Cavity-confined Au@Cu₂O yolk-shell nanoreactors enable switchable CH₄/C₂H₄ selectivity

Abstract

The regulation of product selectivity in electrochemical CO₂ reduction (ECO₂R) remains fundamentally constrained by the dynamic equilibrium between intermediate transport and surface coverage. In this study, we report a progress in catalytic architecture through precision-engineered Au@Cu₂O yolk-shell tandem nanoreactors featuring dual-tunable parameters: cavity confinement dimensions and shell thickness gradients. This structural modulation enables dynamic control over both *CO intermediate enrichment and reaction pathway bifurcation. ECO₂R performance evaluations demonstrate significant product selectivity switching at −1.31 V (vs. reversible hydrogen electrode (RHE)). The Faradaic efficiency (FE) for CH₄ exhibits significant architectural dependence, increasing from 43.02% (thick-shell/large-cavity) to 65.54% (medium-dimension) and then decreasing to 23.26% (thin-shell/small-cavity). Conversely, the FE for C₂H₄ demonstrates an inverse structural correlation, improving from 6.68% (medium-dimension) to 38.73% (thin-shell/small-cavity). The spatial domain-limiting mechanism of the yolk-shell structure directly controls the transition between protonation-dominated CH₄ formation and coupling-driven C₂H₄ production. This work establishes a pioneering paradigm for dynamically steering catalytic selectivity through purely geometrical modulation, bypassing traditional compositional tuning limitations, thereby opening avenues for precision design of advanced electrocatalytic systems.

Product-selective regulation of electrochemical carbon dioxide reduction is governed by the dynamic equilibrium between intermediate transport and surface coverage. Here, the authors report a yolk-shell nano-catalyst that achieves dynamic regulation of reduction products.

Introduction

Electrochemical CO₂ reduction (ECO₂R), a pivotal technology for converting greenhouse gases into value-added hydrocarbons, represents a critical pathway toward achieving carbon neutrality and enabling renewable energy storage^1–4. The realization of this vision hinges on the development of electrocatalysts with superior activity and selectivity. Copper (Cu)-based catalysts, renowned for their moderate adsorption strength toward *CO intermediates, remain central to ECO₂R systems^5–7. However, the intricate reaction network on Cu surfaces (such as the competition between *CO hydrogenation and C–C coupling) imposes significant challenges in precisely controlling product selectivity.

Conventional modification strategies, including alloying^8,9, doping^10,11, crystal facet engineering^12,13, and defect engineering^14,15, have demonstrated efficacy in enhancing Faradaic efficiency (FE) for specific products. However, these approaches lack the capacity to dynamically regulate reaction pathways. Dynamic switching to specific products has been possible by manipulating the oxidation state on the surface of Cu nanoparticles¹⁶ and by inducing compressive strain on the surface of Cu nanocrystals¹⁷. However, constructing tandem catalysis between Cu and a second metal is still the most widely used method^18–20. This method can spatially separate CO2 activation (e.g., at Au/Ag sites) from *CO deep conversion (e.g., at Cu sites). However, inherent limitations persist in such open architectures: (1) gas-phase-diffused *CO struggles to accumulate at Cu active sites, and (2) rapid *CO desorption under high CO₂ flux severely restricts hydrogenation or coupling processes.

Recent advances in cavity-structured catalysts have highlighted their unique potential in tailoring reaction microenvironments via spatial confinement^21–23. By physically restricting key intermediates (e.g., *CO), these structures enable directional manipulation of reaction kinetics. While cavity engineering has been shown to enhance product selectivity in Cu-based systems^24–26, existing studies predominantly focus on optimizing selectivity for predefined products, neglecting mechanistic insights into dynamically adjusting *CO coverage through structural parameters (e.g., cavity dimensions, shell thickness) to achieve product switching. Notably, subtle variations in *CO coverage critically dictates the ECO₂R pathway bifurcation between hydrogenation and C–C coupling^27–29. In addition, the transformation of the catalyst structure is usually accompanied by a significant change in the local pH at the catalyst-solution interface during the ECO₂R process, which has a dramatic effect on the selectivity of final products^30,31.

Herein, we proposed a “structure engineering-driven reaction path switching” strategy and designed a series of Au@Cu₂O nanoreactors (Au@Cu₂O NRs) with tunable cavity sizes and shell thicknesses. Structural evolution from thick-shelled large cavities to thin-shelled small cavities (Au@Cu₂O-L → Au@Cu₂O-M → Au@Cu₂O-S) systematically modulated reaction microenvironments, shifting the dominant ECO₂R pathway from *CO hydrogenation to C–C coupling. Further investigation into Au@Cu₂O-M NR loading revealed that excessive catalyst aggregation disrupted selectivity due to inter-nanoreactor crosstalk, underscoring the necessity of spatial isolation for confined catalysis. Through in situ spectroscopy, finite element modeling (FEM) simulations, and density functional theory (DFT) calculations, we elucidated how shell thickness and cavity size governed four pivotal processes: (1) the localized proton supply at the surface of catalysts, (2) CO₂ diffusion kinetics to Au cores, (3) *CO intermediate accumulation/spillover within cavities, and (4) spatial confinement-mediated *CO hydrogenation or coupling. This work establishes a universal paradigm for achieving product-switchable ECO₂R via purely structural modulation–without altering chemical composition–and advances the development of adaptive electrocatalytic systems for renewable energy applications.

Results

Synthesis and characterizations

Au@Cu₂O NRs were synthesized through a two-step methodology. Initially, uniform decahedral gold nanocrystals (Au NCs) with an average size of 77.4 ± 1.2 nm were fabricated via a one-pot polyol reduction process (Supplementary Fig. S1)³². The successful preparation of Au NCs, as well as the predominant exposure of (111) plane was verified by X-ray diffraction (XRD) analysis (Supplementary Fig. S2). As depicted in Fig. 1, during the subsequent preparation stage, the as-synthesized Au NCs were dispersed in a mixed aqueous solution containing Cu(NO₃)₂·3H₂O and (NH₄)₂SO₄, followed by dropwise addition of NaOH. The introduction of (NH₄)₂SO₄ effectively suppressed the spontaneous nucleation of Cu(OH)₂, facilitating the reverse conversion to [Cu(H₂O)₆]²⁺ complexes³³ that preferentially adsorbed onto Au NCs as nucleation sites. This directed the heterogeneous growth of Cu(OH)₂ aggregates on the surfaces of Au NCs. Subsequent reduction by L-ascorbic acid rapidly transformed the surface-bound Cu(OH)₂ into Cu₂O. Given the fast reaction kinetics, the initial product predominantly comprised metastable spherical aggregates of Cu₂O composed of numerous small crystallites and amorphous phases. Notably, the interior microcrystalline/amorphous phases exhibited higher surface energy than their exterior counterparts, rendering them more susceptible to dissolution³⁴. Under continuous mechanical stirring, Ostwald ripening³⁵ process was significantly accelerated in the presence of NH⁴⁺ ions, leading to progressive dissolution of internal particles and their recrystallization into larger, well-defined crystalline phases within the outer shell. This mechanism ultimately induced cavity formation. To precisely control the structural evolution, the reaction was deliberately terminated at 13 min to prevent inadequate growth of Cu₂O shells due to insufficient reaction time (Supplementary Fig. S3a), as well as excessive maturation and collapse of Cu₂O shells due to prolonged reactions (Supplementary Figs. S3c, d).

Fig. 1. Schematic illustration of the synthesis of Au@Cu₂O-x NRs.

Different amounts of Au NCs were dispersed in a mixed aqueous solution containing Cu(NO₃)₂·3H₂O and (NH₄)₂SO₄, and NaOH was slowly added. Subsequently, L-ascorbic acid was added to convert the Cu(OH)₂ generated during the reaction into Cu₂O. Under continuous mechanical stirring, NH₄⁺ accelerated the Ostwald ripening process and formed a hollow structure.

Varying the addition of Au NCs profoundly influenced the density of Cu(OH)₂ nucleation sites, thereby dictating the final dimensions of Au@Cu₂O NRs. Specifically, the Au@Cu₂O-L NCs with minimal Au NCs addition (0.1 mg) exhibited limited nucleation sites, resulting in pronounced overgrowth of Cu(OH)₂ on individual Au core. This yielded nanoreactors with an average diameter of 416.3 ± 5.3 nm (Supplementary Figs. S4c, d). In contrast, control experiments without Au NCs produced significantly larger Cu₂O nanospheres (Cu₂O NSs; 789.9 ± 6.0 nm, Supplementary Figs. S4a, b and S5). Gradual increases in Au NCs addition to 0.5 and 0.9 mg decreased the average sizes of Au@Cu₂O NRs to 284.9 ± 1.8 (Au@Cu₂O-M NRs) and 206.1 ± 3.1 nm (Au@Cu₂O-S NRs), respectively (Supplementary Figs. S4e–h). We further hypothesize that variations in Au NCs addition concurrently modulate both cavity dimensions and Cu₂O shell thickness, as will be quantitatively demonstrated below.

