Catalyst Preparation and Characterization

Catalysts were prepared by a two-step calcination process with acidified g-C₃N₄ (a-C₃N₄) and Cu(NO₃)_2·3H₂O as precursors, as illustrated in Figure 1a. Acidified g-C₃N₄ was obtained by exfoliating bulk g-C₃N₄ in concentrated H₂SO₄. Detailed synthesis steps are provided in Supporting Information. It has been reported that acid treatment could induce the generation of hydroxyl groups on the surface of bulk g-C₃N₄,^[23] which is also confirmed by our FT-IR analysis. As shown in Figure S1, compared with primary bulk g-C₃N₄, new peaks appeared in the FT-IR spectra of a-C₃N₄ at 1536, 1194, and 1130 cm⁻¹, which correspond to the stretching vibration of OH, ─NO, and ─CO, respectively, indicating the presence of these functional groups in a-C₃N₄. Hydroxyl groups contribute to the transformation of bulk C₃N₄ into a thin-layer structure, increasing the specific surface area.^[23] Meanwhile, the hydroxyl groups can also coordinate with metal ions, which helps to anchor more metal ions and enhance the metal loading amount. Cu₄/Cu₁@CN was obtained when the mass ratio of Cu(NO₃)₂ to a-C₃N₄ was 30%, whereas Cu₁@CN was obtained when the ratio was 3%. Acidified g-C₃N₄, calcined under the same conditions, was used as control sample and named as CN. The mass percentage of Cu in Cu₄/Cu₁@CN and Cu₁@CN were determined to be 19.0% and 0.14%, respectively, by inductively coupled plasma mass spectrometry (ICP-MS).

The morphologies of the samples were investigated by high resolution scanning electron microscopy (HRSEM) and transmission electron microscopy (HRTEM). All three samples show a porous layered structure; the number of pores increase with the increase of Cu content (Figures 1b and S2–S4). BET analysis shows that Cu₄/Cu₁@CN has the largest specific surface area (71.82 m² g⁻¹) (Figure S5 and Table S1); such porous structure with large specific surface area could provide more reactive sites and enhance light harvesting as well as mass transfer during catalysis.^{[24, 25]} No particles are found in HRTEM images of Cu₄/Cu₁@CN and Cu₁@CN. The energy-dispersive X-ray spectroscopy (EDS) mapping image shows that Cu, C, N, and O are uniformly dispersed in these two catalysts. The power X-ray diffraction (XRD) patterns of the three samples consist of one strong diffraction peak at 27.4° and one weak peak at 13.0°, corresponding to the orderly stacked (002) interlayer-plane and (100) in-plane ordering of tri-s-triazine units, respectively(Figure S6).^{[26, 27]} Notably, the higher the amount of Cu used, the weaker is the peak at 13.0°, indicating that Cu loading disrupts the tri-s-triazine structure, leading to poor in-plane ordering. There are no peaks corresponding to Cu metal or other Cu compound in XRD, indicating the absence of Cu-based nanoparticles in Cu₄/Cu₁@CN and Cu₁@CN. The atomic-resolution aberration-corrected HAADF-STEM images of Cu₄/Cu₁@CN show that the density of Cu atoms on the majority of the support is remarkably high (Figure 1c), primarily attributable to the ultrahigh Cu loading, as determined by ICP-MS. In the relatively thinner regions with lower Cu content, both isolated Cu single atoms and clusters are present. Upon magnification, it is evident that most of the clusters are formed by the close-packed 4 Cu atoms (Figure 1d). The distance between adjacent Cu atoms within these clusters ranges from 1.5 to 2.3 Å (Figure S7), signifying the formation of Cu─Cu bonds. Notably, it seems that there is always a Cu single atom next to a Cu₄ cluster. Elemental mapping analysis shows that Cu, C, O, and N are uniformly dispersed in the sample (Figure 1e). On the other hand, Cu₁@CN consists of many Cu single atoms (orange circles), as shown in Figure S8.

