Precise molecular sieving in polycrystalline covalent organic framework membranes

Abstract

Covalent organic frameworks (COFs) unlock opportunities to unite rapid mass transport with molecular-level sieving through membranes. However, achieving long-range crystallographic order within these membranes remains a central challenge that limits their separation precision. Here, we report a monomer-driven strategy for the ambient-condition fabrication of COF polycrystalline membranes featuring definite lattice order and polyhedral textures, reminiscent of benchmark metal-organic framework counterparts. Using a rigid, π-conjugated pyrene-centered tetraamine, we enable nonepitaxial intergrowth of faceted COF crystals on a porous support, yielding a 350-nanometer-thick membrane with a record-high Brunauer-Emmett-Teller surface area of 2121 square meters per gram among COF separation membranes of comparable pore size. The resulting COF membrane preserves its crystallographic order and polyhedral texture even after 1 month of solvent exposure while uniquely combining increased surface stiffness with high tensile ductility. Notably, we demonstrate its superior liquid-phase molecular selectivity, outperforming COF analogs with considerably smaller pore apertures, and elucidate an unambiguous correlation between crystallographic order and permselectivity. This work positions crystallographic ordering as a guiding principle for COF membranes, opening more possibilities at the intersection of framework materials and separation technologies.

INTRODUCTION

Covalent organic frameworks (COFs) are a versatile class of porous, crystalline polymers constructed through the reticulation of organic building blocks via robust covalent bonds (1–3). Their modular synthesis and predictable topologies afford precise control over pore size, geometry, and functionality (4), leading to high surface areas, uniform and tunable nanopores, and exceptional chemical and thermal stability. These features underpin their broad applications in energy storage (5), catalysis (6), separation (7), and beyond (8).

COF membranes have garnered substantial attention as a platform for high-precision and energy-efficient separations (9, 10). Their ordered pore networks and chemical tunability enable the integration of multiple separation mechanisms, including size exclusion, charge repulsion, and host-guest interactions (11–13). To exploit these attributes, numerous membrane fabrication strategies have been developed, such as interfacial crystallization (14), solvothermal growth (15), electrosynthesis (16, 17), and nanosheet assembly (18, 19). These approaches have enabled COF membranes to deliver unconventional performance in water desalination (20, 21), organic solvent nanofiltration (22–24), and gas separations (25, 26).

A key determinant of COF membrane performance is the formation of thin, continuous, and selective layers, wherein the framework structure dictates permselectivity (27). Although considerable efforts have been devoted to tuning pore size (28), morphology (29), orientation (30, 31), and functionality (32), the crystallinity of COF layers, which is one of the defining characteristics distinguishing COFs from amorphous porous polymers, remains critically underexplored. Unlike analogous metal-organic framework (MOF) membranes, which often exhibit long-range lattice order and faceted morphologies (33–35), most state-of-the-art COF membranes show ambiguous x-ray diffraction features and smooth morphologies reminiscent of semicrystalline or amorphous polymers (Fig. 1) (36). This lack of crystallographic order disrupts pore periodicity and directional mass transport, ultimately limiting molecular sieving accuracy and membrane usability. Recent advances in the growth of COF single crystals have deepened our understanding of framework crystallization (2, 3, 37), but translating crystallographic order into practical membrane architecture remains unresolved. Addressing this gap is essential for unlocking the potential of organic framework materials in membrane science and addressing persistent concerns on COF membrane technologies (36, 38, 39).

Fig. 1. Illustration of composite separation membranes with selective layers built of conventional polymers, conventional COFs, and polyhedral COF crystals.

