Boosting proton reduction via isolated Cu atoms in Bi lattice for efficient electroreduction of CO2 to formate


 Current Bi-based catalysts suffer from low current density for electroreduction of CO2 to formate due to the high energy barrier of H+ reduction to *H on Bi sites. Here we report a unique BiCu single-atom alloy catalyst (SAAC) that can deliver a ultrahigh formate partial current density (jformate) of 434 mA cm–2, the highest among the reported Bi-based electrocatalysts to date, with a formate Faradaic efficiency (FEformate) of 96.5% at –0.55 V (vs. RHE) in a flow cell, while BiCu alloy catalyst containing Cu nanoclusters can only deliver a jformate of 48.5 mA cm–2 with a FEformate of 37.3% under an identical condition. Mechanism investigations reveal that the isolated single-atom Cu in BiCu SAAC can dramatically reduce the energy barrier of H+ reduction to *H on Cu site for boosting the reduction of CO2 to formate. Our work provides a new strategy for engineering unfavourable energy barrier of electrocatalysts to promote CO2 reduction.


Introduction
The excessive emission of carbon dioxide (CO 2 ) has caused severe climate problems. To alleviate the global warming and fossil fuel depletion, an appealing route is to convert CO 2 into fuels and industrial feedstocks with renewable electricity 1,2 . Among the various products in electrocatalytic CO 2 reduction reaction (CO 2 RR), formate/formic acid (HCOO -/HCOOH) has received growing concern due to its competitive economic value for its excellent hydrogen storage capacity in fuel cell applications and extensive demands in chemical and pharmaceutical industries 3,4 . To date, several metals, such as Pd, In, Sn, Pb and Bi, have been reported for e cient electroreduction of CO 2 into formate in aqueous solution [5][6][7][8] .
In particular, by virtue of the low cost, low toxicity and high Faradaic e ciencies for formate (FE formate ), Bi-based electrocatalysts have attracted increasing interest 9,10 . Bene ting from the semi-metal feature of Bi, the surface electron accessibility is signi cantly limited in aqueous solution, thus suppressing the competing hydrogen evolution reaction (HER) 11 . However, the semi-metal feature of Bi also leads to low current density (low production rate) for electroreduction of CO 2 to formate, which greatly hinders the commercial application of Bi-based electrocatalysts. To solve this problem, several strategies such as integrating catalysts with conductive substrates 10 , increasing the density of active sites by optimizing the morphology of catalyst or raising catalyst loading 12,13 , have been reported. Alloying strategy has also been proved to be an effective pathway to regulate the unsatisfactory intrinsic activity of Bi via introducing another component with complementary characters, thereby realizing better catalytic performance than that of the individual constituent 14 . For example, a bimetallic catalyst featuring BiSn alloy core with BiSnO x shell shows a formate partial current density (j formate ) as high as 74.6 mA cm -2 at -1.38 V versus reversible hydrogen electrode (vs. RHE), with a FE formate of 72.1% in a gas diffusion electrolyser (GDE), whereas monometallic Sn/SnO x core/shell structure can only achieve a j formate of 15 mA cm -2 under identical conditions 15 . Compared with pure Bi, the Bi-SnO x surface structure can effectively lower the energy barrier of *HCOO formation, whereas the energy barrier of H* formation on Bi-SnO x is even higher than that on SnO x , contributing to a more effective suppression of HER. However, the j formate signi cantly decreases to 20.9 mA cm -2 with the FE formate over 90%. Therefore, simultaneously achieving high FE formate and high j formate still remains a great challenge for Bi-based catalysts.
