Main

Nanoporous atomically thin membranes (NATMs) have emerged as a promising platform for effectively separating molecules and ions, exploiting mass, size and affinity differences1,2,3,4,5,6. Despite progress on the fundamental aspects of NATM, including transport mechanism, pore incorporation and membrane fabrication, this field faces an outstanding critical challenge on the scalability of NATMs. The most popular and studied NATM is porous single-layer graphene (PG). The successful incorporation of carbon dioxide (CO2)-selective pores in PG has led to attractive performance for carbon capture2,7,8,9. This involves selective separation of CO2 from nitrogen (N2). It is crucial to develop high-performance carbon capture membranes that reduce the energy consumption and the associated cost of carbon capture from point-emission sources. Indeed, techno-economic assessments of the capture process based on high-performance PG membranes indicate that the energy efficiency of capture can be substantially improved compared with the commercial amine-based absorption process10,11. This is mainly because the membrane processes do not require thermal energy but rather rely on electrical energy for feed compression and permeate evacuation12,13. This is also because PG membranes yield extremely high CO2 permeance resulting from their atom-thin selective layer. This minimizes the required membrane area and the capture process footprint. This, combined with the exceptional chemical stability of graphene, makes capture based on PG membranes advantageous for carbon capture compared with the state-of-the-art membranes based on polymers14,15, zeolites16,17, metal–organic frameworks18,19, covalent organic frameworks20,21 and carbon molecular sieves22,23. A low-footprint capture process is attractive for application in the transportation sector, especially the international shipping industry, responsible for 2–3% of global CO2 emissions.

An early proof-of-concept study on the selective transport of CO2 from PG was demonstrated by Bunch and coworkers in 2012 (ref. 2). This was a fundamental study from a micrometer-sized exfoliated graphene with only a couple of pores responsible for gas transport. However, for practical membranes, one must prepare macroscopic films. For this, polycrystalline graphene, produced by chemical vapor deposition (CVD) on Cu foil, has become a standard material24,25,26. Yet, it is challenging to prepare a large-area PG membrane. Several aspects of the PG membrane must be debottlenecked to become commercially attractive for carbon capture. This includes the reduction of membrane element cost, the scalable incorporation of CO2-selective pores in graphene and the crack-free fabrication of membranes. These challenges are elaborated below.

Currently, the cost of commercial CVD graphene is extremely high for its application in carbon capture membranes. It is unclear whether low-cost Cu foils (approximately US$10 m−2), which contain micrometer-sized particulate contaminations from rough processing, can be suitable for producing graphene for CO2-selective membranes. This is mainly because strict control of substrate smoothness and particulate contaminations is needed to achieve attractive CO2/N2 selectivity given the sub-ångström difference in the size of the two molecules. Literature studies often use expensive Cu foils (US$500–10,000 m−2) or delicate processing methods (for example, electropolishing) to prepare high-performance graphene membranes for carbon capture7,9,24,27,28,29. This high raw material cost is prohibitive for the carbon capture application. By comparison, polymeric membranes for carbon capture have a membrane element cost of US$20–50 m−2 (refs. 30,31). Incidentally, the major cost of CVD graphene is related to the expensive Cu foil. However, Cu foils are also produced cheaply (approximately US$10 m−2). An important challenge here is that low-cost Cu foils have surface imperfections in the form of micrometer-sized particles and large grooves and scratches. These imperfections are detrimental to the fabrication of high-quality graphene membranes for carbon capture because they compromise the integrity of the suspended graphene by generating cracks7,32,33.

The second challenge concerns the lack of demonstration of CO2-selective pore incorporation in graphene over a large area. Pores in graphene are carbon vacancy defects incorporated by removing atoms from the basal plane. Chemical routes involving the gasification of the lattice by oxidation are intrinsically more scalable than carbon-knockout routes using energetic beams. For oxidation, oxidative plasma34, oxygen (O2) (refs. 35,36), ultraviolet light/ozone (O3) (ref. 33) or O3 (refs. 9,29) have been demonstrated. However, they involved sophisticated setups that are challenging to scale up. The uniformity of pore incorporation over a large area is not demonstrated.

