Introduction

Of versatility, energy efficiency, and cost effectiveness, electrochemical CO2 reduction (eCO2RR) has been deemed a potential strategy to mitigate excessive anthropogenic CO2 emissions1,2,3,4,5,6,7,8. From a variety of catalysts for eCO2RR, those copper-based have drawn much attention1,6, since Cu is the one which can electrochemically reduce CO2 molecules to economically favorable and energy-dense C2 products, such as C2H4 and C2H5OH5,9. However, the typically low product selectivity hinders the catalysts from industry applications. There, accordingly, have been numerous studies on improving the selectivity by means of nanostructuring10,11, bimetallic systems12,13,14, and oxidative modifications15,16,17,18. Among them oxidative modifications have received considerable attention, since they could significantly improve the selectivity of Cu-based catalysts toward valuable C2 products.

Nevertheless, despite showing improvements, these oxidative modifications still cannot meet the decent selectivity for industry applications, 90% as Jouny et al. pointed out5. It has been noticed that these studies, whether knowingly or otherwise, suggested that Cu(I) may play a critical role in improving the selectivity toward C2 products, but the chemical state of Cu in the catalysts were varied drastically during electrochemical reduction by using a chronoamperometry15,18. This ineffective control of the chemical compositions and/or chemical states leads to the resulting low selectivity toward C2 products, and a fair speculation could be that a stable chemical composition of Cu species during electrolysis is indispensable to enhance the related C2 product selectivity. For instance, Nogami et al. showed that an oxidation process stimulated by switching applied potentials in corresponding time intervals could bring an improvement in the product selectivity in their eCO2RR system19. Recent studies also agreed with the integrated oxidation process for improving CO2RR product selectivity20,21,22,23. For example, the employment of pulses voltage has been revealed to enhance the selectivity toward the C2 products on polycrystalline Cu21,23,24. Recently, we have justified the existence of another competing reaction (i.e., the spontaneous oxidation of Cu(0) in aqueous electrolyte) that significantly governs the chemical state of active centers of Cu, establishing a strong correlation between the chemical state under reaction conditions and the CO2RR product profile25. A similar effect has been also revealed in the case of Fe single atom catalyst, in which the reduction activity toward CO is evidently dominated by the chemical state of metal centers26.

However, some studies brought a different perspective that the active catalyst is metallic copper because no significant concentration of residual oxide was detected in the oxide-derived samples; ambient pressure X-ray photoelectron spectroscopy (XPS) and electron energy loss spectroscopy (EELS) proved the absence of residual copper oxide in the reduced electrocatalyst27,28. Regarding this controversial issues, we have to point out a paramount subject concerning a significant mismatch in time-scale of acquiring spectrum. For conventional X-ray absorption spectroscopy (XAS), the spectroscopic acquiring time is approximately several 10 min (normally 20–30 min). In the cases of XAS and EELS, both of them would require a time-scale in minutes at least for spectrum acquisition. Notably, the chemical state of metal centers could reach a steady state in a few minutes under the eCO2RR; as revealed by in situ Raman, the CuOx precursor would achieve a steady-state condition due to the co-occurrence of electrochemical reduction and of spontaneous oxidation caused by the trace oxidant in the electrolyte25. Consequently, a time-resolved identification of the characteristics in chemical state of metal centers should be highly desirable, especially for a time-scale of few seconds for revealing the real dominated factor in CO2 reduction electrocatalysis.

Thus far, there have been various types of time-resolved XAS developed. For example, the fixed energy X-ray absorption voltammetry (FEXRAV) has demonstrated its capability to provide valency information of a material under time-dependent electrochemical working conditions29,30,31. Another technique in the time-resolved XAS family is energy-dispersive XAS (EDXAS). The parallel acquisition of all data points in a XAS spectrum using EDXAS allows its time resolution down to the level of ms and, if possible, to ps32. While the FEXRAV and the EDXAS are of time resolution, certain drawbacks make both of them possibly not suitable for the present study. In the case of the FEXRAV, it fails to offer the information regarding chemical surroundings, which has been reported to influence the selectivity of a catalyst33,34 and normally can be derived using EXAFS spectroscopy. These technical limitations can be well removed by using quick-XAS. The basic setup for quick-XAS is shown in Fig. 1a. Via a quick-scanning monochromator, an acquisition time of a single XAS spectrum, including X-ray absorption near edge structure (XANES) and extended X-ray absorption structure (EXAFS), can be cut down to a few seconds, reaching seconds-resolution. Moreover, its sequential detection mode also renders the operando detection achievable. Overall, the advantage of data collection and interpretation makes quick-XAS (TR-XAS hereafter) the time-resolved technique of choice in present study.

Fig. 1: Experimental setup and catalyst characterization.
figure 1

a The schematic diagram of the setup of operando time-resolved XAS experiments. b The cartoon representation of customized operando XAS cell. c TEM image of Cu nanocubes. d TEM image of CuOx. Scale bars are 50 nm. e HR-TEM image and SAED pattern (inset) of CuOx. f First derivatives of the Cu K-edge XANES spectra of CuOx and of commercial Cu, Cu2O, and CuO as references. g, h Wavelet transform-EXAFS of as-prepared CuOx and the references.

