Facet-Dependent Selectivity of Cu Catalysts in Electrochemical CO2 Reduction at Commercially Viable Current Densities.

Despite substantial progress in the electrochemical conversion of CO2 into value-added chemicals, the translation of fundamental studies into commercially relevant conditions requires additional efforts. Here, we study the catalytic properties of tailored Cu nanocatalysts under commercially relevant current densities in a gas-fed flow cell. We demonstrate that their facet-dependent selectivity is retained in this device configuration with the advantage of further suppressing hydrogen production and increasing the faradaic efficiencies toward the CO2 reduction products compared to a conventional H-cell. The combined catalyst and system effects result in state-of-the art product selectivity at high current densities (in the range 100-300 mA/cm2) and at relatively low applied potential (as low as -0.65 V vs RHE). Cu cubes reach an ethylene selectivity of up to 57% with a corresponding mass activity of 700 mA/mg, and Cu octahedra reach a methane selectivity of up to 51% with a corresponding mass activity of 1.45 A/mg in 1 M KOH.

Although in the last decades several advances have been recorded in the field of renewable energy, these sources are discontinuous and the energy generated cannot easily be stored for long periods of time.[1−3] In this scenario, the electrochemical CO2 reduction reaction (CO2RR) represents an appealing process that can be conveniently integrated with various renewable energy systems for producing carbon-based chemical feedstocks and fuels.[4,5] However, this method must satisfy several requirements in order to become an economically valuable solution. At present, CO2RR suffers from poor efficiency due to the CO2 thermodynamic stability and to the reaction kinetic impediments, which result in the need for a large overpotential to activate and convert this molecule to more energy-dense products.[6−8]

Among the transition metals, copper is the only one capable of driving CO2RR toward longer chain hydrocarbons and alcohols at reasonable faradaic efficiencies (FEs).[9−16] Several studies highlight that the obtained reduction products depend on multiple factors, including the environmental pH,[14,15] the nature of the electrolyte,[17−19] the applied potential,[10,20] the diffusion mechanism of CO2,[21,22] as well as the chemical and morphological characteristics of the catalyst itself.[23,24] These conclusions have been mostly obtained by experiments performed in the traditional H-cell, where CO2 is dissolved in an aqueous electrolyte (the most common being 0.1 M KHCO3) and the catalysts are primarily deposited on flat glassy-carbon electrodes.[25] While these testing conditions can provide valuable information about the intrinsic activity of different catalytic materials, current densities are limited by the low catalyst loading allowed on the flat electrode and, more importantly, by the low solubility of CO2 in water and the long CO2 diffusion pathway to the catalyst surface.[21,22,26]

Recently, several studies have performed CO2RR in gas-fed reactors capable of sustaining high current densities with acceptable faradaic efficiencies.[21,22,26,27] Using this approach, a very high surface area porous catalyst layer can be employed, so that CO2 mass limitations to this surface are greatly reduced, resulting in substantially higher current densities at lower overall potentials.[28,29] Several architectures have been proposed and realized to achieve these goals, including flow cell configurations based on gas-diffusion electrodes (GDE).[26,30−32] With this design, conversion efficiencies for ethylene up to 70% at a potential of −0.55 V vs RHE (reversible hydrogen electrode) and current densities up to 300 mA/cm2 have been reported in the presence of a highly concentrated alkaline electrolyte (10 M KOH).[33]

Investigations on Cu single crystals in an H-cell have revealed that the (100) surface is more selective for C2H4, whereas the Cu (111) surface is more selective toward CH4 under CO2RR conditions.[34−37] Nanoparticle-based studies, including our own, have also demonstrated that such facet-dependent structural selectivity is maintained down to the nanoscale and can be further tuned through size effects.[38−42] Decade-long studies on nanocatalysts for the oxygen reduction reaction, however, have evidenced that tests under idealized conditions do not predict the catalytic activity and stability under conditions relevant for commercial-scale reactors.[43,44] For this reason, the impressive performance in fundamental studies has never been reproduced into a membrane electrode assembly used in fuel cells.[43,44]