The third challenge relates to fabricating large-area PG membranes while avoiding cracks in graphene. Cracks typically manifest during the transfer of CVD graphene from the Cu foil to porous support or during pressurization of the PG membrane. Wafer-scale24 and roll-to-roll26 transfer have been demonstrated in fabricating graphene devices for electronic applications. However, while these applications are tolerant to macroscopic cracks in graphene, gas separation membranes are not. Crack-free PG is essential for selective transport between similarly sized gas molecules such as CO2 and N2 under pressurized conditions. Centimeter-scale PG membranes have been demonstrated for nanofiltration after stitching cracks using interfacial polymerization33,37,38. However, this approach has not been demonstrated for gas separation. Prior works have shown that cracks in PG can be avoided by using gas-permeable mechanically reinforcing support film (MRF). Typically, MRF is deposited on PG, and the Cu foil is removed by wet-chemical etching, after which the floating composite PG–MRF is scooped on the desired porous support24,39,40,41. MRF can be a nanoporous carbon (NPC) film7, highly permeable polymers such as poly[1-(trimethylsilyl)-1-propyne] (PTMSP)8, polydimethylsiloxane42 or a composite film such as multiwalled carbon nanotube deposited on NPC43. However, the success rate of achieving centimeter-scale gas-selective membranes (defined as the number of selective membranes normalized by the number of attempts), even using the MRF strategy, remains low. This is mainly because the transfer protocol involves floating and scooping steps where cracks can easily develop in PG. Transfer methods involving careful scooping of floating PG are also extremely challenging to scale up. Therefore, a key challenge is to develop a membrane fabrication route avoiding careful scooping of graphene, and where the success rate is nearly 100%.

Here, we present several advances toward scaling up PG membranes for carbon capture. We demonstrate the successful use of a low-cost Cu foil (US$10 m−2) to prepare PG membranes with attractive CO2/N2 separation performance with a near 100% success rate. A simple protocol was developed to eliminate large contaminant particles on the surface of low-cost Cu foil. A large reactor for pore incorporation was implemented, which allowed the incorporation of uniform pores in large-area graphene (500 cm2) in a reproducible manner. A systematic study of graphene oxidation revealed that O3 mass transfer (velocity), as opposed to the commonly studied reaction kinetics (temperature and time), dominates pore formation kinetics. A facile and user-friendly transfer protocol was developed, completely avoiding the floating step, resulting in a highly reproducible synthesis. These interventions yielded attractive performance with a near 100% success rate from PG membranes.

Results

High-quality graphene from low-cost Cu foil

For synthesizing high-quality large-area graphene on a low-cost Cu foil, a CVD system consisting of a 160-cm-long furnace with a uniform heating zone of 110 cm, a gas delivery and a pressure management subsystem was developed to achieve a uniform temperature and to manipulate gas composition and pressure in the reactor from vacuum to 1 bar (Fig. 1a,b and Supplementary Fig. 1). Two dumbbell-shaped blocks made of Cu are placed at the end of the reactor as radiation shields to prevent excessive heating of flanges sealing the reactor. Two large coupons of Cu foils, each with a size of 11 × 26 cm2, could be introduced for graphene synthesis (Fig. 1c).

Fig. 1: Synthesis of large-area graphene on low-cost Cu foil.

a, A home-built CVD reactor for the synthesis of large-area graphene. b, Three-dimensional cross-section model of the reaction chamber equipped with radiation shields, an alumina tube and a 110-cm-long sample plate inside the heating zone. c, Two 11 × 26 cm2 single-layer graphene coupons synthesized in a single batch in the CVD reactor. d, The side view of the CVD reactor hosting a 16-cm-diameter alumina tube. e–h, SEM images of graphene surface grown on as-received (e) and acid-treated (f) Cu foil and the corresponding contamination particle density (for as-received (g) and acid-treated (h) Cu foil). The error bar refers to the standard deviation in particle density across three different spots for different particle sizes. The center of the error bar represents the average particle density calculated from the SEM analysis. i–n, Optical microscope images of as-received (i and j) and acid-treated (k and l) Cu foil before (i and k) and after (j and l) graphene growth. Profilometry mapping images of as-received (m) and acid-treated (n) Cu foil after graphene growth. o, The Raman spectrum of graphene transferred on a Si/SiO2 wafer. p, An AC-HRTEM image of single-layer graphene synthesized in this work. q,r, Raman mappings of D and G peak intensity ratio (ID/IG, q) and 2D and G peak intensity ratio (I2D/IG, r) across an area of 48 × 80 µm2. RMS, root mean square.

