How local electric field regulates C–C coupling at a single nanocavity in electrocatalytic CO2 reduction

Nano-confined Raman spectroscopy measurementsTo attain high−resolution spectroscopic analysis of single nanocavities on Cu catalyst during CO2RR, we develop a nano-confined hyperspectral Raman microscopy based on surface plasmon resonance-coupled Raman scattering29. As shown in Fig. 1a, the optical setup includes an inverted high numerical aperture oil immersion objective below and an upright water immersion objective above. A glass slide is placed between two objectives, on which a 47-nm-thick Au film is coated. A collimated 661.0-nm p-polarized laser beam is directed onto the gold film with an appropriate incident angle to generate surface plasmons, further exciting the Raman scattering on the surface. To form the Raman image, a 664 nm long-pass filter stops the Rayleigh scattering from the excitation laser, and only Raman signals enter the spectrometer, which is recorded by a slit-free monochromator equipped with an EMCCD (details in Methods).Fig. 1: Schematic illustration of the construction of single-site-resolved Raman spectroscopy apparatus.a Optical setup of the nano-confined hyperspectral Raman microscopy. SPP is the surface plasmon polaritons. WE, RE, and CE are working electrodes, reference electrodes, and counter electrodes, respectively. b Schematic illustration of the construction of nanogaps on Cu substrate. The Cu, C, H, and O atoms are presented in brown, black, white, and red balls, respectively. c Top (xy-plane) view of the theoretical electromagnetic enhancement factor, |Eloc/E0|4, in the nanogap. The scale bar is 2 nm. d, e Zero- (d) and first-order (e) Raman mapping of CO2RR products at single nanocavities in nanogaps when applied a potential of −0.6 V (vs. Ag/AgCl). Grating center wavelength: 0 (d) and 750 nm (e). Scale bar is 10 μm. Int indicates the intensity. f Converted Raman spectra of the nanocavities marked in d, e.To enable the investigation of uniform nanocavities for catalytical studies, we conduct the deposition of atomically flat Cu monolayer on an Au film-coated cover slide using the underpotential deposition (UPD) method (Supplementary Fig. 1)21,31,32,33. Following this, we drop-cast monodisperse Cu-coated gold nanoparticles (Au@Cu NPs) onto the gold film (Supplementary Figs. 2, 3). The Au@Cu NPs and the substrate are separated by cetyltrimethyl ammonium bromide (CTAB) molecules, generating a nano-confined hotspot for Raman signal enhancement (Fig. 1b)28,29. It is worth noting that the CTAB molecules are stable when exposed to the laser beam due to the low power density (Supplementary Fig. 4). This setup allows for the exclusive investigation of the nanocavity within the nanoscale hotspot (Fig. 1c). By applying a constant voltage (–0.6 V vs Ag/AgCl) in 0.1 M CO2-saturated KOH, CO2 is reduced at the nanocavity, resulting in the formation of various products, such as CO and C2. The zero-order (mirror mode) and first-order (grating mode) Raman spectral images of individual nanocavities are presented in Fig. 1d, e, respectively. The clear scattering spots in the zero-order image indicate the locations of individual nanocavities, and the dispersions containing bright dots in the first-order image provide spectral information. These dispersions could be directly converted into Raman spectra by ref. 29.$$\nu=1/{\lambda }_{{{{\rm{ex}}}}}-1/(k{d}_{{{{\rm{eff}}}}}+{\lambda }_{0})$$
(1)
where \(\nu\) is the Raman shift, \({d}_{{{{\rm{eff}}}}}\) = \({{{\rm{\alpha }}}}({{{\rm{x}}}})d\) is the effective separation (in pixel) between Raman bands in first-order dispersions and corresponding zero-order scattering spots, \({{{\rm{\alpha }}}}({{{\rm{x}}}})\) is the correction coefficient for separations \({d}\), \(k\) is calibrated to be 0.2071 nm pixel−1, \({\lambda }_{0}\) is 750 nm, which is determined by the grating center wavelength, and \({\lambda }_{{{{\rm{ex}}}}}\) = 661.0 nm is the excitation laser wavelength.By applying Eq. (1) to the first-order dispersions, we obtain Raman spectra of the CO2RR products