Introduction

Covalent organic framework (COF) materials exhibit significant potential in gas separation because of their highly ordered porous structure, designable organic building units, and exceptional chemical and thermal stability1,2,3. Compared with zeolites and metal‒organic frameworks (MOFs), the fabrication of dense and thin COF membranes on porous substrates remains a significant challenge using conventional in situ and secondary-growth methods owing to the inherently limited dynamic covalent reversibility, weak seed/substrate interactions, stringent crystallization requirements with sluggish kinetics, and large molecular dimensions of COF precursors4,5,6.

The interfacial polymerization (IP) method is the most popular method for fabricating dense and ultrathin COF membranes with exceptional gas separation performance under mild conditions7,8. However, the mechanical strength of the resulting COF films is still limited owing to their inherent brittleness, making them susceptible to cracking or even breaking apart9. Additionally, transferring the membrane onto substrates is technically challenging and difficult to control, and combined with its weak mechanical stability, these issues severely limit the practical application of the interfacial polymerization approach10. Therefore, developing an innovative COF membrane preparation strategy that combines exceptional separation performance with triggered stability represents a critical technological challenge.

Most recently, COF-mixed matrix membranes (COF-MMMs) have attracted both academic and industry attention. By combining the advantages of the processability of polymers with ordered mass transfer channels and abundant functional groups of COFs, great achievements have been obtained in the field of gas separation via COF-MMMs11,12. Currently, the dominant strategy for preparing COF-MMMs is dispersing pre-synthesized COF crystallites directly into polymer matrices13,14,15. However, micron-scale COF particles exhibit a limited interfacial contact area with the polymer matrix due to their large particle size, resulting in weak interfacial adhesion and the formation of non-selective defects16. Notably, Jin and coworker17 achieved the preparation of an ultrathin (<100 nm) mixed-matrix membrane (MMM) with high filler loading (80.4 vol%) through a solid-solvent processing strategy. The key to this strategy lies in using the polymer as a solid solvent to first form a metal salt@polymer precursor layer, followed by the controlled in situ conversion of the metal salt into a metal-organic framework (MOF). Throughout this process, the polymer segments tightly combine with the MOF particles, resulting in an intact MOF-polymer interface. The resultant membrane exhibits remarkable H2/CO2 separation performance with a selectivity of 76.1 and a hydrogen permeance of 3640 GPU. Similarly, Tan and coworker18 developed an in situ polymerization approach for fabricating COF-MMMs. Uniformly distributed 2D-COFs were synthesized by mixing COF precursors with polyethylene oxide (PEO) monomers, followed by a microwave-assisted Zincke reaction. The resulting COF-MMMs exhibited a remarkable permeability of 803.9 barrer for CO2 and a selectivity of 15.0 for CO2/H2. In situ polymerization effectively overcomes traditional challenges such as COF aggregation and poor interfacial compatibility. However, the low gas separation selectivity still limits the development and exploration of COF-MMMs as selective membranes for gas separation. Therefore, a highly controllable and reproducible method for fabricating COF-MMMs with highly dispersed COF nanocrystals for gas separation is highly desirable.

Non-solvent induced phase separation (NIPS) is a classic method for preparing polymer membranes19,20. Typically, polymer membranes prepared via NIPS exhibit an asymmetric structure comprising a skin layer with molecular sieve selectivity and finger-like pore channels that enhance gas transportation21. Polyether sulfone (PES) has excellent chemical stability, stable mechanical properties, a high glass transition temperature (~225 °C) and a high thermal decomposition temperature (~300 °C), making it a commonly used matrix material for NIPS22. PES membranes prepared via NIPS demonstrate both robust mechanical strength and temperature-resistant gas selectivity23,24, which fully exemplifies the advantages of NIPS technology for producing high-performance separation membranes.

The COF-MMMs with highly dispersed COF nanocrystals extending over the centimeter scale for highly efficient H2/CO2 separation were fabricated via a facile strategy, denoted induced phase separation (NIPS)-triggered in situ interfacial polymerization (IP). The COF-MMMs achieved an excellent H2/CO2 selectivity of 88.8 ± 2.46 with a high H2 permeance of 2738 ± 58.02 GPU (1 GPU = 3.35 × 10−10 mol m−2 s−1 Pa−1). Furthermore, grand canonical Monte Carlo (GCMC) simulations were systematically employed to investigate the adsorption behaviors of H2 and CO2 in TpPa-1, revealing a synergistic separation mechanism involving Knudsen diffusion in the PES matrix, selective adsorption by the COF domains, and molecular sieving through the interlayer spacing of COFs. This study provides a practicable pathway for fabricating COF-MMMs for gas separation.