When we consider the development of catalysts for CO2 electrolysis, the intrinsic nature of the reaction warrants a similar degree of scepticism. In particular, in order to reach substantial current densities due to the mass transport limitations of CO2 in aqueous electrolytes, a catalyst must be incorporated into either a GDE or a membrane electrode assembly. Under these conditions, the catalyst layer now functions as a porous electrode due to the electrolyte/ion pathway on one side of the catalyst and the CO2 diffusion pathway on the other.[21] Such configurations are then expected to provide more surface area and reach higher current densities at lower applied potentials, which can impact not only potential-dependent product formation but also the restructuring and stability of facets on metal surfaces. Further, as current densities are increased, the reaction environment becomes substantially different from those achievable within controlled H-cell systems.[21] Indeed, multiphysical transport modeling of mesostructured silver electrodes has already shown that the improved selectivity toward CO compared to flat silver electrodes is purely a result of mass transport effects.[45] Similar effects to those described above could play a role when faceted Cu nanoparticles are integrated into a GDE. A recent study of (100) faceted Cu nanoparticles in alkaline gas diffusion electrolyzers has revealed that the systems were similarly selective toward ethylene as in an H cell.[46] Nevertheless, an increased local pH as a result of high current densities in a GDE would also promote formation of ethylene,[15,28,33] and so a definitive mechanism for the apparent similarity in selectivity is still not clear. Therefore, assessing various shaped Cu nanoparticles in different electrode architectures can help to provide further insights to the influences of surface facets, surface area, and the reaction environment on catalytic selectivity.

Having this in mind, we sought to directly compare the facet-dependent activity and selectivity of differently shaped Cu nanoparticles at much higher current densities than those achievable in an H-cell. By making this direct comparison, we can determine if the observed performance is maintained or altered in a GDE configuration compared to an aqueous H-cell and thus provide meaningful insights to the CO2 reduction community that will shed light on an open debate.

Cu nanocrystals (NCs) of different shapes were synthesized according to colloidal methods previously reported by our group (see Experimental Section).[38,39,47]Figure provides an overview of the morphological and structural characterization of the NCs. Figure a–c reports the transmission electron microscopy (TEM) images of the as-obtained Cu spheres of 6 nm (Cusph), Cu cubes of 44 nm (Cucub), and Cu octhaedra of 150 nm (Cuoh), which all possess high uniformity in size and shape. These structures were chosen as they were found to be separately optimally selective for ethylene, methane, and a mixture of C1–C2 products, respectively, in H-cell tests.[38−42] X-ray diffraction (XRD) patterns in Figure d show the characteristic preferential orientation for the Cucub and Cuoh, which is along the (100) and (111) directions, respectively, whereas the Cusph’s exhibit both orientations and a peak ratio closer to the bulk copper reference. More extensive characterization of the same NCs is reported in our previous work.[38,39,47] The as-synthesized NCs were spray-coated or drop-casted onto the GDL to form the gas diffusion electrode (GDE), as described in the Experimental Section. No particular difference was observed between the two deposition techniques upon optimization (Figure S1). Figure e and 1f reports representative in-plane scanning electron microscopy (SEM) images of the Cucub and Cuoh NCs deposited on the GDL for the highest loadings used in this work, revealing a high degree of coverage.

Figure 1

(a, b, and c) TEM images of the as-synthesized Cusph, Cucub, and Cuoh, respectively. (d) XRD patterns of the obtained materials along with the Cu reference pattern (PDF no. 04-0836). (e and f) SEM images of Cucub (440 μg/cm2) and Cuoh (200 μg/cm2) NCs which were spray-coated and drop-casted, respectively, on a Sigracet BC39 GDL.

The catalytic performance of the as-prepared GDEs was tested in a gas-fed flow cell with 1 M KOH as the supporting electrolyte (Experimental Section, Table S1). This setup was reproduced from previous literature, and the alkaline electrolyte was chosen because of the reported record FE toward ethylene of 36% and 66% with partial current densities of 150 and 184 mA/cm2 in KOH 1 and 10 M, respectively, at an applied potential of around −0.55 V vs RHE.[28,33]Figure reports the faradaic efficiencies (FEs) obtained for the three NCs in the gas-fed flow cell along with the partial mass activities and current densities normalized by the electrochemically active surface area (ECSA) at three representative values of iR-corrected potentials. The current densities normalized by the geometric area are shown in Figure S3; the corresponding CO2 conversion efficiency and cell voltages are reported in Figure S4.