Source data

Imperfection in Cu foil consists of surface contaminations (particles) and grooves. High-purity expensive Cu foils have a low concentration of particles and have been reported to synthesize graphene after simple solvent washing. They are also free of large grooves and scratches. However, their cost is prohibitive for membrane application (Supplementary Table 1). For low-cost Cu, micrometer-sized contamination particles are present on the surface of the Cu foil and are probably deposited during the process of rolling Cu foils. Another source of contamination, irrespective of Cu foil, is the reaction between Cu vapor and the reactor wall made of a fused quartz tube, where SiOx particles are formed. These particles eventually deposit on the Cu foil, affecting graphene quality44 (Supplementary Fig. 2a). These were largely eliminated by placing a high-purity (99.7%) alumina tube (diameter ∅ = 16 cm) inside the quartz tube, which screened off the quartz tube from Cu vapor7,8,43,45 (Fig. 1d).

We chose a low-cost Cu foil (down to US$10 m−2; Supplementary Table 1) and developed a simple protocol for removing contamination, grooves and scratches. Micrometer-sized particles formed by several elements (Ca, K, Al or Cl) could be observed on the as-received Cu foils (Supplementary Fig. 2b). A facile protocol involving submerging Cu foil in a dilute nitric acid solution was developed46,47. Dissolution of the Cu surface in the nitric acid effectively detached the particles from the surface of the Cu foil. An underlying challenge here is the tradeoff between removing particles and pittings of Cu foil as a function of treatment time and acid concentration. A treatment time of 10 min and an acid concentration of 4 wt% were found to be optimal. Scanning electron microscopy (SEM)-based mapping of a relatively large area (~0.2 mm2; Supplementary Figs. 3 and 4) of graphene was used to analyze the particle density on the surface (Fig. 1e–h). Acid treatment significantly reduced the density of particles. The population of particles smaller than 1 µm was reduced by 70%. Importantly, particles larger than 1 µm were eliminated. This is crucial for reducing cracks during graphene transfer because the macroscopic particles lead to pinhole defects during transfer. The acid treatment also smoothened grooves and scratches on the Cu foil. Optical microscopy images (Fig. 1i–l) and profilometry mapping analysis (Fig. 1m,n and Supplementary Fig. 5) demonstrate that graphene produced on acid-treated Cu foil has a significantly lower root-mean-square surface roughness (150 nm) compared with that on as-received Cu foil (261 nm). Therefore, the two interventions, namely acid pretreatment and using the alumina tube in the CVD reactor, effectively reduced the surface contamination of graphene.

As-synthesized graphene was characterized by Raman spectroscopy. Figure 1o shows a spectrum where the two characteristic peaks of graphene, G and 2D, could be observed at 1,589 and 2,678 cm−1, respectively. The D peak at ~1,350 cm−1, attributed to lattice defects, was negligible, confirming that high-quality graphene can be synthesized on a low-cost Cu foil. A 2D/G peak intensity ratio of ~2.6 confirms that the graphene was a single layer. High-resolution aberration-corrected transmission electron microscopy (AC-HRTEM) further confirmed that graphene was a single layer (Fig. 1p). Raman mappings (48 × 80 µm2 area) of the peak intensity ratio for D/G (ID/IG: 0.05 ± 0.02; Fig. 1q) and 2D/G (I2D/IG: 2.15 ± 0.35; Fig. 1r) confirmed uniformity at this length scale. Characterization of graphene at several locations along the 55-cm-long synthesis zone in the CVD reactor confirmed macroscopic uniformity (Supplementary Fig. 6).

Scalable incorporation of CO2-selective pores in graphene

Oxidation of graphene in an O3 flow leads to the formation of energy-minimizing O clusters on graphene48,49. These clusters yield a pore at their core upon gasification50. Recently, a simplified process of decoupling pore nucleation and expansion involving exposing graphene to O3 followed by heat treatment or light exposure was reported9,50,51. However, this approach has been limited to preparing small coupons. Implementing this concept for large-area PG requires developing a large-area reactor with a uniform reaction zone and the development of a dedicated protocol for incorporating uniform pores.

