Bipolar membrane electrolyzers enable high single-pass CO2 electroreduction to multicarbon products

In alkaline and neutral MEA CO2 electrolyzers, CO2 rapidly converts to (bi)carbonate, imposing a significant energy penalty arising from separating CO2 from the anode gas outlets. Here we report a CO2 electrolyzer uses a bipolar membrane (BPM) to convert (bi)carbonate back to CO2, preventing crossover; and that surpasses the single-pass utilization (SPU) limit (25% for multi-carbon products, C2+) suffered by previous neutral-media electrolyzers. We employ a stationary unbuffered catholyte layer between BPM and cathode to promote C2+ products while ensuring that (bi)carbonate is converted back, in situ, to CO2 near the cathode. We develop a model that enables the design of the catholyte layer, finding that limiting the diffusion path length of reverted CO2 to ~10 μm balances the CO2 diffusion flux with the regeneration rate. We report a single-pass CO2 utilization of 78%, which lowers the energy associated with downstream separation of CO2 by 10× compared with past systems.

C O 2 RR for C 2+ production requires the simultaneous achievement of high production rate and high energy efficiency 1,2 Table 1) and membrane-electrode assemblies (MEAs, B in Table 1) have reached industrially relevant levels (ethylene partial current density > 100 mA cm −2 ) 3 ; however, the energy penalty associated with low single-pass CO 2 utilization (SPU: the fraction of the CO 2 converted to the total input CO 2 ) has yet to be reduced to practical levels (SPU > 40%) 4 . Carbonate formation and crossover in typical CO 2 RR electrolyzers limit the SPU to ≤25% for C 2+ (details in SI1), imposing energy penalties of 280-480 GJ in alkaline-media, and of 80-130 GJ in neutral-media, for the production of each ton of ethylene 5,6 . Analysis of the CO 2 and carbonate mass balance in neutral-media electrolyzers (SI1) indicates that achieving high SPU requires that (bi)carbonate not cross the membrane to the anode, and that (bi) carbonate formed at the cathode must revert to CO 2 and remain available to participate in CO 2 RR by returning to the cathode. Bipolar membranes (BPMs) have been used to block CO 2 crossover and convert (bi)carbonate back to CO 2 7-9 . In CO 2 RR electrolyzers, commercial BPMs inhibit CO 2 loss 10 , but the acidic cation-exchange layer (CEL) degrades the cathode's CO 2 RR selectivity 10 .

. The current densities in flow cells (A in
To address cathode acidification, one may use a buffering catholyte (e.g., KHCO 3 ) between CEL and cathode 9,11,12 ; this approach ( Fig. 1a) provides a pH at the CEL surface of~3, and keeps the cathodic local pH >12. However, in a buffering catholyte, (bi)carbonate reverts to CO 2 near the surface of the CEL (solid black line in Fig. 1a), slowing mass transfer of the reverted CO 2 and reducing reactant availability for CO 2 RR. This leads to the best SPUs reported of~15% in prior BPM-based electrolyzers in C 2+ electroproduction (SI3 and SI5). When the catholyte is flowing, the SPU is even lower,~6% according to previous reports, because flowing catholyte removes reverted CO 2 11 . Prior BPM-based electrolyzers (Table 1) have, as a result, not exceeded the 25% SPU limitation for C 2+ electroproduction (SI3).
We show that a stationary, non-buffering catholyte with welldesigned catholyte layer thickness (Fig. 1b) addresses these masstransfer limitations while simultaneously providing the needed high local cathode pH (Fig. 1a). BPM-generated protons migrate farther in non-buffering catholyte than in buffering catholyte, since the latter consumes protons at a higher rate. Here we show further quantitative analysis of the design requirements to achieve this target.

Results
Finite-element numerical simulations of the stationary catholyte (SC)-layer. The composition and thickness of the catholyte layer influence the local pH, the efficiency of CO 2 regeneration and, thereby, the overall cell performance. We applied a onedimensional multiphysics model in COMSOL to investigate the catholyte layer in BPM-based CO 2 RR electrolyzers.
The CO 2 reactant is provided by two sources: the inlet CO 2 flow (gas) and the regenerated CO 2 (dissolved form, aq.) in the catholyte. To achieve high SPU, it is necessary to restrict the gaseous CO 2 feed 13,14 . Under a restricted gaseous CO 2 availability, the cathode CO 2 supply relies more on regeneration (SI1): in an ideal case with 100% SPU and 100% C 2+ selectivity, regeneration contributes 75% of the consumed CO 2 . Thus, the mass transport of regenerated CO 2 is most critical, and that transport is governed by catholyte composition and thickness.
