Controllable synthesis of a chemically stable molecular sieving nanofilm for highly efficient organic solvent nanofiltration

A chemically stable molecular sieving nanofilm with Janus pore structure was synthesized with high permeability in both polar and non-polar solvent.


Introduction
Organic solvent nanoltration (OSN) has attracted increasing interest in a wide range of applications in chemical processes, such as solvent recycling, concentration and solute extraction. [1][2][3] The OSN membrane can effectively fractionate solutes with a molecular weight (MW) between 200 and 1000 Da from organic liquids. [2][3][4] This requires the membrane to be tolerant to harsh organic media in a complex background (strong acid/ base, high temperature, aggressive solvent, and polar/nonpolar solvents). [4][5][6] Thus, membranes with a versatile separation capability and a strong chemical resistance are highly desirable for organic solvent separations.
Thin lm composite (TFC) membranes are typical organic solvent nanoltration membranes, consisting of a selective layer, a polymeric substrate and a non-woven fabric support. 7 Due to their stability in organic solvents and superior separation efficiencies as well as the designable pore architecture of the skin selective layer, TFC membranes are preferred. 8 However, conventional polyamide membranes developed from monomers of m-phenylenediamine (MPD) and trimesoyl chloride (TMC) through interfacial polymerization (IP) are particularly unsuitable for non-polar solvents, such as n-hexane, due to their highly cross-linked chemical structure and intrinsic hydrophilicity prohibiting solvent transport. 9,10 In applications, organic solvent systems are sophisticated as they contain both polar and non-polar solvents. 4,10 Currently, most OSN membranes are only suitable for either polar or non-polar organic solvents. 8,[10][11][12][13] Incorporation of uorine and silicon in the polyamide top layer can effectively endow the membrane with hydrophobic properties. 9 However, the graing barrier layer inevitably decreases the solvent permeability due to the resulting higher resistance. Furthermore, the dense structure of the top selective layer still poses a challenge. Recently, molecule-level design strategies have been shown to be promising in manipulating the pore structure and its physicochemical properties towards highly permeable OSN membranes. 5,10-12,14-16 For example, Chung et al. constructed Janus pathways with both hydrophobic and hydrophilic channels using cyclodextrins (CDs) as building blocks. 10 The hydrophobic inner cavities of CDs facilitate transport of non-polar solvent molecules and the hydrophilic outer space provides superhighways for polar solvent molecules. However, the large pore size of the CDs could only enable membrane rejection of large solute molecules. Liang et al. prepared conjugated microporous polymer (CMP) membranes via a surface-initiated polymerization strategy through C-C coupling reactions. 14 The rigid backbones of the CMP enable ultra-fast organic solvent nanoltration. Nevertheless, the complex molecular manipulation poses challenges in industrial application. A facile method to construct a well-dened pore structure may open up new opportunities to design advanced OSN membranes for ltration of both polar and non-polar solvents. In addition, it is still a formidable challenge to achieve a membrane possessing high chemical and structural stability under harsh conditions. Recently, Liu et al. prepared a novel nanoltration membrane by blending m-xylylenediamine (m-XDA) and polyethyleneimine as the aqueous co-monomer. 17 The addition of m-XDA endows the NF membrane with good chlorine-tolerant properties. The methylene of the m-XDA reduces the activity of amide bonds and improves the chemical resistance of the thin lm.
Here we propose a new synthetic approach to designing molecular sieving nanolms with Janus pore structures. An aromatic diamine with an "elongated arm" of the methylene moiety was used as the aqueous monomer for the interfacial polymerization. The methylene moiety introduced in the polyamide network endows the nanolms with Janus pores containing both hydrophilic parts and hydrophobic parts. The surface morphology, thickness and hydrophilicity of the nano-lms were effectively controlled by varying the monomer concentrations. The solvent-resistant aramid nanobrous hydrogel was used as a "green" support without further crosslinking. Nanoltration of polar and non-polar organic solvents with the composite membranes was investigated. The excellent chemical stability of the composite membranes was also evaluated under both aggressive acidic and strongly basic conditions. This work demonstrates that interfacial synthesis using monomers of extended alkanes provides polymeric nanolms with tunable structural diversity and chemical stability.

