Graphene Electrode for Studying CO2 Electroreduction Nanocatalysts under Realistic Conditions in Microcells

The ability to resolve the dynamic evolution of electrocatalytically induced processes with electrochemical liquid‐phase electron microscopy (EM) is limited by the microcell configuration. Herein, a free‐standing tri‐layer graphene is integrated as a membrane and electrode material into the electrochemical chip and its suitability as a substrate electrode at the high cathodic potentials required for CO2 electroreduction (CO2ER) is evaluated. The three‐layer stacked graphene is transferred onto an in‐house fabricated single‐working electrode chip for use with bulk‐like reference and counter electrodes to facilitate evaluation of its effectiveness. Electrochemical measurements show that the graphene working electrode exhibits a wider inert cathodic potential range than the conventional glassy carbon electrode while achieving good charge transfer properties for nanocatalytic redox reactions. Operando scanning electron microscopy studies clearly demonstrate the improvement in spatial resolution but reveal a synergistic effect of the electron beam and the applied potential that limits the stability time window of the graphene‐based electrochemical chip. By optimizing the operating conditions, in situ monitoring of Cu nanocube degradation is achieved at the CO2ER potential of −1.1 V versus RHE. Thus, this improved microcell configuration allows EM observation of catalytic processes at potentials relevant to real systems.


Introduction
Electrochemical liquid-phase electron microscopy (ec-LPEM) is used to study solid-liquid interfacial processes that are activated by the application of a potential. [1]In transmission electron microscopy (TEM), ec-LPTEM has been implemented with remarkable findings on various electrochemical and electrocatalytic systems, including the highly prevalent activation pathways of CO 2 DOI: 10.1002/adma.202311133electroreduction (CO 2 ER) nanocatalysts. [2,3]t the same time, the need to bridge the scales for a better understanding of electrocatalytic phenomena has led to advancements in scanning EM instrumentation, which has also enabled the realization of electrochemical liquid-phase SEM (ec-LPSEM) experiments with valuable results. [4,5]In both cases, the challenging aspects of the techniques are similar; they mainly involve limitations in the spatial resolution, due to electron scattering caused by the presence of the liquid and the liquid-sealing membranes, and means of application of the potential, due to the geometrical and fabrication constraints of the substrate electrodes, which also contribute to the thickness of the cell. [6]Overall, there has been significant progress in controlling the liquid thickness and thus improving the spatial resolution of imaging samples in liquids. [7,8]However, in the case of electrochemical LPEM experiments, advances in system functionality have lagged behind.
The configuration of the electrochemical liquid-containing microcell for EM studies has remained mostly the same since its invention. [9]In detail, the typical liquid enclosure comprises of a pair of Si-based chips coated with a thin layer of liquid/gas impermeable SiN x membrane, which bulges under the pressure differential inside the high-vacuum column of the electron microscopes.To apply the potential, one of these chips, the top chip, is patterned, using standard microelectromechanical systems (MEMS) fabrication methods, with, typically, three electrodes: the working (WE), which is where imaging/spectroscopy of the sample's evolution is performed, the counter and the reference (CE and RE, respectively).Between these two chips, microchannels allow for liquid wetting on and circulation around the electron transparent area and are patterned on the bottom spacer chip.Under this configuration, the thickness (since the measurements are in projection) of the liquid is difficult to evaluate, however, control of the liquid thickness is critical because it directly affects the resolution of the acquired images and further complicates post-processing methodologies applied to extract meaningful information. [6]The substrate/electrode material adds to the overall electron scattering of the probed area degrading the spatial resolution while its electrochemical stability window needs to include potentials used to drive the reaction of interest. [10]To date, electrode materials are available for most of the reactions studied with LPEM, however, the spatial resolution remains a challenge.
Focusing on this setup, to improve the spatial resolution via controlling the liquid thickness, Keskin et al. connected a pump to the fluidic tubing of the liquid cell holder that was used to decrease the pressure difference between the inside of the liquid cell and microscope chamber. [7]While the resolution was improved, this method is primarily suitable for static liquid conditions and can affect the wetting properties of the substrate.Serra-Maia et al. proposed a technique that employed electrochemical water splitting for the generation of a gas bubble in the viewing region and consequently calculated the liquid to be ≈30 nm in thickness by performing electron energy-loss spectroscopy. [8]Earlier studies reported on a similar approach and stated that control over the liquid thickness when a gas bubble is formed is difficult as a result of intense electron-beam-induced radiolysis, [11] and therefore not suitable for sensitive samples.Also, this approach can severely interfere with studies on electrochemical reactions taking place at potentials other than that of the hydrogen or oxygen evolution reactions (HER or OER) but even in the case of OER or HER, the varying liquid thickness can influence the real-time imaging from the beginning of the processes.
Undoubtedly, high-resolution imaging in LPEM is primarily driven by the implementation of membrane materials in the place of SiN x for sample encapsulation.Graphene liquid cells are widespread nowadays. [12]In both single-crystal and polycrystal states, graphene has a Young's modulus value up to 3 times greater than the SiN x membranes. [13]Hence, graphene cells of different generations are reported to exhibit less bulging and enable thinner liquid layers, allowing high-resolution imaging of materials in liquid media. [7,14][16] Recently, Tan et al. added graphene as an electrode material, on top of the full SiN x membrane-based ec-LPTEM cell and demonstrated its use for Cu electroplating and stripping. [17]ith the aim to advance the functionality of ec-LPEM experiments, herein, we implement graphene in the microcells as membrane and working electrode material.We begin by describing the fabrication of a single-electrode electrochemical chip patterned with a single Pt electrical contact and hole-array on the SiN x membrane.Thereafter, tri-layer graphene was fabricated and transferred on the chip using wet transfer methods.To electrochemically evaluate the functionality and stability of the graphene electrode, we used Cu 2 O nanoparticles for CO 2 ER experiments.We further tested the cumulative effect of the ebeam and applied potential and employed the system for in situ SEM imaging of Cu nanocatalyst degradation under CO 2 ER conditions.