at these single nanocavities (Fig. 1f). While most of the nanocavities exhibit similar shape of Raman spectra, we still observe the high heterogeneity in the products among these nanocavities, which are usually obscured in conventional in situ Raman methodologies. In contrast, the control group immersed in CO2-free electrolyte shows no distinguishable Raman peaks except for a characteristic Cu−OH peak during voltage scanning (Supplementary Fig. 5)16. Furthermore, regions located outside the nanocavities do not show any products, even when subsequently positioned within the nanocavities (Supplementary Fig. 6). This compellingly suggests that the collected Raman signals originate from the nanocavities rather than neighboring areas. Further control experiments are carried out with Cu-free AuNPs and Au film. As shown in Supplementary Fig. 7, a broad peak ranging from 560–613 cm–1 was observed when applied a negative potential, attributed to the Au-OH/AuOx structure34. Furthermore, peaks near 2100 cm–1 attributed to CO* and very weak peaks around 1540 cm–1 associated with carbonate were also observed. This reduction in Raman signal of carbonates may be attributed to the extensive release of CO. The Raman peak associated with C2 could not be detected due to the inability of Au to undergo C–C coupling.Evolution of C–C coupling intermediates at single nanocavitiesTo observe the evolution of C–C coupling intermediates, we examine the potential−dependent behavior of intermediate species involved in CO2RR. As shown in Fig. 2a, we conduct a step potential experiment from 0 to –1.0 V (vs. Ag/AgCl). The time-resolved Raman spectra of a single nanocavity are visualized as a heat map, where y axis represents the Raman wavenumber, x axis denotes the time and step potential, and z axis corresponds to the Raman scattering intensity. By acquiring Raman spectra in the interval of 0.5 s, we achieve dynamic monitoring of the active species changes over time and under different potential steps at this nanocavity. Figure 2b presents snapshots of time-resolved Raman spectra at different potentials to facilitate the identification of the peak associated with the products. The dynamic tracing of Raman spectra provides rich details, either the potential or time-resolved evolution of the reaction intermediates formed at the nanocavity. When the potential is at 0 V, only a set of Raman peaks around 1480−1580 cm−1 are visible. The origin of these peaks is still controversial in the reported literature, and we tentatively assigned them to carbonates and *CO2−22,35,36,37. As the potential steps to –0.2 V, a distinctive peak appears at 1070 cm−1 which can be attributed to the vibration mode of carbonates15,17,22,38. When the potential reaches –0.4 V, Raman peaks associated with CO2RR become distinguishable. Initially, peaks at 1325 cm−1 and 1558 cm−1 show a notable rise. Previous studies have identified them as *CO2−, crucial precursors for CO generation15. The rapid accumulation of reactants facilitated the formation of *CO located around 1854 cm−1 and 2100 cm−1, corresponding to bridge-bonded CO and linearly-bonded CO at the interface, respectively16,18. In the low wavenumber region, a broad peak located at 270–350 cm−1 emerges which can be assigned to Cu-C. Of particular note are two prominent Raman peaks at 973 cm−1 and 1278 cm−1, which simultaneously appear and exhibit remarkable similarity in the Raman pattern. Tentatively, these can be attributed to initial intermediates in C–C coupling, denoted as O*CCHO39.Fig. 2: Time-resolved Raman spectra at a single nanocavity.a The top panel illustrates the applied step potential from 0 to –1.0 V (vs. Ag/AgCl). The bottom panel shows the time-resolved Raman spectra at a single active site in 0.1 M CO2-saturated KOH. b Snapshots of the time-resolved Raman spectra in a when the applied potential scans to OCP, −0.2 V, −0.4 V, −0.6 V, −0.8 V, and OCP again, respectively. OCP is open circuit potential.Upon further decreasing the potential to –0.6 V, the Raman peak at 1325 cm−1 is replaced by a peak at 1373 cm−1 and eventually