As an intersection of single-atom catalysts (SACs) and alloy catalysts, single-atom alloy catalysts (SAACs) have shown exceptional performances in various catalytic reactions, such as selective hydrogenation 16, 17 , O 2 activation 18 , selective oxidation 19 , CO oxidation 20 , C-H activation 21 , and electrocatalytic water oxidation 22 . The precise surface structures of SAACs provide excellent platforms for understanding the structure-performance relationship in heterogeneous catalytic process on atomic scale. Moreover, unlike the single metal atoms anchored on the metal oxides or carbon materials which usually serve as supports only (although the charges of the metal centres are signi cantly in uenced by the supports) 23 , the isolated metal in SAAC may catalyse reactions synergistically with the neighbouring metal atoms in the metal support 24 . This feature endows SAACs with unique behaviours in the adsorption and activation of reactants or key intermediates, thereby contributing to enhanced catalytic performance. Taking a deep insight into the process of CO 2 -to-HCOOconversion, the reduction of H + plays a signi cant role in the formation of the key intermediate *HCOO 25 , which is, however, often ignored in Bi-based catalysts. The H + reduction is very di cult on metallic Bi because of its low a nity to H + , and thus greatly reduces the j formate for Bi-based catalysts. According to the volcano plot of the exchange current density as a function of the Gibbs free energy of adsorbed atomic hydrogen for various metals 26 , the non-noble metal Cu has a suitable hydrogen adsorption energy that enables not only a favourable hydrogen adsorption compared with Bi, Ag and Au, but also an easier hydrogen desorption than that on Pt, Ir and other metals for the subsequent H transfer to *CO 2 . However, aggregation of Cu atoms would lead to undesired HER and lower the selectivity for the formation of formate. Therefore, constructing BiCu SAACs, in which the isolated Cu atoms are embedded in Bi lattice, would effectively promote the H + reduction and the formation of *HCOO, with an e cient suppression of HER, and eventually achieve both high FE formate and high j formate .
Herein, we report the design, synthesis and electrocatalytic performance for the reduction CO 2 to formate of a BiCu nanosheet SAAC, with the isolated Cu atoms embedded in Bi lattice. As expected, the optimized pathway for the reduction of CO 2 to formate enables BiCu nanosheet SAAC achieving both high j formate and high FE formate in CO 2 RR, with a j formate of 77 mA cm -2 and a FE formate of 91.6% in an H-type cell, as well as a j formate as high as 434 mA cm -2 and a FE formate of 96.5% in a GDE. Moreover, the FE formate can still remain over 80% even in 10% CO 2 , where the CO 2 concentration is close to that of the industrial CO 2 emissions without energy-consuming CO 2 capture process. Our results strongly demonstrate the great potential of the BiCu SAAC for CO 2 RR in commercial applications.

Results
Fabrication and characterization of BiCu SAAC. The BiCu-x alloy nanosheets were obtained by in situ electroreduction of BiOI-Cu-x nanosheets in Ar-saturated 0.5 M NaHCO 3 ( Supplementary Fig. 1). BiOI-Cu-x precursors with a sheet-like morphology were synthesised via a hydrothermal method . The molar ratios of Bi/Cu were carefully tuned by controlling the amount of CuCl 2 during the synthesis of BiOI-Cu-x nanosheets, and inductively coupled plasma-optical emission spectrometry (ICP-OES) results reveal that the molar contents of Cu in BiCu-x alloy catalysts are 0.2%, 0.5%, 1% and 3% (denoted as BiCu-0.2, BiCu-0.5, BiCu-1, and BiCu-3, respectively), close to those for the corresponding BiOI-Cu-x precursors (Supplementary Table 1). For comparison, metallic Bi nanosheets were also prepared by in situ electroreduction of BiOI nanosheets under the same conditions as those of BiCu-x nanosheets. Xray diffraction (XRD) patterns show only metallic Bi crystallographic phase in BiCu-x nanosheets ( Supplementary Fig. 5), suggesting the total conversion of BiOI-Cu-x precursors to metallic Bi and BiCu-x alloys, which is further veri ed by the X-ray photoelectron spectroscopy (XPS) results ( Supplementary Fig.   6).