Scanning electron microscopy (SEM) images (Fig. 2a–c) demonstrate the spherical morphology and significant size uniformity of Au@Cu₂O-x NRs. The characteristic yolk-shell architecture was confirmed through combined high-angle annular dark-field scanning transmission electron microscopy (HAADF-STEM) images and energy-dispersive X-ray (EDX) elemental mapping (Fig. 2d–f). Spatial resolution of elemental distributions revealed that Au signals localized in the core region, homogeneous the elements of Cu and O distribution throughout the shell, and well-defined cavities between core and shell components. Transmission electron microscopy (TEM) analysis (Fig. 2g–i and Supplementary Figs. S6–S8) provided statistically significant cavity dimensions and shell thicknesses. Quantitative microstructural analysis validated our hypothesis regarding Au addition-dependent structural parameters. Specifically, the Cu₂O shell thickness decreased systematically from 111.3 ± 5.0 (Au@Cu₂O-L NRs) to 66.6 ± 5.1 (Au@Cu₂O-M NRs) and 46.1 ± 3.7 nm (Au@Cu₂O-S NRs), while corresponding cavity sizes diminished from 192.2 ± 8.9 to 151.2 ± 7.9 and 112.6 ± 5.8 nm, respectively. These trends originate from differential Ostwald ripening extents. Higher Au additions promote distributed nucleation, limiting individual particle growth time before ripening termination.

Fig. 2. Microstructure of Au@Cu₂O-x NRs.

SEM, EDX mapping images and TEM of (a, d, g) Au@Cu₂O-L NRs, (b, e, h) Au@Cu₂O-M NRs, and (c, f, i) Au@Cu₂O-S NRs.

HAADF-STEM images of the Au@Cu₂O-M NRs show (Supplementary Fig. S9) that the shells exhibit a loose and highly porous structure. Furthermore, porous channels traversing the shell layer are prevalent throughout the Au@Cu₂O-M NRs (Supplementary Figs. S9a1–j1). This permeable shell facilitates the diffusion of reactants and products, enabling CO₂ entry and the spillover of reaction products. To further demonstrate the permeation of CO₂ by the shell layer of Au@Cu₂O-x NRs, we compared the CO₂ adsorption isotherms (Supplementary Fig. R11) of commercial solid Cu nanoparticles (Supplementary Fig. S10a) with those of Au@Cu₂O-M NRs. The results demonstrate that the CO₂ adsorption capacity of Au@Cu₂O-M NRs reached 32.57 cm³ g⁻¹, exceeding that of the commercial solid Cu nanoparticles (10.49 cm³ g⁻¹) by more than threefold. Given that both materials possess comparable average sizes (commercial solid Cu nanoparticles: 283.8 ± 20 nm; Au@Cu₂O-M NRs: 284.9 ± 1.8 nm) and exhibit smooth spherical morphologies, the significantly enhanced adsorption capacity indicates that a substantial portion of CO₂ molecules permeated the shell layer and entered the internal cavities of the Au@Cu₂O-M NRs. These results create prerequisites for the subsequent cavity-confined tandem catalytic process of Au@Cu₂O-x NRs.

High-resolution TEM (HR-TEM) analysis of a representative Au@Cu₂O-M NRs (Fig. 3a and Supplementary Fig. S12) revealed lattice spacings of 0.246, 0.213, and 0.151 nm in the shell region, corresponding to the (111), (200), and (220) planes of cubic Cu₂O^36,37, respectively. XRD patterns (Fig. 3b) exhibited diffraction peaks assignable to both Cu₂O and Au. The gradual attenuation of Au peak intensities from Au@Cu₂O-S to Au@Cu₂O-L specimens directly correlates with decreasing Au content and concomitant shell thickening, consistent with our structural measurements.

Fig. 3. Phase composition of Au@Cu₂O-x NRs before and after reduction.

a HR-TEM images of Au@Cu₂O-M NRs. The red, green and blue frames are the (200), (220) and (111) facets of Cu₂O, respectively. b XRD patterns. c O 1 s, (d) Cu 2p, (e) Cu LMM, and (f) Au 4 f XPS spectra. In situ Raman spectra of (g) Au@Cu₂O-L NRs, (h) Au@Cu₂O-M NRs and (i) Au@Cu₂O-S NRs from 250–600 cm⁻¹. All potential values were recorded without iR correction. Source data are provided as a Source Data file.

X-ray photoelectron spectroscopy (XPS) survey spectra (Supplementary Fig. S13) confirmed exclusive presence of Cu, O, and C elements, with no detectable Au signal. Considering that the thickness of the Cu₂O shell is greater than 40 nm across all specimens and the characteristic 5–10 nm XPS sampling depth^38,39, this observation demonstrates complete encapsulation of the Au cores. This is also evidenced by the absence of significant Au 4 f peaks for all catalysts in Fig. 3f. Deconvolution of O 1 s spectra (Fig. 3c) resolved two components: lattice oxygen in Cu-O bonds (530.5 eV) and physiosorbed H₂O (533.6 eV)⁴⁰. High-resolution Cu 2p spectra (Fig. 3d) exhibited characteristic satellite peaks at 935.1 and 955.1 eV, indicative of partial Cu²⁺ presence⁴¹. However, Cu LMM Auger analysis (Fig. 3e) confirmed Cu⁺ as the predominant oxidation state (Supplementary Table S1), with Cu²⁺ possibly arising from superficial air oxidation.

It is well known that Cu₂O is inherently unstable under prolonged reductive conditions. To identify the active components during ECO₂R, we examined the morphological and compositional evolution of Au@Cu₂O-M-100 NRs after 1 h of electrolysis at −1.31 V (vs. RHE). Notably, varying catalyst loading (up to 100 μg cm⁻²) primarily influenced product selectivity without altering phase morphology or composition, as elaborated later. Post-reduction characterization revealed that the Au@Cu₂O-M-100 NRs retained their initial morphology without structural remodeling or Au cores exfoliation (Supplementary Figs. S14 and S15). XRD analysis (Supplementary Fig. S16) confirmed the emergence of a Cu phase with preferential (111) orientation, accompanied by the disappearance of the Cu₂O diffraction peaks, demonstrating that the shells of the reduced Au@Cu₂O-M-100 NRs consist of Cu⁰. XPS further corroborated the reduction of Cu²⁺ to Cu⁰, as evidenced by the disappearance of Cu²⁺ peaks at 935.1 and 955.1 eV (Supplementary Fig. S17b). Residual weak satellite peaks and Cu LMM Auger spectra (Supplementary Fig. S17c) suggested partial oxidation of Cu⁰ to Cu⁺ and Cu²⁺ upon air exposure. HR-TEM (Supplementary Fig. S18) revealed a lattice spacing of 0.209 nm, consistent with the Cu(111) plane⁴². These findings collectively demonstrate the rapid reduction of the Cu₂O shell to Cu⁰ with (111) facet orientation during ECO₂R, while the structural integrity of the nanoreactors remained intact, ensuring sustained catalytic performance.

The reduced Cu state of Au@Cu₂O-x NRs during ECO₂R was further confirmed by in situ Raman spectroscopy (Supplementary Fig. S19). The characteristic Cu₂O band at 528 cm⁻¹ diminished under negative potentials and vanished at −1.6 V (vs. RHE) (Fig. 3g–i). Time-resolved in situ Raman spectroscopy of Au@Cu₂O-M NRs was further performed at a constant potential of −1.4 V (V vs. RHE) (Supplementary Fig. S20). When the reaction was carried out for 30 min, the characteristic peak of Cu₂O completely disappeared, indicating that Cu₂O was deeply reduced to Cu⁰ during the ECO₂R⁴³, consistent with the XRD and HR-TEM results (Supplementary Figs. S16 and S18). The results further demonstrated that the residual Cu⁺ signals detected by XPS (Supplementary Figs. S17b, c) originated from rapid surface oxidation upon air exposure, aligning with prior reports^44,45.