At steady-state, electrolysis creates a pH gradient through the catholyte layer: the pH is high near the cathode and low near the CEL. The protons and (bi)carbonate ions recombine in the catholyte, forming CO 2 (aq.) that diffuses, in response to a concentration gradient, to the Cu catalyst.
Simulations resolve the local cathode environment as a function of dimensions, electrolyte, and running conditions (Figs. 1a, c,  and SI5). We selected 250, 125, 65, and 16 μm as the modelled thicknesses to correspond to commercially available materials. Use of a buffering catholyte (e.g., KHCO 3 ) leads to a thick CO 2 (aq.) diffusion layer close to the catholyte thickness, since the CO 2 (aq.) is generated near the CEL surface, as shown in Fig. 1a (details in SI5). This effect reduces the CO 2 (aq.) mass-transfer efficiency. The experimental results presented and discussed in SI6 show similar trends to the simulation results.
In contrast, Fig. 1a, c illustrate that with a non-buffering catholyte layer (e.g., 0.5 M K 2 SO 4 ) with thicknesses of 250, 125, and 65 μm, the local pH values near the cathode are greater than 11, which is sufficient to promote selectivity towards CO 2 RR over HER 13 . Reducing the SC-layer thickness to 16 μm results in a cathode pH of 8.7, implying a lower selectivity toward CO 2 RR. Figure 1c, d shows the simulated concentration profiles of CO 2 (aq.) in the non-buffering SC-layer. At steady-state, the CO 2 (aq.) is continuously supplied to the cathode to participate in CO 2 RR, forming a concentration gradient (the boundary was defined here as the position where CO 2 concentration is 1% lower than the saturated concentration) to the cathode surface. Prior studies have termed the zone between the cathode and this boundary the diffusion layer 15 . The thickness of the diffusion layer controls the efficiency of CO 2 (aq.) mass transport 15 . According to the simulations, the thicknesses of the diffusion layers are 75, 35, 12, and 5 μm for the catholyte layers with the thicknesses of 250, 125, 65, and 16 μm, respectively (marked in Fig. 1d). For reference, the CO 2 (aq.) diffusion layer thickness in H-cells (all CO 2 supplied in dissolved form) is typically 40-100 μm, and this does not support current densities exceeding 100 mA cm −2 . 16 We expect that diffusion layers <40 μm, and a corresponding catholyte thickness <150 μm, are required for sufficient mass transport in a nonbuffering catholyte. To achieve similar mass transport in a buffering catholyte, the total thickness could not exceed 12 μm, and the cathodic pH would not be sufficiently alkaline for selective CO 2 RR (SI5).
The simulation results suggest the following design principles for the catholyte layer in a BPM-based electrolyzer: the local cathode pH and the diffusion layer thickness of the regenerated CO 2 increase as the catholyte thickness increases; the buffering capacity of the catholyte increases the diffusion layer thickness and reduces transport. Precise control of the thickness of a nonbuffering catholyte should thus offer a route to high SPU, CO 2 RR selectivity, and reaction rate.
System design for high SPU of CO 2 feedstock. Guided by the above analysis, we focused on a stationary catholyte bipolar membrane electrode assembly (SC-BPMEA) electrolyzer and incorporated a judiciously-designed catholyte layer and BPM (Fig. 1a).
The cathode was prepared by spraying Cu nanoparticles onto a hydrophobic carbon gas-diffusion layer for CO 2 RR (Fig. 2a). The anode was IrO 2 supported on Ti felt for the oxygen evolution reaction (OER). A BPM (SEM in Fig. 2b) under reverse bias was employed with the anion exchange layer (AEL) contacting the anode and the CEL contacting the SC-layer (porous support saturated with electrolyte). The cathode was compressed onto the porous layer, and the anode and cathode flow-field plates sandwiched the system.
The BPM employed in this work sandwiched TiO 2 nanoparticles as the water dissociation catalyst 17 . This custom BPM can lower the cell voltage by~1 V compared with commercial BPMs (e.g., Fumasep, details in SI2). The full cell voltage of such custom BPM-based electrolyzers is close to that of anion exchange membrane (AEM)-based systems.
