Main

We achieved the bottom-up synthesis of MFI nanosheets by hydrothermal growth of MFI seeds approximately 30 nm in diameter (Fig. 1) in the presence of bis-1,5(tripropyl ammonium) pentamethylene diiodide (hereafter denoted as dC5; see Extended Data Fig. 1a). The tetrapropyl ammonium cation (TPA; see Extended Data Fig. 1e) is the most effective structure-directing agent (SDA) for MFI. Its dimers, like dC5 (the dimer with five methylene carbons connecting the two nitrogen atoms), are known to yield distinct crystal morphologies depending on the length of the nitrogen-bridging alkyl chains6,19. dC5, in particular, can direct the formation of plate-like MFI6,20 with the thin crystal dimension along the b axis, a morphology that is considered favourable for membrane applications, owing to the presence of straight micropores along this direction21.

Figure 1: Growth stages of MFI nanosheets from seed.
figure 1

a, Schematic and the corresponding bright-field TEM (BF-TEM) images, representing different stages of growth of MFI nanosheets starting from seeds approximately 30 nm in diameter. b, Schematic showing the preferential growth of a seed along the b axis. c, The corresponding TEM image of the seed in b (left panel); the HR-TEM images overlaid with the crystal structure model along the [100], [010] and [001] zone axes (middle panels) confirming the MFI-type zeolite structure; and diffraction pattern of the seed taken along the [001] zone axis (right panels) confirming elongation of the seed along the b axis. d, Schematic of a nanosheet growing out of the seed. e, High-angle annular dark-field scanning transmission electron microscopy (HAADF-STEM) images with different magnifications for the seed in d (left and middle panels); and image intensity linescan along the red dotted line (‘Position’ indicates the distance from the left edge of the red dotted line) in the second HAADF-STEM image (right panel), showing a thickness step in the nanosheet that corresponds to a ratio of 5 pentasil chains to 3 pentasil chains from left to right. f, Schematic showing partial wrap of a b-axis-oriented nanosheet around an a-axis-oriented seed. g, BF-TEM image (left panel) and corresponding [100] zone axis diffraction patterns (right panels) from the seed (red ring) and [010] zone axis diffraction pattern from the sheet (yellow ring). Scale bars from left to right in a are 20 nm, 20 nm, 50 nm, 100 nm, 100 nm and 500 nm; in c are 20 nm and 1 nm−1; in e are 50 nm and 20 nm; and in g are 100 nm, 1 nm−1 and 1 nm−1.

PowerPoint slide

Full size image

However, a bottom-up (not based on exfoliation) synthesis of single nanosheets has not been achieved until now owing to the formation of orthogonal intergrowths (MFI twins). In fact, the propensity for twin formation in the presence of dC5 has been elegantly exploited to create MFI crystals with hierarchical porosity20 (Extended Data Fig. 1b). Earlier, we suggested that the formation of MFI and other zeolite rotational intergrowths10,22 are preceded by the intergrowth of a related zeolite structure with different symmetry. In the case of MFI, the orthogonal intergrowth is probably caused by the incorporation of the higher-symmetry framework type MEL10 (Extended Data Fig. 1c). Insertion of MEL segments within MFI has been detected by electron microscopy23, and it has been described from crystallography and crystal growth mechanism standpoints on the basis of the placement of silicate chains (called pentasil chains; see Extended Data Fig. 1d) along the common c axis of MFI and MEL24. MFI is formed when the pentasil chains grow so that an inversion centre exists along the a axis, whereas MEL is formed when the pentasil chains relate by a mirror plane.

From the above arguments (made in ref. 24) regarding the arrangement of pentasil chains and our proposal that MEL domains trigger orthogonal intergrowths10, we hypothesized that the emergence of orthogonal intergrowths is favoured on flat, well developed, crystal facets on which MEL-forming pentasil chain arrangements can extend along the common c axis of MEL and MFI. Therefore, to avoid intergrowths, we investigated seeded growth with nanometre-sized seeds that do not have well defined facets and, therefore, cannot support orthogonal intergrowth at early stages of growth (see Extended Data Fig. 1f). Indeed, after extensive variation of synthesis conditions, we identified a synthesis window that produces zeolite nanosheets free of rotational intergrowths.

Figure 1a shows schematic illustrations and the corresponding transmission electron microscopy (TEM) images of seed crystals (depicted in red) and nanosheets (depicted in yellow) from different growth stages. For an extended time (typically 20–40 h), the MFI nanocrystal seeds grow slowly to acquire a near-cylindrical shape (depicted in Fig. 1b). Figure 1c shows a typical cylindrical nanocrystal grown to 140 nm in length with the corresponding electron diffraction pattern and also high-resolution TEM images along the three MFI axes of similar nanocrystals. In accordance with an earlier report6, we determined the long axis of the cylindrical crystals to be the b axis of MFI (see also Extended Data Fig. 2a–f). Examination of several crystals at this stage of growth does not yield any evidence of rotational intergrowth, and the lack thereof supports the underlying hypothesis of this work. In most crystals examined at this stage, owing to slight misorientation with respect to the remainder of the crystal, the original 30-nm seed can be made visible using dark-field TEM imaging (Extended Data Fig. 2g–i). This observation indicates that the cylindrical crystals evolve from the 30-nm seeds by epitaxial growth, and that they are not newly nucleated crystals.