Here, we report the fabrication of polycrystalline COF membranes with crystallographic order and polyhedral crystal textures, achieved via ambient-condition growth using a rigid, π-conjugated pyrene-based tetraamine. This building block promotes the nonepitaxial intergrowth of faceted COF crystals on a hydrolyzed polyacrylonitrile (HPAN) substrate, forming a thin selective layer with a record Brunauer-Emmett-Teller (BET) surface area of 2121 m² g⁻¹ among COF membranes with comparable pore sizes. This membrane maintains its structural integrity and crystallinity after prolonged exposure to solvents, exhibiting a unique combination of mechanical stiffness and flexibility. The polycrystalline COF membrane shows exceptional molecular selectivity, surpassing COF membranes with much smaller pore apertures, and establishes a clear correlation between crystallinity and permselectivity. These findings position crystallographic ordering as a central principle for designing COF membranes toward precise separations.

RESULTS

Fabrication of polycrystalline Py-1P membranes

The rigidity (40) and π-conjugation (41) of building blocks have been recognized as key determinants governing the morphology and crystallinity of COFs. Guided by this principle, we examined a series of four-connected, square tetraamines, including 4,4′,4″,4‴-(ethene-1,1,2,2-tetrayl)tetraaniline (ETTA), 4′,5′-bis(4-aminophenyl)-[1,1′:2′,1″-terphenyl]-4,4″-diamine (BATD), 5′,5″-bis(4-aminophenyl)-[1,1′:3′,1″:3″,1‴-quaterphenyl]-4,4‴-diamine (BAQD), and 1,3,6,8-tetrakis(4-aminophenyl)pyrene (TAPPy) (Fig. 2A). To quantify their conformational rigidity, we calculated the ratio of rotatable to total covalent bonds, establishing the trend ETTA > BAQD > BATD > TAPPy (fig. S1), indicative of increasing molecular rigidity. This trend is corroborated by thermogravimetric analysis (TGA), which shows a corresponding increase in decomposition temperatures, with TAPPy exhibiting the highest thermal stability at 450°C (Fig. 2B). Variable-temperature ¹H nuclear magnetic resonance (NMR) spectra provided additional insights into their structural rigidity. At 100°C, the aromatic protons on the aniline-substituted phenyl ring of ETTA exhibited a more pronounced downfield shift than those of TAPPy (Fig. 2C), suggesting that TAPPy affords a more rigid conformation in solution. To assess their π-conjugation–driven interactions, we performed density functional theory (DFT) simulations to determine interlayer binding energies. The interlayer binding energies become increasingly negative from ETTA to BAQD and TAPPy (Fig. 2D), indicating progressively stronger π-π stacking interactions that are conducive to ordered framework assembly (41). To correlate these structural features with microscopic film structures, we synthesized COF films on silicon wafers via an acetic acid–catalyzed Schiff-base reaction between the tetraamines and terephthalaldehyde (TPA). Fourier transform infrared (FTIR) spectroscopy confirmed the presence of imine linkages in all films (fig. S2). Field-emission scanning electron microscopy (SEM) revealed a clear morphological evolution from planar sheets in the ETTA-TPA film to well-defined faceted textures in the BAQD-TPA and TAPPy-TPA (termed Py-1P) films (Fig. 2, E to H). Consistent with this morphological change, powder x-ray diffraction (PXRD) patterns further corroborate the progressive increase in film crystallinity (fig. S3). This structural progression recognizes the role of elevating monomer rigidity and π-conjugation in directing the growth of COF films with polyhedral crystal textures. The formation of such polyhedral texture also relies on the use of dichloromethane (DCM), as evidenced by contrast experiments with other organic solvents (fig. S4). This result is likely associated with the suitable monomer solubility in DCM, which governs interfacial diffusion and nucleation-growth kinetics.

Fig. 2. Monomer-driven structures of COF films.

(A) Molecular structures and (B) TGA curves of the tetraamines. (C) Variable-temperature ¹H NMR spectra of ETTA and TAPPy. ppm, parts per million. (D) Interlayer interaction energies of the tetraamines. Surface SEM images of the (E) ETTA-TPA, (F) BATD-TPA, (G) BAQD-TPA, and (H) Py-1P films grown on silicon wafers.

