Introduction

The electrochemical reduction reaction of CO2 (CO2RR) to higher-value chemicals holds great promise for mitigating climate change and transitioning the chemical industry towards sustainability1,2. Among various heterogeneous electrocatalysts, copper (Cu) was found uniquely capable of producing multicarbon products, such as ethylene, ethanol, and propanol, at significant rates3,4. While previous research5,6,7,8 mainly aimed at improving the activity and selectivity of Cu catalysts, their stability under reaction conditions is at least equally important and has drawn increasing research attention9,10. Some studies demonstrate the stable operation of Cu catalysts for tens or hundreds of hours11,12,13. In contrast, others show substantial performance fluctuations within an hour or less14,15. The existence of such a discrepancy in literature is puzzling since essentially the same material, copper, was being tested. In addition to the possible causes associated with electrochemical reactors (e.g., flooding)16, it is widely acknowledged that Cu catalysts undergo significant reconstructions during the electrocatalytic reactions that may also affect its stability17,18,19. Buonsanti and colleagues20, as well as other researchers21,22,23 showed that this potential-induced reconstruction is the primary deactivation mechanism for shape-controlled Cu catalysts. However, in contrast, other studies by Jung et al. 24, by Choi et al. 25, and more recently by Yang et al. 26 suggested that the reconstruction can create active sites in Cu catalysts.

Experimentally studying the reconstruction of Cu under CO2RR conditions has been challenging at least partially due to the susceptibility of Cu to be oxidized27. Recently, many in situ characterization methods, such as in situ electron microscopy23,26,28,29, in situ scanning probe microscopy17,18,30, in situ X-ray scattering31 were developed and applied to study the reconstruction of Cu catalysts. However, the underlying causes and the impact of such a phenomenon on catalyst stability remain still elusive. For instance, some researchers13,24 attributed it to the reduction of copper oxides, while others showed that CO32,33 or H234,35 played important roles. The limited understanding of this reconstruction behavior of Cu catalysts in the cathode made it difficult to clarify the structural-property relationship, possibly also fueling debates regarding issues such as the identity of the active sites in Cu catalysts36 and whether oxide-derived Cu is intrinsically better than just Cu8,37,38.

In this work, we track the structural evolution of Cu nanocubes under relevant CO2RR reaction conditions using a combination of identical location TEM (IL-TEM), cyclic voltammetry (CV), in situ X-ray Absorption Fine Structure Spectroscopy (XAFS), Scanning Transmission Electron Microscopy (STEM), Density Functional Theory (DFT) and grand-potential Ab Initial Molecular Dynamics (AIMD) simulations. Our results suggest that Cu catalysts reconstruct via a hitherto unexplored yet critical pathway - alkali cation-induced cathodic corrosion39, when the electrode potential is more negative than an onset value (e.g., −0.4 VRHE when using 0.1 M KHCO3). Having alkali cations in the electrolyte is found critical for this phenomenon to take place. Our experiments and modeling suggest that Cu can be etched out by possibly forming intermediate ternary hydride40,41 with alkali cations, which can then be redeposited as metallic Cu under cathodic conditions, leading to experimentally observed reconstruction of the Cu catalyst and the formation of smaller Cu particles during the CO2RR reaction.

Since CO2RR usually requires highly cathodic conditions and having alkali cations in the electrolyte42,43,44,45, Cu catalysts will inevitably undergo surface reconstructions that lead to dynamic morphologies. We will argue that having this dynamic morphologies does not necessarily preclude stable operation using Cu catalysts as the catalyst morphology may potentially reach an equilibrium. However, because of the alkali cation-induced cathodic corrosion, many approaches towards engineering the morphology of Cu pre-catalyst may unlikely bring long-term selectivity/activity benefits46,47,48. Alternatively, this observation suggests that strategies can be designed to mitigate selectivity changes by reducing the extent to which the cathodic corrosion of Cu can occur. To this end, by operating Cu catalysts at less negative potentials than the threshold (e.g., −0.4 VRHE) in the CO electrochemical reduction reaction (CORR)49,50,51, we demonstrate that the cathodic corrosion can be switched off, allowing Cu nanocubes to maintain a stable selectivity advantage over spherical Cu nanoparticles.