Scanning electron microscopy (SEM) images show that the Bi and BiCu-x samples remain the twodimensional morphology with rough surfaces ( Fig. 1a and Supplementary Fig. 7). High-angle annular dark-eld scanning transmission electron microscopy (HAADF-STEM) image of BiCu-0.5 shows small nanoparticles with an average size of ~5 nm dispersed uniformly over the surface of the nanosheet, indicating abundant boundaries and interfaces in BiCu-0.5 (Fig. 1b). High-resolution transmission electron microscopy (HRTEM) image for BiCu-0.5 displays clear lattice fringes of rhombohedra Bi (012) facet, with an interplanar spacing of 0.33 nm for both nanoparticle and substrate (Fig. 1c). Fast Fourier transform (FFT) pattern further demonstrates the polycrystallinity of the BiCu-0.5 nanosheets ( Supplementary Fig. 8). High-resolution energy-dispersive X-ray spectroscopy (EDS) analysis for BiCu-0.5 shows that Bi and Cu elements are homogeneously distributed over the catalyst (Fig. 1d), suggesting the formation of BiCu alloy. Notably, no distinguishable aggregation of Cu signals was observed in the XRD pattern ( Supplementary Fig. 5), the HRTEM image ( Fig. 1c) or the EDS elemental mapping image (Fig.  1d), indicating the possible atomic dispersion of Cu atoms in BiCu-0.5 nanosheets.
Evaluation of CO 2 RR performances. CO 2 RR performances of the Bi and BiCu-x catalysts were evaluated in an H-type cell lled with 0.5 M NaHCO 3 aqueous solution, and the liquid product was analysed by 1 H-NMR and ion chromatograph ( Supplementary Fig. 9). As revealed by the linear sweep voltammetry curves ( Fig. 2a and Supplementary Fig. 10), at low potential (below -1.0 V), all the current densities in CO 2saturated solution are higher than those in Ar-saturated solution, indicating the occurrence of CO 2 RR on each catalyst. With the potential more negatively shifted, the current density of BiCu-0.5 in CO 2 -saturated solution keeps higher than that in Ar-saturated solution (Fig. 2a), whereas reverse results were observed for the other catalysts ( Supplementary Fig. 10). A total current density of 63.6 mA cm -2 was achieved by BiCu-0.5 at -1.58 V vs. RHE in CO 2 atmosphere, surpassing the current densities of all the other Bi and BiCu-x alloy catalysts. These results highlight the best electrocatalytic performance of BiCu-0.5 for CO 2 RR among the Bi and BiCu-x catalysts, especially under high current density. The FE formate of Bi and BiCu-x catalysts at different potentials are shown in Fig. 2b and Supplementary Fig. 11. For the Bi nanosheets, the FE formate reaches a maximum value of 84.7% at -1.18 V, and quickly decreases to less than 50% at the potentials negative than -1.48 V, where HER becomes dominant. On the contrary, the BiCu-0.5 catalyst keeps high selectivity for formate with FE formate reaching above 94.0% in a wide potential window (-0.88 V to -1.28 V), and the maximum FE formate reaches 95.7% at -0.98 V. The FE formate can maintain over 85.0% even at -1.48 V, indicative of the e cient suppression of HER on BiCu-0.5. Notably, the maximum FE formate of BiCu-3 decreases to 89.7% at -1.08 V and sharply drops to 39.9% at -1.28 V. A volcano-type relationship between FE formate and the Cu content was observed, and BiCu-0.5 displays the maximum FE formate . Fig. 2c presents the j formate for Bi and BiCu-0.5 at various potentials. Distinctly, as the overpotential increases, the disparity in j formate between Bi and BiCu-0.5 becomes increasingly signi cant, and a j formate of 45.1 mA cm -2 was achieved at -1.48 V for BiCu-0.5, which is 3.2 times higher than that for Bi. The Tafel slop of formate production for BiCu-0.5 is 107 mV dec -1 , much lower than that for Bi (133 mV dec -1 ), which con rms the favourable CO 2 RR kinetics for BiCu-0.5 ( Supplementary Fig. 12). To further explore the intrinsic activity of the BiCu-0.5, electrochemically active surface area (ECSA) was estimated ( Supplementary Fig. 13), and used to normalize the geometrical current density. The ECSAcorrected j formate of BiCu-0.5 still shows the highest value among the ve catalysts, 2.8 times higher than that of Bi at -1.48 V ( Supplementary Fig. 14). Thus, it can be concluded that incorporating Cu in Bi can actually enhance the catalytic current for CO 2 RR, and 0.5% is the optimal molar percentage of Cu.
