligible interaction between FeN4 and NiP4 sites, a relatively far Fe-Ni distance was simulated to avoid the interference. As shown in Fig. 2d, the charge density distribution of FeN4 and NiN4 sites for FeN4-C-NiN4 show similar intensity and scope due to the same N coordination. While introducing four P ligands to Ni center, the charge distribution scope of NiP4 is largely pronounced with increased electron density around Ni center, resulting in electron-rich feature of NiP4 sites. Comparatively, the FeN4 sites is electron-deficient with depressed charge distribution. Thus the combination of FeN4 and NiP4 sites for FeN4-C-NiP4 shows an obviously enhanced surface charge polarization than N-coordinated FeN4-C-NiN4. Similarly, the spin gap in FeN4-C-NiP4 also becomes wider in comparison with FeN4-C-NiN4. As shown in Fig. 2e, the NiN4 indicates a higher spin state (0.12) compare to that of the FeN4 (-0.09). While introducing four P ligands into Ni center, the maximum spin density of NiP4 further uplifts to 0.19, due to the contribution of 3p lone-pair electrons from P. As a result, the FeN4-C-NiP4 possess higher spin NiP4 sites and relativity lower spin FeN4 sites, illustrating that the heterogeneous coordination could further enlarge the spin difference of dual-metal sites. Moreover, the d band center of FeN4 and NiP4 sites in FeN4-C-NiP4 exhibit a larger difference in comparison with those in FeN4-C-NiN4. As shown in Fig. 2f, the PDOS location follows the order of FeN4 (-0.208 eV) > NiN4 (-1.956 eV) > NiP4 (-2.951 eV). The FeN4 with electron-deficient feature is endowed with an up-shifted d band center that closer to the Fermi level, while the d band center of NiP4 is much lower than NiN4, which is far away from Fermi level. This indicates the P ligands could further downshift the d band center of Ni center, resulting in enlarged difference in d band center of FeN4 and NiP4 sites for FeN4-C-NiP4. Overall, above results demonstrate the asymmetrical charge, spin, and PDOS deployment in FeN4-C-NiP4, which could simultaneously provide two atomic sites with electron accumulation and deficiency, or two atomic sites with high and low spin, as well as two atomic sites with close and far d band center towards Fermi level. Since these electronic properties are closely related to the adsorption behavior of oxygen-contained intermediates and thus determine the intrinsic activity of metal center23, 24, the hetero-coordinated FeN4-C-NiP4 with stronger charge polarization, wider spin gap, and larger PDOS difference is promising for enlarging the difference in catalytic activity of atomic Fe and Ni sites.
Experimental and theoretical oxygen catalytic mechanism on of FeN 4 -C-NiP 4 . Since the MN4-C moiety has been affirmed to follow the activity order of Fe > Co > Ni for ORR, Ni > Co > Fe for OER, and Co > Ni > Fe for HER in alkaline medium25–28, the FeN4-C-NiP4 dual-atom sites were applied to bifunctional ORR and OER in 0.1 M KOH. To identify the role of separated FeN4 and NiP4 sites and shed lights on the hetero-coordination for enhancing electro-catalytic selectivity, mono-component FeN4, NiP4, and NiN4 were employed as references (structural details shown in Supplementary Figs. 18, 19, 12, respectively). The FeNi/Ni2P nanoparticles synthesized without using SiO2 template (Supplementary Fig. 20) and hollow C spheres without any metal doping were also used for comparison. As the linear sweep voltammetry (LSV) curves in Fig. 3a, the FeN4-C-NiP4 display similar onset potential (Eonset=1.02 V) and half-wave potential (E1/2=0.89 V) compare to those of the FeN4 (1.01, 0.89 V), which are much better than those of Pt/C (0.95, 0.83 V) and NiP4 (0.85, 0.72 V). At 0.9 V, the FeN4-C-NiP4 exhibit the jk of 1.44 mA cm− 2, which is similar to that of the FeN4 (1.53 mA cm− 2) and much better than Pt/C and NiP4 (Fig. 3b). The trend of Tafel slope is consistent with activity, that the FeN4-C-NiP4 and FeN4 show faster ORR kinetics than Pt/C and NiP4 (Supplementary Fig. 21a). Besides ORR, the FeN4-C-NiP4 also present outstanding OER activity, while the activity trend of reference samples are reversed. As shown in Fig. 3c, to reach a current density of 10 mA cm− 2, FeN4-C-NiP4 require an overpotential (η) of 247 mV, which is slightly higher than NiP4 (218 mV) and much lower than NiN4 (313 mV), RuO2 (350 mV), and FeN4 (406 mV). At η = 300 mV, the FeN4-C-NiP4 reach a current density of 53.29 mA cm− 2, which is similar to that of the NiP4 (49.24 mA cm− 2) and much higher than the other samples (Fig. 3d). The trend of Tafel slope is consistent with activity, that the FeN4-C-NiP4 and NiP4 show much lower Tafel slope than NiN4, RuO2, and FeN4, revealing the fast OER kinetics over FeN4-C-NiP4 and NiP4 (Supplementary Fig. 21b). In addition, the hollow C spheres without any metal doping exhibit feeble activity for both ORR and OER, as shown in Supplementary Fig. 22. This illustrates that the hollow carbon spheres in FeN4-C-NiP4 are inactive for ORR and OER, verifying the pivotal role of Fe-N4 and Ni-P4 as active sites. The FeNi/Ni2P nanoparticles synthesized without using SiO2 template exhibit obviously degenerative activity for ORR and OER, in comparison with FeN4-C-NiP4 (Supplementary Fig. 23). This verifies the importance of atomic metal sites for enhancing the ORR and OER activity.
