Introduction

The electrocatalytic reduction of CO2 to useful chemicals on metallic electrodes has become an increasingly important area of research [1,2,3,4]. While many metals have been studied for this process [5,6,7,8,9,10], currently no metal has proved to be both highly active and selective for long periods of time [11,12,13,14,15]. Normally in aqueous media, gold cathodes exhibit high CO selectivity (current efficiency 16–91%) at relatively low overpotentials [16,17,18], with formate found as a minor product (current efficiency < 10%) [5, 6, 9, 19,20,21]. As the other major product from CO2 reduction of gold is hydrogen (from the parallel hydrogen evolution reaction), if even the selectively towards CO cannot reach 100%, the CO and H2 produced can be used as a feedstock for other process requiring syngas.

Despite the high selectivity of Au cathodes towards CO, it is clear that this is not maintained during continuous reduction [15, 22]. This deactivation is also found on other metallic cathodes, and while some have developed methods to maintain CO2 reduction activity on these other metals [11, 13, 14, 23,24,25,26], there has been limited studies focusing on understanding or preventing the deactivation of gold during electrocatalytic CO2 reduction [15, 27]. In general, the deactivation during electrocatalytic CO2 reduction is normally attributed to the poisoning of the cathode by metallic [14, 22, 28] or organic [14] impurities within the electrolyte. Other studies have shown that this deactivation is unavoidable during continuous CO2 reduction due to the accumulation of surface bound intermediates from the reaction [13, 17, 29,30,31,32,33], and in some cases the influence or significance of metallic impurities has been specifically discounted [13, 32, 34].

Preventing electrode deactivation has been investigated by either applying periodic anodic treatments to the cathode [11, 13, 15, 27] or by removing the impurities using pre-electrolysis of the electrolyte [14]. While pre-electrolysis has proved feasible in the laboratory environment [14], others suggest that this approach is ineffective and energy and time consuming [11, 13, 35, 36], and seems unlikely to be successful on an industrial scale due to the ultra-low level of impurities which appear to cause deactivation. While periodically interrupting the CO2 reduction process to strip off either metallic or carbonaceous poisons via anodic treatments has also been successful for Cu, Ag and gold cathodes, little attention has been focused on the long-term stability of the cathodes due to anodic dissolution of the metal during these treatments. For example, it is clear that these anodic treatments will alter the surface morphology of gold cathodes due to gold dissolution [27, 37, 38], which occurs at high anodic potentials upon the formation of gold oxide, as well as during the reduction of gold oxide back to metallic gold [39,40,41,42]. While this dissolution could arguably improve the performance of bulk gold cathodes (by enhancing the roughness factor), if gold nanoparticles are used for CO2 reduction, the anodic treatment used to prevent deactivation will cause dissolution and/or growth of the nanoparticles [43, 44], especially considering that the anodic oxidation of gold nanoparticles occurs more readily than the oxidation of bulk gold [45, 46]. Given these factors and considering that the previous studies did not provide significant insight on the deactivation mechanism and catalyst recovery, it is clear that a systematic study of the electrochemical treatment protocol for gold cathodes during CO2 reduction needs to be developed, with the specific aim of avoiding gold dissolution. Thus, the influence of various periodic cleaning protocols (cyclic voltammetry, open-circuit potential steps and anodic pulses) is investigated in terms of their ability to prevent the deactivation of gold cathodes for CO2 reduction while avoid gold dissolution.