Full size image

Herein, utilizing TR-XAS with a small X-ray incident angle, we develop an operando methodology to realize the seconds-resolved near-surface investigations of materials under CO2RR working conditions (Fig. 1a, b). This allows us to in situ characterize the chemical nature of catalytic surfaces. An oxide-derived Cu electrocatalyst with great potentials to produce hydrocarbons is employed as a model to provide scientific insights to the roles of metal states in CO2RR. We can conclude that chemical state of copper ions can reach a steady-state in ~5 min toward the metallic Cu(0) under a conventional chronoamperometry. Most interestingly, by utilizing a potential switching approach, the Cu(I)/Cu(0) redox could achieve a steady state of half-and-half during a cathodic CO2RR electrolysis, producing an asymmetric C2 product – C2H5OH – with fairly high selectivity in a wide potential range. Furthermore, a theoretical computation revealed that a surface composed of Cu–Cu(I) ensembles could have dual CO molecules coupled asymmetrically, which potentially enhanced the catalyst’s CO2RR product selectivity toward C2 products35. Operando TR-XAS gives conclusive evidence that chemical state of copper significantly dominates the catalytic behavior and the product profile of CO2RR. Our results can offer understandings of the fundamental chemical state and insights into the establishment of selective eCO2RR.

Results

Catalyst synthesis and characterization

A two-stage wet chemical synthesis was employed to realize our model catalyst. Uniform Cu nanocubes were first synthesized with an average size of 38.0 ± 4.2 nm (Fig. 1c, and Supplementary Figs. 1 and 2). To investigate the individual role of Cu(0) and Cu(1+), 1-hexadecylamine (HDA) was utilized to mediate the oxidation rate of targeted metallic nanomaterials and to gently oxidize the as-prepared Cu nanocubes36, which protected the morphologies of the as-prepared nanomaterials from being crushed and further suppressed the effects caused by morphologies of materials37. TEM image of the oxidation modified Cu nanocubes (CuOx hereafter) clearly demonstrated that the morphology of CuOx stayed a cube-like after the oxidation modification (Fig. 1d). Regarding the chemical composition of the as-prepared CuOx, selected area electron diffraction (SAED) was characteristic of two domains, metallic Cu and oxide Cu2O species, in the CuOx (Fig. 1e). These observations confirm the formation of our desired material made of Cu–Cu(I) ensemble. Electrochemical analyses also agreed with the electron microscopic analysis. Shown in Supplementary Fig. 3a, the LSV indicated at −0.45 V (vs RHE), the system had a strong cathodic response, suggesting the existence of oxide species in CuOx. To identify species of such oxides, as depicted in Supplementary Fig. 3b, the CV scanning in a cathodic direction showed that in the first cycle (green line), a noticeable reduction peak which was preceded by a slight oxidation peak could be corresponding to a reduction from Cu+ to Cu, suggesting the oxidized copper atoms in our catalyst were nearly cuprous38. In the second cycle of the CV measurement (indicated in red), the original slight oxidation peak grew clearer as shown in the inset of Supplementary Fig. 3b. This growth supported the idea that our catalysts contained Cu+ ions, since the following reduction phenomenon also occurred and represented the reduction of Cu+ to Cu.

To further confirm the chemical composition, X-ray absorption spectroscopy (XAS) of Cu K-edge for the CuOx was conducted (Fig. 1f and Supplementary Fig. 4). As depicted in Fig. 1f, the normalized 1st derivative of X-ray absorption near edge structure (XANES) of CuOx and references suggested that no detectable Cu(II) species existed in the as-prepared CuOx39. On the other hand, a strong peak overlapping that of Cu2O at ~8980 eV indicated the presence of Cu(I) species in the CuOx catalyst. Notably, a slight difference between the normalized absorbance at 8981 eV of CuOx and that of Cu2O might be attributed to a couple of reasons, such as a dilution effect from metallic Cu species and the catalysts’ nanoscale25. Wavelet transform was employed to clarify the coordination environment near by the Cu atom and demonstrate the atomic dispersion of Cu in the CuOx sample. As shown in Fig. 1g, a strong wavelet transform signal focused at 7.4 Å−1 derived from Cu–O contribution can be observed, while another stronger signal focused at 8.8 Å−1 can be attributed to a contribution of Cu–Cu. The Fourier-transformed k3-weighted spectra of CuOx and corresponding references reveal the characteristic peaks at 2.3 Å, which correspond to the Cu–Cu scattering, implying the existence of metallic Cu (Fig. 1h). Note that a significant Fourier-transformed peak at 1.6 Å can be referred to the scattering path of Cu–O, which implied the presence of the Cu oxide with a similar nature to that of Cu2O. The wavelet transform contour plot of CuOx also further clarified the neighboring environment around Cu, which is consistent with all observations above and indicated the fact that the copper species in the as-prepared CuOx were nearly of Cu and Cu(I).