Results

A hitherto unexplored yet critical reconstruction pathway for Cu catalysts

We use Cu nanocubes (NCs) as our model catalysts due to their well-defined initial structure, enabling us to closely track the structural evolution of the catalysts. Cu nanocubes were prepared following a previously reported protocol (see “Methods” section)7. As shown in Fig. 1a, the as-synthesized NCs are largely uniform with an average size of approximately 40 nm, although occasionally NCs of different sizes and spherical Cu particles can also be found. The metallic nature of the NCs is confirmed by electron diffraction (inset of Fig. 1a and Fig. S1). Despite our efforts to limit air exposure (see “Methods” section), we still anticipate the presence of thin surface oxide shells on the Cu NCs. Through scanning transmission electron microscopy (STEM) imaging and electron energy loss spectroscopy (EELS), we show that the surface oxide layers on Cu NCs that were subjected to our typical sample handling procedures are about 2 nm in thickness (Fig. S2). As part of our control experiments, we have also considered extreme cases wherein Cu catalysts were deliberately stored or exposed to air for extended periods (Figs. S3 and S4), where the Cu oxide layers were found no thicker than 5 nm. Based on these observations, it is reasonable to assume that the oxide layer on the Cu NCs that will be discussed in this manuscript will typically be around 2 nm and unlikely to exceed around 5 nm.

Fig. 1: IL-TEM characterization of Cu nanocubes under CO2RR at −1.1 VRHE.
figure 1

a Representative TEM bright field images of Cu nanocubes about 40 nm in size. The inset shows a selected area electron diffraction pattern, confirming the metallic Cu crystal structure. b Schematic demonstration of the Identical location TEM (IL-TEM) to study Cu catalysts at cathodic conditions, using a standard RDE electrode with a custom-made PTFE cap and Au TEM grid with Formvar, coated with carbon on both sides. (see “Methods” section). c–f Identical location TEM bright field images before and after treating the catalysts at −1.1 VRHE (iR-corrected) using in a CO2-saturated 0.1 M KHCO3 electrolyte (Table S1, entry 2) for 20 min. The morphology changes before (red) and after (blue) CO2RR are highlighted using outlines, which are significantly beyond the thickness of the surface oxide layer. The current recorded during the IL-TEM experiment can be found in Fig. S7.

Full size image

As-synthesized Cu NCs were tested for CO2RR at −1.1 VRHE using a typical H-cell setup and a CO2-saturated 0.1 M KHCO3 electrolyte for 10 h (Table S1, entry 1). Significant changes in reaction selectivity among gas-phase products were observed (Fig. S5), as were the morphological changes of the Cu catalysts and the formation of smaller Cu nanoparticles (Figs. S5 and S6). Identical location TEM (IL-TEM) was applied (Fig. 1b) to follow individual NCs before and after treating the catalysts at desired electrode potentials, using a method modified from the work by Mayrhofer et al.52 (see “Methods” section). A pair of representative IL-TEM bright field images of the catalyst before and after treating the catalysts at −1.1 VRHE in a CO2-saturated 0.1 M KHCO3 electrolyte (Table S1, entry 2) for 20 min were shown in Fig. 1c. Evidently, almost all Cu particles within this view exhibit clear signs of morphological changes after the treatment, irrespective of their size and shape. Moreover, as shown in Fig. 1d–f, it is apparent that the morphological changes of the Cu NCs extended deep into the core of the particle, significantly surpassing the estimated thickness of the earlier discussed oxide crusts. This suggests that the reduction or dissolution of surface oxides is unlikely the sole contributor to the morphological changes of the Cu catalysts.

Intriguingly, when we removed CO2 gas from the reaction and repeated the IL-TEM experiments at −1.1 VRHE using Ar-saturated 0.1 M KHCO3 electrolyte (Table S1, entry 3), a very similar catalyst reconstruction was observed (Fig. 2a, b). In contrast, the morphological changes observed on nanocubes subjected to electrochemical treatment under 0 VRHE (Table S1, entry 4) or slightly more negative potential of −0.36 VRHE (Table S1, entry 25) in Ar-saturated 0.1 M KHCO3 electrolyte, are primarily confined to less than 5–6 nm (Figs. S8–S10), which matches the estimated thickness of oxide layers. When we replace the electrolyte with Ar-saturated 0.1 M K2SO4 (Table S1, entry 5) or 0.1 M K2CO3 (Table S1, entry 6) and applied −1.1 VRHE, significant catalyst reconstruction will again be observed (Fig. S11). These results suggested that CO-related mechanisms32 are likely not the only possible mechanism responsible for observed morphological changes in Cu catalysts.

