Electrochemical CO₂ Reduction on Au Cluster-based Electrodes: Investigating the Role of Nafion Ionomer

The clusters were deposited on carbon black (>99%, Cabot Vulcan XC-72R) following the previously reported methods. ^33,34 Briefly, 2 g Vulcan carbon was dispersed in dichloromethane (DCM) (ca. 40 ml) via sonication and stirred under a nitrogen atmosphere for 2 h in a Schleck flask. [Au₉(PPh₃)₈](NO₃)₃ clusters were dissolved in DCM (ca. 20 ml). The calculated amount of DCM solution was added to the stirring dispersion to achieve a 10% monolayer of Vulcan carbon surface (Fig. SI4a). The dispersion was stirred for 2 h; the clusters were observed to be deposited on the carbon support, as the colour of the mother liquor changed towards colourless. The solvent was then evaporated under reduced pressure using a Schlenk line, and the product was collected and stored in the fridge, avoiding exposure to light.

The cathode for electrochemical CO₂ reduction studies was prepared by spraying an electrocatalyst ink onto a carbon felt (AvCarb C100, area 3.141 cm²). ²⁴ The ink suspension was prepared by sonicating the ethanol and water mixture containing 1 mg ml⁻¹ Au₉/C catalyst and Nafion ionomer (5 wt%, 10 wt%, 20 wt%, and 30 wt%). The calculated amount of Nafion ionomer (Equation SI1) was added from a 10 mg ml⁻¹ stock solution in ethanol. The catalytic ink was sprayed at 0.15 ml min⁻¹ using an ultrasonic atomiser with a constant flow of compressed air onto a carbon felt placed on a hotplate at 120 °C (Fig. SI4b). The carbon felt was further heated at 120 °C for 30 min after spraying to remove traces of solvent. This procedure formed a uniform electrocatalytically active layer on carbon felt with different Nafion content. A uniform sprayed layer onto the carbon felt was obtained with a catalytic loading of 0.84 ± 0.06 mg cm⁻² (measured gravimetrically).

The cathodes for electrochemical CO₂ reduction studies were prepared by spraying carbon black onto a carbon felt followed by drop-casting of [Au₉(PPh₃)₈](NO₃)₃ clusters from DCM (Fig. SI4c). Briefly, the sprayed layer of carbon black onto a carbon felt was prepared, following the abovementioned method. The carbon felt was heated at 120 °C for 30 min after spraying to remove traces of solvent. The carbon felt was allowed to cool to room temperature, and Au₉ clusters were then drop-casted from a 3 mg ml⁻¹ solution of Au₉ clusters in DCM. The electrode thus prepared was dried in a vacuum overnight to remove traces of DCM. These electrodes will be referred to as the Au₉ DC electrode.

Electrochemical reduction of CO₂ was carried out on a custom-made three-electrode flow cell. The cell consists of an Au clusters-based cathode, a Pt plate as the counter electrode, and a Ag|AgCl (saturated KCl) reference electrode. Constant potential electrolysis of CO₂ was carried out at −1.3 V vs Ag|AgCl and −1.7 V vs Ag|AgCl reference electrode using a Gamry Reference 3000 potentiostat. 0.2 M KHCO₃ (Sigma-Aldrich, 99.7%), used without any purifications and prepared with 18.2 MΩ cm deionised water, was used as catholyte. The electrolyte was bubbled with CO₂ gas at 20 ml min⁻¹ during the experiment using a mass flow controller Alicat, MC-20SCCM-D. To ensure that the electrolyte was saturated with CO₂, purging was started almost 30 min before the experiment. The CO₂ saturated KHCO₃ catholyte was recirculated from a 100 ml reservoir at 20 ml min⁻¹ through carbon felt, and the flow channels machined on the graphite block using a peristaltic pump (LongerPump^® YZ1515x). The counter electrode and anolyte (1 M KHCO₃) were separated from the working compartment with a fine porous frit. The reference electrode was held in a lugging capillary placed in the anode chamber. Nafion 117 membrane was used to separate the anode and the cathode chamber along with the flow of the proton. Electrochemical impedance spectroscopy was studied to determine the resistance between the cathode and reference electrode; this resistance was then compensated using positive feedback during potentiostatic electrochemical reduction. All the electrochemical measurements were carried out at room temperature.

The headspace of the electrolyte reservoir was vented directly to a gas chromatograph (SRI 8610C) equipped with a 6" haysep-D column, TCD, and methanizer-FID detectors to quantify H₂ and CO during the electrolysis. Samples were injected into the column through the sampling loop continuously every 15 min. The liquid products of CO₂ electrolysis were analysed by High-Performance Liquid Chromatography (Thermo Scientific™ UltiMate™ 3000) using the electrolyte after completion of the experiment.