Results and Discussion
The lack of reports on the integration of graphene into electrochemical MEMS-based Si/SiN x chips both as electrode and membrane material presumably stems from the complexity of its transfer on the WE in the typical three-electrode system.The geometrical proximity of the WE to the CE and RE requires precise control of the transfer to avoid short circuits.Additionally, the integrity of the graphene layer during transfer needs to be retained in a way that ensures the sealing of the cell.To circumvent these challenges, we utilized an electrochemical SEM holder (Figure 1a) that includes bulk RE and CE [5] and allows for a single working electrode to be fabricated on the electrochemical chip.

Microfabrication of Graphene Working Electrode (Gr WE) on Si/SiN x Substrates
To integrate graphene as the WE for ec-LPSEM measurements, we started with a Si substrate having a suspended 50 nm SiN x membrane.A single Pt thin-film electrical contact was then patterned on the chip away from the liquid-containing area to eliminate its possible participation in the reactions upon operation, Figure 1b.The subsequent MEMS fabrication and graphene transfer step on the electron transparent area are illustrated in the top view and cross-sectional schematics in Figure 1c.For graphene's added function as the liquid-sealing membrane on the electrochemical chip, we first patterned an array of holes on the suspended region of the SiN x membrane.The approach of retaining the SiN x membrane and etching holes on it provides improved durability of the microcells due to the small area of graphene exposed to high vacuum conditions.The size of holes ranged from 1 to 3 μm, as shown in the SEM/TEM images of Figure 1e-g, which allowed their patterning by conventional photolithography tools.Subsequently, the graphene sheets were transferred onto the Si/SiN x chip via a solution-based method (see Experimental Section for more details) ensuring contact with the Pt film, as shown in Figure 1d.More specifically, to have sufficient spatial resolution and durability, tri-layer graphene was used, as confirmed by the high-resolution TEM (HRTEM) image and corresponding fast Fourier transform (FFT) shown in Figure 1h,i.Previous studies that used graphene to seal the liquid TEM cells have demonstrated the stability of graphene at a pressure differential of 3-6 bar while remaining well-adhered at liquid flow rates of up to 120 μL min −1 for three stacked layers and above. [14]

