Nutrient diffusion-inspired catalysts with self-reinforced concentration gradient for sustainable electroreduction of dilute CO₂

Abstract

The electrocatalysis of flue gas into CO in membrane electrode assembly (MEA) provides a sustainable route for realizing practical CO₂ electrolysis technology but suffers from restricted CO₂ mass transport due to thick gas boundary layer (GBL) and weak concentration gradient. Inspired by nutrient diffusion mechanism in plant, we introduce the concept of self-reinforced CO₂ concentration gradient, which is realized via porous carbon nanosheets (PC) as soil for enriching CO₂ and single-atomic Ni-doped carbon nanotubes (Ni-CNTs) as rhizome for electro-catalyzing CO₂. A combined experimental and simulation study reveals optimal length of Ni-CNTs on PC reduces the GBL thickness and spontaneously enhances CO₂ concentration gradient, synergistically breaking the limitation of CO₂ transport. Consequently, the CO Faradaic efficiency attains >90% with varying CO₂ concentration of 4-15 vol. % CO₂ in MEA. Further through incorporation of an O₂-adsorption packed column before MEA, we realize the stable and selective conversion of O₂-containing flue gas into CO.

Introduction

Electrochemical carbon dioxide reduction reaction (CO₂RR) provides a sustainable technology to convert waste CO₂ into valuable chemicals and fuel using renewable electricity^1,2,3,4. Carbon monoxide (CO), as a product of the two-electron CO₂RR, is suggested to be one of the most economically viable since it is an important carbon building block to synthesize long-chain industrial reagents^5,6. In this respect, many efforts have been devoted to boosting the CO₂RR-to-CO performance, including catalyst design and reactor upgrade^7,8,9,10. In particular, the zero-gap membrane electrode assembly (MEA) electrolyzers based on gas diffusion electrode (GDE) enable the gaseous CO₂ to directly reach the cathode catalyst layer (CL) through the microporous layer (MPL), which can circumvent the mass transport limited by the poor CO₂ solubility and possess low ohmic loss^10,11. Benefiting from these features, industrially relevant current density (>200 mA cm^–2) with impressive CO Faradaic efficiency (>90%) has been attained in gaseous feedstock MEA by using CO-selective catalysts such as silver^3,12 and metal single atoms^{8,13,14,15,16}, pushing the CO₂ electrolysis technology into the realm of industrial applications.

Despite the notable advances, a majority of studies have focused on concentrated CO₂ streams, but the flue gas from combustion of fossil fuels typically contains 3–15% of CO₂ (expressed as volume percentage unless specifically noted) primarily diluted by nitrogen (N₂), meaning that it is necessary to first capture CO₂ before flowing into electrolyzer as feed for CO₂RR¹⁷. The mature technology for the production of pure CO₂ from point sources requires absorption in amine/alkaline solution, followed by a regeneration process involving heating or pressure reduction, which are environmentally and economically unfavorable, sacrificing the viability of CO₂RR technology^18,19,20,21. In this context, it can be envisioned that integration of CO₂ capture sites and CO₂RR active sites into one single catalyst system for direct electrocatalysis of a dilute CO₂ stream is more sustainable since it can avoid the cost caused by CO₂ separation and enrichment. Nevertheless, the direct intake of low-concentration CO₂ as gaseous feed suffers from serious limitations in mass transport due to the heavier CO₂ molecule than N₂, which causes slow mass transport speed based on the well-known Graham’s law of effusion²², thus leading to multiple challenges for low-concentration CO₂RR. First, CO₂ navigates through the cathode channel and approaches the catalyst surface via the gas boundary layer (GBL)²³. At this point, the slow transport rate of CO₂ in the bulk phase, coupled with the strong adhesion force of N₂, jointly leads to a thicker GBL (Supplementary Note 1). This, in turn, results in a lower CO₂ concentration on the gas-liquid-solid surface, which thermodynamically favors the competitive hydrogen evolution reaction (HER) over CO₂RR (Fig. 1a). Second, according to Fick’s Law^24,25,26, a decrease in the feeding concentration of CO₂ weakens the concentration gradient effect within the CL, which directly incurs a weak CO₂ flux to the catalytic sites, resulting in sluggish CO₂RR kinetics and high overpotential (Fig. 1b and Supplementary Note 2).

Fig. 1: Schematic illustration of the challenges in low-concentration CO₂RR from the thermodynamical and dynamical perspectives, as well as our proposed strategy inspired by natural nutrient diffusion from plants to address the issues.

a Dilute CO₂ stream causes an increased thickness of gas boundary layer (δ_GBL), inversely proportional to Reynolds number (Re), thereby thermodynamically favors HER. b Dilute CO₂ stream diminishes the CO₂ concentration gradient (∂c/∂x) effect within catalyst layer (CL) that results in sluggish CO₂RR kinetics. c Schematic of nutrient diffusion-inspired hierarchical electrocatalyst with ultramicropore-induced self-reinforced CO₂ concentration gradient for low-concentration CO₂RR.

To address the challenges in low-concentration CO₂RR, limited attempts have been implemented on enriching CO₂ in the vicinity of CO₂RR active sites via the introduction of amino groups or physical confinement with optimal pore structures^{27,28,29,30,31}. Beyond improving the CO₂ adsorption ability, the gaseous CO₂ transport behavior within the GBL and CL fundamentally determines the low-concentration CO₂RR performance in MEA, which is seriously neglected. Additionally, how the CO₂ capture sites and catalytic sites guide the CO₂ mass transport behavior during the dilute CO₂RR process remains unexplored; however, which is very critically important for CO₂RR³², especially from industrial point sources. For these reasons, very few works on dilute gaseous CO₂RR can realize the high conversion of CO₂ into CO in MEA when CO₂ concentration is below 10%, which cannot fully meet the requirement of CO₂ from the iron and steel industry (CO₂ concentration: 3–5%)³³. Therefore, it is imperative to design an advanced electrocatalyst system that can simultaneously promote CO₂ transport at the GBL and CL for improving dilute CO₂RR performance.

Herein, mimicking natural nutrient diffusion that enables swift nutrient conveyance based on the difference in nutrient concentration between roots in soil and rhizome, we propose a hierarchical structure composed of nickel single atoms anchored on N-doped carbon nanotubes (Ni-CNTs) growing from porous carbon nanosheets (Ni-CNTs/PC), affording a self-reinforced CO₂ concentration gradient effect, which facilitates CO₂ transport to the Ni active sites for low-concentration CO₂RR (Fig. 1c). The PC nanosheets featuring ultramicropores can capture and concentrate CO₂ from CO₂/N₂ mixture, while the Ni single atoms can serve as active sites for CO₂RR. Experimentally, Ni-CNTs/PC was synthesized through chemical vapor deposition (CVD) growth on swelled Ni²⁺-crosslinked aerogel, where the Ni-CNTs length can be simply controlled via the growth time. With optimal length of Ni-CNTs grown on PC NSs, the Ni-CNTs/PC is found to thin the GBL and establish a self-reinforced CO₂ concentration gradient within CL in MEA electrolyzer, thereby significantly enhancing the CO₂ transport ability. Such a nutrient diffusion-mimicking structure generates a CO Faradaic efficiency surpassing 90% with varying CO₂ concentrations from 4 to 15%. Notably, the Ni-CNTs/PC enables the efficient conversion of dilute CO₂RR containing 3% of CO₂ with 87.7% of CO Faradaic efficiency. Further integrating an oxygen-adsorption packed column (OAPC) before the MEA reactor, the Ni-CNTs/PC enables the direct conversion of simulated flue gas containing 3–10% O₂. This work provides a sustainable route for converting flue gas by introducing a self-reinforced concentration gradient, paving the way for the practical implementation of CO₂ electrocatalysis technology.

Results

Material synthesis and characterization

The hierarchical structure of Ni-CNTs/PC was synthesized via CVD growth on the swelled Ni²⁺-crosslinked carboxymethylcellulose sodium (Ni-CMC) aerogel, as illustrated in Fig. 2a. Briefly, nickel chloride (NiCl₂) was first added into the CMC solution for producing a green homogeneous colloidal solution based on the interaction between Ni²⁺ and OH^–/–COO^– groups in CMC (Supplementary Fig. 1). After molding on the frozen copper billet, the as-prepared solid was immersed in cold ethanol bath containing sodium hydroxide (NaOH) to form Ni-CMC hydrogel through H–bond interaction³⁴ (Supplementary Fig. 1), followed by swelling treatment in deionized (DI) water and freeze-drying to obtain Ni-CMC aerogel (Fig. 2b), which presents porous networks comprising vertically arranged CMC layers with pore sizes of ~65 μm (Supplementary Fig. 2). Notably, the abundant porous structures rely on not only the Ni²⁺ crosslinking and NaOH activation, but also the employment of ice template (Supplementary Figs. 3, 4). Afterward, CVD growth was conducted on the well-ordered Ni-CMC aerogel with acetonitrile (CH₃CN) as feedstock at 750 °C, where the Ni metal nanoparticles can catalyze the CH₃CN decomposition into C–N fragments, forming Ni–N bonds and directly incorporating into the carbon lattice (Supplementary Fig. 5)⁶. Meanwhile, the Ni-CMC aerogel is converted into porous carbon nanosheets (PC NSs) during Ni, N co-doped CNTs growth (Supplementary Fig. 5 and Supplementary Table 1), forming Ni-CNTs/PC heterostructures, as evidenced by the scanning electron microscope (SEM) images in Fig. 2c and Supplementary Figs. 6, 7. Notably, the CNTs vertically grow on PC NSs due to a well-known tip-growth mode³⁵, and thus the residual Ni nanoparticles can be easily removed by acid leaching (Supplementary Fig. 8). It is found that the Ni-CMC formation is the key factor for forming the well-defined Ni-CNTs/PC heterostructures (Supplementary Figs. 9, 10). Gram-scale purified product can be collected in one batch synthesis after etching in hydrochloric acid, followed by filtration (Supplementary Fig. 11), meeting the requirement of high mass loading for MEA devices.

