In Operando investigations of oscillatory water and carbonate effects in MEA-based CO 2 electrolysis devices

In Operando investigations of oscillatory water and carbonate effects in MEA-based CO2 electrolysis devices Asger B. Moss , Sahil Garg a, , Marta Mirolo, Carlos A. Giron Rodriguez , Roosa Ilvonen, Ib Chorkendorff, Jakub Drnec, Brian Seger * Surface Physics and Catalysis (Surf Cat) Section, Department of Physics, Technical University of Denmark, 2800 Kgs. Lyngby, Denmark Experimental Division, European Synchrotron Radiation Facility, Grenoble, France Joint first authors: Asger B. Moss; Sahil Garg *Corresponding author: Brian Seger, Email: brse@fysik.dtu.dk


Introduction
Storing renewable energy into chemical bonds/fuels via CO2 electrolysis (CO2E) is a promising approach to creating a carbon-neutral cycle while also curbing net CO2 emissions. At present, copper-based membrane electrode assembly (MEA) electrolyzers allow a high rate of CO2E to ≥ C2+ products. 1 However, the lack of stability, mainly seen as CO2 reduction reaction (CO2RR) being suppressed by the competitive hydrogen evolution reaction (HER), is a major factor delaying the deployment of CO2E technology on a commercial scale. [2][3][4][5][6] Therefore, for the future development of these electrolyzers, it is crucial to investigate the exact cause of HER increase over time.
In recent years, many attempts have been made to determine what can cause the observed gradual selectivity change from CO2RR to HER in a CO2 electrolyzer. 2,3,[6][7][8] Overall, it is believed that the issues arise from improper water management and the local reaction environment in the vicinity of the cathode. 9 For example, sufficient water is necessary to drive the CO2E, but excess water can flood the gas diffusion electrode (GDE), which restricts CO2 mass transfer, and shifts selectivity to HER. 5 The reductive potentials needed for CO2RR lead to electrowetting of the GDE which provides a flooding mechanism. 2,7,10,11 Moreover, the highly alkaline environment created from the cathodic reactions will also equilibrate with CO2 to form bicarbonate/carbonate (HCO3⁻/CO3 2 ⁻) ions that easily permeate through anion exchange membranes (AEMs). 12,13 Besides lowering the CO2 utilization, the CO3 2 ⁻ formation also exacerbates any CO2 mass transport limitation at the cathode. 12,14 In addition, the hydroxyl groups produced at the cathode (from CO2RR) can attack hydrophobic supports of the AEM, entailing potential durability issues. All these factors complicate the investigation of catalytic properties as it entangles CO2 mass transport limitations, ionic conductivity, and water management.
Most of the cathode flooding mechanisms and concepts are based on fuel cell works [15][16][17] with only a few CO2E studies showing indirect evidence of flooding in a flow cell (i.e. flowing catholyte between GDE and membrane) and zero-gap MEA-based CO2 electrolyzers. 2,3,18,19 For instance, Leonard et al. 2 demonstrated both HER and electrochemical double-layer capacitance of the cathode GDE increased with the amount of charge passed through the GDE, where a higher electrochemical double-layer capacitance was attributed to a higher rate of cathode flooding.
Although there is a consensus that cathode flooding plays a role in affecting the CO2RR/H2 selectivity, an in operando visual inspection of a cathode GDE only allows the back surface to be analyzed. To the best of our knowledge, there has yet to be a direct in operando approach to show the flooding effect and its relation to product performance. Obtaining in operando results is especially important in CO2E since the local reaction environment has been shown to vary with time and electrolysis conditions. 12 Furthermore, how a catalyst behaves and whether it changes due to these variations has also yet to be analyzed in-operando.
Another issue affecting CO2RR/H2 selectivity relates to salt depositions in the cathode GDE, which block gas flow to the catalyst. 20 There is a current debate as to if the salt deposits as alkali metal cation (K + /Cs + ) carbonate [21][22][23] or bicarbonate, 20 however, this has only been investigated by computational models 23 or post mortem analysis. 24 Furthermore the three-way relationship between flooding, salt deposition, and variations in selectivity has rarely been investigated in depth.
In this work, we couple high flux 4 th generation synchrotron X-ray source with superior phase contrast of diffraction phenomena to visualize the interior of an operational GDE to not only understand the relationship between selectivity, flooding, and salt deposition, but also to determine how this affects potential (at a constant current) and ion selectivity crossing through the membrane to the anode. Furthermore, this work analyzes structural changes within the Cu catalyst layer, such as how quickly it transitions from a surface oxide to metallic and whether it ripens, gets strained, or migrates within the GDE. It should be noted that the Bazylak group 18,19 has previously attempted to study both MEA and catholyte-based CO2 electrolysis devices by X-ray radiography with a focus on gas bubble formations in the MEA and electrolyte layer respectively, however, they were unable to follow the evolution of different phases in the catalyst and to concomitantly analyze products, thus limiting the scope of their work.
One additional parameter that this work investigates is the oscillations that are often seen at CO2 electrolysis devices devolving from CO2RR to HER selectivity at the cathode.
Oscillations typically occur at conditions on the borderline of stability, and thus while not typically consistent, they do provide transient conditions that allow for a much easier understanding of the processes leading to a loss in CO2RR selectivity.