Fig. 2: IL-TEM and CV characterization of Cu nanocubes with different reaction conditions.
figure 2

a–d Representative IL-TEM bright field images of Cu nanocubes before and after 20 min of reaction at about −1.1 VRHE (iR-corrected) in Ar-saturated 0.1 M KHCO3 (Table S1, entry 3). The morphology changes of individual nanocubes before (red) and after (blue) the reaction are highlighted using outlines, which are significantly beyond the surface oxide layer. The current recorded during the IL-TEM experiment can be found in Fig. S7. e Voltammetric profiles of the Cu nanocubes recorded after 20 min (black curve) and 65 min (red curve) of reactions at about −1.05 VRHE in Ar-saturated 0.1 M KHCO3 (Table S1, entry 3). f Voltammetric profiles of the Cu nanocubes recorded after 20 min (green curve) and 65 min (red curve) of reaction at about −1.05 VRHE in Ar-saturated 0.05 M H2SO4 (Table S1, entry 7). The purple curve represents the case where the Cu nanocubes first undergo 20 min of treatment in Ar-saturated 0.05 M H2SO4, followed by subsequent 45 min of reaction at about −1.05 VRHE (purple curve) after adding additional Ar-saturated K2SO4 (0.3 M) (Table S1, entry 8) to the electrolyte while withholding the electrode potential. 0.05 M H2SO4 was replaced by 0.05 M H2SO4 & 0.3 M K2SO4 through successive electrolyte replacement several times with half the volume of the electrolyte replaced each time. The arrows highlighted the changes at the region representing the {100} terraces and the regions representing surface defects on Cu, respectively.

Full size image

To address the limited sampling capacity of TEM, we implemented cyclic voltammetry (CV) as a complementary technique to provide an overall assessment of the structural evolution of Cu NCs53. CV is done in an electrochemical cell with a three-electrode configuration as in the IL-TEM setup, except that more catalysts were used to enhance OH- adsorption signals (see “Methods” section). Figure 2e show two voltammograms, recorded in 0.1 M NaOH after the catalysts were treated at about −1.05 VRHE in Ar-saturated 0.1 M KHCO3 (Table S1, entry 3) for 20 min and 65 min, respectively. This two-stage comparison helps to exclude the influence of surface oxide crusts, which should be reduced quickly (e.g., within 10 min for 40 nm thick Cu2O11) under such a highly cathodic condition. It is therefore reasonable to assume the surface oxides are completely reduced within 20 min under our experimental conditions. The voltammogram of the 65-min treated catalyst differs notably from its 20-min treated analogs in two main aspects: it exhibits a reduced feature associated with the {100}-specific OH-adsorption at around −0.1 V, and an enhanced feature related to surface defects Cu at above +0.3 V. These disparities observed are notably more substantial than those arising from the experimental uncertainties inherent between individual CV experiments (Fig. S12). This reinforces the suggestion that the reduction of surface oxides cannot be the sole reason for the reconstruction of Cu catalysts. It is also important to highlight that morphological changes occur on {100} facets and not just on the particle corners. As shown in Fig. 1c, we also observe significant morphological transformations occurring on spherical particles. These findings suggest that the shape of the particles and the potential field concentration are at least not the determining factors influencing the reconstruction process.

Considering the potential contribution of the hydrogen evolution reaction (HER) to the reconstruction of Cu under CO2RR relevant conditions34,35, we carried out additional CV experiments to study the catalysts being treated at about −1.05 VRHE using 0.05 M H2SO4 as the electrolyte (Table S1, entry 7). In this case, we observed only minor changes between voltammograms (Fig. 2f) associated with catalysts treated by 20 min and by 65 min. This suggests that in scenarios where only the HER occurs without the presence of alkali cations, the catalyst reconstruction is considerably less pronounced. Interestingly, changes in the voltammogram that correspond to significant reconstructions of NCs could be observed again once 0.3 M K2SO4 was added into the electrolyte, indicating that the presence of alkali cations in the electrolyte may promote the catalyst reconstruction. IL-TEM results of the catalysts being treated at about −1.1 VRHE in acidic electrolyte without (Table S1, entry 7) and with K+ (Table S1, entry 8) are shown in Figs. S13a and S12b, respectively, which are consistent with the CV observations. It’s important to emphasize that in our experiments, the electrolyte was thoroughly de-aerated to eliminate oxygen, which, if present, could lead to further dissolution of Cu, as reported in previous studies54. To experimentally demonstrate this, we intentionally reintroduced oxygen into the H2SO4 electrolyte by blowing air (Table S1, entry 9), which then resulted in significant copper dissolution observable through both IL-TEM and CV methods (Figs. S13c and S12d). These findings indicate that the further reconstruction of Cu catalysts, occurring after the initial surface oxides are removed, is unlikely due to additional reduction or dissolution of oxides. Increasing the K+ concentrations to 0.5 M and 1.5 M (Table S1, entries 10–11) led to apparently more obvious structural changes (Fig. S14). Significant reconstruction was observed when the catalysts were treated in electrolyte with other alkali cations including Li+, Na+, and Cs+ (Table S1, entries 12–19) both at constant voltage (Fig. S15) and at constant current (Fig. S16). This observation is consistent with the cathodic corrosion behavior reported for noble metals40.

