## Introduction

Owing to the high energy storage density and zero carbon emission, hydrogen (H2) fuel from water electrolysis has been regarded as the most promising alternative to fossil fuels1,2. Strikingly, the hydrogen evolution reaction (HER) plays an essential role in electrochemical water splitting for energy conversion. Various water electrolyzers demand different pH values of the electrolyte, such as proton exchange membrane electrolysis in strong acid, seawater electrolysis in neutral medium, and commercial water electrolysis in strong base3. To meet the above requirements, pH-universal HER catalysts with superior performance in both acidic and alkali media are highly regarded; however, they are barely accessible4. Platinum (Pt) and Pt-based catalysts are still the best-known pH-universal HER electrocatalysts, but their limited availability and high cost hinder their large-scale applications5. Therefore, exploring non-precious metal-based electrocatalysts with Pt-like pH-universal HER activity is highly desired, yet challenging.

To date, numerous earth-abundant HER electrocatalysts including oxides, hydroxides, alloys, phosphides, nitrides, sulfides, and their hybrids, have been identified as promising HER catalysts3,6,7,8,9,10,11. However, satisfactory Pt-like activity has been seldomly achieved and only a few of them can be simultaneously active in both acidic and alkaline media3. Recently, single-atom catalysts (SACs) with nearly 100% atom economy and unique electronic properties compared to their regular nanoparticle (NPs) counterparts have attracted immense scientific attention in the field of photo/electro/thermo-catalysis12,13,14. Most SACs contain isolated single metal sites coordinated with the neighboring nitrogen atoms in carbon matrix (M-NC), which are only capable of catalyzing simple elementary reactions15. Due to the simplicity of the single-atom center, the possibilities for further modification of the active site in SACs are extremely limited, hindering their wide range of applications16. In response to this, recent exploration suggests that tuning the coordination site to sulfur/phosphorus or by introducing secondary metal atom to construct metal–metal dual atom sites (single-atom dimer: SAD) can further modulate the electronic structure of SACs and boost their intrinsic activity, attributed to the unique atomic interface and synergistic effect of dual-metal site17,18,19. Recently, Fe-Co, Zn-Co, and Ni-Fe dual-metal sites have been demonstrated as efficient bifunctional oxygen electrocatalysts (oxygen reduction reaction (ORR)/oxygen evolution reaction (OER) and for CO2 reduction reaction (CO2RR)20,21,22. Zhang et al.23 synthesized noble metal-based Pt-Ru dimer using the advanced atomic layer deposition technique and showed comparable HER performance to commercial Pt in acidic media. However, the evidence for the formation of a single metal–metal bond from X-ray absorption spectroscopy (XAS) was unclear due to the existence of additional atomic clusters in the sample23. Although SADs are explored towards ORR/OER/CO2RR, a generalized cost-effective and versatile strategy to fabricate pH-universal low-cost HER catalyst with targeted dimeric sites at atomic precision along with appropriate identification of the dimeric structure and deeper understanding of the dual-metal atom synergism has never been achieved and remain elusive.

Herein, we report a transition metal-based SAD (TM-SAD) atomic interface, which can efficiently catalyze complex HER in a wide pH range (0–14). At first, systematic density functional theory (DFT) screening reveals that among various TM-SADs, the synergistic interaction between the Ni-Co at the atomic level in the SAD configuration can significantly upshift the d-band center, thereby accelerating water dissociation and boosting pH-universal HER activity. Motivated by DFT prediction, we develop a facile methodology to synthesize NiCo-SAD on N-doped carbon (NiCo-SAD-NC) via in situ trapping of targeted metal ions in the polydopamine sphere followed by annealing with precisely controlling the N-moieties. State-of-the-art techniques including X-ray absorption near edge structure (XANES), extended X-ray absorption fine spectra (EXAFS), aberration-corrected scanning transmission electron microscopy (AC-STEM), and X-ray photoelectron spectroscopy (XPS) along with theoretical calculation are employed to analyze the detailed structure of the NiCo-SAD-NC, which reveal the emergence of Ni-Co bond with strong electronic coupling at the atomic level. The as-prepared NiCo-SAD-NC exhibits an exceptional pH-universal HER activity, which requires only 54.7 and 61 mV overpotential at −10 mA cm−2 in acidic and alkaline media, respectively, outperforms the NiCo-NP and monoatomic Ni/Co-SACs. The activity of NiCo-SAD-NC is comparable/superior to commercial Pt-C/Pt-SAC, as well as superior to most of the recently reported TM-based single-atom electrocatalysts.