Inert Potential Range and Stability of the Gr WE under CO 2 ER Conditions
To evaluate the electrochemical properties of the Gr WE, we first performed cyclic voltammetry (CV) in CO 2 -saturated 0.1 M KHCO 3 , and compared it to the classical substrate of choice for CO 2 ER studies, an inhouse-made glassy carbon (GC) WE.Gr and GC chips were pretreated with hydrogen and air plasma, respectively, to functionalize their surfaces.Using the bulk RE and CEequipped ec-LPSEM holder, the measurements were carried out on the bench to eliminate possible instrumental interferencerelated effects on the wetting behavior inside the microscope.Figure 2a shows the background CV response of the SiN x membrane (i.e., with no substrate electrode or catalyst loaded) and bare Gr and GC electrodes (i.e., with no catalyst loaded).SiN x exhibits the characteristics of an insulator where the current density is constant around zero, as can be better seen at the inset of   2a) as a cut-off value above which the electrodes' response is no longer considered to be inert. [10]The inert cathodic potential range for GC WE reaches to −0.52 V versus RHE while the Gr WE stays electrochemically inert up to −0.69 V versus RHE, indicating that it is suitable for a substrate electrode for cathodic electrocatalysis applications.The results are in good agreement with previous studies reporting on the electrochemical behavior of graphene electrodes where multi-layer graphene was reported to exhibit slower heterogeneous electron transfer rates than other pyrolytic graphite electrodes [18] and the latter was shown to exhibit lower electrochemical activity than GC. [19]The origin of this behavior has been associated with the major sites that partake in fast electron transfer kinetics.In graphitic materials, the edges dominate the kinetics while in graphene there is a substantial contribution of basal planes compared to edges.In glassy carbon, the presence of graphite-like nanostructures and voids results in even higher activity than in graphitic materials.This general trend is well preserved in the electrochemical performance of our Gr and GC WEs.
Finally, to test the electrodes' electrochemical stability at −1.1 V versus RHE, which is one of the most commonly reported potential values for realistic CO 2 ER studies of Cu-based catalysts, we performed chronoamperometry (CA).Figure 2b compares the SiN x , Gr, and GC substrates under CA for 10 min.The Gr WE exhibits lower absolute current density than GC WE while its value is decreasing over time.Experiments on several bare graphene chips show that the current density is similar, confirming the re-producibility of the transfer process (Figure S1, Supporting Information).In addition, we observed that after wet etching of the Cu foil (substrate layer for graphene growth), there can be Cu residues on the graphene that are responsible for the subtle redox features in the CVs (Figure S1a, Supporting Information).However, this trace amount of Cu does not have a significant effect on the current density profile of graphene (Figure S1b, Supporting Information).The current fluctuations are attributed to gas bubble formation, which in turn could stem from the surface roughness that in the case of the GC WE is expected to be higher than that of graphene.As bubbles form and grow on the surface of the electrode, they cause fluctuations in the active surface area of the electrode. [20]With the collapse of bubbles, the drastic increase in the surface area of the electrode exposed to the electrolyte induces spikes in the current profile. [21]Overall, graphene's electrochemical performance including its wide inert potential range, compared to the state-of-the-art GC electrode, and smooth current density profile, confirms its stability at potentials lying beyond that of HER.

Charge Transfer Properties of the Gr WE on Cu 2 O
Apart from the stability of Gr WEs across the inert potential range, their efficacy toward electrocatalyst studies depends on the charge transfer properties at the graphene-catalyst interface when catalysts are loaded on the electrode.To test this, we used the redox features of cuprous oxide nanocubes (NCs) drop-casted on the Gr WE.The TEM and selected area electron diffraction (SAED) characterization of the Cu 2 O NCs are depicted in Figure 3a,b.The average size of the cubes is 40 nm.plots the CV response in CO 2 -saturated 0.1 m KHCO 3 of bare Gr WE and NCs loaded-Gr WE.The CV of catalyst-loaded Gr WE shows a decrease in the onset potential value with respect to one of the bare Gr WE, which is due to the active CO 2 ER and HER processes occurring at Cu-based catalysts, as opposed to the bare Gr WE.Notably, the observed increase in the cathodic current density of the catalyst-loaded Gr WE corroborates the efficient electron transfer from the graphene electrode to Cu 2 O NCs, The results further indicate the quasi-reversible Cu redox reaction and Cu precipitation.Starting from OCV toward negative potential, the CV curve shows two subsequent cathodic peaks corresponding to the reduction of Cu 2+ to Cu + (peak C 1 ) and further Cu + to Cu 0 (peak C 2 ).There is a third cathodic peak (C 3 ) that takes place at more negative potentials and this is most likely associated with the precipitation of dissolved Cu species. [22]Meanwhile, the positive scan depicts a single anodic peak, attributed to the Cu 0 to Cu + oxidation (A 1 ) (Figure 3c, inset).We note that these observations are in good agreement with experimental results obtained for Cu 2 O NCs on GC WE (Figure S2, Supporting Information), although, due to the inevitable variations/uncertainty in the amount of catalyst loading, the current densities between catalyst drop-casted Gr and GC WEs are not comparable.In ad-dition, CA measurements shown in Figure 3d indicate instabilities in the current profile recorded at −1.1 V versus RHE that are related to fluctuations in the electrochemically active surface area of catalyst and/or catalytic reactions. [20]The current spikes are related to gas bubble generation/release reported in earlier studies. [21]Control experiments (i.e., CV and CA measurements) with Cu 2 O NCs drop-casted on bare SiN x chip show close to zero current response for applied potential in accordance with the insulating nature of the SiN x substrate, which confirms that the detected current response in our experiments corresponds to the graphene and catalyst system (Figure S3, Supporting Information).Overall, the electrochemical response of Cu 2 O NCs loaded Gr WE at CO 2 ER-relevant conditions confirms the effectiveness of charge transfer at the graphene-catalyst interface.

