Main

Electrochemical energy conversion driven by renewable energy is a cost-effective, environmentally friendly route to convert undesired substances (such as CO2) into valuable chemicals and fuels. A challenge for the practical application of complex electrochemical processes, such as the CO2 reduction reaction (CO2RR) over copper-based catalysts, is control of selectivity1,2,3,4,5. One of the critical parameters affecting the catalyst’s properties and function—its oxidation state6,7—can be conveniently manipulated in situ by choosing appropriate applied potentials. In particular, under pulsed reaction conditions the desired structural motifs and preferred oxidation state can be (re)generated8,9,10,11.

Previous investigations have provided conflicting explanations for the enhancement of catalytic properties under pulsed CO2RR conditions. Studies employing polycrystalline copper electrodes8,9,12,13,14,15,16,17,18, single crystals10 and preoxidized copper nanocubes11,19 revealed enhanced C2+ product formation, suppression of the competing hydrogen evolution reaction (HER) and increased stability of catalytic activity under pulsed CO2RR. These findings were attributed either to the formation of oxidized copper species10,11,18, to catalyst surface morphology changes10,19,20, to the removal of poisoning hydrogen and carbon species from the catalyst’s surface8,17, to an enhanced adsorption of OH species and stabilization of CO2RR intermediates16, or to the local modulation of the electrolyte pH and CO2 concentration17,19,21,22. The dominant mechanism depends on the parameters of the applied potential (for example, its value, pulse shape and duration of the working and anodic pulses).

In this article we focus on the regime in which the selected anodic potential (Ea = +0.6 V versus the reversible hydrogen electrode (RHE)) and the duration of the restoring pulse (Δta > 0.5 s) are sufficient for the generation of copper oxides. Thus, changes in the catalyst surface structure and composition could play a decisive role in the reaction pathway and selectivity. By controlling the type and amount of copper oxide formed during the restoring pulses we steer the CO2RR selectivity of a Cu2O nanocube (NC)-derived catalyst. In particular, ethanol formation is doubled (compared with that under stationary conditions) within a narrow range of pulse durations, where a balance between metallic copper and distorted copper oxide species on the catalyst surface is achieved, as revealed by time-resolved operando X-ray absorption spectroscopy (XAS) and high-energy X-ray diffraction (XRD), and supported by quasi in situ X-ray photoelectron spectroscopy (XPS). Our study demonstrates the great potential of pulsed electrolysis for tailoring the catalyst performance and highlights the role of operando investigations for the mechanistic understanding of next-generation catalysts operating under dynamically changing reaction conditions.

Results

Selectivity under static and pulsed reaction conditions

The size-selected (30 nm) and shape-selected Cu2O NCs employed in this study (Fig. 1a) were shown to be catalytically active and selective for CO2RR to C2+ products under potentiostatic conditions19,23,24, but are also well-suited for synchrotron-based operando investigations. Details of the sample preparation and characterization are given in Supplementary Note 1 and Supplementary Fig. 1.

Fig. 1: Evolution of the catalyst morphology and selectivity under static and pulsed CO2RR.
figure 1

ac, Ex situ TEM images of as-prepared Cu2O NCs (a), NCs after CO2RR for 60 min under static −1.0 V (RHE) potential (b) and NCs after pulsed CO2RR for 60 min (Δtc = Δta = 10 s, Ec = −1.0 V and Ea = +0.6 V) (c). Scale bars correspond to 20 nm. dg, Changes in the FE under pulsed CO2RR with respect to that under a static −1.0 V potential: excess FE for EtOH (d), C2H4 (e), CO (f) and CH4 (g) under pulsed CO2RR with different lengths of anodic and cathodic pulses (Δta and Δtc, respectively). Regions A–D correspond to distinct regimes with different catalyst surface composition, as explained in the text.

Under static CO2RR in CO2-saturated 0.1 M KHCO3 at −1.0 V, a transformation of the crystalline Cu2O phase to metallic copper was observed by XRD, while XAS showed nearly complete copper reduction at −1.0 V (Supplementary Figs. 27)23,24. This process is relatively slow: even after 60 min under CO2RR, peaks of the Cu2O phase were still present in the XRD pattern (Supplementary Fig. 4), and several hours were required until no further changes were visible in the XAS data (Supplementary Fig. 3 and Supplementary Note 2). The reduction under static CO2RR also resulted in changes in the catalyst morphology, namely, a partial loss of the cubic shape (Fig. 1b and Supplementary Fig. 1). The selectivity of the reduced catalyst under static conditions was in line with previous reports24,25,26. Ethylene (C2H4), methane (CH4) and ethanol (EtOH) were detected as the main CO2RR products, with Faradaic efficiencies (FEs) of 40%, 20% and 10%, respectively. The FE of H2 from the competing HER was only 17% (Supplementary Note 3 and Supplementary Table 1).