ANF hydrogel substrate preparation
The casting solution was prepared as reported previously. 2,17 The ANF hydrogel membrane was produced via non-solvent phase inversion in water. The casting solutions were allowed to settle to remove air bubbles and undissolved precipitates before casting the membrane. The non-woven polypropylene/ polyethylene (PP/PE) or polyolen (PO) fabric was placed on a clean glass plate. The casting solutions were cast onto the PP/ PE support using a casting knife set to a thickness of 250 mm at a temperature of 20-25 C. The hydrogel membranes precipitated from the water aer phase inversion and were stored overnight in deionized (DI) water.

Freestanding nanolm and m-XDA/TMC composite membrane preparation
A freestanding m-XDA/TMC molecular sieving nanolm was prepared on the support-free water/hexane interface via interfacial polymerization. The m-XDA/TMC molecular sieving nanolm was in situ fabricated on the ANF hydrogel substrate by interfacial polymerization as follows. Firstly, the hydrogel support membrane was purged with nitrogen gas to remove the water on the membrane surface. Subsequently, the aqueous diamine solution containing m-XDA and 0.5 wt% TEA was poured onto the hydrogel membrane surface and kept for 6 min. Aer draining the aqueous solution, the residual droplets were absorbed gently using a tissue paper and the membrane was dried for 1-3 min. Then, the organic hexane containing TMC was poured on the membrane surface to form a selective polyamide layer. The organic solution was drained aer 1 min of reaction and the membrane surface was rinsed with pure hexane. Finally, the resulting membrane was cured in an oven at 90 C for 1 min to enhance the crosslinking reaction and immediately placed in DI water for storage. The as-prepared membrane was denoted as m-XDA/TMC-x/y, where x represents the concentration of m-XDA in the aqueous solution and y indicates the concentration of TMC in the organic phase.

Characterization
The morphology of the membrane surface and of the crosssection was observed using a XL30 FEG eld-emission scanning electron microscope (FE-SEM, the Netherlands). Each sample was sputter-coated with a 1.5-2 nm Au layer before testing. Atomic force microscopy (AFM) was performed under ambient conditions using a Dimension 3100D AFM (Bruker) to obtain the membrane surface morphology, roughness and membrane thickness. The thickness of the freestanding nanolms was measured by depositing the nanolm on a silica substrate. AFM images were obtained in tapping mode using SSS-NCHR probes from NanoAndMore GmbH. The cantilever was made of Si with a spring constant of 40-50 N m À1 and a nominal tip apex radius of <5 nm. The AFM images were attened with order 1 aer scanning. Aer attening, the roughness parameter was calculated using the ISO 25178-2 standard. The elemental compositions of the m-XDA/TMC composite membrane were determined using X-ray photoelectron spectroscopy (XPS, AXIS Ultra DLD, Kratos Analytical, Japan). The acquisition time was 321 s, the anode was mono (aluminium (mono); 45 W), the step was 50.0 meW and the dwell time was 200 ms. The charge neutralizer was set at: current 1.8 A, balance 3.3 V and bias 1.0 V. X-ray diffraction (XRD) analysis was utilized to examine the crystalline structure of the prepared composite membranes using a Philips PW1830 diffractometer with a Bragg/Brentano q-2q setup and CuKa radiation at 45 kV to 30 mA on a 173 mm goniometer circle. The chemical structures of the composite membrane surfaces were analyzed by attenuated total reectance Fourier transform infrared spectroscopy (ATR-FTIR, NEXUS670). The contact angles were measured with a 3.0 mL DI water drop using the sessile drop method on a video contact angle system (OCA20, Dataphysics, German). The average value of six measurements at random positions for each sample was reported. The surface charge of the membranes was measured using the zeta potential (SurPASS 3, Anton Paar, Australia) with a 1 mM KCl electrolyte solution; the pH was adjusted with 0.05 mol L À1 HCl and NaOH and the gap height was xed at 100 mm. The thin lm made from m-xylylenediamine (m-XDA) and trimesoyl chloride (TMC) from an interfacial reaction is presented in Fig. S1. † The degree of network cross-linking (DNC) is the fraction of fully cross-linked segments in the polymer thin lm as dened in eqn (1) and (2): 12 where X and Y are the crosslinked segments and the linear segments in the polyamide network. According to the chemical formula of fully cross-linked (C 21 The O/N element ratios were determined from the respective peak areas from XPS narrow scan elemental peaks. Based on eqn (3), the DNC of the m-XDA/TMC thin lm was calculated.