Results and discussion
Reactor performance Figure 1 shows the reactor design and the measurement design: the synchrotron X-ray beam (5 μmvertical x 20 μmhorizontal) probed from the edge of the membrane to deep within the cathode GDE (~ 150 μm in total) by a continuous series of vertical line scans allowing for a comprehensive understanding of electrolyte, salt, and copper as a function of depth within the GDE. To accelerate durability issues related to a change in selectivity from CO2RR to HER and limited time availability at the synchrotron, we did not use a Teflon gas diffusion layer 25 and low anolyte electrolyte concentrations 26 (e.g. 0.01 M), but rather a moderately hydrophobic carbon-based gas diffusion layer and a high concentration anolyte (0.1 M). Similarly, we used dry CO2 instead of humidified CO2. 6,27,28 Furthermore, we performed a series of operando experiments at different current densities (100, 150, 200, and 250 mA·cm -2 ) to show that these trends were comprehensive and not a function of a given current density regime. The complexities of CO2 electrolysis reactor designs entail a benchmark test is necessary to compare CO2RR performance, especially given the X-ray cell in Figure 1 that was used for synchrotron work. Thus, without X-ray irradiation, experiments were done to show the performance of the cell. Without the need for beam alignment, this allowed us to test the cell immediately and produced clear and consistent results. Figure 1C shows concentrations of gaseous products from one of these preliminary tests with sputtered Cu using a conventional reactor design (used in our earlier studies) 13,29 at 200 mA·cm -2 over a ~ 6 hr test period. With liquid product analysis only being taken at the end of the experiment, the total faradaic efficiency is a convolution of the entire experiment. Nevertheless, it gives some insight into the electrochemical performance of the reactor, and Figure S1 shows the average faradaic efficiency for all products. It should be noted that while other works have proposed X-ray irradiation may influence CO2RR performance, 19 we do not see any significant influence on performance related to X-ray irradiation ( Figure S2). 30 Figure 1C notably demonstrates a slow HER increase and C2H4 and CO decrease during the first 3 hr of the experiment. While methane increases slightly, its low overall faradaic efficiency entails it is difficult/unwise to analyze this further. Given that the only hydrophobic barrier used was the GDE, this decrease in overall CO2RR was expected. 25 However beyond 3 hr, we found an oscillating trend in FE of gaseous CO2RR products and HER, correlating well with the cell potential. These oscillations at the increased time were observed in approximately half of all samples tested. With oscillations often related to stability limits, this typically entails an inconsistent behavior: oscillations would occur after some time, but the frequency and intensity (in terms of voltage and productivity changes) would often vary.
To investigate both the increasing HER selectivity and the oscillating trends, we performed a series of in operando diffraction measurements using high-energy X-rays. By setting the current density and simultaneously measuring voltage, cathode, and anode selectivity, the state of Cu (oxidation level, micro-structure, particle size, and location), water content, and presence of crystalline KHCO3 and K2CO3 via a variety of techniques, we gain unprecedented insights into the internal functioning of CO2 electrolyzers. While we do expect flooding to be the main cause of degradation, we wanted to ensure that the loss in CO2RR selectivity did not arise from a change in the catalyst itself. To confirm this, we provide a detailed analysis in Supplementary Note 1 (including Figures S3 to S9) on how the Cu catalyst did not change (only native Cu2O reduced to metallic Cu) and there was no IrO2 crossover from the anode side to the cathode during CO2 electrolysis.