ECO₂R performance

ECO₂R was carried out using an H-cell. The performance assessment was operated in 0.5 M KCl by chronopotentiometry and products were analyzed using gas chromatography (GC) (Supplementary Figs. S21 and S22). To elucidate the cavity confinement effects, the catalyst loading on gas diffusion layers (GDLs) was deliberately reduced to 10 µg cm⁻² (denoted as Au@Cu₂O-x-10 NRs), effectively minimizing interparticle crosstalk. Preliminary evaluation of bare Au NCs (Supplementary Fig. S23 and Table S2) revealed exclusive H₂ and CO production, with CO faradaic efficiency (FE_CO) reaching 80% at −1.12 V (vs. RHE). Notably, FE_CO maintained above 55% across the tested potential range (−1.12 to −1.45 V (vs. RHE)), confirming the intrinsic capability of Au as a CO generator for downstream hydrogenation processes. Electrochemically active surface areas (ECSAs) were quantified via cyclic voltammetry-derived double-layer capacitance (C_dl) measurements (Supplementary Fig. S24). The calculated C_dl values followed the sequence: Au@Cu₂O-L-10 NRs (1.22 mF cm⁻²) > Au@Cu₂O-M-10 NRs (1.19 mF cm⁻²) > Au@Cu₂O-S-10 NRs (1.04 mF cm⁻²) > Cu₂O NSs (0.96 mF cm⁻²; Supplementary Fig. S25), indicating superior active site densities of Au@Cu₂O-x-10 NRs⁴⁶. Linear sweep voltammetry (LSV) tests were performed in a CO₂-saturated electrolyte (Supplementary Fig. S26). The catalysts exhibited similar ECO₂R response current densities. At high reduction potentials, Au@Cu₂O-x-10 NRs exhibited higher current densities than Cu₂O NSs-10, suggesting that they may have higher ECO₂R activity⁴⁰.

The ECO₂R capability of Au@Cu₂O-x-10 NRs was tested (Supplementary Table S2). Au@Cu₂O-M-10 NRs exhibited high CH₄ selectivity, achieving 65.54% FE_CH4 at −1.31 V (vs. RHE) with a partial current density of −19.77 mA cm⁻² (Fig. 4b and Supplementary Fig. S27c). This catalyst maintained above 60% FE_CH4 within the −1.22 to −1.39 V potential window. During 10 hours stability testing at −1.31 V (vs. RHE), total current density remained stable while FE_CH4 persisted above 50% (Fig. 4d). Post-test SEM characterization (Supplementary Fig. S28) revealed structural degradation in deactivated catalysts (red circles), where collapsed yolk-shell architectures likely caused intermediate leakage and subsequent FE_CH4 decline. A detailed comparison demonstrates that Au@Cu₂O-M-10 NRs outperforms for ECO₂R to CH₄ compared to previously reported catalysts (Supplementary Fig. S29 and Table S3). At −1.31 V (vs. RHE), the FE_CH4 of Au@Cu₂O-L-10 and Au@Cu₂O-S-10 NRs reached 43.02% and 23.26%, respectively (Fig. 4a, c), corresponding to CH₄ partial current densities of −13.04 and −7.15 mA cm⁻². These two catalysts exhibited significantly different product selectivity compared to Au@Cu₂O-M-10 NRs. Specifically, Au@Cu₂O-L-10 NRs demonstrated more vigorous hydrogen evolution reaction (HER) activity, while Au@Cu₂O-S-10 NRs showed higher selectivity toward C₂H₄ and CO production. At −1.22 V (vs. RHE), the FE_H2 of Au@Cu₂O-L-10 NRs reached 26.81%, corresponding to a partial current density of −5.17 mA cm⁻² (Supplementary Fig. S27a). In contrast, the FE_H2 values for Au@Cu₂O-M-10 and Au@Cu₂O-S-10 NRs were 14.29% and 16.18%, respectively, with corresponding partial current densities of −2.68 and −3.25 mA cm⁻². The Au@Cu₂O-S-10 NRs achieved FE_C2H4 and FE_CO of 38.73% and 10.44%, respectively, corresponding to partial current densities of −7.52 and −2.22 mA cm⁻² (Supplementary Figs. S27b, d). In comparison, the FE_C2H4 values for Au@Cu₂O-L-10 and Au@Cu₂O-M-10 NRs were 8.45% and 6.68%, respectively, with partial current densities of −1.57 and −1.30 mA cm⁻². The corresponding FE_CO values were 10.44% and 9.00%, with partial current densities of −1.98 and −1.80 mA cm⁻². Figure 4e visually compares the FE ratios (FE_(CH4/C2H4)) of CH₄ to C₂H₄ for the four catalysts across different potentials. At −1.45 V (vs. RHE), Au@Cu₂O-M-10 NRs exhibited the highest FE_(CH4/C2H4) value of 24.38, which was 4.02 and 39.97 times higher than those of Au@Cu₂O-L-10 (6.06) and Au@Cu₂O-S-10 NRs (0.61), respectively. The significant difference in FE_(CH4/C2H4) values strongly suggests that the spatial geometry of the nanoreactor played a crucial role in modulating the selectivity of ECO₂R products. The shell thickness and cavity size had a significant effect on the catalytic activity and product distribution.

Fig. 4. ECO₂R performance of Au@Cu₂O-x NRs.

FEs of various reduction products at^80–93 different potentials tested in an H-cell with 0.5 M KCl electrolyte for (a) Au@Cu₂O-L-10, (b) Au@Cu₂O-M-10, and (c) Au@Cu₂O-S-10 NRs, respectively. d Long-term stability test of Au@Cu₂O-M-10 NRs at −1.31 V (vs. RHE). Gray shading indicates areas with less than 50% FE_CH4. e FE_(CH4/C2H4) over Cu₂O-10 NSs and Au@Cu₂O-x-10 NRs. f A summary of the regulating performance of reported and current catalysts for electrocatalytic products. Both blue (in a flow-cell) and red (in an H-cell) shading indicate that this study is superior to the literature in terms of the ability to switch the CH₄/C₂H₄ product. FEs of various reduction products at different potentials tested in a flow-cell with 1 M KOH electrolyte for (g) Au@Cu₂O-L-10, (h) Au@Cu₂O-M-10, and (i) Au@Cu₂O-S-10 NRs, respectively. The CO₂ flow rate was 20 sccm. Error bars represent the standard deviation of three independent measurements. All potential values were recorded without iR correction. Source data are provided as a Source Data file.

To demonstrate that the unique yolk-shell structure of Au@Cu₂O-x-10 NRs is the primary factor governing product selectivity differences, we prepared pure Cu₂O NSs without Au NCs using the same synthetic method and evaluated their ECO₂R activities. As shown in Supplementary Fig. S30, the product distribution of Cu₂O-10 NSs was consistent with that of Au@Cu₂O-x-10 NRs but exhibited a substantial increase in FE_H2 and a decrease in ECO₂R activity. At −1.31 V (vs. RHE), the FE_H2, FE_CH4, and FE_C2H4 values were 38.46%, 40.16%, and 5.88%, respectively. The ECO₂R performance of Cu₂O-10 NSs closely resembled that of Au@Cu₂O-L-10 NRs but differed significantly from Au@Cu₂O-M-10 and Au@Cu₂O-S-10 NRs, which demonstrated high selectivity for CH₄ and C₂H₄, respectively. Compared to Cu₂O-10 NSs, it suggests that a portion of CH₄ on Au@Cu₂O-M-10 NRs was generated directly on the oxide-derived Cu shells. A portion of its overproduced FE_CH4 (25.38%) may be generated either by the unique yolk-shell structure modulation between the Au cores and the Cu₂O shells or by the interaction between the Au NCs and the Cu₂O NSs. To rule out the latter, Au NCs and Cu₂O NSs were physically mixed at the same Au:Cu molar ratio (1:46.6) as that of Au@Cu₂O-M-10 NRs, resulting in phase-separated catalysts with dispersed Au NCs (indicated by red circles in Supplementary Fig. S31a). As shown in Supplementary Fig. S31b, the phase-separated catalysts exhibited relatively lower FE_CH4 and higher FE_CO, along with pronounced HER activity, compared to Au@Cu₂O-M-10 NRs across the tested potential range. This observation confirms that simple physical mixing of Au NCs and Cu₂O NSs is insufficient to facilitate effective tandem reactions. The unique spatial geometry of the Au cores and Cu₂O shells, along with the dual reaction surfaces on both sides of the shell, collectively contribute to the diversification of the reduction products.

To clarify the structural effects of nanoreactors on product selectivity while eliminating the influence of active site quantity, we calculated the ECSA-normalized CH₄ partial current densities for Au NCs-10, Cu₂O NSs-10, and Au@Cu₂O-M-10 NRs, along with the ECSA-normalized C₂H₄ partial current densities for Au NCs-10, Cu₂O NSs-10, and Au@Cu₂O-S-10 NRs. We defined the performance differential as Δj = j_Au@Cu2O-x-10 − (j_{Cu2O NSs} + j_{Au NCs}). For CH₄ production, Au@Cu₂O-M-10 NRs exhibited superior ECSA-normalized partial current density compared to Cu₂O NSs (Supplementary Fig. S32a). Notably, within the operational potential window of −1.12 to −1.45 V (vs. RHE), Au NCs showed no catalytic activity toward CO hydrogenation, confirming that all CH₄-related current densities originated exclusively from Cu₂O NSs. Similarly, in C₂H₄ generation, Au@Cu₂O-S-10 NRs displayed significantly enhanced ECSA-normalized partial current densities (Supplementary Fig. S32b). Given the inherent inability of Au NCs to facilitate C-C coupling, all ECSA-normalized partial current densities directed to C₂H₄ must likewise originate from Cu₂O NSs-mediated processes. The comparative analysis revealed positive Δj_CH4 and Δj_C2H4 values (Supplementary Figs. S32a, b), demonstrating that both Au@Cu₂O-M-10 and Au@Cu₂O-S-10 NRs outperformed their individual Cu₂O-10 NSs counterparts. This enhanced performance directly correlates with the incorporation of Au cores. The tandem catalytic mechanism inherent to Au@Cu₂O-x-10 NRs enables efficient utilization of CO intermediates generated at Au cores through subsequent reactions on Cu₂O shells. This synergistic interaction significantly amplifies the production of reduced hydrocarbons, particularly enhancing both CH₄ and C₂H₄ yields through distinct structural configurations (Au@Cu₂O-M-10 NRs for CH₄ and Au@Cu₂O-S-10 NRs for C₂H₄ optimization).