Evaluation of OSN performance
The separation performance of prepared membranes was evaluated in a stainless-steel dead-end membrane module (HP4750) with an effective testing membrane area of 14.6 cm 2 . The separation equipment was operated with a constant stirring of 600 rpm to minimize concentration polarization at 20-25 C. The organic solvent nanoltration performance of the membrane was evaluated in various organic solvents using different dyes with a concentration of 20 mg L À1 at 4 bar. In all the OSN experiments, the membranes were ltered with pure DMF for 10 min to activate the membrane, followed by ltration of pure acetone for 10 min before testing. The solvent permeability (P, L m À2 h À1 bar À1 ) was calculated from the following equation: where V is the volume of the collected permeate (L), A is the effective area of the testing membrane (m 2 ), t is the interval time (h) and DP is the applied transmembrane pressure (bar). The solute (dye or salt) rejections (R, %) were calculated using the following equation: where C p and C f are the solute concentrations in the permeate and feed solutions, respectively. The dye concentrations were determined using a PerkinElmer lambda 12 UV-vis spectrophotometer.

Chemical stability test
The long-term ltration stability test of the membrane in aggressive acetone was evaluated using methyl orange (327 g mol À1 ) as the solute with a concentration of 100 mg L À1 at 4 bar and 600 rpm in dead-end ltration in batch mode for seven cycles. For each cycle, the membrane was tested for 8 hours and allowed to stand undisturbed for 16 hours. The feed solution was 200 mL and the permeate was recycled when collecting 100 mL. As the feed organic stream in practical applications generally involves highly harsh conditions, 1,5,18 the chemical stability of the m-XDA/TMC composite membranes under harsh acidic and basic conditions was also investigated. The membranes were cut into disks ( Table  S1. † Aer this, the OSN performance of the treated membrane was evaluated using methanol and acetone as the feed solution and using dye molecules as the solute with a concentration of 20 mg L À1 in dead-end ltration at a pressure of 4 bar and 600 rpm. The membrane was rinsed with DI water for 5 min to remove the residual solution mixture and ltered with DMF for 10 min before testing.

