Sample preparationZIF-8 framework was synthesized using 2-methylimidazole (4.92 g, 99% Sigma-Aldrich) and zinc-nitrate hexahydrate (4.24 g, 98%, Acros Organics), dissolved in methanol (500 mL). This solution was heated to 60 °C and stirred for 24 h under reflux, followed by centrifugation and thorough washing with methanol and ethanol. The resulting powder was dried at 60 °C in air. The obtained ZIF-8 powder was pyrolyzed for 1 h in Ar flow (100 mL min−1) at 1000 °C, resulting in the metallization and evaporation of Zn species within the ZIF-8 framework. The remaining porous N-doped carbon structure was acid-washed for 24 h in 20 w% HNO3 (≥ 65%, Carl Roth). The acid treatment removes the remaining crystalline Zn-containing species. The resulting N-C precursor sample was washed thoroughly with ultrapure water.To prepare Cu-N-C catalysts, the N-C precursor was impregnated with Cu2+ species. For this purpose, N-C was dispersed in 6 mM Cu(NO3)2 ∙ 3 H2O (Sigma-Aldrich, > 99%) solution in isopropanol (≥ 99.8%, Sigma-Aldrich) by sonicating for 2 h at ~ 40 °C. The suspension was then stirred for another 2 h at room temperature (RT). The Cu-N-C precursor was collected by centrifugation, air dried at 60 °C and heated in Ar (100 mL min−1) at 700 °C for 1 h. The obtained Cu-N-C was stirred in 20 w% HNO3 for 24 h and washed with ultrapure water. Finally, the catalyst was dried in air at 60 °C.Electron microscopySTEM images were obtained using a probe-corrected JEM-ARM 200 F (JEOL, Japan) operated at 200 kV and equipped with a cold field emission gun (CFEG). HAADF signals were acquired from an electron probe with a 14.2 mrad convergence semi-angle and a 90–370 mrad collection semi-angle. The beam current was kept at 11 pA, resulting in an electron dose scaled by the pixel size. For the samples after the reaction, the catalysts were collected after 4000 s under static or pulsed CO2RR. The samples were removed from the glassy carbon by sonication in ethanol and drop-casted onto Au TEM grids.Operando QXAFS measurementsThe main XAFS spectra measurements in QXAFS mode were carried out at the SuperXAS beamline at the SLS synchrotron (Villigen, Switzerland) using a LN-cooled channel-cut Si(111) monochromator for energy selection43. The monochromator was oscillating with 1 Hz frequency. The intensity of the incoming X-rays was measured using an N2-filled ionization chamber. Rh-coated collimating and focusing mirrors were used to reduce the heat load and reject higher harmonics. XAFS measurements were performed in fluorescence mode at the Cu K-edge (8979 eV) using a PIPS detector.Operando QXAFS data for pulsed CO2RR with Δta = Δtc = 30 s were collected at the P64 beamline at PETRA III (Hamburg, Germany)44,45 using settings similar to those at SLS SuperXAS.Supplementary operando XAFS data for pulsed CO2RR (Δta = 1 s, Δtc = 1 s or 4 s; and Δta = 4 s, Δtc = 1 s or 4 s) were collected at CryoEXAFS endstation at KMC-3 beamline at BESSY II synchrotron (Berlin, Germany)46. In these measurements the acquisition of each XAFS spectrum took several minutes, therefore, the analysis of these data shows the chemical state and particle sizes, averaged over many potential pulses. The Cu K-edge fluorescence data were collected using a 13-element Si drift detector.For operando measurements, we used an in-house built single-compartment cell (Supplementary Fig. 25)47. The samples were spray-coated on a carbon electrode, which was used as a working electrode while acting also as a window for the incident and fluorescent X-ray photons. A Pt mesh was used as a counter-electrode, and a leak-free Ag/AgCl electrode was used as a potential reference. Measurements were done in a CO2-saturated 0.1 M KHCO3 electrolyte (pH = 6.8), with CO2 continuously bubbling through the cell. The electrolyte volume in the cell is 35 ml. Circulation of the electrolyte was ensured by a peristaltic pump. The potential was controlled by a BioLogic potentiostat. The potentials reported in the text are given with respect to the reversible hydrogen electrode (RHE), using the following equation:$${E}_{{{{{\rm{RHE}}}}}}={E}_{{{{{\rm{Ag}}}}}/{{{{\rm{AgCl}}}}}}+0.242 \, {{{{\rm{V}}}}}+0.059{{\cdot pH}}.$$
(1)