Flooding of cathode GDE
The electrolyte distribution in the different parts of the cell, mainly GDE, is followed by deconvoluting the liquid phase diffraction pattern from the diffractograms. [31][32][33] We take advantage of the fact that the other solid amorphous phases do not change during the experiment and the change in the q-range 2.45 to 2.46 Å -1 of diffractogram is mainly due to the electrolyte signal. Investigations of individual patterns confirm that the changes observed in the full q-range background match well with water. As an example, we show some diffraction patterns ( Figure S10) and change in the background ( Figure S11) at a single region in the GDE during different electrolysis time.
As seen in Figure 2, the overall electrolyte content in GDL increases slowly until ~100 min where it suddenly decreases. Given that the selectivity change seems to be correlating with the electrolyte content in the GDL, this slow increase of electrolyte content in the GDL is most likely the explanation for why we see a slow degradation in CO2RR performance over time.
For example, with GDL pores slowly filled with electrolyte, the CO2 diffusion distance towards the catalyst layer increases, reducing CO2 availability and causing a gradual shift from CO2RR to HER. Interestingly, the electrolyte content is observed further toward the flow field than we expected. Recent studies reported that loss of GDE hydrophobicity during electrolysis (in addition to the formation of liquid products affecting the physical properties of the GDE) can cause the electrolyte to penetrate deeper into the GDE. 2,3,7 The electrolyte content seems to be highest in the outer layers of the GDL (> 50 µm), whereas there seems to be a relatively unchanged region closer to the catalyst layer ( Figure 2). A similar trend is observed for all GDEs, and we expect this to be caused by the differences in PTFE content (25% vs. 5%) and pore size (<1 µm vs. 25-200 µm) of the micro-and macro-porous layers of the Sigracet 39 BB GDL resulting in different water management properties. 34,35 The degree of penetration of the microporous layer into the macroporous layer as well as GDL compression are unknown, thus an exact determination of the microporous/macroporous transition cannot be determined.
Nevertheless, this order of magnitude thickness does align with a relatively dry area between the catalyst and outer GDL layers. Initially, the slow increase in electrolyte content and the cell potential follow until around 90 min and both seem to correlate with a steady increase in HER indicating that slow flooding is causing this performance degradation. From around 90 min to 100 min the HER seems to increase faster, but at this point, the potential starts to decrease while the electrolyte content continues to increase. The initial correlation seems to break at this point indicating that a different mechanism is accelerating the performance degradation. Once the potential reaches a local minimum (~100 min), the potential starts to increase again. Shortly after (~105 min), the water content drastically starts to decrease. The oscillation repeats in roughly 20 min cycles with a 3-5 min delay between the minimal potential and maximum electrolyte content.
Such oscillations, both in terms of cell potential and electrolyte content in GDL could be caused by changes in the anode potential, e.g., from anode degradation or bubble formation, or could be due to changes in gas pressure and temperature in the electrolyzer. There are, however, no indications that any of this is the cause. We measured the anode potential by putting a reference electrode in the inlet flow channel of the anolyte. The anode potential remained stable throughout the CO2 electrolysis time, indicating that any change happening in the cell potential is either from changes in the cathode potential or membrane ohmic losses ( Figure S14). Small periodic variations in the cathode inlet pressure ( Figure S15) were observed, but with different periodicity (caused by a slight overpressure in the GC). Also, no variation in cathode flow-field temperature was observed during the electrolysis ( Figure S15). It should be noted that the cells used for experiments shown in Figure 1C and Figure 2 were both tested with the same parameters, but their oscillations and CO2RR selectivity trends varied significantly, pointing to the chaotic nature of the processes driving these fluctuations. Figure 2C shows the faradaic efficiency of the cathodic gas products ( Figure S13 shows the exact CO2RR product breakdown from Figure 2). The FE of H2 gradually increases during the first 80 min before further rapid increase at 100 min. For the next 15 min, the rate of H2 selectivity increase slows down slightly before the start of the second cycle of acceleration, and then the deceleration of the H2 selectivity increase. The slow sampling rate of the gas chromatograph dilutes the trends, but there is a notable correlation between the electrolyte content in the GDE and the level at which HER increases.
The aforementioned shift from a high device potential during primarily CO2RR to a lower potential during the spike of H2 selectivity provides an intriguing result as this is contrary to what is seen in the first 80 minutes, where as HER increases, there is a gradual increase in operating potential. At the local pH's produced during high current density CO2RR, 36 it is well established that CO2RR has a lower onset potential than HER (otherwise CO2RR would not occur [37][38][39]. While this argument is in agreement with the first 80 minutes of the experiment, during the fluctuations there must be an additional factor in play causing a higher HER and lower voltage. For example, an ohmic resistance decrease in the membrane (or other places within the device) or fundamental variation in catalytic activity could potentially be the cause of lower cell potential and increased HER. The conductivity of the ion exchange membrane is well known to be a function of the ion being transferred. For instance, the conductivity of a Sustainion membrane X-24 membrane in KHCO3 is 24 mS·cm -1 whereas the same membrane in KOH shows a conductivity of 66 mS·cm -1 . 40 While this work uses a different version of Sustainion, we expect the trends to still hold. It is well established that analyzing the anodic CO2 to O2 ratio in CO2 electrolysis devices works as a proxy to determine what species (i.e. HCO3⁻, CO3 2 ⁻or OH ⁻ ) is transferring through the membrane. 12,13,41 A CO2 to O2 ratio of 4 means HCO3⁻ transfer, a ratio of 2 means CO3 2 ⁻, and a ratio of 0 means OH ⁻ transfer.
Since carbonates crossing over to the anode reacts quickly with protons and release CO2, we cannot calculate the migration of carbonates by measuring the anolyte concentration. Instead, to understand if the membrane could be contributing to potential oscillations, we looked into the anodic CO2 to O2 ratio. To show these oscillations are current independent, the experiment was done at 150 mA·cm -2 , and the results are shown in Figure 3 ( Figure S16 shows a detailed version of Figure 3B). Again, we see the link between electrolyte content, cell potential, and faradaic efficiency of H2, albeit much clearer. Figure S17 shows detailed cathodic gas product distribution from the GC whereas Figure S18 shows cathodic and anodic gas products from a mass spectrometer that was operated concurrently with the GC. In Figure 3D we can now see that the CO2 to O2 ratio is initially near 2 entailing carbonate transfer, which is expected of a device with a well-functioning cathode. 12,13 However at the point where we start seeing a change in potential, we see a drop in the CO2 to O2 ratio entailing a partial switch from a carbonate transfer to a hydroxide transfer. Given that OH⁻ conducts more efficiently through the membrane, this at least partially explains the drop in potential. Furthermore, at the point when the membrane again becomes more carbonate-rich (i.e. CO2:O2 ≈ 2) the potential also increases again. It should be noted that the CO2:O2 ratio is a function of both anolyte equilibration time and GC measurement time, thus it is hard to quantify the lag between the CO2:O2 ratio and potential (also there was a lag of 9 min between the cathode and anode gas injection to the GC).
Another point to note is that during CO2 electrolysis the surface is covered by a certain percentage of CO, and thus this is known to decrease H2 evolution catalytic activity substantially. 37 In the case where the catalyst is CO2 deficient, this will entail enhanced activity and should result in a lower device potential. 37 Given that a CO2 deficiency will result in both an increase in membrane conductivity and a catalytic improvement for H2 evolution, these two parameters are intrinsically coupled, and thus concluding the relative impact of each is beyond the scope of this work (especially since this involves a chaotic oscillatory system being analyzed). If the fundamental issue was mainly flooding of the cathode, this would dilute the concentration of cathodically produced hydroxyl ions (OH⁻). The dilution of OH⁻'s would lower the pH, entailing a shift towards HCO3⁻ instead of CO3 2 ⁻, and an increasing CO2/O2 ratio approaching 4. 41 However, Figure 3D shows just the opposite trend, this decrease can only occur by a lack of CO2, entailing OH ⁻ transfer through the membrane since this case would have a CO2/O2 ratio of 0. Another quite interesting observation, though less clearly visible, is that the potential and shift in selectivity seem to be preceding the flooding slightly. This effect was seen in all synchrotron experiments where fluctuations were clearly visible (see Figures S19-S22 for additional experiments). As it is well known that salt precipitation occurs during CO2E and degrades performance, 20,22 our following experiments focused on monitoring salt formation and whether we could conclude that this was the instigator of the oscillations.