These experimental findings indicate the existence of a previously unexplored yet critical pathway of reconstruction for Cu catalysts under CO2RR-related reaction conditions. The promoting effect of alkali cations makes the process closely resembling to the phenomenon of alkali cation-induced cathodic corrosion, which is known to cause the reconstructions of many metal electrodes, such as platinum, rhodium, and gold39. Koper and colleagues40,55 proposed that the presence of alkali cations can lead to the formation of intermediate ternary hydride with the cathode metal. Such species leave the electrode but are unstable and will be quickly redeposited onto the cathode, leading to a reconstruction. Although the detailed mechanism of cathodic corrosion is still not fully understood, likely due to the difficulties of identifying these transient species, researchers have already harnessed this phenomenon for applications, such as nanoparticle synthesis56. It is important to note that, to the best of our knowledge, alkali cation-induced cathodic corrosion has not been previously reported for Cu catalysts under CO2RR reaction conditions27.

Identifying an onset potential for the cathodic corrosion of Cu catalysts

Another important characteristic of cathodic corrosion is that it occurs only when the electrode potential is more negative than a certain onset potential55. This was found to also be the case for Cu catalysts. Figure S17 shows that the changes in CV features associated with {100} facets and defects become suppressed as the electrode potential was adjusted to more positive values of −0.82 VRHE, −0.62 VRHE, and −0.42 VRHE (Table S1, entries 21–23). We then repeated the experiment to ensure that the total charge transferred is approximately equivalent while varying the electrode potentials. As shown in Fig. 3a, a brief 3-min treatment at about −1.02 VRHE caused much more significant changes in the voltammogram compared to a 120 min additional treatment at a more positive potential of −0.36 VRHE (Table S1, entries 25–26). This indicates a more significant reconstruction in the case of −1.02 VRHE treated catalysts compared to the −0.36 VRHE treated, notwithstanding that the total amount of charge transferred in both cases was nearly identical (i.e., about 0.4 C, Fig. S18). The repeated CV experiments show consistent results as shown in Fig. S20. These results from the CV experiment also agree well with the ex-situ TEM (Fig. S19) and the IL−TEM results (Fig. 3b and Fig. S21) which show that the Cu NCs morphology remains largely intact after 150 min at −0.36 VRHE. Note that the images of the Cu nanocubes before the reaction shown in Fig. 3b have thin layers of surface oxides, making them differ slightly from those generating the signal of reference CV (30 min) in Fig. 3a since the 30 min of treatment under −0.36 VRHE or more will remove the surface oxides. We then repeated the experiment, this time comparing the CV of Cu nanocubes experienced a 60-min treatment at −0.36 V and the CV of Cu nanocubes with a 25-min treatment at −0.36 VRHE plus a 30-s treatment at about −1 VRHE using 1 M KOH as the electrolyte (Table S1, entries 23–24), and a similar trend was observed (Fig. S22). In sum, we found that there exists an onset potential of cathodic corrosion for Cu catalysts of approximately −0.4 VRHE under our experimental conditions.

Fig. 3: CV and IL-TEM characterization of Cu Nanocubes reacted at −0.36 VRHE.
figure 3

a Voltammograms of Cu nanocubes after 30 min (black curve) and 150 min (red curve) of reactions at around −0.36 VRHE (iR-corrected) in 0.1 M KHCO3 are presented (Table S1, entries 20 and 25). The purple curve depicts the profile after an initial 30-min reaction at around −0.36 VRHE, followed by a 3-min reaction at approximately −1.01 VRHE (Table S1, entry 26). Highlighted regions show changes related to defects and the {100} terraces. The suppressed {100}-related charge of catalysts treated at −1.01 VRHE versus −0.36 VRHE suggests a decreased population of {100} facets. The defects-related peak at about +0.37 V shows a marked enhancement compared to the results after 30 and 150 min at −0.36 VRHE. b Representative IL-TEM bright field images of Cu nanocubes before and after 150 min of reaction at −0.36 VRHE in 0.1 M KHCO3.