Fig. 2: Synthesis and characterization of Ni-CMC aerogel and Ni-CNTs/PC.

a Schematic for the preparation of hierarchical Ni-CNTs/PC. b Schematic for synthesizing Ni-CMC hydrogel through the sol-gel strategy followed by swelling treatment. c Top-view SEM images and d HRTEM image of Ni-CNTs30/PC (the white circle in d refers to the location of the etched Ni particle). e Aberration-corrected HAADF-STEM image of Ni-CNTs. f EDS elemental mappings of C, N, and Ni. g Schematic for the tip-growth mode of Ni-CNTs. h Swelling ratio of Ni-CMC hydrogel and Ni nanoparticles (NPs) size versus swelling time. i The thickness of PC NSs and the length of Ni-CNTs in Ni-CNTs/PC versus growth time. Error bars represent the standard deviation of three independent measurements. Source data are provided as a Source Data file.

Transmittance electron microscope (TEM) images were carried out after 30 min of CVD growth and confirm the vertical growth behavior of Ni-CNTs from the CMC-derived PC NSs (Fig. 2d). Aberration-corrected high-angle annular dark-field scanning transmission electron microscope (HAADF-STEM) was employed to discern the distribution of Ni single atoms across the region of CNTs and PC NSs, respectively. As shown in Fig. 2e and Supplementary Fig. 12, the bright spots corresponding to Ni atoms are uniformly dispersed on the CNTs profile, without any observed clusters or nanoparticles. Conversely, extensive HAADF-STEM images reveal the absence of Ni species on the PC NSs (Supplementary Fig. 13). This contrast can be further supported by the energy-dispersive spectroscopy (EDS) mappings in Fig. 2f and Supplementary Fig. 14, where Ni species only homogeneously distribute on the CNTs skeleton. The observation can be explained by the tip-growth mode of CNTs^36,37, where the Ni nanoparticles detach from the CMC-derived carbon substrate during CNT growth and can dope into the CNTs through Ni–N interaction⁷, as schematically shown in Fig. 2g. Since the Ni seeds are located at the ends of CNTs, they are susceptible to being etched away, leaving rare Ni nanoparticles or clusters on Ni-CNTs.

It is noted that the swelling treatment before free-drying of Ni-CMC hydrogel is very critical for growing CNTs arrays and avoiding the residual of Ni nanoparticles, possibly due to the increased interval of Ni species by water uptake that prevents the heavy Ni aggregation into large nanoparticles in the subsequent high-temperature growth (Fig. 2h, i and Supplementary Figs. 15–17)³⁸. It is found that the swelling ratio reaches equilibrium after 48 h, where the Ni nanoparticles are the smallest with sizes of ~5.5 nm. Accordingly, well-defined CNTs with diameters of ~100 nm are successfully synthesized on PC NSs, which can effectively avoid the Ni aggregation as compared to other swelling times. Moreover, the length and coverage of CNTs can be easily adjusted by controlling the growth time while the thickness of PC NSs remains unchanged (Fig. 2i and Supplementary Fig. 18), which provides rich models to study the effect of CO₂ transport distance and concentration gradient effect on dilute CO₂RR. Specifically, CNTs with growth time of 10, 30, 60, 90, and 180 min, denoted as Ni-CNTs10/PC, Ni-CNTs30/PC, Ni-CNTs60/PC, Ni-CNTs90/PC, and Ni-CNTs180/PC, correspond to the length of ~100, ~500, ~1100, ~1300, and ~1800 nm, respectively. Besides, the inductively coupled plasma mass spectroscopy (ICP-MS) analyses reveal the Ni content in Ni-CNTs10/PC, Ni-CNTs30/PC, Ni-CNTs60/PC, Ni-CNTs90/PC, and Ni-CNTs180/PC was ~0.75, ~2.00, ~2.14, ~2.34, and 2.75 wt% (Supplementary Table 2), respectively, suggesting that the Ni content slightly increases as growth time was over 30 min. All these results demonstrate the successful construction of hierarchical structures composed of Ni, N co-doped CNTs, and PC nanosheets, where the Ni single atoms exclusively distribute on the CNTs and the length of CNTs can be readily adjusted.

The Ni-CNTs/PC samples were further analyzed by X-ray diffraction (XRD), all of which display a distinct peak at ~26° (Fig. 3a), assigned to the plane (002) of graphitic carbon. The (002) diffraction peak becomes sharper as growth time proceeds, ascribed to the increased CNTs coverage. The absence of characteristic Ni species collaborates with the isolated Ni single atoms, consistent with the HAADF-STEM results. X-ray photoelectron spectra (XPS) analyses were performed to reveal the chemical compositions and valence states of Ni-CNTs/PC. As shown in Fig. 3b, the Ni 2p peaks of the four samples are all higher than Ni foil, but lower than NiO, which suggests that the valence state of Ni species in Ni-CNTs/PC is situated between Ni⁰ and Ni²⁺. In addition, the high-resolution N 1s XPS spectra can be deconvoluted into three peaks at binding energies of 398.4, 399.4, and 401.0 eV (Fig. 3c), corresponding to pyridinic N, Ni–N, and graphitic N, respectively^13,39, which confirms the formation of Ni–N bonds via a direct CVD approach. The N configurations and their ratios show an inapparent difference (Supplementary Table 3), which can exclude their effect on CO₂ adsorption and conversion. Besides, pure PC NSs were also prepared without introducing CH₃CN during annealing of the Ni-CMC aerogel, followed by acid etching, which shows negligible Ni and N signals (Supplementary Figs. 19, 20).

Fig. 3: Coordination environment characterization of Ni single atoms for CO₂RR.

a XRD patterns of Ni-CNTs30/PC, Ni-CNTs60/PC, Ni-CNTs90/PC, and Ni-CNTs180/PC. Corresponding high-resolution of Ni 2p XPS spectra (b) and N 1s XPS spectra (c). d Ni K-edge XANES spectra. e k²-weighted Fourier transform of the EXAFS spectra at the Ni K-edge of Ni foil, Ni-CNTs30/PC, and NiO. f EXAFS fitting curve for Ni sites in Ni-CNTs30/PC. In situ ATR-FTIR spectra recorded at –0.75 V vs. RHE in a mixture of CO₂/N₂ (v/v = 15/85) on (g) Ni-CNTs30/PC and (h) PC NSs. i Schematic illustration of the CO₂RR process on Ni-CNTs/PC, where the Ni–N₃ on CNTs are the active sites for forming *COOH intermediate and low-concentration CO₂RR. Source data are provided as a Source Data file.

To delve deeper into the coordination structures of Ni single atoms, X-ray absorption near-edge structure (XANES) and extended X-ray absorption fine structure (EXAFS) analyses were conducted. As illustrated in Fig. 3d, the Ni K-edge position of Ni-CNTs30/PC is located between that of NiO and Ni foil, suggesting a positive Ni valence state between 0 and +2. The EXAFS spectrum of Ni-CNTs30/PC displays a dominant peak at around 1.41 Å (Fig. 3e), which can be attributed to the Ni−N path scattering. Meanwhile, the coordination of Ni−Ni is negligible, evidencing that the Ni species mostly exist in the form of single atoms. Furthermore, quantitative EXAFS fitting was conducted to extract the structure parameter. The coordination number of N and C for Ni-CNTs30/PC in the first coordination shell was around 3 and 1, with the average scattering length of 1.85 and 1.86 Å, respectively, reflecting that the proposed local structure is Ni−N₃C (Fig. 3f, Supplementary Fig. 21, and Supplementary Table 4).

Following the identification of Ni species, in situ electrochemical attenuated total reflectance Fourier transform infrared spectroscopy (ATR-FTIR) was carried out to elucidate the contribution of Ni−N₃ to CO₂RR. The experiments were performed at a potential of −0.75 V vs. reversible hydrogen electrode (RHE) in 0.5 M potassium bicarbonate (KHCO₃) using a home-made ATR-FTIR setup under dilute CO₂ stream (CO₂/N₂ = 15/85), as shown in Supplementary Fig. 22a. A characteristic peak located at 1356.8 cm^–1 appears in Ni-CNTs30/PC after 5 min of catalysis and increases gradually with reaction time (Fig. 3g). The characteristic peak can be ascribed to the vibration of C−O stretching of the *COOH, a widely recognized key intermediate for CO₂RR to CO⁴⁰. In sharp contrast, the *COOH intermediate cannot be detected on pure PC NSs without Ni-CNTs, indicating CO₂RR barely happens (Fig. 3h). While for Ni-CNTs180/PC, the appearance of *COOH signal takes a longer time with a weaker intensity in comparison with Ni-CNTs30/PC (Supplementary Fig. 22b), manifesting that the high coverage of CNTs, even with a higher loading of Ni single atoms, has an adverse effect on CO₂ activation, possibly due to insufficient CO₂ supply. These findings imply that Ni single atoms in CNTs are the active sites for CO₂RR rather than PC NSs, as shown in Fig. 3i.