Given the use of PVP in catalyst synthesis, we systematically investigated its influence on product selectivity in Au@Cu₂O-x-10 NRs. Control samples including PVP-free Cu₂O NSs (denoted as non-Cu₂O NSs) and PVP-free Au@Cu₂O-x NRs (non-Au@Cu₂O-x NRs) were synthesized for comparative analysis. SEM characterization (Supplementary Fig. S33) revealed significant morphological alterations in PVP-free systems. The non-Cu₂O NSs exhibited severe particle agglomeration with irregular size distribution, while non-Au@Cu₂O-x NRs completely lost their spherical architecture, forming heavily aggregated nanoparticles with no discernible structural variations among the three catalyst types. This structural degradation directly impaired their ECO₂R performance in 0.5 M KCl electrolyte (Supplementary Fig. S34). All four catalysts maintained relatively high methane selectivity. While non-Cu₂O-10 NSs showed comparable product selectivity to standard Cu₂O-10 NSs, non-Au@Cu₂O-x-10 NRs demonstrated significantly increased FE_CO without achieving product selectivity switching. This phenomenon is attributed to the collapse of the tandem catalytic mechanism following the loss of yolk-shell architecture. The non-Au@Cu₂O-x NRs showed only superimposed ECO₂R activities from isolated Au NCs and Cu₂O NSs rather than synergistic effects. These findings conclusively demonstrate the critical role of PVP in controlling the nanocatalyst size, morphology preservation and structural stability during the assembly of Au@Cu₂O-x NRs.

To evaluate the practical application of the Au@Cu₂O-x-10 NRs under high current density conditions, a conventional flow-cell using gas diffusion electrodes was employed to perform ECO₂R tests in 1 M KOH (Supplementary Fig.35 and Table S4). Supplementary Fig.36 showed the LSV curve obtained using the flow cell setup. Beneficial from the improved mass transfer, a much higher current density was achieved. As shown in Fig. 4g–i, the HERs of Au@Cu₂O-x-10 NRs all decreased significantly and maintained the same dynamic switching of the products as in H-cells. Specifically, at a current density of −200 mA cm⁻², the FE_CH4 of Au@Cu₂O-M-10 NRs reached 61.52% and the FE_C2H4 of Au@Cu₂O-S-10 NRs reached 41.43%. Figure 4f summarized the comparison between the present study and the reported ability of the catalysts for selective dynamic switching of specific products (CH₄ and C₂H₄), where Au@Cu₂O-x-10 NRs showed superior CH₄ and C₂H₄ product selective modulation in both H-cells and flow-cells.

In situ characterizations of the catalysts

The coverage of *CO intermediates on the surface of Au@Cu₂O-x NRs were obtained by the Raman signature bands between 2010–2100 cm⁻¹ in the in situ Raman spectrum^47,48. As shown in Fig. 5a-c, distinct *CO complex peaks emerged on all three catalysts as the reduction potential shifted negatively. The *CO signals of Au@Cu₂O-S NRs were markedly stronger than those of the other two catalysts, indicating the highest *CO surface coverage, which facilitates subsequent C–C coupling. Notably, the *CO peak positions differed among the catalysts. To resolve these differences, the *CO peaks at −1.6 V (vs. RHE) were deconvoluted (Supplementary Fig. 37), revealing three components⁴⁹: (1) bridged-CO ( ~ 2015 cm⁻¹), (2) low-frequency linear CO (LFB-CO, ~ 2060 cm⁻¹), and (3) high-frequency linear CO (HFB-CO, ~ 2090 cm⁻¹). Au@Cu₂O-L NRs exhibited the weakest composite peaks, predominantly classified as LFB-CO. The composite peaks of Au@Cu₂O-M NRs and Au@Cu₂O-S NRs were stronger, with a larger proportion of HFB-CO peaks in the former (40.45%) and LFB-CO peaks in the latter (51.44%). Supplementary Fig. S38 compared the values of HFB-CO/LFB-CO at −1.6 V (vs. RHE), and the Au@Cu₂O-M NRs were 1.229, while the Au@Cu₂O-S NRs were only 0.435. Previous studies associate^50–53 LFB-CO with active intermediates for C–C coupling and C₂H₄ formation, while HFB-CO correlates with gaseous CO release. Thus, a higher HFB-CO fraction promotes free *CO availability for methanation, whereas a dominant LFB-CO population drives Au@Cu₂O-S NRs toward C–C coupling. To demonstrate that Au NCs serve as a key *CO generator role in the catalytic process, in situ Raman characteristic bands were investigated within the spectral range of 2010–2100 cm⁻¹. As shown in Supplementary Fig. S39, the Au NCs gradually exhibited *CO peaks attributed to HFB-CO starting from −0.8 V (vs. RHE) and increasing in intensity with more negative potentials. This suggests that Au NCs can serve as an *CO source to provide a large amount of free-state *CO for subsequent tandem catalysis.

Fig. 5. In situ capture of reaction intermediates.

In situ Raman spectra of (a) Au@Cu₂O-L NRs, (b) Au@Cu₂O-M NRs and (c) Au@Cu₂O-S NRs from 1900–2200 cm⁻¹. Fitting of in situ SERS in (d) Au@Cu₂O-L NRs, (e) Au@Cu₂O-M NRs and (f) Au@Cu₂O-S NRs. g The ratio of HCO₃⁻/CO₃²⁻ in Au@Cu₂O-x NRs. In situ ATR-SEIRAS spectra of (h) Au@Cu₂O-L NRs, (i) Au@Cu₂O-M NRs and (j) Au@Cu₂O-S NRs. All potential values were recorded without iR correction. Source data are provided as a Source Data file.

In situ surface-enhanced Raman spectroscopy (SERS) in 1 M KOH further demonstrated nanoreactors control over reaction pathways via local microenvironment modulation and evaluated the HER competition during ECO₂R. As shown in Fig. 5d–f, signals recognized from SERS at ~ 1010 and ~ 1070 cm⁻¹ were ascribed to the adsorbed bicarbonate (HCO₃⁻) and carbonate (CO₃²⁻), respectively. CO₃²⁻ peaks were observed on the surface of all three nanoreactors, but the significant Raman response of HCO₃⁻ was only present in Au@Cu₂O-L NRs and Au@Cu₂O-M NRs, and they showed different peak intensity ratios. In common, during ECO₂R, the acidulous CO₂ will neutralize with electrolyte (KOH) to generate CO₃²⁻ and over-neutralize into HCO₃⁻ (CO₂ + 2KOH → K₂CO₃ + H₂O; K₂CO₃ + H₂O + CO₂ → 2KHCO₃). Thus, the localized proton concentration in solution around catalysts during ECO₂R can be evaluated via monitoring the ratio of adsorbed HCO₃⁻ and CO₃²⁻, where the higher ratio of HCO₃⁻/CO₃²⁻ means the higher localized proton concentration around catalysts^54,55. Quantitative analysis showed (Fig. 5g) that the localized proton concentrations around Au@Cu₂O-L NRs and Au@Cu₂O-M NRs was higher than that of Au@Cu₂O-S NRs within a potential window of −0.2 V to −1.8 V (vs. RHE). The lower proton donor density of the Au@Cu₂O-S NRs created a localized alkaline condition, which contributed to C–C coupling^56–58. In contrast, the higher proton donor density around Au@Cu₂O-L NRs and Au@Cu₂O-M NRs promoted the *CO protonation pathway and accelerated CH₄ production⁵⁹. These results demonstrate the modulation of local pH by the geometric parameters of Au@Cu₂O-x NRs: thick shells restrict proton diffusion, leading to higher H₊ concentration at the catalyst surface. Conversely, thin shells accelerate proton depletion, facilitating the creation of an alkaline local microenvironment. Notably, the local proton concentration of Au@Cu₂O-M NRs was consistently higher than that of Au@Cu₂O-L NRs in the more negative potential interval (−1.0 V to −1.8 V (vs. RHE)), which was correlated with the stronger methanation capacity of the former (Fig. 4b, h). However, the excessively high localized proton concentration also exacerbated HER when *CO coverage was insufficient, which explained the higher selectivity of Au@Cu₂O-L NRs for H₂ (Fig. 4a, g). Subsequently, we will discuss in depth the mechanism by which the nanoreactor structure regulates the *CO coverage.