Experimental

Electrochemical experiments were performed using gold disc (99.999%) mechanically polished to a mirror finish using silicon carbide paper and alumina slurries (down to 0.05 μm). The working electrode with a geometric area of 2.01 cm2 was used in a two-compartment gas-tight glass electrochemical cell with a graphite counter electrode, with Nafion 115 used to separate the anolyte and catholyte chambers. The potential of the cathode was measured with respect to an Ag|AgCl (saturated KCl) reference electrode held in a Luggin capillary. All the potentials in this study are referenced to the Ag|AgCl unless specifically stated. CO2 gas (99.995%) was continuously purged at a rate of 40 mL min−1 for 30 min before each experiment to saturate the electrolyte and then the flow rate reduced to 20 mL min−1 during the experiment by using Alicat mass flow controller. The current efficiency (percentage of the total current going towards the formation of a certain product) was determined by gas chromatography (SRI instruments, methaniser FID and TCD detectors, Haysep-D column) every 15 min. High performance liquid chromatography (HPLC, equipped with a SUPELCOGELTM C-610H column) was employed for post-electrolysis analysis of reduction products present within the electrolyte. In all cases, 0.2 M KHCO3 (99.7% ACS reagent, Sigma-Aldrich) was used directly as the electrolyte without any purification. All the experiments were conducted at room temperature (23 °C) using a Gamry Reference 3000 potentiostat. Experiments were carried out at constant current densities for 4 h, with electrochemical impedance spectroscopy (EIS) measurements (100 kHz–10 Hz, 5 mV rms amplitude) every 15 min to measure the electrolyte resistance and correct the measured potential. Periodically measuring the electrolyte resistance is important during long-term CO2 electroreduction experiments as the electrolyte resistance can decrease with time due to selective K+ ion transportation from the anolyte to the catholyte [47].

To determine the concentration of metallic impurities within the electrolyte, samples were analysed by ICP-MS (Agilent 7500cx). ICP-MS was also conducted on electrolyte samples after electrolysis to quantify the gold dissolution due to the cleaning procedures.

The impurities deposited on the surface of the gold cathodes during CO2 reduction were examined by XPS. In all cases, after 4 h of electrolysis, the electrodes were removed from the electrochemical cell, rinsed thoroughly with DI water and dried under room temperature. The XPS spectra were obtained on a Kratos Axis Ultra DLD instrument using monochromated Al Kα X-rays (1486.69 eV). The anode was set at 15 kV with a 10-mA emission current and a base pressure of 5 × 10−9 Torr. The survey scans were collected over the range 1300 to − 5 eV in 1-eV steps with a 13-min acquisition time.

Results and Discussion

Continuous Galvanostatic CO2 Reduction on Gold

During the galvanostatic reduction of CO2 on gold cathodes, the main gaseous products were found to be CO and H2 (Fig. 1) which is consistent with previous reports [6, 20, 21, 27, 48]. Small amounts of formic acid were also produced with a current efficiency of < 10%. As expected, the current efficiency for CO production was found to decrease over 4 h of reduction (Fig. 1a), with this drop in current efficiency for CO production accompanied by the cathode potential becoming less negative (Fig. 1b). The deactivation of electrodes during CO2 reduction has been reported elsewhere for gold cathodes [15] and other metals [11, 14, 49] and is normally attributed to the deposition of metals from electrolyte impurities [22] or surface-bound intermediates [13, 34]. These deposited poisons either block the active sites for CO production or produce sites that favour the hydrogen evolution reaction (HER). As the cathode potential becomes more positive, the selectivity towards CO decreases and the HER increases; this indicates that the activity of the cathode for the HER increases during the deactivation process. This would be expected if elements like Fe or Cu are deposited on the gold surface, as these are more selective for the HER compared with gold [17, 50]. While the bulk pH of the electrolyte does change during CO2 reduction due to the selective transport of K+ ions across the membrane, it seems unlikely that this contributes to the observed changes in CO vs H2 selectivity as the bulk pH only increases from around 6.8 to 7.2 during a 4-h CO2 reduction experiment.

Fig. 1
figure 1

a Current efficiency towards gas products and b potential during CO2 reduction at a gold cathode in 0.2 M KHCO3 saturated with CO2 at − 5 mA cm−2

Full size image

To investigate possible changes in the surface of the gold during the deactivation process, cyclic voltammetry was conducted before and immediately after 4 h of galvanostatic CO2 reduction, with significant differences in the voltammograms being observed (Fig. 2).