Full size image

In situ X-ray absorption fine structure (XAFS) spectroscopy studies were carried out in order to monitor the oxidation states of Cu during the CO2RR process. To eliminate the possible self-absorption effect in fluorescence mode57,58,59 and thus to have proper comparisons with a Cu foil standard, the transmission mode was adopted for the study. A schematic view of the in situ XAFS cell is shown in Fig. 4a. The electrochemical cell has a three-electrode configuration, with a total electrolyte thickness of 1.5 mm on one side of the electrode. This design ensures adequate transmission of X-ray photons. (see “Methods” section) To ensure that XAS captures the maximum amount of information about the sample surface, we employed the smallest attainable particle size of 7 nm7. This size represents the lower limit of controllable synthesis in our laboratory while maintaining a sufficiently large specific surface area. Figures 4b, c show Cu K-edge X-ray absorption near edge fine structures (XANES) recorded in 0.1 M KHCO3 as the Cu nanoparticles being treated at about −1.05 VRHE and −0.4 VRHE, respectively (Table S1, entries 29–30). At open circuit potential (OCP), Cu was found in a partially-oxidized state. After 15 min of cathodization at −1.05 VRHE, the spectra from the catalyst largely resembled the Cu foil reference, indicating oxide reduction to metallic Cu. Intriguingly, further cathodization at −1.05 VRHE led to a redshift for the shoulder feature relative to the Cu foil reference (Fig. 4b). Similar redshifts on the Pt L edge60 or Pd K-edgeS1, entries 24) to better match with the simulation parameters. The results, presented in Fig. S17d, reveal that the catalyst undergoes significant reconstruction at −1.3 VRHE, mirroring the behavior observed at −1.1 VRHE. This consistency validates the choice of parameters used in our modeling. We then conducted constant-potential Blue-Moon AIMD41,62,63,64,65,66 on snapshots at 300 K taken from a copper dissolution trajectory generated by slow-growth AIMD (see “Methods” section). Here, the Blue-Moon AIMD simulation samples microstates of each window along the dissolution-reaction trajectory. Figures 5a, b show the statistical free energy gradient (purple) and numerically integrated free energy (green) as functions of the reaction coordinate of the copper dissolution. The reaction coordinate represents an approximate distance along with the copper atom gets away (in other words, dissolves) from the metal surface.

Fig. 5: AIMD simulation with constant-potential thermodynamic integration for Cu dissolution at different potentials.
figure 5

Free energy gradient (purple) and free energy (green) as functions of Cu distance from the surface along the coordinate of the dissolution reaction. The gradient shows an overall monotonically increasing trend at −0.35 VRHE (a) while it shows negative local minima at −1.31 VRHE (b). Inset schematic in (a) shows that copper dissolution is inhibited at the less negative voltage of −0.35 VRHE due to the increasing gradient and high free energy barrier. Inset in (b) shows the atomic structures (with Cu distance from the surface of 0 Å, 1.5 Å, and 2.5 Å) of the copper dissolution reaction that happens at the more negative voltage of −1.31 VRHE.

Full size image

At −0.35 VRHE (Fig. 5a), the free energy gradient shows a monotonic increase up to a distance 1.5 Å far from the basal plane of the copper surface reaching a gradient ~1.3 eV/Å, and the gradients stay over 1 eV/Å at 2.0 Å. This trend leads to an increase in free energy, which reaches ~1.2 eV (115.8 kJ/mol) at 2.0 Å. The increasing gradient and high energy indicate that copper dissolution is energetically unfavored and inhibited. Notably, the structure at 2.5 Å from prior slow-growth AIMD (shown in the inset of Fig. 5a on the right) becomes unstable and cannot be converged by the constrained constant-potential AIMD. This is an indication that the copper atom leaving the surface is unstable at positions far away from the surface (deep into the solvent) and at more positive potentials, e.g., −0.35 VRHE. On the other hand, at more negative potentials −1.31 VRHE (Fig. 5b), the free energy gradient does not show an increasing trend, and multiple local minima below 0 eV/Å are observed. The trend results in a rather flat free energy curve around 0 eV and indicates energetically favored copper dissolution. The inset in Fig. 5b shows atomic structures along the dissolution trajectory (at 0 Å, 1.5 Å, and 2.5 Å Cu distance from the surface) generated by slow-growth AIMD. At 0 Å, the potassium cation appears partially desolvated to three water molecules and adsorbed at the copper surface. The potassium and copper species remain coordinated during the dissolution process. It is observed that copper species forms hydride with two protons (Cu-H bond lengths of ~1.64 and ~1.57 Å) when it dissolves at a distance of ~2.5 Å from the surface (right-most inset structure in Fig. 5b).

These modeling results show that the potential is indeed an important factor for the dissolution of copper. From these results, we can infer a plausible mechanism41 of the cathodic corrosion of Cu: at an electrode potential more negative than the onset potential (i.e., about −0.4 VRHE at our conditions) and with the presence of alkali cations in the solution, Cu forms soluble intermediate ternary hydride species. They may react quickly with adsorbed hydrogen species or water from the solution and get deposited back onto the electrode, leading to the observed reconstruction of the Cu catalyst and the formation of smaller Cu particles, and the reconstruction of Cu catalysts.