To the pre-reduced catalyst, we applied a series of oxidizing (anodic) pulses and working (cathodic) pulses with durations Δta and Δtc, respectively. Figure 1d–g shows the difference in the FEs of the products measured under pulsed CO2RR with varied Δta and Δtc, with respect to the FEs obtained under a static cathodic potential (−1.0 V). The potential values for cathodic and anodic pulses in all cases were Ec = −1.0 V and Ea = +0.6 V, respectively. The latter is a potential that is expected to oxidize Cu(0) to Cu(I) (Supplementary Note 4 and Supplementary Fig. 8)10. The striking feature in Fig. 1 is the distinct dependency of the ethanol FE on the pulse duration. At Δta > 2 s and Δtc < 2 s, the formation of ethanol is suppressed. Furthermore, an enhancement in the ethanol FE is observed for Δta < 2 s and Δtc values between approximately 1 s and 10 s, with a maximum at Δta = 0.5 s and Δtc = 4 s, where the ethanol FE is 1.7 times higher than that under static CO2RR (Fig. 1d). At the same time, the C2H4 formation (Fig. 1e) is suppressed under pulsed CO2RR at all Δta and Δtc. The latter is accompanied by a boost in the C1 product formation (CO, Fig. 1f and CH4, Fig. 1g), and the suppression of the H2 (Supplementary Fig. 9a).

Supplementary Note 3, Supplementary Tables 2 and 3 and Supplementary Figs. 915 provide additional details on pulse-length dependencies of the catalyst selectivity. In particular, Supplementary Fig. 9a shows a remarkable independency of the H2 selectivity on the pulse duration. This indicates that the enhanced ethanol formation observed for Δta < 2 s and 1 s < Δtc < 10 s cannot be attributed to changes in the hydrogen coverage or local pH. These effects are considered in the prior literature8,17,19,21 to explain the changes in CO2RR under pulsed conditions, but should also affect the HER. While we recognize their importance for pulsed CO2RR in general, they do not play a decisive role in the ethanol selectivity for the pulse lengths and oxidative potential values used here. We also note here that an oxidation of reaction products and intermediates by anodic pulses cannot be ruled out, but is expected to have only a minor impact (Supplementary Note 1). The FE changes observed in our system thus must stem from two different effects: periodic dynamic changes induced in the catalyst by potential pulses, and the irreversible evolution of the catalyst morphology. Indeed, the ex situ transmission electron microscopy (TEM) images in Fig. 1c and Supplementary Fig. 1 show that exposure of the catalyst to pulsed CO2RR conditions for 1 h results in a partial loss of the cubic morphology and the formation of a peculiar granular structure, not observed under static CO2RR. We attribute these to the dispersion/dissolution–redeposition processes that we recently revealed for cubic copper nanostructures under potentiodynamic CO2RR-related conditions using electrochemical electron microscopy27,28,29.

To distinguish the relative contribution to the selectivity of dynamic transformations from the concomitant irreversible structural changes also induced by the pulses, the pulse sequence was interrupted after 1 h and the catalyst selectivity was subsequently measured under static CO2RR conditions after another hour. As shown in Supplementary Figs. 9 and 10, under these conditions the high FE for CH4 is preserved, while the FE for C2H4 remains low. Thus, the enhancement in CH4 formation and suppression of C2H4 must stem from the irreversibly increased surface roughness and granular morphology. Nonetheless, such morphological changes cannot explain the increase in the CO FE, the suppression of HER and the pulse-length-dependent ethanol FE. In fact, when the pulses are interrupted, the FEs of CO and H2 mostly return to the level observed for a fresh sample under static CO2RR conditions. Moreover, the ethanol FE under pulsed CO2RR appears to be a result of two competing mechanisms: irreversible morphological transformations and dynamic processes. The former have a detrimental effect, similarly to the case of C2H4 (Supplementary Fig. 9e,f). The latter, within a narrow range of Δta and Δtc values, results in enhanced ethanol formation, and are probably associated with changes in the copper chemical state. Their transient nature requires us to apply operando methods for their investigation.

Catalyst evolution under pulsed reaction conditions

Periodic reversible changes in the catalyst structure and composition under pulsed CO2RR with Δta = Δtc = 10 s and Ea = 0.6 V are evidenced by both XAS (Fig. 2a,b) and XRD (Fig. 2c). In particular, linear combination analysis (LCA) of X-ray absorption near-edge spectroscopy (XANES) data (Fig. 2d) shows that the Cu(I) fraction increases upon applying the anodic pulse, and decreases during the cathodic pulse. An opposite trend is observed in the Cu(0) concentration, while the Cu(II) concentration varies much less (Supplementary Fig. 16). The variations in the fraction of the metallic copper phase and the size of the crystalline copper domains are captured also by the changes in the metallic Cu(311) Bragg peak area and width, respectively. In particular, the amount of metallic copper and the size of the copper domains are smaller under pulsed CO2RR with respect to those during the static CO2RR at −1.0 V. Similar variations in the catalyst composition and structure were observed for longer (Δta = Δtc = 30 s) and shorter (Δta = Δtc = 1 s) pulses (Supplementary Figs. 1618).