Fig. 2
figure 2

Cyclic voltammetry at the scan rate of 100 mV s−1 of a gold electrode in 0.2 M KHCO3 saturated with CO2a before and b after 4 h of CO2 reduction at − 5 mA cm−2

Full size image

The voltammogram of the gold cathode before CO2 reduction exhibits the normal gold–gold oxide redox features, with peaks C and D corresponding the formation and reduction of gold oxide, respectively (Fig. 2a). These peaks are still observed after 4 h of CO2 reduction, but are accompanied by three new peaks (Fig. 2b). The large anodic peak at the potential around − 0.13 V vs Ag|AgCl (peak A) is attributed to the oxidation of CO [51,52,53,54], formed from the CO2 reduction reaction. The size of this peak decreases significantly if the potential is held at − 0.75 V for 1 min prior to cyclic voltammetry, suggesting that this CO is either weakly adsorbed on the cathode or dissolved CO within the diffusion layer (and thus diffuses into the bulk during this pause at − 0.75 V). The peak size also decreases on the subsequent voltammetric cycles due to the CO stripping reaction and diffusion away from the surface. It was also observed that the potential of this CO oxidation peak shifts to slightly higher potentials between the 1st and 2nd cyclic voltammetry cycles which we attribute to the high interfacial pH (which forms during CO2 reduction [47, 55]) decreasing back to the bulk pH value during the cyclic voltammetry. Peaks B and E (0.27 V vs Ag|AgCl) only appear after prolonged reduction, suggesting these are related to the species responsible for deactivation. Others have shown that both Cu and Zn deposit on gold cathodes, with a Zn stripping peak found at − 0.5 vs Ag|AgCl and Cu stripping peaks found at − 0.05 and 0.24 V vs Ag|AgCl [22].

From the Cu Pourbaix diagram [56] at pH = 7, it is clear that multiple solid-state reactions can occur if Cu was deposited during CO2 reduction. These include:

$$ 2\mathrm{Cu}+{\mathrm{H}}_2\mathrm{O}\rightleftharpoons \mathrm{C}{\mathrm{u}}_2\mathrm{O}+2{\mathrm{H}}^{+}+2{\mathrm{e}}^{-}{E}^0=-0.143\ \mathrm{V}\ \mathrm{vs}\ \mathrm{AgAgCl} $$
(1)
$$ \mathrm{Cu}+{\mathrm{H}}_2\mathrm{O}\rightleftharpoons \mathrm{Cu}\mathrm{O}+2{\mathrm{H}}^{+}+2{\mathrm{e}}^{-}{E}^0=-0.044\ \mathrm{V}\ \mathrm{vs}\ \mathrm{AgAgCl} $$
(2)
$$ \mathrm{Cu}+2{\mathrm{H}}_2\mathrm{O}\rightleftharpoons \mathrm{Cu}{\left(\mathrm{OH}\right)}_2+2{\mathrm{H}}^{+}+2{\mathrm{e}}^{-}{E}^0=-0.005\ \mathrm{V}\ \mathrm{vs}\ \mathrm{AgAgCl} $$
(3)
$$ {\mathrm{Cu}}_2\mathrm{O}+{\mathrm{H}}_2\mathrm{O}\rightleftharpoons 2\mathrm{CuO}+2{\mathrm{H}}^{+}+2{\mathrm{e}}^{-}{E}^0=+0.055\ \mathrm{V}\ \mathrm{vs}\ \mathrm{AgAgCl} $$
(4)
$$ {\mathrm{Cu}}_2\mathrm{O}+2{\mathrm{H}}_2\mathrm{O}\rightleftharpoons 2\mathrm{Cu}{\left(\mathrm{OH}\right)}_2+2{\mathrm{H}}^{+}+2{\mathrm{e}}^{-}{E}^0=+0.133\ \mathrm{V}\ \mathrm{vs}\ \mathrm{AgAgCl} $$
(5)