Figure 2

(a) Faradaic efficiencies vs potential for Cusph (200 μg/cm2), Cucub (250 μg/cm2), and Cuoh (50 μg/cm2) deposited on a GDL and measured in the gas-fed flow cell in 1 M KOH. These loadings were chosen to ensure similar catalyst coverage of the GDL, similar ECSA (Figure S2, Table S2), and conditions far from a mass transport-limited regime. Detailed discussion is reported in the SI. Black dots in a represent the geometric current density for each case (right axis). Higher values of current densities were not achievable with our current setup. (b–d) Mass activities (left axis) and partial current density normalized by the ECSA (right axis) for each of the detected gas products vs potential for the three NCs studied. All potentials are iR corrected (see SI for details).

At a current density of 100 mA/cm2 at −0.69 V vs RHE, Cusph NCs produce 28% CO and 20% ethylene. At a higher current density (200 mA/cm2, −0.79 V vs RHE), the FE for CO drops to 16%. Concomitantly, ethylene becomes the main CO2RR reaction product (FE ≈ 31%), which suggests that more CO molecules undergo coupling. Finally, when the Cusph’s are investigated at 300 mA/cm2 and −0.86 V vs RHE, HER increases (FE ≈ 43%), CO further decreases (FE ≈ 4%), as well as C2H4 decreases (FE ≈ 25%). The mass activities in Figure b follow the same described trend. Overall, the behavior of the Cusph (Figure a and 2b) resembles that of polycrystalline copper tested in a similar device configuration.[21,29−31,33]

When turning to the Cucub’s, they exhibit a much higher selectivity toward ethylene compared to the spheres across all of the potentials (Figure a). This result points at the fact that the exposed (100) facets do play a role in directing selectivity, even under these more extreme conditions. Specifically, conversion of CO2 to C2H4 ranges from 55% at 100 mA/cm2 and −0.65 V vs RHE to around 60% at 200 mA/cm2 and −0.70 V vs RHE. Finally, at 300 mA/cm2 and −0.75 V vs RHE, the observed conversion to ethylene is ∼57%. The corresponding ethylene mass activities are reported in Figure c and vary between 200 and 700 mA/mg, higher than the mass activities of the other products across the whole potential range. These efficiencies overcome those previously reported for randomly shaped Cu NCs in 1 M KOH, which were 36% at −0.58 V vs RHE at a similar ethylene partial geometric current density (150 mA/cm2) and with lower ethylene mass activities of around 176 mA/mg.[28]

As for the Cuoh’s, methane is the main hydrocarbon product, in line with the presence of the exposed (111) facets. The highest FE of ∼53% is obtained at 100 mA/cm2 and −0.91 V vs RHE, with the corresponding lowest H2 production (FE ≈ 22%). Ethylene is also present as a product but only with an ∼10% conversion efficiency at this current density, giving a methane:ethylene ratio of almost 5:1 compared to the roughly 1:20 ratio for the Cucub’s. At the higher current densities and potentials (200 mA/cm2 and −0.93 V vs RHE; 300 mA/cm2 and −0.96 V vs RHE), HER increases substantially and a general decrease of C1 and C2 products is observed, most likely due to the reaction entering a CO2-depleted regime. Yet, methane does still remain the major CO2RR product with FE around 40%. The mass activities in Figure d are consistent with the selectivity trend, i.e., the methane and hydrogen current densities are higher than those for CO and ethylene across the whole potential range. To the best of our knowledge, the catalyst’s mass activity toward methane is the largest reported to date and varies from 1.00 to 2.5 A/mg for potentials in the range from −0.91 to −0.96 V vs RHE.

Compared to the results obtained from the same NCs tested in an H-cell (Figure S5), the major CO2RR products are preserved yet the hydrogen production is lower in the gas-fed flow cell at all potentials. The high alkaline conditions may explain this result as the CO2 reduction reaction remains constant on the SHE scale while the hydrogen evolution reaction does not.[48] As CO2 will interact with the alkaline electrolyte over time, it is also important to ensure that these changes do not impact the conclusions made over the length of the experiment.[49] For this reason, the pH of the bulk electrolyte was measured before and after the experiment as shown in Table S3. As the pH remains highly alkaline even at the end of the experiment, the slow degradation of the KOH electrolyte with CO2 was not considered as a factor in these specific experiments.

Considering the notable differences in CH4 vs C2H4 activity obtained with the faceted NCs, the following discussion will only focus on the Cucub and Cuoh NCs and dive further into the influences of catalyst loading and operating conditions for these two catalysts.