## Results

### Synthesis and structural characterization

Transmission electron microscopy (TEM) image of the NiCo-NP-NC revealed that the NiCo-alloy NPs (diameter: 15–20 nm) were uniformly encapsulated into the carbon matrix (Supplementary Fig. 8a). The high-resolution TEM image along with the corresponding selected area electron diffraction pattern confirmed the lattice spacing of 0.21 nm correspond to the (111) plane of NiCo-alloy (Supplementary Fig. 8b). The STEM high-angle annular dark-field (HAADF) image with EDS elemental map of the NiCo-NP-NC showed uniform distribution of the Ni, Co, and N (Supplementary Fig. 8c). Contrarily, no obvious NiCo-alloy NPs were spotted after the introduction of a sufficient amount of N, suggesting both Ni and Co species were atomically dispersed in the NiCo-SAD-NC (Supplementary Fig. 8d–f). In addition, the Raman spectra of both NC and NiCo-SAD-NC showed the characteristics D and G band, consistent with their corresponding XRD results (Supplementary Fig. 9)21. Aberration-corrected HAADF-STEM image in Fig. 3b clearly demonstrated the existence of isolated Ni-Co dimer sites (marked by the yellow square) with coordination between Ni and Co at atomic level along with some isolated Ni or Co atoms (marked by the orange circle). The homogeneously distributed bright dual dots marked by the yellow squares confirmed the existence of Ni-Co dual sites, verified using the intensity profile and corresponding electron energy loss (EEL) spectra (Fig. 3c, d). The bright Ni-Co dual dots were clearly identified in the intensity profiles and corresponding EEL spectrum, suggesting the possible formation of metal–metal bonds with an average dimer distance of 0.241 ± 0.024 nm, obtained from the statistical analysis over multiple dimer sites (Fig. 3e and Supplementary Fig. 10). The ratio of dimer structure was around 78%, indicating a significant amount of this type of structure in the prepared NiCo-SAD-NC material (Supplementary Fig. 11a). Meanwhile, HAADF-STEM and EDS elemental mapping revealed that N, Ni, and Co atoms were homogeneously dispersed in the NiCo-SAD-NC, rather than any possible aggregations in the form of NPs (Fig. 3f and Supplementary Fig. 11b).

### Spectroscopic characterizations

We further employed XPS, XANES, and EXAFS measurements to investigate the electronic state and local coordination chemistry of Ni/Co atoms in the catalysts. The C 1s high-resolution XPS spectra of NiCo-SAD-NC and NiCo-NP-NC were similar to that of NC, suggesting the absence of chemical bonding between metal atoms with C (Supplementary Fig. 12a–c). Contrarily, compared to NC and NiCo-NP-NC, the N 1s XPS spectra of NiCo-SAD-NC were dominated by the pyridinic-N along with porphyrin-like moieties at 399.2 eV, corresponding to the metal-nitrogen (Ni/Co-N) coordination (Supplementary Figs. 12d and 13a)22. Both Ni and Co 2p XPS spectrum of NiCo-SAD-NC, NiCo-NP-NC, Ni-SA-NC, and Co-SA-NC exhibited the characteristics 2p3/2 and 2p1/2 peaks (Supplementary Fig. 13b, c). Compared to NiCo-NP-NC, the binding energies for Ni-SA-NC and Co-SA-NC were positively shifted after introducing N to trap the single-atom sites, revealing Ni-N and Co-N bond formation28. However, after forming the NiCo dimer sites, the Ni 2p3/2 XPS spectra of NiCo-SAD-NC showed a positive shift with Ni oxidation state of +1.73 eV compared to that of Ni in Ni-SA-NC (+1.57), whereas the Co 2p3/2 XPS spectra of NiCo-SAD-NC exhibited a negative shift with Co oxidation state of +1.39 compared to Co in Co-SA-NC (+1.67), suggesting that the electron transfer occurred from Ni to Co site at the atomic interface of NiCo-SAD-NC, probably due to single Ni-Co bond formation at the atomic level (Supplementary Fig. 14).