ec-LPSEM Studies of Cu NCs under CO 2 ER Conditions
For in situ ec-LPSEM studies, the electrochemical applicability of the graphene membrane/electrode design was investigated for the CO 2 ER-active Cu NC catalysts, Figure 4a.The SAED pattern in Figure 4b confirms the metallic structure of the pristine cubes.Prior to the operando experiments, we tested the stability of the catalysts/microcell under e-beam irradiation and we evaluated the combined effect of the e-beam and applied potential on the system.

Properties of the In Situ Microcell
Cu NCs were dropcasted on the electrochemical Gr singleelectrode chip and the microcell was formed by stacking the electrochemical chip with a windowless spacer chip.The liquid electrolyte, CO 2 saturated 0.1 m KHCO 3 , was flown after insertion of the stage into the SEM high-vacuum chamber.A schematic illustration of the cross-section of the cell inside the SEM is depicted in Figure 4c.In this configuration, the achieved spatial resolution is expected to be affected by the membrane material(s) and the liquid thickness.The total liquid thickness can be determined by the spacer size and the bulging of the SiN x /Gr membrane.Concerning graphene membranes, it has been previously reported that their bulging increases with the increase of hole diameter. [14]Thus, for the 1-3 μm diameter-sized holes, we expect the graphene membrane to have little influence on the bulging of the microcell, which will be defined by the properties of the SiN x membrane.However, in the case of secondary electron (SE) detection in SEM, the spatial resolution is primarily dependent on the thickness of the membrane whereas the absolute liquid thickness does not influence the SEM resolution, which becomes limiting in transmission electron studies.To illustrate this effect, we performed Monte Carlo simulations of the SiN x and the trilayer graphene membranes (Figure S4, Supporting Information), and the calculations reveal the broadening of the incident electron beam upon impact on thicker membranes.Moreover, the excited SEs at the top surface of the cell have to travel through the membrane to the SE detector.Compared to the 50 nm thick SiN x , the suspended tri-layer Gr membrane obstructs imaging much less (i.e., the SEs have a high escape probability), leading to better spatial resolution. [23,24]We also note that there is an observed improvement in the image quality of the nanocatalysts that are attached to the membrane upon liquid infusion as opposed to dry cell conditions (Figure S5, Supporting Information).This is probably due to the high scattering of the primary electrons in the liquid that leads to the minimization of artifacts from backscattered produced SEs. [25]Hence, for SE imaging of nanoparticles in liquid, the spatial resolution is expected to be dominated by the thickness of the membrane and not degrade with the increase of the liquid thickness in the cell.Figure 4d demonstrates this effect and depicts the clear improvement of the spatial resolution in the liquid-containing cell under the Gr membrane as opposed to the SiN x /Gr region.The achievable resolution was 18 nm ± 4 nm calculated by the particle edge profile method (Figure S6, Supporting Information).Figure 4e shows the evolution of the region in Figure 4d at 5 kV and 32 pA after 5 min of continuous irradiation.There is no visible change of the Cu NC agglomerates under the graphene-covered region apart from an increase in their separation, which could result from bulging (i.e., expanding) of the graphene membrane during imaging.Meanwhile, the Gr/SiN x region shows electron beam-induced dissolution and redeposition of Cu NCs, possibly due to trapping of SEs under the SiN x membrane.