CO2RR performance evaluation

To realize the steady chemical state of the as-prepared CuOx and clarify the real dominated factor in CO2 reduction electrocatalysis, we adopted the method of switching potential (redox shuttle; R.S. hereafter). Specifically, we applied a constant anodic potential (0.5 V), which was in light of the above discussion regarding the oxidation of CuOx, and various cathodic potentials (from −0.7 to −1.15 V). The time interval of each anodic and cathodic potential cycle was set for 10 s (see Supplementary Fig. 6 for the potential map of R.S.). As stated above, the technique provides the electrocatalytic materials with a cycle of oxidation and reduction, thereby having the potential to stabilize the chemical nature of materials. Figure 2a exhibited the overall CO2RR performance of the as-prepared CuOx catalyst. Surprisingly, until −0.9 V, the C2H5OH molecule was the only identified CO2RR product in our system while the rest product was hydrogen gas generating from proton reduction, and the corresponding F.E. (Faradaic Efficiency) reach a maximum value of ~12.9 % @ −0.75 V. With increasing the cathodic potential through the redox shuttle, the F.E. of ethanol gently declined. Note that the ethanol was still the only product from CO2RR. In contrast to the case of redox shuttle, the CO2 reduction product through a conventional chronoamperometry (CA hereafter) at the same potential of −0.75 V was carbon monoxide (Supplementary Fig. 5), which was characteristic of a copper-like behavior with the only product of CO from CO2RR at such potential1,37. The CuOx catalyst in present study gave a copper-like CO2RR performance once applying a higher cathodic potentials (higher than −0.95 V) where the dominant CO2RR product was CH4. We could say that even though the increasing potential would bring a decline in F.E. toward ethanol, the single product from CO2RR still could be realized without presence of other common CO2RR products, such as C2H4, CO, and CH4. This finding further validates the fact that the redox shuttle can achieve the desired single selectivity of CO2RR. Note that we kept anodic potential at 0.5 V (vs RHE), referred to the oxidation potential of as-prepared CuOx (as revealed in Supplementary Fig. 3), and varied the cathodic potentials from −0.7 to −1.15 V (vs RHE). Anodic and cathodic potentials were switched in an interval of 10 s (Supplementary Fig. 6). Apparently, it seems that a large cathodic potential was able to result in significant increasing the current density of CO2RR. Nevertheless, the large cathodic potential was failed to achieve a highly selective CO2RR production since that was going to cause a remarkably generation of metallic Cu0 (vide infra). As indicated by the CV results (Supplementary Fig. 7 and Supplementary Note 1), the oxidation variations of Cu species between applied cathodic potentials were relatively subtle through using R.S. and therefore highly suggested the technique enabled the composition of Cu species in the CuOx to be well preserved. Interestingly, it is well known that the formation of C2H4 and C2H5OH may share the same path way of CO2RR through a hydroxyl containing intermediates40,41,42. As a result, more detail analyses in following sections are essential to address why C2H5OH is the favorable product rather than C2H4 in our system.

Fig. 2: Operando time-resolved XAS characterization.
figure 2

a Potential dependence of CO2RR product selectivity on the CuOx in 0.5 M CO2-saturated KHCO3 under the employment of redox shuttle. The error bars represented standard deviations based on three individual measurements. b Time-resolved variations using Redox Shuttle (R.S.) of Cu species in CuOx and the corresponding electrochemical responses during CO2RR at −0.75 V and c the corresponding time-resolved EXAFS spectra without phase-correction of CuOx under CO2RR. d Time-resolved variations using chronoamperometry (CA) of Cu species in CuOx and the corresponding electrochemical responses during CO2RR at −0.75 V and e the corresponding time-resolved EXAFS spectra without phase-correction of CuOx under CO2RR. The spectra were depicted in a 2D contour plot with interatomic distance (Å) on x-axis and with time (sec) on y-axis, while normalized Fourier transform magnitude served as corresponding intensity. Quantification of time-resolved chemical composition information extracted from operando Quick Cu XAS for CuOx under eCO2RR using redox shuttle (f) and conventional chronoamperometry (g). Each error bar indicated an estimated standard deviation based on the curve fitting of EXAFS spectra. Both cathodic potentials were −0.75 V and electrolyte was 0.5 M CO2-saturated KHCO3.

Full size image

Operando XAS characterization

To assess the chemical composition of the CuOx under CO2RR, operando XAS was performed, due to its capability to probe the chemical composition of the catalysts under realistic reaction conditions26,43,44,45. As we mentioned above, conventional XAS only offers the information which is referred to a steady condition because of time requirement in spectrum acquirement (typically several tens mins). Thus, it inaccessibly provides a convincing time-resolved information for an electrochemical system where the reactions are typically being steady on a timescale of seconds or few minutes46. With the aim of addressing this timescale mismatch, the operando time-resolved X-ray absorption spectroscopy (operando TR-XAS) with a sub-second time-resolution can be reached by equipping a continuously oscillating crystal on a monochromator47. Operando time-resolved-XANES spectra for CuOx during CO2RR at −0.75 V (vs RHE) using R.S. and those of using CA were displayed in Supplementary Figs. 8 and 9, respectively. A linear model developed by Vitova et al.48 was utilized to extract the pristine and time-resolved compositional information from acquired XANES spectra. As shown in Supplementary Fig. 10, the result suggested that the composition of the pristine CuOx was of equal Cu(I) and Cu. This not only supported again our previous characterization that Cu(I) and Cu coexisted in the CuOx (Fig. 1c–h), but also proved the feasibility of the linear model to quantitatively analyze the composition of Cu species, as Vitova et al. suggested. In this regard, we used the linear model to extract the chemical nature information of CuOx from the operando time-resolved-XANES spectra. Extracted time-resolved chemical composition information combined with the electrochemical responses was shown in Fig. 2 (2b for R.S. and 2d for CA) and in Supplementary Fig. 11 (the magnified portion of Fig. 2b by which one can track the footprint of XAS acquisition under the R.S. working condition). During the CO2RR, R.S. restricted the chemical composition of CuOx in around a composition of half Cu and half Cu(I), while the CA transformed the chemical composition of CuOx into that Cu-dominated. Following the 1-hour reaction time, XANES spectrum of the CuOx after the CO2RR using R.S. remained nearly unchanged, further indicative of R.S. having the ability to preserve the chemical composition. On the other hand, that of the CuOx after the electrolysis using CA clearly showed an electrochemical reduction toward metallic Cu (Supplementary Fig. 10).