For the calibration of the QXAFS data, at the beginning of each scan a Cu foil spectrum was collected in transmission mode. Calibration of the QXAFS data was done using beamline-specific software48. Spectra averaging, background subtraction, and XANES data processing were carried out using a set of in-house built Wolfram Mathematica scripts. Extraction and fitting of EXAFS data was carried out using LARCH49 and FEFFIT50 codes.For averaging XAFS spectra collected at the same time moments after the onset of each respective potential pulse (Fig. 3a), XAFS spectra μ(t) with the same φ = (t mod T) are gathered and averaged, where t is the time at which the respective spectrum was collected (with t = 0 corresponding to the onset of pulsed CO2RR), T = Δtc + Δta is the total duration of potential cycle, and “mod” denotes modulo operation. Only the spectra corresponding to the steady state, i.e., collected a sufficiently long time after the onset of the pulsed CO2RR, are included in the averaging (t > 2000 s).Interpretation of XANES and EXAFS dataLinear combination fits of XANES spectra were performed using the Cu K-edge XANES spectrum for a Cu foil and a spectrum for the as-prepared Cu-N-C catalyst as references. This approach thus neglects the dependency of the XANES spectra for metallic Cu on the particle size, and the changes in the XANES spectra for singly dispersed Cu species during the exposure of these sites to CO2RR conditions. The concentration of singly dispersed Cu sites xSAC can be estimated from LCA-XANES results simply as the weight of the corresponding reference spectrum in the linear combination. The concentration of metallic Cu is then given by 1 – xSAC.Fitting of the EXAFS spectra χ(k)k2 was carried out in R-space in the range from 1.0 up to 2.8 Å (up to 2.0 Å for samples that did not contain metallic Cu clusters). Fourier transform was carried out in the k range from 2.0 up to 8.0 Å-1. We included in the fit Cu-O and Cu-Cu paths. Here, Cu-O paths account also for the possible Cu-C and Cu-N bonds, since these cannot be discriminated easily by EXAFS analysis. For both paths, their average interatomic distance R, coordination number N, and disorder factor σ2 were refined. Furthermore, the correction to the photoelectron reference energy ΔE0 was also fitted. Amplitude reduction factors S02 = 0.85 for the Cu-Cu bond and 0.68 for the Cu-O bond were determined from the EXAFS fits of reference materials (Cu foil and CuO).The apparent coordination numbers obtained in EXAFS fits NCu-O and NCu-Cu are averages over all Cu species in the sample. Since singly dispersed Cu sites do not have another Cu atom in their first coordination shell, and the metallic Cu species are not expected to form strong bonds with C, N, or O atoms, this allows us to obtain an independent estimate of the concentrations of singly dispersed and metallic Cu. Assuming that singly dispersed Cu sites remain 6-coordinated during the CO2RR, the concentration of singly dispersed Cu sites can be estimated as xSAC = NCu-O/6. The validity of this assumption can be questioned. However, the observed good agreement between xSAC values obtained independently from XANES and EXAFS data analyses validates our approach.Furthermore, the true Cu-Cu coordination numbers for the metallic Cu phase can now be obtained as \({\widetilde{{{{{\rm{CN}}}}}}}_{{{{{\rm{Cu}}}}}-{{{{\rm{Cu}}}}}}={{{{{\rm{CN}}}}}}_{{{{{\rm{Cu}}}}}-{{{{\rm{Cu}}}}}}/(1-{x}_{{{{{\rm{SAC}}}}}})\). The latter quantity is directly linked to the particle’s volume-to-surface ratio, and, hence, the average particle size47. Following one simple approach, the relation between the size of the spherical particle with fcc-type structure and true metal-metal coordination number can be approximated with the following formula51:$$\frac{{\widetilde{{{{{\rm{CN}}}}}}}_{{{{{\rm{Cu}}}}}-{{{{\rm{Cu}}}}}}}{{{{{{\rm{CN}}}}}}_{{{{{\rm{fcc}}}}}}}=\left[1-\frac{3}{4}\left(\frac{R}{{R}_{{{{{\rm{nano}}}}}}}\right)+\frac{1}{16}{\left(\frac{R}{{R}_{{{{{\rm{nano}}}}}}}\right)}^{3}\right]$$
(2)
Here, R is the interatomic distance (2.56 Å for copper), Rnano is the radius of the particle, and CNfcc = 12 is the coordination number in the corresponding bulk metal (see Fig. 2f, inset).Measurements of electrocatalytic propertiesTo extract the electrocatalytic properties, an H-type cell with two compartments separated by an anion exchange membrane (Selemion, AMV, AGC Inc.) was used. The membrane is stored in a wet state, and prior to the measurement, it was rinsed with water. One compartment contains the working electrode, where both sides of a 0.5 cm2 carbon paper piece (Toray Carbon Paper, GGP-H-60) were exposed to the electrolyte, while only one side was spray-coated with the catalyst, making a geometric area of 0.5 cm2. A leak-free Ag/AgCl reference electrode (LF-1, Alvatek) is placed near the working electrode. The reference electrode was checked against a reversible hydrogen electrode (Gaskatel) in CO2 sat. 0.1 M KHCO3, pH 6.8, see Eq. (1), and a possible drift in the potential of the Ag/AgCl reference electrode