The evolution of intermediates on the Au@Cu₂O-x NRs during the ECO₂R process was further investigated using in situ attenuated total reflectance surface enhanced infrared absorption spectroscopy (ATR-SEIRAS; Supplementary Fig. S40). As shown in Fig. 5h–j, bands at ~ 1440 cm⁻¹ was attributed to *COOH⁶⁰, a critical precursor for *CO formation⁶¹. The Au@Cu₂O-L NRs exhibited notably lower overall peak intensity (Fig. 5g), indicative of their poor ECO₂R activity, consistent with the dominance of HER observed in Fig. 4a, g. In Fig. 5i, bands at ~ 1060 cm⁻¹, ~1300 cm⁻¹, and ~1730 cm⁻¹ correspond to *CHO^33,62, the key intermediate triggering CH₄ formation via ECO₂R^63,64. Notably, an O–H stretching band at ~1640 cm⁻¹, assigned to water⁶⁵, appeared only in Fig. 5i. This observation, combined with the catalytic performance of the three catalysts, confirms that Au@Cu₂O-M NRs promote H₂O activation, thereby supplying abundant active hydrogen (*H) for *CO methanation⁶⁶. This conclusion is further supported by the intense *OCH₃ peak at 1145 cm⁻¹ (Fig. 5i)⁶⁷, which serves as a critical intermediate for *CHO protonation. In contrast, Au@Cu₂O-S NRs displayed distinct bands at ~ 1230 and ~ 1590 cm⁻¹, attributed to *OCCOH⁵⁴ (Fig. 5j), a potential key intermediate for C₂H₄ generation via C–C coupling^68,69. The absence of these bands in Au@Cu₂O-M NRs aligns with their poor C–C coupling activity, consistent with ECO₂R selectivity trends. Additionally, a *COCOH peak at ~ 1180 cm^−1 70 was observed across all three catalysts, suggesting its limited role in C₂H₄ production. Combining the ECO₂R activities of Au@Cu₂O-x-10 NRs and the above spectroscopic results, we propose the following generation pathways for CH₄ and C₂H₄:

$${\mathrm{CH}}_4:{\mathrm{CO}}_2{\to }^{*}\mathrm{COOH}{\to }^{*}\mathrm{CO}{\to }^{*}\mathrm{CHO}{\to }^{*}{\mathrm{OCH}}_3\to {\mathrm{CH}}_4.$$

$${\mathrm{C}}_2{\mathrm{H}}_4:{\mathrm{CO}}_2{\to }^{*}\mathrm{COOH}{\to }^{*}\mathrm{CO}{\to }^{*}\mathrm{OCCO}{\to }^{*}\mathrm{OCCOH}\to {\mathrm{C}}_2{\mathrm{H}}_4.$$

Mechanistic Insights into the Local *CO Flux Modulated ECO₂R

The results of in situ spectroscopy demonstrated that the differences in the reduction products of Au@Cu₂O-x NRs originated from distinct intermediates generated during the ECO₂R process. However, the fundamental mechanism governing these key intermediate variations remained elusive. Because the three nanoreactors have same physical phase compositions, the difference in product selectivity can be attributed to the spatial geometry effects of catalysts.

To elucidate the CO₂ transfer limitation effect of Au@Cu₂O-x-10, we conducted the galvanostastic step experiments (Supplementary Fig. S41)⁷¹. The current was instantly stepped from a reductive current to an oxidative one. The appeared characteristic platform correlates with the oxidation of the localized intermediates⁷². When the intermediates are more locally concentrated, the platform lengthens accordingly. Au@Cu₂O-M-10 NRs did not show a characteristic platform under Ar atmosphere (Supplementary Fig. S41a), indicating that the reaction intermediates of ECO₂R did not appear at this time. As shown in Supplementary Fig. S41b, the galvanostastic step curves under CO₂ atmosphere showed a longer platform as the thickness of the Au@Cu₂O-x-10 shell decreased. Therefore, we hypothesize that the thicker shells hinder the CO₂ transfer to the internal cavity, which subsequently affects the concentration of the reaction intermediates.

FEM simulations were performed to investigate how variations in shell thickness and cavity size of Au@Cu₂O-x NRs influence ECO₂R product distributions. As illustrated in Supplementary Fig. S42, three planar models with distinct Cu shell thicknesses (blue regions) and cavity dimensions (gray regions) were constructed based on the structural data from Fig. 2g–i and Supplementary Figs. S6–S8, corresponding to the three Au@Cu₂O-x NRs. While previous discussions have established the catalytic activity of the oxide-derived metallic Cu shells/sites in ECO₂R, our computational model specifically examines two critical aspects: (1) the kinetic behavior of CO generation within Au cores, and (2) the confinement effects inherent to yolk-shell architectures. The proposed reaction pathway follows three sequential stages: First, CO₂ molecules permeate through the derived Cu shell to access Au cores, where adsorption and subsequent reduction to *CO intermediates occur. Second, *CO desorbs from Au surfaces with two possible trajectories: either undergoing cavity-mediated enrichment and secondary reactions, or directly diffusing outward through the shell to form gaseous CO. Though this modeling approach accentuates the Cu-Au tandem synergy by emphasizing intermediate confinement and transport dynamics, it intentionally preserves the fundamental catalytic mechanisms and observed trends. This strategic simplification allows focused investigation of structure-activity relationships without compromising the validity of mechanistic interpretations. Spatial distributions of CO₂ and *CO across the three Au@Cu₂O-S NRs are shown in Fig. 6a–c. For Au@Cu₂O-L NRs (thickest shells), restricted CO₂ diffusion through the dense shell layer limited reactant availability at the Au cores (Fig. 6a), leading to minimal *CO enrichment in the cavity (Fig. 6d) and the lowest ECO₂R activity (Fig. 4a, g), which corroborates the results of Supplementary Fig. S41. Conversely, Au@Cu₂O-S NRs (thinnest shells) allowed rapid CO₂ diffusion (Fig. 6c), resulting in substantial *CO accumulation within the cavity and partial spillover through the shell (Fig. 6f). This high *CO concentration promoted C–C coupling, elevating FE_C2H4, while spillover increased FE_CO, aligning with Fig. 4c, i. Au@Cu₂O-M NRs (intermediate shells) balanced CO₂ diffusion (Fig. 6b) and *CO retention (Fig. 6e). Their moderate cavity size maintained sufficient *CO levels for hydrogenation without triggering C–C coupling, thereby maximizing FE_CH4 (Fig. 4b, h).

Fig. 6. Simulation calculations reveal reaction mechanisms.

Computed concentration distribution of CO₂ and CO species of (a, d) Au@Cu₂O-L, (b, e) Au@Cu₂O-M, and (c, f) Au@Cu₂O-S. Concentrations are shown by color bar in millimoles. g nonlinear scaling relations between *CO coverage and formation energy of *CHO and *OCCOH on Cu(111) surface. Gibbs free energy profiles for the (h) CH₄ and (i) C₂H₄ generation pathways on the Cu(111) surface with a *CO coverage of 3/16 and 6/16 ML. Insets are the atomistic structure of *CHO and *OCCOH under various *CO coverages. Color codes: Cu, orange; C, gray; O, red; H, pink. (j) Schematic diagram of the cavity-confined catalytic mechanism of Au@Cu₂O-x NRs. Source data are provided as a Source Data file.

DFT calculations were further employed to elucidate the transport trends of CO₂ gas and key intermediates through the Cu shells in Au@Cu₂O-x NRs, and to mechanistically reveal the product-selective modulation related to *CO coverage. Previous characterization of pore channels and CO₂ adsorption isotherms indicated that gas can freely access both the shell and core of the nanoreactor. However, the intrusion hindrance of Cu shells to CO₂ gas and the active retention ability of CO seepage are still to be further verified. Therefore, we calculated the adsorption energy (E_ads) of CO₂ and CO on Au(111) and Cu(111) surfaces, respectively (Supplementary Fig. S44). The results show that the E_ads of CO₂ on Au(111) (−0.21805 eV) was more negative than on Cu(111) (−0.20904 eV). This energy difference drove CO₂ migration towards the Au cores under the influence of the concentration gradient. Similarly, the E_ads of CO on Cu(111) (−0.64788 eV) was more negative than on Au(111) (−0.48549 eV). Consequently, adsorbed *CO desorbed and diffused outward towards the shell layer, following the gradient of reduced adsorption energy. This bidirectional selective diffusion facilitated the *CO hydrogenation and C–C coupling processes. These results further demonstrate the theoretical feasibility of the cavity-confined domain tandem catalytic model within the yolk-shell structure of Au@Cu₂O-x NRs.