Fig. 2: Evolution of the operando Cu K-edge XAS and XRD data under pulsed CO2RR.
figure 2

ac, Changes in XANES (a), EXAFS (b) and XRD (c) data under an anodic pulse. Data corresponding to different times since the onset of the anodic pulse are shown (see colour bar). Panels a and b show raw spectra and the differential data, where the XANES or the magnitude of the Fourier-transformed (FT)-EXAFS corresponding to the onset of the anodic pulse (t = 0) were subtracted. Panel c shows differential XRD patterns, where the XRD intensities corresponding to t = 0 were also subtracted. Inset: zoomed-in region corresponding to the (311) Bragg peak for metallic copper. d, Time dependencies of the concentrations of Cu(0) and Cu(I) species from XANES analysis, and changes in the area of the Cu(311) Bragg peak as a measure of the amount of metallic copper, and in the coherence length of the metallic copper domains estimated using the Scherrer equation from the width of the Cu(311) Bragg peak. Time dependencies of the applied potential E(t) and measured current I(t) are also shown. w0 and CL0 are the concentration and coherence length, respectively, of metallic copper species at t = 0. Pulse parameters: Δtc = Δta = 10 s, Ec = −1.0 V, Ea = +0.6 V.

Furthermore, for all pulse durations with Δta = Δtc, the periodic variations of the Cu(I) and Cu(0) concentrations were not accompanied by accumulation of oxide: all oxide species generated during the anodic potential pulse were removed during the subsequent cathodic pulse. The removal of the electrochemically grown oxide species is quick, which contrasts with the sluggish initial reduction of the Cu2O phase under the static CO2RR. This implies that only the near-surface layers of the catalyst (up to 0.5 nm, as estimated from the changes in coherence length) are reoxidized during the anodic pulse, while the metallic core structure is maintained. The structure and properties of this electrochemically formed surface oxide probably differ noticeably from the properties of the bulk-like Cu2O phase in the as-prepared NCs, making it easier to reduce. The formation of a thicker oxide layer can be achieved by increasing the anodic potential value (Supplementary Fig. 19e and Supplementary Note 5)19. The catalyst properties in this regime of higher Ea values might differ strongly from those of the system exposed to Ea = 0.6 V pulses, and are beyond the scope of this paper.

Characteristic time scales of catalyst transformations

To better understand the periodic changes in the catalyst structure, we averaged the XAS and XRD data points collected at the same times after the onset of each pulse cycle (Fig. 3 and Supplementary Figs. 2024). The LCA-XANES method and extended X-ray absorption fine structure (EXAFS) data fitting for the averaged data (Supplementary Fig. 20) both revealed the oxidation of copper species and Cu–O bond formation during the anodic pulse, and are in excellent quantitative agreement (Fig. 3a and Supplementary Fig. 22). The periodic reoxidation of the catalyst is also observed in XRD, as evidenced by the changes in the metallic copper phase (Fig. 3b and Supplementary Fig. 23), but also by the appearance of Bragg peaks from a Cu2O-like phase (Fig. 3c).

Fig. 3: Variations in catalyst structure and composition during potential pulse.
figure 3

a, LCA-XANES analysis results (grey open circles) for averaged data for NCs under CO2RR with pulse lengths Δta = Δtc = 10 s, Ec = −1.0 V and Ea = 0.6 V. Results of EXAFS data fitting (filled black and red circles show the Cu(0) and total Cu(I) and Cu(II) concentrations extracted from Cu–Cu and Cu–O coordination numbers, respectively). Solid lines display the result of fitting concentration profiles for Cu(I) (red line) and Cu(II) (green line) with an analytical model. The concentration profile for Cu(II) is multiplied by 2 for clarity and to account for the twofold larger Cu–O coordination number in CuO than in the Cu2O bulk structure. Depicted error bars reflect the standard errors of LCA-XANES fitting results. The time dependencies of the potential E(t) and the current I(t) are also shown (vertical bars show the scale for different datasets). b, Evolution of the Cu(311) Bragg peak parameters. The relative changes in the Bragg peak area, the copper coherence length and the lattice strain are depicted. c, Evolution of the Cu2O-like Bragg peak parameters during the anodic pulse. The total area of the (111) and (220) Bragg peaks, average coherence length extracted from the (111) and (220) peak broadenings, and the lattice strain for the (111) and (220) lattice spacings are shown. The solid lines in b and c are guides for the eye (exponential fits).

Although the spike in the current I(t) detected upon changing the applied potential and mostly related to double-layer charging is relatively short (<1 s, Fig. 3a), the transformations of the catalyst structure and composition are much slower. Moreover, a strong asymmetry between the oxidation and reduction processes was observed, with the latter being significantly faster. Indeed, although 10–20 s of the anodic potential with Ea = 0.6 V are required for the oxidation to reach its stationary value, the complete removal of the oxide species generated is achieved within 1–2 s of the cathodic pulse. Similarly, the intensity of the XRD peaks corresponding to the metallic copper lattice decreases within 10–20 s upon application of the anodic pulse, and increases back rapidly, within 1–2 s, upon the onset of the cathodic pulse (Fig. 3b and Supplementary Fig. 23). When comparing the results for different pulse lengths, we observe that the characteristic reduction and oxidation times are independent of the pulse length (Supplementary Fig. 24), and thus can be used as descriptors of a particular sample under a given oxidizing potential (Ea). The XAS and XRD results are also confirmed by quasi in situ XPS measurements (Supplementary Fig. 25 and Supplementary Note 6). In particular, at all pulse lengths, the surface-sensitive XPS method shows much higher variations in the cationic copper concentration due to the potential pulse than the bulk-sensitive XAS, providing important evidence that catalyst oxidation is limited to its surface.