was accounted for. The counter electrode in the second compartment consists of a platinum gauze (MaTecK, 3600 mesh cm-2). A defined amount of previously purified (cation-exchange resin, Chelex 100 Resin, Bio-Rad) 0.1 M KHCO3 (Alfa Aesar, 99.7%) was filled into the cell and saturated with CO2 (4.5 N) with a constant flow of 20 ml min-1 during CO2RR. CO2RR was performed with an Autolab potentiostat (PGSTAT 302 N, Metrohm). The acquired data was collected with the Nova 2.1.5 software and processed via OriginPro®2023b software. The electrolyte was prepared immediately before the experiment and saturated with CO2 for at least 15 min. Ohmic drop resistance, usually between 6 and 14 Ω, was measured via the i-interrupt method prior to the reaction. The CO2 flow towards both the cathode and the anode compartment was controlled with a mass flow controller (Bronkhorst, EL-Flow). The correct flow was ensured with a flow meter (Agilent, ADM, G66991A) to be 20 ml/min prior to each measurement.Online gas product detection was performed after 60 s and every 15 min during CO2RR with a gas chromatograph (GC, Agilent 7890B) for a total time of 4000 s at each pulse condition. Our GC featured a thermal conductivity detector (TCD) for H2 detection and a flame ionization detector (FID) for carbon products. The liquid products were detected with a liquid GC (L-GC, Shimadzu 2010plus), equipped with a fused silica capillary column and an FID detector. Acetate and formate were detected with high-performance liquid chromatography (HPLC, Shimadzu Prominence), equipped with a NUCLEOGEL SUGAR 810 column and a refractive index detector (RID).Measurements were done by using a fresh sample for each pulse condition (when both liquid and gaseous products were quantified, Fig. 4a), or by reusing one sample several times with one pulse sequence following another, with 30 m rest at open circuit potential in between. The electrolyte was not changed in between the different pulse conditions, thus, only gaseous products were monitored. Here, the selectivity and partial current densities are reported instead of the Faradaic Efficiency.The measured currents are dominated by non-Faradaic contributions for pulses with Δtc values lower than 10 s, hindering the systematic evaluation of the charge provided for the Faradaic processes. Nonetheless, the currents during the cathodic pulses for all Δtc values match well with each other over the whole course of pulsed CO2RR (Supplementary Fig. 21), allowing the assumption that both, the capacitive and the faradaic contributions are comparable. Even though the capacitive current dominates the currents of very short pulse durations, non-neglectable amounts of products were found.The Faradaic efficiency of each product was calculated by adapting the previously reported procedure20. In order to deconvolute the capacitive from the faradaic currents, specifically for very short cathodic pulses where capacitive currents dominate over the faradaic currents, we first assumed that the detected products sum up to a total Faradaic efficiency of 100% \(({{{{{\rm{FE}}}}}}_{{{{{\rm{total}}}}},{{{{\rm{ass}}}}}.})\). Thus, the partial current density \({j}_{{{{{\rm{partial}}}}},{{{{\rm{x}}}}}}\) for each product \(x\) was calculated as:$${j}_{{{{{\rm{partial}}}}},{{{{\rm{x}}}}}}=\frac{\dot{V}\, {C}_{x}\, {z}_{x} \, F}{A \, {V}_{M}\frac{\Delta {t}_{c}}{\Delta {t}_{c}+\Delta {t}_{a}}{{{{{\rm{FE}}}}}}_{{{{{\rm{total}}}}},{{{{\rm{ass}}}}}.}}$$
(3)
with\(\dot{V}\): CO2 gas flow rate / L s-1\({C}_{x}\): Volume-fraction of the product \(x\) detected by GC\({z}_{x}\): Electrons transferred for reduction to product \(x\)\(F\): Faradaic constant / C mol−1\(A\): Geometric area of the electrode / cm−2\({V}_{M}\): Molar volume / 24.5 l mol−1\(\frac{\Delta {t}_{c}}{\Delta {t}_{c}+\Delta {t}_{a}}\): Factor to account for the absolute time actual CO2RR while pulsing\({{{{{\rm{FE}}}}}}_{{{{{\rm{total}}}}},{{{{\rm{ass}}}}}.}\): assumed total Faradaic Efficiency / %.Afterwards, the obtained total current densities were used to calculate the Faradaic Efficiency of each product \(x\) as:$${{{{{\rm{FE}}}}}}_{x}=\frac{{j}_{{{{{\rm{partial}}}}},x}}{{\sum }_{i=x}^{j}\,{j}_{{{{{\rm{partial}}}}},x}}*100$$
(4)
\({{{{{\rm{FE}}}}}}_{{{{{\rm{x}}}}}}\): Faradaic Efficiency of product \(x\)\({j}_{{{{{\rm{total}}}}}}\): Total current density, calculated as \({\sum }_{i=x}^{j} \, {j}_{{{{{\rm{partial}}}}},x}\) / A cm−2FEs for liquid products were calculated as discussed in our previous work20.
Reversible metal cluster formation on Nitrogen-doped carbon controlling electrocatalyst particle size with subnanometer accuracy