CO₂/N₂ adsorption selectivity in Ni-CNTs/PC catalyst

Since the low-concentration CO₂RR highly depends on the adsorption capability of CO₂, we proceeded with a detailed study of the CO₂ adsorption behavior in the catalysts composed of PC NSs with varying growth periods of Ni-CNTs from 30 to 180 min, namely Ni-CNTs30/PC, Ni-CNTs60/PC, Ni-CNTs90/PC, and Ni-CNTs180/PC. Furthermore, bare PC NSs and Ni-CNTs30 were also studied for comparison. The uptake capacity of CO₂ was evaluated from adsorption isotherms at 298.15 K. As shown in Fig. 4a and Supplementary Table 5, PC NSs achieve a remarkable CO₂ uptake of 1.36 mmol g^–1 at 1.00 bar. When Ni-CNTs grow on the surface, the CO₂ uptake shows a negative correlation to the CNTs coverage. Without PC NSs, the bare Ni-CNTs present a poor CO₂ uptake capability (~0.11 mmol g^–1) at 1.00 bar, which is close to that of Ni-CNTs180/PC. The differences in adsorption capacity among the six samples provide a strong hint that the PC NSs fundamentally determine the CO₂ adsorption ability of Ni-CNTs/PC catalyst, while Ni-CNTs growth may block the pores in PC NSs, which affects the CO₂ capture. Under conditions of low CO₂ concentration (0.15 bar), the CO₂ adsorbed amount is calculated to be 0.29, 0.07, 0.03, 0.02, and 0.02 mmol g^–1 for Ni-CNTs30/PC, Ni-CNTs60/PC, Ni-CNTs90/PC, Ni-CNTs180/PC, and Ni-CNTs30, respectively. This observation aligns perfectly with the electrochemically active surface area (Supplementary Fig. 23). Combining with the ATR-FTIR results in Fig. 3g–i, it can be concluded that the PC NSs are responsible for CO₂ adsorption in the dilute stream and facilitate the access of CO₂ to Ni−N₃ active sites, jointly realizing dilute CO₂RR. Impressively, with an ultralow CO₂ pressure of 0.03 bar, the Ni-CNTs30/PC still exhibits 0.094 mmol g^–1 of CO₂ uptake, which is expected to realize ultralow-concentration CO₂RR.

Fig. 4: Study of CO₂/N₂ gas adsorption and separation behavior.

a CO₂ adsorption isotherm at 298.15 K on PC NSs, Ni-CNTs30/PC, Ni-CNTs60/PC, Ni-CNTs90/PC, Ni-CNTs180/PC, and Ni-CNTs30. b Corresponding CO₂/N₂ IAST selectivity at 298.15 K and 1.0 bar for the binary mixture of CO₂/N₂ (15/85). c Pore-size distribution curves for PC NSs, Ni-CNTs30/PC, Ni-CNTs60/PC, Ni-CNTs90/PC, Ni-CNTs180/PC, and bare Ni-CNTs30. d Schematic illustration of the column breakthrough experiments, analogous to the MEA cell experiments. e Experimental column breakthrough curves for dry condition and 60% of RH CO₂/N₂ (v/v = 15/85) separations of PC NSs, Ni-CNTs30/PC, and Ni-CNTs180/PC. f Schematic of a two-bed VSA process by Aspen adsorption. The Aspen simulation results of CO₂ recovery and purity (g), CO₂ velocity and corresponding h LDF coefficient, and the i total energy consumption for the CO₂ separation process in Ni-CNTs30/PC, Ni-CNTs180/PC, and Ni-CNTs30. Source data are provided as a Source Data file.

To quantify the adsorption selectivity of CO₂ and N₂, we employed the ideal adsorbed solution theory (IAST) to predict the adsorption equilibrium for the CO₂/N₂ mixture with a volume ratio of 15/85^41,42, based on a set of pure N₂ and CO₂ adsorption isotherms data which were described by the Quadratic and BET model (Supplementary Note 3 and Supplementary Figs. 24–29), respectively. As shown in Fig. 4b and Supplementary Table 5, the calculated IAST selectivity for CO₂ in the six catalysts decreased in the order of PC NSs (40.36) > Ni-CNTs30/PC (16.17) > Ni-CNTs60/PC (10.28) > Ni-CNTs90/PC (7.73) > Ni-CNTs30 (6.94) > Ni-CNTs180/PC (6.34), evidencing that PC NSs are responsible for the physical adsorption of CO₂ and dense Ni-CNTs coverage sacrifices the CO₂ capture capability.

To make it clear, the specific surface area and pore distribution of the as-prepared catalysts were measured based on the CO₂ gas adsorption–desorption isotherms at 273.15 K. As depicted in Fig. 4c and Supplementary Table 6, PC NSs possess an average pore size of 0.5 nm, which just falls into the region of ultramicropore (pore size <1 nm, especially those ranging from 0.3 to 0.6 nm)⁴¹, indicating the swelling-assisted Ni-CMC aerogel enables the formation of a well-defined ultramicroporous structure. Accordingly, the total ultramicropore volume (<1 nm) is calculated to be ~0.051 cm³ g^–1, testifying to the presence of abundant ultramicropores in PC NSs responsible for CO₂ adsorption. In comparison, the bare Ni-CNTs30 without PC NSs have almost no pore size distribution in the ultramicropore region, and the corresponding volume is quite small (only 0.002 cm³ g^–1). When CNTs gradually grow on the PC NSs, there is an obvious decline in porosity. The total ultramicropore volume of Ni-CNTs30/PC is 0.039 cm³ g^–1, which is higher than that of Ni-CNTs60/PC (~0.012 cm³ g^–1) and Ni-CNTs90/PC (~0.008 cm³ g^–1). While for Ni-CNTs180/PC, no porous structure can be observed, which is similar to bare Ni-CNTs, affirming our conclusion that the as-grown Ni-CNTs cover the ultramicropores of PC NSs. Noted that the gas adsorption–desorption isotherms of Ni-CNTs30/PC were also conducted in the N₂ atmosphere, which presents a narrow pore size distribution over a wide range of 0.60 to 2.30 nm (Supplementary Fig. 30), alongside a larger fitting standard deviation as compared to that in the CO₂ atmosphere. This can be explained by the larger kinetic diameter of N₂ (3.64 Å) than that of CO₂ (3.30 Å), which induces a steric effect⁴¹ only available for the smaller CO₂ molecules diffusion into the ultramicropores in PC NSs, thus greatly contributing to the superior adsorption selectivity of CO₂ in the dilute CO₂ stream.

To simulate and reflect the CO₂ diffusion ability in the cathode channel of MEA, column breakthrough experiments were further carried out to examine the feasibility of the ultramicroporous PC NSs for the separation of CO₂/N₂ (v/v = 15/85) with 60% of relative humidity (RH) at ambient temperature and pressure (Fig. 4d). As shown in Fig. 4e, the breakthrough time of N₂ and CO₂ on bare PC NSs is 0.11 and 3.26 min g^–1, respectively, serving as direct evidence that the PC NSs are capable of separating N₂ and CO₂. When Ni-CNTs30 covers PC NSs, the breakthrough time decreases to 0.11 min g^–1 for N₂ and to 1.30 min g^–1 for CO₂, respectively, apparently retaining the separation ability, largely due to the partial exposure of ultramicropores. The longer breakthrough time of CO₂ than N₂ further corroborates that the PC NSs can physically adsorb CO₂ and delay the CO₂ from penetrating through. While for Ni-CNTs180/PC, the breakthrough times for CO₂ (0.03 min g⁻¹) and N₂ (0.02 min g⁻¹) are very close, ascribed to the almost complete concealing of ultramicropores by highly dense Ni-CNTs, in good agreement with the observation in Fig. 4a–c.

To quantitatively evaluate the CO₂ transport rate on Ni-CNTs30/PC in MEA, a two-bed vacuum pressure swing adsorption (VSA) process was conducted through the Aspen Adsorption simulation⁴³. 15% of CO₂ balanced by N₂ serves as the feedstock with a total feed rate of 1.45 kmol h^–1 (Fig. 4f, Supplementary Note 4, Supplementary Figs. 31–36, and Supplementary Tables S7–11). During the adsorption process at 1.00 bar, the CO₂ loading on Ni-CNTs30/PC is ~0.34 mol kg^–1, approximately sixfold higher than that on Ni-CNTs180/PC (0.06 mol kg^–1) and Ni-CNTs30 (0.05 mol kg^–1) (Supplementary Table 12). On account of superior adsorption capability, Ni-CNTs30/PC achieves 94.8% of recovery and 79.5% of purity for CO₂, both of which are higher than those of bare Ni-CNTs30 and Ni-CNTs180/PC (Fig. 4g). Besides, the Ni-CNTs30/PC attains a remarkable CO₂ velocity of 0.19 kmol h^–1 within the bed, corresponding to a high CO₂ linear driving force (LDF) coefficient of 2.02 × 10^–2 s^–1 (Fig. 4h). By comparison, Ni-CNTs180/PC and bare Ni-CNTs30 exhibit an almost identical CO₂ velocity (0.12 kmol h^–1) with low LDF coefficient (3.24 × 10^–3 s^–1), much lower than those of Ni-CNTs30/PC. Benefiting from the high adsorption capacity and transport rate for CO₂, Ni-CNTs30/PC costs a lower energy consumption of $2.59\,{{{\rm{MJ}}}}\,{{{{{\rm{kg}}}}}_{({{{{\rm{CO}}}}}_{2})}}^{{-}1}$, which is only half that of Ni-CNTs180/PC ($4.64\,{{{\rm{MJ}}}}\,{{{{{\rm{kg}}}}}_{({{{{\rm{CO}}}}}_{2})}}^{{-}1}$) and Ni-CNTs30 ($4.43\,{{{\rm{MJ}}}}\,{{{{{\rm{kg}}}}}_{({{{{\rm{CO}}}}}_{2})}}^{{-}1}$). Furthermore, COMSOL Multiphysics simulations were employed to model the CO₂ velocity within the MEA channel under a dilute CO₂ stream (CO₂/N₂ = 15/85). As depicted in Supplementary Fig. 37, Ni-CNTs/PC achieves a velocity of 0.0045 m s^–1 at the channel outlet, which is ~1.2 times higher than that observed in Ni-CNTs. Taken together, the Ni-CNTs30/PC with opportune Ni-CNTs coverage integrates the complementary functions of CO₂ capture and activation as well as enhanced CO₂ transport kinetics, ensuring feasibility for the dilute CO₂ reduction in MEA.