Catalyst loading has been shown to affect the electrochemical performance due to the reconstruction of the deposited particles as well as by affecting transport at the mesoscale.[50,51] Furthermore, it is unknown whether the whole amount of catalyst loaded onto a GDL, or only part of it, reacts with CO2.[21,33] For these reasons, it is important to investigate the effect of increasing catalyst loading on CO2 depletion and availability while maximizing the electrochemically active surface area to reduce applied potentials. It is worth noting that reductive stripping of the native ligands from the catalyst surface occurs (Figure S6). Therefore, similar to what was concluded in comparative experiments done in an aqueous H-cell, the ligands do not have any major impact on the electrochemical performance.[52,53]

Figures , 4, and S7 summarize the results related to the effect of loading for the Cucub’s, while those for the Cuoh’s are reported in the Supporting Information (Figures S8–S10). Figure a shows a representative cross-sectional SEM image for the Cucub NCs deposited on the GDL with a loading of 440 μg/cm2, and Figure b illustrates the cross-section energy-dispersive X-ray spectroscopy (EDX) colored maps for the same NCs at different loadings. The images evidence that the as-deposited NCs are uniformly distributed through the whole GDL thickness as proved by the presence of a NC layer on the fibers of the backing paper support.

Figure 3

(a) Cross-sectional SEM images of 440 μg/cm2 Cucub NCs loaded on the GDL together with a magnification of the same. (b) EDX colored map cross-section of Cucub for different loadings: 100, 250, and 440 μg/cm2 from left to right. (c) In-plane SEM images of the samples in b with an inset of the same at higher magnification.

Figure 4

(a) Geometric current density vs iR-corrected potentials in 1.0 M KOH for Cucub NCs on a GDL at varying loading. (b–d) Corresponding C2H4, CH4, and H2 faradaic efficiencies. Error bars indicate the standard deviation of three independent samples.

In Figure b, one can notice that a surface layer builds up on the top of the GDL. Figure c shows the top-down SEM images of CucubNCs with the same loadings of Figure b. As the loading increases, the Cucub’s form an increasingly more compact layer on the top of the GDL. Very similar observations were made for the CuohNCs, though formation of such a compact top layer occurred at lower loadings (200 μg/cm2) than the CucubNCs (440 μg/cm2), most likely because of the bigger NC size (Figuress S8 and S9).

The CO2RR performance of the Cuoh and of the Cucub NCs was evaluated at loadings in the range of 50–1000 μg/cm2 (Figures , S7, S10, and S11 and Tables S4 and S5). Figure a shows that the required potential can be driven down at higher loading as the catalytic surface area increases. At lower current densities, the CucubNCs exhibit similar slopes for all loadings, indicative of a similar level of intrinsic activity. As the current increases above 100 mA/cm2, the slope also changes. Such change is more pronounced for 440 and 1000 μg/cm2, which are higher loadings than those utilized in Figure (250 μg/cm2). The deviation from a linear dependence of the current density on the potential in this high-current/high-loading region is likely a combined result of mass transport effects and changes in overall activity due to modifications of the local environment. Similar observations were made for the Cuoh (Figure S10), though with even more drastic effects. Indeed, above 200 μg/cm2, which is when the thick top layer forms, only vigorous gas bubbling from hydrogen production was observed on the GDE surface (Figures S8 and S9).

To gain further understanding about the accessibility of the electrolyte to the NCs, EDX analysis of the potassium through the GDE thickness was performed (Figure S12). It is reasonable to assume that the presence of potassium indicates a wetted electrode and that, therefore, all of the NCs are in the condition to be potentially active. We found that the increasing loading is accompanied by a decreasing concentration of potassium in the GDE up to the point of not detecting any potassium inside when the top compact layer forms. These results suggest that the top compact layer effectively prevents the electrolytes from penetrating into the pores, thus effectively impeding the reaction between CO2 and water.

In agreement with this compositional analysis, the ECSA for the Cucub notably increases with a loading of up to 440 μg/cm2 while the change becomes less significant between 440 and 1000 μg/cm2 (Figure S2). Instead, the results for the Cuoh evidence that 50 μg/cm2 is already enough to maximize the active surface area, which we speculatively assign to the bigger size of the octahedra (Figure S2).