Moreover, even at Ea = 0.6 V, not only Cu(I) species but also some Cu(II) is created during the anodic pulse, as evident from XANES, but also confirmed by observed variations in the Cu–O bond length in the EXAFS data (Supplementary Fig. 26). The presence of these Cu(II) species is attributed to the unique surface morphology adopted by the NCs under the harsh pulsed reaction conditions. Cu(II) species are likely to be unstable, as noted in cyclic voltammetry experiments where different Cu oxidation states were tracked by XAS (Supplementary Note 4) and the generation of Cu(II) coincided with the loss of copper fluorescence intensity. Indeed, inductively coupled plasma mass spectrometry (ICP-MS) analysis of the copper concentration in electrolyte after CO2RR shows partial catalyst dissolution under all pulsed regimes investigated (Supplementary Note 7 and Supplementary Table 4), signalling the appearance of unstable species during the anodic pulse. Due to their continuous regeneration, the importance of these species for catalytic activity cannot be excluded. We observe no delay between the formation of Cu(I) and Cu(II) species. In fact, it appears that the generation of Cu(II) species starts slightly before that of Cu(I) species. This suggests that instead of the gradual oxidation [Cu(0) → Cu(I) → Cu(II)], the generation of Cu(II) species is a separate, independent process [Cu(0) → Cu(II)]. The independence of the Cu(0) → Cu(II) and Cu(0) → Cu(I) processes is also observed in the cyclic voltammetry experiments (Supplementary Note 4). This leads to a situation that during the short anodic pulse, the concentrations of the Cu(I) and Cu(II) species are comparable, although at large Δta values the concentration of Cu(I) is much higher than that of Cu(II). Indeed, in the dataset corresponding to 1 s pulses (Supplementary Fig. 22a) the variations in Cu(II) species concentrations seem to be as pronounced as those in the Cu(I) concentration. With increasing Ea, the contribution of Cu(II) species increases strongly (Supplementary Fig. 22d). The generation of Cu(II) species at Ea = 1.0 V is clearly faster than that of Cu(I) species at Ea = 0.6 V, resulting in a steeper drop in the Cu(0) concentration when an anodic pulse is applied (Supplementary Note 5).

We conclude that under CO2RR pulses with Ea = 0.6 V, a layer of oxide with mixed valence state (Cu(I)/Cu(II)) is formed on the surface of the copper NCs. Further insights into the structure of this oxide can be extracted from the XRD measurements. The area of the Bragg peaks attributed to Cu2O shows similar kinetics (with opposite sign) to that of metallic copper (Fig. 3b,c). The coherence length of the CuOx phase increases within the entire 10 s anodic pulse, reaching 3 nm at the end of the pulse. At the same time we observe a gradual expansion of the oxide lattice, as determined from the position of the Cu2O(111)-like Bragg peak. Importantly, the Cu2O(220)-like Bragg peak position follows a different trend: the (220) lattice expands during the first 2–3 s upon the application of the anodic pulse, and then slightly shrinks. This suggests a distortion of the Cu2O-like phase in the early stages of the anodic pulse, which we associate with the growth of (111)-oriented Cu2O on the metallic Cu(100) surfaces30,31. The lattice mismatch between Cu(100) and Cu2O(111) induces the in-plane expansion and a relatively larger Cu2O-like (220) lattice spacing. The reduced (111) lattice spacing in the early stages of the anodic pulse might, in turn, be a consequence of a tilted growth axis which differs slightly from the surface normal and leads to anisotropic compressive strain in the oxide lattice30,31,32. As the oxide layer grows thicker, the distortion gradually diminishes, leading to a rather bulk-like Cu2O after 10–20 s. Changes in the lattice spacing in metallic copper under pulsed CO2RR are also observed, but are quite different from those in the oxide phase. In particular, the copper lattice expansion upon the application of the anodic pulse is faster than the changes in the Cu2O(111) Bragg peak position (Fig. 3b,c), and the expansion magnitude practically does not depend on the anodic pulse length (Supplementary Fig. 24). This expansion appears to be independent from the copper oxidation process and reflects the charging of the Cu/CuOx interface.