Results

Synthesis and characterization of COF-MMMs

The fabrication process of COF-MMMs via the NIPS-triggered in situ IP strategy is illustrated in Fig. 1. The experimental section (Supplementary Fig. 1) provides additional methodological details, with the names of all the fabricated membrane materials comprehensively listed in Supplementary Data 1. Typically, the Tp@PES solution was prepared by dissolving the COF precursor 1,3,5-triformylphloroglucinol (Tp, organic aldehyde monomer) directly into a PES solution with 1-methyl-2-pyrrolidone (NMP) as the solvent. A glass fiber (GF) served as the supporting substrate, where the Tp@PES solution was uniformly deposited onto the GF substrate surface (Tp@PES@GF), followed by immersion into an aqueous phase containing p-phenylenediamine (Pa-1) and p-toluenesulfonic acid monohydrate (PTSA). During the NIPS process, the polymer solution immersed in the non-solvent bath (water) undergoes bidirectional diffusion driven by the solvent/non-solvent concentration gradient. The rapid outflow of solvent and influx of non-solvent disrupt thermodynamic equilibrium, triggering liquid‒liquid phase separation that simultaneously generates a polymer‒rich continuous matrix phase and dispersed domains dominated by solvent‒nonsolvent21,25. The rapid diffusion and precipitation of solvent at the surface cause the polymer to undergo instantaneous phase separation and deposition, forming a dense skin layer24,26. The polymer-rich phase solidified into a porous framework through chain entanglement or phase transition, establishing a stable solid‒liquid interface with the non-solvent bath. Concurrently, during the initial liquid-state stage of NIPS, the rapid solvent/non-solvent interdiffusion and the polymerization of Pa-1 in the aqueous phase facilitated the migration of Tp precursors toward the PES surface. These Tp precursors then underwent an in-situ interfacial polymerization (IP) reaction with Pa-1 monomers that had diffused to the PES surface from the aqueous phase. At the same time, a portion of the Tp monomers was carried by the solution to the internal surfaces of the finger-like pores. As the polymer phase gradually solidified, the fluidity of the entire system dropped drastically. The diffusion of Tp monomers trapped within the polymer chain network became severely restricted. However, this did not mean the reaction stopped. The Tp monomers that had reached or were in close proximity to the pore surfaces could still react with Pa-1 monomers that continuously diffused in from the aqueous phase, forming a dense, ultrathin COF membrane. The Tp monomers encapsulated within the polymer network likely formed COF nanocrystallites through local chain relaxation or via very short-range diffusion and rearrangement within microscopic pores. In the fabrication of COF-MMMs, the polymer matrix serves as both a solid solvent and a reactive site for subsequent in situ IP.

Fig. 1
Fig. 1
Full size image

Fabrication of COF-MMMs via the NIPS-triggered in situ IP strategy.

The PES membrane prepared directly on the GF (PES@GF) appeared white to the naked eye and exhibited a smooth surface, which was the typical structural morphology of the PES film fabricated by NIPS on the basis of the SEM images (Fig. 2a). The COF-MMMs prepared after 72 h of crystallization (TpPa-1@PES@GF-72) appeared reddish-brown. SEM revealed the formation of a membrane composed of irregular COF nanoparticle aggregates on the surface (Fig. 2b). The appearance of small pores in the COF membrane was attributed to the overly mild reaction conditions as well as the fact that the COF membrane consisted of small and irregular nanoparticles27. With prolonged polymerization time, COF nanoparticles formed on the membrane surface, leading to a denser and porous structure (Supplementary Fig. 2), accompanied by a distinct color change from white to reddish-brown (Supplementary Fig. 3). Atomic force microscopy (AFM) characterization revealed porous features of the PES@GF membrane (Fig. 2c), whereas the AFM image of the TpPa-1@PES@GF-72 membrane confirmed the uniform distribution of COF nanoparticles on the polymer matrix surface (Fig. 2d).

Fig. 2: Characterizations of COF-MMMs.
Fig. 2: Characterizations of COF-MMMs.
Full size image

a SEM images of the PES@GF membrane surface (inset: photographic image of the PES@GF membrane). b SEM images of the TpPa-1@PES@GF-72 membrane surface, (inset: photographic image of the TpPa-1@PES@GF membrane). c AFM image of the PES@GF membrane. d AFM image of the TpPa-1@PES@GF-72 membrane. e–h Cross-sectional SEM images of the TpPa-1@PES@GF-72 membrane in different regions. i, j EDS elemental mapping of the cross-section of the TpPa-1@PES@GF-72 membrane. k–p Cross-sectional HRTEM images of the TpPa-1@PES@GF-72 membrane at different magnifications. Photographic images of the TpPa-1@PES@GF-72 membrane before (q) and after (r) treatment in a DMF solution. Photographic images of the TpPa-1@PES@GF-72 membrane before (s) and after (t) detachment with tape.