A 12-cm-diameter tubular reactor was implemented to fabricate a large-area PG (Fig. 2a; experimental setup shown in Supplementary Fig. 7). This reactor could house an 11-cm-wide and 55-cm-long graphene coupon. The reactor was connected to an O3 generator, yielding 8% O3 in O2. An evacuation and gas delivery system was designed to rapidly exchange the gas environment (argon (Ar), hydrogen (H2) and O3) and to attain control over the reaction time. Several critical steps were identified and implemented in sequence. These are the removal of atmospheric contaminations on graphene, oxidation of graphene, gasification for pore generation and, finally, reduction of CuO formed during the O3 exposure step. The corresponding conditions are detailed in Fig. 2b.

Fig. 2: Scaled-up reactor for pore incorporation in graphene.

a, A schematic illustration of the ozone functionalization setup, where T indicates the sample temperature. b, Oxidation schemes with corresponding temperature profiles used in the process. c,d, Photos of graphene resting on Cu before (c) and after (d) oxidation, where the color change of oxidized Cu can be visualized. e, Raman spectra of graphene under various temperatures for oxidation. The spectra have been normalized to the G-band intensity. f, Raman mapping of D/G peak intensity ratio of graphene oxidized at 90 °C across an area of 20 × 40 µm2.

Source data

Atmospheric contaminations deposit on the graphene surface when they are stored. If not removed, they hinder the oxidation of graphene. More importantly, depending on the coverage of contaminations, they make it challenging to achieve reproducible results. We found this to be an extremely critical step. Therefore, graphene was cleaned by heating under a reducing atmosphere of H2/Ar mixture at 600 °C. Although a lower temperature (500 °C) was also effective, improved reproducibility was observed when heating to at least 600 °C. Storage of CVD graphene also tends to generate CuO on the Cu foil, which makes the surface rough52. Heating under a reduced atmosphere helped to smoothen the surface (Supplementary Fig. 8). This step was also crucial to obtain a smooth Cu/graphene surface, which was important for successful membrane fabrication. It should be noted that this step can be avoided when pore formation is implemented immediately after the synthesis of graphene.

For oxidation, the reactor temperature was lowered to the oxidation temperature, and the H2/Ar atmosphere was exchanged with O3. Samples were exposed to O3 for a given period of time, after which they were cooled down to room temperature. O clusters are formed on graphene during this step. The extent of oxidation was increased by increasing the reaction time, temperature or O3 velocity (the latter is discussed in ‘Reaction kinetics versus mass transfer in oxidation reactor’ section). To incorporate pores in the O cluster, the oxidation with O3 was stopped, and the reactor was heated to 150 °C to facilitate the gasification of the clusters.

We observed that the oxidation step also resulted in the oxidation of Cu foil, which could be easily identified by a change in color to an intense red-orange (Fig. 2c,d and Supplementary Fig. 9). This increased the foil’s surface roughness, making graphene transfer challenging. Therefore, a reduction step was implemented at 600 °C.

Pore incorporation in PG is indicated by a monotonic increase in ID/IG to 1.38 ± 0.08 when the oxidation temperature was increased from 40 °C to 90 °C (Fig. 2e). The intensity of the 2D peak decreased as a function of oxidation consistent with the formation of high-density O clusters, which makes graphene nanocrystalline. As oxidation progressed, an obvious D′ peak appeared on the shoulder of the G peak with a D′/G peak ratio higher than 0.8, which confirmed the formation of carbon vacancies in graphene. This was also confirmed by gas permeation studies (see ‘Crack-free transfer of large-area graphene to porous support’ section).

The Ramam mapping (20 × 40 µm2) of ID/IG from the oxidized sample shows that oxidation was uniformly implemented at this length scale (Fig. 2f). A systematic study of the mass transfer led to uniform oxidation in the entire oxidation zone (see ‘Reaction kinetics versus mass transfer in oxidation reactor’ section). Overall, the home-built reactor allowed the oxidation of large graphene coupons (500 cm2) to form large-area PG.

Crack-free transfer of large-area graphene to porous support

The MRF approach has been reported to address the crack formation in graphene during its transfer (Supplementary Fig. 10). Although centimeter-scale membranes have been reported, the success rate has been low (Supplementary Note 1) because of the stress generated in the film during wet-chemical etching of the film where the film is floated. To address this issue, we developed a facile transfer strategy involving a membrane module architecture (Fig. 3a,b). This involved coating an MRF (PTMSP) on PG with a target thickness close to 1 µm (Fig. 3c). The resulting Cu–PG–MRF was placed on a porous membrane support (polyethersulfone (PES), ~0.2-µm pores; Fig. 3d) resting on a macroporous stainless-steel (SS; Fig. 3e) mesh. Step-by-step assembly of the module with a stacking order of Cu–PG–MRF–PES–SS mesh is illustrated in Fig. 3f. The module was sealed by two rubber gaskets (Fig. 3b) and was compressed by two cover plates (Fig. 3f(i,ii)).