Salt precipitation in the cathode GDE
Salt depositions are in crystalline form, and thus can easily be isolated via their diffraction pattern (See supporting information for details). By operating another experiment, this time at 100 mA·cm -2 , we were able to achieve distinct oscillations while monitoring electrolyte and carbonate content as shown in Figure 4. Unfortunately, we do not have cathodic GC or MS data for this synchrotron experiment (due to a technical issue), however, the anode CO2 to O2 ratio ( Figure S23) shows trends consistent with previous experiments. One of the most interesting characteristics of Figure 4 is that salt precipitation occurs preceding the substantial increase in electrolyte content. These results are in contrast to previous reports that assumed salt precipitation is a result of massive electrolyte flooding. 7,8 Even more interesting is that KHCO3 was found (as validated by seeing all its XRD peaks -see Figure   S24), but no XRD peaks attributed to K2CO3 were found. Previous reports [21][22][23]42 have indicated seeing K2CO3 at the cathode, and given that it is known that CO3 2 ⁻ is primarily transferring through the membrane (see Figure 3 or Figure S23), it would be expected that any liquid electrolyte at the cathode would be in the form of K2CO3. While it is not clear why KHCO3 deposits to begin with instead of K2CO3, one hypothesis is that there become localized areas of more neutral pH's, due to higher CO2 concentrations near the flow field, thus allowing the formation of KHCO3 and since the solubility of KHCO3 is less than half that of K2CO3, 43 this leads to the salt deposition.
As we start (before electrolysis) with no potassium at the cathode and the solubility limit of both KHCO3 (3.62 M at 25 • C) 43  To exclude that such changes in anolyte concentration can cause oscillations, we performed a control experiment with 0.01 M KHCO3 as the anolyte. The experiment shows similar cell potential and selectivity behavior during the first 3 h as seen in Figure S25. As expected, oscillations occurred, though slightly later than what was observed in 0.1 M KHCO3 experiments indicating any change in potassium concentration (irrespective of the initial concentration) in the anolyte would not cause the observed oscillations.
While finding a clear reason for the formation of KHCO3 is challenging, there are highly plausible reasons for its disappearance once the electrolyte reaches the salt crystals. The most obvious is that the aqueous solution simply dissolves it, but a more subtle point is that this electrolyte is expected to be highly alkaline and lacking carbonates (as denoted by OH⁻ rather than CO3 2 ⁻ crossing over the membrane). This alkaline environment (i.e. KOH based) would entail that the KHCO3 rapidly converts to the much more soluble K2CO3, which would help in solubilizing the KHCO3 crystals and preventing the dissolved salt to precipitate even when the electrolyte level (in the GDE) is once again lowered.
To further confirm that salt precipitation is triggering the observed oscillations in CO2RR, we did multiple CO reduction (COR) experiments. As expected, we do not see any oscillating phenomenon in cell potential and COR products/HER selectivity ( Figure S26). In addition, no signs of salt precipitation were found in the cathode flow field post-COR.