Given the polyhedral crystal texture and intergrown architecture of the Py-1P film, we next pursued its fabrication as a separation membrane, leveraging its intrinsic high-order pores as molecular sieving channels (Fig. 3A). The separation membrane was prepared using a biphasic strategy (42), wherein an aqueous solution of acetic acid (6 M) was overlaid atop a DCM phase containing TAPPy and TPA (Fig. 3B). Slow diffusion of acetic acid across the interface modulated nucleation and growth kinetics, enabling the formation of Py-1P polycrystals in the organic phase (fig. S5). A porous HPAN membrane was then used as the substrate for the growth of Py-1P selective layers. The formation of polycrystalline Py-1P membranes proceeds via nucleation followed by crystal growth, as revealed by time-dependent SEM imaging (fig. S6). Hydrolysis of cyano groups on the pristine polyacrylonitrile (PAN) substrate yielded carboxyl functionalities that acted as nucleation sites (fig. S7), which are decisive for the intergrowth of Py-1P crystals to produce a continuous membrane. A control experiment showed that unmodified PAN yielded only sparse, discontinuous Py-1P crystals due to insufficient nucleation sites (fig. S8), akin to nucleation-dominated growth of polycrystalline MOF membranes (43).

Fig. 3. Fabrication of the Py-1P polycrystalline membrane.

(A) Schematic of the chemical structures and the condensation between TAPPy and TPA to form Py-1P polycrystals. (B) Schematic of the biphasic fabrication of the Py-1P membrane on a porous HPAN substrate.

Structural analysis of polycrystalline Py-1P membranes

The resulting composite membranes were termed Py-1P-x-y/HPAN, where x and y represent the molarity (M) of acetic acid and fabrication time (hours), respectively. The Py-1P-6-72/HPAN membrane was comprehensively characterized to reveal its structural attributes. The attenuated total reflectance FTIR (ATR-FTIR) spectrum of the HPAN substrate exhibits characteristic -OH stretching vibrations at 3355 cm⁻¹ (Fig. 4A), matching the hydrolysis of the PAN backbone. Upon the growth of Py-1P crystals, the ATR-FTIR spectrum of the Py-1P-6-72/HPAN membrane presents characteristic stretching vibrations from imine bonds at 1628 cm⁻¹, providing evidence for the formation of imine-linked frameworks. X-ray photoelectron spectroscopy (XPS) and solid-state ¹³C NMR analyses further validate the growth of Py-1P frameworks (figs. S9 and S10). PXRD elucidates the structural evolution during membrane growth, revealing improved crystallographic order as fabrication time increases (fig. S11). The HPAN substrate exhibited broad diffraction peaks at 2-theta positions of 17.5°, 22.5°, and 25.7° (Fig. 4B), which are attributed to the short-range ordered packing of PAN and polyethylene terephthalate chains (fig. S12). The Py-1P-6-72/HPAN membrane showed sharp diffraction peaks at 3.6° and 7.4°, corresponding to the (110) and (220) lattice planes of Py-1P, respectively (44). Notably, the (110) diffraction peak yields a full width at half maximum of ∼0.24°, surpassing the peak sharpness of the reported crystalline COF films (45). Higher-order diffraction peaks from the (200), (310), and (330) planes of Py-1P were also observed (inset in Fig. 4B). The PXRD pattern of the support-free Py-1P-6-72 membrane closely matches that of Py-1P-6-72 polycrystals (figs. S13 and S14). Unlike most COF membranes, which show weak or poorly defined COF diffraction characteristics (46), our polyhedral Py-1P membrane demonstrates unambiguous crystallographic order.

Fig. 4. Characterization of the Py-1P polycrystalline membrane.

(A) ATR-FTIR spectra of the PAN, HPAN, and Py-1P-6-72/HPAN membranes. (B) PXRD patterns of the HPAN and Py-1P-6-72/HPAN membranes. The inset shows a magnified view. (C) 77 K N₂ sorption isotherms of the Py-1P-6-72 membrane. The inset shows the pore size distribution. STP, standard temperature and pressure. (D) BET surface areas of the reported COF separation membranes and the Py-1P-6-72 membrane as a function of pore diameter. Surface (E) SEM and (F) AFM images, (G) GIWAXS pattern, and (H) false-color cross-sectional SEM image of the Py-1P-6-72/HPAN membrane. (I) HRTEM image of the Py-1P-6-72 membrane. The inset shows the FFT image.