Fig. 3: Crack-free direct transfer of graphene inside the membrane module.

a,b, A schematic illustration of the graphene transfer strategy (a) and the architecture of the membrane module (b). c, An optical microscope image of the MRF transferred on a Si/SiO2 wafer. Inset: the film thickness characterization. d,e, SEM images of commercial PES support (d) and SS mesh (e). Insets: pictures of the two supports. f, Pictures of stacked membrane assembly hosting Cu–PG–MRF–PES–SS mesh (i and ii), etching setup for Cu (iii) and as-prepared graphene membrane module after etching Cu foil (iv). g–i, Optical (g) and SEM (with high (h) and low (i) magnification) images of graphene surface after removal of Cu. j,k, A ~8 × 12 cm2 graphene coupon (j) is cut into twenty-four 2 × 2 cm2 small coupons (k). l, All 24 coupons in j lead to successful 1-cm-scale membranes. Half of the membrane modules were assembled with transparent cover plates to reveal the sealing.

The stacking order, Cu–PG–MRF–PES–SS mesh, exposes Cu foil on one side of the module, allowing one to etch and remove Cu directly from the assembled module. This effectively eliminated the need to float graphene. After sealing the module, the exposed Cu was placed in contact with a cell hosting a Cu etchant (1 M FeCl3; Fig. 3f(iii)). During etching, PG reinforced by MRF was secured in the module. The thickness of MRF film played a crucial role. It was optimized as a function of the concentration of the PTMSP solution in toluene and spin-coating speed (Supplementary Fig. 11). A 1.2-µm-thick MRF film markedly improved the transfer of graphene. This is mainly because it allows residual SiOx contaminants (<1 µm in size) to be covered on graphene.

Optical microscope (Fig. 3g) and SEM (Fig. 3h,i) images of the graphene surface exposed after removing Cu reveal the absence of any visible cracks. The white particles observed in Fig. 3h are a combination of Cu foil impurities and residues from the etching of Cu foil. They have a negligible effect on the membrane performance (Supplementary Fig. 12). The reproducibility of this transfer strategy was probed by cutting a ~8 × 12 cm2 graphene coupon (Fig. 3j) into 24 pieces of 2 × 2 cm2 coupons (Fig. 3k) and fabricating membranes from each coupon (Fig. 3l). An unsuccessful transfer typically leaves yellow residues on the white PES support because the Cu etchant (FeCl3) penetrates the broken graphene (Supplementary Fig. 13). This is a useful and quick method to check the success of the transfer. We did not observe the yellow stain in these samples, indicating a successful transfer. Selective gas permeation from all membranes confirmed that the success rate of the gas transfer was 100% (Supplementary Table 2). The variation in CO2 permeance in these samples is attributed to nonoptimized O3 velocity for these samples (see ‘Reaction kinetics versus mass transfer in oxidation reactor’ section).

The simple design of this module allowed the upscaling of the membrane element. However, the circular disk design of the above module limits the ability to achieve the practical cross-flow configuration. Therefore, larger decimeter-scale modules were designed by including cross-flow permeation channels (Fig. 4). The membrane stacking order and Cu etching strategy were identical (Fig. 4a). The module consisted of a symmetrical body frame and two identical cover plates to pack two 5-cm2-sized membrane elements in a single module to increase the packing density. A cross-flow channel was created using slits on the side of the module (Fig. 4b). Cu foil in the assembled module could be removed by flowing the etchant through the cross-flow slits (Fig. 4c). This exposed graphene and generated a feed channel for gas permeation experiments. The cover plate on both sides had a central opening serving as a permeate window. This module could be further scaled into a larger one capable of hosting two 5 × 10 cm2 membrane elements (Fig. 4d). Detailed pictures of a step-by-step assembly are shown in Supplementary Figs. 14 and 15. No yellow stain was observed in the transfer to this large module, which indicated a successful transfer.