Comprehensive analysis of oscillations
Based on the preceding analysis, we can link many of these effects to provide a comprehensive mechanism for the oscillatory nature seen during CO2 electrolysis. Figure 5 summarizes the mechanism we feel is plausible based on this analysis. It is important to note that often some of these processes are taking place simultaneously and the order in which they are taking place may be slightly different than proposed. Nevertheless, this does give an overview of the major processes taking place.
Initially, we observe a small amount of electrolyte deep into the GDL and we see the formation of KHCO3 crystals. This causes a blockage in the GDE preventing sufficient CO2 access. We base the hypothesis that the salt crystals are blocking the CO2 access on the fact that we see both a decrease in CO2 reduction selectivity in favor of HER and a decreased carbonate crossing over the membrane. Thus, from a pure mass transfer standpoint, there appears to be a loss of CO2 reaching the catalyst/membrane interface. This leads to anions transporting through the membrane shift to OH⁻, causing a drop in membrane ohmic losses and thereby decreasing the overall cell potential. Additionally, the lack of CO2 and its intermediate such as CO bound to the catalyst also increases catalytic activity and decreases the potential. Beyond this point, the electrolyte then floods the cathode. Given that the incoming cathode gas is dry CO2, this electrolyte/water must be coming from the anode, but the reason for this increased water penetration is still unclear. One factor that could lead to increased water penetration is the fact that membranes in the OH⁻ form are more hydrated than in the CO3 2 ⁻, 44 and this increased swelling could potentially lead to an increased water transfer from the anode to the cathode. This hypothesis is supported by clear observations of local water content increase inside the membrane (electrolyte heatmaps in Figures 2 to 4). However, more experiments would be needed to verify this effect. Another factor is that osmotic drag will naturally pull water from the anode to the cathode to hydrate the salts. Given the complexity of the system, there could be additional factors not accounted for, thus currently all that can be determined for certain is that water does return to the cathode. Figure 5. Proposed mechanism to illustrate the oscillating phenomenon of salt precipitation and dissolution causing a change in electrocatalytic performance of the CO2 electrolyzer. It should be noted that the illustration only describes the oscillations. An underlying performance degredation due to slow flooding and potassium build up at the cathode should be added. The total observed phenomena would therefore be more like a degrading spiral.