It is also possible that Cu carbonates can form on any Cu deposits in KHCO3 electrolytes [57]. Any solid Cu species can also dissolve to Cu+ and Cu2+ following:

$$ 2{\mathrm{Cu}}^{+}+{\mathrm{H}}_2\mathrm{O}\rightleftharpoons \mathrm{C}{\mathrm{u}}_2\mathrm{O}+2{\mathrm{H}}^{+}\log \left(\mathrm{C}{\mathrm{u}}^{+}\right)=-0.84-\mathrm{pH} $$
(6)
$$ {\mathrm{Cu}}^{2+}+{\mathrm{H}}_2\mathrm{O}\rightleftharpoons \mathrm{CuO}+2{\mathrm{H}}^{+}\log \left({\mathrm{Cu}}^{2+}\right)=7.89-2\mathrm{pH} $$
(7)
$$ \mathrm{Cu}\rightleftharpoons {\mathrm{Cu}}^{+}+{\mathrm{e}}^{-}{\mathrm{E}}^0=0.320+0.0591\ \log \left(\mathrm{C}{\mathrm{u}}^{+}\right)\mathrm{V}\ \mathrm{vs}\ \mathrm{AgAgCl} $$
(8)
$$ \mathrm{Cu}\rightleftharpoons {\mathrm{Cu}}^{2+}+2{\mathrm{e}}^{-}{\mathrm{E}}^0=0.137+0.0295\ \log \left(\mathrm{C}{\mathrm{u}}^{2+}\right)\mathrm{V}\ \mathrm{vs}\ \mathrm{AgAgCl} $$
(9)

As peaks B and E appear only after prolonged electrolysis and appear at potentials similar to those given in Eq. 5, 8 and 9, together with a previous report [22], it is reasonable to suggest that peaks B and E are due to the presence of Cu species deposited on the electrode from impurities present in the electrolyte. This was confirmed by ICP-MS analysis, which revealed that the 0.2 M KHCO3 electrolyte contained approximately 0.06 μM Cu, 0.06 μM Zn and 0.2 μM Fe. To examine if these elements altered the selectivity of the CO2 reduction reaction on gold, three additional experiments were performed in which 1 μM of Cu2+, Fe2+ and Zn2+ was individually added to the electrolyte prior to CO2 reduction. By performing cyclic voltammetry after a period of prolonged CO2 reduction, it was found that the addition of Cu2+ caused peaks B and E to increase in size (Fig. 3d), confirming that Cu impurities are the source of these peaks. No additional peaks could be found by cyclic voltammetry when Fe2+ and Zn2+ were added to the electrolyte (the standard potential of the Fe2+/Fe and Zn2+/Zn reactions are − 0.64 and − 0.96 V vs Ag|AgCl [56]). However, by following the current efficiency of the CO2 reduction reaction in the presence of either 1 μM of either Cu2+, Fe2+ or Zn2+, it was found that the presence of Fe2+ and Zn2+ had the most dramatic impact on the deactivation of the cathode towards CO production, a corresponding rapid increase in the current efficiency towards H2 (Fig. 3a). For the experiments containing additional Fe2+, this seems reasonable given the much higher selectivity and activity that Fe has towards H2 compared with gold [17, 21]. Interestingly, the addition of Cu2+ did not alter the selectivity towards HER, but did cause CO to decrease more rapidly compared with the control experiment (Fig. 3b). Liquid samples taken over the course of these experiments revealed that the drop in current efficiency towards CO in the presence of Cu2+ can be explained by the increase in formate production rate (Fig. 3c). The increasing formate production rate, and thus activity of the electrode for formate production over the electrolysis time, suggests that the surface is continuously accumulating Cu. In the case of Fe2+ additions, no changes in the formate production rate could be observed, which again is consistent with the known behaviour of Fe cathodes [21] while the additional 1 μM of Zn2+ caused both H2 and formate production to increase. As this analysis suggests that Fe, Zn and Cu impurities present in the electrolyte may play a role in the deactivation of gold cathodes, XPS was used to the examine whether these elements are deposited onto the surface of the gold during CO2 reduction. This analysis (Fig. 4) revealed that small amounts of Zn and Cu were found on the surface of the gold in agreement with others [22]. Surprisingly, despite the clear evidence that Fe2+ in the electrolyte causes rapid deactivation of the gold cathode towards CO2 reduction, no evidence for Fe on the surface of the gold cathodes was found by XPS. One possible explanation for this is that the small quantity of Fe deposited on the gold during CO2 reduction quickly re-dissolves in the short period at open-circuit potential prior to the electrode being removed from the electrolyte.