To eliminate the overpotential caused by the resistance of the electrolyte and show the accurate electrocatalytic performance, ohmic resistance compensation was conducted. The maximum FE formate of 98.3% was achieved with a j formate of 27.0 mA cm -2 at an overpotential of only 680 mV, and j formate reaches a maximum value of 76.6 mA cm -2 with a FE formate of 91.6% at -1.13 V ( Supplementary Fig. 15), further con rming the outstanding CO 2 RR performance of BiCu-0.5. Long-term stability test was also conducted for BiCu-0.5. As shown in Fig. 2d, BiCu-0.5 delivers a steady current density of about 40 mA cm -2 with FE formate kept over 90% during the 40 h CO 2 RR measurement. Besides, neither phase transformation nor morphology change was observed for the BiCu-0.5 after 40 h electrolysis, as revealed in Supplementary Fig. 16.
Insight of the enhanced CO 2 RR performance. It is natural to ask what factors enable the outstanding CO 2 RR performance of BiCu-0.5. The result of Nyquist plots reveals that the incorporation of Cu in BiCu-0.5 contributes to an improved mass transportation over the catalyst/electrolyte interface during the CO 2 RR (Supplementary Fig. 17) 27 . Another key parameter that greatly in uences the CO 2 RR performance is the pH at the surface of catalyst, since the reduction of H + to *H plays an important role in both formation of formate and H 2 28 . During the electrolysis, the pH near the cathode would arise owing to the rapid consumption of proton. To monitor the pH change during electrolysis, we conducted in situ Raman spectroscopy measurements, where the Raman signals of HCO 3 -(the peak at 1015 cm -1 ) and CO 3 2-(the peak at 1068 cm -1 ) can act as indicators to re ect the change in pH 29,30 . The spectra were recorded after 600 s electrolysis at each potential in CO 2 -saturated 0.5 M NaHCO 3 solution (Fig. 3a,b). Evidently, the peak intensities of CO 3 2for both Bi and BiCu-0.5 gradually increase as the potential negatively shifts, indicative of the increase in pH value. Besides, a more pronounced enhancement in the peak intensities of CO 3 2was observed for BiCu-0.5, implying that the pH value is more elevated at the surface of BiCu-0.5 catalyst, which can be attributed to its higher current densities and more rapid consumption of the proton.