As shown in Fig. 2c, e, operando time-resolved EXAFS verified that the intensity of bond pairs regarding Cu–O at 1.6 Å and Cu–Cu at 2.3 Å stayed highly stable under the employment of R.S., whereas the intensity under the case of using CA showed a decline at 1.6 Å and an increase at 2.3 Å that could be corresponded to a transformation from Cu2O to metallic Cu0. To further quantify such chemical composition information from operando EXAFS, we obtained the Cu–Cu path from both Cu2O (Cu–Cu(Cu2O)) and metallic Cu (Cu–Cu(Cu)) extracted by a standard fitting procedure (detailed results shown in Tables 1–2 and Supplementary Figs. 12–31). We note here that in order to have quality spectra for EXAFAS fitting, the related XAS acquisition time was extended from 5 to 60 s. While the time resolution could not match that of seconds (Fig. 2b–e), the minutes-resolved EXAFS fitting results were also in line with those seconds-resolved (see below). The quantification was displayed in terms of coordination numbers and, more straightforward, in terms of the percentage of both Cu(0) and Cu(I) according to Eqs. (1) and (2):

$$\frac{{CN_{({\mathrm{experimental}}\,{\mathrm{Cu}} - {\mathrm{Cu}}@2.5{\AA})}}}{{CN_{({\mathrm{theorectical}}\,{\mathrm{metallic}}\,{\mathrm{Cu}})}}} \times 100\% = Cu(0){\mathrm{composition}}(\% )$$
$$\frac{{CN_{({\mathrm{experimental}}\,{\mathrm{Cu}} - {\mathrm{Cu}}@3.0{\AA})}}}{{CN_{({\mathrm{theorectical}}\,{\mathrm{Cu}}2{\mathrm{o}})}}} \times 100\% = Cu(I){\mathrm{composition}}(\% )$$

where CN(experimental Cu-Cu@2.5 Å) and CN(experimental Cu-Cu@3.0Å) represent the coordination numbers (CNs) of the Cu–Cu path at 2.5 Å and 3.0 Å, respectively. CN(theoretical metallic Cu) and CN(theoretical Cu2O) are equal to those of metallic Cu and Cu2O, namely 12. As depicted in Fig. 2f, g, time-resolved fitting results fairly agreed with both analyses of XANES and EXAFS. Using R.S., the CNs of Cu(I) and Cu(0) in the as-prepared CuOx stayed relatively steady during CO2RR (Fig. 2f). By contrast, under the employment of CA, the CNs of Cu(I) and Cu(0) in the CuOx gradually reduced and increased, respectively, suggesting the electrochemical reduction of the CuOx (Fig. 2g). Consequently, the results of time-resolved operando XAS proved that under CO2RR, the chemical composition of CuOx could remain in a steady state of half-and-half Cu and Cu(I) through using R.S.

Table 1 Operando EXAFS fitting parameter for the CuOx at −0.75 V under CO2RR using redox shuttle approach.
Full size table
Table 2 Operando EXAFS fitting parameters for the CuOx at −0.75 V under CO2RR using conventional chronoamperometry.
Full size table

By contrast, under the employment of CA, the chemical composition of CuOx continued a downward trend toward Cu(0). A similar result as illustrated in Fig. 3 was also revealed by the conventional XAS, showing a significant reduction of Cu+ to generate metallic Cu0. XANES data elucidated as a cathodic potential was applied from −0.05 to −1.25 V (Fig. 3a, b). The XANES features as illustrated in Fig. 3b were denoted as three distinct peak-like structures, while no remarkable pre-edge feature that caused by a dipole transition from 1s to the mixing states of 4p and 3d could be visible and suggested a centro-symmetry nature of copper37,49. An additional small feature of the excitation in the 3d10Lhole state at a lower energy region (as indicated by green arrow) gradually intensifies with increasing cathodic voltage, this indicates the formation of metallic Cu (Fig. 3b). Furthermore, the shifts in the edge position could conclude the chemical states in CuOx were reduced into copper-like and the content of metallic Cu(0) species in the CuOx evidently increased. Consistent with XANES spectra, the analysis of extended X-ray absorption fine structure (EXAFS) revealed that as an applied cathodic potential increased, the metallic Cu–Cu bond pair at 2.3 Å, matching a standard metallic Cu, intensified significantly (Fig. 3c). Worth noting is that like cases in literature39, the as-prepared CuOx exhibited a resistance to electrochemical reduction, as the intensity of a Cu–O bond pair at ~1.6 Å, referring to a Cu2O standard, did not entirely disappear even under a cathodic environment. Despite having this resistance to the electrochemical reduction, however, the as-prepared CuOx could not remain its initial chemical nature under CO2RR when the CA was adopted. This observation can refer to the fact that the catalytic sites and the chemical states can fail to reach a steady situation under the conventional chronoamperometry. These results also can explicate the finding from numerous reports that the products profile of CO2RR are commonly characteristic of a mixture nature.