Based on HR-TEM (Supplementary Fig. S18) with in situ Raman spectroscopic characterization (Fig. 3g–i), Cu(111) surface models with different *CO coverage (2/16, 3/16, 4/16, 5/16, 6/16 ML) were constructed (Supplementary Fig. S45). Figure 6g revealed a distinct nonmonotonic relationship between *CO coverage and intermediate formation energies. The formation energy (E_formation) of *CHO intermediates exhibited a characteristic pattern of significant initial decrease followed by subsequent increase as *CO coverage increased from 2/16 to 6/16 ML. Conversely, the E_formation of *COCOH intermediates demonstrated an inverse trend of increasing and then decreasing. These computational results elucidated two distinct catalytic regimes: under low *CO coverage conditions (2/16–3/16 ML), the preferential stabilization of *CHO intermediates (reaching a minimum E_formation of 0.282 eV at 3/16 ML, corresponding to Au@Cu₂O-M NRs) thermodynamically favored subsequent methanation pathways. When *CO coverage exceeded 4/16 ML (characteristic of Au@Cu₂O-S NRs), we observed enhanced thermodynamic stability of the C-C coupling pathway through *COCOH intermediates, with E_formation values decreasing from 0.675 eV to 0.590 eV with increasing coverage. This *CO coverage-dependent energetic preference directly correlated with the experimentally observed product selectivity shift, i.e., optimal methanation activity at intermediate coverages versus predominant C₂H₄ production through enhanced *CO dimerization capacity at higher coverages. Based on in situ ATR-SEIRAS spectroscopic analyses (Fig. 5h–j) and the evolution of E_formation for key intermediates (Fig. 6g), we further constructed free energy (ΔG) step diagrams to quantitatively compare CH₄ and C₂H₄ generation pathways under low (3/16 ML) versus high (6/16 ML) *CO coverages. The adsorption configurations of the reaction intermediates were displayed in Supplementary Figs. S46–49. For the methanation pathway (Fig. 6h), the ΔG barrier of the *OCH₃ → *CH₄ step exhibited a dramatic increase from 2.55 (3/16 ML) to 4.44 eV (6/16 ML), indicating that reduced *CO coverage thermodynamically favored CH₄ formation. Conversely, in the C₂H₄ synthesis pathway (Fig. 6i), both critical steps demonstrated progressive ΔG reductions with increasing coverage: The *CO → *OCCO coupling barrier decreased from 1.02 to 0.85 eV, while the subsequent *OCCOH → *C₂H₄ barrier declined from 2.31 to 2.06 eV. This dual-barrier attenuation mechanism demonstrated that elevated *CO coverage promoted C₂H₄ generation through enhanced CO dimerization kinetics and stabilized C-C coupling intermediates.

The consistency between these computational predictions and experimental selectivity measurements strongly supported our proposed mechanism of *CO coverage-mediated pathway control in Au@Cu₂O-x NRs. Combined FEM and DFT analyses produced the mechanistic diagram in Fig. 6j, illustrating how cavity-confined catalytic mechanism regulate *CO spillover, hydrogenation, and dimerization pathways.

This study maintained a constant Au@Cu₂O-x NR loading of 10 µg cm⁻² to mitigate interparticle interactions. To assess the effects of loading on selectivity, a gradient series (1, 10, 50, 100, 250, 500, 750, and 1000 µg cm⁻²) was tested using Au@Cu₂O-M NRs. At 1 µg cm⁻², SEM imaging on GDL substrates failed to resolve individual Au@Cu₂O-M NR due to background interference (Supplementary Fig. S50a). Switching to silicon wafers revealed sparse dispersion of Au@Cu₂O-M NRs (Supplementary Fig. S50b). At 10 µg cm⁻², Au@Cu₂O-M NRs were uniformly distributed on GDL with micrometer-scale spacing (Supplementary Fig. S50c), ensuring isolated catalytic domains. Higher loadings induced progressive agglomeration (Supplementary Figs. S50d–i), compromising particle independence.

LSV curves in CO₂-saturated electrolyte (Supplementary Fig. S51) showed increasing current density with loading, reflecting increased number of active sites. Product FE analysis (Supplementary Fig. S52 and Table S5) revealed that at 1 µg cm⁻², selectivity mirrored that of Au@Cu₂O-M-10 NRs, with 55.82% FE_CH4 at −1.31 V (vs. RHE). However, excessive GDL exposure intensified HER (FE_H2 = 33.34%; Supplementary Fig. S52a). The FEs of Au@Cu₂O-M-y NRs with different loadings toward CO, CH₄ and C₂H₄ were intuitively illustrated by the fitted 3D colormap (Fig. 7a). Examination of the 3D colormap projection along the FE axis (Fig. 7b) confirmed that Au@Cu₂O-M-y NRs exhibited high FE_CH4 ( ≥ 50%) at −1.31 V (vs. RHE) for loadings ≤ 50 µg cm⁻², indicating dominance of tandem catalysis by individual Au@Cu₂O-M NR. Above 100 µg cm⁻², agglomeration (Supplementary Figs. S50e-g) weakened individual Au@Cu₂O-M NR effect, shifting selectivity toward CO and C₂H₄. Beyond 500 µg cm⁻², clustered Au@Cu₂O-M NRs (Supplementary Figs. S50h, i) eliminated single-particle behavior, resulting in FE_C2H4 exceeding 24% at −1.22 V (vs. RHE) due to inter-shell interactions forming new active sites. These results demonstrate that isolated Au@Cu₂O-x NRs critically govern product selectivity, while low loadings suppress HER and optimize intrinsic catalytic performance.

Fig. 7. Effect of catalyst loadings on product selectivity.

a 3D colormap surface plot and (b) corresponding color maps for FEs towards different products (i.e., C₂H₄, CO and CH₄) versus mass loadings of Au@Cu₂O-M-y NRs and applied cathodic potentials. The CO₂ flow rate was 20 sccm. All potential values were recorded without iR correction. Source data are provided as a Source Data file.

Discussion

In summary, we demonstrated a structural engineering strategy of catalysts for ECO₂R through the rational design of Au@Cu₂O NRs. The precisely constructed architecture with Au cores and Cu₂O shells enabled dynamic modulation of product selectivity via spatial confinement effects. In situ spectroscopic analysis demonstrated that the unique yolk-shell structures of Au@Cu₂O-x NRs could modulate the localized proton supply at the surface of catalysts and influence the generation of reaction intermediates, thus enabling the targeted modulation of ECO₂R products. FEM simulations and DFT calculations confirmed that tailoring shell thickness and cavity dimensions governed CO₂ permeability while regulating *CO coverage within the confined spaces, thereby steering the ECO₂R pathway. By strategically increasing shell thickness (46.1 ± 3.7 → 66.6 ± 5.1 nm) and cavity size (112.6 ± 5.8 → 151.2 ± 7.9 nm), we achieved restricted C–C coupling process that boosted FE_CH4 from 23.26% to 65.54% at −1.31 V (vs. RHE), accompanied by a marked decrease in FE_C2H4 (33.87% → 6.51%). Excessive geometric constraints (shell thickness of 111.3 ± 5.0 nm, and cavity size of 192.2 ± 8.9 nm) impeded CO₂ diffusion, shifting the dominant process to HER (FE_H2 = 28.63%). Furthermore, loading-dependent experiments unveiled two operational regimes: single-particle confinement prevailed at low catalyst loadings (≤ 50 µg cm⁻², FE_CH4 > 50%), whereas interparticle communication at high loadings (> 500 µg cm⁻²) promoted C–C coupling (FE_C2H4 > 24%). This study established quantitative correlations between nanoreactor geometry (shell thickness/cavity size), intermediate surface coverage, and product selectivity, providing fundamental insights into spatial confinement mechanisms that governed electrocatalytic pathways.

Methods

Chemicals

Hydrogen tetrachloroaurate(III) trihydrate (HAuCl₄·3H₂O, 99.9%), polyvinylpyrrolidone (PVP, Mw = 58,000), L-ascorbic acid (C₆H₈O₆, ≥99.7%), and copper nitrate trihydrate (Cu(NO₃)₂·3H₂O, 99%) were purchased from Shanghai Aladdin Bio-Chem Technology Co., Ltd. Ammonium sulfate ((NH₄)₂SO₄, AR), Diethylene glycol (DEG, C₄H₁₀O₃, AR), ethanol (C₂H₆O, AR), Sodium hydroxide (NaOH, AR), and potassium chloride (KCl, AR) were purchased from Sinopharm Chemical Reagent Co., Ltd. Nafion solution (5 wt.%) and Nafion-117 perfluorinated membrane were sourced from Alfa Aesar Chemical Co., Ltd. and DuPont Company, respectively. Carbon dioxide (CO₂, ≥ 99.999%), Nitrogen (N₂, ≥ 99.99%) and argon (Ar, ≥ 99.999%) were supplied by Shaanxi Kuaite Gas and Equipment Co., Ltd. All chemicals were used as received without further purification. Deionized water (18.2 MΩ cm) was prepared using an EPED-S2-D water purification system.