The formation of the COF membrane was confirmed by X-ray diffraction (XRD) and attenuated total reflectance Fourier transform infrared spectroscopy (ATR-FTIR). XRD analysis revealed the gradual emergence of characteristic diffraction peaks at 4.7° and 28° (2θ) upon prolonged reaction, which is consistent with reported TpPa-1 COF crystalline structures28 (Supplementary Fig. 4a). Upon COF membrane formation, the broad diffraction peak of the original GF substrate in the 17–37° range was obscured by characteristic PES peaks, confirming complete coverage of the GF surface by the mixed matrix. ATR-FTIR spectroscopy further provided evidence for the formation of COF membrane. The characteristic C = O stretching vibration peak at 1637 cm−1, initially present in Tp@PES@GF membrane and attributable to Tp monomers, progressively diminished during the reaction, indicating complete monomer conversion28 (Supplementary Fig. 4b). The Si-O-Si peaks of GF at 785 cm−1 and 1016 cm−1 disappeared in the COF-MMMs, whereas the newly observed S = O peak at 1240 cm−1 confirmed successful coverage by the PES matrix.

The commercial GF substrate features a unique internal fiber network structure (Supplementary Fig. 5). Cross-sectional SEM characterization revealed that the prepared COF-MMMs exhibited an asymmetrical multilayered architecture (Fig. 2e), comprising a dense COF membrane on the top, a foam-like porous layer in the media, and a bottom layer with finger-like pore channels, similar to those of a pure PES membrane prepared by NIPS24 (Supplementary Fig. 6). X-ray photoelectron spectroscopy (XPS) provided direct evidence for the formation of a dense COF membrane on the top. The strong characteristic peak observed at 399.8 eV in the N 1 s spectrum was unambiguously assigned to nitrogen in the imine bond (-C = N-), directly confirming the successful condensation reaction between Tp and Pa-1. Meanwhile, the S 2p signal of PES at 168.0 eV was significantly attenuated in the TpPa-1@PES@GF membrane, demonstrating the formation of a dense COF covering membrane (Supplementary Fig. 7). The COF-MMMs prepared by the conventional blending method exhibited only an extremely weak nitrogen signal, providing further evidence that the COF membrane was successfully constructed on top of the PES substrate rather than through simple physical blending. Furthermore, the SEM results proved that the Tp@PES solution penetrated into the interior of the GF substrate (Fig. 2f) and formed a robust mechanical interlocking porous structure with the GF fibers (Fig. 2g). Notably, the walls of the finger-like pores in the polymer matrix were extensively covered with irregular COF structures (Fig. 2h), demonstrating that COF formation occurred not only at the top surface but also throughout the internal surface of the polymer matrix. The Tp@PES solution formed numerous foam-like pores within the GF substrate and filled the fiber pores (Fig. 2g). Energy dispersive X-ray spectroscopy (EDS) elemental mapping provided further evidence for the penetration of the COF mixed polymer matrix into the GF interior. Distinct N signals (representative of COF nanoparticles29) were detected on the cross-section of the TpPa-1@PES@GF-72 membrane (Fig. 2j), whereas negligible N signals were observed on both the surface and cross-section of the Tp@PES@GF membrane (Supplementary Fig. 8). Additionally, trace Si detected on the surface of both the Tp@PES@GF and TpPa-1@PES@GF-72 membranes confirmed the complete coverage of the COF mixed polymer matrix over the GF surface (Supplementary Fig. 8).