Results and discussion
3.1 Fabrication of m-XDA/TMC membranes Fig. 1a illustrates the interfacial polymerization (IP) process on the aramid nanober (ANF) hydrogel substrate. Due to the interconnected porous structure, the nanobrous substrate provides a homogeneous distribution of the aqueous solution compared to a conventional polymeric dense substrate. 2 In addition, the non-covalent interactions between water and the macromolecular chain of nanobers controlled the diffusion of the aqueous solution. 2,19 A defect-free selective layer with an ultra-thin thickness is expected based on the ANF substrate. A molecular sieving nanolm was prepared via IP between mxylylenediamine (m-XDA) in aqueous solution and trimesoyl chloride (TMC) in the hexane phase. Triethylamine (TEA) was used as the additive in the aqueous solution to absorb the byproduct of HCl produced during IP, thereby enhancing the cross-linking degree of the selective layer. 20,21 m-XDA as an aromatic diamine was used as the aqueous monomer for the synthesis of the OSN membrane. It has a similar chemical structure to the commonly used m-phenylenediamine (MPD) (Fig. S2 †). However, the methylene moiety (-CH 2 -) situated between the benzene ring and the amide group is expected to endow the crosslinked polyamide (PA) nanolm with hydrophobic properties (Fig. 1b). Besides, the elongated side chain could produce a larger membrane pore size compared to that of the nanolms prepared from MPD/TMC. Fig. 1c displays the freestanding nanolm held by a 4 cm diameter steel ring. Fig. 1d shows the SEM image of a crumpled and defect-free nanolm prepared from the support-free water/hexane interface, followed by transfer onto a copper mesh. Freestanding nanolms prepared from different ratios of m-XDA/ TMC could also be easily obtained (Fig. S3 †). In view of industrial applications in pressure-driven organic solvent nano-ltration, a composite membrane containing a selective nanolm and a porous substrate was fabricated. As shown in Fig. 1e, the intertwined aramid nanobers were used as "building blocks" to create a robust 3D interconnected structure for the membrane substrate. The solvent-resistant brous substrate has a great advantage over conventional polymeric supports as it requires no cross-linking agent. 2 The molecular sieving nanolm with the support of ANF hydrogel yields a colorful surface, indicating the successful interfacial polymerization (Fig. 1f). 2,19 The IP process on the hydrogel substrate was manipulated on a molecular level by varying the monomer concentration of m-XDA and TMC. As shown in Fig. 1g, the m-XDA/TMC-3/0.15 composite membrane exhibited a rough surface with a "leaf-like" structure and a high thickness of 457 nm (Fig. S4 †). In addition, a smooth and ultra-thin selective layer with a thickness down to 27 nm was also obtained for the m-XDA/TMC-0.5/0.025 membrane (Fig. 1h).
In order to precisely control the membrane thickness, the monomer concentration was varied, manipulating the condensation rate and cross-linking degree. In this work, molecular sieving nanolms were prepared by xing the concentration ratio of m-XDA/TMC at 20 but varying the monomer concentrations (3/0.15, 2/0.1, 1.5/0.075, 1/0.05, and 0.5/0.025). The corresponding nanolm thickness was between 27 and 457 nm. The nanolm thickness was found to be inversely proportional to concentration levels (Fig. S4 †). Moreover, the membrane surface morphology greatly varied depending on the monomer concentration (Fig. 2b-f and S5 †). The ANF substrate was found to have a relatively rough surface (Fig. 2a). Aer interfacial polymerization, a dense and smooth thin lm membrane covered the substrate (Fig. 2b). In addition, a growing number of crystalline particles were observed on the membrane surface, indicating an increased condensation reaction rate, as shown in Fig. 2b-e. 22 Notably, the membrane surface produced well distributed hollow bubbles at a monomer concentration of 3/0.15 (Fig. 2f). These bubbles are generated in  situ by the heat released during interfacial polymerization due to the fast reaction rate at high monomer concentrations. The surface morphology and roughness of the composite membranes were also analyzed by atomic force microscopy (AFM), as shown in Fig. S6. † Table 1 shows that the m-XDA/ TMC-3/0.15 membrane has the highest root-mean square (RMS) surface roughness of 42.7 nm, as compared to those scoring between 14 and 21 nm. These results are in accordance with the SEM images (Fig. 2).