## Discussion

In summary, we developed a facile strategy to obtain earth-abundant SAD sites via in situ trapping of the targeted metal ions (Ni, Co) followed by pyrolysis with precisely controlling the N-moieties for pH-universal HER. The detailed structural analysis of the obtained NiCo-SAD sites was carried out by XAS, AC-STEM, and XPS, which revealed that the NiCo-SAD-NC contains Ni-Co bond at atomic level stabilized by the N coordination. More notably, the synergistic interaction at the Ni-Co atomic interface in the SAD structure can significantly upshift the d-band center closer to the Fermi level and accelerated water dissociation, boosting pH-universal HER, as predicted by DFT calculations. Consistently, the obtained NiCo-SAD-NC delivered exceptional pH-universal catalytic kinetics towards HER, outperformed the NPs counterpart and comparable/superior to commercial Pt-C/Pt-SA, additionally surpassed most of the recently reported TM-based single-atom electrocatalysts. Our findings provide a rational design strategy for fabricating an earth-abundant metal-based SAD catalyst with atomic precision for both fundamental and practical research as well as for the deeper understanding of the bimetal synergistic effect for future energy-related applications.

## Methods

### Chemicals

Tris-buffered Saline (Sigma-Aldrich), Dopamine hydrochloride ((HO)2C6H3CH2CH2NH2·HCl; Sigma-Aldrich), Nickel(II) nitrate hexahydrate (Ni(NO3)2·6H2O; Sigma-Aldrich, ≥99%), Cobalt(II) nitrate hexahydrate (Co(NO3)2·6H2O; Sigma-Aldrich, ≥99%), Iron(III) nitrate nonahydrate (Fe(NO3)3.9H2O; ≥99%), Manganese(II) nitrate tetrahydrate (Mn(NO3)2·4H2O; Sigma-Aldrich, ≥99%), Dicyandiamide (NH2C(=NH)NHCN; Sigma-Aldrich, ≥99%), Chloroplatinic acid hexahydrate (H2PtCl6·6H2O; Sigma-Aldrich, ≥99%), potassium hydroxide pellets (KOH; Sigma-Aldrich, ≥85%), Sulfuric acid (H2SO4; Sigma-Aldrich, ≥99.99%), ethanol (C2H5OH; Sigma-Aldrich, ≥99.9%), Toray CFP/Ni foam (Alfa Aesar), and the nafion perfluorinated resin solution (5 wt.%, Sigma-Aldrich) were used without further purification.

### Synthesis of Ni2+-Co2+@Polydopamine (precursor)

In a typical procedure, Tris-buffer (1.21 g) was dissolved in 135 mL of distilled water (DI water) followed by drop-wise addition of aqueous solution (5 mL) containing metal salts (2 mg/mL, Ni(NO3)2·6H2O: Co(NO3)2·6H2O = 1 : 1). Then dopamine hydrochloride (70 mg) was quickly added in the above suspension and the polymerization was kept under magnetic stirring for 24 h. The resultant precipitate was collected via filtration and washed two times with DI water and ethanol, respectively, and dried at 60 °C overnight. For control samples, Ni2+-Co2+@Polydopamine precursor with different ratio of Ni(NO3)2·6H2O: Co(NO3)2·6H2O (1 : 2 and 2 : 1) as well as only Ni2+@Polydopamine (2 mg/mL, Ni(NO3)2·6H2O solution), Co2+@Polydopamine (2 mg/mL, Co(NO3)2·6H2O solution), Pt4+@Polydopamine (2 mg/mL, H2PtCl6·6H2O solution), Co2+-Fe3+@Polydopamine (2 mg/mL, Co(NO3)2·6H2O: Fe(NO3)3·9H2O = 1 : 1), Co2+-Mn2+@Polydopamine (2 mg/mL, Co(NO3)2·6H2O: Mn(NO3)2·4H2O = 1 : 1), and polydopamine were also synthesized.

### Synthesis of NiCo-NP-NC

For the synthesis of NiCo-NP-NC, a certain amount of Ni2+-Co2+@Polydopamine powder was placed in a vacuum furnace and heated at 800 °C for 2 h with a heating rate of 5 °C min−1.