The impact of cathodic corrosion of Cu on its catalytic performance

Previous studies have highlighted the critical roles alkali cations played in CO2RR, as they can stabilize key reaction intermediates43,67 and suppress the hydrogen evolution reaction (HER)42,45,68. In addition, Cu catalysts are commonly operated at about −1 VRHE or more cathodic conditions to enhance selectivity towards multicarbon products4. Therefore, if the alkali cation-induced cathodic corrosion is in play, Cu catalysts will inevitably undergo reconstructions under a typical CO2RR condition and the selectivity advantages brought by morphology control in the pre-catalyst stage will unlikely to be stable. This is exactly what we found. As shown in Figs. S5, 6a, and Fig. S25, although Cu NCs initially offer high selectivity towards C2H4, the catalyst morphology and catalytic performance evolve over time, producing a lot more CH4 after 10 h (Table S1, entries 1). We then conducted another 10-h CO2RR experiment, this time employing 25 nm spherical Cu nanoparticles (Table S1, entries 31). These spherical Cu nanoparticles generate comparable amounts of C2 products (mainly C2H4) towards the end of the 10 h reaction, mirroring the performance of aged Cu nanocubes.

Fig. 6: The catalytic performance of Cu nanocubes and 25 nm spherical Cu nanoparticles with or without alkali cation-induced cathodic potential in play.
figure 6

The comparison of Cu nanocubes and 25 nm spherical Cu nanoparticles when used for 10-h tests of a CO2RR at −1.1 VRHE (iR-corrected) in 0.1 M KHCO3 (Table S1, entries 1 and 31) and b CORR test at −0.37 VRHE (iR-corrected) in 1 M KOH (Table S1, entries 36 and 40). Liquid product distributions of the CORR experiments as a function of time for c Cu nanocubes and d 25 nm spherical Cu nanoparticles. The error bars show the standard error of three independent measurements. Insets show ex-situ TEM bright field images of the corresponding sample in each case before and after the CORR experiments.

Full size image

Given that controlling catalyst morphology is a major method to adjust the intrinsic activities of electrocatalysts5,6,69,S34). A PTFE electrochemical cell with a three-electrode configuration was employed, using a carbon rod through a Luggin capillary as the counter electrode to isolate the oxygen gas bubbles. For cathodization in acid, Hg/HgSO4 was used as a reference electrode. Hg/HgO was used as a reference electrode in alkaline electrolytes. Before cathodization, electrodes were connected to a potentiostat, and potentiostatic was run so that protective cathodic potential was applied to the copper surface immediately when it contacts with the electrolyte to avoid significant oxidation/dissolution of the copper surface at OCP. After cathodic treatment, copper surfaces were transferred and electrochemically characterized at room temperature by cyclic voltammetry in argon-saturated 0.1 M NaOH solutions at a sweep rate of 50 mVs−1. Transfer process takes several seconds from the cathodization cell to the characterization cell, during which the copper surface was mostly kept within the cell flowed with Ar to decrease the air exposure. For voltammetric, a PTFE electrochemical cell with a three-electrode configuration was employed, using a carbon rod as the counter electrode and an Hg/HgO electrode as the reference electrode. The voltammetries were normalized to yield identical double-layer (DL) thicknesses within a potential region where the voltammetries exhibited capacitive behavior (+0.2 V < E < +0.28 V, +0.02 V < E < +0.08 V)81. All electrolytes were pre-electrolyzed for 3–5 h on the clean carbon paper to remove electrolyte impurities. All electrolytes were bubbled with Argon for 1 h before measurement and kept being bubbled during measurement. In this work, all potentials are referred to RHE. The electrode potential was controlled using a CHI760E potentiostat.

Identical location transmission electron microscopy (IL-TEM)

Identical location TEM was achieved by positioning copper particles on an electrically conductive and corrosion-resistant Au grid that can be used both as a working electrode in an electrochemical cell and as a sample holder during TEM (Fig. 1b). A custom-made PTFE cap with a hole in the center was used to fix the TEM grid onto the surface of glassy carbon on rotation disk electrode. Both sides of the gold grid were covered by carbon deposition to block gold surface from electrochemical reaction and copper particles supported on the carbon film were observed under TEM. By taking overview images at progressively increasing magnification and using tracking markers, locations of interest in the untreated particles are selected and recorded. This step was repeated after an ex-situ electrochemical test, allowing the evaluation of the identical particles. After the electrochemical test, the grid was taken out and dried on a blotting paper for seconds before it was put into a glovebox and wait for the characterization. The electrochemical reaction was performed in the electrochemical cell with a three-electrode configuration, with a carbon rod through a Luggin capillary as the counter electrode and Hg/HgSO4 for the acidic electrolyte, and Hg/HgO for the alkaline electrolyte as reference electrode. The electrolyte was bubbled with Argon or CO2 for 1 h before measurement. Before immersion of the working electrode into the electrolyte, electrodes were connected to potentiostat, and potentiostatic was run so that protective cathodic potential was applied to the copper surface immediately when it contacts with the electrolyte to avoid further oxidation/dissolution of copper surface in the electrolyte at OCP. After electrolysis for the targeted time, the copper surface was extracted from the electrolyte with holding potential. For Ex-situ TEM imaging, a TEM grid was pressed onto the glassy carbon with a drop of hexane atop to collect the reacted nanoparticles after electrolysis.