Synergistic Effect of Electron Beam and Application of Potential
Next, we evaluated the combined effect of the e-beam irradiation and the applied potential to determine the conditions that enable operando ec-LPSEM experiments of the Cu NCs in CO 2saturated 0.1 m KHCO 3 .SEM imag at 5 kV and 32 pA was performed under a linear sweep voltammetry (LSV) protocol to imitate the start condition of the catalytic cell, followed by potentiostatic hold at varying cathodic potentials of −0.8, −0.9, and −1.1 V versus RHE to replicate the cell operation conditions.Figure 5a depicts the potential and current profile for LSV and CA curves at −0.8 V versus RHE together with SEM images corresponding to the beginning and the end of the measurement.The stability of the graphene electrode while holding this potential for 10 min is attained while the dissolution of the primary particles is also observed (Movie S1, Supporting Information).For increasing values of the cathodic potential, there is noticeable degradation of the graphene electrode within 4 min (Figure 5b), which becomes more enhanced at more cathodic potentials (Figure 5c) while no apparent transformation of particles is present at these potentials.Degradation of graphene begins at ≈2 min (Movie S2, Supporting Information) and 1 min 30 s (Movie S3, Supporting Information) for −0.9 V versus RHE and -1.1 V versus RHE, respectively, and reveals an apparent dependence on applied potential during SE imaging.It is worth to note that degradation begins once the set potential value is reached (i.e., during chronoamperometry).In addition, for all potential values, the Cu NCs at non-electron beam exposed regions undergo more pronounced dissolution/redeposition, as shown in Figure 5d,e.[28] In addition, another study has shown an increase in the intensity of characteristic D peak in graphene's Raman spectra after e-beam irradiation in the presence of water that was attributed to the interaction of graphene with radiolytic products. [29]Considering that the instability of graphene was mainly observed in imaged regions under applied potential, we can conclude that there is a synergistic effect of the electron beam and the applied cathodic potential on the electronic properties of graphene, which appears to influence the observed nanocatalyst transformations.Nevertheless, this e-beam irradiation/potentialinduced effect can be minimized either by decreasing the applied  potential or by further decreasing the probe current (i.e., electron dose).

Operando CO 2 ER of Cu NCs at −1.1 V versus RHE
For operando ec-LPSEM, we lowered the probe current to 16 pA at 5 kV, to minimize the effect of the electron beam irradiation on the graphene electrode for the duration of the 3 min LSV-CA experiment.Figure 6a depicts the overview SEM image of the Cu NCs at the beginning of the experiment.The potential profile applied to the Cu NCs along with the detected current response is shown in Figure 6b.During LSV, we observed a single cathodic peak that corresponds to Cu + → Cu 0 reduction.This behavior is associated with the tendency of surface Cu 0 to oxidize to Cu + under OCV conditions. [2]Figure 6c shows a sequence of SEM images depicting nucleation and subsequent growth process of secondary Cu particles.Temporal synchronization of the electrochemical response and recorded image sequence revealed that the onset for reprecipitation of dissolved Cu species to secondary particles takes place at potentiostatic hold (Movie S4, Supporting Information).It has been reported that the reduction of OCV oxidized Cu nanospheres under cathodic potential scan at −0.8 V versus RHE proceeds via mild dissolution of the outmost surface layer of Cu particles and redeposition of the dissolved Cu + species under reducing cathodic potential hold. [2]In our case, the limited spatial and temporal resolution during the SEM imaging does not allow the evaluation of a possible area loss of the primary particles, whose dissolution rate has been reported to initially proceed rapidly by the reduction of the surface oxide layer. [4]It has also been reported that the dissolution of metallic Cu is slower than in cupric oxides, [30] and considering that we do not observe a noticeable change in the Cu nanocube morphology, we can conclude that the inevitable OCV-induced Cu NC oxidation is limited to a thin layer (i.e., not observable under the conditions used).In addition, the discernible redeposition of secondary particles, during the in situ SEM experiments at the cathodic potential of −1.1 V versus RHE, suggests that the simultaneous dissolution of the primary particles continues with a moderate amount of dissolved species, which is dependent on the applied, cathodic potential. [22,31]Post-mortem scanning (S)TEM characterization of the chip reveals that the primary nanocubes underwent dissolution which according to bright field (BF)-TEM image (Figure S7a, Supporting Information) was initiated at the edges of the particles as they have the highest surface energy among the facets of the cubes.The size of secondary particles varies from as large as primary cubes to as small as a few nanometers which in turn could result from several degradation mechanisms reported earlier such as redeposition of secondary particles accompanied by Ostwald ripening and nanoclustering/fragmentation of primary particles. [32]The oxidized nature of all particles to the cuprous oxide phase (Cu 2 O) due to air exposure is confirmed by the post-mortem SAED pattern and elemental map (Figure S7b,c, Supporting Information) that include both primary and secondary particles and the localized FFT analysis from high-resolution TEM images (Figure S7d-f, Supporting Information).