X-ray absorption spectroscopy (XAS) measurements on clusters were carried out in fluorescence mode at the Australian Synchrotron in Clayton, Victoria, Australia, using a multipole wiggler XAS beamline (12-ID) operating with an electron beam energy of 3.0 GeV and a beam current of 200 mA (maintained using the top up mode). The Au-L₃ data were collected using a Si(111) monochromator and focusing optics, and the measurements were performed at 10 K helium cryostat.

Raw data obtained at the beamline was converted using Sakura software ³⁵ and processed using Athena software ³⁶ by extracting the EXAFS oscillation χ(k) as a function of photoelectron wave number (k) following standard procedures. For all the clusters, the k-range was chosen from 2.5 to 15 Å^–1. Fittings of EXAFS were performed using a fuzzy degeneracy approach, which allows grouping the path length in a bin of user-defined width. ^37,38 The contribution of smaller hydrogen atoms in the EXAFS spectra is expected to be very weak due to the low scattering power of this species; thus, it is ignored in the FEFF calculations. ³⁹ More details on the analysis of the XAS spectrum are given in the supplementary information.

Atomically precise clusters ([Au₉(PPh₃)₈](NO₃)₃) were synthesised following the reported procedure and deposited onto Vulcan carbon support. ³³ More details on the synthesis, purification and characterisation of pure clusters are provided in the supplementary information. The electrodes were prepared by spraying Au₉/C catalyst with 5 wt% Nafion (Fig. SI4b), with the carbon fibres found to coat uniformly with a layer of Au₉/C aggregates (Fig. 1a). This uniform deposition over the carbon fibres within the felt allows for easy electrolyte penetration and gas release during CO₂ electrolysis. The SEM image (Fig. 1a, inset) clearly shows that nanometer-sized carbon aggregates containing Au₉ clusters which are embedded in the Nafion matrix forming a uniform network conducive to electron transport.

Zoom In Zoom Out Reset image size

The activity of Au₉/C catalyst deposited carbon felt was assessed using potentiostatic reduction at −1.3 V vs Ag|AgCl (sat. KCl) in a three-electrode system. CO and H₂ were major products for the electrolysis (Fig. 1c), along with a small fraction of formate, consistent with other gold-based cathodes. ^7,40–43

The overall current density and CO Faradaic efficiency were observed to improve over time, reaching almost 75% after 12 h of electrolysis (Figs. 1b and 1c). The increase in both current density and selectivity towards CO suggests the availability of more active sites over time. The increase in activity over the electrolysis run is contrary to the previous reports, ^44–46 where the catalyst undergoes deactivation during electrolysis due to various reasons such as metal impurities in the present aqueous electrolyte, ^47–49 surface poisoning due to chemical species, ^50–52 change in surface morphology of catalyst over time. ^46,53 It should be noted that in the beginning, the total current efficiency was not closer to 100% (Fig. 1c); this can be the result of the lower current density (< 1 mA cm⁻²). The production rate for the products will be low for such current densities to be detected by Gas Chromatography efficiently.

Due to ultrasmall size Au₉ clusters (ca. 0.8 nm), it is expected that most of the Au atoms are on the surface (ideally, 100% Au atoms are on the surface from the known crystal structure); hence a bigger Au oxide reduction behaviour was expected. Interestingly, the as-made electrode does not show typical Au oxide reduction behaviour ⁵⁴ in cyclic voltammogram (Fig. 1d), likely due to considerably different oxidation behaviour of molecule-like Au clusters compared to bulk gold or inaccessibility of electrolyte to Au clusters, which may be located in pores of Vulcan XC-72R carbon black. ^24,55 However, after 12 h of electrolysis at −1.3 V vs Ag|AgCl, the typical Au/AuO_x peak can be seen, suggesting the sintering of at least a fraction of the Au clusters into bulk-like Au nanoparticles (which are known to exhibit the Au/AuO_x redox feature), in line with the previous reports. ^24,56

Unusually, although Au particles are growing in size during CO₂ electrolysis, the catalytic activity is found to increase, suggesting that despite the lower dispersion of gold atoms in the electrode, the activity of the gold sites is increasing. Clearly, the performance of an electrocatalytic layer depends not only on the number of catalytic particles (or active sites) but also on their accessibility to the reactants. Previously it has been found that sulfonate groups in Nafion ionomers get adsorbed onto the Pt surface, resulting in blockage of active sites hence an increased overpotential or poor activity. ^29,57–59 Here, the increase in catalytic performance could result from a favourable interface restructuring during prolonged electrolysis.