Conclusion
While the oscillatory effects seen in this work are clearly not desired from an applied standpoint, these oscillations enabled us to gain deep fundamental scientific insight into the processes that lead to loss of selectivity to CO2 products by analyzing these oscillations via monitoring a wide variety of parameters such as potential, cathode and anode products, Cu oxidation state, as well as location and intensity of Cu, electrolyte content and salt deposition in the GDE. Although our results showed that Cu is relatively stable (at least at the microscale), there is a highly interconnected relationship between electrolyte/water penetration into the GDE, salt deposition, membrane ion conduction (CO3 2 ⁻ vs. OH⁻), and overall device potential at a given current density. Though this work opens new questions, it does clarify that mitigating salt deposition throughout the gas diffusion layer is essential to obtaining long-term stability. Thus, further works on durability should focus on enhanced reactor designs, salt management processes, and more resistant gas diffusion layers to help resolve these issues.

Electrode preparation and zero-gap MEA setup
The cathode GDEs for all the experiments were prepared by sputtering a 150 nm layer of 6N Cu onto a Sigracet 39BB carbon paper (Fuel Cell Store) in a vacuum environment (10 -5 ~ 10 -6 Torr) at a deposition rate of ~ 1 Å s -1 under 5 sccm Ar, total pressure 3 mTorr, and at room temperature. The anode GDE was a commercial IrO2-coated GDE (Dioxide Materials).
All electrochemical CO2E tests were performed in a 0.64 cm 2 MEA-based CO2 electrolyzer.