Conclusion
We have demonstrated that free-standing tri-layer graphene is mechanically robust and electrochemically inert, and can be used as a viewing membrane and substrate electrode for electrochemical liquid-phase EM experiments on the CO 2 ER of Cu nanocubes.It was shown that the stability of graphene during in situ SEM experiments is determined by the synergistic effect of electron beam and applied potential, with the effect increasing at high negative potentials.Using a low probe current, in situ SEM monitoring of Cu nanocube degradation and secondary particle redeposition was achieved at the operating CO 2 ER potential of −1.1 V versus RHE.Thus, customized solutions for liquid encapsulation and electrochemical stimuli for operando EM studies, aided by the advances in liquid cell electrochemical holders, can enable high spatial resolution imaging of electrochemically driven processes that approach realistic catalytic cell conditions.

Experimental Section
Microfabrication of Electrochemical Chip: Electrochemical chips were fabricated using a double-side polished 200 μm thick Si wafer covered with a 50 nm thick SiN x layer on both sides.On the back side of the wafer, SiN x was photolithographically patterned (1.2 μm thick AZ ECI 3007 positive tone resist; Süss MicroTec MA6 mask aligner) and etched with reactive ion etching (RIE, SPTS APS dry etcher).The patterned SiN x served as a hard mask for Si wet etching in potassium hydroxide solution (20% KOH, 80 °C, 2 h 15 min) that resulted in suspended 50 nm SiN x membranes on the front side of the wafer.On-chip electrical contact was obtained via double layer lift-off method, where the electrode pattern was obtained photolithographically (0.4 μm LOR/1.1 μm AZ 1512 HS, Heidelberg Instruments MLA150 laser writer).The electrical contact was deposited with ebeam evaporation (Leybold Optics LAB 600H evaporator) of Ti (adhesion layer)/Pt (5 nm Ti/45 nm Pt) on the wafer.Excess photoresist and evaporated metal were lifted-off by keeping the wafer in a positive photoresist stripping solution (Remover 1165, MICROPOSIT) for 24 h.The hole patterning of the suspended SiN x membrane was done utilizing either RIE for hole patterning on a wafer scale or focused ion beam (FIB; Zeiss Cross-Beam 540 and Zeiss NVision 40) for single chip scale hole patterning.Photolithography (0.6 μm thick AZ ECI 3007 positive tone resist; Heidelberg Instruments VPG200 laser writer) was used to pattern a hole array of different designs and RIE (SPTS APS dry etcher) was utilized to obtain a micron-sized hole array in 50 nm thick SiN x support membrane.For some of the chips, FIB was used to obtain the hole as it provided similar results and could be used to produce smaller-sized openings.
The glassy carbon working electrode was custom-fabricated using pyrolysis.First, SU-8 2000.5 epoxy resin (Kayaku Advanced Materials, 800 nm thick) was patterned on a 200 μm thick Si wafer deposited with a 50 nm low-stress SiN x membrane.Then the wafer with the patterned electrodes was annealed in forming gas (3% H 2 in N 2 ) at 900 °C for 30 min.The thickness of the pyrolyzed epoxy was reduced by a factor of 10 that resulted in an 80 nm thick glassy carbon electrode pattern.
Tri-Layer Graphene Fabrication: The stacking of three graphene layers followed an earlier reported method. [14]In detail, monolayer graphene grown on Cu foil (Graphenea) was spin-coated with polymethyl methacrylate (PMMA A4) to serve as a supporting layer for the wet transfer.The backside of PMMA-coated Cu foil was then treated with air plasma (100% power, 10 sccm, 30 s) to remove possible graphene residues for effective Cu wet etching.Then, the PMMA-coated graphene was floated onto a 0.1 m ammonium persulfate solution (Thermo Scientific Chemicals, 98%) to etch the Cu foil.Thereafter, the PMMA-coated graphene was transferred to a deionized (DI) water bath for 15 min to rinse the residues of etchant.This process was repeated 3 times.A glass slide (rinsed with acetone and IPA) was used to transfer the graphene from one bath to another.At the next step, the graphene was scooped with another piece of monolayer graphene-covered Cu foil (pretreated with air plasma; see