Fig. 3: Conventional operando XAS characterization.
figure 3

a Potential dependence of operando Cu K edge XANES spectra without phase-correction of the CuOx under CO2RR in 0.5 M CO2-saturated KHCO3 using chronoamperometry (CA) and the corresponding 1st derivative spectra (b). c Potential dependence of operando EXAFS spectra of the CuOx under eCO2RR in 0.5 M CO2-saturated KHCO3 using CA. d Potential dependence of operando Cu K edge XANES spectra without phase-correction of the CuOx under CO2RR in 0.5 M CO2-saturated KHCO3 using R.S. and the corresponding 1st derivative spectra (e). f Potential dependence of operando EXAFS spectra of the CuOx under eCO2RR in 0.5 M CO2-saturated KHCO3 using R.S.

Full size image

To extend the investigation further, we also performed conventional operando XAS analyses for the CuOx under CO2RR using R.S. in various potentials (Fig. 3d–f). Only slight variations on XAS results across the various applied potentials, the chemical composition of CuOx could remain in a steady condition of the composition of Cu and Cu(I) through using R.S. as illustrated in Fig. 3d, e. Note that, as compared with those of Fig. 3b, no additional feature of the excitation in the 3d10Lhole state at a lower energy region can be found with increasing cathodic voltage, this further indicates the chemical nature of CuOx can reach steady-state under CO2RR using R.S. (Fig. 3e). In the case of operando EXAFS spectra, the final condition from various applied potentials were similar except for the situations in high cathodic potentials (<−0.95 V vs RHE). This finding corroborated the observation of potential-dependent product profiles in Fig. 2a, in which the CO2RR products have become a metallic copper-like profile with a mixed feature. Therefore, these results became solid evidence to conclude the existence of the steady composition of Cu species under CO2RR using R.S. Aside from spectroscopic analysis, the post electron microscopy analysis also supported the validation of R.S. As shown in Supplementary Fig. 32, the TEM image of CuOx after a 6-h CO2RR electrolysis at −0.75 V using R.S. demonstrated that the morphology of the CuOx was similar to those before the electrolysis (Fig. 1d). This consistence fairly helped rule out the possibility that the observed CO2RR performance in the present study was brought by morphology change37. Meanwhile, the HR-TEM analysis, as well as related SAED, of the post-catalysts in Supplementary Fig. 33 also confirmed the chemical nature of the CuOx was the same as the as-prepared catalyst (Fig. 1e). Furthermore, in terms of electrochemical stability, we also evaluated the electrochemical active surface area (ECSA) of CuOx before and after the CO2RR electrolysis. The double layer capacitance (Cdl) was used for the assessment, since it is directly proportional to the ECSA of materials50,51. Using a cyclic voltammetry (CV) method, the Cdl values of CuOx before and after the electrolysis were 77.27 and 76.60 μF cm−2, respectively (Supplementary Figs. 34 and 35). The result of nearly the same Cdl values reflected the fairly stable ECSA of CuOx after the electrolysis and highly suggested the stability of using R.S. Altogether, results of the operando XAS analyses, the electron microscopy analyses, and electrochemical studies convincingly proved the validation of R.S. as a steady-state approach.

Through operando TR-XAS analyses, we correlated the chemical nature of CuOx with its unique selectivity toward asymmetric C2H5OH molecules, which has evidently concluded a paramount role played by the composition of the Cu–Cu(I) ensemble. Moreover, the preceding analyses also revealed that different CO2RR products may result from catalytic surfaces with different compositions of Cu and Cu(I); a Cu-dominated surface synthesized CO, while the surface with halves of Cu and Cu(I) gave rise to ethanol (Supplementary Fig. 10). This rather suggests a delicate Cu(I)/Cu(0) interface is the likely active center for CO2 to asymmetric products and motivates future work regarding such interfacial engineering. Analogous interfacial engineering strategies have been applied to enhance product selectivity in electrochemical reaction systems52,53. The study by Feng et al., for example, has demonstrated a linear correlation between the boundary densities within their as-prepared Cu nanoparticles and the corresponding catalytic behavior of reducing CO to multi-carbon products, i.e., ethanol and acetate54. Furthermore, using scanning electrochemical cell microscopy (SECCM), Mariano and coworkers provided a submicrometer-resolved study on gold films55. The study with such spatial resolution clearly demonstrated that the electrocatalytic activity of a gold film for CO2RR depends on the grain boundary types and densities on the film.

Theoretical understanding of the CO2RR selectivity

As follows, we make an attempt to elucidate the unique selectivity in our reaction system thorough theoretical computation. By using density functional theory (DFT) calculations, we further evaluated the stability of oxygen-vacancy generation/removal with various compositions of Cu and Cu(I) at the top layer of Cu2O. All details regarding the structural simulation are addressed in Methods, while the model coordinates are listed in Supplementary Table 1. Such a geometric stability during the oxygen removal process can potentially facilitate the regeneration of the acting-catalytic environment by the redox-shuttle approach. These structural simulation results demonstrated that the equal numbers of Cu and Cu(I) on the top layer could provide an ideal environment to host the C–C bond formation intermediate (OCCO) (Supplementary Figs. 36 and 37). All studies in following parts, therefore, will focus on the surface composed of the equal numbers of Cu and Cu(I) for subsequent modeling (defined it as a Cu–Cu(I) ensemble). Afterwards, we characterized C-C coupling behavior on the Cu–Cu(I) ensemble, and the configurational results—calculated absorption energy and geometrically optimized structures of absorbed CO molecules at the gas-solid interface – were shown in Supplementary Fig. 37. The ensemble configured dimerized CO molecules to an asymmetric “end-side” configuration; a CO molecule was absorbed in the side-on direction and coupled with an end-on CO molecule, which was consistent with previous work56.