Direct electrocatalysis of dilute CO₂ in MEA

Based on the simulation results, the porous carbon nanosheets in Ni-CNTs30/PC can accelerate the outlet velocity of CO₂ towards CL ascribed to the high uptake capacity for CO₂, which is supposed to thin the GBL (Supplementary Note 1), as schematically illustrated in Fig. 5a. In this case, the local CO₂ concentration in the vicinity of active sites can be enhanced, thus favoring CO₂ reduction even at flue gas concentration. To this end, we evaluated the electrocatalytic performance of dilute CO₂ with a volume ratio of 15% (balance gas: N₂) with varying inlet CO₂ flow rates at 100 mA cm^–2 in MEA. The Ni-CNTs30/PC shows a consistently high CO Faradaic efficiency (FE_CO) of ~95%, irrespective of the dilute CO₂ stream flow rate varying from 10 to 120 sccm (Fig. 5b and Supplementary Fig. 38), providing powerful experimental evidence that Ni-CNTs30/PC with self-accelerated outlet CO₂ velocity can maximally thin the GBL, which is barely affected by CO₂ feed rate. In this context, the accessible CO₂ molecule to the Ni–N₃ sites are markedly increased, thus favoring dilute CO₂ reduction over competing HER. Conversely, the FE_CO on bare Ni-CNTs30 increases from 74.2 ± 4.5% to 85.9 ± 0.6% as the CO₂ feed rate increases from 10 sccm to 120 sccm, assuring the dilute CO₂ stream indeed forms a thick GBL, which can be narrowed by increasing CO₂ flow rate, in good accordance with Supplementary Equations 1-2. On the other hand, the CO selectivity on Ni-CNTs30/PC is consistently higher than that of Ni-CNTs30 even at a high CO₂ feed rate, strongly confirming that the porosity of the catalyst critically influences the thickness of GBL, which directly affects the dilute CO₂RR selectivity. With the same CO₂ concentration (15%) and flow rate (60 sccm), we further compared the FE_CO of CO₂RR on Ni-CNTs30/PC and bare Ni-CNTs30 under various cell voltages. As shown in Fig. 5c and Supplementary Fig. 39, Ni-CNTs30/PC and Ni-CNTs30 exhibit similar FE_CO under low cell voltage (<3.0 V) that corresponds to the CO₂RR reaction kinetics-limited region¹³, indicating comparable intrinsic CO₂RR activity derived from their Ni–N₃ sites. In the mass transport-limited region (>3.0 V), Ni-CNTs30/PC achieves the maximum FE_CO of 98.3 ± 0.6% at 3.5 V. In sharp contrast, the FE_CO over bare Ni-CNTs30 without PC NSs rapidly decreases to 79.9 ± 0.6% at 3.6 V, largely ascribed to the limited CO₂ transport caused by the thick GBL. Meanwhile, the Ni-CNTs30/PC exhibits a consistently high partial current density of CO (j_CO) over the entire cell voltage with a positively increasing trend as cell voltage (Fig. 5d). Specifically, the j_CO over Ni-CNTs30/PC reaches 196.5 ± 2.2 mA cm^–2 at 3.1 V, double that of bare Ni-CNTs30 (~83.8 ± 1.9 mA cm^–2). The above comparison experimentally collaborates the ultramicropore structure in Ni-CNTs30/PC guarantees CO₂ capture/enrichment and breaks the dilute CO₂ mass transport limit, thus preferentially converting CO₂ into CO even at high current densities. Besides, the calculated energy conversion current density (ECCD) of Ni-CNTs30/PC in Supplementary Fig. 40 is consistently higher than of Ni-CNTs30 over the measured cell voltage²⁹, indicating the reinforced CO₂ concentration gradient effect can lower the energy consumption, consistent with the Aspen simulation in Fig. 4i.

Fig. 5: CO₂RR performance in dilute CO₂ stream.

a Schematic to illustrate the dependence of GBL thickness on CO₂ concentration. b The comparison of CO Faradaic efficiency for Ni-CNTs30 and Ni-CNTs30/PC at 100 mA cm⁻² with dilute CO₂ flow rates ranging from 10 sccm to 120 sccm. Cell voltage-dependent c CO Faradaic efficiency and d j_CO on Ni-CNTs30 and Ni-CNTs30/PC. e Schematic to illustrate the effect of CL length and CO₂ concentration in the concentration gradient. The comparison of CO Faradaic efficiency for f Ni-CNTs30/PC and g Ni-CNTs180/PC measured in MEA under different CO₂ concentrations. h CO and H₂ Faradaic efficiency for Ni-CNTs30/PC in a 5 cm² MEA eletrolyzer. i The comparison of CO Faradaic efficiency over Ni-CNTs30, Ni-CNTs30/PC, and Ni-CNTs180/PC with the dilute CO₂ feeding from 15 to 3% at a current density of 100 mA cm^–2. j Comparison of the dilute CO₂RR performance metrics (e.g., FE_CO, j_total, cell voltage, CO₂ concentration, and CO₂ adsorption capacity) in this work with previous works on low-concentration CO₂RR with gaseous feedstocks^27,29,31. k Schematic illustration of introducing an O₂-adsorption pack column before CO₂RR MEA electrolyzer for direct electrocatalysis of O₂-containing simulated flue gas (CO₂/N₂/O₂ = 15/82/3). l, m The corresponding CO Faradaic efficiency with the OAPC system and the stability test of Ni-CNTs30/PC at 100 mA cm^–2. Error bars correspond to the standard deviations based on triplicate electrochemical measurements. Source data are provided as a Source Data file.

According to Supplementary Note 1 and Supplementary Equation 21, the narrowed GBL accelerates CO₂ molecular transport toward Ni–N₃ sites. To validate this, we further measured the j_CO and HER on Ni-CNTs30/PC and bare Ni-CNTs30 under a cell voltage of 3.2 V at varying temperatures from 298.15 to 338.15 K and calculated the activation energy (${E}_{a}$) for dilute CO₂RR and HER by Arrhenius-fitting j_CO and ${j}_{{H}_{2}}$ over 1/T, respectively²⁸. As shown in Supplementary Fig. 41, Ni-CNTs30/PC displays a substantially lower ${E}_{a}$ for CO₂ conversion into CO (3.50 kJ mol^–1), as compared to that for HER (83.19 kJ mol^–1). In contrast, for bare Ni-CNTs30 without the CO₂ enrichment effect, the ${E}_{a}$ for CO₂RR (11.61 kJ mol^–1) and HER (20.59 kJ mol^–1) differ little, highlighting the crucial role of PC NSs in enhancing CO₂RR reaction kinetics through accelerating CO₂ mass transfer velocity within GBL. Besides forming a thick GBL, the dilute CO₂ feed causes a weak concentration gradient effect within the CL based on the well-known Fick’s law²⁴ (Fig. 5e), which diminishes the CO₂ transport ability and slows down the CO₂RR kinetics. Since the CO₂ concentration gradient shows a strong dependence on both the feeding concentration of CO₂ and the length of CL, we compared the CO₂RR performance using Ni-CNTs/PC with varying lengths of Ni-CNTs and different concentrations of CO₂ with the cathode geometric area of 1 cm². As shown in Fig. 5f and Supplementary Fig. 42, it is surprising to find that the FE_CO in Ni-CNTs30/PC shows an irrelative trend to CO₂ concentration, all exceeding 90% with 15, 50, and 100% purity of CO₂ as feedstock over a wide current density from 50 to 250 mA cm^–2 in MEA. For Ni-CNTs180/PC, the FE_CO is consistently higher than 90% with pure CO₂, indicating the high activity of Ni–N₃ sites. However, when N₂ was introduced to dilute the CO₂, the FE_CO sharply decreases, especially at high current densities (Fig. 5g and Supplementary Fig. 42), which can be ascribed to the limited CO₂ transport. The marked contrast aligns with our hypothesis that a higher length of Ni-CNTs would lead to a weak CO₂ concentration gradient effect, which induces weak transport ability and low local CO₂ concentration on the top of CNTs, particularly during the low-concentration CO₂RR, resulting in sluggish CO₂RR kinetics (Supplementary Figs. 43–45). While for Ni-CNTs30/PC, the presence of both porous carbon nanosheets for concentrating CO₂ and optimal length of Ni-CNTs for converting CO₂ synergistically contributes to the establishment of the self-reinforced CO₂ concentration gradient effect. This effect is analogous to the nutrient diffusion process from soil to rhizome and leaves in plants, which can, in turn, promote CO₂ diffusion and accelerate CO₂RR kinetics.