Fig. 3
figure 3

The effect of intentionally added electrolyte impurities on a CO current efficiency, b H2 current efficiency and c formate production rate and d cyclic voltammetry after 4 h CO2 reduction at − 10 mA cm−2 in CO2-saturated 0.2 M KHCO3 with and without the addition of 1 μM Cu2+. Each impurity was added in individual experiments to the 0.2 M KHCO3 electrolyte

Full size image
Fig. 4
figure 4

XPS 2p spectra of a Cu and b Zn on the gold electrode after 4 h CO2 electroreduction at the current density of − 5 mA cm−2 in CO2-saturated 0.2 M KHCO3. The vertical lines indicate the binding energies associated with the metallic species [22]

Full size image

While this analysis cannot identify which sites are more easily poisoned by the deposition of metallic impurities, others have shown that step or defect sites are favoured locations for the nucleation and growth of Cu deposits on gold [58, 59]. As these same types of sites are often proposed as the most active sites for CO2 reduction [60,61,62,63], it is likely that the step or defect sites are those which are first poisoned by the metallic deposits.

Preventing Deactivation by Using Periodic Anodic Treatments

To avoid the observed deactivation of gold electrodes during CO2 reduction, the common approach is to periodically apply relatively positive potentials to the gold electrode in order to anodically strip the poisons responsible for deactivation. Shiratsuchi et al. [27] applied a repeating square-wave potential protocol of − 1.5 V vs Ag|AgCl for 5 s (CO2 reduction step) followed by 1 V vs Ag|AgCl for 5 s (cleaning step). Similarly, Kedzierzawski and Augustynski [15] used a periodic activation process wherein the potential of the gold electrode was swept from the CO2 reduction potential (− 1.65 to − 0.95 V vs Ag|AgCl) to 1.1 V vs Ag|AgCl at 50 mV s−1 every 15 min in order to anodically strip any poisons off which may accumulate on gold during CO2 reduction. Both of these approaches successfully prevent deactivation, with the current efficiency towards CO production found to be almost stable at around 60% for 4 h (Fig. 5). In the case of the square-wave potential protocol proposed by Shiratsuchi et al. [27], in agreement with their work, it is found that while this stabilises the electrode for CO production, the net rate of H2 production is found to be very low and the total charge balance no longer holds. This is likely due to the oxidation of a portion of the evolved H2 when the potential is pulsed to 1 V vs Ag|AgCl. As the kinetics of hydrogen oxidation are faster than CO oxidation [64], it is not surprising that this oxidation appears to only effect the net H2 evolution and not CO production. This oxidation of hydrogen does not seem to occur when following the method proposed by Kedzierzawski and Augustynski [15], as the oxidising or stripping potentials are obtained by sweeping the potential from the CO2 reduction conditions (− 1 V) up to 1.1 V at 50 mV s−1 during which time the majority of the evolved hydrogen can diffuse away from the electrode. While these methods clearly prevent or slow the deactivation of the gold electrode, both approaches will result in the formation of gold oxide and thus lead to gold corrosion and/or severe surface restructuring [15, 27, 37, 39,40,41,42]. Given that surface restructuring is expected to alter the selectivity [65], this corrosion and restructuring will make long-term CO2 reduction on gold problematic, especially if gold nanoparticles are used [43, 44]. In fact, Shiratsuchi et al. pointed out that their periodic cleaning procedure resulted in gold corrosion and an increase the roughness factor of the gold cathode [27]. To confirm that gold corrosion occurs during these cleaning procedures, electrolyte samples taken after 4 h of CO2 electrolysis were analysed by ICP-MS, with 450 ppb gold found in the electrolyte when the method of Shiratsuchi et al. was used, and 50 ppb gold found after 4 h of electrolysis following the procedure of Kedzierzawski and Augustynski. The lower level of gold in the electrolyte following the Kedzierzawski and Augustynski method is due to the shorter time this method spends at potentials capable of anodic gold dissolution.