Characterization
X-ray diffraction (XRD) was used to examine the chemical structure of the nanolms. The results are shown in Fig. 3a. In comparison with the XRD pattern of the pristine silica substrate, the XRD pattern of the nanolms shows two strong peaks at 2q values of 7.8 and 15.3 . This indicates that the nanolm has a crystalline structure. 22 The m-XDA/TMC-2/0.1 membrane and the m-XDA/TMC-1.5/0.075 membrane show a higher degree of crystallinity with values of 98.1% and 95.8%, respectively (Table S2 †). However, the m-XDA/TMC-3/0.15 has the lowest crystallinity of 81.8%, which indicates a higher content of amorphous regions in the nanolm. Karode et al. developed a comprehensive model for interfacial polymerization and found that lm growth by nucleation is dominant as the organic phase concentration decreases. 23 Thus, the lower crystalline fraction of the m-XDA/TMC-3/0.15 membrane might be caused by the high organic phase concentration. In addition, a sufficiently high reactant concentration also leads to more rapid heat release from the exothermic polycondensation reaction, producing a more pronounced amorphous structure. 24 To demonstrate the successful interfacial polymerization between m-XDA and TMC, the membrane surface chemistry was examined by Fourier transform-infrared spectroscopy (FT-IR) (Fig. 3b). The m-XDA/TMC composite membranes show FT-IR spectra similar to that of the ANF hydrogel substrate, except for the two new peaks appearing at 2842 cm À1 and 2910 cm À1 , which originate from the C-H bending vibration of methylene (-CH 2 -). This indicates that the -CH 2was successfully introduced into the polyamide structure. The N-H stretching vibration at 3320 cm À1 is ascribed to the amine group. 3 The characteristic peaks at 1644 cm À1 (C]O stretching vibration, amide I band) and 1540 cm À1 (N-H in-plane bending and C-N bending vibration, amide II band) demonstrate the presence of functional -NHCObonds. 2,19 The elemental composition, chemical bonding, and cross-linking degree of the molecular sieving nanolm were analyzed by X-ray photoelectron spectroscopy (XPS). The XPS survey spectra of composite membranes presented in Fig. 3c conrm the presence of carbon (C 1s), nitrogen (N 1s) and oxygen (O 1s) elements at the membrane surfaces ($10 nm depth). The cross-linking degree was calculated from the ratio of O/N (Table 1). 12 It was found that the m-XDA/TMC-3/0.15 membrane has the highest crosslinking degree (56.4%), followed by the m-XDA/TMC-2/0.1 membrane (52.1%). The synthetic components aer deconvolution along with the background and tted curves are shown in Fig. 3d-f. The C1s spectrum of the m-XDA/TMC-3/0.15  membrane without TEA addition (Fig. 3d) was deconvoluted into four peaks at 284.2 eV (C-C), 284.8 eV (C-C and C]C), 285.3 eV (C-N) and 287.8 eV (C-O and O-C]O). 2 The N 1s spectrum showed the amide bond at 399.5 eV (N-C]O) (Fig. 3e). 2 The O1s spectrum of the nanolm was deconvoluted into two peaks at 531.2 eV and 532.6 eV, corresponding to the amide bond (N-C]O) and the carboxylic acid group (O-C]O), respectively (Fig. 3f). 25,26 The narrow scan results of the XPS spectra of composite membranes with the addition of TEA are shown in Fig. S7. † Plausible species for the polyamide composite membranes with the addition of TEA were determined from the deconvolution of C1s, O1s and N1s core level XPS spectra (Table S3 †).