Impressively, the FE_CO on Ni-CNTs30/PC reaches 95.4 ± 1.2% at 250 mA cm^–2, far surpassing those on Ni-CNTs60/PC (73.8 ± 3.4%), Ni-CNTs90/PC (36.1 ± 1.5%), and Ni-CNTs180/PC (20.0 ± 0.9%) (Fig. 5f, g and Supplementary Fig. 46). The negative correlation of FE_CO to Ni-CNTs length further collaborates with the weakened CO₂ concentration gradient with increasing the length of Ni-CNTs as CL. Accordingly, the Ni-CNTs30/PC requires the smallest cell voltage to drive the same magnitude of current density, which is also irrelevant to CO₂ concentration (Supplementary Fig. 47). A similar changing trend was also observed in H-type cell, where the Ni-CNTs30/PC display the FE_CO higher than 90% over the potential ranging from –0.57 to –0.85 V, with the highest value of 97.0 ± 1.9% at –0.75 V, and shows independence on CO₂ concentration (Supplementary Fig. 48). With respect to bare PC NSs, CO can be scarcely detected in either H-type cell (FE_CO <1%) or MEA device (FE_CO <1%), further validating the active sites for dilute CO₂RR originate from Ni-CNTs rather than porous carbon nanosheets, which aligns well with ATR-FTIR results (Supplementary Figs. 49, 50).

Considering the variance in Ni content across different lengths of Ni-CNTs, the turnover frequency (TOF) was calculated to evaluate the pristine activity of Ni-CNTs/PC. Even the Ni-CNTs10/PC with the strongest CO₂ concentration gradient demonstrates the highest TOF at different current densities, its FE_CO rapidly decreases to 44.3 ± 9.4% at 250 mA cm⁻² (Supplementary Figs. 51, 52), largely due to the insufficient active sites. On the contrary, Ni-CNTs30/PC with suitable CO₂ concentration gradient and active sites achieves a high TOF of 1.31 ± 0.02 × 10⁵ h^–1 at 250 mA cm^–2 (Supplementary Figs. 52, 53 and Supplementary Table 13), which is slightly lower than that Ni-CNTs10/PC (1.62 ± 0.34 × 10⁵ h^–1) and around eightfold enhancement in comparison with Ni-CNTs180/PC (1.99 ± 0.09 × 10⁴ h^–1). These contrasts substantiate that the low-concentration CO₂RR performance primarily depends on the CO₂ concentration gradient, especially when the active site density reaches a certain level (i.e., 2 wt%). Meanwhile, the Ni-CNTs30/PC shows a negligible change in both FE_CO (~90%) and cell voltage (~3.3 V) during 40 h of continuous electrocatalysis under dilute CO₂ stream (15%) at 100 mA cm^–2 (Supplementary Fig. 54), demonstrating the superior CO₂RR stability, consistent with the high recovery rate in Aspen simulation. This can be further supported by the postmortem analyses, where Ni-CNTs30/PC can well preserve its original morphology with uniform distribution of Ni and N throughout the CNTs profile (Supplementary Figs. 55, 56). In contrast, the FE_CO on bare Ni-CNTs30 and Ni-CNTs180/PC both rapidly decrease below 50% after only 1 h of CO₂RR operation under the same conditions. Notably, the low CO₂ concentration also significantly reduces the formation rate of (bi)-carbonate species, thereby extending the time to reach KHCO₃ or K₂CO₃ saturation (Supplementary Figs. 57, 58).

To assess the scalability of Ni-CNTs30/PC for the CO₂RR to CO under dilute conditions, we enlarged the cathode geometric area to 5 cm², maintaining a constant dilute CO₂ stream flow rate of 60 sccm (CO₂/N₂ = 15/85). As the total current incrementally rises from 0.25 to 1.00 A, the FE_CO consistently exceeds 90%. Even when the applied current reaches 1.25 A, the system sustains a FE_CO of 86.2 ± 3.7% with a CO partial current surpassing 1.00 A (Fig. 5h and Supplementary Fig. 59), demonstrating the great potential of the self-reinforced concentration gradient strategy for industrial application. Inspired by the high selectivity on Ni-CNTs30/PC for 15% purity of CO₂, we further decreased the CO₂ concentration and compared the selectivity of CO₂-to-CO at 100 mA cm^–2 in MEA. Remarkably, the Ni-CNTs30/PC catalyst achieves a FE_CO of 92.3% at 5% CO₂ feed concentration and retains 87.7% even under an ultralow CO₂ concentration of 3% (Fig. 5i and Supplementary Figs. 60), which realizes the ultralow concentration for studying CO₂ electroreduction as reported so far^{9,27,28,31,44,45,46,47} (Supplementary Figs. 61 and Supplementary Table 14). Considering the key performance metrics, including FE_CO, j_total, CO₂ concentration, and CO₂ adsorption capacity, this work explicitly stands out among the reported works^27,29,31,48, (Fig. 5j), demonstrating the superiority of the self-reinforced CO₂ concentration gradient effect under dilute CO₂ conditions. By contrast, a rather low FE_CO of 26.5 and 48.2% was observed under 5% of CO₂ on Ni-CNTs180/PC and Ni-CNTs30, respectively. The above phenomenon elucidates that the narrow GBL and self-reinforced CO₂ concentration gradient in Ni-CNTs30/PC ensures both high CO₂RR selectivity and good stability due to enhanced mass transport ability, which is expected to realize direct electrocatalytic conversion of CO₂ from the ultra-dilute CO₂ point sources.

Furthermore, we examined the feasibility of a self-reinforced concentration gradient strategy for converting O₂-containing simulated flue gas (15% CO₂ and 3–10% O₂, balanced by N₂) using different O₂ removal systems⁴⁹. By connecting the oxygen reduction reaction (ORR) and CO₂RR MEA electrolyzers in series (Supplementary Fig. 62), the flue gas flows into the first reactor for ORR with oxidized CNTs (O-CNTs) as a catalyst (Supplementary Figs. 63, 64), followed by entering the second reactor for CO₂RR. The tandem ORR reactor not only enables the on-site production and utilization of H₂O₂, but also mitigates the adverse effect of O₂ on CO₂RR performance by consuming the O₂ in the flue gas (Supplementary Figs. 65–68). With a 5 cm² cathode for ORR, the H₂O₂ yield reaches 3.34 ± 0.70 mol g_cat^–1 h^–1 at 0.50 A, which not only rapidly removes the organic effluent in 20 min but also attains a FE_CO of 77.7 ± 10.7% at 200 mA cm⁻² (Supplementary Fig. 69). To further improve the practical viability, we introduced a home-made O₂-adsorption packed column (OAPC) containing sodium dithionite (Na₂S₂O₄) and quartz sand before CO₂RR MEA electrolyzer (Fig. 5k). The OAPC exhibits an ultra-high O₂ removal efficiency (~99.4%), leading to a high FE_CO of 91.5 ± 2.0% at 200 mA cm⁻² over the Ni-CNTs30/PC (Fig. 5l and Supplementary Figs. 70, 71). Inspiringly, the OAPC strategy sustains an O₂ removal efficiency of 93% during 24 h of continuous operation, where the FE_CO consistently exceeds 90% with a stable cell voltage of ~3.1 V at 100 mA cm⁻² (Fig. 5m and Supplementary Fig. 72). For the higher-concentration O₂-containing flue gas (e.g., 5–10% typically from ore roasting and coal combustion processes)⁵⁰, the OAPC treatment provides >99% of O₂ removal efficiency along with a FE_CO above 90% (Supplementary Fig. 73). The coupling of OAPC with the self-reinforced CO₂ concentration gradient strategy enables favorable economic viability (CO cost = 378.3 $ ton_CO⁻¹), which significantly outperforms the conventional amine-based flue gas capture methods (Supplementary Fig. 74 and Supplementary Note 6).

Fundamental principles and mechanisms of self-reinforced CO₂ concentration gradient

To further understand the origin of self-reinforced CO₂ concentration gradient within the CL as well as its role in boosting CO₂ transport, we employed Ni-CNTs/PC with varying CNTs lengths to regulate the CO₂ concentration gradient for comparing the local CO₂ concentration and CO₂RR kinetics. Notably, the CO₂ concentration is fixed at 15% with N₂ as a balance gas in the following study. The difference in hydrophobicity among PC NSs, Ni-CNTs30/PC, Ni-CNTs60/PC, Ni-CNTs90/PC, and Ni-CNTs180/PC was first excluded by contact angle measurements, all of which show hydrophobic character with contact angles higher than 130° (Supplementary Fig. 75). Since the acidic CO₂ dissolution typically results in a pH decrease, a rotating ring disc electrode (RRDE) assembly was employed to determine the local pH change for further evaluating local CO₂ concentration during dilute CO₂RR⁵¹. As shown in Supplementary Fig. 76, the peak shift separation induced by pH (∆E_pH-shifted), which is a linear response to the pH change, is the least noticeable in Ni-CNTs30/PC across a wide potential range. While for Ni-CNTs180/PC, the ∆E_pH-shifted is obviously the largest and sharply increases at high potential range, mainly because of the limited CO₂ transport. The above result reflects that the Ni-CNTs30/PC with a suitable length of CNTs can provide sufficient CO₂ for electroreduction in the vicinity of the catalyst layer.