The porous nature of the membrane was probed by N₂ sorption at 77 K. The Py-1P-6-72 polycrystalline membrane exhibited significant N₂ uptake at low relative pressures (Fig. 4C), confirming the presence of accessible nanopores analogous to Py-1P polycrystals (fig. S15). DFT model fitting determined a pore diameter of ∼2.2 nm (inset in Fig. 4C), consistent with the theoretical value of Py-1P (44). The BET surface area of this Py-1P polycrystalline membrane reaches ∼2121 m² g⁻¹, markedly surpassing that of state-of-the-art COF separation membranes with similar pore sizes (Fig. 4D and table S1). These results confirm that the developed Py-1P membrane exhibits remarkable crystallinity and porosity, comparable to those of MOF polycrystalline membranes (33).

Morphological and microstructural analyses by a combination of microscopy techniques delineated the membrane architecture. Figure 4E shows the surface SEM image of the Py-1P-6-72/HPAN membrane, with a unique polyhedral crystal texture analogous to those observed in the well-known UiO-66 and ZIF-8 membranes (33, 47). Py-1P crystals with a lateral size of ∼200 nm are densely intergrown, forming a continuous, defect-free selective layer in sharp contrast to the smooth HPAN substrate (fig. S16). By comparison, Py-1P-3-72/HPAN and Py-1P-12-72/HPAN membranes show discontinuous surfaces despite their high crystallinity (fig. S17). This indicates that an appropriate acetic acid concentration is critical for balancing nucleation and growth kinetics, thereby promoting the formation of continuous crystal membranes. Atomic force microscopy (AFM) imaging revealed a root mean square roughness (R_q) of ∼22.6 nm and a roughness factor (r) of ∼1.24 (Fig. 4F and fig. S18), reflecting an increased surface area relative to the projected area. The corresponding AFM phase image further showed uniform contrast across individual crystals (fig. S19), suggesting a single-phase crystalline structure. To examine crystallographic orientation, we performed grazing incidence wide-angle x-ray scattering (GIWAXS) tests on the Py-1P-6-72/HPAN membrane. The crystalline selective layer showed a sharp diffraction ring at q_xy = 0.27 Å⁻¹, corresponding to the (110) reflection of Py-1P, whereas the HPAN substrate only exhibited a diffuse halo at q_z = 1.3 to 1.8 Å⁻¹ (Fig. 4G). The ring feature in the GIWAXS pattern signifies the nonepitaxial growth of Py-1P crystals within the membrane, indicating an isotropic distribution without a preferred crystallographic orientation. Cross-sectional SEM confirmed tight intergrowth among Py-1P crystals and seamless integration with the HPAN substrate (Fig. 4H), attributed to the compatible polymeric nature of both components. The Py-1P selective layer exhibited a uniform thickness of ∼350 nm (fig. S20), which is thinner than typical inorganic crystal membranes (48) and comparable with conventional COF membranes (20, 21). High-resolution transmission electron microscopy (HRTEM) corroborated the high-quality polycrystalline nature of the Py-1P crystal layer, with clear lattice fringes observed throughout the image (Fig. 4I and fig. S21). Fast Fourier transform (FFT) analysis of the HRTEM data reveals distinct ring patterns corresponding to the (110), (220), and (330) planes of Py-1P (inset in Fig. 4I), which align with the PXRD and GIWAXS results and confirm the robust crystallinity of the Py-1P-6-72 polycrystalline membrane. These structural features set a previously unknown benchmark for COF-based separation membranes and are anticipated to facilitate high-precision molecular separations.