Electron microscopy characterization

Scanning Transmission Electron Microscopy High angle annular dark field (HAADF) imaging, X-ray Energy Dispersive Spectroscopy (X-EDS), and Electron Energy Loss Spectroscopy (EELS) studies were conducted on a JEOL JEM ARM 200CF in the National University of Singapore, equipped with an Oxford Instrument X-ray Energy Dispersive Spectrometer and a Gatan Image Filter (GIF) system. In a typical STEM-EELS measurement, a probe current of around 100pA was utilized, with a dwell time of approximately 20 ms per pixel. The convergence half angle is about 29 mrad and the collection half angle is about 36 mrad for the STEM-EELS measurement. To generate the elemental map, background removal, signal integration, and multiple linear least square fitting were performed using Gatan Digitalmicrograph software.

X-ray absorption fine structure spectroscopy

XAFS spectra around the Cu K-edge (8979 eV) were collected at the XAFCA beamline of Singapore Synchrotron Light Source (SSLS) in transmission mode82. The double-crystal Si (111) monochromator was used for measurements. The size of the beam at the sample position was ca. 2 mm (h) × 1 mm. All XAFS data analyses were performed with the Athena software package to extract XANES83. The electrochemical setup with three electrodes and a Polyetheretherketone electrochemical cell was used for the reaction. The electrochemical setup consisted of an Ag/AgCl reference electrode, a carbon counter electrode, and a catalyst-modified carbon cloth (ca. 10 mm × 10 mm geometric area) working electrode. The catalyst ink was drop** cast onto carbon cloth. The electrolyte was first bubbled with CO2 gas for 30 min and then a constant potential was applied. The XAFS measurements were recorded simultaneously while performing the chronoamperometric measurements.

CO2RR measurements

The experiments were implemented at an electrochemical workstation (CHI760E) under ambient pressure and temperature. A gas-tight two-compartment cell configuration was employed with a counter electrode of Pt mesh and reference electrode of Ag/AgCl electrode and electrolyte of 0.1 M KHCO3. The electrolyte was prepared by dissolving KHCO3 in DI water and stored in a plastic volumetric flask. A piece of anion-exchange membrane with 2.5*2.5 cm2 area and 70um thickness was served to separate the counter and working electrodes in the two compartments. Working electrodes were prepared by evaporating hexane solution containing around 50ug Cu NCs within a squared area of 1cm2 on the glassy carbon substrates. The loading was calculated by multiplying the Cu concentration of the ink with the volume of the ink dropped onto the substrate. Here, the loading amounts were chosen to yield a decent Faradaic efficiency towards multicarbon products. The working electrodes were then held at a constant bias of −1.1 V vs. RHE using chronoamperometry for a set time of up to 10 h. During electrolysis, CO2 was constantly bubbled through the electrolyte at a flow rate of 20 sccm. The flow rate of CO2 was controlled with a mass flow controller. All potentials were corrected for iR-loss compensation by electrochemical impedance spectroscopy (EIS) and converted to potentials versus the reversible hydrogen electrode (RHE) using the following equation (1): \(E\left({vs}.{RHE}\right)=E\left({vs}.{Ag}/{AgCl}\right)-{iR}+ 0.197V+ 0.0591V\times {pH}\). The gaseous products were monitored by an online gas chromatograph, equipped with a thermal conductivity detector (TCD) detector for H2 and a flame ionization detector (FID) detector for hydrocarbon quantification. High-purity He gas (99.999%) was used as the carrier gas for all compartments of the GC. A GC run repeats every 29.5 minutes. The GC was calibrated using calibration curves of analytical gas standards. The calculation of Faradaic efficiency (FE) for gaseous products based on the following equation (2): \({{FE}}_{i}\left(\%\right)={j}_{i}/{j}_{{total}}\times 100={zF}\times \left(Q{P}_{0}{V}_{i}/{RT}\right)/{j}_{{total}}\). Here, the total current was measured by the potentiostat, and the partial currents were obtained from the areas of GC chromatogram peaks where Vi is the volume concentration of gas product i based on the calibration of GC. Q is the flow rate of CO2, z is the number of electrons transferred for reduction to product i, F is the Faradaic constant, P0 is atmosphere pressure, T is room temperature, and R is the ideal gas constant, 8.314 J·mol/K. 1 ml electrolyte was collected from both cathode and anode chamber at 1 h and 10 h of reaction and the containing liquid products were analyzed by 1H Nuclear Magnetic Resonance (NMR) spectrum (Bruker 400 MHZ system). The Faradic Efficiency (FE) of each product was calculated by following equation (3): \({{{{{\rm{FE}}}}}}\left(\%\right)={amount\; of\; the\; product}\times n\times F/C\times 100\), where n represents the number of electrons transferred, F represents the Faradaic constant, and C represents the total Coulomb number.