HER performance of Bi and BiCu-0.5 was also evaluated in Ar-saturated 0.5 M Na 2 SO 4 solution ( Supplementary Fig. 18). Distinctly, a decrease of 80 mV in overpotential was achieved by BiCu-0.5 at an ECSA-corrected current density of 0. For BiCu-x alloy catalysts, the arrangements of the surface atoms play a critical role in the catalytic performance. To further understand the outstanding CO 2 RR performance of BiCu-0.5, atomic resolution aberration-corrected HAADF-STEM was performed (Fig. 4a). The darker dots marked by red circles can be assigned to Cu atoms according to the disparity in Z-contrast between Cu and Bi atoms 32 . The line intensity pro le in Fig. 4b also validates the isolated Cu atom in Bi lattice. As a step further, synchrotron radiation-based X-ray absorption ne structure (XAFS) spectroscopy was conducted to resolve the local structure of Cu atoms. Fig. 4c shows the Cu K-edge X-ray absorption near-edge structure (XANES) spectra for BiCu-x nanosheets (x = 0.5, 1 and 3), as well as the spectra for standard Cu foil, Cu 2 O and CuO as references. The absorption edge positions of three BiCu-x catalysts situate between those of Cu and CuO, suggesting that the Cu atoms in BiCu-x are partially oxidized 14 . The Fourier-transformed extended XAFS (EXAFS) spectra are shown in Fig. 4d. The BiCu-0.5 displays a dominant peak of Cu-O at 1.94 Å, and a shoulder peak at 2.24 Å probably to Cu-Cl introduced by the CuCl 2 precursor 33 . For BiCu-3, however, the peak of Cu-Cu at 2.57 Å becomes dominant, and the Cu-O peak signi cantly declines. In BiCu-1, both peaks of Cu-O at 1.94 Å and Cu-Cu at 2.57 Å can be observed, and the peak of Cu-O is dominant (Fig.   4d). The wavelet transform (WT) plots for BiCu-0.5 and BiCu-1 both show the maximum value at 5 Å -1 , which corresponds to Cu-O bond, whereas the maximum of the WT plot for BiCu-3 is at 8 Å -1 , close to that for Cu foil, which corresponds to Cu-Cu bond (Fig. 4e). The absence of Cu-Cu bond in BiCu-0.5 demonstrates that the Cu atoms are single-atom isolated, whereas the Cu atoms in BiCu-1 are mainly single-atom isolated accompanying with small amount of Cu clusters, and the Cu atoms in BiCu-3 are mainly in aggregation state. The Cu aggregation would lead to the promoted H 2 evolution 34 , accounting for the poor formate selectivity of BiCu-3 ( Supplementary Fig. 11c). Besides, the XAFS data of the BiCu-0.5 after 40 h electrolysis suggests that the ne structure of BiCu-0.5 is well retained (Supplementary Fig.  21). Notably, the Bi L 3 -edge XANES spectra show that the Bi atoms in BiCu-x are also severely oxidized ( Supplementary Fig. 22), which is consistent with the results of XPS and ex situ Raman spectrum ( Supplementary Fig. 23). The existence of oxidized Cu and Bi in BiCu-0.5 can be attributed to the high susceptibility of Bi and isolated Cu to O 2 35 . To unearth the actual valence states of the metal elements during CO 2 RR, in situ Raman spectroscopy was carried out (Fig. 4f). For BiOI-Cu-0.5 precursor, the peaks at 87 and 151 cm -1 can be assigned to the A 1g and E g stretching modes of Bi-I bond, respectively 36 .
After cyclic voltammetry (CV) treatment, only peaks at 71 and 97 cm -1 were observed for BiCu-0.5, which are known as the E g and A 1g modes for metallic Bi 37 . This result strongly demonstrates the total conversion of BiOI-Cu precursor to metallic Bi in BiCu-0.5. Moreover, a similar metallic Bi Raman spectrum was recorded at -0.88 V in CO 2 -saturated NaHCO 3 solution, which con rms the maintenance of Bi 0 during CO 2 RR. According to the standard reduction potentials ( E 0 (Bi 3+ /Bi) = 0.317 V; E 0 (Cu 2+ /Cu) = 0.340 V; E 0 (Cu + /Cu) = 0.520 V) 38 , the reduction of Cu 2+ and Cu + to Cu 0 is more thermodynamically favoured than the reduction of Bi 3+ to Bi 0 . Fig. 4g displays the CV curves for Bi and BiCu-x catalysts. The reduction peak at 0.08 V can be assigned to the reduction of Bi 3+ to Bi 0 35 . Owing to the extremely low concentration of Cu elements in BiCu-0.5, no extra reduction peak was observed. Nonetheless, a new peak at 0.15 V emerged in BiCu-3, corresponding to the reduction of Cu 2+ to Cu 0 39 . Therefore, we can deduce that both Cu and Bi atoms in BiCu-x are in zero valence state during the CO 2 RR. With all the analysis above, we can nally con rm the successful fabrication of BiCu SAAC with atomically dispersed Cu atoms in Bi lattice for CO 2 RR.