The solvent stability of the Py-1P-6-72/HPAN polycrystalline membrane was evaluated by immersing it in water and ethanol for 30 days. N₂ sorption analyses at 77 K confirmed that both porosity and pore size distribution remained largely intact (Fig. 5A). PXRD patterns further revealed that the membrane retained significant diffraction peaks after solvent exposure (Fig. 5B), indicating that the crystallographic order was well preserved. This finding is corroborated by SEM images (fig. S22), which display intact polyhedral crystal textures. The polycrystalline membrane also exhibited high-pressure resistance, retaining its crystallinity and polyhedral morphology after treatment at transmembrane pressures of 1 to 5 bar (fig. S23). Together, these results demonstrate the excellent structural robustness of the polycrystalline membrane for liquid-phase separations. Moreover, TGA demonstrated its outstanding thermal stability, with decomposition temperatures up to 415°C in air and 450°C in N₂ (fig. S24). This pronounced thermal stability, derived from its tetraamine building block, exceeds that of previously reported COF membranes (49). The surface mechanical robustness of our membrane was examined using nanoindentation (fig. S25). Figure 5C shows the nanoindentation load-displacement curves of the HPAN and Py-1P-6-72/HPAN membranes. The HPAN substrate exhibited a surface hardness (H) of 2.7 MPa and an elastic modulus (E) of 65.4 MPa. After Py-1P crystal growth, the composite membrane displayed an enhanced H value of 4.5 MPa while maintaining a similar E value (∼62.0 MPa), suggesting improved surface stiffness without largely compromising elasticity (50). Uniaxial tensile testing was conducted to capture the full stress-strain response (fig. S26). Figure 5D reveals a clear transition from elastic to plastic deformation, followed by fracture. Notably, the Py-1P-6-72/HPAN membrane exhibited a slightly higher fracture strain of 15.5% compared to the HPAN substrate (12.3%). This indicates enhanced mechanical resilience under tensile stress without loss of ductility (51). These results reveal that Py-1P crystals impart a desirable balance between surface stiffness and bulk flexibility (52), which is promising for pressure-driven separation processes where membranes must endure mechanical stress while maintaining operational resilience.

Fig. 5. Physicochemical characteristics of the Py-1P polycrystalline membrane.

(A) 77 K N₂ sorption isotherms and (B) PXRD patterns of the membrane after immersion in ethanol and water for 30 days. The inset in (A) shows the pore size distributions. a.u., arbitrary units. (C) Load-displacement and (D) stress-strain curves of the HPAN and Py-1P-6-72/HPAN membranes. (E) Dynamic water contact angles and (F) zeta potential curves of the HPAN and Py-1P-6-72/HPAN membranes.

Next, surface wettability and charge properties, which are critical for liquid-phase separations, were characterized by water contact angle and zeta potential measurements. Figure 5E depicts dynamic water contact angles of the HPAN and Py-1P-6-72/HPAN membranes. The Py-1P-6-72/HPAN membrane exhibited a relatively low initial water contact angle of ∼53°, which is significantly lower than that of typical COF membranes with similar aromatic backbones (22). The water contact angle decreased markedly over time, reflecting rapid water penetration into the porous polyhedral structure and consistent with the Wenzel model, where surface roughness intensifies inherent wettability (fig. S27) (53). Zeta potential results showed that the membrane surface is negatively charged, with a value of −8.7 mV at pH = 7.1 (Fig. 5F), potentially facilitating molecular separation through electrostatic repulsion in addition to size exclusion.