In a typical procedure, a certain amount of Ni2+-Co2+@Polydopamine (precursor) was mixed with dicyandiamide (organic molecule: OM) in a ratio of 1 : 7 by grinding in a mortar. The mixture was annealed in a vacuum furnace at 800 °C for 2 h with a heating rate of 5 °C min−1 to yield NiCo-SAD-NC. Other control samples were also synthesized via the same procedure by varying the ratio of Ni2+-Co2+@Polydopamine (precursor) to OM and denoted as NiCo-NC (1 : 1), NiCo-NC (1 : 3), and NiCo-NC (1 : 5). For comparison, NiCo-SAD-NC (1 : 2 and 2 : 1), Ni-SA-NC, Co-SA-NC, CoFe-SAD-NC, and CoMn-SAD-NC were also synthesized following the similar procedure by only changing the starting precursor (Ni2+-Co2+@Polydopamine with different ratio of Ni(NO3)2·6H2O: Co(NO3)2·6H2O (1 : 2 and 2 : 1), Ni2+@Polydopamine, Co2+@Polydopamine, Co2+-Fe3+@Polydopamine, and Co2+-Mn2+@Polydopamine).

### Synthesis of NC

For the synthesis of NC, a certain amount of polydopamine was mixed with dicyandiamide in a ratio of 1 : 7 by grinding in a mortar, followed by annealing in a vacuum furnace at 800 °C for 2 h with a heating rate of 5 °C min−1.

### Synthesis of Pt-SA

For the synthesis of Pt-SA, a certain amount of Pt4+@Polydopamine precursor was mixed with dicyandiamide in a ratio of 1 : 20 by grinding in a mortar, followed by annealing in a vacuum furnace at 800 °C for 2 h with a heating rate of 5 °C min−1.

### Material characterization

The XRD measurements were carried out using a Rigaku Ultima IV powder X-ray diffractometer with Cu Kα radiation at λ = 0.15405 nm. FESEM images and EDS spectra were obtained using a JEOL 7500F FESEM. The Raman spectra were obtained using a Renishaw RM 1000-Invia micro-Raman system with excitation energy of 2.41 eV (514 nm). The XPS measurements were carried out on a Thermo VG Microtech ESCA 2000, with a monochromatic Al-Ka X-ray source at 100 W. The binding energy scale was calibrated by referencing C 1s to 284.5 eV. The XPS data were background corrected by the Shirley method and the peaks were fitted using Fityk software, with Voigt peaks containing 80% Gaussian and 20% Lorentzian components to get the valence states. TEM images were recorded using a JEOL JEM-2100F with an accelerating voltage of 200 kV. The aberration-corrected HAADF-STEM was performed using a Thermo Fisher Themis Z TEM equipped with a double Cs corrector, an electron-beam monochromator, and Gatan Image Filter (GIF, model Quantum 965) at Seoul National University. The acceleration voltage was set to 200 kV. EEL spectra were all acquired with a 5 mm EELS aperture corresponding to a collection angle of 45 mrad, a probe with a convergence angle of 49 mrad, and a beam current of ~75 pA. The EELS spectrometer was set to 0.25 eV per channel dispersion. The ICP-AES measurements were done using OPTIMA 4300 DV. XANES and EXAFS data were collected on BL10C beamline at the Pohang light source (PLS-II) with top-up mode operation under a ring current of 250 mA at 3.0 GeV. The monochromatic X-ray beam could be obtained using liquid-nitrogen cooled Si (111) double crystal monochromator (Bruker ASC) using intense X-ray photons of multipole wiggler source. The X-ray absorption spectroscopic data were recorded for the uniformly dispersed powder samples with a proper thickness on the polyimide film, in fluorescence mode with N2 gas-filled ionization chamber (IC-SPEC, FMB Oxford) for incident X-ray and passivated implanted planar silicon detector (PIPS, Canberra, Co.). Higher-order harmonic contaminations were eliminated by detuning to reduce the incident X-ray intensity by ~30%. Energy calibration has been simultaneously carried out for each measurement with reference metal foils placed in front of the third ion chamber. The data reductions of the experimental spectra to normalized XANES and FT radial distribution function were performed through the standard XAFS procedure using IFEFFIT package. Also, Morlet wavelet-transformed EXAFS spectra have been obtained with proper values of η and σ in the equation as follows;

$$\varphi \left(t\right)=\frac{1}{\sqrt{2\pi \sigma }}({e}^{i\eta t}-{e}^{-{\eta }^{2}{\sigma }^{2}/2}){e}^{-{t}^{2}/2{\sigma }^{2}}$$
(1)

in which the η is the frequency of the oscillation functions and the σ is the half-width.