Theoretical study on the mechanism for enhanced formate production. To shed light on the correlation between the isolated Cu atom and the enhanced CO 2 RR performance, density functional theory (DFT) calculations were carried out. The (012) surface is adopted to construct the theoretical model. On the basis of our experimental results, we reasonably proposed a reaction mechanism involving H + reduction for the formate generation on the surface of Bi and BiCu-0.5 catalysts, as illustrated in Fig. 5a, b, respectively. The corresponding free energy diagrams are shown in Fig. 5c. The maximum free energy barrier (ΔG) on Bi surface is from H + to adsorbed *H (0.66 eV), indicating that the H + reduction is the rate-determining step (RDS). In contrast, the ΔG of H + reduction is reduced to 0.25 eV on the isolated Cu sites in BiCu-0.5. The projected density of states (PDOS) calculation results reveal that the d-band centre of the isolated Cu atoms (-2.42 eV) is closer to the Fermi level than that of Bi (-2.75 eV), which is bene cial to the hydrogen adsorption ( Supplementary Fig. 24). The above results were con rmed by the experiments of the kinetic isotope effect (KIE) of H/D over Bi and BiCu-0.5 catalysts. As shown in Supplementary Fig.  25 To gain in-depth information for the *H transfer step, the transition state (*TS) for the formation of *HCOO from adsorbed *H and *CO 2 was further simulated. As shown in Fig. 5c, the formation of *TS becomes the RDS on BiCu-0.5, with a maximum ΔG value of 0.42 eV, which can be attributed to the stronger interaction between *H and Cu atoms. Nonetheless, it is still lower than the ΔG of the RDS on Bi Extended assessment on CO 2 RR performances. Encouraged by the unique surface structure and outstanding CO2RR performance of BiCu-0.5, we move forward to investigate its potential in commercial utilization. Gas diffusion electrolyser (GDE) was employed to overcome the mass-transfer limitation at the electrode/solution interface ( Supplementary Fig. 26), thereby producing high current density to meet the industrial requirements 4 . As shown in Fig. 6a, b, a current density of 150 mA cm -2 with a FE formate of 94.8% was achieved at -0.48 V vs. RHE, and the current density and FE formate remained virtually unchanged after electrolysing 240 min, suggesting the remarkable stability of BiCu-0.5 under high current density. Moreover, both an extremely high current density of 450 mA cm -2 and a high FE formate of 96.5% can be achieved at -0.55 V vs. RHE. In sharp contrast, BiCu-3 only achieved a current density of 130 mA cm -2 with a FE formate as low as 37.3% at -0.55 V vs. RHE; the FE formate was further reduced to 30.6% at -0.70 V vs. RHE, accompanied with a current density of 320 mA cm -2 ( Supplementary Fig. 27). The low FE formate of BiCu-3 can be attributed to the existence of Cu nanoclusters on the surface of Bi nanosheets, which are favourable for H 2 rather than formate production. This result further demonstrates the crucial role of single-atom Cu in BiCu-0.5 for promoting CO 2 electroreduction to formate with both high FE formate and high j formate . Fig. 6c displays a comparison of BiCu-0.5 with the reported state-of-the-art Bi-based electrocatalysts for reduction of CO 2 to formate (detailed information is listed in Supplementary Table 2).
To our knowledge, the BiCu-0.5 SAAC exhibits the highest j formate values among the reported Bi-based electrocatalysts with excellent FE formate in both H-type cell and GDE. Another issue that hinders the utilization of the electrocatalytic technique is the requirement of high-purity CO 2 as a feedstock for achieving high Faradaic e ciency, whereas the concentration of CO 2 in ue gas is only about 10% 41,42 .