Performance evaluation of polycrystalline Py-1P membranes

To elucidate solvent transport behavior in the Py-1P-6-72 polycrystalline membrane, both solvent vapor sorption and liquid permeation tests were performed. As shown in Fig. 6A, the membrane exhibited similar saturation vapor uptakes of ∼75% for acetone and butanol at P/P₀ = 0.95, indicating comparable overall adsorption capacities due to the abundance of accessible pores. At low relative pressures, acetone uptake increased sharply, attributed to its smaller kinetic diameter and higher vapor pressure, which enabled faster diffusion and stronger interactions with the pore walls. However, butanol exhibited a slower, more gradual adsorption profile due to its larger size and viscosity. Pronounced hysteresis in the butanol isotherm points to restricted diffusion, confinement effects, and capillary condensation within the nanopores. Consistently, liquid permeation showed that acetone permeance exceeded that of butanol by over an order of magnitude (Fig. 6B). In addition, a clear inverse relationship was observed between solvent viscosity and permeance across a range of solvents, consistent with laminar flow behavior through the Py-1P cylindrical channels described by the Hagen-Poiseuille law (54).

Fig. 6. Separation performance of the Py-1P polycrystalline membrane.

(A) Vapor uptakes of acetone and butanol by the Py-1P-6-72 polycrystalline membrane. (B) Pure solvent permeance as a function of viscosity. (C) Molecular rejection curves of the membrane in water and ethanol. (D) Ultraviolet-visible absorption spectra for the separation of an MO/NP mixture in water. (E) Schematic illustration of the molecular sieving by Py-1P. (F) MWCO values of COF separation membranes as a function of pore size. (G) PXRD patterns of the Py-1P-6-72 and Py-1P-0.5-72 membranes. (H) Water permeance and MWCO of the Py-1P-6-72 and Py-1P-0.5-72 membranes.

The Py-1P-6-72/HPAN polycrystalline membrane delivered permeances of ∼54.6 and ∼45.0 liters m⁻² hour⁻¹ bar⁻¹ for water and ethanol, respectively. Molecular separation performance was assessed via rejection tests with solutes ranging from 139 to 961 g mol⁻¹ (fig. S28). The rejection profiles determined molecular weight cutoff (MWCO) values of ∼242, ∼760, and ∼900 g mol⁻¹ in water, ethanol, and methanol, respectively (Fig. 6C and figs. S29 to S31). The steeper rejection profile in water than in ethanol indicates that aqueous solute transport is primarily governed by electrostatic effects, with an essential contribution from size sieving within the negatively charged, well-ordered Py-1P channels (16). Motivated by the disparate rejection rates of methyl orange (MO; molecular weight = 327.3 g mol⁻¹) and nitrophenol (NP; molecular weight = 139.1 g mol⁻¹) in water, we conducted a binary sieving experiment. Ultraviolet-visible absorption spectra in Fig. 6D reveal that the maximum absorption peaks of MO and NP appear at 463 and 318 nm, respectively. Although both MO and NP are present in the feed spectrum, only NP appears in the filtrate spectrum. This spectral result suggests that MO is largely rejected by our membrane, whereas most NP passes through with water, as illustrated in Fig. 6E. The high molecular selectivity of our membrane was further evidenced by the selective separation of mordant orange 1 (MO1; molecular weight = 287.2 g mol⁻¹) from NP (fig. S32). Comparative analysis reveals that the polycrystalline Py-1P-6-72 membrane exhibits a lower MWCO value than previously reported COF membranes, despite its much larger nominal pore size (Fig. 6F and table S2). Given that most COF membranes feature pores with similar negative charges, our findings demonstrate that optimizing the crystallographic order of COF selective layers is an effective strategy for improving the precision of molecular separation in aqueous media.