Linking the catalyst’s structural evolution to its selectivity

Considering that the oxidation and reduction rates do not depend on the pulse length, we estimate them from the XAS data collected during pulsed CO2RR with the largest Δta and Δtc values (30 s), and propose a simple kinetic model to predict the time dependencies of the concentrations of Cu(0), Cu(I) and Cu(II) species at arbitrary Δta and Δtc (Supplementary Note 8, Supplementary Fig. 27 and Supplementary Table 5). Knowing the latter, we can now focus on: (1) the average concentration of copper oxide species present during the cathodic pulse, (2) the ratio of the Cu(II) and Cu(I) content during the cathodic pulse and (3) the concentration of oxide species remaining after the cathodic pulse is completed (Fig. 4a–c). Based on these results and the catalytic selectivity data in Fig. 1d–g, we identify four regimes (A–D, Fig. 4a–c) with distinct catalytic properties, which stem from distinct catalyst surface compositions. The latter are schematically depicted in Fig. 4d. In particular, the enhanced ethanol production at low Δta and intermediate Δtc values (region B) is attributed to an optimal balance between oxidized and reduced copper species on the catalyst surface, and the presence of a distorted copper oxide phase.

Fig. 4: Effect of the durations of the anodic (Δta) and cathodic (Δtc) pulses.
figure 4

ac, Results obtained from the analysis of XAS data for copper NCs during pulses with Ea = 0.6 V, showing the average concentration of the oxide species present during the cathodic pulse (a), the ratio of Cu(II) and Cu(I) concentrations during the cathodic pulse (b) and the concentration of oxide species that remains after the nth cathodic pulse is completed (c). d, Schematic depiction of the catalyst structure and composition during a cathodic pulse extracted from XAS and XRD data. The ranges of Δta and Δtc values corresponding to each regime are indicated in ac.

At larger Δtc (region A), in the absence of the effect of an irreversible morphology change, the FEs for all CO2RR products become similar to those observed under static CO2RR. Here the catalyst is on average completely metallic, and dynamic changes associated with the pulsed protocol take place for proportionally shorter times and cannot affect the catalyst performance noticeably. However, at shorter Δtc, the oxide species created on the catalyst surface during the anodic pulse have a pronounced impact. Note that the formation of a complete monolayer of surface oxide on cubic nanoparticles 30 nm in size is equivalent to an increase in the oxide concentration by 3%, which is in good agreement with the maximal increase in the oxide fraction observed in Fig. 4a for pulses with Δta Δtc, suggesting that the catalyst surface is completely oxidized during the anodic pulse in this case. During the subsequent cathodic pulse, the surface is completely (region C) or partially (region D) re-reduced, recovering the metallic copper sites active for CO2RR. The presence of metallic islands on the catalyst surface in direct contact to copper-oxide-rich surface regions during the working pulse has a beneficial effect for CO formation. In particular, the pulse-duration dependency of the excess FE for CO (Fig. 1f) has a similar profile as the pulse-duration dependency of the oxide fraction present during the working pulse (Fig. 4a), where an increase in the oxide fraction results in an enhanced CO formation. The latter observation is in agreement with the results obtained under pulsed CO2RR for a copper foil, where the CO production was favoured over HER when significant accumulation of oxides was observed33. While the pulse parameters in ref. 33 differ from those in our work, this study provides additional evidence that strongly oxidized copper surfaces can lead to efficient CO2 conversion to CO.

The coexistence of metallic and oxidized Cu species is also the key factor for enhanced ethanol formation. As predicted by density functional theory modelling11,34,35 and corroborated also by recent experimental infrared spectroscopy and in situ XAS data36, the asymmetry between CO adsorption energies on metallic and oxidized copper sites facilitates CO dimerization and is essential for C2 product formation34. Nevertheless, the more complex shape of the pulse duration dependency for ethanol production as compared with CO suggests that not only the amount but also the type of copper oxide formed during the anodic regime play a role in the increased ethanol yield. The enhancement in ethanol is observed only for the shortest anodic pulses (Δta < 1 s, region B), where the distorted oxide structure (as seen from our XRD and XAS data) is formed, featuring also some contribution of Cu(II) species (oxides or hydroxides, Fig. 4b). We consider these distorted oxide or hydroxide species, in close contact with metallic copper, to be the crucial sites for the ethanol production. Importantly, an increase of Δta results in an increased thickness of the oxide layer that adopts a structure that increasingly resembles the structure of bulk Cu2O (region C). The latter results in the loss of the enhanced ethanol production. Similarly, the reduction of Δtc below 1 s leads to an incomplete reduction of the oxide layer during the cathodic pulse (Fig. 4c), resulting in the gradual accumulation of the oxide and the formation of a thicker Cu2O shell with bulk-like structure (region D). As a consequence, the enhancement in the ethanol production is not pronounced for Δtc < 1 s. The observed peak of the ethanol formation at Δtc ≈ 4 s and Δta < 1 s corresponds to the optimal ratio between distorted oxidic and metallic copper sites.

Our advanced operando X-ray studies thus revealed the complex interplay between a (reversible) oxide formation and the catalytic function of oxide-derived cubic copper nanocatalysts. We show the crucial role of the catalyst’s surface repopulation with distorted multivalent copper oxide species that boost ethanol production. In contrast, extensive bulk-like copper-oxide formation prevents CO dimerization and rather facilitates CO formation. The observed dependency of the catalyst selectivity on the parameters of the applied pulsed CO2RR protocol provides opportunities for steering the catalyst’s selectivity on demand.