Overall, the ORR activity follows the order of FeN4-C-NiP4 ≈ FeN4 > Pt/C > FeNi/Ni2P > NiP4 > hollow C spheres, while the OER activity follows the order of NiP4 ≈ FeN4-C-NiP4 > FeNi/Ni2P > NiN4 > RuO2 > FeN4 > hollow C spheres. Accordingly, following conclusions could be deduced: i) the mono-component FeN4 sites are highly active for ORR but poor for OER, meanwhile the mono-component NiP4 sites are highly active for OER but poor for ORR. This illustrates that the FeN4-C-NiP4 could provide highly active FeN4 and NiP4 sites for ORR and OER, respectively; ii) the ORR activity of FeN4-C-NiP4 and FeN4 is similar, so as the OER activity of FeN4-C-NiP4 and NiP4. This illustrates no obviously synergistic effect is existed between FeN4 and NiP4 sites, which is possibly due to a relative far metal-metal distance; iii) the OER activity of NiP4 atomic sites are much better than NiN4 atomic sites, testifying the prominent role of four P ligands for promoting OER than conventional N coordination.
Inspired by the high ORR and OER activity, the FeN4-C-NiP4 exhibit a small potential gap of 0.59 V, surpassing most of the state-of-art bifunctional catalysts (Supplementary Fig. 24 and Fig. 3e). The ORR and OER stability of FeN4-C-NiP4 were evaluated by accelerated durability tests (ADTs) and chronoamperometry (Fig. 3f). The FeN4-C-NiP4 dual-atom sites manifest a remarkable long-term stability with negligible decay of E1/2 (5 mV) for ORR and η10 mA cm−2 (8 mV) for OER after 5,000 cycles. Chronoamperometry also confirms the durability of FeN4-C-NiP4, exhibiting a small degradation of 3.4% and 5.8% after 18,000 s for ORR and OER, respectively. The practicability of FeN4-C-NiP4 as bifunctional catalysts was examined by Zn-air battery (Fig. 3g and Supplementary Fig. 25). The FeN4-C-NiP4-assemblied Zn-air battery could drive the LED light array for more than 24 h and exhibit a small charge-discharge voltage gap (1.18 V@10 mA cm− 2). After 600 cycles (~ 200 h), it shows up an even better voltage gap to 1.02 V, outperforming the Pt/C + RuO2 air-cathode (Supplementary Fig. 26).
DFT calculations were carried out to explain the origin of high ORR and OER activity on FeN4-C-NiP4 and unravel the explicit role of hetero-coordination. Figure 4a shows the DFT-optimized adsorption configuration of reaction adsorbates on FeN4 and NiP4 sites. The adsorption of O2* is stronger at FeN4 sites with bonding length of 2.337 Å, which is much shorter than NiP4 sites (2.539 Å). On the contrary, the adsorption of OH* is stronger at NiP4 sites with bonding length of 1.901 Å, which is shorter than FeN4 sites (1.947 Å). These declare that the FeN4-C-NiP4 dual-atom sites possess two metal centers with contrasting adsorption behavior. Figure 4b displays the Gibbs free energy profiles on FeN4 and NiP4 sites, in an effort to identify the main active sites in FeN4-C-NiP4 that determine the ORR and OER, respectively. From left to right, the FeN4 sites have more significantly exothermic proton-electron transfer steps than NiP4 sites, indicating a favorable ORR pathway at FeN4 sites. Moreover, the FeN4 sites show a more favorable protonation of O2* (ΔGO2*→OOH*= -0.82 eV) than NiP4 (-0.59 eV), further authenticating a faster ORR kinetics on FeN4 sites. For OER, the free energy difference between GOOH* and GOH* (ΔGOOH*-OH*)is found to be a key reaction descriptor, with the ideal value of 2.46 eV29. The value of GOOH*-GOH* for NiP4 sites (2.75 eV) is closer to 2.46 eV than that of FeN4 sites (3.16 eV), indicating a favorable OER pathway at NiP4 sites. Overall, above results suggest the oxygen catalytic selectivity of separated FeN4 and NiP4 sites in FeN4-C-NiP4, which are highly active for ORR and OER, respectively, thus their combination could balance the competition between rate-limiting steps of oxygen reduction/evolution, responsible for both high ORR and OER activity.