X-ray absorption fine structure (XAFS) analysis was used to determine the coordination environment and local structure of copper in Cu₄/Cu₁@CN and Cu₁@CN. Figure 2a presents the Cu K-edge X-ray absorption near-edge structure (XANES) spectra of the two catalysts as well as Cu foil, copper phthalocyanine (CuPc), and CuO as references. Notably, the near-edge absorption peaks of Cu₄/Cu₁@CN and Cu₁@CN are located between that of Cu foil and CuO, indicating that the Cu atoms in Cu₄/Cu₁@CN and Cu₁@CN carry positive charges with oxidation state between 0 and +2. Figure 2b shows the Fourier-transformed extended X-ray absorption fine structure (EXAFS) spectra of these samples. The R space positions (prior to phase correction) of the Cu─Cu peak in copper foil is at 2.22 Å, the Cu─O scattering in the first shell of CuO and Cu─N scattering in the first shell of CuPc are at the R-space distances of 1.53 and 1.50 Å, respectively. Thus, the peak at 1.52 and 2.14 Å in Cu₄/Cu₁@CN are attributed to Cu─N/O and Cu─Cu. This is also confirmed by the wavelet transform contour plots of Cu K-edge weighted EXAFS as shown in Figure 2d. Figures 2c and S9a, as well as Table S2, show the corresponding EXAFS fitting results, which reveal that each Cu atom coordinates with 2 N atoms with bond length of 1.95 Å, 1 O atom with bond length of 1.80 Å, and 3 Cu atoms with bond length of 2.56 Å, demonstrating the coexistence of Cu₄ cluster, Cu─N₂, and Cu─O in Cu₄/Cu₁@CN.

Similarly, the peaks at 1.47 Å in Cu₁@CN is attributed to Cu─N/O. The fitting results show that one Cu atom coordinates with 2 N and 1 O atom, with the bond length of 1.96 and 1.81 Å, respectively (Figure S9b,c and Table S2). Combining the result of HAADF-STEM and wavelet transform contour plot (Figure S9d), it can be deduced that Cu single atoms with Cu─N₂ and Cu─O coordination environment exist in Cu₁@CN.

Based on the XAFS results and HAADF-STEM images, density functional theory (DFT) calculations were conducted to model the Cu coordination in Cu₄/Cu₁@CN and Cu₁@CN. The optimized structures are inserted in Figures 2c and S9b, respectively. Cu₄/Cu₁@CN consists of Cu₄ clusters anchored on CN via three Cu─N₂ bonds and the adjacent Cu single atom is coordinated with 2 N and 1 O atoms. Cu₁@CN consists of isolated Cu single atoms anchored on CN via Cu─N₂ and Cu─O bonds.

X-ray photoelectron spectroscopy (XPS) was used to reveal the chemical nature of Cu₄/Cu₁@CN and Cu₁@CN. Figure S10a shows the Cu 2p spectrum of Cu₄/Cu₁@CN; the main characteristic peaks appear at 932.9 and 952.5 eV, corresponding to Cu⁺ or Cu⁰ 2p_3/2 and 2p_1/2, respectively. (Note that the peaks of Cu⁰ and Cu⁺ are located at similar positions in the XPS spectrum.)^[28] The oxidation states of Cu have been further confirmed by Auger electron spectroscopy (AES) (Figure S10b). There are two deconvolved peaks of Cu⁺ (916.5 eV) and Cu⁰ (918.5 eV). Thus, the XPS analysis is consistent with XAFS result. On the other hand, XPS as well as the corresponding AES are unable to explicitly determine the oxidation states of Cu in Cu₁@CN due to the relatively low signal-to-noise ratio associated with the low copper content.^[29]