With this in mind, we performed the CO 2 RR using 10% CO 2 as the feedstock. As shown in Fig. 6d, the FE formate can even approach over 80% in a wide potential window, and a value as high as 86.9% is achieved at -0.98 V, highlighting the great potential of BiCu-0.5 SAAC for CO 2 RR in commercial applications.

Discussion
In summary, we have successfully synthesized the BiCu nanosheets SAAC featuring isolated Cu atoms in Bi lattice. Bene ting from the unique surface structure, the BiCu-0.5 exhibits an outstanding electrocatalytic activity toward CO 2 reduction to formate with a current density up to 450 mA cm -2 and a was achieved owing to the promoted H 2 production on Cu clusters, in contrast to *H formation on isolated Cu sites in BiCu-0.5. Our work provides an ideal platform to understand the roles of both metals of "isolated single-atom" and support in SAACs during catalytic process, and opens up new perspectives for rational design of SAACs for highly e cient electroreduction of CO 2 to value-added products.

Methods
Preparation of BiCu-x catalysts. BiOI-Cu-x nanosheets were rst synthesized according to a modi ed reported method as following 9 Fig. 26). The gas diffusion electrode (GDE) was fabricated by depositing BiOI-Cu-0.5 sample onto a 3 × 3 cm 2 gas diffusion layer (Sigracet 29 BC) with a mass loading of 1 mg cm -2 , followed by the in situ electrochemical reduction as described above. During the CO 2 RR test, the exposed area of GDE was xed to 1 cm 2 by a shaped gasket. A Ag/AgCl electrode and a 3 × 3 cm 2 Pt foil were used as the reference electrode and the counter electrode, respectively. KOH aqueous solution (1.0 M) was used as the electrolyte with a ow rate of 2.5 mL min -1 , and the ow rate of CO 2 was xed at 30 sccm by a mass ow controller. Ohmic resistance compensation with a level of 80% was applied for the electrolysis.
In situ Raman spectroscopy measurements. In situ Raman spectroscopy measurements were performed using a WITec alpha300 R confocal Raman imaging system with a 633 nm laser as the excitation source.
A standard silicon wafer (520 cm -1 ) was employed to calibrate the Raman system. A CCD camera with 1650 × 1650 pixels was used to photograph the sample surface and collect the spectra. In a specially designed two-compartment cell, the sample was spread on a glassy carbon electrode with its planar surface perpendicular to the incident laser. The electrode surface was covered by a thin layer of electrolyte (0.5 M NaHCO 3 ). Each compartment of the cell was sealed with a quartz window, and equipped with two channels for gas ow.
Computational details. The Vienna ab initio simulation package (VASP) combined with the projected augmented wave (PAW) method was used to perform all the calculations. The generalized gradient approximation (GGA) was adopted, with the Perdew-Burke-Ernzerh (PBE) chosen as the exchangecorrelation functional. An energy cutoff of 450 eV was set for the plane wave basis set. The van der Waals interactions were described using the DFT-D3 method with Becke-Jonson damping. A 3 × 3 × 1 Monkhorst-Pack k-point was employed. The convergence thresholds for energy and force were set to 10 -5 eV and 0.02 eV Å -1 , respectively.
In this work, a 2 × 2 supercell of bismuth bulk was optimized and the (012) surface with six atomic layers was constructed as the surface slab. During the calculations, the top two layers were fully relaxed. To avoid the interaction between two units, the vacuum layers were set to be 15 Å. The Gibbs free energy of each intermediate in the simulated pathway was calculated as follows: Where E DFT is the electronic energy directly obtained from DFT calculations, E ZPE is the zero-point vibrational energy, ∫C V dT is the heat capacity, T is the temperature (298.15 K), and S is the entropy.

Declarations
Data availability The data that support the plots within this paper and other ndings of this study are available from the corresponding author upon reasonable request.