Because of the hydrolysis of residue acyl groups into carboxyl groups and the intrinsic hydrophilicity of polyamide, conventional TFC membranes prepared from MPD/TMC always have a hydrophilic surface, especially aer the solvent activation of DMF. 9,19 This explains why PA membranes have no/extremely low permeance for non-polar solvents. As shown in Fig. S8, † compared with an m-XDA/TMC composite membrane, the polyamide thin lm surface shows good wettability with a water contact angle (CA) of 33.2 . However, interestingly, introducing hydrophobic moieties into the cross-linked polyamide could effectively balance the hydrophilicity-hydrophobicity. It was found that reducing the m-XDA/TMC concentration could shi the composite membrane surface from hydrophilic to hydrophobic. As shown in Fig. 4a, the ANF substrate has a CA of 67.2 . Because of the higher content of the carboxyl groups (>2%) and amide groups (related to the higher cross-linking degree) (Table  1), the membranes fabricated from concentrations higher than 1.5/0.075 show hydrophilic properties with a CA around 60 . Further decreasing the m-XDA/TMC concentration leads to a decrease in the content of carboxyl groups ($1.6%) and an increase of the CA (Fig. 4b). Consequently, the m-XDA/TMC-1/ 0.05 and m-XDA/TMC-0.5/0.025 membranes exhibit a water CA of 103.1 and 98.3 , respectively, indicating their hydrophobic characteristics (Table 1). Livingston et al. prepared polyamide nanolms by manipulating the monomer concentrations in a wide range of 0.0025-0.5 wt%. 12 It was found that even the nanolm prepared from an extremely low TMC concentration of 0.0025 wt% still maintained a hydrophilic surface with a CA smaller than 60 . By comparison, the introduced methylene endows the m-XDA/TMC membrane with tunable physicochemical properties. The balance between the hydrophilic parts and the hydrophobic parts gives rise to a different hydrophilic/ hydrophobic feature of the m-XDA/TMC membrane. All these composite membranes were found to have a negatively charged surface (Fig. 4c) with a zeta potential between À45 and À24 mV at neutral pH ( Table 1). The carboxyl (-COOH) content in the PA thin lm was conrmed from the O1s core level XPS spectra (Table S3 †). The relationship between the zeta potential and the carboxyl group content is plotted in Fig. S9. † An overall trend could be found where a higher carboxyl content exhibits a lower zeta potential for the nanolms except for the m-XDA/TMC-2/0.1 membrane, which has a lower zeta potential.

Organic solvent nanoltration performance
There is a linear relationship between the permeance of the organic solvents and the reciprocal of the thickness of the selective PA layer (Fig. 5a). High permeances of 70.6 L m À2 L À1 bar À1 and 54.5 L m À2 L À1 bar À1 were achieved for DMF and acetone, respectively, when the PA thickness was 27 nm. In contrast, the permeances for the two aggressive solvents decreased to 9.1 L m À2 L À1 bar À1 (DMF) and 8.2 L m À2 L À1 bar À1 (acetone) when the PA thickness was 457 nm. These results demonstrate that controlling the monomer concentration allows us to effectively control the thickness of the molecular sieving nanolm, determining the solvent permeability. Table 2 compares the organic solvent nanoltration performance of m-XDA/TMC composite membranes. The m-XDA/TMC-3/0.15 membrane was found to have an almost complete rejection of rose Bengal (RB, 1017 g mol À1 , 100%) and methyl orange (MO, 327 g mol À1 , 99.6%) in acetone. The m-XDA/TMC-2/0.1 membrane had an acetone permeance two times higher than that of the m-XDA/TMC-3/0.15 membrane because of the decreased nanolm thickness. In addition, the membrane still retained a high rejection for RB (99.9%) and a considerable rejection for MO (90.7%). The rejection for MO decreased and the permeability increased signicantly when further deceasing the membrane thickness, indicating that the cross-linking degree of the selective layer was decreased. As shown in Table  1, a higher monomer concentration leads to a higher crosslinking degree but also results in a larger nanolm thickness and vice versa. The interplay among the cross-linking degree, thin lm thickness and separation performance (permeability and molecular selectivity) is regulated by manipulating the monomer concentration in the IP reaction. The solvent resistance of the m-XDA/TMC composite membrane was evaluated by immersing the membrane in various polar aprotic solvents, polar protic solvents and non-polar solvents for one week. It was found that dissolution of the crystalline particles on the membrane surface occurred in dimethyl sulfoxide and a strong swelling of the membrane in 1-methyl-2-pyrrolidinone and tetrahydrofuran was observed (Fig. S10 †). However, the surface morphology of these composite membranes in mild alcoholic solvents (methanol, ethanol, and isopropanol), acetone and non-polar solvents (toluene and n-hexane) was maintained aer the treatment. The membrane permeance versus solvent viscosity is plotted in Fig. 5b. An inverse non-linear relationship between solvent permeance and solvent viscosity was observed, which is in accordance with the results reported elsewhere. 