Measurements of the CO 2 /O 2 ratio in the anode gas stream show that the SC-BPMEA effectively prevents CO 2 crossover, as required for high SPU (SI1) 8,11 . In agreement with the previous studies 8, 11 , the AEM-based MEA (AEMEA) showed an anode CO 2 /O 2 ratio of~2 for current densities ranging from 100 to  300 mA cm −2 (Fig. 2c). In conventional AEMEAs, the anionic charge carriers are CO 3 2− , and thus suffer the loss of one molecule of CO 2 for every two electrons transferred. The anode CO 2 /O 2 ratio in the SC-BPMEA (0.06 at 200 mA cm −2 ) is one order of magnitude lower. Control experiments confirm that the CO 2 detected in the anode is not due to acidification of anolyte (using 0.1 M K 2 SO 4 instead of 0.1 M KHCO 3 resulted in a similar CO 2 /O 2 ratio, Supplementary Fig. 3). The anode CO 2 /O 2 ratio decreases as the operating current density increases, an effect we ascribe to an increased flux of protons toward the cathode. This flux decreases the pH at the CEL surface and reduces the diffusion of CO 2 and HCO 3 − /CO 3 − in the CEL 18,19 .
Impact of the thickness of the SC-layer on CO 2 RR. As predicted from simulations, the thickness of the stationary catholyte has a major impact on cell voltage. The cell voltage of the SC-BPMEA decreases as the thickness of the SC-layer decreases ( The simulations (Fig. 1d) indicate that the thicker SC-layer results in longer transport distances for dissolved CO 2 . The CO 2 regeneration rate inside the SC-layer also depends on the current density, and for thicker SC-layers (e.g., >125 μm), CO 2 bubbles are more prone to form near the CEL. These bubbles obstruct ion migration, increasing the ohmic resistance of the SC-BPMEA. Electrochemical impedance spectroscopy measurements (Supplementary Fig. 13d, e) also support this finding. An applied current of 200 mA cm −2 resulted in an insignificant change to the highfrequency resistance (HFR) of the SC-BPMEA with a 65 μm-thick SC-layer; while, in contrast, the HFR of the SC-BPMEA with a 125 μm-thick SC-layer increased by 120% after applying 200 mA cm −2 for 20 min, leading to a cell voltage 0.6 V higher than for the 65 μm SC-layer.
The cell voltage of the SC-BPMEA with a 65 μm SC-layer operating at 200 mA cm −2 is 3.8 V, comparable to the AEMbased neutral-media MEAs operating at similar conditions (difference <±0.05 V) [20][21][22] . This result demonstrates that the cell voltage of a BPM-based CO 2 RR electrolyzer can be as low as that of an AEM-based electrolyzer with a current density of up to 200 mA cm −2 , while suppressing unwanted crossover and providing high SPU. Figure 3b shows the breakdown of the 3.8 V cell voltage, determined using methods reported previously 13,21 (SI7). The factors making up the cell voltage include the thermodynamic potential, cathode overpotential, anode overpotential, ohmic loss, and Nernstian/concentration overpotential (i.e., due to pH gradient) 21 . The sum of these factors accounts for most of the cell voltage, suggesting that the water dissociation overpotential at the AEL/CEL interface of the BPM is small in the SC-BPMEA at 200 mA cm −2 , in agreement with previous reports employing BPMs fabricated in this way (SI2 and ref. 17 ).
The thickness of the SC-layer also affects selectivity towards CO 2 RR. With thicknesses of 65, 125, and 250 μm, the H 2 Faradaic efficiencies (FEs) are consistent (~20% at 200 mA cm −2 , Fig. 3c-e), confirming that high local pH conditions are maintained the cathode in these cases (Fig. 1c). However, reducing the thickness to 16 μm increases the H 2 FE to 88% at 200 mA cm −2 ( Supplementary Fig. 14), consistent with a cathodic pH that is reduced due to fast proton transport through a thin Assessment of SPU in SC-BPMEA. By suppressing the crossover of CO 2 (e.g., <0.5% of total CO 2 input at 200 mA cm −2 , Figs. 2c and 4e), the SC-BPMEA surpasses the SPU of conventional CO 2to-C 2+ electrolyzers, in which carbonate is the dominant charge carrier. Measuring the CO 2 SPUs with a restricted CO 2 flow rate is a direct approach to determining the upper bound of SPU in the CO 2 RR electrolyzers 13 . As the inlet CO 2 flow rate decreased, the C 2+ FE of the SC-BPMEA at 200 mA cm −2 decreased, accompanied by an increase in the H 2 FE (Fig. 4a-c). With SC-layer thicknesses of 65 μm (Fig. 4c), as the input CO 2 flow rate decreases from 1.17 to 0.58 and 0.29 sccm cm −2 , the C 2+ FE decreases from 49 to 48% and 34%, while the H 2 FE increases from 23 to 31% and 64%. This shift is consistent with a CO 2 mass transport limitation 13,20,21 . The exploration of the CO 2 RR performance and energy intensity of SC-BPMEA with restricted reactant availability. All the measurements were conducted at 35°C and 200 mA cm −2 , and the data were collected after 2 h of continuous operation. a-c The FE distributions and the CO 2 requirements (total CO 2 converted to products) of the SC-BPMEAs with different catholyte thickness and input CO 2 flow rates (sccm normalized by electrode area). C 1 refers to CO, formate, and methane. C 2+ refers to ethylene, ethanol, acetate, and n-propanol. d The total CO 2 single-pass utilization (the CO 2 -to-ethylene single-pass conversion see Supplementary Fig. 13a) for the SC-BPMEAs with different catholyte thickness and input CO 2 flow rates. e Carbon balance in SC-BPMEA with 65 μm 0.5 M K 2 SO 4 at different input CO 2 flow rates. See Supplementary Fig. 13b for plots on a logarithmic scale. f The comparison of the energy cost distributions among state-of-art CO 2 -to-ethylene electrolyzers. The SC-BPMEA case refers to the input CO 2 flow rate of 1.17 sccm cm −2 . The error bars represent the standard deviation of three measurements.