The N 1s XPS spectra were also fitted to compare the coordination environments of the catalysts (Figure S10c). The N 1s XPS spectra exhibit a broad feature with binding energy ranging from 397 to 402 eV. For Cu₄/Cu₁@CN, this feature can be deconvoluted into three peaks centered at 398.4, 398.9, and 401.02 eV, which can be assigned to pyridinic (C─N═C), tertiary (─N < or M─NC₂), and pyrrolic (─NH) nitrogen, respectively.^[29-31] Compared with CN, Cu₄/Cu₁@CN exhibits a much more pronounced feature associated with the tri-coordinated N (M─NC₂), consistent with the expected Cu─N coordination. On the other hand, there is no obvious difference in N 1s XPS spectra between Cu₁@CN and CN, which should be attributed to the low Cu content in Cu₁@CN. Moreover, the additional peak centered at 404.4 eV for the three samples should be ascribed to terminal nitrate groups, charging effects, or π excitations.^[32-34] The high-resolution O 1s spectra of Cu₄/Cu₁@CN (Figure S9d) can also be deconvoluted into three peaks centered at 530.6, 531.7, and 532.7 eV, corresponding to Cu─O, vacancy O, and adsorbed O, respectively.^{[35, 36]}

Photocatalytic CO₂ Reduction

Photocatalytic CO₂ reduction was conducted in a solid–gas reaction system under irradiation with Xenon lamp. CO₂ saturated with water was purged into the reactor for 30 min before irradiation. Detailed experimental procedures are provided in Supporting Information. The products formed from Cu₄/Cu₁@CN, Cu₁@CN, and CN and their production rates are presented in Figures 3 and S11–13, as well as Table S3. Among the three catalysts, Cu₄/Cu₁@CN displayed significantly higher activity and selectivity for CH₃CH₂OH formation. Ethanol production increased linearly with time with rate of 154 µmol g⁻¹ h⁻¹ and selectivity of 98% (Figures 3a and S13); such performance is among the best in terms of both ethanol production rate and selectivity, compared with the reported CO₂ to ethanol photocatalysts (Figure 3b).^{[16, 18, 37-42]} Traces of CH₃OH, CH₄ and CO were also detected. Notably, the yield and selectivity of ethanol were almost unchanged in four consecutive cycles (Figure 3d). Additionally, TEM image, XRD patterns, and Cu 2p XPS spectra were almost the same as before the reaction (Figures S14–16), demonstrating the profound catalytic stability of Cu₄/Cu₁@CN. In contrast, Cu₁@CN had the highest activity for CO and negligible activity to form ethanol. For CN, no ethanol was produced, only trace amounts of CH₄ and CO were detected. The experiment with Cu₄/Cu₁@CN as catalyst under Ar also gave negligible products (Table S3). The apparent quantum efficiency for ethanol production was determined to be 0.012% after irradiating at monochromatic light of 350 nm for 20 h.

To verify the origin of ethanol, isotopic labeling experiments with ¹³CO₂ were performed and the GC-FID spectra of gaseous products are shown in Figure S17. GCMS of gaseous products from use of ¹²CO₂ show peaks at m/z = 16, 29, 31, 44, 45, and 46 in the mass spectrum, which are attributed to ¹²CH₄, ¹²CH₃¹²CH₂, ¹²CH₂OH, ¹²CO₂, ¹²CH₃¹²CH₂O, and ¹²CH₃¹²CH₂OH, respectively (Figure 3c). The peak at m/z = 28 should be attributed to N₂ and ¹²CO. When ¹³CO₂ was used, the peaks at m/z = 17, 29, 31, 45, 47, and 48 in the mass spectrum are assigned to ¹³CH₄, ¹³CO, ¹³CH₃¹³CH₂, ¹³CO₂, ¹³CH₃¹³CH₂O, and ¹³CH₃¹³CH₂OH. The concentration of methanol in the gas phase product was extremely low, such that no corresponding peaks were detected in the GCMS analysis. The above results indicate that ethanol was produced from CO₂ reduction.