COMSOL Multiphysics simulations were further employed to visually comprehend the CO₂ concentration distribution on bare PC NSs and Ni-CNTs, as well as Ni-CNTs/PC heterostructures in MEA under a cell voltage of 3.0 V. As shown in Fig. 6a, it is found that the abundant ultramicropores in PC NSs effectively concentrate CO₂ near the CL, consistent with the conclusion from adsorption selectivity in Fig. 4. The admirable CO₂ capture capacity of Ni-CNTs/PC enables the formation of a pronounced concentration gradient along with the Ni-CNTs spanning a length of ~800 nm. Beyond this length, the CO₂ concentration stabilizes along the CNTs, and a significant concentration gradient ceases to exist. This well explains the strong dependence of dilute CO₂RR selectivity on CNTs length. For Ni-CNTs30/PC, the CO₂ concentration gradient distributes on the entire CNTs, which drives CO₂ transport from porous carbon to Ni-CNTs and guarantees a high flux of CO₂ accessible to the Ni–N₃ sites based on Fick’s law, thus accelerating dilute CO₂RR kinetics and rapid conversion. In contrast, the bare Ni-CNTs exhibit no discernible concentration gradient, which is responsible for the inferior CO₂RR performance. Meanwhile, the Ni-CNTs/PC with self-reinforced CO₂ concentration gradient induces a concentration gradient for CO product (Supplementary Fig. 77), which ensures prompt CO desorption and avoids the blocking of active sites by CO. For Ni-CNTs with a length of ~1000 nm, without the presence of PC NSs, results in a notably low CO₂ concentration, which in turn aggravates the HER and leads to poor CO₂RR selectivity, underscoring that the CO₂ enrichment is the key premise for dilute CO₂ reduction.

Fig. 6: The self-reinforced CO₂ concentration gradient effect within CL.

a Comparison of modeled CO₂ concentration along the CL for PC NSs (left), Ni-CNTs/PC (middle), and bare Ni-CNTs (right). The color bar represents the concentration of CO₂. DRT plots for b Ni-CNTs30/PC, c Ni-CNTs60/PC, and d Ni-CNTs180/PC in MEA. e Time-dependent local CO₂ concentration during the low-concentration CO₂RR on Ni-CNTs30/PC, Ni-CNTs60/PC, and Ni-CNTs180/PC. f Proposed mechanism in Ni-CNTs/PC for enhancing CO₂ transport across GBL and CL in low-concentration CO₂RR through a self-reinforced CO₂ concentration gradient. Source data are provided as a Source Data file.

To assess the improved CO₂ transport ability via the self-reinforced concentration gradient, the distributed relaxation time (DRT) was acquired through deconvolving the electrochemical impedance spectroscopy (EIS) during dilute CO₂ electroreduction in MEA at a cell voltage from 2.8 to 3.6 V (Fig. 6b–d and Supplementary Figs. 78, 79). The DRT plots exhibit three distinct peaks over a timescale (τ) ranging from 10⁻⁵–10⁻³, 10⁻³–10⁻¹, and 10⁻¹–10, corresponding to contact resistance, charge transfer, and CO₂ mass transport, respectively^52,53. As Ni-CNTs length decreases, the CO₂ mass transport process in the timescale region of 10⁻¹ to 10 s appears high-frequency shifts along with decreased impedance, corroborating Ni-CNTs30/PC can ameliorate the problem associated with mass transport. Accordingly, the charge transfer process in the timescale of 10^–3 to 10^–1 is accelerated in Ni-CNTs30/PC as compared to Ni-CNTs60/PC and Ni-CNTs180/PC, which aligns well with the Tafel slopes (Supplementary Fig. 80), declaring the improved dilute CO₂RR kinetics by the self-reinforced CO₂ concentration gradient. A comparable trend was observed in the H-type cell (Supplementary Figs. 81, 82), where Ni-CNTs180/PC shows much higher resistance associated with both mass and charge transfer processes. This observation further demonstrates that a weak concentration gradient impedes CO₂ mass transport, consequently resulting in diminished reaction kinetics.

To quantify the mass transport rate of CO₂ within CL, the CO₂ mass transport coefficient (k) was introduced, which can be calculated using the following equation^54,55:

$$k=\frac{{j}_{CO}/nF}{{[C{O}_{2}]}_{0}-{[C{O}_{2}]}_{i}}$$

(1)

where n, F, [CO₂]₀, and [CO₂]_i represent the electron transfer number for CO₂ to CO, Faradaic constant (96485 C mol^–1), and the initial and instantaneous CO₂ concentration near the CL during CO₂RR, respectively. For the dilute CO₂RR in this work (CO₂/N₂ = 15/85), [CO₂]₀ is calculated to be 6.12 mM. [CO₂]_i is dynamically monitored by in situ fluorescence electrochemical spectroscopy (FES) during the CO₂RR process (Supplementary Fig. 83). As shown in Fig. 6e, Ni-CNTs30/PC shows a shorter relaxation time and higher equilibrium concentration of [CO₂]_i of 4.17 mM, as compared to Ni-CNTs60/PC (3.68 mM) and Ni-CNTs180/PC (3.24 mM). Accordingly, k was calculated to be 1.24, 0.98, and 0.59 cm s^–1 for Ni-CNTs30/PC, Ni-CNTs60/PC, and Ni-CNTs180/PC, respectively. The results strongly support our conclusion that the self-reinforced CO₂ concentration gradient accelerates the CO₂ transport rate and elevates the CO₂ concentration at the interface between the catalyst and the electrolyte.

Based on the above simulated and experimental results, we justifiably propose the mechanism of the self-reinforced CO₂ concentration gradient for low-concentration CO₂ to CO conversion in MEA, as schematically in Fig. 6f. In the case of dilute CO₂ system, PC NSs features abundant ultramicroporous architectures that preferentially adsorb and concentrate CO₂, which effectively reduces the thickness of GBL and facilitates the access of CO₂ to active sites (Supplementary Fig. 25). Under these circumstances, the Lewis basic sites of Ni–N₃ in Ni-CNTs chemically adsorb the accessible CO₂ molecules (Supplementary Fig. 84), forming the self-reinforced CO₂ concentration gradient along the CNTs profile, which weakens as the CNTs length increases. With a suitable Ni-CNTs growth time on porous carbon, the enhanced gradient concentration can continuously drive CO₂ transport to the Ni–N₃ active sites, which breaks the CO₂ mass transport limit and enables rapid conversion.

Discussion

In summary, we constructed heterostructures composed of porous carbon nanosheets and Ni single atoms anchored in CNTs (Ni-CNTs/PC) to mimic the biological function of nutrient diffusion in plants. Specifically, the ultramicroporous feature of the PC NSs serves a role analogous to soil, enabling selective CO₂ adsorption, while the rhizome-like Ni-CNTs incorporating Ni–N₃ active sites efficiently transport and catalyze the captured CO₂. This appealing architecture fosters a complementary effect between the PC and Ni-CNTs for dilute CO₂RR, where the porous carbon nanosheets enrich CO₂ and the Ni-CNTs catalyze CO₂, allowing to establish a self-reinforced concentration gradient during the dilute CO₂RR. The combined experimental and theoretical analyses reveal that the self-reinforced CO₂ concentration gradient effect can narrow the gas boundary layer and strengthen the concentration gradient within the catalyst layer. This enhancement in CO₂ mass transport accelerates dilute CO₂RR kinetics and suppresses the HER. Consequently, the Ni-CNTs30/PC achieves a high CO Faradaic efficiency of 95.4 ± 1.2% under 15% CO₂ at an industrial-level current density of 250 mA cm^–2. When enlarging the MEA size, the CO partial current density reaches 1.08 ± 0.05 A. Notably, the integrated OAPC-CO₂RR MEA electrolyzer achieves >90% of CO selectivity from simulated flue gas containing 3–10% of O₂. This self-reinforced CO₂ concentration gradient strategy offers a promising solution to overcome the challenges of CO₂ mass transport in the conversion of dilute CO₂ from industrial point sources, marking a substantial step toward practical application.

Methods

Synthesis of the Ni-CMC hydrogel

The swelled Ni²⁺-crosslinked carboxymethylcellulose sodium (Ni-CMC) hydrogel was fabricated through a combination of the sol-gel method and directional freeze-casting method. Typically, 1.0 g of CMC (Sigma-Aldrich, with a viscosity ranging from 1500–3000 mPa·s, USP grade) and 0.2 g of NiCl₂ (Alfa Aesar, 98%) were dissolved in 50 mL of deionized water to form a green homogeneous colloidal solution. Then, the uniform sol was poured into a plastic mold (15 mm × 15 mm × 15 mm) and placed on a frozen copper billet (frozen in liquid nitrogen), enabling the upper sol surface to freeze solid. The as-prepared solid was immersed in an ethanol/NaOH (Sigma-Aldrich, 98%) bath (–30 °C) until the ice crystals completely dissolved in ethanol to form Ni-CMC hydrogel.

Synthesis of the Ni-CMC aerogel

The as-fabricated Ni-CMC hydrogel was soaked in deionized water (18.25 MΩ cm) for swelling, with the water being replaced every 2 h. After a 48-h swelling period, the same directional freeze-casting method was replicated. Subsequently, the sample underwent freeze-drying treatment at –60 °C for 48 h, and the Ni-CMC aerogel was ultimately obtained.