Synthesis of decahedral Au nanocrystals (Au NCs)

The synthesis of decahedral Au NCs followed a modified protocol from Seo et al.³² Briefly, 2.0 g of PVP was dissolved in 25 mL of DEG under reflux for 5 min. Subsequently, 2.0 mL of a DEG solution containing 20 mg of HAuCl₄·3H₂O was injected into the boiling mixture, followed by refluxing for 10 min. After cooling to room temperature (25 ± 5 °C), the product was diluted with ethanol, centrifuged at 6000 rpm for 30 min, and washed four times with ethanol to yield a homogeneous dispersion of decahedral Au NCs. The final dispersion was stored at 4 °C for subsequent use.

Synthesis of Au@Cu₂O Nanoreactors (Au@Cu₂O NRs)

In a typical synthesis³⁴, different amounts of the prepared Au NCs (0.1 mg, 0.5 mg, 0.9 mg) were dispersed in 50 ml of 2 mM aqueous PVP solution under vigorous stirring for 10 min. Next, 28.6 mg of Cu(NO₃)₂·3H₂O and 57.8 mg of (NH₄)₂SO₄ were added and stirred for 5 min to form the growth solution. Next, 2 mL of 0.2 M NaOH solution was introduced dropwise (1.2 mL min⁻¹) into the growth solution, followed by 2 min of stirring. Subsequently, 3.5 mL of 0.1 M L-ascorbic acid solution was added dropwise (0.4 mL min⁻¹), and the reaction proceeded for an additional 13 min. The precipitate was collected via centrifugation (6000 rpm, 30 min), washed four times with ethanol, and redispersed to obtain Au@Cu₂O NRs. Samples were labeled Au@Cu₂O-L, Au@Cu₂O-M, and Au@Cu₂O-S based on the increasing mass of Au NCs (0.1, 0.5, and 0.9 mg, respectively). Pure Cu₂O nanospheres (Cu₂O NSs) were synthesized using an identical protocol without Au NCs.

Preparation of electrodes

The catalyst ink was prepared by diluting Au@Cu₂O NRs ethanol dispersion to the desired concentration and mixing with 5% Nafion solution at a 24:1 (v/v) ratio. The mixture was sonicated for 30 min to ensure homogeneity. For the H-cell, the electrode size was 1 × 2 cm². For the flow-cell, the electrode size was 2.5 × 2.5 cm². The ink was drop-cast onto a gas diffusion layer (GDL, YLS-30T) and air-dried to form the working electrode. The coated electrode was trimmed to expose a 1 × 1 cm² active area of catalytic layer. The electrodes were denoted as Au@Cu₂O-x-y (x = L, M or S). y represents the catalyst mass loading (μg cm⁻²). The final loading was determined by the product of the diluted ink concentration and the drop volume.

Characterization

Transmission electron microscopy (TEM) and high-resolution transmission electron microscopy (HR-TEM) were performed using a JEOL-JEM-2100F microscope. A ThermoFisher Talos-F200X microscope operating at 300 kV was used to conduct the high-angle annular dark-field scanning transmission electron microscopy (HAADF-STEM) and energy-dispersive X-ray (EDX) elemental mapping studies. Surface morphology was analyzed via scanning electron microscopy (SEM, TESCAN). X-ray diffraction (XRD, Shimadzu) and X-ray photoelectron spectroscopy (XPS, ESCALAB Xi + ) were employed to determine crystallographic structures and chemical states, respectively. The C 1 s peak (284.8 eV) served as a reference for XPS calibration. CO₂ adsorption isotherms were measured at 298 K by using automatic volumetric adsorption equipment (MicrotracBEL-BELSORP MAX II) after a degassed process at 120 °C for 12 h.

Electrochemical measurements

ECO₂R measurements were implemented in H-cell and flow cell with a standard three-electrode system controlled by a potentiostat (CHI-1140c). All performances were evaluated at room temperature (25 ± 5 °C) and atmospheric pressure (1.01 × 10⁵ Pa). All potential values in this experiment were recorded without iR correction.

1. H-cell: An H-cell consists of two glass compartments (50 mL) separated by proton exchange membrane (Nafion 117, 2.5 cm diameter). The proton exchange membrane was soaked in 0.5 M KCl for 48 h and rinsed with deionized water before use. Each compartment contained 0.5 M KCl electrolyte that was utilized within 72 h of preparation. The electrolyte was stored in volumetric flasks under ambient temperature conditions (25 ± 5 °C). Electrochemical measurements were performed in a three-electrode configuration using the Ag/AgCl electrode (saturated KCl) as the reference electrode and Pt foil (1.5 × 1.5 cm) as the counter electrode. CO₂ gas was delivered at an average rate of 20 sccm. CO₂ flow rate accurately controlled by a mass flow meter (ACU10FD-LC). Before the ECO₂R, the cathode electrolyte was purged with CO₂ for 30 min to form CO₂-saturated 0.5 M KCl solution (pH = 4.2 ± 0.2). All potentials were calibrated to the potential relative to the reversible hydrogen electrode (RHE) based on the following equation:

   $$\mathrm{E}\left(\mathrm{Vvs}.\mathrm{RHE}\right)=\mathrm{E}\left(\mathrm{Vvs}.\mathrm{Ag}/\mathrm{AgCl}\right)+0.0591\times \mathrm{pH}+0.199$$ 2

2. Flow-cell: A flow cell made of polytetrafluoroethylene consisted of an Au@Cu₂O-x-10 GDE, an anion-exchange membrane, a Hg/HgO reference electrode, and a platinum mesh counter electrode (1 × 1 cm). The cathode and anode electrolytes were both 1.0 M KOH aqueous solution. The electrolyte was stored in volumetric flasks under ambient temperature conditions (25 ± 5 °C). The anion exchange membrane (2 × 2 cm²) was soaked in 1 M KOH for 48 h and rinsed with deionized water before use. The flow rate of the cathode and anode electrolytes were controlled at 10 and 7.5 mL min⁻¹, respectively, using two peristaltic pumps. The CO₂ flow rate was 20 sccm controlled by a mass flow controller. Before the ECO₂R, the cathode electrolyte was purged with CO₂ for 30 min to form CO₂-saturated 1.0 M KOH solution (pH = 13.7 ± 0.2). All potentials were calibrated to the potential relative to the RHE based on the following equation:

$$\mathrm{E}\left(\mathrm{Vvs}.\mathrm{RHE}\right)=\mathrm{E}\left(\mathrm{Vvs}.\mathrm{Hg}/\mathrm{HgO}\right)+0.0591\times \mathrm{pH}+0.098$$ 2

The ECO₂R gas products were detected using an in-line gas chromatography system (GC 7920, Beijing China Education Au-light Co., Ltd.) equipped with a thermal conductivity detector (TCD) and two flame ionization detectors (FIDs). The TCD was employed for H₂ detection while the FIDs for detecting CO, CH₄, C₂H₄ and C₂H₆. The ECO₂R electrolyte was collected and the liquid phase products such as CH₃OH, C₂H₅OH and n-C₃H₇OH were detected by an out-line gas GC system equipped with an FID and a high-temperature gasification inlet (set at 250 °C). HCOOH and CH₃COOH were quantitatively analyzed using ¹H nuclear magnetic resonance (NMR, BRUKER AVANCEIIIHD500) spectroscopy with H₂O suppression. 500 µL electrolyte mixed with 10 µL dimethyl sulfoxide (0.4 μL mL⁻¹) and 100 µL D₂O was used as the internal standard.

Faradaic efficiency (FE) of the products was calculated by the following equation:

$$\mathrm{FE}\left(\%\right)=\mathrm{e}\times \mathrm{F}\times \mathrm{n}/\left(\mathrm{I}\times \mathrm{t}\right)$$ 3

where e is the number of electrons transferred, F is the Faraday constant (96485 C mol^–1), I is the current (A), t is the running time (s) and n is the total amount of product (mol).

All electrochemical measurements were performed in the same three-electrode system as used in ECO₂R process. Linear sweep voltammetry (LSV) was conducted in the potential range from 0 to −1.8 V (vs. RHE) at a scan rate of 20 mV s⁻¹. The electrochemically active surface area (ECSA) was determined by the double layer capacitance (C_dl). To obtain the C_dl value, cyclic voltammetry (CV) was carried out in the potential range from 0.75 to 0.85 V (vs. RHE) at various scan rates (10, 20, 30, 40, 50, 60, 70, 80, 90 and 100 mV s⁻¹). The anodic and cathodic current density difference of each CV curve at 0.80 V (vs. RHE) was then plotted as a function of scan rate, with the slopes used to calculate C_dl^73,74.

The galvanostastic step experiment was performed in a three-electrode system. A time sequence of two steps was involved. The current densities in the two steps were set as −30 mA cm⁻² and 10 mA cm⁻², respectively. The switch of two steps was set at 10 s. The time interval of collected potentials was set as 0.8 ms.