On the other hand, the explicit water solvation effect has been demonstrated to well achieve a realistic modeling in free energy calculations. As depicted in Fig. 4a, the free energy barriers (ΔG‡) for the C–C bond formation under the explicit water solvation indicated the overall OCCO formation on the ensemble was thermodynamically accessible at 0.73 eV, as the corresponding reaction free energy ∆G was −1.22 eV (for computation details, refer to Supporting Information). Furthermore, Bader charge analysis also suggests that the mixed-valence ensemble constructively stabilize the asymmetrical OCCO intermediate though a favorable electrostatic interaction, in which the highly polar OCCO (+0.50 and −1.68 for the positive and negative fragments of (O(1)C(1))δ+(C(2)O(2))δ−, respectively) is stabilized by the presence of Cu(0) and Cu(I) on the surface, respectively (Fig. 4b). All computational results demonstrated the potential introduced by the mixed-valence ensemble could produce an asymmetric OCCO intermediate in CO2RR, which might considerably facilitate the formation of C2 products from CO2RR.

Fig. 4: DFT calculation results and proposed selectivity mechanism scheme.
figure 4

a The profiles of the potential energy surface for the dimerization of CO on Cu–Cu(I) ensembles. b The Bader chargers of the OCCO intermediates. c The accessible carbonyl-stabilized intermediates on the Cu–Cu(I) ensemble. The relative free energy in eV of each hydrogenation stage is denoted in the parentheses. d The schematic representations of the determined factor for the selective ethanol synthesis.

Full size image

The plausible mechanistic steps of ethanol formation are computationally explored with considering the various hydrogenated intermediates on the Cu–Cu(I) ensemble, being simulated by the mixed-valence Cu2O surface model, as highlighted as 1H to 6H in Fig. 4c. The calculated electrochemical step started from the hydrogenation step of CCO (1H), being generated from the dehydration step of the experimental observed OCCOH intermediate54. The relative free energies in respect to the minimum geometries (all marked as 0 eV) at each hydrogenation stage were labeled in Fig. 4c. It should be noted that all of the hydroxyl containing intermediates (CCOH, HCCOH_OH, HCCHOH_OH, H2CCOH_OH, and H2CCHOH_OH) were found to be less stable by 0.82, 0.43, 0.65, 1.08, and 0.79 eV than the corresponding minima, respectively, where the presence of hydroxyl groups within the hydrocarbon species could lead to the pathway of ethylene formation via a proton-coupled-electron dehydration process. Although HCCOH (crossed out in grayscale) was proposed as a necessary intermediate for C2H4 formation55, it became inaccessible due to its unfavorable precursor – HCCHO.

By contrast, OH species at the boundary of Cu–Cu(I) ensemble could play a critical role in facilitating the ethanol formation. Such a mixed-valance region favored the formation of surface hydroxyl given the electron-rich Cu region enhances the basicity of the boundary oxygen sites of Cu(I) region. Theoretically, OH species at the Cu–Cu(I) boundary was commonly identified along the hydrogenation pathway of CO2 reduction, and that resulted in the attractive electrostatic interactions between surface-OH dipole and the dipole of carbonyl group of hydrocarbon intermediates. This carbonyl stabilization effect could protect the oxygen end from being protonated as well as occupy the valence electron of carbon to avoid the CC double bond formation. Consequently, managing the boundary oxygen sites during the electrochemical CO2 reduction could be the determined factor of ethanol generation. The unique selectivity toward C2H5OH could be attributed to the fact that the CuOx could maintain well balance of Cu and Cu(I) during eCO2RR using R.S. and offer an asymmetrically OCCO adsorbing site with the subsequent carbonyl group stabilization effect by the OH groups at the boundary of Cu–Cu(I) ensemble, eventually synthesizing an asymmetric oxygenated product (Fig. 4d).

Discussion

In summary, using redox shuttle manner, the results of product characterization demonstrated that in a wide potential range, the as-prepared CuOx exclusively produced C2H5OH over other CO2RR products. Using the operando time-resolved XAS, we tracked the evolution of its chemical nature under CO2RR. Quantitative XANES and EXAFS analyses evidently revealed that the steady chemical composition of the material could be achieved via the usage of the redox shuttle approach, while the employment of conventional chronoamperometry drastically altered the chemical nature of the material. Collectively, a strong correlation between the stabilized ensembles and the corresponding unique selectivity of CO2RR products has been established. More importantly, the carbonyl stabilization mechanism by the boundary OH species was proposed to prevent the protonation of the terminal oxygen sites of hydrocarbons, and such a hedge device produced an oxygenated CO2RR product (ethanol in our case) with the unique selectivity. Our studies provided the first empirical real time information, correlating the chemical nature of Cu-based catalysts with their corresponding selectivity toward CO2RR products.

Methods

Chemicals

The copper(I) bromide (CuBr, 98%), copper(I) chloride (CuCl, 99%), trioctylphosphine (TOPO, 99%), 1-hexadecylamine (HAD, 90%), oleylamine (80-90%), 1, 5-pentanediol (98%), and nafion solution (20 wt. %) were purchased from Acros Organics. The potassium bicarbonate (KHCO3, 99%) was purchased from Fisher Scientific. The gases CO2 (99.999%) and N2 (99.999%) were purchased from Shen-Yi Gas Co. These chemicals were used without further purification.