Synthesis of the Ni-CNTsX/PC

The Ni-CNTsX/PC was synthesized via the chemical vapor deposition method, where X represents the growth time. Specifically, for the Ni-CNTs30/PC sample, the Ni-CMC aerogel was placed in the center of a tube furnace and heated to 750 °C with a heating rate of 5 °C min^–1 under an Ar atmosphere (Air Liquide Tianjin Co. Ltd, 99.9999%). After reaching this temperature, CH₃CN (Sigma-Aldrich, 99.9%) was introduced by Ar to initiate the growth of Ni-CNTs for 30 min. Then the system was cooled to 25 °C. Finally, 1.0 M HCl (Gaofeng, AR) was used to etch the residual Ni species. For comparison, PC NSs were synthesized by the same process except for the introduction of CH₃CN.

Characterization

Field emission SEM (FE-SEM) images were captured using JSM-7900F. FTIR spectroscopy was performed utilizing a Nicolet iS50 (Thermo Fisher Scientific). XPS spectra was recorded using Thermo Scientific K-Alpha with aluminum Kα source (1486.6 eV) and a collimator at 12 kV and 72 W. N₂ adsorption–desorption isotherms were obtained at 77 K using a Micromeritics 3Flex system, with surface areas calculated via the Brunauer–Emmett–Teller (BET) method. The porosity distribution was analyzed using the density functional theory (DFT) method, employing CO₂ adsorption isotherms measured on a Micromeritics 3Flex instrument at the temperature of 273.15 K. Prior to conducting the tests, all samples underwent a degassing process at 573.15 K for a duration of 8 h. TEM equipped with EDS mappings was carried out in the Talos F200X G2. HAADF-STEM images were collected by JEM-ARM200F. XRD was performed using SmartLab 9KW with 2θ ranging from 5˚ to 80˚, where Cu Kα radiation (λ = 1.54 Å) was operated at 40 kV and 150 mA. XAS measurements were conducted at the XAS Beamline of the Australian Synchrotron, located in Melbourne. The analysis of the Ni K-edge was performed using Si (111) crystal monochromators. The contact angles were measured at ambient temperature utilizing the Dataphysics OCA 25, employing a probe consisting of 10 μL of water. ATR-FTIR analyses were carried out using a Thermo Fisher Scientific Nicolet iS50 FTIR spectrometer. The electrolyte of 0.5 M KHCO₃ was bubbled with a mixture gas (CO₂/N₂ = 15/85) for 30 min before the test. Electrochemical curves were recorded during 20 min of continuous electrolysis while maintaining a constant potential of –0.75 V versus RHE.

Electrochemical measurements and product quantification

The electrochemical measurements were conducted using an Ivium electrochemical workstation at room temperature (25 °C). In the case of the H-type cell, 5 mg of catalyst was sequentially dispersed in 500 μL of isopropyl alcohol (Sigma-Aldrich, 99.9%), 480 μL of DI water, and 20 μL of Nafion D520 binder solution (5 wt%, SCI Material Hub). Then, the mixture was sonicated for 30 min to achieve a homogenous ink. The resulting ink was dropped onto AvCarb P75T (20% PTFE treated, SCI Material Hub) to form a working electrode with a catalyst loading of ~0.5 mg cm². SELEMION anion exchange membrane (size: 2 cm × 2 cm, thickness: 100 μm, SCI Material Hub) was used to separate the cathode and anode chambers, which was presoaked in deionized water for 24 h before being employed. Platinum (Pt) gauze (size: 15 mm × 15 mm, SCI Material Hub) served as the counter electrode. The gas flow controller was used to continuously supply various concentrations of a CO₂/N₂ mixture gas into electrolyte of 0.5 M KHCO₃ with pH of 7.0 ± 0.2 (Sigma-Aldrich, 99.99%), including 15, 50, and 100% of CO₂ with a total flow rate of 30 sccm (CO₂ and N₂, 99.9999%, Air Liquide Tianjin Co. Ltd). The potential, measured against the saturated calomel electrode (SCE, SCI Material Hub), was transformed into RHE values with an iR-correction using the following equation:

$$\,E(vs.{{{\mathrm{RHE}}}})=E(vs.{{{\mathrm{SCE}}}})+0.244V+0.0591V\times pH-iR$$

(2)

The iR compensation was implemented during the electrolysis experiments at each applied potential in the H-type cell. The ohmic resistance was determined through electrochemical impedance spectroscopy (EIS) measurements, which were conducted by applying an AC voltage with an amplitude of 5 mV across a frequency range spanning from 100 to 0.01 kHz.

During the electrochemical reaction, the gas products were analyzed utilizing a gas chromatograph (GC-2014), where H₂ was detected by a thermal conductivity detector (TCD) and CO was detected by a flame ionization detector (FID). The Faradaic efficiency (FE) for gas production (H₂ and CO) was calculated as follows:

$${{{\mathrm{FE}}}}=\frac{x\cdot v\cdot \frac{2{P}_{0}F}{{{{\mathrm{RT}}}}}}{{j}_{{{{\mathrm{total}}}}}}\times 100\%$$

(3)

where x represents the volume fraction of H₂ or CO which determined by GC, v represents gas flow rate (mL min^–1), F represents Faraday constant, P₀ is 101.325 kPa, T represents temperature (K), R represents gas constant, and j represents total current density at a specific voltage. The corresponding partial current density was calculated by following:

$${j}_{CO}={j}_{{{{\mathrm{total}}}}}\times {{{\mathrm{FE}}}_{{\mathrm{CO}}}}$$

(4)

The energy conversion current density (ECCD) for the electrochemical reduction of dilute CO₂ to CO (ECCD_CO) were determined according to the following equations²⁸:

$${{{\mathrm{ECC}}}}{D}_{{{{\mathrm{CO}}}}}=\frac{{E}_{{{{\mathrm{OER}}}}}-{E}_{{{{\mathrm{CO}}}}}}{{E}_{{{{\mathrm{cell}}}}}}\times {j}_{{{{\mathrm{CO}}}}}$$

(5)

where E_OER and E_CO represent the thermodynamic equilibrium potential for OER and CO₂RR, respectively.

Additionally, TOF (h^–1) for CO was calculated by the following formula:

$${{{\mathrm{TOF}}}}=\frac{{j}_{{{{\mathrm{CO}}}}}/2F}{{m}_{{{{\mathrm{cat}}}}}\cdot \omega /{M}_{Ni}}\times 3600$$

(6)

where m_cat represents the weight of catalysts (g), ω represents the metal loading of Ni, which was detected by ICP-MS, M_Ni represents the atomic mass of Ni (58.69 g mol^–1).

As for the dilute CO₂RR test in MEA electrolyzer, the Sustainion X37-50 Grade FA anion exchange membrane (2 cm × 2 cm, thickness 50 μm, SCI Material Hub) served as the separator, which was soaked in 1.0 M KOH (Sigma-Aldrich, 95%) at least 24 h before being employed. In the case of the MEA cell, 20 mg of catalyst with 0.2 mg PTFE powder (~200 nm, Teflon, SCI Material Hub) was sequentially dispersed in 1 mL of isopropyl alcohol, 0.9 mL of deionized water, and 0.1 mL of Nafion D520 binder solution. Then, the mixture was sonicated for 30 min to achieve a homogenous ink. The resulting ink was air-spray coated onto the Sigracet 39BB gas diffusion layer electrode (size: 1.0 cm × 1.0 cm, thickness: 0.3 mm, SCI Material Hub) to form a working electrode with a catalyst loading of ~1.0 mg cm². A Pt/Ti mesh (size: 1.5 cm × 1.5 cm, thickness: 0.3 mm, Pt coating thickness: 0.5 μm, SCI Material Hub) with the IrO₂ (mass loading: ~2.0 mg cm⁻²) served as the anode catalyst. Detailly, 20 mg IrO₂ (Sigma-Aldrich, 99.9%) was sequentially dispersed in 2 mL of isopropyl alcohol and 0.1 mL of Nafion D520 binder solution, followed by 30 min of sonication to achieve a homogenous ink. The resulting ink was air-spray coated onto the Pt/Ti mesh. The MEA measurements used a freshly prepared 1 M KOH electrolyte (pH = 13.6 ± 0.2), which was replaced daily to ensure consistency. In this work, all electrolytes are stored in a thermostat at 25 °C between experimental sessions. All cell voltage measurements were conducted without iR compensation.

Direct conversion of O₂-containing simulated flue gas

The tandem ORR-CO₂RR MEA electrolyzer system was designed to operate with simulated flue gas containing 15% CO₂, 3% O₂, and 82% N₂ at a controlled total flow rate of 60 sccm. The 5 cm² ORR-MEA electrolyzer utilized 0.5 M KHCO₃ as the anode electrolyte with a pH of 7.1 ± 0.1 to minimize carbon loss while maintaining stable H₂O₂ production. The cathode was fabricated by casting a homogeneous slurry of O-CNTs with 1 wt% PTFE (Teflon, SCI Material Hub) onto a 5 cm² reinforced 39BB GDL. IrO₂/Pt/Ti mesh (area: 5 cm²) served as the anode catalyst. The cathode and anode chambers were separated by a Sustainion X37-50 Grade FA anion exchange membrane. The oxygen removal efficiency (ORE) of the ORR-MEA electrolyzer can be determined using the following expression:

$${{{\rm{ORE}}}}=\frac{{{{{\rm{c}}}}}_{{O}_{2},in}-{c}_{{O}_{2},out}}{{c}_{{O}_{2},in}}\times 100\%$$

(7)

where co₂, in and co₂, out represent the O₂ volume fraction at the electrolyzer inlet and outlet, respectively, which can be detected by GC.