becomes invisible, possibly due to overlapping of Raman peaks. Considering subtle kinetic difference from O*CCHO, it is reasonable to assign the band at 1378 cm−1 to the C=C stretching vibration of *OCHCHO*, which is formed via protonation of O*CCHO13. As switching potential to −0.8 V, the peaks at 973 cm−1, 1278 cm−1 and 1373 cm−1 become more evident. All these intermediates’ peaks begin to gradually weaken when switching potential to −1.0 V, resulting from the competition of electrocatalytic hydrogen evolution reaction (HER). The overwhelming HER reaction generates substantial bubbles at the interface which inhibit the CO2RR process. Intriguingly, after the termination of the reducing potential, the peak appears at 1230 cm−1, which is assigned to the C−OH stretching mode of OC**COH according to the previous report18. Probable reason is that OC**COH is a kind of long−lived oxygen−containing intermediate in CO2RR38, which is formed elsewhere and diffuses to this nanocavity.To further confirm the formation of these intermediates, we perform isotopic substitution experiments. In the D-substituted experiment, all the primary peaks of O*CCHO* (973 and 1278 cm−1), *OCHCHO* (1373 cm−1) and OC**COH (1230 cm−1) underwent a red-shift, indicating the influence of at least one C−H in their vibrational modes (Supplementary Fig. 8). Meanwhile, 13C-substitution experiments demonstrate conspicuously red-shifts in the peaks at 973 cm−1, 1278 cm−1, and 1373 cm−1, relocating to 943 cm−1, 1246 cm−1, and 1325 cm−1, respectively (Supplementary Fig. 9a). Besides, the peak assigned to C−OH at 1230 cm−1 shifts to 1210 cm−1 (Supplementary Fig. 9b) and the peak assigned to Cu−C at 331 cm−1 shifts to 323 cm−1 (Supplementary Fig. 10). All these peak shifts are in line with the theoretical result of mass formula (see Methods)15,18,40. More evidences are obtained by statistical results of plenty of single nanocavities. Supplementary Fig. 11 illustrates the wavenumber distributions of the above four peaks at more than 40 nanocavities in the 13C-substitution experiments. The isotopic experiments provide evidence supporting our identification of the CO dimers.Heterogeneity of single nanocavitiesAs mentioned earlier, different C–C coupling pathways may occur among these single nanocavities. To acquire a more comprehensive understanding of the evolution of C–C coupling intermediates, we conduct the time-resolved Raman spectroscopy for two presentive nanocavities when applied a potential of –0.6 V. For the nanocavity in Fig. 3a, the typical CO emerged around 2100 cm–1 is weak initially and then disappears accompanied by the appearance of O*CCHO (973 and 1278 cm–1) and *OCHCHO* (1373 cm–1), indicating the complete depletion of CO during the vigorous C–C coupling process. For another nanocavity shown in Fig. 3b, a swift accumulation of CO (2100 cm–1) is evident once the reduction potential is applied, coinciding with the appearance of OC**COH (1230 cm–1). It is worth noting that the band at 1230 cm–1 intermittently switches to 1278 cm–1, suggesting a competing relationship between the two types of C–C coupling pathways. After the band of CO reaches maximum, the band located at 1230 cm–1 vanishes entirely, supplanted by the observable appearance of the O*CCHO (1278 cm–1). Following this transition, the Raman peak of CO exhibits a pronounced attenuation characterized by similar features as observed earlier.Fig. 3: Cavity-specific C–C coupling pathways revealed by the time-resolved Raman spectra.a, b Two types of the evolution of Raman spectra at single nanocavities when applied a potential of –0.6 V (vs. Ag/AgCl). c–f The representative reaction pathways during CO2RR process. The Cu, C, H, and O atoms are presented in brown, black, white, and red balls, respectively. c The adsorption of carbonate and bicarbonate occurs at −0.2 V (vs. Ag/AgCl). d Substantial production of CO is observed at −0.4 V (vs. Ag/AgCl). e, f Two potential pathways leading to the formation of C2 species at potentials at −0.6 V (vs. Ag/AgCl). g The vibrational modes of the relevant intermediates.Two