The stationary catholyte thickness affects the SPU of the SC-BPMEA. The SPU gradually increases up to 21, 61, and 78% for the SC-BPMEAs with SC-layer thicknesses of 250, 125, and 65 μm, respectively (Fig. 4d). These results demonstrate that high CO 2 conversion efficiencies are possible using SC-BPMEAs with SC-layer thicknesses of 125 and 65 μm.
For a given CO 2 flow rate, a thicker SC-layer produces a lower SPU (Fig. 4d). In the SC-BPMEA, reactant CO 2 is available from the inlet gas stream and regeneration in the SC-layer. With unrestricted CO 2 supply (Fig. 3c-e), the H 2 FEs are similar for different stationary cathode layer thicknesses, indicating that both the CO 2 availability and local pH are unaffected by catholyte thickness under excess supply conditions. The simulations suggest that the thicker SC-layer results in a lower dissolved CO 2 flux to the cathode due to the smaller concentration gradient (Fig. 1d). Compared to the SC-BPMEAs with thinner SC-layers, CO 2 availability with thicker SC-layers decreases more significantly with reducing CO 2 flow rate, leading to a more dramatic increase in H 2 FE (Fig. 4a-c).
The experimental trends are generally consistent with those of the simulations. The SC-BPMEA with a dissolved CO 2 diffusion layer thicker than 75 μm (representing a 250 μm SC-layer) fails to surpass the SPU limit because of insufficient mass transfer. In contrast, a 65 μm SC-layer facilitates efficient mass transport of the regenerated CO 2 (diffusion layer thickness of 12 μm) and simultaneously promotes high local cathode pH.
As demonstrated in SI10, SC-BPMEAs using acidic and alkaline electrolytes achieve carbon efficiencies comparable to those using neutral electrolytes. The compatibility of SC-BPMEAs with a range of electrolytes offers flexibility in the selection of cathode and anode catalysts. In contrast, acidic CO 2 -to-C 2+ electrolyzers have only been demonstrated with precious metal anodes 13,14 .
As shown in Supplementary Fig. 18, the SC-BPMEA shows > 50-h stability operating at 200 mA cm −2 with limited CO 2 availability (CO 2 input flow rate of 1.42 sccm cm −2 ). This operating stability is competitive with that of the neutralelectrolyte-based CO 2 -to-C 2+ electrolyzers 23,24 .
Can a cation-exchange membrane replace the BPM in SC-BPMEA? We attempted to extend the SC-layer strategy in a CEM-based MEA cell (i.e., SC-CEMEA, Fig. 5a) using an acidic anolyte with pH < 2.4, expecting a lower cell voltage than the SC-BPMEA while maintaining high SPU. We found that in the SC-CEMEA, the CO 2 crossover was essentially eliminated. This observation is ascribed to the lower pH near the stationary catholyte layer/CEM interface, as shown in Fig. 5a.
SC-CEMEA shows a lower full cell voltage (Fig. 5b) compared to the SC-BPMEA presented, partly due to the lower resistance of the CEM and the absence of water dissociation overpotential. Meanwhile, it has a reasonable CO 2 RR selectivity over HER (Fig. 5c) due to the cation effect and high local pH induced by the presence K + in the SC-layer (Fig. 5a). However, this design is not amenable to steady-state operation without continuous addition of acid and salt to the anolyte, as the initial pH gradient will be eliminated due to co-ion transport and neutralization. We found the CO 2 RR selectivity decreases over time and approaches 100% H 2 after~3 h.