In situ Raman spectroscopy measurements

In situ Raman spectra were carried out with a Raman spectrometer (InVia Qontor). A visible light laser (Renishaw, RL532, λ = 532 and 633 nm) was used as the excitation source. The ECO₂R was conducted in a custom-built cell made of polytetrafluoroethylene. The cell was equipped with an Ag/AgCl reference electrode and a graphite counter electrode, while the working electrode with the catalyst was drop-casted on carbon paper. Multiple spectra were collected in 0.5 M KCl at OCP and potentials ranging from 0.0 V to −1.6 V (vs. RHE). Each potential was applied for a minimum of 5 min before collecting the spectra. In situ surface-enhanced Raman spectroscopy (SERS) measurements were performed in 1.0 M KOH.

In situ ATR-SEIRAS measurements

In situ attenuated total reflectance surface enhanced infrared absorption spectroscopy (ATR-SEIRAS) measurements were performed on an INVENIO R Fourier transform infrared (FT-IR) spectrometer (Bruker). The reactor is a homemade cell containing a single Si crystal, where a thin Au film was chemically deposited on the surface of Si crystal in advance according to a reported method⁷⁵. Specifically, the well-polished Si crystal was immersed in the mixture of 6 mL H₂SO₄ (98%) and 2 mL H₂O₂ (30%) for 1 h, and then washed with deionized water and ethanol. 21 mg NaOH was dissolved in 13.4 mL deionized water and mixed with 0.6 mL HAuCl4 (0.1 mg mL⁻¹) to obtain an orange solution A. 330.9 mg Na₂S₂O₃·5H₂O, 71.3 mg NH₄Cl and 514.8 mg Na₂SO₃ were dissolved in 16 mL deionized water to obtain solution B. Solution A and B were mixed and stirred for 4 h to obtain a clear solution C. Thereafter, the surface of Si crystal was immersed in NH₄F (40%) for 2 min and then immersed in a mixture of 2 mL solution C and 17 μL HF (40%) at 55 °C for 3 min. After washing the Si crystal with deionized water and ethanol, a thin Au film was chemically deposited on the surface of Si crystal. The same ink (30 μL) as used in electrochemical measurements was dropped onto the Au film to be the working electrode. An Ag/AgCl electrode (saturated KCl) and a Pt plate served as the reference electrode and the counter electrode, respectively. 0.5 M KCl was employed as the electrolyte bubbled by CO₂ continuously. Then the in situ ATR-SEIRAS spectra were recorded in the potential range from 0.0 to −1.6 V (vs. RHE).

Finite Element Method (FEM) simulations

FEM simulations were conducted utilizing the COMSOL Multiphysics software package. The “Chemistry” module was used to define the ECO₂R intermediate steps. As shown in Supplementary Fig. S42, based on the statistically significant cavity size and shell thickness results of Supplementary Figs. S6-S8, three different planar models of derived Cu shell thickness (blue parts) and cavity size (gray parts) were constructed to match the three Au@Cu₂O-x NRs with different structural characteristics. Specifically, for the Au@Cu₂O-L NRs, the cavity size was set to 193 nm, the shell size was set to 11.5 nm. For Au@Cu₂O-M NRs, the cavity size was set to 151 nm, and the shell size was set to 67.0 nm. For Au@Cu₂O-S NRs, the cavity size was set to 113 nm, and the shell size was set to 46.0 nm. All the dimensions of the Au NCs were set to 77.0 nm. For detail input parameters, the temperature was set to 298 K, the pressure was set to 101325 Pa, the concentration of CO₂ was initially set to 40 mol m⁻³, based on the solubility of CO₂. The concentration of CO was initially set to 0 mol m⁻³. For detail input parameters, the temperature was set to 298 K, the pressure was set to 101325 Pa, the concentration of CO₂ was initially set to 40 mol m⁻³, based on the solubility of CO₂. The concentration of CO was initially set to 0 mol m⁻³. In the module dedicated to “Chemistry”, two types of chemical substances and corresponding three surface reactions were established, aiming to simulate the intermediate steps of CO₂ electroreduction. Two chemical substances, CO₂ and CO, each of which is present in the bulk solution and adsorbed onto surfaces, defined as respectively representing the CO₂ feedstock and *CO intermediate. Three reactions were defined: two surface adsorption-desorption equilibria of two chemical species and an irreversible reaction for the CO₂ reduction to CO.

Two surface adsorption-desorption equilibrium reactions for the two chemical species.

$${\mathrm{CO}}_2\rightleftharpoons {\mathrm{CO}}_2\text{-}\mathrm{ad}$$ 4

$$\mathrm{CO}\rightleftharpoons \mathrm{CO}\text{-}\mathrm{ad}$$ 5

An irreversible reaction for the CO₂ reduction to CO.

$${\mathrm{CO}}_2\to \mathrm{CO}$$ 6

Subsequently, an analysis of mass transport for the two species was conducted employing the “Transport of Diluted Species” module. After fitting to the experimental data, the equilibrium coefficient for the surface adsorption-desorption was set to 0.03, the rate constant for CO production was set to 100.

DFT computational details

All DFT calculations were performed with the Vienna Ab Initio Simulation Package (VASP) code⁷⁶. The Perdew–Burke–Ernzerhof (PBE) was employed for electron exchange-correlation⁷⁷. Projector Augmented Wave (PAW) potentials were used to describe the ionic cores⁷⁸. The geometry optimizations were performed with a plane-wave cutoff of 400 eV. The convergence criteria for the electronic steps were set to 1 × 10⁻⁶ eV/atom, and for the ionic steps, the maximum force was set to be less than 0.05 eV/Å. Dipole corrections were included in all the calculations to minimize the inaccuracies in the total energy due to the simulated slab interactions. The dipole moment was calculated parallel to the z-direction. The empirical correction in Grimme’s method with Becke-Jonson damping (DFT-D3(BJ)) was used to describe the van der Waals interactions.

The lattice constant of Cu was optimized to be 3.63 Å in its fcc crystal structure. The Cu (111) surface was constructed with 4 × 4 × 4 supercell with two bottom layers fixed as shown in Supplementary Fig. S43. Two topmost layers and adsorbates were free to move in all directions. The vertical separation between periodically repeated images was set to be at least 15 Å in all cases, to ensure no interaction between images. The coverage of *CO is expressed in monolayer (ML) units, defined as the number of adsorbed *CO molecules (n, n = 2, 3, 4, 5 and 6 in this simulation) divided by the total number of copper surface atoms (16). The adsorption energy was calculated by the following equation:

$$E_{\mathrm{ads}}=E_{\mathrm{final}}-E_{\mathrm{substrate}}-E_{\mathrm{adsorbate}}$$ 7

where E_ads, E_final, E_substrate, and E_adsorbate were the adsorption energy of the adsorbate on the substrate, total energy of the adsorbate on the substrate, total energy of the substrate, and total energy of the adsorbate, respectively.

The formation energy of intermediates (e.g.,*CHO and *OCCOH) under such *CO coverage then are calculated as follows:

$$E_{\mathrm{formation}}=E_{\mathrm{total}}-\left(E_{\mathrm{slab}+\mathrm{n}*}\mathrm{CO}+E_{\mathrm{H}2}/2\right)$$ 8

Where E_total is the energy of the total system with adsorbate, E_slab+n*CO is the energy of Cu slab with adsorption of n *CO molecules. E_H2 is the energy of a hydrogen molecule, which is −6.76 eV.

The thermally correction of Gibbs reaction free energy was calculated by the VASPKIT code with the equation⁷⁹:

$$\mathrm{\Delta }\mathrm{G}={\mathrm{\Delta }E}_{\mathrm{ads}}+{\mathrm{\Delta }E}_{\mathrm{ZPE}}\mathrm{-}{\mathrm{T}\mathrm{\Delta }\mathrm{S}}_{\mathrm{ads}}$$ 9

where ΔE_ads represents the electronic adsorption energy, ΔE_ZPE represents the zero-point energy difference between the adsorbed and gaseous species, and TΔS_ads represents the corresponding entropy difference between these two states (T was set to 298 K). For all calculations, the optimized structures are provided in Supplementary Data 1.

Author contributions

Z.K.Z., X.Y.W., and S.J.L. designed and conceived the experiment. Z.K.Z. and S.J.L. performed the catalyst synthesis, structural characterization, and ECO₂R electrocatalytic measurements. M.T.L. and H.G. conducted the DFT calculations. Z.K.Z. and S.H.G. performed the in situ Raman Spectroscopy and ATR-SEIRAS measurements and analyses. M.T.L. and W.Y. contributed to the FEM simulations and analyses. Z.K.Z., H.G., and S.J.L. wrote and revised the manuscript with inputs from all authors. H.X. supervised the project.

Peer review

Peer review information

Nature Communications thanks Lu Liu, Jian Yang, and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Data availability

The authors declare that all data generated in this study are provided in the Supplementary Information/Source Data file. Source data are provided with this paper.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-025-62875-8.