As the interfacial rearrangements triggered by electrolysis are expected to be accelerated at a higher electrolysis current, the activity of the catalyst was accessed before and after the electrochemical treatment of the electrode at a significantly more negative potential. The electrochemical CO₂ reduction was performed at −1.3 V vs Ag|AgCl, then the applied cathode potential was decreased to −1.7 V vs Ag|AgCl for 2 h, and finally, the potential was stepped back again to −1.3 V vs Ag|AgCl (Fig. 2a).

Zoom In Zoom Out Reset image size

The electrochemical activation of the Au₉/C electrode is clearly seen to improve both the current density and CO Faradaic efficiency (Figs. 2b and 2c). One explanation for these results is that this activation procedure alters the morphology of the Au₉-Nafion interface, thus enabling better reactant access to or less blocking of the active sites. This increase in active site availability is then preserved with the potential returned to −1.3 V vs Ag|AgCl, and thus the restructured catalyst exhibits higher activity. It is worth noting that while the CO Faradaic efficiency before and after electrochemical activation is almost the same irrespective of the amount of Nafion in the catalyst layer, the increase in current density after electrochemical activation was less for the higher amount of Nafion in the electrode (Figs. 2b and 2c). This would be expected for the activation process, which relies on "un-blocking" sites poisoned by Nafion, as a higher amount of Nafion in the layer would make almost certainly make the activation process more difficult.

In order to investigate possible interactions between the clusters and Nafion ionomer, methanol solutions containing the clusters and ionomer were prepared. Upon mixing, the characteristic colour of the Au₉ clusters in methanol was reduced, and a precipitate was obtained within seconds. UV–vis spectra of the supernatant were recorded to quantitively estimate the fraction of precipitate with a different mass ratio of Nafion ionomer and Au₉ clusters (Fig. 3a).

Zoom In Zoom Out Reset image size

The Au₉ clusters show UV–vis absorption peaks at 443, 375, 352, and 314 nm, consistent with the previous reports. ^60,61 After mixing the Au₉ clusters with the Nafion ionomer, while the supernatant showed UV–vis absorption peaks at the same wavelengths as that of the pure Au₉ cluster solution, the peak intensities decreased. The absorbance at 314 nm (which is proportional to the concentration of Au₉ clusters in the solution) reveals that the Nafion: Au₉ ratio had a strong influence on the concentration of Au₉ clusters that remain in the solution (Fig. 3b). This suggests that the interactions between Au₉ clusters and Nafion ionomer lead to the formation of a precipitate, whilst a fraction of clusters remains in solution and that this interaction is strongest at the Nafion:Au₉ weight ratio of 0.75:1. This change in the UV–vis spectra of Au₉-Nafion mixture is significantly different from previously reported Au₂₅-Nafion system, ²⁶ where the authors observed that one peak shifts to higher wavelength and some of the peaks disappear upon mixing cluster and Nafion ionomer, and this attributed these changes to alteration or agglomeration of intact Au₂₅ clusters.

The precipitate obtained from the Au₉ clusters and Nafion ionomer was studied using X-ray absorption spectroscopy. The X-ray absorption near edge structure (XANES) spectra of the Au₉ cluster and the Au₉-Nafion precipitate are significantly different compared to a metallic Au foil (Fig. 3c). The absorption edge of the Au₉ clusters is shifted to a higher energy, and clusters show a higher white line intensity compared to metallic gold, which is in line with previous reports on ultra-small clusters. ^62–66 A more detailed discussion comparing XANES of the Au₉ clusters with bulk metal has been given in the supplementary information. Interestingly, XANES of the precipitate shows an absorption edge (determined by the peak in the dI/dE spectrum, Fig. 3c, inset) at a lower energy compared to that of the Au₉ clusters, and the "white line" for the Au₉-Nafion precipitate is less intense compared to Au₉ clusters. This is most likely because the interactions between positively charged Au₉ clusters (i.e. [Au₉(PPh₃)₈]³⁺) and negatively charged Nafion ionomer would lower the overall charge per Au atom, resulting in a shift in the absorption edge and lowering the "white line" intensity. ⁶⁷ The change in electronic properties of clusters was previously observed as the origin of an XPS 4f_7/2 shoulder when Au₂₅ clusters mixed with Nafion ionomer. ²⁶ The difference becomes clearer in EXAFS (Fig. 3d), which suggests that the local environment around the Au atoms has changed in the precipitate formed after mixing. The EXAFS of the precipitate was fitted with one Au–O, one Au–P, and two Au–Au bonding environments, and the result suggests a bonding between cluster core and O from the Nafion ionomer (Table I and Fig. SI8). The structure of the Au species in the precipitate has notably distorted compared to pure Au₉ clusters; a detailed description of fitting the clusters is given in the supplementary information. Previously, based on the change in XPS and UV–vis spectra, Andrews et al. proposed that mixing ultra-small clusters with Nafion may have resulted in coordination between clusters and the sulphonate group. ²⁶