Fig. 5
figure 5

Current efficiency of gas products from CO2 reduction at a gold cathode in 0.2 M KHCO3 saturated with CO2 using periodic activation procedures. a Repeating sequence of − 5 mA cm−2 for 5 s and 1 V vs Ag|AgCl for 5 s (Shiratsuchi et al. [27]), b CO2 reduction at – 5 mA cm−2, with sweeps from − 1 to 1.1 V at 50 mV s−1 every 15 min (Kedzierzawski and Augustynski [15])

Full size image

Preventing Deactivation Without Gold Dissolution

From the previous section, it is clear that gold electrodes will undergo significant dissolution and very likely severe surface restructuring due to the anodic treatments applied to remove the poisons responsible for the deactivation during CO2 reduction. As this dissolution and restructuring is due to the formation and reduction of gold oxide [39,40,41,42], here the possibility of removing the poisons without the formation of gold oxide is investigated by performing cyclic voltammetry every 15 min during galvanostatic CO2 reduction. In this case, the potential was swept at 100 mV s−1 between − 0.5 V vs Ag|AgCl and various upper potential limits. As the onset of gold oxide formation is 0.65 V vs Ag|AgCl in 0.2 M KHCO3 saturated with CO2, the upper potential limit was selected to be between 0 and 0.5 V vs Ag|AgCl to ensure that no gold oxide formation or gold dissolution would occur. Thus, upper potential limits of 0, 0.15, 0.35 and 0.5 V vs Ag|AgCl were selected to be more positive than the CO stripping peak, just below the stripping of Cu (peak B = 0.28 V), just after the Cu stripping potential and well after the Cu stripping peak but below the gold oxide formation potential. While there is still some deactivation of the CO2 reduction towards CO when the upper potential limit of the stripping voltammetry is only 0 V, sweeping the potential to at least 0.15 V appears to stabilise the current efficiency for CO production (Fig. 6a). This stabilisation is highlighted by the measured potential during the galvanostatic CO2 reduction periods, where it is clear that upper potential limits of 0.35 and 0.5 V vs Ag|AgCl ensure that the potential remains essentially constant over 4 h of CO2 reduction. For an upper potential limit 0.15 V vs Ag|AgCl, it appears that the electrode is still being deactivated as the potential during the CO2 reduction continuously increases, as does the potential when the upper potential limit is 0 V. The increasing electrode potential coupled with a drop in the current efficiency for CO production indicates that the cathode is becoming more active towards the HER, which is consistent with the accumulation of Cu on the cathode. This is not surprising given that cyclic voltammetry reveals that Cu is not stripped from the gold until the potential is at least 0.28 V vs Ag|AgCl. This stripping can be observed on the cyclic voltammograms used to clean the electrode, with the charge associated with the Cu stripping process (peak B) found to decrease over the five cycles suggesting that less Cu remains on the gold electrode after this cleaning process (Fig. 7). It was also found that when the upper potential limit of the stripping voltammetry was 0.5 V vs Ag|AgCl, the formate production rate was constant at around 3 nmol s−1 (compare with Fig. 3c) which confirms that this approach successfully prevents the continuous accumulation of Cu on the surface during CO2 reduction. While the deactivation is essentially prevented when the upper potential limit of the stripping voltammetry is at least 0.35 V vs Ag|AgCl, the current efficiency for CO production is more stable (compared with Fig. 1) even when the upper potential limit is below the Cu stripping potential. This suggests all of the stripping voltammograms used in this work have removed other poisons in addition to Cu. Given that the electrolyte also contains dissolved Fe and Zn and that these both lead to deactivation (Fig. 6b), it seems clear that Fe and Zn are also dissolved off the gold surface during the stripping cyclic voltammetry. This is consistent with the expected potential required to strip Fe and Zn off the gold (− 0.64 and − 0.96 V vs AgAgCl). This was confirmed by XPS which revealed that no Fe, Zn and Cu impurities were found on the gold surface when the upper potential limit of voltammetry is 0.35 V vs Ag|AgCl. Finally, unlike the two previously report cleaning processes [15, 27], ICP-MS analysis showed that no significant gold corrosion occurs during the cleaning process that has been applied in this work, suggesting that this approach can be used to prevent the deactivation during long-term CO2 reduction processes.