To shed lights on the role of hetero-coordination for enhancing the catalytic selectivity of atomic Fe and Ni sites, the ORR pathway at FeN4 and FeP4 sites (Fig. 4c and Supplementary Fig. 27), and OER pathway at NiN4 and NiP4 sites (Fig. 4d and Supplementary Fig. 28) were compared. For ORR, the FeN4 and FeP4 sites show very similar proton-electron transfer pathways, suggesting that for Fe center, selecting four N or four P coordination cause unconspicuous effect for promoting ORR. On the contrary, for OER, the NiP4 and NiN4 sites show disparate proton-electron transfer steps, wherein the first three electron-transfer steps at NiN4 sites are too fast. As a result, the ΔGOOH*-OH* for NiN4 (3.13 eV) is much far away from 2.46 eV than NiP4 (2.75 eV). This verifies that for Ni center, four P coordination are more favorable for optimizing OER pathway than N coordination. Combining with the electronic properties illustrated in Fig. 2d-f, the correlation between ligands, electronic structure and oxygen reduction/evolution activity could be clarified. The PDOS is used to explain the improved ORR activity experimentally observed in FeN4, as shown in Fig. 4e. According to the d band center theory, a closer d-band center to the Fermi level represents a strong binding ability for surface adsorbed O2* species30, 31. Accordingly, the electron-deficient FeN4 sites with up-shifted d-band center are more favorable for adsorbing O2* as reactant and possess a faster O2* protonation pathway (as evidenced by the short O2* bonding length of 2.337 Å and ΔG O2*→OOH*= -0.82 eV) compare to both NiN4 and NiP4 sites, resulting in higher ORR activity. The spin density is served to explain the strongly enhanced OER activity experimentally observed in NiP4, as shown in Fig. 4f. For OER, the ΔGOOH*-OH* is largely restricted by the stability of O* adsorption, that either too strong or too weak O* adsorption leads to an increase in ΔGOOH*-OH*32. Significantly, the high spin NiP4 sites could stabilize the unpaired electron of O* through the exchange interactions, balancing the too fast and too slow transformation to OOH* at NiN4 and FeN4 sites, respectively, resulting in an ideal ΔGOOH*-OH* of 2.75 eV and thus responsible for the outstanding OER activity. Overall, above results indicate the role of heterogeneous N, P coordination for respectively customizing the electronic structure of atomic Fe and Ni sites and thus enhancing the selectivity for oxygen catalytic reaction than conventional N coordination. Specifically, the FeN4 sites with closer d-band center towards the Fermi level are responsible for ORR, while the NiP4 sites with higher spin are more favorable for OER than both FeN4 and NiN4 sites, as schematic shown in Fig. 4g.
Generality of the synthesis of hetero-coordinated dual-atom sites. Above concept of using heterogeneous ligands to severally customize the electronic structure of dual atom sites could be extended to perform Janus-distributed CoN4-C-NiP4. The structural features was confirmed by HADDF-STEM image (Fig. 5a-b), EDX-mapping (Fig. 5c), HRTEM, XRD, and XPS analysis (Supplementary Figs. 29, 30). The dominated active sites for HER and OER in CoN4-C-NiP4 were first analyzed by DFT calculations (Fig. 5d). The free energy of hydrogen adsorption at CoN4 is as low as 0.07 eV, which is much closer to 0 eV than that of the NiP4 (1.35 eV), indicating more favorable HER kinetics at CoN4 sites. The other side, the Gibbs free energy profiles for OER show a smaller ΔGOOH*-OH* value and a faster proton-electron transfer pathway at NiP4 sites, in comparison with CoN4 sites, suggesting a more favorable OER kinetics at NiP4 sites. Thus the combination of separated CoN4 and NiP4 sites in CoN4-C-NiP4 is theoretically promising for generating high HER and OER activity, simultaneously. The experimental HER and OER performance of CoN4-C-NiP4 were examined by polarization curves obtained in 0.1 M KOH (Fig. 5e-f), and the trends were summarized in Fig. 5g. The η10 mA cm−2 for HER follows the order of Pt/C (96 mV) > CoN4-C-NiP4 (117 mV) ≈ CoN4 (124 mV) > NiP4 (249 mV). Meanwhile the η10 mA cm−2 for OER follows the order of CoN4-C-NiP4 (206 mV) ≈ NiP4 (218 mV) > RuO2 (350 mV) > CoN4 (370 mV). These experimentally confirm that the separated CoN4 and NiP4 sites in CoN4-C-NiP4 possess different selectivity for HER and OER, well consistent with the DFT calculation. As a result, the CoN4-C-NiP4 dual-atom sites achieve a small voltage difference of 1.53 V to reach 10 mA cm− 2 (Supplementary Fig. 31), surpassing most of the state-of-art non-precious bifunctional catalysts (Supplementary Table 4). The practicability of CoN4-C-NiP4 was evaluated by applied to bifunctional electrodes for water splitting. As shown in Fig. 5h, the CoN4-C-NiP4 electrodes require a cell potential of 1.544 V to afford 10 mA cm− 2, which is even equal to that of the Pt/C//RuO2, as well as exhibit a long-range stability operated for 24 h (Supplementary Fig. 32).