High-resolution transmission electron microscopy (HRTEM) characterization further revealed the asymmetrical multilayered architecture detected by SEM. A dense, ultrathin COF membrane with a thickness of 15–30 nm was fabricated on the surface of the PES skin layer (Fig. 2k). This was mainly related to the solvent exchange and aggregation-induced effects promoting the migration of Tp to the polymer surface during the NIPS process, where it underwent in situ polymerization and formed a dense, ultrathin COF membrane at the top (Fig. 2l, Supplementary Fig. 9). Scattered COF nanocrystals were also observed on the internal surface of the foam pores within the polymer matrix (Fig. 2l), which was attributable to concurrent solvent/non-solvent interdiffusion that induced localized Tp aggregation and polymerization along the pore walls. Remarkably, highly dispersed COF nanocrystals (4–8 nm) were identified adjacent to the COF membrane (Fig. 2m and Supplementary Fig. 10). Similarly, the same feature was observed in the TpPa-1@PES@GF-24 (Supplementary Fig. 11). The likely explanation was that the formation of a dense COF membrane on the surface slowed solvent/non-solvent diffusion, leading to an increased Tp concentration in the foam-pore layer region near the membrane surface, thereby triggering in situ polymerization and the formation of a nanocrystal-enriched layer. Notably, the inner surface of the finger-like pore walls also contained uniformly distributed COF nanocrystals (Fig. 2n, Supplementary Fig. 12). The slower solvent exchange rate within the finger-like pores of the polymer matrix led to the formation of dispersed COF nanocrystals, which were distributed along the inner surface of the dense skin layer of the polymer (Fig. 2o). Most importantly, the HRTEM images clearly revealed that the COF nanocrystals were enveloped by worm-like polymer chains, forming a seamless COF‒polymer interface with almost no intergranular defects (Fig. 2p). This can be attributed to the formation of hydrogen bonds between the ether (-O-) and sulfone (-SO2-) groups in the PES matrix and the hydroxyl (-OH) groups in the Tp precursors. This interaction was further corroborated by XPS analysis. The O 1 s spectrum revealed a significant intensity enhancement of the C–O characteristic peak near 533.0 eV, attributed to phenolic hydroxyl and ether oxygen groups, indicating the successful incorporation of Tp molecules into the PES matrix. More critically, the binding energy of the O = S = O component in the sulfone group shifted positively from 531.4 eV to 531.7 eV, while the binding energy difference between this component and the ether oxygen peak decreased from 1.93 eV to 1.45 eV. These systematic binding energy shifts and the observed energy gap narrowing can be attributed to the decreased electron density around oxygen atoms induced by hydrogen bonding formation, thereby providing direct evidence for the existence of hydrogen bond interactions between PES and Tp (Supplementary Fig. 7). The precursor Tp was completely soluble in the PES solution (Supplementary Fig. 13). During the NIPS process, Tp migrated outward with the diffusion of the NMP solvent. Since Tp was insoluble in the aqueous phase, it became entrapped and concentrated on the surface of the solidifying PES matrix, where it underwent in situ polymerization with Pa-1. This process occurred uniformly across the entire forming interface, effectively preventing the macroscopic aggregation often observed in physical blending methods18. Subsequent in situ IP creates covalent linkages between COF precursors within the crosslinked polymer network, establishing tight interfacial bonding between the COF and the polymer matrix that eliminates interface defects caused by poor compatibility between fillers and polymers via traditional methods. Additionally, the AFM force-distance curve (F-D) measurements revealed a notably high interfacial adhesion force of 2.863 nN, which can be ascribed to the abundant and effective weak van der Waals interactions between PES and the COF (Supplementary Fig. 14). Therefore, the NIPS-triggered in situ IP strategy first achieves a uniform distribution of COF nanocrystals in the polymer matrix to effectively avoid the particle aggregation issues of traditional methods. Second, a perfect COF‒polymer interface was successfully constructed through multiple synergistic interactions, forming a dense, defect‒free ultrathin COF membrane on the COF-MMMs surface for high separation selectivity.

An asymmetrical multilayered architecture analysis of the PES@GF and TpPa-1@PES@GF-72 membranes was performed on the basis of their pore size distributions. The significant difference between the average pore size (0.30 μm) and the most likely pore size (0.016 μm) of the PES@GF membrane indicated the formation of an asymmetrical multilayer structure consisting of foam-like and finger-like pores via NIPS (Supplementary Fig. 16a). The decrease in the average pore size of the TpPa-1@PES@GF-72 membrane to 0.2287 μm (Supplementary Fig. 16b) was caused by the presence of COF nanocrystals on both the topmost polymer surface of the COF-MMMs and on the PES matrix surface within the GF substrate, as confirmed by SEM and HRTEM observations. N2 adsorption‒desorption isotherms revealed that the specific surface areas of the GF, PES@GF and TpPa-1@PES@GF-72 membranes were 2.2319 m2/g, 7.1130 m2/g, and 9.8237 m2/g, respectively (Supplementary Fig. 17). The gradual increase in surface area provided clear evidence for the formation of COF nanocrystals on the PES surface within the GF substrate.

The COF-MMMs exhibited deformation resistance without crease formation under external stress (Fig. 2q). Remarkably, the structural integrity of the COF-MMMs was retained even after the PES matrix was completely dissolved by immersion in DMF, despite the occurrence of stress-induced increases (Fig. 2r). These results confirmed that the polymer matrix not only served as both a solid solvent and a polymerization platform but also provided physical support that enhanced membrane flexibility. For comparison, the COF-MMMs fabricated without the GF substrate were also treated with DMF to selectively remove PES. A complete and free-standing COF membrane was obtained (Supplementary Fig. 18). A prolonged reaction time enhanced membrane integrity, resulting in characteristics comparable to those of liquid‒liquid interfacial polymerized COF membranes (Supplementary Fig. 19). In addition, after the TpPa-1@PES@GF membrane was subjected to external damage tests, such as adhesive tape tearing (Fig. 2s-t), the structural integrity of the membrane layer remained intact. To quantitatively assess the functional integrity of the membrane layer after mechanical damage, we compared the gas separation performance before and after the adhesive tape tearing test. The H2/CO2 selectivity of the membrane after testing was measured at 74.3 (Supplementary Fig. 20). Although this reflects a reduction from the original value of 85.3, it remains significantly higher than that of the pristine PES@GF membrane (17.4) and the selectivity of the COF-MMMs fabricated via the conventional blending method (33.2). This indicates that the core separation layer remained intact under physical damage, and the minor performance variation likely stems from a slight increase in non-selective pores within the support layer. The above experimental results confirmed that the TpPa-1@PES@GF membrane had high mechanical strength and stability.