We also observed that the SC-CEMEA design periodically ejects electrolyte from the cathode flow channel, likely due to poor water balance. On the anode, the OER generates one proton per one electron transfer. The charge carriers across the CEM are primarily H + , although neutral ion pairs will diffuse as well. At the cathode K + makes up the electrochemical double layer at the Cu surface, and the steady-state K + profiles are governed by the electric and chemical-potential gradients that develop under operation, which usually takes tens of seconds 19 . H + migrates to the cathode and combines with OH − (or CO 3 2− /HCO 3 − ), producing water at the cathode. The protons also drag water molecules (~1 per proton) 25 by electro-osmosis. We accordingly calculate the water balance for different cathode products as listed in Table 2. The water generated and transported to the cathode appears to dilute and push out the electrolyte in the stationary catholyte layer, of which the volume is small (ca. 10 μL per cm 2 electrode area). This phenomenon results in flooding of the cathode (as confirmed experimentally) and loss of supporting electrolyte, thus degrading performance. In the BPMEA design, it is likely that the BPM slows co-ion transit across the membrane, compared to the CEM, by the large outward flux of OH − and H + from the water dissociating junction.
Energy assessment of the SC-BPMEA with optimal SC-layer. The energy costs (measured in gigajoules per tonne of the target product, GJ/t) for a CO 2 -to-C 2+ electrolyzer include the electrolysis electrical energy, cathodic stream separation, and anodic stream separation 13 . CO 2 RR performance metrics of importance include cell voltage, target product FE, SPU, and CO 2 crossover 5 . High SPU and high energy efficiency have not been accomplished simultaneously in C 2+ electroproduction. In SC-BPMEAs, a higher SPU reduces the energy required for cathode separation, but the accompanying decrease in the ethylene selectivity (Fig. 4c) elevates the specific energy requirement 5 . We carried out a total energy assessment of the SC-BPMEA and other state-of-art CO 2to-ethylene electrolyzers and summarized the results in Table 3, Fig. 4f, and Supplementary Table 2.
The energy consumption of an alkaline CO 2 RR electrolyzer 2 is listed in Supplementary Table 2. In such systems, CO 2 and OH − react to form carbonate continuously. This carbonate has to be recovered to maintain the CO 2 RR performance of such a system, consuming 5.5 GJ per tonne CO 2 5 . In the alkaline CO 2 RR electrolyzers, ca. 63 tonne of CO 2 transforms to carbonate to produce 1 tonne of ethylene, representing an energy penalty of 350 GJ 5 . This costs at least $1900 per tonne of ethylene, while its market price is $800-1000 per tonne 26 . The alkaline electrolyzers thus do not allow for ethylene electrochemical production to be yet profitable.
In neutral-media CO 2 RR electrolyzers, recovering the CO 2 from the anodic gas stream results in significant energy costs. In the context of highly selective conversion (i.e., CO 2 -to-ethylene with unity selectivity), the recovery process requires an energy input of 52 GJ to produce every tonne of product. In practice, due to non-unity product selectivity, the process is even more prohibitive, i.e., requiring an energy penalty of 80-130 GJ for producing one tonne of ethylene 5 .
As the SPU increases from 4 to 35%, we found a dramatic decrease in energy associated with cathode separation-from 85 to 15 GJ/t ethylene (Table 3), with the ethylene FE reduced by only 2%. Further increasing the SPU beyond 35% does not substantially reduce the energy cost associated with cathodic separation (Supplementary Table 2). This finding agrees with a recent energy analysis that in a (bi)carbonate-free CO 2 -to-C 2+ electrolyzer, improving SPU over 40% offers an insignificant benefit to the downstream separation cost 4 . Pursuing an SPU > 35% decreases ethylene FE by more than 4% when using the SC-BPMEA, and thus the increased input electricity cost exceeds the savings in the cathodic separation (Table 3 and Supplementary  Table 2). Therefore, 35% SPU is the most favourable condition for the present SC-BPMEA.
The energy intensity of producing ethylene in SC-BPMEA is 30% lower than that in conventional neutral-electrolyte-based CO 2 electrolyzers ( Fig. 4f and Table 3). In conventional neutralelectrolyte CO 2 -to-ethylene electrolyzers, the CO 2 crossover (at least 70%) 21 costs 60-90 GJ per ton of ethylene to recover CO 2 from the anodic O 2 stream 5 . Notably, this energy penalty cannot readily be reduced, independent of optimizing catalysts and operating conditions (e.g., input CO 2 flow rates, reaction rates, operating temperature, and pressure) 6 . In contrast, crossover CO 2 in SC-BPMEA is < 0.5% of the total CO 2 input, minimizing the energy cost of anodic separation.