Fig. 6
figure 6

a Current efficiency for CO production and b potential vs time during CO2 reduction at a gold cathode in 0.2 M KHCO3 saturated with CO2 at – 5 mA cm−2 with cyclic voltammetry every 15 min between − 0.5 V and an upper potent limit of (right-pointing pointer) 0 V, (plus sign) 0.15 V, (asterisk) 0.35 V and (white circle) 0.5 V vs Ag|AgCl

Full size image
Fig. 7
figure 7

Stripping cyclic voltammetry on a gold cathode during CO2 reduction in 0.2 M KHCO3 saturated with CO2 at – 5 mA cm−2. The stripping voltammetry was performed every 15 min at 100 mV s−1. The anodic current between 0.5 and − 0.25 V during the scan in the negative direction is most likely due to oxidation of residual CO not completely oxidised during the scan in the positive direction

Full size image

Using Periodic Open Circuit or Anodic Treatments to Prevent Deactivation

While using cyclic voltammetry over an appropriate potential range is an effective method to prevent the deactivation of gold cathodes during CO2 reduction, using a method requiring a reference electrode is unlikely to be applied in an industrial electrolysis process. Given that the gold cathode is poisoned by Zn, Fe and Cu, it is proposed that any simple two-electrode method which results in the potential of the gold electrode being more positive than − 0.96, − 0.64 or 0.28 V vs Ag|AgCl will be enough to remove the Zn, Fe and Cu poisons respectively. Therefore, interrupting galvanostatic CO2 reduction with periods of open-circuit or low anodic currents has been investigated as possible simple cleaning methods.

By interrupting CO2 reduction every 15 min with 4 min at open circuit, the deactivation of the gold cathode is found to be less severe than without the interruptions (Fig. 8a). While this approach could be very easily applied on an industrial CO2 electrolyser, it does have the disadvantage that both the cathode and anode potentials would be uncontrolled and thus could lead to unintended corrosion. To test this, during each open-circuit interruption, the cathode potential was continuously measured and is found to almost reach the same potential after 4 min (Fig. 8b). This potential is well below the onset of gold oxidation and thus it is unlikely that the cathode would undergo corrosion during these open-circuit periods. Here, the small improvement in the stability of the current efficiency for CO production is most likely due to the removal of Fe and partial removal of Zn (both of which should occur at potentials above − 0.64 V). Clearly, as the maximum potential during the open-circuit interruptions below that required to remove the Cu, the remaining deactivation is likely due to the accumulation of Cu, with cyclic voltammetry and XPS analysis confirming that Cu indeed remains on the surface after these open-circuit periods. In addition to Cu, XPS also showed the presence of Zn on the gold despite the fact that the achieved potential during open-circuit potential periods is more positive than the Zn/Zn2+ potential. As Zn passivates at pH = 7 and − 0.36 V vs Ag|AgCl [56], it is likely that the Zn remaining on the gold surface is ZnO rather than metallic Zn. While this passive ZnO forms during OCP interruptions, we propose that the formation of this passive ZnO is avoided during periodic stripping voltammetry cleaning process. This is because the pH-potential window where the passive ZnO can form is very small [56] and the surface pH will be higher than 7 immediately after a period of CO2 reduction [47, 55].