To enhance the residence time of simulated flue gas, the O₂-adsorption packed column system (OAPC) are helical with inner and outer diameters of 15 and 100 mm, respectively. The Na₂S₂O₄ (Sigma-Aldrich, AR, 90%) and quartz sand (Sigma-Aldrich, 40 mesh) are packed under an anaerobic environment with a volume ratio of 1:1. In this OAPC system, Na₂S₂O₄ serves as the primary O₂ adsorbent while the quartz sand provides dual functions of increasing the gas-solid interfacial contact area and preventing the agglomeration of Na₂S₂O₄ powder. To reflect the O₂ removal ability of the OAPC, the ORE can be calculated by Eq. (7).

Quantification of the accumulation rate of K salt

The quantification of accumulation K salt rate was performed using 2-h CO₂RR measurements with Ni-CNTs30/PC as catalyst at 100 mA cm^–2 under different CO₂ concentrations (3, 15, 50, and 100%) in MEA using 1.0 M KOH as anolyte. When the tests finished, the cathode channels were thoroughly and repeatedly rinsed with DI water while remaining the MEA assembled. The resulting solution was analyzed via ICP-MS to quantify the formation rate of the accumulated salt.

Single-component adsorption experiments and IAST prediction

The adsorption isotherm curves of CO₂ and N₂ on 100 mg of the PC NSs, Ni-CNTs30/PC, Ni-CNTs60/PC, Ni-CNTs90/PC, Ni-CNTs180/PC, and bare Ni-CNTs30 were measured using the Micromeritics 3Flex equipment at 298.15 K. The BET adsorption isotherm model and the quadratic model was employed to fit the adsorption isotherms of CO₂ and N₂, respectively. Meanwhile, IAST was utilized to predict the mixture adsorption isotherm based on the pure CO₂ and N₂ isotherms^41,42. The BET adsorption isotherm model was used to describe the adsorption behavior of CO₂ with three parameters as follows:

$$q=\frac{I{P}_{1}\cdot {P}_{i}}{(1+I{P}_{2}\cdot {P}_{i})\cdot (1-I{P}_{3}\cdot {P}_{i})}$$

(8)

where IP₁, IP₂, and IP₃, represent the isotherm parameters of the BET model, P_i represents the corresponding equilibrium pressure (bar). The quadratic adsorption isotherm model was used to describe the adsorption behavior of N₂ with three parameters as follows:

$$q=I{P}_{1}\frac{(I{P}_{2}+2\cdot I{P}_{3}\cdot {P}_{i})\cdot {P}_{i}}{1+I{P}_{2}\cdot {P}_{i}+I{P}_{3}\cdot {{P}_{i}}^{2}}$$

(9)

where IP₁, IP₂, and IP₃ represent the isotherm parameters of the quadratic model, P_i represents the corresponding equilibrium pressure (bar). For the calculation of IAST, see Supplementary Note 3 for more details.

CO₂ chemical adsorption experiments

The CO₂ chemical adsorption behavior was studied through temperature programmed desorption (TPD) measurements, conducted using an AutoChem1 II 2920. For a detailed assessment of CO₂-TPD, 100 mg of catalyst powder was placed in a reaction cube and subsequently heated to 523.15 K at the rate of 10 K min^–1 and maintained for 1 h for pre-drying under a helium (He) atmosphere. Once cooling down to 323.15 K, the adsorption of CO₂ was conducted in mixed gas (10% CO₂ and 90% He) flow for 1 h. Following this, He flow (99.999%) was introduced to purge the residual CO₂ from the catalyst surface for an additional hour. Finally, the reaction cube was gradually heated from 323.15 to 1073.15 K at a rate of 10 K min^–1. The desorbed gas was detected using the TCD.

CO₂/N₂ breakthrough experiments

The CO₂/N₂ breakthrough experiments were conducted using a multicomponent competitive adsorption equipment (BSD-MAB) at 298.15 K. The stainless-steel column with an inner diameter of 10 mm and a length of 70 mm served as the fixed-bed reactor to minimize wall-flow effects and associated breakthrough curve tailing. The weight of the sample packed in the column was 500–800 mg. Prior to the breakthrough experiment, all samples were activated via purging He flow at 573.15 K for 2 h. Subsequently, the mixed gas containing 15% CO₂ and 85% N₂ was introduced at a flow rate of 10 sccm. The outlet gas from the column was analyzed using online mass spectrometry (INFICON). For humidified CO₂ tests, the feed gas was conditioned to 60% relative humidity via water vapor saturation, accurately reflecting the humidified gas for MEA operation.

Local pH measurement

Local pH environments of various samples were measured using RRDE (RRDE-3A, ALS). This device comprises a catalyst disc electrode and a Pt ring electrode, with SCE and Pt foil functioning as the reference and counter electrode, respectively. Prior to testing, cyclic voltammetry (CV) was employed to purify the Pt ring under N₂ atmosphere, sweeping from 0.45 to 1.15 V at a rate of 100 mV s^–1. Subsequently, the same technique was utilized to activate the catalyst. Operating at a rotation speed of 1600 rpm, a constant potential ranging from –0.68 to −0.98 V was held on the catalyst disc to perform the CO₂RR process Simultaneously, potential cycling from 0.17 to 1.36 V was applied to the Pt ring electrode for detecting the peak of CO oxidation. Consequently, the difference value between the concentration-dependent peak of CO (E_CO) and the actual oxidation peak (E_obs) observed is designated as the pH-induced peak shift (∆E_pH-shifted), a direct indicator of the local pH level⁵¹.

Operando EIS experiments

Operando EIS measurements were conducted in both an H-type cell and an MEA at 298.15 K utilizing the Ivium electrochemical workstation. The impedance data were galvanostatically recorded, employing an amplitude of 10 mV across a frequency range of 0.01 to 100 kHz. The cell voltages in the MEA were adjusted from 2.8 to 3.6 V in increments of 0.2 V, while the potentials in the H-type cell were adjusted from –0.68 to –0.98 V in increments of 50 mV. The EIS data were analyzed using the DRT method, which were discretized by employing radial basis functions of Gaussian type⁵⁶. Fitting without inductance was adopted to address the inductive features:

$${Z}_{DRT}={R}_{\infty }+{\int ^{+\infty }_{-\infty }}\frac{\gamma ({{\mathrm{ln}}}\,\tau )}{1+i2\pi f\tau }d({{\mathrm{ln}}}\,\tau )$$

(10)

where R_∞ represents Ohmic resistance, γ(lnτ) represents the DRT function that describes the time relaxation characteristics of the electrochemical system studied, τ represents relaxation time and f represents frequency.

The regularization parameter λ utilized in Tikhonov regularization was optimally adjusted by employing the regularization method of mGCV.

In situ fluorescence electrochemical spectroscopy (FES)

A fluorescence spectrophotometer (QM/TM/NIR) connected with a CH Instruments 760E was utilized to monitor local CO₂ concentrations. Initially, a pH-sensitive fluorescent probe was synthesized by adding 8-hydroxypyrene-1,3,6-trisulfonic acid (HPTS, ACMEC, 98%) to 1 mL of a 0.5 M tetraoctylammonium hydroxide solution (TH, Macklin, 20 wt% in methanol), followed by the incorporation of an additional 2.5 mL of methanol to create the fluorescent probe solution. Subsequently, 30 μL of this solution was applied to the reverse side of the gas diffusion layer. The excitation and emission wavelengths for HPTS were set at 485 and 520 nm, respectively. Experiments were conducted in a custom-built cell using Ni-CNTs30/PC or Ni-CNTs180/PC as the cathode, a Pt foil (SCI Material Hub) as the anode, and 1 M KOH as the electrolyte, at a current density of 50 mA cm^–2.

Aspen adsorption simulation

A two-bed vacuum swing adsorption (VSA) was employed to evaluate the performance of Ni-CNTs30/PC, Ni-CNTs180/PC, and Ni-CNTs30 in separating CO₂ from a dilute CO₂ stream consisting of 85% N₂ and 15% CO₂ under 298.15 K^41,43. Aspen Adsorption was utilized for simulating and calculating the above process. During the simulation process, CFD4 was employed for partial differential equations. Detailed information regarding the mass balance model, momentum balance model, dynamic hypotheses, adsorption equilibrium models, bed parameters, VSA simulation methodology, and performance evaluation metrics for the VSA process can be found in Supplementary Note 4.

COMSOL simulation

A multiphysics model, implemented using COMSOL 6.1, was developed to elucidate the distribution and CO₂ transport dynamics within the local environment during the low-concentration CO₂RR process. This model was constructed based on an MEA electrolyzer, comprising anode and cathode catalyst layers, an electrolyte chamber, and a gas channel. The cathode catalysts were modeled using three configurations: 200 nm-thick PC NSs, 1-μm length of Ni-CNTs with anchored Ni–N₃ active sites, and a heterarchical structure of Ni-CNTs/PC integrated with PC NSs and Ni-CNTs. The total ultramicropores volume parameters of these model catalysts were derived from Supplementary Table 5. The Butler–Volmer equation was used to correlate the relationship between current density and applied electrode potential, and Faraday’s law was applied to convert the current density to the generation/consumption rates of chemical species in the system, which were used as the source/sink terms in the convection-diffusion-reaction equation. Detailed information can be found in Supplementary Note 5.