Crystallinity-performance correlation of Py-1P membranes

Elucidating the role of crystallinity in COF membrane performance is critical but remains unresolved. To probe this structure-performance relationship, we deliberately reduced the crystallinity of the Py-1P selective layer by preparing a Py-1P-0.5-72/HPAN membrane under suboptimal fabrication conditions. This membrane exhibited a morphology similar to that of conventional COF membranes and lacked polyhedral crystal textures (fig. S33). PXRD patterns in Fig. 6G differentiate the crystallinity of Py-1P-6-72/HPAN and Py-1P-0.5-72/HPAN membranes, with the latter showing significantly attenuated diffraction features indicative of diminished crystallographic order. This structural deficiency translated into compromised separation performance. The Py-1P-0.5-72/HPAN membrane exhibited reduced water permeance of ∼22.1 liters m⁻² hour⁻¹ bar⁻¹, less than half that of the Py-1P-6-72 membrane. This decline in permeability likely results from a reduced density of effective mass-transport channels, as corroborated by 77 K N₂ sorption analysis (fig. S34). Moreover, this lower crystallinity also impaired molecular selectivity. The Py-1P-0.5-72/HPAN membrane showed compromised rejection of solutes with molecular weights below 600 g mol⁻¹, yielding an MWCO value of ∼620 g mol⁻¹ (Fig. 6H and figs. S35 and S36). This MWCO value substantially exceeds that of the Py-1P-6-72/HPAN membrane, suggesting a considerable deterioration in molecular selectivity. These findings demonstrate that the absence of a well-ordered pore network not only hinders water permeance but also compromises molecular selectivity, highlighting the fundamental importance of crystallographic order in governing the separation performance of COF-based membranes.

DISCUSSION

We have developed MOF-like polyhedral COF crystal membranes via ambient-condition condensation of a rigid, π-conjugated pyrene-centered tetraamine (TAPPy) with TPA, achieving precise molecular sieving in a thin, well-faceted Py-1P polycrystalline membrane. Departing from conventional amorphous or nontextured COF analogs, our Py-1P polycrystalline membrane delivers a record-high BET surface area among COF membranes with similar pore sizes, without the need for harsh solvothermal conditions. The membrane also exhibits permanent porosity, pronounced crystallinity, and excellent chemical robustness, retaining its structural and sorption characteristics after prolonged exposure to solvents. Notably, the isotropic intergrowth of Py-1P crystals on the HPAN substrate enhances surface stiffness while preserving tensile ductility. These integrated features result in exceptional molecular selectivity, even surpassing that of COF membranes with substantially smaller pore apertures. We further establish a clear correlation between crystallographic order and separation performance, demonstrating that high lattice order synergistically enhances both permeability and selectivity. By bridging framework chemistry and membrane science, this work sets a compelling design paradigm that leverages the polycrystalline architecture of COFs for precise membrane separations. Ultimately, with the rapid progress of single-crystal COFs, membranes derived from atomically ordered COF single crystals are expected to deliver unparalleled separation performance.

MATERIALS AND METHODS

Fabrication of COF films on silicon wafers

The fabrication of the Py-1P film was used as a representative procedure. TAPPy (11.3 mg, 20 μmol) and TPA (5.4 mg, 40 μmol) were dissolved in DCM (10 ml) by ultrasonication at 50°C to form a homogeneous organic solution. After cooling to room temperature, the organic solution was transferred to a glass bottle (3.5 cm in diameter) preloaded with a cleaned silicon wafer. An equal volume of aqueous acetic acid solution (6 M, 10 ml) was then gently added to form an organic-aqueous biphasic interface. The bottle was sealed and left undisturbed at room temperature for 72 hours to allow film growth on the silicon wafer. After the reaction, the resulting Py-1P film was washed with acetone and ethanol and stored in ethanol for further use. Other COF films were synthesized using the same procedure with the same molar amount (20 μmol) of the corresponding tetraamine monomers.