Fig. 8
figure 8

The effect of interrupting CO2 reduction at – 5 mA cm−2 with 4 min of open-circuit potential every 15 min. a Current efficiency for CO production, and b the open-circuit potential during the CO2 reduction interruption at various times over the 4-h period of CO2 reduction

Full size image

As the open-circuit periods were not found to remove all of the Zn and Cu poisons, the CO2 reduction was instead interrupted every 15 min with short periods of anodic current. The currents and treatment times were based on the charge associated with the CO oxidation peak and Cu stripping peak found by cyclic voltammetry (Fig. 7). By interrupting the galvanostatic CO2 reduction every 15 min with a 25 μA cm−2 anodic current for between 10 and 50 s (250–1250 μC cm−2 of anodic charge), it was found potential of the gold electrode could reach between − 0.15 and 0.55 V (Fig. 9a) which is enough to partially strip Fe and Zn and at higher potentials Cu. As with the open-circuit interruptions, this resulted in an improved stability, but did not completely prevent deactivation with the current efficiency for CO production dropping from an average maximum of 53% to an average of 46% (Fig. 9b). XPS revealed that no Fe or Cu was on the gold surface following this cleaning protocol (when the achieved potential is around 0.3 V vs Ag|AgCl), and again it is suggested that the Zn remaining on the surface is in the form of ZnO which is resistant to corrosion at the pH and potentials found during this cleaning process.

Fig. 9
figure 9

The effect of anodic treatments applying various charges on a potential during treatment procedure and b current efficiency for CO production. The anodic treatments were applied every 15 min during CO2 reduction at – 5 mA cm−2 in 0.2 M KHCO3 saturated with CO2

Full size image

Finally, an interestingly observation made during this work was that electrodes were generally found to be more stable for CO production when the electrodes were less active for CO production. While the initial differences in the current efficiency for CO production between experiments are likely to be due to slight variations between the electrode themselves (e.g. surface roughness factor or crystal facets), no clear trend between the initial electrode structure (based on cyclic voltammetry recorded before the CO2 reduction) and the initial current efficiency could be found. Despite this, the observation that less active electrodes were less susceptible to poison may suggest that only certain sites on the gold electrodes are readily poisoned.

Conclusions

Electrochemical CO2 reduction on gold can produce CO, H2 and formate, although the current efficiency towards CO decreases over time due to the deactivation of gold cathode. This deactivation results in higher H2 production and a continuously increasing formate production rate. XPS analysis shows that this deactivation is caused by the deposition of Cu and Zn from trace amounts of these elements in the electrolyte. Surprisingly, no Fe could be found on gold cathode by XPS after prolonged CO2 reduction, despite the fact that the Fe concentration in the electrolyte is higher than that of Zn and Cu and that intentional addition of Fe2+ to the electrolyte leads to severe deactivation. Thus, it is concluded that Zn, Cu and Fe all contribute to the deactivation of gold cathodes for CO2 reduction. This deactivation can be prevented by periodically increasing the potential of the gold electrode to anodically strip these poisons from the surface. Unfortunately, this treatment can lead to significant gold dissolution if the potential exceeds approximately 0.65 V vs Ag|AgCl at pH 7, which would make long-term electrolysis very difficult. However, by periodically interrupting CO2 reduction with cyclic voltammetry every 15 min between − 0.5 and 0.5 V vs Ag|AgCl, the metallic impurities can be stripped and the deactivation prevented without leading to gold dissolution. Partial prevention of the CO2 reduction deactivation can also be achieved with periodic open-circuit periods or short anodic current treatments, although further work needs to be conducted to optimise these simple cleaning protocols.