Recently, CO 2 -to-ethylene conversion has been achieved in acidic electrolytes in both flow cell 13 and MEA configurations 14 . These systems enabled CO 2 SPUs exceeding 75% and also mitigated the energy cost associated with anodic separation (Table 3). Owing to the strongly acidic environment, the flow cell enables an ethylene FE of 28% at a full-cell potential of 4.2 V. The acidic MEA used an anion-exchange ionomer coating on the catalyst layer to promote CO 2 RR over HER. The modification of the surface with the anion exchange ionomer resulted in a higher ohmic loss, and thus the cell required potentials of 3.8 and 4.4 V at 100 and 200 mA cm −2 , respectively. These devices thus eliminated the anodic CO 2 /O 2 separation energy but at the penalty of larger cell voltages and/or lower ethylene FEs. In contrast, SC-BPMEA shows a cell voltage of 3.8 V at 200 mA cm −2 with an ethylene FE of 42%-voltages and selectivities comparable to the best conventional neutral-electrolyte CO 2 -to-ethylene MEAs 21 . Compared to acidic systems, the energy intensity of the SC-BPMEA is 36% and 12% lower than acidic flow cell and acidic MEA, respectively ( Fig. 4f and Table 3).

Discussion
We demonstrate a BPM-based CO 2 -to-C 2+ MEA, with a judiciously-designed SC-layer between catalyst and BPM, that overcomes the (bi)carbonate-formation reactant loss issue without compromising performance. The composition and thickness  of the SC-layer determine the CO 2 RR performance and SPU via a strong influence on the local pH and the chemistry and transport of CO 2. The buffering capacity and the thickness of the SC-layer determine the efficiency of the regeneration, the transport, and the availability of reactant CO 2 . These effects were predicted in simulations and supported by experiments. The SC-BPMEA design largely eliminates the energy penalty associated with the CO 2 loss in electrochemical CO 2 reduction. The performance of the SC-BPMEA might be further improved using, for example, ionic liquid or other organic salts as the catholyte, and by optimizing the porosity, structure, and hydrophobicity of the porous support layers. The CO 2 RR performance of the SC-BPMEA might be improved with new cathodic catalysts, optimizing the loading and processing of the catalyst layer, and by implementing BPMs with further-lowered water-dissociation voltage loss. Broadly, the SC-BPMEA is a useful platform for evaluating CO 2 RR catalysts operating with high CO 2 utilization. The strategy and findings presented here are also relevant to the electrochemical systems such as nitrate reduction and (bi)carbonate reduction, where controlling dissimilar microenvironments near each electrode is useful, and the exchange/transport of species (other than OH − or H + ) between cathode and anode is problematic.
Electrode preparation. For the CO 2 RR, we prepared the gas diffusion electrodes (GDEs) by spray-depositing a catalyst ink dispersing 1 mg mL −1 of Cu nanoparticles and 0.25 mg mL −1 of Nafion TM 1100 W in methanol onto a hydrophobic carbon paper. The mass loading of Cu NPs in the GDE was kept at 1.5 mg/cm 2 . The GDEs were dried in the air overnight prior to experiments.
The OER electrode preparation procedure involves: etching the Ti felt in hydrochloric acid at 70°C for 40 min; rinsing the etched Ti felt with DI water; immersing the Ti felt into an Ir(IV) chloride hydrate solution; drying and sintering the Ir-loaded Ti felt. The loading, drying, and sintering steps were repeated until a final Ir loading of 1.5 mg cm −2 was achieved.
Assembly of the stationary catholyte membrane electrode assembly (SC-BPMEA). The MEA set (5 cm 2 ) was purchased from Dioxide Materials. A cathode was cut into a 2.1 cm × 2.1 cm piece and placed onto the MEA cathode plate with a flow window with a dimension of 2.2 cm × 2.2 cm. The four edges of the cathode were sealed by Kapton tape, which also made the flow window fully covered. The exposed cathode area was measured every time before the electrochemical tests, in the range of 3.1 to 4.2 cm 2 . Onto the cathode, a porous support layer (2 cm × 2 cm with various thicknesses, 250 μm was stacking two 125 μm-thick PVDF) saturated with desirable electrolyte (sonicated in electrolyte for 15 min to degas) was carefully placed. This porous support layer serves as the 'stationary catholyte layer (SClayer).' The considerations of membrane selection can be found in SI2 and SI4 of the Supplementary Information. When using the custom BPM, a TiO 2 -coated Nafion membrane was placed onto the SC-layer with the TiO 2 layer facing up, then covered by a Piperion (5 cm × 5 cm) membrane. When using Fumasep BPM, the membrane was placed with its cation-exchange layer (CEL) facing the cathode side. An IrO 2 loaded Ti felt (2 cm × 2 cm) was placed onto the anion-exchange layer (AEL) of the BPM.