Methods

Catalyst synthesis

Cu2O NCs were prepared by adapting a synthesis recipe described in a previous work37. Briefly, 4 ml 0.1 M CuSO4 (Sigma-Aldrich, >98%) was diluted in 366 ml H2O and stirred vigorously. Then 14 ml 1 M NaOH (Alfa Aesar, >97%) was added to start the nucleation process, while the reduction and NC growth were initialized by adding 16 ml of 0.25 M l-ascorbic acid (Sigma-Aldrich, reagent grade) after 10 s. The solution was stirred further for 13 min. The obtained sample was washed two times with an EtOH:H2O mixture (1:1) and once with EtOH. The clean sample, with a copper concentration of 1.2 mg ml−1, as determined by ICP-MS, was stored in 20 ml of EtOH. The electrodes were prepared by drop-casting 42 μl of the former solution on both sides of 1 cm2 carbon paper (Alfa Aesar, Toray Carbon Paper, GGP-H-60). The total copper loading was 50 µg.

Electrocatalytic measurements

Electrocatalytic measurements were performed with an Autolab (Metrohm) potentiostat in a H-type cell equipped with an anion-exchange membrane (Selemion AMV, AGC) separating the cathodic and the anodic compartments. A leak-free Ag/AgCl reference electrode (LF-1, Alvatek) was set close to the working electrode in the cathodic compartment. A platinum gauze electrode (MaTecK, 3,600 mesh cm−2) served as the counter-electrode in the anodic compartment. As electrolyte, 0.1 M KHCO3 (Alfa Aesar, 99.7–100.5%) was purified with a cation-exchange resin (Chelex 100 Resin, Bio-Rad) and saturated with CO2 (99.995%) for at least 20 min. Online products were determined for the static CO2RR conditions by taking an aliquot of the electrolyte every 15 min.

After applying a linear sweep voltammogram from the open circuit potential to −1.0 V at 20 mV s−1, a potential pulse protocol was carried out with fixed cathodic potential Ec = −1.0 V, anodic potential Ea = +0.6 V and different lengths of cathodic and anodic pulses. The linear sweep step was omitted for the data points with low Δtc and Δta values, since in this case it contributed non-negligibly to the final yields of reaction products. The pulse sequence was repeated for 4,000 s. For comparison, after pulsing, another chronoamperometric measurement at constant −1.0 V was performed for 4,000 s.

Gas products were detected and quantified after 1 min and every 15 min by online gas chromatography (GC, Agilent 7890B), equipped with a thermal conductivity detector and a flame ionization detector. Liquid products were analysed after each measurement with a high-performance liquid chromatograph (Shimadzu Prominence), equipped with a NUCLEOGEL SUGAR 810 column and a refractive index detector, and a liquid GC (Shimadzu 2010 plus), equipped with a fused silica capillary column and a flame ionization detector. Each presented data point was measured at least three times, and we report the average values.

All catalytic results in this study are shown in terms of FE. The FE of the gas products was calculated as:

$${\mathrm{FE}}_x = \frac{{\dot V \times C_x \times z_x \times F}}{{A \times V_{\mathrm{M}} \times j_{\mathrm{total}} \times \frac{{{{{\mathrm{{\Delta}}}}}t_c}}{{{{{\mathrm{{\Delta}}}}}t_c + {{{\mathrm{{\Delta}}}}}t_a}}}} \times 100{{{\mathrm{\% }}}}$$
(1)

The FE for liquid products (which cannot be measured online, but only after a certain time under pulsed CO2RR) was calculated as \({\mathrm{FE}}_x = \frac{{V \times {{{\mathrm{{\Delta}}}}}C_x \times z_x \times F}}{{{{{\mathrm{{\Delta}}}}}Q}} \times 100\%\). The factor \(\frac{{{{{\mathrm{{\Delta}}}}}t_c}}{{{{{\mathrm{{\Delta}}}}}t_c + {{{\mathrm{{\Delta}}}}}t_a}}\) is used in the FE expression for the gas products to weight jtotal according to the total time spent by the catalyst in the Faradaic (cathodic) potential regime. In contrast, for liquid products ΔQ already contains the contribution of the pulse times by integrating the cathodic charge. Here FEx is the FE of product x, \(\dot V\) is the CO2 gas flow rate (l s−1), Cx is the volume fraction of the product x detected by GC, zx is the number of electrons transferred for reduction to product x, F is the Faraday constant (C mol−1), A is the geometric area of the electrode (cm2), VM is the molar volume (22.4 l mol−1), jtotal is the total current density during CO2 bulk electrolysis (A cm−2), \(\frac{{{{{\mathrm{{\Delta}}}}}t_c}}{{{{{\mathrm{{\Delta}}}}}t_c + {{{\mathrm{{\Delta}}}}}t_a}}\) is a factor to account for the effective CO2RR time of jtotal while pulsing, ΔCx is the accumulated concentration of product x detected by HPLC or liquid GC (mol l−1), ΔQ is the total charge transfer during the electrolysis at constant potential or current (C) and V is the volume of the electrolyte (l).