Electrochemical measurements
A dry CO2 gas (for non-X-ray irradiated samples at DTU ( Figure 1C) Linde, 5N, for X-ray irradiated samples at ESRF Air Products, 5N) was supplied to the cathode flow field with a flow rate of 30 sccm using mass flow controller (Vöegtlin red-y smart series), while, the anode was fed with 0.1 M KHCO3 (Sigma-Aldrich, 99.99% metal basis). Cathode inlet pressure was measured using a pressure transducer (Brooks Instrument, SolidSense II Series). A constant volume of 60 ml anolyte was recirculated through the anode with a digitally controlled diaphragm pump (KNF NF1.5TTDCB-4) set to 25% corresponding to approximately 7.5 ml/min. The anolyte reservoir was purged with Ar (at 30 sccm) to carry the gases formed at the anode for gas chromatography analysis. All electrochemical measurements were performed under galvanostatic mode with a potentiostat (Biologic SP-300 and SP-240) and the cell voltages are reported without any iR correction. A leakless Ag|AgCl reference electrode was introduced before the anolyte inlet at the anode side to measure the anodic potential.

Product analysis
The outlet cathode gas stream was passed through a gas wash to extract the liquid CO2RR products. All cathodic gas products such as H2, CO, CH4, C2H4, and C3H6 products were

In-operando X-ray measurements
Synchrotron wide-angle X-ray scattering (WAXS) measurements were performed at the ID31 beamline of the European Synchrotron Radiation Facility (ESRF) in Grenoble, France. The high-energy X-ray beam (68 keV) was focused (5μmvertical x 20 μmhorizontal) on the MEA in grazing incidence mode (beam parallel to the MEA layers), and the scattered signal was collected using a Dectris Pilatus CdTe 2M detector positioned 710 mm behind the sample. The energy, detector distance, and tilts were calibrated using a standard CeO2 powder and the 2D diffraction patterns were reduced to the presented 1D curves using the pyFAI software package. 45 To avoid the high, spotty, and randomly distributed signal arising from the gasket in the cell, iterative filtering has been implemented in the azimuthal integration (referred to as "sigma_clip" integration in the azimuthalIntegrator Module in the pyFAI package).
Rietveld refinements of the WAXS patterns were performed to extract the crystallite sizes, lattice parameters, and microstrains using the Fm3m structure of Cu metal and the GSAS-II software. 46 The instrumental parameters were determined by the refinement of a CeO2 standard sample. As repetitive vertical scans across the cathode GDE and the membrane were performed during the experiment, the background subtraction consisted of removing the GDE signal from the catalyst signal.

Electrode characterization
The surface composition of the used cathode GDEs was determined using a Kratos Axis Ultra DLD X-ray photoelectron spectrometer (XPS). All the measured spectra were acquired using a monochromatized Al Kα radiation at 15 kV and 15 mA. The binding energy of the acquired spectra was calibrated using C 1s binding energy to 284.8 eV. Ion-scattering spectroscopy profiles of the cathode GDEs were also done to identify any contamination after CO2 electrolysis.

Supplementary information
The authors declare that the data supporting the findings of this study are available within the paper and its Supplementary Information files. The processed data for all figures are available at https://figshare.com/s/164b8df7d20e8332c15d including integrated X-ray data. Raw X-ray data generated at the European Synchrotron Radiation Facility (ESRF) large-scale facility are available at https://doi.esrf.fr/10.15151/ESRF-ES-439498753 from 2024. Alternatively, this data can be available from the corresponding author upon request. and the Villum Center for the Science of Sustainable Fuels and Chemicals grant 9455. We acknowledge the European Synchroton Radiation Facility (ESRF) for the provision of synchrotron radiation using beamline ID 31.

Declaration of interests
The authors declare no competing interests.