Scanning electron microscopy (SEM). Images of cathode and custom BPM were captured by an FEI Quanta FEG 250 environmental SEM.
Electrochemical measurements. Throughout all experiments, CO 2 flowed to the cathode side at 10 sccm cm −2 unless otherwise specified, while the anode side was fed with neutral 0.1 M KHCO 3 at 10 mL/min by a peristaltic pump unless otherwise specified. The electrochemical measurements were performed with a potentiostat (Autolab PGSTAT204 with 10A booster). The cell voltages reported in this work are not iR corrected. The system was allowed to stabilize at the specific conditions for > 1000 s before recording the results. All the error bars represent standard deviations based on three measurements.
Product analysis. The CO 2 RR gas products, oxygen, and CO 2 were analyzed by injecting the gas samples into a gas chromatograph (Perkin Elmer Clarus 590) coupled with a thermal conductivity detector and a flame ionization detector. The gas chromatograph was equipped with a Molecular Sieve 5A Capillary Column and a packed Carboxen-1000 Column with argon as the carrier gas. The volumetric gas Crossover of CO 2 in the acidic flow cell, acidic MEA, and SC-BPMEA are each lower than 0.5% of input CO 2 . Therefore, we assume the anodic separation energy to be 0. All the energy costs are normalized per ton of ethylene produced. The energy intensities of the SC-BPMEA operating at other CO 2 input flow rates are listed in Table S3 of Supplementary Information.
flow rates in and out of the cell were measured with a bubble column. The FE of a gas product is calculated as follows: Where x i is the volume fraction of the gas product i, V is the outlet gas flow rate in L s −1 , P is atmosphere pressure 101.325 kPa, R is the ideal gas constant 8.314 J mol −1 K −1 , T is the room temperature in K, n i is the number of electrons required to produce one molecule of product F is the Faraday Constant 96485 C mol −1 , and J is the total current in A. The liquid products from the cathode side of the SC-BPMEA were collected using a cold trap cooled to 0°C. The collected liquid was combined with anolyte (some crossover liquid product) for quantifying by the proton nuclear magnetic resonance spectroscopy ( 1 H NMR) on an Agilent DD2 500 spectrometer in D 2 O using water suppression mode and dimethyl sulfoxide (DMSO) as the internal standard. For each plot of liquid product quantification, fresh anolyte was used, and the duration of the collection was 30 min. The FE of a liquid product is calculated as follows: Where m i is the quantity of the liquid product i in mole, t is the duration of product collection (1800 s).
The CO 2 SPU calculation is detailed in SI1 of Supplementary Information.
COMSOL one-dimensional modelling. The electrochemical reaction model was performed by COMSOL Multiphysics version 5.5. This simulation was built upon previous modelling work [27][28][29][30] . The local pH and different species concentrations were simulated for different catholyte thicknesses (16,65,125, and 250 μm). Two different catholytes (K 2 SO 4 and KHCO 3 ) were used in the simulation. All the chemical reactions between species were considered in this one-dimensional modelling. The simulation (Fig. 6) included a 50 μm thick gas diffusion layer (GDL), a 0.1 μm thick Cu cathode catalyst (CL), a catholyte region with various thicknesses indicated above, and a cation exchange layer (CEL) boundary. Constant concentration (Dirichlet) boundary conditions were used. Specifically, a constant concentration 37.8 mM of CO 2 was assumed within the GDL layer, as this region is in direct contact with the input CO 2 flow and thus assumed to be at equilibrium with gas phase CO 2 over this region for the purposes of the simulation. The BPM was interpreted as a boundary with a constant species concentration (1 M H 3 O + at the CEL surface) 18,31 , because it was assumed to generate protons as the dominant ionic charge carrier at a constant rate under constant current density (200 mA cm −2 ).
A user-controlled mesh is employed in the COMSOL simulation. Edge type of mesh is used for GDL, CL, catholytes, respectively. Specifically, the mesh distribution is predefined with an interval of 500 nm for GDL and catholytes, and an interval of 5 nm for CL.