distinct reaction pathways occur alternately at each nanocavities. This phenomenon arises from the unique properties of the nanocavities, which creates different local microenvironments. Figure 3c–f detail the mechanistic implications and conclusions resulting from our time-resolved kinetic Raman spectral analysis. The vibrational modes of the relevant intermediates in Fig.3c–f are displayed in Fig. 3g. The mechanism of the first pathway is the proton−assisted C–C coupling, which is widely acknowledged in CO2RR4. CO2 is initially reduced to CO and adsorbed *CO molecules on Cu undergoes hydrogenation to yield *CHO, which rapidly combines with *CO to generate O*CCHO, followed by a proton−coupled electron transfer step leading to *OCHCHO* (Fig. 3e)5,13,41. Therefore, it is challenging to collect the Raman signal of *CHO (~1720 cm−1)42,43,44, which usually manifests as weak shoulder peak after the entirely disappearance of *CO (Fig. 3a).For those nanocavities exhibiting a high accumulation of CO, the direct dimerization pathway of CO is highly favored4, leading to the discernible characteristic Raman peak at 1230 cm−1. We assign it to the C−OH stretching vibration mode of OC**COH, which is the first hydrogenation product of OC**CO (Fig. 3f)45,46. OC**COH is also the most stable intermediate after the first hydrogenation step of CO dimers, in line with our experimental results and previous computational results38. This phenomenon implies that this pathway is more prevalent at the nanocavities that can produce CO at very low overpotentials. The variation in overpotentials for CO generation at these nanocavities leads to different rates of localized CO accumulation, providing undeniable evidence of the significant impact of local environment in nanocavities on the selectivity of C–C coupling pathway.Mechanistic insightsOne plausible explanation for the deviation in the overpotential among these nanocavities is the cation effect, which is the most widely accepted mechanism for enhancing CO2RR47,48,49,50,51,52,53,54. Nanostructured electrocatalysts can produce local high electric fields that concentrate electrolyte cations, especially K+, which in turn leads to a strong dipolar field. The strong dipolar field effectively stabilizes intermediates involved in CO2 conversion to CO and modulates the coverage and adsorption strength of *CO at the interface, thereby promoting the production of C2 products48,49,51,53. To estimate the impact of the electric field on the surface-adsorbed K+ concentration in nanocavities, we use Gouy–Chapman–Stern model to simulate the surface-adsorbed K+ ion density. As shown in Supplementary Fig. 12, the simulated results indicate a significant increase in the surface-adsorbed K+ ion concentration in nanocavities due to locally enhanced electrostatic field. However, the K+ concentration control experiments cannot support this hypothesis. Despite the fact that increasing electrolyte concentration enriches the gap region with a higher concentration of cations and reactants, however, the onset potential distribution of CO shifts towards lower potentials (Supplementary Fig. 13). Therefore, the cation effect, at least in this work, is not the main influence factor.The unexpected results promote us to seek another dominant mechanism to elucidate the dissimilar electrocatalytic behaviors of individual nanocavities. Recent studies have shown that the electric field and charge density at the catalyst interface have a fundamental impact on the stability of intermediates, catalyst activity, and even product selectivity55,56,57,58,59. For instance, Arenz et al. proposed the mechanism of overlapping EDLs as the cause for the distance-dependent enhancement of the oxygen reduction reaction (ORR) current in nanogaps60. The overlapping EDLs in nanogaps between neighboring nanoclusters may modify the potential drop within the compact layer, leading to a significant increase in ORR activity. This inspires us to develop an explanation for the nanocavity-specific electrocatalytic activity of CO2RR. Although the gap distance at each nanocavity