Gas-separation performance of COF-MMMs

The gas separation performance of PES@GF and COF-MMMs were evaluated using pure H2, CO2, N2, and CH4 at 298 K and a feed pressure of 1 bar. The PES@GF membrane showed a H2/CO2 selectivity of 17.4 with a H2 permeance of 4111 GPU (Fig. 3a). The gas separation mechanism of the PES membrane followed the Knudsen diffusion mechanism, achieving separation via gas‒wall collisions through nanoscale pore diffusion in the dense skin layer24,30. The PES matrix provided low-resistance transport pathways for rapid gas permeation, ensuring high gas flux. The COF-MMMs prepared by a NIPS-triggered in situ IP strategy exhibited improved gas selectivity and stable gas permeability. A prolonged polymerization time markedly increased the H2/CO2 selectivity of the TpPa-1@PES@GF-X membranes. Moreover, the CO2 permeability decreased sharply, primarily due to the selective adsorption of CO2 by the COF31. The introduction of COF led to the formation of more tortuous gas transport pathways and caused a decrease in permeability. However, the porous structure of PES provided fast gas permeation channels for the membrane, stabilizing the H2 permeance at approximately 2700 GPU. The TpPa-1@PES@GF-72 membrane achieved optimal separation performance after 72 h of reaction, with an H2/CO2 selectivity of 88.8 ± 2.46 and an H2 permeance of 2738 ± 58.02 GPU (1GPU = 3.35 × 10−10 mol m−2 s−1 Pa−1), exceeding the 2008 Robeson upper bound (Fig. 3a). The effect of the operating temperature on the H2/CO2 separation performance of TpPa-1@PES@GF-72 was investigated (Fig. 3b). The elevated gas permeance and reduced H2/CO2 selectivity as the temperature increased to 423 K could be attributed to the more intense molecular motion and weakened CO2 adsorption by the COF at elevated temperatures. Even at an elevated temperature of 423 K, the TpPa-1@PES@GF-72 membrane maintained a high H2/CO2 selectivity of 40.9 ± 0.92, surpassing the 2008 Robeson upper bound, along with H2 permeability of 4919 ± 180.46 GPU (Fig.3b). During the continuous heating test, the H2 and CO2 permeances of the TpPa-1@PES@GF-72 membrane increased sharply with increasing temperature, whereas the corresponding H2/CO2 selectivity decreased. The TpPa-1@PES@GF-72 membrane exhibited remarkable thermal stability, as evidenced by its nearly consistent separation performance when the temperature was cycled between 423 K and room temperature (Fig. 3c).

Fig. 3: Gas separation performances.
Fig. 3: Gas separation performances.
Full size image

a H2/CO2 separation performance of the PES@GF and TpPa-1@PES@GF-X membranes as a function of reaction time (Data are presented as mean values ± s.d. (standard deviation), n = 3). b H2/CO2 separation performance of the TpPa-1@PES@GF-72 membrane mixed at different temperatures. c Temperature‒gradient evaluation of the H2/CO2 separation performance of the TpPa-1@PES@GF-72 membrane mixed. d Gas separation schematic diagram of traditional COF-MMMs, where COF nanoparticles are distributed on the cross-section of the membrane. e Gas separation mechanism of COF-MMMs prepared via NIPS-triggered in situ IP strategy, where the COF membrane is distributed on the surface of the polymer matrix. f Long-term stability test of the TpPa-1@PES@GF-72 membrane under continuous operation. g Digital photograph of the scaled-up fabricated COF-MMMs. h Mixed H2/CO2 separation performance of 9 portions.

To demonstrate the superiority of the NIPS-triggered in situ IP strategy, we fabricated COF-MMMs via the conventional blending approach by admixing the presynthesized COF nanoparticles into the PES polymerization mixture (Fig. 3d). The TpPa-1/PES/GF membrane prepared by the traditional blending method exhibited a H2/CO2 selectivity of 33.2 and a H2 permeance of 3060 GPU (Supplementary Fig. 21, Supplementary Data 1). This inferior performance is likely attributable to the unavoidable aggregation of pre-synthesized COF nanoparticles and the poor interfacial adhesion, which lead to non-selective defects, as commonly encountered in traditional MMMs. And most COF nanoparticles are located at the cross-section (interior) of the COF-MMMs. Increasing the COF loading causes difficulties during fabrication because the high COF content significantly increases the viscosity of the doped solution, which cannot be cast into a film26. In contrast, NIPS-triggered in situ IP strategy ensures the in situ formation of highly dispersed COF nanocrystals and a seamless COF-polymer interface, which is the key to the 162.05% greater selectivity, with a mere 11.76% permeance reduction (Fig. 3e), exceeding the 2008 Robeson upper bound. Moreover, the TpPa-1@PES@GF-72 membrane exhibited operational stability during 40 h of continuous operation (Fig. 3f), with negligible fluctuations in H2 and CO2 permeances. The long-term stability can be attributed to the high chemical stability of the COF and the good COF‒polymer compatibility of the COF-MMMs, as well as the mechanical interlocking structure. We also explored the scaled-up fabrication of COF-MMMs (Fig. 3g), and the separation performance of each part was stable, demonstrating the potential of the NIPS-triggered in situ IP strategy for large-scale applications (Fig. 3h).