In order to investigate the change in the Au₉/C electrodes (prepared using 5 wt% Nafion ionomer) during electrolysis, the Au L₃ edge X-ray absorption spectra of the electrodes in as-made form, after potentiostatic (−1.3 V vs Ag|AgCl) CO₂ reduction and after electrochemical activation were studied. The gold species in all the electrodes tested had absorption edges (determined by the peak of the derivative spectra) at higher energies compared to the bulk gold but lower than the edge energy observed for the pure Au₉ clusters (Fig. 3c). The shift in absorption edge for the electrodes relative to bulk gold suggests the presence of a fraction of Au₉ cluster-like species. Moreover, the systematic decrease in the absorption edge shift (compared to bulk gold) of the electrode after electrolysis and after electrochemical activation suggests a decrease in the population of Au cluster-like species in samples. ^67–69 The "white line" of all the electrodes is more intense than that of the bulk gold but less intense than that of Au₉ clusters (Figs. 4b and SI9a), suggesting that gold species within the electrodes have a higher density of unoccupied d states compared to the bulk gold. ^69,70 The XANES features of fcc gold (regions c and d) are visible in the spectra of all three electrodes, but they are attenuated compared to the bulk gold, which is due to the lower average coordination number of Au atoms in the electrode of interest compared to the bulk gold (i.e. CN = 12). ⁶⁷ The intensity of XANES features at c and d increases after electrolysis and increases further after electrochemical activation, suggesting an increase in bulk gold-like species in the sample population. As previously demonstrated, ³⁴ linear combination analysis using Au₉ clusters and bulk gold as the main component can be used to study the agglomeration of clusters (Fig. SI10). Here, we observed that even the as-made electrodes have a fraction of bulk-like gold (∼20%), and the fraction of bulk-like gold (based on the linear combination fitting of the XANES spectra) was observed upon electrolysis, with the electrochemical activation further increasing the bulk-like gold features to 40% (Fig. SI10d). The agglomeration of clusters into bulk-like nanoparticles after CO₂ electrolysis is in line with the appearance of the Au/AuO_x peak in the cyclic voltammogram after the constant potential electrolysis experiment (Fig. 1b).

Zoom In Zoom Out Reset image size

One explanation for the as-made electrode containing a significant fraction of bulk-like gold is that during the electrode fabrication process, the electrodes are heated to 120 °C in order to dry the ink sprayed onto the electrode substrate. Even though it is known that the Au₉ clusters in crystalline form remain stable up to 200 °C (Fig. SI3), ⁷¹ our results suggest that a fraction of Au₉ clusters deposited onto carbon starts agglomerating at 120 °C, which is in line with a report from Longo et al. ²⁵ The EXAFS of the three electrodes is also significantly different from the as-made clusters, with these electrodes exhibiting a peak around 2.7 Å, corresponding to the bulk-like gold features ²⁴ (Fig. 4b). The EXAFS spectra of electrodes were fitted with two Au–Au scattering environment groups, similar to the EXAFS spectra of Au₉ clusters and one Au–P/O bond (Fig. SI11). The EXAFS fitting parameters shown in Table II suggest that the clusters upon electrolysis undergo structural changes without significant change in Au–P/O coordination, and Au–P/O coordination decreases significantly upon electrochemical activation (Table II). Although the as-made electrode is significantly agglomerated, it is expected that the clusters will lose some phosphine ligands, but surprisingly it maintains the Au–P/O coordination number closer to the crystalline Au₉ clusters (i.e., CN = 0.889, see Table SIIII). This suggests that a fraction of Au atoms bind with the sulfonate group from the Nafion ionomer, as observed in the physical mixing of the clusters and ionomer. While this data suggests that the activity of the electrode scales inversely to the fraction of Au cluster-like species within the electrode (the as-made electrode has the highest fraction of intact Au cluster-like species but is the least active), it is important to consider the availability of the gold sites for the reaction. It can be seen that the CN for the Au-P/O decreases with electrolysis and is lowest after activation. In principle, agglomeration of clusters to larger particles would avail lesser surface atoms for not only ligand coordination but also for the reactant. This should result in a lower activity upon electrochemical activation. Contrary to this, we observed higher activity which could be explained by the hypothesis that the bulky sulfonate group bonded to the metallic core can block access of reactants to the active sites, but when the electrochemical potential is applied, the Au₉-Nafion matrix realigns itself, and Au–P/O coordination is lost, allowing the reactants to reach the Au core, and resulting in the higher conversion. Adsorption of the Nafion ionomer onto the surface and blockage of active sites has been extensively documented in the literature. ^26,28–32