The electrochemical surface area after 1 h reduction at −1.0 V was obtained by double-layer capacitance with cyclic voltammetry in the non-Faradaic region with variable scan rates. The resulting capacitance of 0.23(2) mF cm−2 corresponds to a 8.5-fold higher surface area compared to the copper foil (0.027 mF). Cyclic voltammetry was carried out between +0.3 and +0.45 V versus RHE, with scan rates of 20, 40, 60, 80, 100, 120 and 140 mV s−1. The amount of dissolved copper in the electrolyte after CO2RR was determined by ICP-MS (iCAP RQ, Thermo Fisher Scientific). See Supplementary Note 1 for more details on the analysis of electrocatalytic measurement data.

Characterization of catalyst morphology and composition

The morphology of the catalysts before and after CO2RR was characterized by transmission electron microscopy (TEM). TEM measurements were performed using a Thermo Fisher 300 kV Titan microscope. The catalysts after CO2RR were sonicated in 200 µl isopropanol to detach the sample and drop-cast on a nickel lacey carbon grid.

The surface composition of the copper catalysts was determined by quasi in situ XPS in the as-prepared state of the catalyst, after 1 h reduction under static CO2RR and after 1 h under pulsed CO2RR. Following the approach established in our previous work to determine the oxidation state at the cathodic and anodic pulse10, the catalytic procedure was interrupted at the respective potential, the sample was washed with water at open circuit potential and directly transferred into an ultra-high-vacuum system for XPS measurements without exposure to air. The ultra-high-vacuum system system was equipped with an X-ray source (XR 50, SPECS) and a hemispherical electron analyser (Phoibos 100, SPECS, Epass = 15 eV) and an aluminium anode (Al Kα,  = 1,486.6 eV, 300 W).

Operando XAS measurements and data analysis

To collect X-ray absorption fine structure (XAFS) data, operando time-resolved XAS experiments at the Cu K-edge (8,979 eV) were carried out in quick XAFS (QXAFS) mode at the SuperXAS beamline at the SLS synchrotron (Switzerland). Additional experiments were carried out at the P64 beamline of the PETRA III synchrotron (Germany). In both cases the experiments were carried out in fluorescence mode using a home-made in situ cell for electrochemical measurements. A schematic of the cell is shown in ref. 38. A platinum mesh and leak-free Ag/AgCl electrode were used as counter-electrode and reference electrode, respectively. The sample was deposited on carbon paper (Sigracet 29 BC, SGL Carbon), with the other side of the substrate covered with Kapton tape. The sample loading used in our operando XAS experiments was 500 µg cm2: this is the maximal loading of the sample that allows us to collect data of good quality, but is still sufficiently low to avoid self-absorption effects in fluorescence XAS data. The sample was mounted in the electrochemical cell with its Kapton-covered side acting as a window for incident and fluorescent radiation, while the side coated with the catalyst was in contact with the electrolyte. As electrolyte we used CO2-saturated 0.1 M KHCO3, which was continuously circulated through the cell using a peristaltic pump. CO2 flow (20 ml min−1) was ensured through the cell. The applied potential was controlled with a BioLogic (in the experiments at SuperXAS) or an Autolab (in experiments at DESY) potentiostat.

In the SuperXAS set-up, the super-bending magnet was used as X-ray source. A silicon-coated collimating mirror and platinum-coated toroidal refocusing mirror were used to remove higher-order harmonics, control the beam size and reduce heat load on the monochromator. A Si(111) channel-cut monochromator designed for high-speed scans was used for energy selection. The monochromator oscillated with a 5 Hz frequency, resulting in five XAFS spectra being collected every second. The beam size was 30 μm × 2.5 mm. At P64, a tapered undulator was used as the X-ray source, a Si(111) channel-cut monochromator was used for fast energy selection, and the beam size was less than 2 mm × 2 mm. The monochromator oscillation frequency was 1 Hz in this case. In both cases a PIPS detector was used for fluorescence intensity measurements. The intensity of the incident radiation was monitored by a gas chamber filled with nitrogen.

For measurements in QXAFS mode, calibration of the acquired data is essential. For this purpose, we mounted on the surface of our cell a thin CuO pellet, the fluorescence signal from which was used as a reference. Each scan began with the pellet being in the beam, then the sample stage was moved to bring our sample into the beam. To align the spectra, the energy scale for the collected QXAFS spectra was then shifted so that the maximum of the first derivative for the first QXAFS spectrum, which corresponds to the signal from the CuO pellet, is at the same energy as the maximum of the first derivative for the standard CuO spectrum, collected independently at the same beamline. Data calibration was performed using dedicated beamline-specific software (Pro-QEXAFS at SuperXAS, and JAQ at P64)39. Further data alignment, background subtraction, normalization, averaging and linear combination fitting of XANES spectra were performed using a set of Wolfram Mathematica scripts. EXAFS data extraction from averaged data was performed using Athena software40.

EXAFS data fitting for the first coordination shell was performed using the FEFFIT code40. Theoretical phases and amplitudes were obtained in self-consistent ab initio calculations with FEFF8.541 for bulk copper, Cu2O and CuO materials. The complex exchange-correlation Hedin–Lundqvist potential and default values of muffin-tin radii as provided within the FEFF8.5 code were employed.