Gas-separation mechanisms of COF-MMMs

The gas separation mechanism was investigated through single gas permeation tests (H2, CO2, N2, and CH4) using the TpPa-1@PES@GF-72 membrane (Fig. 4a). The H2, N2, and CH4 permeances significantly surpassed the CO2 permeance, which was attributed to the flexibility of the framework and CO2 adsorption hindering diffusion through the COF channels. The ideal selectivities of H2/CO2, H2/N2, and H2/CH4 were 85.53, 59.57, and 52.57, respectively, significantly exceeding the Knudsen diffusion values (4.69, 3.73, and 2.83)9. The mixed-gas selectivities of 87.07, 57.63 and 54.83 showed minimal deviation from the ideal selectivity, revealing competitive adsorption among gas molecules within the COF framework. According to literature reports, CO2 adsorption can block the pores of COFs, and competitive adsorption occurs within the reduced pores, hindering the transport of H232. For COF TpPa-1 with a free pore size of approximately 18 Å28, the selective adsorption of CO2 molecules created steric hindrance that slowed subsequent rapid CO2 permeation. For smaller H2 molecules, the weak affinity of the COF for H2 and the effective pore size enabled faster diffusion, thereby improving selectivity. Further research revealed that the CH4/CO2 separation factor (4.0) of the mixed gas exceeded the ideal selectivity (3.77) owing to the stronger adsorption affinity of TpPa-1 for CO2. The synergistic mechanism of selective adsorption and competitive diffusion in H2/CO2 separation was confirmed33. In addition, the porous structure of the PES matrix also plays an important role in the separation process. Larger molecules experience greater transport resistance, whereas smaller molecules permeate more rapidly. The significant increase in the gas diffusion difference and synergistic selective adsorption of COFs jointly achieve a balance between the high permeability and high selectivity of membrane materials to meet the requirements of industrial gas separation (Fig. 4b)34. Moreover, the influence of feed pressure on the separation performance of the TpPa-1@PES@GF-72 membrane was also explored. Notably, the membrane’s ultra-high permeance presented a technical challenge for maintaining stable high feed pressures (>1 bar) due to rapid pressure equalization across the membrane. However, at a moderately elevated feed pressure of 0.02 MPa (0.2 bar), the H2 permeance surged to approximately 8000 GPU, underscoring the immense gas flux capability of the membrane. Concurrently, the H2/CO2 selectivity stabilized at around 15. The observed reduction in selectivity may be attributed to enhanced non-selective viscous flow and subtle structural adjustments under applied pressure. Nevertheless, the selectivity remained significantly above the Knudsen limit, confirming the persistence of selective transport mechanisms. The COF-MMMs achieved an optimal balance between ultra-high permeance and considerable selectivity, highlighting their promising potential for high-flux separation applications (Supplementary Fig. 22). Compared with the most advanced H2/CO2 separation membranes (Fig. 4b, Supplementary Data 1), the TpPa-1@PES@GF membrane exhibited reasonable separation performance. Furthermore, the reproducibility demonstrated by tests on ten independent membranes highlights the advantages of the NIPS-triggered in situ IP strategy in fabricating high-performance, highly stable COF-MMMs (Supplementary Table 1).

Fig. 4: Gas Separation Performance and Summary of Literature Data.
Fig. 4: Gas Separation Performance and Summary of Literature Data.
Full size image

a Kinetic diameter-dependent single-gas permeance through the TpPa-1@PES@GF-72 membrane and corresponding gas pair selectivity (Data are presented as mean values ± s.d. (standard deviation), n = 3). b Comparison of the separation performance of previously reported results with those of the TpPa-1@PES@GF membrane (detailed in Supplementary Data 1). The solid black line represents the 2008 upper bound for polymeric membranes for H2/CO2 separation41, with permeability converted to permeance assuming a 0.1 μm thick selective membrane.