As shown in Figure S18 and Table S4, O₂ with the yield rate of 449 µmol g⁻¹ h⁻¹ was generated in the system using Cu₄/Cu₁@CN, which is equivalent to 1795 µmol g⁻¹ h⁻¹ electrons and H⁺ based on the reaction of water oxidation to O₂. This is nearly equal to the mole of electrons and H⁺ consumed to produce 154 µmol g⁻¹ h⁻¹ ethanol from CO₂ reduction (1844 µmol g⁻¹ h⁻¹), demonstrating that the holes generated by light irradiation were used to oxidize water to O₂ and provide H⁺ for CO₂ reduction.

Reaction Mechanism

The light adsorption property and charge separation efficiency of the catalysts were investigated by spectroscopic methods. The UV–visible diffuse reflectance spectra (UV–vis DRS) (Figure S19) of Cu₁@CN and CN show a light absorption band from 300 to 700 nm, whereas Cu₄/Cu₁@CN exhibits a broader and enhanced light absorption from 300 to 800 nm. According to the Tauc plots (inset in Figure S19), Cu₄/Cu₁@CN has the narrowest bandgap (2.14 eV) compared with Cu₁@ CN (2.68 eV) and CN (2.71 eV).

In addition, Mott–Schottky plots were employed to determine the conduction bands of the catalysts. The valence band of samples was calculated using the formula E_g = E_VB − E_CB. As shown in Figures S20 and S21, the band structures of all catalysts indicate their capability of oxidizing H₂O to O₂ and reducing CO₂ to CO, CH₄, CH₃OH, and C₂H₅OH.

Transient photocurrent responses (Figure S22a) and electrochemical impedance spectroscopy (EIS) Nyquist plots (Figure S22b) were conducted to explore the separation of the photogenerated carriers. Cu₄/Cu₁@CN (1.52 µA cm⁻²) exhibited the highest photocurrent density in comparison to Cu₁@ CN (0.15 µA cm⁻²) and CN (0.02 µA cm⁻²), and EIS Nyquist plots show that Cu₄/Cu₁@CN has the smallest semicircle radius, indicating the lowest transfer impedance. These results show that Cu₄/Cu₁@CN exhibits the optimal effect on the separation of photogenerated carriers. Steady-state photoluminescence (PL) spectra (Figure S22c) and time-resolved photoluminescence (TRPL) spectra (Figure S22d) further reveal the charge transfer dynamics of the catalysts. According to the intensity of the emission peak, Cu₄/Cu₁@CN has the weakest signal peak in the entire wavelength range compared with the other two catalysts, meaning that the recombination of photoinduced carriers could be effectively inhibited.^[43] The TRPL spectra also show that the average emission lifetime of Cu₄/Cu₁@CN (1.85 ns) is longer than those of Cu₁@CN (0.81 ns) and CN (0.85 ns). The increased PL lifetime of Cu₄/Cu₁@CN indicates a longer retention time of charge carriers. The excellent charge separation ability provides ample electrons for photocatalytic CO₂ reduction.

In situ XAFS was used to probe into the alterations in the valence state of Cu during photocatalytic process. As shown in Figure 4a, upon the introduction of CO₂, a remarkable increase in the valence state of Cu was observed. This phenomenon implies that adsorption of CO₂ onto Cu is accompanied by the transfer of electrons to *CO₂. Subsequently, once the photocatalytic reaction commenced, the valence state of Cu was further elevated, signifying that a greater number of electrons were transferred from Cu to the reactants during the reaction course. In other words, CO₂ was reduced under the catalysis of Cu.

In situ attenuated total reflection surface-enhanced infrared absorption spectroscopy (ATR-SEIRAS) was applied to monitor the possible intermediates in photocatalytic CO₂ reduction. As shown in Figures 4b and S23, *OCH₂CH₃ (1353 cm⁻¹),^[44-46] *HCO₃⁻ (1433 cm⁻¹),^[47] m-CO₃²⁻ (1275,1510 cm⁻¹),^{[47, 48]} *COCO (1545, 1750 cm⁻¹), *COCHO (1571 cm⁻¹),^{[18, 49]} *COOH (1605 cm⁻¹),^{[18, 49, 50]} and *CO (2008, 2033, 2047 cm⁻¹)^[49] are observed. *COOH is the signature intermediate to form CO, *COCO is an essential intermediate for C₂₊ products, and *COCHO is a key intermediate to form ethanol. The presence of *OCH₂CH₃ is another important evidence that ethanol is produced in the reaction. The above results indicate that photocatalytic CO₂ reduction to ethanol on Cu₄/Cu₁@CN occurs via *COCO to *COCHO and then to ethanol.