is difficult to be controlled with atomic-level precision, we can measure the gap distance using the plasmonic coupling scattering of single Au@Cu NPs and correlate it with the Raman spectra61.The varied gap distances between individual Au@Cu NPs are depicted in Fig. 4a. Three Au@Cu NPs (labeled as 1–3) exhibit increasing scattering intensities, indicating different gap distances. Once the supporting molecules are removed (see Methods), the three particles directly contact the surface and display comparable scattering patterns, suggesting their uniform dimensions (Fig. 4b). It is worth noting that the scattering intensities of these particles are significantly reduced compared to before, which can be attributed to the mismatch in scattering peak wavelength with the incident light (Supplementary Fig. 14). We subsequently investigate the onset potential of CO at the nanocavities under these Au@Cu NPs prior to the removal of supporting molecules, as it is closely associated with the site-specific pathways. Figure 4c illustrates that the Au@Cu NP with lower scattering intensity has a higher onset potential of CO (see Supplementary Fig. 15 for the complete spectra). This result promotes us to thoroughly investigate the relationship between different gap distances and the onset potential of CO. To achieve this, we develop a finite-difference time-domain (FDTD) model that describes how scattering intensity varies with gap distance, using a fixed incident laser wavelength of 661.0 nm (see Methods). Figure 4d illustrates two inverse relationships between scattering intensity and gap distance on both sides of the 1.5 nm gap. The highest scattering intensity corresponds to a scattering peak wavelength that aligns with the incident laser, with longer wavelengths observed on the left side61. The statistical analysis of the single-particle scattering spectra shows that the coupling scattering peak wavelength of CTAB-capped Au@Cu NPs is primarily concentrated ~670 nm. Replacing CTAB with polyvinyl pyrrolidone (PVP) or removing CTAB results in blue or red shifts in the scattering spectra, respectively (Supplementary Fig. 14). Since we utilize a laser with a fixed wavelength of 661.0 nm, most of the gap distances fall within the range of 0–1.5 nm. This approach allows for qualitative analysis of the gap distance based on scattering intensity. The statistical results of nanogaps are summarized in Fig. 4e, clearly demonstrating a strong dependence of the gap distance on the measured onset potential of CO. Smaller gap distances correspond to higher onset potentials of CO.Fig. 4: Correlation of onset potential of CO and the gap distance.a, b Dark-field scattering image of single Au@CuNPs before (a) and after (b) removal of gaps. Scale bar is 1 μm. c Potential-dependent Raman spectra of marked Au@Cu NPs in a. Vertical lines indicate the onset potential of CO. d FDTD simulations of the scattering intensities of single Au@Cu NPs versus the gap distances. e Correlation of the measured onset potential and the calculated gap distance. The dashed line indicates the observed trend.Further insights into the gap distance-dependent onset potentials are obtained by performing computational modeling focusing on interactions between the overlap of EDL and the associated effect on the potential distribution in the compact layer. The extent of EDL overlap is determined by the ion concentration and gap size, prompting separate finite element analysis simulations to investigate the influence of different degrees of EDL overlap on the potential of the compact layer. Initially, the impact of Debye length on EDL overlap is examined. The Debye length \({x}_{D}\) for a monovalent binary electrolyte can be described by the concentration \({c}_{{bulk}}\) and solvent relative permittivity \({\varepsilon }_{r}\): \({x}_{D}=\sqrt{\frac{{RT}{\varepsilon }_{r}{\varepsilon }_{0}}{2{F}^{2}{c}_{{bulk}}}}\). As the electrolyte concentration increases, the Debye length decreases. Figure 5a, b exemplify the potential distribution at the compact layer for two configurations with different