3,14 Due to the strong swelling effect of the hydrogel substrate, the membrane had a signicantly higher permeance for DMF (21.6 L m À2 h À1 bar À1 ) than for other solvents. 2 Because of the lower viscosity of acetonitrile (0.37 cP), acetone (0.32 cP) and methanol (0.59 cP), the permeances for these three polar solvents ranked the highest with 22.9, 16.7 and 12 L m À2 h À1 bar À1 , respectively. The membrane also showed a high permeance for a non-polar solvent, i.e., toluene (4.2 L m À2 h À1 bar À1 ), which is over six times higher than that of commercial membranes (Starmem 122 and Starmem 144). 13,27 This permeance also surpasses that of most recently reported membranes, such as AlO x /PIM-1, 28 cross-linked PVDF, 29 and epoxysilicone composite membranes. 30 To further demonstrate their potential application in non-polar solvents, the composite membranes were applied in n-hexane. It was found that the composite membrane had an increasing permeance for nhexane from 0.6 to 2.6 L m À2 h À1 bar À1 (Fig. 5c). Notably, due to the hydrophilicity, a conventional polyamide membrane has a superior permeability for polar solvents, such as acetone and methanol, but extremely low permeances for non-polar , eosin Y (EY, 648 g mol À1 ), Sudan black B (SBB, 457 g mol À1 ), methyl orange (MO, 327 g mol À1 ) and disperse orange 3 (DO3, 242 g mol À1 )) in acetone and methanol, respectively, with a molecular concentration of 20 mg L À1 . Table 2 Organic solvent nanofiltration performance of m-XDA/TMC composite membranes. Methyl orange (327 g mol À1 ) and rose Bengal (1017 g mol À1 ) dissolved in acetone with a concentration of 20 mg L À1 were used as the feed solution to study the solute rejection. Membranes were activated with DMF filtration for 10 min followed by filtration with a feed solution of acetone. The permeate was collected after the membrane was filtered for 30 min and the rejection was determined by UV-vis spectrophotometry. Nanofiltration experiments were performed in a dead-end stirred cell (600 rpm) at room temperature and 4 bar

Composite membrane
Pure acetone permeance (L m À2 h À1 bar À1 ) Rose bengal (1017 g mol À1 ) Methyl orange (327 g mol À1 ) Permeance (L m À2 h À1 bar À1 ) solvents, for example, n-hexane (<1 L m À2 h À1 bar À1 ). 1,2,9,12,13 The comparison of the non-polar solvent (toluene and hexane) permeances of the m-XDA/TMC TFC membrane with those of other commercial OSN membranes and reported composite membranes in the literature are listed in Table S4. † The elevated non-polar solvent permeance for the m-XDA/TMC composite membrane can be attributed to the introduction of the alkyl (-CH 2 -) group into the polyamide network, which has a good affinity towards n-hexane and toluene. In addition, the alternating distribution of the hydrophilic part and the hydrophobic part endows the membrane pores with Janus pathways, where the former represent the residual hydrolyzed carboxyl groups and amide groups, and the latter represents the introduced alkyl groups and the benzene rings (Fig. 5d). As a consequence, the Janus pathways facilitate the transport of both polar and non-polar solvents. To validate the interaction differences between the cross-linked polyamide polymer with different solvent molecules, the solubility parameter differences for the polymer and solvent are calculated, as shown in Tables S5 and  S6. † Compared with conventional MPD/TMC polyamide, the m-XDA/TMC polyamide not only shows strong interactions with polar solvents (methanol, ethanol, and acetone) but also has better interactions with non-polar solvents (n-hexane and toluene). These results indicate that the m-XDA/TMC composite membranes are promising for use in nanoltration of polar and non-polar solvents. The molecular weight cut-off (MWCO) of the m-XDA/TMC composite membranes was evaluated based on the rejection of different dye molecules in specic solvents (methanol and acetone). Dye molecules with different charges, molecular sizes and molecular weights are listed in Fig. S11. † A plot of rejection of dye molecules against molecular weight is presented in Fig. 5e. The composite membrane showed a similar MWCO between 242 and 327 g mol À1 in acetone and methanol, respectively. The UV-vis measurement results and the photographs of the feed/permeate in acetone are provided in Fig. S12. † Tables S7 and S8 † also provide the permeance and rejection information of the MWCO experiment.