Wang and Li^[22] have demonstrated through density functional theory (DFT) calculations that Cu₄ clusters with three Cu─N₂ bonds anchored on g-C₃N₄, which are consistent with the structure of the Cu₄ clusters reported here, are capable of catalyzing the reduction of CO₂ to C₂₊ products. In contrast, Cu SACs anchored on g-C₃N₄ are unable to drive this reaction due to the high energy barrier for CO─CO coupling. Their conclusion is in line with our experimental results, in which no ethanol or other C₂₊ products were observed with Cu₁@CN as catalyst, whereas CH₃CH₂OH was produced with high rate in the system using Cu₄/Cu₁@CN.

We first calculated CO formation on both clusters and single-atom sites. As shown in Figure 4c, the free energy change (ΔG) for *COOH formation on both cluster sites (Cu₄ pathway) and Cu single atom sites (Cu₁ pathway) are very low, indicating that *CO is readily produced on both sites. To achieve C₂₊ products, C─C coupling is usually considered as the rate determine step. The C─C coupling on the Cu single atom site has been studied; however, *COCO was not stable on this site, indicating that Cu single atoms alone are not the active sites for C─C coupling. This is consistent with the experimental results that only CO and CH₄ were obtained when Cu₁@CN was used as the catalyst. Thus, two other pathways are considered for the C─C coupling on Cu₄/Cu₁@CN in our study: i) two *CO are both located on clusters (Cu₄ pathway), and ii) two *CO coupling in a synchronized manner where one *CO is located on Cu₄ cluster and the other one is on Cu single atom site (Cu₄₊₁ pathway). For the Cu₄ pathway, *CO is firstly absorbed on the vertex site of Cu₄ cluster due to its more negative Gibbs free energy of *CO absorption (−1.51 eV, Figure S24), and the adsorption of a second CO molecule to form *CO−*CO is thermodynamically favored with ΔG = −1.19 eV. However, C─C coupling by the two adjacent *COs needs to overcome an energy barrier of 1.94 eV. For the Cu₄₊₁ pathway, the energy barrier of C─C coupling between *CO located on Cu single atom site and the *CO on the nearby Cu₄ cluster is 1.53 eV, lower than that in Cu₄ pathway, indicating that the synergistic effect between Cu cluster and single atom promotes C─C coupling reaction. *COCHO can be readily produced by the hydrogenation of *COCO with ΔG = −0.56 eV. *COCHO is further reduced to *CHOHCH₂OH with ΔG of 0.9 eV and then to the product *CH₃CH₂OH with an additional ΔG of 0.26 eV.

Effect of Cu Loading

To investigate the effect of Cu loading on photocatalytic performance, another two samples with mass ratios of Cu(NO₃)₂ to a-C₃N₄ of 15% and 60% were synthesized, and their photocatalytic performances were compared with the above reported 3% (Cu₁@CN) and 30% (Cu₄/Cu₁@CN) samples. As depicted in Figure S25, ethanol production demonstrated a volcano-type dependence on Cu(NO₃)₂ addition, peaking at 30% before declining. This trend is rationalized by a dual mechanism: incremental Cu loading could increase active site density, whereas excessive Cu deposition obscured the C₃N₄ matrix, compromising light absorption efficiency and photogenerated carrier separation. These observations are corroborated by complementary spectroscopic and electrochemical characterizations, including UV–vis DRS, transient photocurrent response, PL, and TRPL spectroscopy (Figures S26 and S27).