concentrations, clearly revealing the dependence of the potential distribution on the Debye length. Figure 5c demonstrates the variation in potential at the compact layer with different electrolyte concentrations. Switching from 0.1 M to 2.0 M causes a 100-mV increase in the calculated potential of the compact layer. The simulated results strongly validate that the dominant factor for the overpotential is the EDL overlap rather than the cation effect.Fig. 5: Simulation of the impact of overlapping EDL on the compact layer potential.a, b Schematic illustration of the EDL (top) and simulated potential distribution (bottom) within a 1.0 nm gap immersed in 0.1 (a) and 2.0 M (b) electrolyte. c Simulated compact layer potentials \({V}_{{{{\rm{c}}}}}\) (open circles) and the measured distribution of CO onset potential (boxes) as function of electrolyte concentration. d, e Schematic illustration of the EDL (top) and simulated potential distribution (bottom) within a 0.5 (d) and 2 nm (e) gap immersed in 0.1 M electrolyte. f Simulated compact layer potentials as a function of gap distance. In all cases a constant reduction potential of −0.4 V (vs. Ag/AgCl) is applied at the surfaces of the substrate.We further conduct simulations on nanogaps with varying sizes while maintaining a fixed electrolyte concentration of 0.1 M, resulting in a relevant Debye length of approximately 0.96 nm (Methods). Figure 5d, e illustrates the model schemes and the corresponding potential mappings for different gap distances. When there is a larger gap distance (d = 2 nm), there is no overlap of EDLs, thus leaving the potential at the compact layer unaffected. In this scenario, the potential distribution at the compact layer is equivalent to that of an isolated nanosphere (~–200 mV). However, for gap distances smaller than the Debye length (d = 0.5 nm), the Debye screening is less efficient, resulting in EDL overlapping, which leads to an increase in the potential at the compact layer to approximately –250 mV. Figure 5f presents the variation of the potential at the compact layer for various gap distances, showing a trend similar to the onset potential for CO in Fig. 4d, thus validating our hypothesis.By carefully correlating the simulation outcomes with experimental results, we can deduce that the overlap degree of EDLs can serve as a reliable indicator of the selectivity of C–C coupling pathways. A higher degree of EDL overlapping will promote a significant accumulation of CO, making the direct dimerization pathway of *CO highly favored. To verify it, a comparison is made between C–C coupling intermediates at a specific nanocavity with and without EDL overlapping. Figure 6a illustrates the Raman spectra of a single nanocavity in 0.1 M CO2 saturated KOH, where the overlapping of EDL results in a decreased overpotential and a pronounced CO buildup at the nanocavity when a potential of −0.4 V is applied. Concurrently, the appearance of a peak at 1230 cm–1 suggests the occurrence of the direct dimerization pathway of *CO. However, upon the addition of 1.0 M CO2-saturated KOH, the higher ionic strength reduces the Debye length, causing the separation of overlapping EDLs. Consequently, only a weak CO peak is observed at the same nanocavity until a lower potential (−0.6 V) is applied (Fig. 6b). The emergence of peaks at 973, 1278, and 1373 cm–1 indicates the presence of CHO–CO coupling pathway. This clear contrast inspires the design of highly selective catalyst materials tailored to promote specific C–C coupling pathways and downstream products.Fig. 6: Tuning of C–C coupling pathways by changing the Debye length.a Raman spectra of a single nanocavity in 0.1 M CO2-saturated KOH when applied a potential of −0.4 and −0.6 V, respectively. The inset shows the schematic illustration of overlapping EDLs. b Raman spectra of the single nanocavity in 1.0 M CO2-saturated KOH when applied a potential of −0.4 and −0.6 V, respectively. The inset shows the schematic illustration of separated EDLs. The Cu, C, H, and O atoms are presented in brown, black, white, and red balls, respectively.