We started with single-shell fitting of the copper foil and Cu2O data to obtain the values of the amplitude reduction factors \(S_0^2\). The obtained values were used later for the fitting of the experimental EXAFS data for nanocatalysts. Fitting of spectra for reference materials was carried out in the same ranges in k- and R-spaces as the ones later used for the nanocatalysts, to partially compensate for systematic errors due to the limited signal length in k-space. Fitting of EXAFS spectra χ(k)k2 thus was carried out in R-space in the range from Rmin = 1.1 Å up to Rmax = 2.8 Å. Fourier transform was carried out in the k range from 2.0 Å−1 up to 11.5 Å−1. Fitting parameters were the coordination numbers N and interatomic distances R and disorder factors σ2 for Cu–O and Cu–Cu bonds, and the corrections to photoelectron reference energies ΔE0.

Operando high-energy XRD measurements and data analysis

Operando time-resolved high-energy XRD experiments were conducted at the Swedish Materials Science Beamline (P21.2) at Petra III. P21.2 is an undulator beamline equipped with a cryogenically cooled double Laue Si(111) monochromator. The X-ray energy was set to 67 keV (λ = 0.18505 Å) and a working distance of 0.895 m was calibrated using the diffraction pattern of a CeO2 standard. The diffraction patterns were recorded using a Dectris Pilatus X CdTe 2 M, which is a large-area detector especially suitable for time-resolved high-energy XRD experiments. The acquisition time of the individual detector images was set to 10 s under stationary conditions and to 0.127 s for the time-resolved pulse experiments. Primary data treatment (calibration, masking, detector image integration) was conducted using the software package pyFAI42.

A home-made electrochemical cell based on a three-electrode configuration using a leak-free miniature Ag/AgCl reference electrode (3.4 M KCl, eDAQ ET072) and a platinum counter-electrode43. A depiction of the operando XRD set-up can be found in Supplementary Fig. 2. The electrode potentials were controlled using a Biologic SP-240 potentiostat. The electrode potential and current were recorded together with XRD data to allow precise synchronization of the collected images and the electrode potential. The copper NCs were deposited on the microporous layer of the gas diffusion electrode (Sigracet 29 BC, SGL Carbon) with a loading of 0.3 mgCu cm−2. For the operando high-energy XRD experiments, the sample was mounted and, subsequently, the operando X-ray cell was filled with the electrolyte. For alignment, the cell was centred and the gas diffusion electrode was aligned parallel to the X-ray beam using the beam stop diode. Subsequently, the incident X-ray angle was tuned between 0° and 1° to ensure an optimal Cu2O to graphite Bragg peak ratio, and to ensure a sufficiently small spread of the X-rays on the sample, thus avoiding parasitic peak broadening.

After acquisition of the diffraction pattern under open-circuit conditions, a CO2RR potential (−1 V versus RHE) was applied for at least 90 min to ensure the reduction of the catalyst. CO2 was continuously bubbled through the electrolyte to ensure a stable pH. The electrolyte was frequently exchanged to avoid product accumulation.

The data analysis of the diffraction patterns after integration of the detector images was conducted using a set of GNU Octave scripts to perform background pattern subtraction, peak fitting using Lorentz profiles and averaging of the XRD data collected under pulsed CO2RR. In the latter case, the averaging of the XRD data was conducted after correcting the peak profiles for lo- frequency variations due to, for example, bubble formation.

We limited the analysis to Cu(311), Cu2O(111) and Cu2O(220) Bragg peaks due to their sufficient intensity and minimal overlap with other Bragg peaks. The coherence length (size) of the crystalline domains was estimated from the individual Bragg peaks using the Scherrer equation based on the FWHM with a shape factor of 0.89. The lattice strain was calculated as (d − d0)/d0, where d0 is the corresponding lattice spacing determined for the catalyst under the cathodic pulse from the Bragg peak position. The lattice parameter a of copper was calculated as \(\sqrt {h^2 + k^2 + l^2} d\).

For comparison of pulsed CO2RR conditions realized in different electrochemical cell set-ups, it is critical to ensure that the resistor-capacitor (RC) time constant is shorter than the duration of the studied pulses. In all our cells, we confirmed that this condition if fulfilled and does not limit the studied processes. The RC constants were determined by measuring the double-layer capacitance and the ohmic cell resistance in each cell used upon catalyst pre-reduction. In the H-type cell used for the selectivity measurements, an RC time constant of 38.7 ms (with C = 2.87 mF and R = 13.5 Ω) was determined for catalysts with the highest catalyst loading considered in our study (500 µg cm−2). This value is much shorter than the shortest pulse length applied (0.5 s). Lower catalyst loadings led to similar time constants. In our operando XAS cell, the electrode area is twice as large, and the typical resistance is 20 Ω. For our XRD cell, the resistance is 40–50 Ω. The latter was used with a lower catalyst loading (300 µg/cm²) with an electrode area of 6% of the one used for the XAS measurements.