When analyzing the product selectivities at different loadings (Figure b–d), the FEs of the intermediate loadings (250 and 440 μg/cm2) show an overall more moderate potential dependence. On the contrary, for the lower and higher loadings (50, 100, and 1000 μg/cm2), the FEC2H4 and FEH2 decrease and increase, respectively, while going from lower to higher potentials (Figure b–d). For the samples with the lower loading, the HER increase is accompanied by increased methane as well (Figure d).

In order to explain the observed product distribution, various effects need to be considered. First, there are the aforementioned mass transport resistances. For the 1000 μg/cm2 sample, the increased HER and decreased ethylene can be justified in terms of mass transport limitations, causing CO2RR to be replaced with HER as current densities increase. Nevertheless, between 200 and 250 mA/cm2 HER slightly decreases for both 440 and 1000 μg/cm2, while polarization curves show already a decreasing slope. Thus, mass transport limitations cannot be the only explanation. Mass transport limitations also do not explain the increased CH4 at higher potential for the lower loadings.

One could then consider the effect of polarization. It has been reported that at high overpotentials (<−0.8 V vs RHE), C2H4 and CH4 can form simultaneously from the common intermediate *COH on both Cu (100) and Cu (111) facets.[15,34,35] This behavior however has not been observed during experiments in alkaline media.[15,54,55] We speculate that the negative potential applied to the system to meet the set current density at such low catalyst loadings (lesser exposed surface area) makes CH4 more favorable than C2H4 at these potentials. However, polarization effects do not justify the high methane production of the 100 μg/cm2 compared to the 250 and 440 μg/cm2 for similar applied potentials (−0.77 V for 200 mA/cm2, −0.75 V for 300 mA/cm2, and −0.76 V for 250 mA/cm2).

If we plot the FEs versus the applied potential (Figure S13), it becomes clear that another factor to consider is the optimal electrode potential range for a given product, which indeed changes for different loadings. These data contribute to explain the observations above. Finally, the contribution of the GDL itself to the high hydrogen evolution rate (FE ≈ 60%) in the low loadings may be considered as well (Figure S14).

The same effects (i.e., mass transport limitations, polarization effects, optimal potential range, substrate effects) explain the observed behavior for different loadings of CuohNCs (Figure S13).

Finally, catalyst and device stability is as important as activity and selectivity. TEM and XRD analyses show that the morphological stability of the Cucub is preserved up to 6 h, and it is accompanied by stable ethylene production (Figures S15 and S16). These initial data point at an improvement of the intrinsic stability compared to what we previously observed in an H-cell.[52] The lower potentials needed in the gas-fed flow cell (−0.76/–0.65 V vs RHE with respect to the −1.1 V vs RHE in the H-cell) are one possible reason for the observed behavior. On the contrary, Cuoh did not show high stability (Figures S17 and S18). In the case of the more stable Cucub, after 6 h, the entire device stops operating because of GDL flooding, thus suggesting that engineering solutions to achieve device stability for a longer time are crucial before continuing to further investigate the parameters contributing to catalyst stability.[21,56,57]

In summary, this study presents the integration of colloidally synthesized NCs in a gas-fed flow cell with optimal morphologies for high ethylene and methane conversion efficiencies and production rates. We observed that the Cucub NCs are highly selective toward C2H4, Cuoh are selective toward CH4, and Cusph are not selective toward any specific product. Loading experiments demonstrated that an optimal amount of catalyst must be deposited in order to achieve the best performance in terms of activity and selectivity. In particular, the catalyst layer should uniformly cover the GDL without clogging the porous structure. The latter results in a limited mass diffusion operational regime and promotes the subsequent evolution of hydrogen. Future modeling will help to understand mass transport in more detail.[45,58]

Overall, integration of the Cu NCs in the gas-fed flow electrolyzer enabled achievement of remarkable performance for ethylene and methane in a low-concentration alkaline electrolyte. Initial tests have revealed that these selectivities are preserved also in different electrolytes (Figure S19, Tables S6 and S7). Specifically, ethylene mass activities between 200 and 700 mA/mg in a potential range between −0.65 and −0.75 V vs RHE with a selectivity of ∼57% were obtained in 1 M KOH. The methane current densities were exceptionally large and varied from 1.45 to 2.5 A/mg in the potential range from −0.93 to −0.96 V vs RHE with selectivities between 51% and 41% in 1 M KOH. The latter is a very interesting fundamental result which illustrates the power of catalyst design even in gas-fed electrolyzers, as activity toward methane is expected and has generally been shown to be suppressed under high bulk and local pH conditions.[15,28,33]