CORR measurements

The measurements were conducted in a flow cell, which was assembled from the sequential stacking of a gas chamber, a catalyst-loaded GDE cathode, a catholyte chamber (where an Ag/AgCl reference electrode was located), an anion-exchange membrane, an anolyte chamber, and an IrO2-loaded Ti plate anode. The working electrodes of Cu cube and Cu sphere were prepared as follows: 1 mg Cu cube/Cu sphere power was dispersed in a mixed solution with 1 mL of ethanol and 4 μL of 5 wt% Nafion solution to form a catalyst ink. The ink was dispersed in an ultrasonic machine for more than 10 minutes. And then the ink was sprayed onto carbon fiber (YSL-30T) paper with 1 mg cm−2 to form a smooth catalyst film.

During COR measurements, 21 sccm of CO was continuously fed to the working electrode through the mass flow controller. And the electrolyte (1 M KOH) was pumped to circulate through the flow cell at the rate of 15 mL min−1 by a peristaltic pump. The electrolyte was prepared by dissolving KOH in DI water and stored in a plastic bottle. The Bio-Logic VMP3 multichannel potentiostat/galvanostat was applied for the electrochemical measurements. The reference electrode was Ag/AgCl and the potential was converted to RHE with necessary iR compensation by equation (1). The ohmic-drop correction of the potentials applied was carried out manually using the resistance measured by the electrochemical impedance spectroscopy under open circuit potentials.

Gaseous reduction products were injected into an online gas chromatograph (GC, Shimadzu 2014) to analyze the concentration of each gas product. And liquid products were collected in the centrifuge tube and were identified by 1H Nuclear Magnetic Resonance (NMR) spectrum (Bruker 400 MHZ system). 0.6 ml electrolyte was taken every 1 h during the 10 h CORR from the initial 15 ml circulated catholyte. The Faradic Efficiency (FE) of each product was calculated by equation (3).

Computational investigation

All calculations were performed using Kohn-Sham DFT as implemented in VASP84. The unknown exchange and correlation functional was approximated by the revised Perdew−Burke−Erznzerhof (RPBE)85. Van der Waals (vdW) interactions were treated with the D3 method with zero dam**86. The wavefunctions of the valence electrons were expanded as plane waves up to a cut-off of 400 eV87,88. Projector-augmented wave (PAW) potentials were used for core electrons89. A smearing width of 0.3 eV together with a second-order Methfessel−Paxton scheme90.

To model the dissolution reaction, a three-layer copper slab containing 24 atoms modeling the Cu(001)/water interface was put in direct contact with 26 explicit water molecules followed by a region where the effect of water is described by an implicit electrostatic model. The 26 water molecules fill a 15 Å-long region to maintain the water density ~1 g/cm3. A K ion was initially added 2.14 Å from the Cu(001) surface and then equilibrated. The Cu surface was pre-relaxed with DFT and 26 water molecules were pre-organized with classic molecular dynamics using LAMMPs. The implicit water was applied using VASPsol91,92.

Ab initio molecular dynamics (AIMD) simulations were performed at the Γ point with a time step of 1.0 fs and hydrogen mass of 3 amu. Simulation of a negative voltage to match the experimental electrode potential is realized by adding extra electrons into the cell and determined from the electrostatic potential of implicit water and the system’s Fermi level. Two different voltages were simulated, namely −1.31 VRHE by adding 3 electrons and −0.35 VRHE by adding 0.2 electrons. Temperature was ramped to 300 K during the first 3 ps using the Nose−Hoover thermostat93,94, and then the system was equilibrated at 300 K for 5 ps. The slow-growth method95 was employed to generate Cu extraction trajectory by controlling the distance between the Cu atom and bottom Cu layer, between the K atom and bottom Cu layer, and between the K and Cu atom. A trajectory of 2.5 Å in a period of 5 ps was generated for each system of different voltages. Five windows were taken from the trajectory and their free energy gradient was calculated using constrained constant-Fermi-level Blue-Moon AIMD41,62,63,64,65,66. The reaction coordinate - the Cu distance from the surface - was fixed for each window, and the Fermi level was kept within ±0.05 eV of the target value from the 5 ps equilibration by adding/removing electrons when it gets out of range.