above) and left for drying on the hot-plate at 45 °C for 30 min.The abovedescribed process was repeated one more time to obtain PMMA-covered 3-layer graphene on Cu foil.Smaller pieces of foil were cut and manually transferred on the electrochemical chip.As this process was done manually for each chip, the geometrical area of transferred graphene varied from chip to chip (Figure S8a, Supporting Information).In addition, it was noted that the electrochemically active surface area of the graphene electrode was the one that was in contact with the electrolyte and this was defined by the o-ring used to seal the microcell (Figure S8b, Supporting Information).
Graphene Transfer on Electrochemical Chips: For the electrochemical experiments, rectangular graphene pieces were first cut to size from the prepared PMMA/3-layer graphene covered Cu foil and then the Cu foil was etched following the process described above.The PMMA/graphene suspended on DI water was transferred on the chip covering the Pt electrical contact and membrane region.After drying the chip on a hotplate at 55 °C, PMMA was stripped by dipping the chip in acetone and ethanol (both solvents have analytical grade purity) bath at 55 °C for 10 min each.Once removed from the solvent, the chip was left to dry on the hotplate for another 2 min.
Assembly of the Cell and Electrochemical Measurements in SEM Holder: Before assembling the electrochemical holder, the graphene chip was pretreated with H 2 plasma (50% power, 30 sccm, 30 s) and glassy carbon chip with air plasma (100% power, 10 sccm, 50 sec) to clean possible residual contamination and improve the wettability of the electrode.The bottom, spacer chip (400 nm) was pretreated with air plasma (100% power, 10 sccm, 50 s).Cu 2 O and Cu nanocubes were synthesized with previously reported protocols. [33,34]Dispersions of Cu 2 O nanocubes (NCs, 40 nm) were drop-casted on the plasma-treated graphene and glassy carbon working electrodes that resulted in 0.14 μg of Cu 2 O coverage.A similar procedure was applied for Cu NCs drop-casting, while the loading was reduced to 0.07 μg to avoid agglomeration of particles.The cell was assembled on a liquid cell SEM holder (Hummingbird Scientific Inc.) equipped with bulk reference (Ag/AgCl) and counter (carbon) electrodes.Once assembled, the microcell was filled with CO 2 -saturated 0.1 m KHCO 3 electrolyte.Liquid filling of the microcell was confirmed under an optical microscope and from the stabilization of an open circuit potential around the relevant value.All electrochemical measurements were performed with the BioLogic SP-200 potentiostat equipped with an ultra-low current probe.
Electron Imaging: SEM imaging was performed on a Thermo Fisher Scientific Quattro S Environmental SEM equipped with a SE Everhart-Thornley detector.In situ recordings were acquired at 5 kV with a probe current of 16 to 32 pA, and 1 μs dwell time.
High-resolution TEM imaging was performed on a Thermo Fisher Scientific Titan Themis 60-300 equipped with a spherical aberration image corrector.The measurements were done at 60 kV with an electron beam current of 30 pA.
Post-mortem characterization of Cu NCs including bright field TEM, high-resolution TEM, high angle annular dark field (HAADF) STEM imaging, and elemental mapping was performed on a Thermo Fisher Scientific Talos F200S operated at 200 kV in TEM, and STEM mode.Energy dispersive spectroscopy maps were acquired in HAADF STEM mode with 1 nA probe current.
Data Processing: For Movies S1-S3 (Supporting Information), the image stack was treated with a 3D Gaussian filter in ImageJ.For Movie S4 (Supporting Information), images were denoised using the total variation Chambolle method and further image stack was drift corrected in ImageJ utilizing the StackReg plugin. [35,36]Electrochemical data were synchronized with the recording and animated into frames with a custom-written Python code.Combining the SEM recording and electrochemical curves was done in ImageJ.
Statistical Analysis: Statistical analysis where relevant was performed by calculating the mean value and standard deviation (SD) of a sample size of three.Statistical data was presented as mean ± SD.