Stability test
The long-term ltration of the m-XDA/TMC composite membrane was evaluated in acetone, considered an aggressive solvent (Fig. 6a). The decrease of permeability is probably caused by physical ageing and compaction of the hydrogel composite membrane with DMF treatment. 2,19 Despite this, the m-XDA/TMC-3/0.15 membrane demonstrated an excellent nanoltration separation ability in the long-term ltration experiment, in which a permeability of 5.2 L m À2 h À1 bar À1 was obtained for acetone with a rejection of 99.0% for methyl orange (MO, 324 g mol À1 ). Fig. S13a † shows the UV-vis spectrum of the feed and permeate. The inset photo of the permeate indicates the highly efficient molecular separation capability of the m-XDA/TMC TFC membrane. As shown in Fig. S13b, † the membrane was compacted but no signicant fouling is observed aer the long-term ltration. These results demonstrate that the m-XDA/TMC TFC membrane has excellent stability in harsh organic solvents and has great potential for applications in solvent recovery, solute concentration, etc. The OSN permeances of the m-XDA/TMC membrane are compared with those of other reported TFC membranes, as listed in Table S9. † The m-XDA/TMC TFC membrane developed in this work shows great advantages in terms of the high solvent permeances for both acetone and methanol as well as the excellent rejections for small organic molecules. The cross-linked aromatic framework and the introduced methylene groups endowed the m-XDA/TMC composite membranes with an extraordinary chemical stability, which is essential for OSN performance in harsh environments. The m-XDA/TMC-2/0.1 composite membranes were soaked in a variety of solutions under extreme conditions for 15 days: 0.1 M HCl in water/ethanol, 0.1 M NaOH in water/ethanol, and 0.1 M HCl in water/acetone. The OSN performances of the treated composite membranes were evaluated (Fig. 6b). The rejection of rose Bengal of these composite membranes remained almost 100% and high rejections of over 90% for MO both in ethanol and acetone were observed. Interestingly, the rejection for MO increased signicantly from 90.7% to 99.4% aer treatment with 0.1 M HCl in water/acetone. In contrast, the permeances for acetone remained stable. To investigate the stability of the membrane under more extreme conditions, the m-XDA/TMC-3/0.1 membranes were soaked under stronger acidic and basic conditions for 15 days: 0.5 M HCl in water/ethanol, 0.5 M NaOH in water/ethanol, and 0.5 M and 1 M HCl in water/acetone. As shown in Fig. S14a, † the color of the membrane immediately changed to pink in HCl/ acetone and yellow in NaOH/ethanol but remained the same in HCl/ethanol. Despite the color change, the membrane exhibited excellent molecular rejections for RB (100%) and MO (>98%) in ethanol and acetone, respectively (Fig. S14b †). The surface morphology of these composite membranes aer ltration with MO/solvent was examined and maintained well according to SEM observations (Fig. S14c †). In addition, the chemical stability of the thinnest membrane (m-XDA/TMC-0.5/0.025) was also investigated under harsh conditions (Fig. S15 †). The results show that the m-XDA/TMC composite membrane even with a thickness down to 27 nm can still maintain its separation efficiency under harsh conditions. The chemical stability of the m-XDA/TMC composite membranes can be attributed to the robust nature of