Fabrication of Py-1P/HPAN separation membranes

Before membrane fabrication, glass bottles (3.5 cm in diameter) were pretreated by immersion in an alkaline bath overnight and then thoroughly washed with deionized water and dried. As a representative example, the Py-1P-6-72/HPAN membrane was synthesized as follows. TAPPy (11.3 mg, 20 μmol) and TPA (5.4 mg, 40 μmol) were dissolved in DCM (10 ml) by ultrasonication at 50°C to form a homogeneous organic solution. After cooling to room temperature, the organic solution was transferred to the pretreated glass bottle, which contained a piece of HPAN substrate placed at the bottom. An equal volume of aqueous acetic acid solution (6 M, 10 ml) was then gently added to form an organic-aqueous biphasic interface. The bottle was sealed and left undisturbed at room temperature for 72 hours to allow membrane growth. After the reaction, the resulting Py-1P-6-72/HPAN membrane was removed, washed sequentially with acetone and ethanol, and stored in ethanol for further use. Py-1P polycrystals formed in the bulk solution were collected by centrifugation, washed with ethanol, tetrahydrofuran, and hexane, and stored in hexane. Polyhedral Py-1P films grown on the inner walls of the glass bottle were also collected, washed with the same solvents, and stored in hexane. For the fabrication of Py-01P-x-72/HPAN membranes, the molarity of acetic acid was adjusted to 0.5, 3, or 12 M, whereas all other procedures remained identical.

Low-dose HRTEM and iDPC-STEM tests

Low-dose HRTEM imaging was conducted on a double aberration-corrected Thermo Fisher Scientific Spectra 300 TEM with a Gatan K3 camera at 300 kV. Using counting and correlated double sampling mode of K3 camera, 60 individual image frames were collected with a total exposure of 4 s. These frames were aligned and superimposed using the Gatan Digital Micrograph software to enhance the signal-to-noise ratio and correct for sample drift during exposure. The total accumulated electron doses for the drift-corrected images were controlled within the range of 10 to 20 e⁻ Å⁻² to minimize beam damage to the samples. iDPC-STEM imaging was performed at 200 kV using a segmented dark-field detector, with a convergence semiangle of 14.9 mrad and collection angles of 16 to 61 mrad. The probe current was about 1 pA. A high-pass filter was used to reduce low-frequency noise.

Characterizations

SEM was performed using a JEOL JSM-7610FPlus instrument at an acceleration voltage of 5 kV. Before imaging, samples were sputter coated with a thin layer of platinum using a Cressington 208HR coater (20 mA, 50 s) to minimize charging. AFM was conducted on a Bruker Dimension Icon microscope in tapping mode. PXRD patterns were collected using a Rigaku MiniFlex 600 powder x-ray diffractometer with Cu Kα radiation (λ = 1.5406 Å). It is noted that the membrane was affixed to a silicon wafer using double-sided tape and placed on the PXRD sample stage, ensuring that the COF layer was level with the stage surface. N₂ sorption tests were performed at 77 K using a Micromeritics ASAP 2020 surface area and porosity analyzer. Samples were degassed at 120°C for about 15 hours to reach a high vacuum degree of 4 μmHg before measurements. FTIR spectra were recorded on a Bruker Vertex 70 spectrometer using the ATR method. Solvent vapor sorption isotherms were tested using a DVS Discovery vapor gravimetric analyzer (Surface Measurement Systems). Solid-state ¹³C NMR spectroscopy was performed on a Bruker Avance 400 MHz NMR spectrometer (DRX400). Variable-temperature ¹H NMR spectroscopy was conducted with a Bruker Avance NEO 500 MHz NMR spectrometer equipped with a 5-mm TBI probe. GIWAXS experiments were conducted on a Xeuss 3.0 SAXS/WAXS system using a Cu Kα source with a Dectris Eiger2R 1M detector. XPS was performed on a Kratos Axis Ultra DLD x-ray photoelectron spectrometer with an excitation source of Al Kα = 1486.71 eV. TGA was performed using a Shimadzu DTG-60AH instrument. Samples were heated from room temperature to 800°C at 10°C min⁻¹ under a constant flow of N₂ or air at 50 ml min⁻¹. Water contact angles of membranes were recorded using a contact angle goniometer (DropMeter A100, Maist). Zeta potentials of membranes were tested using an electrokinetic analyzer (Surpass 3 Anton Paar, Austria). Mechanical properties of membranes were measured using a nanoindentation system (Bruker, Hysitron TI 980). Stress-strain curves of membranes were tested using a fiber elongation meter (XQ-IC).

Supplementary Materials

This PDF file includes:

Supplementary Text

Figs. S1 to S36

Tables S1 and S2

References