Figure 1 .
Figure 1.a) Schematic of the liquid cell SEM stage equipped with bulk reference (RE) and counter electrode (CE).b) Optical microscope image of the electrochemical chip with Pt electrical connection and hole-patterned membrane.Scale bar, 500 μm.c) Fabrication process of the graphene chip in plan view and cross-section.Not to scale.d) Optical microscope image of the as-fabricated graphene electrode chip.Grey dashed lines indicate the graphene-covered region on the chip.Scale bar, 500 μm.e,f) SEM images at increasing magnification of the 3 μm hole array patterned on the SiN x membrane.Scale bars, 50 and 10 μm for (e) and (f) respectively.g) TEM image of graphene transferred onto a 3 μm hole in the SiN x .Scale bar, 1 μm.h) High-resolution TEM (HRTEM) image of tri-layer graphene.Scale bar, 2 nm.i) FFT of (h) showing three rotationally shifted hexagonal lattice reflections.Scale bar, 0.2 nm −1 .

Figure 2a .
Figure2a.For the Gr and GC WEs, the current density increases as the potential sweeps from open circuit voltage (OCV) to −1.1 V versus RHE.Following previous protocols, we defined the current density value of −0.02 mA cm −2 (dashed line in Figure2a) as a cut-off value above which the electrodes' response is no longer considered to be inert.[10]The inert cathodic potential range for GC WE reaches to −0.52 V versus RHE while the Gr WE stays electrochemically inert up to −0.69 V versus RHE, indicating that it is suitable for a substrate electrode for cathodic electrocatalysis applications.The results are in good agreement with previous studies reporting on the electrochemical behavior of graphene electrodes where multi-layer graphene was reported to exhibit slower heterogeneous electron transfer rates than other pyrolytic graphite electrodes[18] and the latter was shown to exhibit lower electrochemical activity than GC.[19]The origin of this behavior has been associated with the major sites that partake in fast electron transfer kinetics.In graphitic materials, the edges dominate the kinetics while in graphene there is a substantial contribution of basal planes compared to edges.In glassy carbon, the presence of graphite-like nanostructures and voids results in even higher activity than in graphitic materials.This general trend is well preserved in the electrochemical performance of our Gr and GC WEs.Finally, to test the electrodes' electrochemical stability at −1.1 V versus RHE, which is one of the most commonly reported potential values for realistic CO 2 ER studies of Cu-based catalysts, we performed chronoamperometry (CA).Figure2bcompares the SiN x , Gr, and GC substrates under CA for 10 min.The Gr WE exhibits lower absolute current density than GC WE while its value is decreasing over time.Experiments on several bare graphene chips show that the current density is similar, confirming the re-

Figure 2 .
Figure 2. a) CV of SiN x, Gr, and GC in SEM holder.The dashed line corresponds to the cut-off current density value (−0.02 mA cm −2 ) for determination of the inert potential range.Inset shows an enlarged view of the CV for SiN x .b) CA at −1.1 V versus RHE for SiN x, Gr, and GC chips.

Figure 3 .
Figure 3. a) TEM image of Cu 2 O NCs. Scale bar, 40 nm.b) Corresponding SAED pattern.Scale bar, 2 nm −1 .c) Cyclic voltammetry of bare Gr and Cu 2 O NCs drop-casted Gr taken in the SEM holder.The inset shows an enlarged region of CV with Cu 2 O NCs. d) CA at −1.1 V versus RHE for bare Gr and Cu 2 O NCs drop-casted Gr electrode chips.

Figure 4 .
Figure 4. a) TEM image of Cu NCs.Scale bar, 40 nm.b) Corresponding SAED pattern.Scale bar, 2 nm − 1 .c) Schematic illustrating a cross-sectional view of the assembled microcell loaded with Cu NCs inside the SEM chamber along with the position of the Everhart-Thornley detector used for SE imaging.Not to scale.d,e) SE images of Cu NCs in CO 2 saturated 0.1 m KHCO 3 electrolyte at 0 and 5 min, respectively, of a dose test performed at 5 kV high voltage and 32 pA probe current.Scale bar, 500 nm.White arrows in image (e) indicate particles formed as a result of electron beam-induced degradation taking place on the SiN x -covered part.

Figure 5 .
Figure 5. a) LSV-CA at −0.8 V versus RHE and images at 0 and 10 min.b) LSV-CA at −0.9 V versus RHE and images at 0 and 4 min.c) LSV-CA at −1.1 V versus RHE and images at 0 and 4 min.Blue arrows (in a) point to degraded primary particles.White arrows (in b,c) indicate graphene degradation regions after LSV-CA at −0.9 and −1.1 V versus RHE.SEM images d) before and e) after subsequent LSV-CA measurements at −1.1 and −0.8 V versus RHE of a region that was not exposed to the electron beam.Blue and green arrows (in d,e) point to degraded primary and secondary particles, respectively.Scale bar, 400 nm.

Figure 6 .
Figure 6.a) SEM image of the initial Cu NCs.Scale bar, 400 nm.b) Potential and current profiles for OCV (5 s), LSV (30 s), and CA (140 s) measurements at −1.1 V versus RHE.c) Time-lapse SEM images of primary (blue frame) and secondary (green frame) Cu particles of regions shown in (a).Green arrows indicate the initial redeposition of secondary particles.Scale bar, 100 nm.