The H2/CO2 separation mechanism of TpPa-1 was thoroughly studied through GCMC simulations. The simulated gas adsorption isotherms revealed adsorption energies of −20.71 kcal mol-1 for CO2 and −0.71 kcal mol−1 for H2 (Fig. 5a), demonstrating a significantly stronger CO2 adsorption capability for TpPa-1 than for H2. The adsorption energy of TpPa-1 for H2 essentially remained unchanged after CO2 adsorption equilibrium. The structure of TpPa-1 features topologically planar and ordered networks which undergo stacking along the third dimension, thereby generating 1D void channels35. This anisotropic alignment is beneficial for the effective transport of molecules along the stacked columns and pore channels36. During dynamic CO2 adsorption, the effective pore size of TpPa-1 decreased from an initial 16 Å to 13–16 Å (Fig. 5b) with a concomitant reduction in the interlayer spacing (Fig. 5d), thereby impeding rapid CO2 diffusion. In contrast, weakly adsorbed H2 molecules could still rapidly traverse the spatially constrained pore channels and reduce the interlayer spacing without obstruction, enabling highly efficient H2/CO2 separation. This separation mechanism achieved highly efficient H2/CO2 separation by combining Knudsen diffusion in the PES matrix, selective adsorption by the COF domains, and molecular sieving through the COF interlayer spacing. These three mechanisms are interrelated and work synergistically, collectively overcoming the “trade-off” effects in membrane separation processes, thereby enabling the COF-MMMs to simultaneously possess both high selectivity and high permeance. It should be pointed out that several other types of COFs have been recently synthesized via interfacial polymerization as well. Indeed, further studies on the fabrication of other COF-MMMs for gas separation are currently underway in our laboratory.

Fig. 5: Molecular simulation of gas adsorption behavior and structural dynamics in TpPa-1.
Fig. 5: Molecular simulation of gas adsorption behavior and structural dynamics in TpPa-1.
Full size image

a Simulated adsorption isotherms of H2 and CO2 on the TpPa-1. b Calculated pore size distribution of the TpPa-1 before and during CO2 adsorption. c Snapshots of the TpPa-1 structure: (left) without CO2 adsorption and (right) at CO2 adsorption equilibrium (color code: C, gray; H, white; O, red; N, blue). d Interlayer structural changes before CO2 adsorption and after adsorption equilibrium is reached.

Discussion

We developed a NIPS-triggered in situ IP strategy to successfully fabricate high-performance COF-MMMs. The uniqueness of this method lies in the use of the polymer matrix simultaneously as both a solid solvent and an interfacial polymerization platform, achieving controllable in situ growth of the COF membrane on the polymer matrix surface and effectively solving the problem of filler aggregation in traditional MMMs. The flexible polymer chains tightly adhere to the COF nanocrystals, forming a complete COF‒polymer interface that significantly improves the gas separation performance. The optimized COF-MMM achieved reasonable H2/CO2 selectivity of 88.8 ± 2.46, with a H2 permeance of 2738 ± 58.02 GPU (1 GPU = 3.35 × 10−10 mol m−2 s−1 Pa−1). The selectivity of H2/CO2 increased by 162.05% compared with that of the traditional blending method and surpassed the performance limits of similar materials. Moreover, the membrane maintained a high H2/CO2 selectivity of 39.8 even at 423 K, confirming its remarkable thermal stability. The COF-MMMs also exhibited long-term operational durability with stable performance throughout 40 h of continuous operation. A synergistic gas separation mechanism for COF-MMMs was proposed by combining gas separation experiments and grand canonical Monte Carlo (GCMC) simulations, including Knudsen diffusion in the PES matrix, selective adsorption by the COF domains, and molecular sieving through the COF interlayer spacing. This highly efficient fabrication strategy provides a facile technical pathway for developing high-performance COF-MMMs.

Methods

Materials

All the chemicals and reagents that are commercially available are used without purification. Polyether sulfone (PES, 80 mesh) was obtained from Dongguan Shuangfu Plastics Trading Co., Ltd., 1-methyl-2-pyrrolidone (NMP, Aladdin, AR), 1,3,5-triformylphloroglucinol (Tp, CoChemist, 95%), p-phenylenediamine (Pa-1, Macklin, AR), p-toluenesulfonic acid monohydrate (PTSA, Aladdin, AR), water (homemade), and GF supports with an open porosity of 94.5%, an average pore size of 4.68 μm (Supplementary Fig. 15) and a diameter of 25 mm (binder-free) were purchased from Shanghai Xinya Purification Equipment Co., Ltd. All gases with a minimum mole fraction of 99.99% were provided by Jinghua (Yinchuan) Gas Co., Ltd., including H2, CO2, N2, CH4 and Ar.

Synthesis of the PES solution

The PES solution was prepared by dissolving 1.5 g of PES granules in 8.5 g of NMP under stirring at room temperature for at least 24 h until a homogeneous solution formed. After that, the solution was kept at room temperature without stirring overnight to remove all the bubbles in the solution. Sealed for future use.

Synthesis of the PES membrane

PES membrane was prepared via the NIPS method. The PES solution was cast on the glass panel via a casting knife, followed by immersion in a water bath at room temperature for phase separation. The resulting membrane was thoroughly washed with water and stored in water before use.