Five different electrode reactions were considered at the cathode catalyst layer in this simulation. Specifically, the hydrogen evolution reaction and CO 2 reduction reactions to CO, CH 4 , C 2 H 4, and C 2 H 5 OH occurred at the cathode catalyst layer. In SC-BPMEA, the catalyst layer is immersed in a catholyte. Thus the simulation considers no gas-phase transport in the catalyst layer. The carbonate equilibrium reactions, corresponding catholyte buffer reactions, and a water dissociation reaction were considered in the catholyte region. The electrochemical reaction rates of the specific products were determined from experimental results. They are calculated based on the same manner as previous work 17 .
The electrochemical reactions at cathode catalyst layer: The heterogenous electrochemical reaction rates are determined by the following equations: Where I i represents the partial current density for CO, CH 4 , C 2 H 4, and C 2 H 5 OH occurred at the cathode catalyst layer, respectively. n i represents the number of electrons transferred per mole reactant. F represents faraday's constant.
I total represents the total current density. The FEs for the specific product is determined by the experimental results shown in Fig. 3e. ε represents the catalyst porosity value. L catalyst represents the cathode catalyst length.
The chemical reactions at the catholyte region and the corresponding forward k f rate constants and reverse k r rate constants taken from the literature 32 (see  Supplementary Table 3).
The Transport of Diluted Species physics model was used. The Nernst-Planck set of equations governed the species diffusion, and they were calculated in the same manner as previous work 13,14 . Migration was ignored for simplicity as the experiments were performed in the concentrated electrolyte. The ion species transport is thus calculated by solving the two equations below.
Where J i is the molar flux, and r i represents the heterogeneous electrode reactions for CO 2 reduction that were modelled at the cathode catalyst layer. R i represents the rates of the homogeneous reactions indicated above. The Millington and Quirk model is used to determine the effective diffusivity, D i . ε p represents porosity coefficient. τ F,i represents tortuosity coefficient.
The porosity value of 0.6 was used for the cathode catalyst and the porosity value of 1 for the catholyte region. The species diffusion coefficients are listed in Supplementary Table 4 [33][34][35][36] .
Henry's law and sets of Sechenov equation are applied to calculate the CO 2 concentration. The concentration of CO 2 in electrolytes depends on temperature and pressure. It is estimated in the same manner as previous work 13,14 . The Sechenov coefficients are listed in Supplementary Table 5 37 .
Energy assessment. We evaluated the energy consumptions for electrolyzer electricity, cathodic separation, and anodic separation in the context of ethylene. We consider the state-of-the-art CO 2 RR systems from the literature, including alkaline flow-cell electrolyzers, neutral MEA electrolyzers, acidic flow-cells, and MEAs. This consideration is based on the performance metrics, including selectivity, productivity, and full-cell voltage-the combination reflects as energy intensity of producing multi-carbon products (i.e., ethylene). The proximity of these performance metrics will help refine the effect of anodic and cathodic separation on the energy requirement for producing ethylene. We summarize the input parameters to the model for all the systems. The energy assessment model, as well as the assumptions, are based on the previous work 5 . Ideally, it will be interesting to use experimental/modelling data corresponding to the exact gas composition from the CO 2 -to-C 2+ device. However, at present, there is a gap in published literature. We, therefore, employed one of the most widely used models 38 (i.e., biogas upgrading) as the best approximation for evaluating the energy cost associated with cathode gas separation. The details of calculations for the carbon regeneration (for alkaline flow cell) and cathodic separation (for all the electrolyzers), can be found in previous work 26 . The anodic separation (for neutral MEA electrolyzer) is modelled based on an alkaline capture solvent 39 . The amount of CO 2 crossover to the anode is calculated for one tonne of ethylene produced. The energy required to separate the CO 2 /O 2 mixture is calculated based on a recent report by Carbon Engineering 40 , in which 5.25 GJ/tonne CO 2 thermal energy and 77 kWh/tonne CO 2 are reported to be required to capture CO 2 and release at 1 bar. This energy consumption is a typical value for the alkaline capture process 39 . For acidic flow-cell and MEA electrolyzers, we assume no energy cost associated with the anodic separation considering no CO 2 availability at the anodic gas stream 13 .

Data availability
All the data generated in this study are provided in the Supplementary Information and in the Source Data file. Source data are provided with this paper.  Fig. 6 The schematic of 1D COMSOL modelling. GDL and CL refer to gas diffusion layer and catalyst layer, respectively. ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-022-31295-3 Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/ licenses/by/4.0/.