From the above discussion, it is clear that the Nafion ionomer interacts with Au₉ clusters, which results in the partial blockage of active sites in the Au₉/C catalyst. These interactions are likely to occur in the catalytic ink dispersion, where both the Nafion ionomer and the Au₉/C catalyst are present (as discussed in the experimental part and Fig. SI4a). ³² These interactions can be minimised by first spraying a film of carbon black with Nafion ionomer and then drop-casting the Au₉ clusters to the dry layer, thereby avoiding the solution phase interactions between the Au₉ clusters and Nafion ionomer. In addition, using a low boiling point solvent for drop-casting the active clusters onto a pre-made layer of carbon black will have the advantage of uniform and efficient delivery of clusters throughout the catalytic layer. Thus, the cathodes were prepared by spraying a film of carbon black onto the carbon felt, followed by drop-casting Au₉ clusters from the dichloromethane (DCM) solution to achieve an Au loading of 90 μg cm⁻².

No significant difference in morphology was observed between the carbon black - Nafion layers and the layers on which Au₉ clusters were dropcast (Figs. 5a and 5b). These layers were also similar in morphology to the spray-coated layers produced from the Au₉/C (Fig. 1a). The electronic and structural characteristics of the as-made Au₉ drop-cast electrodes, as observed from XAS, are similar to that of the Au₉ clusters (Fig. SI12), with the XANES spectra of Au₉ clusters and the Au₉ drop-cast electrode being nearly superimposable (Fig. SI12b). Additionally, the EXAFS fitting parameters of the as-made electrode are close to that of the crystalline Au₉ clusters (Fig. SI13 and Table SIV), which is not unexpected due to the very low gold loading (90 μg cm⁻²) on the high surface area support layer. ²⁵ This further suggests that the interactions between Au₉ clusters and Nafion ionomer in dropcasted electrodes were not significant enough to change the geometrical and electronic properties of the clusters.

Zoom In Zoom Out Reset image size

The dropcasted Au₉-based electrodes demonstrated nearly 80% selectivity towards CO₂ to CO conversion at −1.3 V vs Ag|AgCl (for electrodes with 5% Nafion) before electrochemical activation. Interestingly, the activity of the dropcasted electrodes before electrochemical activation is notably higher compared to the Au₉/C-based electrodes. Due to the absence of notable interaction between the sulfonate group of Nafion and the metal core of Au clusters, the blockage of active sites remains low, which can explain the higher activity of the dropcasted electrodes. In addition, unlike the electrodes produced from the Au₉/C catalyst, the dropcasted Au₉ electrodes do not show similar activation behaviour, with the activity before and after the electrochemical activation procedure being almost the same (Figs. 5c and 5d). The improvement in activity of the Au₉/C catalyst is most likely caused by the potential-driven restructuring of the Au₉-Nafion interface where the cluster core was blocked by interaction with the sulfonate group from Nafion. The dropcasted Au₉ clusters-based electrodes do not have such interactions; therefore, no significant change in activity was observed upon electrochemical activation at higher potential. However, based on cyclic voltammetry (Fig. SI14) and the XAS of the electrodes before and after electrolysis (Fig. SI12), it is clear that a fraction of the clusters undergo aggregation after electrochemical CO₂ reduction. ⁵⁴

Interestingly, the activity and selectivity of electrodes formed by dropcasting the Au₉ clusters onto pre-made carbon black–Nafion layers are found to be strongly dependent on the Nafion content of the support layer (Figs. 5c and 5d), suggesting that the dropcast Au clusters still interact with the Nafion, but in a way which is not amenable to activation. When the Au₉ clusters are dropcasted from DCM to the sprayed carbon black-Nafion layer, the clusters in solution are in contact with the Nafion ionomer momentarily, which could be the reason for pronounced interaction at higher Nafion content.