Fig. 4: Graphene membranes prepared in large-area cross-flow modules.

a,b, A three-dimensional model of the upgraded membrane module architecture (a), and the corresponding cross-section view showing the cross-flow slits (b). c,d, Photos of successfully prepared 1 × 5 cm2 (c) and 5 × 10 cm2 (d) graphene membranes.

The integrity of the graphene membranes transferred directly in the module was examined by gas permeation studies (setup shown in Supplementary Figs. 14 and 15). As-synthesized CVD graphene, without the oxidation step, yielded a low CO2 permeance of 29 gas permeation units (GPU; 1 GPU = 3.35 × 10−10 mol m−2 s−1 Pa−1) with a CO2/N2 ideal selectivity of 15.6, consistent with the literature on intrinsic vacancy defects in graphene, confirming that cracks were avoided during transfer (Supplementary Fig. 16 and Supplementary Table 3). Successful crack-free transfer was also obtained from as-synthesized graphene transferred to the 5 cm2 and 50 cm2 cross-flow modules (Supplementary Table 3). This establishes the robustness of this crack-free transfer approach in preparing large-area PG membranes.

Reaction kinetics versus mass transfer in oxidation reactor

The oxidation of graphene was studied as a function of reaction conditions (temperature and oxidation time) and mass transfer (ozone velocity). To understand the effect of reaction conditions on graphene porosity, the gas transport resistance model53 was used to extract permeance from the graphene layer (Supplementary Note 2 and Supplementary Tables 2, 4 and 5). For PG membranes, the increase in porosity due to oxidation was indicated by an obvious increase in gas permeance compared with the membranes based on as-synthesized graphene (Fig. 5a, colored data points).

Fig. 5: Optimization of oxidation kinetics and O3 mass transfer.

a, Gas permeation results of as-synthesized graphene (black) and PG (colored) membranes at 1 cm2 and 50 cm2 scale. The ozone oxidation was optimized by different reaction routes: temperature, processing time and mass transport routes. The permeance of PG is extracted from the membrane using the resistance model. The error bar refers to the standard deviation in CO2/N2 ideal selectivity and CO2 permeance across three (as-synthesized graphene and PG under 85 °C, 1 h, slow O3) and two (rest of the conditions) membranes, respectively. The center of the error bar represents the average selectivity and permeance calculated from graphene membranes. b, COMSOL CFD simulation results from the unmodified reactor: extracted gas velocity 1 mm above the substrate in the middle of the reactor with a sample size of 6 × 16 cm2. c, A picture of the modified ozone oxidation reactor hosting a quartz semi-cylindrical block (12 cm in diameter). Inset: is the side view of the reactor with the quartz block. d, COMSOL CFD simulation results from the modified reactor at the same sample position in b. e, AC-HRTEM images of pores generated under an oxidation condition using fast ozone velocity. f, The density of CO2-permeable pores as a function of the ozone velocity. The error bar refers to the standard deviation in permeable pore density across five AC-HRTEM images. The center of the error bar represents the average pore density calculated from the AC-HRTEM image. g, The size distribution of CO2-permeable pores under fast ozone velocity obtained from the AC-HRTEM images.

Source data

A monotonic increase in CO2 permeance as a function of reaction temperature for a fixed reaction time of 1 h could be observed (Fig. 5a). An average increase in permeance from 1,091 GPU to 2,444 GPU was observed when the temperature increased from 85 °C to 90 °C. The average CO2/N2 selectivity increased from 20.2 to 22.8. This favorable increase in CO2 permeance, as well as CO2/N2 selectivity, can be attributed to a higher density of O clusters achieved at an elevated reaction temperature, given that cluster nucleation follows an energy barrier50 and is consistent with the literature on millimeter-scale PG membranes50,51. Increasing the epoxidation time to 3 h at 85 °C led to a further increase in CO2 permeance of 2,850 GPU with a CO2/N2 selectivity of 19.3.

Previous studies on the oxidation of graphene have focused on reaction kinetics. Mass transfer limitation in the O3-led oxidation reaction is important to consider because of the formation of an O2 boundary layer next to graphene. The O3 generator used in this study produces a dilute composition of O3 (8% in O2). Upon oxidation, a molecule of O2 is produced per molecule of O3, which further limits the concentration of O3 near the graphene surface48. Computational fluid dynamics (CFD) simulations were performed using COMSOL Multiphysics to understand the resulting mass transfer. Supplementary Fig. 17a shows the simulated iso-surface plot of the gas velocity with an inlet flo