### Electrochemical measurements

Electrochemical measurements were conducted using a VMP3 electrochemical workstation (Bio-logic Science Instruments, France) in a typical three-electrode configuration in 1 M KOH and 0.5 M H2SO4 as the electrolyte. Ag/AgCl (3 M KCl) and graphite rod were used as the reference and counter electrode, respectively. The catalyst ink-coated Ni foam or CFP was used as the working electrode. The reference electrode was calibrated in H2-saturated 1 M KOH and all the potentials are converted to a RHE using the Nernst equation.

$$E({{{{\mathrm{RHE}}}}})=E({{{{{\mathrm{Ag}}}}}}/{{{{{\mathrm{AgCl}}}}}})+{E}^{0}({{{{{\mathrm{Ag}}}}}}/{{{{{\mathrm{AgCl}}}}}})+0.059\,\times \,{{{{{\mathrm{pH}}}}}}$$
(2)

Then, 5 mg of catalyst powder was dispersed in 500 µL of ethanol containing 20 µL 5% Nafion and sonicated for 60 min to get a homogeneous ink. Afterward, a certain quantity of the ink was drop-cast onto Ni foam/CFP (loading: 0.8 mg cm2) and left to dry under ambient atmosphere. Before measurements, the catalysts were saturated via cyclic voltammetry (CV) scans at a scan rate of 100 mV s−1. LSV was taken at a slow scan rate of 2 mV s−1 to minimize the capacitive contribution7. Nyquist plot was obtained using electrochemical impedance spectroscopy measurements in the faradaic region to estimate the charge transfer resistance (RCT). Cdl was obtained by collecting CVs at various scan rates of 10, 20, 30, 40, and 50 mV s−1 in the non-faradaic. ECSA was obtained from the Cdl value using a specific capacitance of 0.04 mF/cm2. The durability test was performed using chronopotentiometry. Faradaic efficiency was measured by using the eudiometric method in an air-tight vessel. All the potentials were 85% iR-corrected with respect to the ohmic resistance of the solution unless specified and calibrated to the RHE using the following equation29.

$${E}_{({{{{{\rm{RHE}}}}}})}={E}_{({{{{{\rm{Ag}}}}}}/{{{{{\rm{AgCl}}}}}})}+{E}_{({{{{{\rm{Ag}}}}}}/{{{{{\rm{AgCl}}}}}})}^{0}+0.059\times {{{{{\mathrm{pH}}}}}}\,-85 \% \,{{{{{\mathrm{iRs}}}}}}$$
(3)

### Computational details

All the DFT calculations were carried out in the VASP computational package34. The plane-wave was constructed with the projected augmented wave pseudopotentials35 and the Perdew-Burke-Ernzerhof generalized gradient exchange approximation correlational functional are used for the treatment of the core electrons36. All geometric structures were fully optimized until forces and total energy are converged to 10−5 eV/cell and −0.01 eV/Å, respectively. The vacuum space in the z-direction was set as 15 Ǻ to eliminate interaction between two periodic images, and the cut-off energy was chosen at 450 eV. The Grimme-D3 level was used to describe the long-range van der Waals interactions37,38. The Brillouin zone of k-points is sampled by a 3 × 3 × 1 Monkhorst–Pack grid. A 4 × 4 × 1 supercell model of primitive graphene containing the TM-SAD-N6C was adopted for the surface calculations. The minimum energy path of water dissociation on TM-SAD-N6C surfaces was obtained by the nudged elastic band method with 5 intermediate images used to search for the transition states39,40. Vibrational free energy was calculated by zero-point energy and entropy contribution at room temperature (298 K).