Synthesis of the PES@GF membrane

The PES@GF membrane was prepared via the NIPS method. The GF substrate was secured on the spin coater, and the PES solution was spin-coated at 1000 rpm for 60 s. The coated GF membrane was then immersed in an aqueous solution and left undisturbed for 24 h. The resulting membrane was thoroughly rinsed with deionized water and stored in water prior to use. The prepared membrane was designated PES@GF membrane.

Synthesis of the Tp@PES@GF membrane

Tp@PES@GF membrane was prepared via NIPS method. The Tp@PES solution was prepared by dissolving 0.1 g of Tp in 10 g of PES solution under stirring at room temperature for at least 24 h until a homogeneous solution formed. The GF substrate was secured on the spin coater, and the Tp@PES solution was spin-coated at 1000 rpm for 60 s. The coated GF membrane was then immersed in an aqueous solution and left undisturbed for 24 h. The resulting membrane was thoroughly rinsed with deionized water and stored in water prior to use. The prepared membrane was designated Tp@PES@GF membrane.

Synthesis of the TpPa-1@PES-X membranes

TpPa-1@PES-X membranes were prepared via NIPS-triggered in situ IP. The Tp@PES solution was subsequently cast on the glass panel via a casting knife, followed by immersion in water containing a Pa-1 (97.4 mg) and PTSA (584.4 mg) bath at room temperature. The prepared membranes were abbreviated as TpPa-1@PES-X, where X denotes the polymerization time (e.g., 24 h, 48 h, 72 h, or 96 h). The resulting membranes were thoroughly washed with water and stored in water before use.

Synthesis of the TpPa-1@PES@GF-X membranes

TpPa-1@PES@GF-X membranes were prepared via NIPS-triggered in situ IP. The GF substrate was secured on the spin coater, and the Tp@PES solution was spin-coated at 1000 rpm for 60 s. The coated GF membrane was then immersed in an aqueous solution containing Pa-1 (97.4 mg) and PTSA (584.4 mg) and left undisturbed for varying durations to allow for polymerization. The resulting membrane was thoroughly rinsed with deionized water and stored in water prior to use. The prepared membranes were abbreviated as TpPa-1@PES@GF-X, where X denotes the polymerization time (e.g., 24 h, 48 h, 72 h, or 96 h).

Gas permeation experiments

Gas permeation tests were conducted using a custom-built evaluation system (Nanjing First Experiment Equipment Co., Ltd.). The system incorporated a Wicke‒Kallenbach permeation cell housed in a temperature-controlled oven (Supplementary Fig. 23). Membrane samples (TpPa-1@PES@GF-X) were sealed using perforated aluminum tape, exposing an effective area of 0.5024 cm2. Gas flows were regulated by mass flow controllers (MFCs, SevenStar, D08-4E). The composition of the permeate stream was analyzed by gas chromatography (Panna A91 plus, Changzhou Panna Instruments Co., Ltd.) and its flow rate was measured with a soap bubble flowmeter. A flame ionization detector with a methanizer quantified trace CO2, while H2 was detected via a thermal conductivity detector. The permeate side was at atmospheric pressure, with no applied transmembrane pressure drop. All reported data are the average of at least three measurements.

For the single gas permeation tests, the feed and the argon sweep gas flow rates were both set at 100 mL min−1.

Gas permeance (Pi, 1 GPU = 3.35 × 10−10 mol m−2 s−1 Pa−1) of gas “i” is calculated using Eq. (1).

$${P}_{i}=\frac{{N}_{i}}{A\varDelta {p}_{i}}$$
(1)

where Ni is the permeate rate of gas “i” (mol s−1), A is the effective membrane area (m2), and Δpi is the partial pressure difference of gas “i” (Pa).

Ideal selectivity (IS) is derived from Eq. (2).

$${{IS}}=\frac{{P}_{i}}{{P}_{j}}$$
(2)

where Pi and Pj are the permeances of gas “i” and gas “j”, respectively.

For equimolar (1:1) binary gas mixtures, the total feed flow was 100 mL min−1 with N2 as the sweep gas (100 mL min−1).

Permeance was calculated according to Eq. (1), and the separation factor (SFi/j) is determined using Eq. (3).

$${{SF}}_{i / j}=\frac{{Y}_{i} / {Y}_{j}}{{X}_{i} / {X}_{j}}$$
(3)

where Xi and Yi denote the molar concentration ratios of component “i” on the feed and permeate sides, respectively.

GCMC simulation

GCMC simulations were conducted with the Cassandra code37 to compute CO2 and H2 absorption under TpPa-1 nanoconfinement at 298 K and 101.325 kPa. The configurational bias Monte Carlo method was utilized to increase the efficiency of gas absorption. For all system sorption calculations, equilibration runs of 1 million steps were performed, followed by production runs of 1 million steps. The LJ and electrostatic interactions for the simulation box were truncated at 15 Å. Bond lengths were constrained to their nominal values. The Opls-aa force field38,39 was applied for TpPa-1. CO2 and H2 models were taken from Son et al. 40.