Synthesis for copper nanocubes

The nanocubes were prepared through a polyamine process under an inert atmosphere. CuBr (1 mmol) and TOPO (3 mmol) were firstly dissolved into oleylamine (15 mL) in a three-neck round-bottom flask. After well mixed, the solution was flushed with N2 for 30 mins to remove the oxygen in the apparatus. Under the inter atmosphere, the solution was heated at 80 °C for 5 mins and then up to 230 °C for another 3 h, turning into a reddish-brown clouded solution. Afterwards, the solution was air-cooled to room temperature for purification. The solution was washed with n-hexane (40 mL) and centrifuged at 8000 r.p.m. for 10 mins three times. The purified nanocubes were stored in n-hexane for the later use.

Synthesis for mixed-valence copper nanomaterials

In all, 20 mg HDA were added into 9 mL 1, 5-pentanediol. Then, the solution was preheated at 170 °C for 10 mins. Afterwards, as-prepared copper nanocube solution (1 mL) was transferred into the solution via one-shot injection. The solution was kept at the temperature for a half hour. Eventually, the solution turned yellow, suggesting the formation of mixed-valence copper nanomaterials. The materials were washed with an ethanol-water mixture in ratio 7:3 and centrifuged at 10,000 r.p.m. for 10 min three times. The purified materials were kept in absolute alcohol for the later use.

Catalyst characterization

The morphology of as synthesized materials was characterized using the transmission electron microscopy (TEM, Hitachi H-760 operated at 80 kV) and the scanning electron microscopy (SEM, JSM-6700F). The materials’ crystal structure was studied by using selected area electron diffraction (SAED) in a high-resolution TEM (HRTEM, JEOL JEM-2100F operated at 200 kV). The elemental analysis of the materials was conducted using energy-dispersive X-ray spectroscopy (EDS) inside the HRTEM.

Electrochemical measurements

All electrochemical measurements were carried out using Biologic-VSP instrument and in a three-electrode setup. The working electrode was a glassy carbon electrode (0.782 cm2); the counter electrode, a Pt wire; the reference electrode, an Ag/AgCl electrode (in 3 M KCl). The test cell was a home-made electrochemical cell. The cell consisted of two parts, cathode and anode compartments. The working electrode and reference electrode were placed into the cathode compartment, and the counter was in the anode compartment. Both compartments were separated by an anion exchange membrane (Fumasep FAA-3-PK-130, FUMATECH) to prevent potential contamination resulting from the dissolution of Pt counter. In electrochemical experiments, 10 μL of the solution of as-prepared materials (2.5 M) were loaded on the glassy carbon. Each compartment was filled with 30 mL pre-bubbled CO2-saturated 0.5 M KHCO3, whose pH changed from 8.6 to 7.2, and the CO2 flow was kept at 100 sccm for 10 more mins in the cathode compartment to build an air-free atmosphere for the later detection of as-prepared materials’ electrocatalytic activity for CO2 reduction. Several electrochemical techniques were employed in the study, including cyclic voltammograms (CV), chronoamperometry (CA), and electrochemical redox shuttle (R.S.). CVs were conducted in the potential interval between −0.35 V and −1.05 V at a scan rate 100 mV/s. CA was performed at stated potentials. R.S. was adopted at stated potentials and at stated time intervals (one may refer to Supplementary Fig. 6 for the potential map). Potentials listed in the article were iR-corrected and referenced to the RHE (short for reversible hydrogen electrode) scale by Eq. (3). The pH of 7.2 was used for the conversion, since the value before and after CO2RR treatment stayed constant at 7.2. The iR drop was determined using impedance measurement technique (ZIR) available in the potentiostat.

$$E_{{\mathrm{RHE}}} = E_{{\mathrm{Ag}}/{\mathrm{AgCl}}} + 0.21 + 0.592 \times {\mathrm{pH}} - iR$$

Double layer capacitance (C dl) Measurement

A cyclic voltammetry (CV) method was used to study the Cdl of as-prepared catalysts. CVs were conducted under N2-saturated 0.5 M KHCO3 at various scan rates (10–300 mV/s). A potential window from 0.1 to 0.2 V was adopted due to the absence of redox peaks in this region (Supplementary Fig. 3). Later, the anodic and cathodic current density difference of each CV at the center of the potential window was plotted as a function of scan rate, whose slope will automatically give the Cdl.

CO2RR product analysis

Analytical methods for product characterization here have been outlined previously37. Gaseous products were evaluated using gas chromatography (GC, Agilent 7890B). The GC was equipped with a thermal conductivity detector (TCD) and a flame ionization detector (FID) in series. The TCD was for H2 and CO detection, and the FID was for CH4 and C2H4 detection. As for those products in liquid phase, gas chromatography-mass spectrometry (GC-MS, 5977A) was adopted to detect ethanol and propanol, and nuclear magnetic resonance (NMR, Bruker Advance III 500 MHz) was used for the detection of formic acid. The faradaic efficiency (F.E.) of each product was calculated by Eq. (4):

$${\mathrm{FE}}(\% ) = \frac{{{\mathrm{moles}}\,{\mathrm{of}}\,{\mathrm{target}}\,{\mathrm{products}} \times n \times F}}{C} \times 100\%$$

where n is the number of electrons transferred, F stands for Faraday’s constant (96485 C mol−1), and C represents the total amount of charge passed through a working electrode.

Operando X-ray absorption spectroscopy

Operando X-ray absorption experiments for Cu K-edge measurement were conducted at beamline BL 17C of National Synchrotron Radiation Research Center (NSRRC). A three-electrode arrangement (the same setup as those described in electrochemical measurements) was used for the operando measurements; the measurements were performed in customized Teflon reactors with a Kapton tape window, allowing X-ray transmission.

Operando quick-scanning X-ray absorption spectroscopy

The measurements for Cu K-edge absorption were taken at TPS 44A, Hsinchu, Taiwan. The corresponding data were recorded in total-fluorescence-yield mode. The setup of the experiments was stated in the article. To have a quality spectrum, we applied the Quick-XAS mode in a time-resolution of 5 s for XANES and EXAFS analyses, while for EXAFS fitting the time-resolution was chosen to be 60 s. The X-ray absorption experimental data were collected in total-fluorescence-yield mode, in which the metallic Cu foil was taken as reference to calibrate the energy scale. Using Demeter, all X-ray absorption spectra were processed by subtracting the baseline of pre-edge and normalizing that of post-edge; such data processing was for both X-ray absorption near edge spectra (XANES) and extended X-ray absorption fine structure (EXAFS) analyses. EXAFS analysis was carried out using Fourier transform on k3-weighted EXAFS oscillations to assess the contribution of each bond pair to Fourier transform peak. The curve fitting of EXAFS spectra was conducted using the software, REX2000, with FEFF8 program.

Theoretical analysis

All calculations were performed with the DFT plane-wave method utilizing the Vienna ab-initio simulation package (VASP)57,58,59,60. We used the projector-augmented-wave method (PAW)61,62 in conjunction with generalized gradient-approximation (GGA) calculation, and Perdew-Burke-Ernzerhof (PBE) exchange-correlation functional was applied. For the design of slab models, we used 8, 6, and 12 atomic layers for Cu2O(111), Cu2O(110), and Cu2O(100), respectively, as shown in Supplementary Fig. 36. For the DFT + U calculations, we chose the cutoff energy at 500 eV with 8 × 8 × 8 and 4 × 4 × 1 k-point mesh for bulk and surfaces models, respectively. On-site Columbic correction (U = 5.2 eV) was used to prevent delocalization of Cu d-orbitals, that has been demonstrated for modeling the previous Cu2O system63,25. The bottom two atomic layers were kept frozen with the remaining layers fully relaxed for these Cu2O slabs. In all cases, the slab calculations included a vacuum region of thickness greater than 15 Å, large enough to ensure no interaction between the two slabs. The adsorption energies were calculated based on the following equation,

$$\Delta E{\mathrm{ads}} = E[{\mathrm{slab}} + {\mathrm{adsorbate}}] - (E[{\mathrm{slab}}] + E[{\mathrm{adsorbate}}])$$

in which E[slab + adsorbate], E[slab], and E[adsorbate] were the calculated electronic energies of adsorbed species on the above surfaces, clean surfaces, and free molecules, respectively. The climbing-image-nudged-elastic-band (CI-NEB) method25,64,65,66 was applied to locate transition structures; the profile of the potential-energy surface (PES) was constructed accordingly. Frequency calculations were applied to verify the adsorbed intermediates and the transition states (with only one imaginary frequency). Based on the stepwise oxygen-vacancy formation energy, the Cu2O(110) surfaces was found to contain the most subtle surface reconstruction among the three common Cu2O cases (i.e., (111), (110), and (100)) as illustrated in Supplementary Fig. 36.

Solvent model

To provide an appropriate solvent effect of CO dimerization reaction, we constructed an explicit water layers (64 water molecules) to represent the hydrogen bond network of the aqueous environment. Water molecules were described by SPC-3f(α) models67 using LAMMPS molecular dynamics simulator25. The liquid-solid interfaces were brought to equilibrium with 2 ns pre-equilibrium simulations68 at 350 K while the solid geometry (with adsorbate) remained frozen during the molecular dynamics simulations, being represented by atomic charges (from Bader charges of DFT). The adsorbate and solid structures were also taken from our DFT calculations. After 2 ns of pre-equilibrium, we sampled 10 solvation structures (0.2 ns apart per sample) for the initial state 2CO (IS), C–C bond coupling transition state (TS), and OCCO final state (FS), followed by solvent structure optimization at DFT-level. Finally, we took the average energy of these 10 structures to represent E(IS), E(TS), and E(FS) for CO‒CO coupling process, respectively.

Thermodynamic correction

The relative Gibbs energy in solution (ΔG) for the free energy difference between any adsorbate pair on the surfaces was defined as ΔG = ΔH ‒ TΔS. ΔH was estimated as Δ(E + ZPE + Evib) with E representing the average electronic energy of DFT under explicit solvation treatment, ZPE representing zero-point energy, and Evib representing the internal energy correction of vibrational models (the relative translational and rotational contribution for any adsorbate pair was assumed canceled out)69,70,25. ΔS was predicted by the vibrational partition function using harmonic vibrations. For the CO‒CO coupling process, the vibrational modes of two surface Cu atoms were taken into account. For example, \(\overline {\Delta G^\ddagger } = \overline {\Delta G} ({\mathrm{TS}}) - \overline {\Delta G} (2{\mathrm{CO}}) = [\overline {\Delta H} ({\mathrm{TS}}) - T\overline {\Delta S} ({\mathrm{TS}})] - [\overline {\Delta H} (2{\mathrm{CO}}) - T\overline {\Delta S} (2{\mathrm{CO}})]\) was estimated by each 10 sampled configurations. For the ethanol formation pathway, we did not take into account the explicit solvation effect due to its trivial difference.