After reaching this stage of growth (typically 20–40 h), a relatively rapid transition in crystal growth occurs (typically completed within a few hours). Nanosheets start to appear from one of the corners of the cylindrical crystals (approximately along the [011] direction; Fig. 1d and e) and then continue to grow, encircling the seed to form a faceted nanosheet (Fig. 1a). High-angle annular dark-field (HAADF) imaging of the emerging nanosheets reveals thickness variations between their outermost portion and the part closer to the cylindrical crystal with thickness ratios of 3/4 and 3/5, which are consistent with nanosheet thicknesses of 3, 4 and 5 pentasil chains (Extended Data Fig. 3). As the nanosheet continues to grow, its thickness becomes uniform, and the nanosheet encircles the cylindrical seed and develops well defined facets (Fig. 1a, f and g and Extended Data Fig. 4). Atomic force microscopy (AFM; Fig. 2b) and electron microscopy images (Fig. 2d and e) indicate that the predominant thickness of the nanosheets upon complete encirclement of the seed is 5 nm (that is, five pentasil chains, or 2.5 unit cells thick along the b axis). With prolonged growth, the nanosheets thicken by the nucleation of islands and slow step propagation (Extended Data Fig. 5a and b). High-resolution transmission electron microscopy (HR-TEM) imaging and simulations demonstrate that the nucleated islands are MFI and that rotational intergrowths are not present on the nanosheets (Extended Data Fig. 5c–f).

Figure 2: Characterization of the MFI nanosheet.
figure 2

a, SEM image for MFI nanosheets. Average sizes are about 2.0 μm and 1.2 μm along the a- and c-axes, respectively. b, AFM height image and height profiles for the MFI nanosheets. The height profiles are extracted along the red dashed lines. The seed crystal at the centre is about 100 nm across, and the nanosheet exhibits a very uniform thickness of 4.6 ± 0.5 nm. c, High-magnification SEM image for the MFI nanosheet. The surface roughness of the nanosheets is due to the additional layer that forms on the top of the primary nanosheet. d, Underfocus (+∆f) and overfocus (−∆f) BF-TEM image with overlaid multislice simulations of a 5-pentasil-chain-thick MFI nanosheet section along the b-axis. White lines in the underfocus image and black lines in the overfocus image correspond to pentasil layers in the overlaid crystal structure model. The spacing between the black and white lines is dependent on TEM imaging conditions (defocus) and is not representative of the actual bond distances between the pentasil layers. e, HAADF-STEM image of the nanosheet (bright area) shown in d with a corresponding average intensity scan confirming the thickness of the MFI nanosheet to be 4.7 nm along the b axis. Scale bars in a, b and c are 1 μm and in d and e are 5 nm.

PowerPoint slide

Full size image

The observations described above establish a new crystal growth modality for MFI that produces the desired high-aspect-ratio nanosheet morphology. A key element of this new growth modality is the transition of the crystal growth mode from one of epitaxial slow growth on a seed crystal to one that can support the emergence and rapid growth of a nanosheet. What triggers this transition? To answer this question, we investigated the crystallographic relationship between the seed and the nanosheet. Interestingly, electron diffraction patterns (Fig. 1f and g and Extended Data Figs 4 and 6) reveal an orthogonal rotational intergrowth relationship between them. They both share a common c axis, but their a and b axes are rotated by 90°. When we attempted to grow nanosheets starting from smaller and smoother seeds, which we prepared by disassembly of three-dimensionally ordered mesoporous-imprinted (3DOm-i) MFI crystals25,26, we found a similar outcome, that is, nanosheets emerged after the 3DOm-i seeds grew to a similar cylindrical morphology (Supplementary Fig. 1). These results show that the emergence of nanosheets from the seeds is triggered by a single rotational intergrowth that takes place only after the seeds reach a certain size and shape. It seems that until this size and shape are attained, there are no extended flat surfaces to support the intergrowth. After the rotational intergrowth is triggered, the emerging nanosheet exposes high index (h0l) facets that are highly reactive and favour fast in-plane growth until a well faceted nanosheet is formed. As the facets approach the (001) and (101) faces of MFI, growth slows down. This self-regulation of in-plane growth rates allows for the final symmetric appearance of nanosheets despite their asymmetric genesis.

The proposed kinetically controlled emergence of nanosheets rather than a strain-induced transition is also supported by X-ray diffraction (XRD) data and theoretical structure optimizations. Analysis of powder XRD patterns (Extended Data Fig. 7, top) do not show substantial unit cell differences between seed crystals and nanosheets. This is in agreement with electronic structure calculations that indicate accommodation of dC5 in nearly strain-free MFI nanosheets (Extended Data Fig. 7a–f). Thermogravimetric analysis (TGA, Supplementary Fig. 2) and Ar porosimetry (Supplementary Fig. 3) show that nanosheets exhibit behaviour similar to conventional MFI.

The bottom-up seeded hydrothermal growth method developed here provides nanosheets with increased lateral dimensions and at higher yield, compared to the top-down, exfoliation-based approaches15, allowing the facile preparation of highly oriented coatings (Supplementary Fig. 4). Moreover, their unusual 2.5-unit-cell thickness and large aspect ratio endow these nanosheets with textural properties that lie between those of conventional MFI and exfoliated MFI nanosheets and are advantageous for thin-film formation. For example, in ref. 27, the conventional MFI crystals used as seed layers for membrane formation by gel-less growth had lateral dimensions of 1.5–2.0 μm and a thickness of about 0.5 μm, whereas the 3-nm-thick nanosheets prepared by exfoliation and used for membrane formation using a similar approach in ref. 28 had lateral dimensions that were not larger than 300 nm.

The dC5-MFI nanosheets reported here combine the lateral dimension of conventional MFI crystals (1.5–2.0 μm; see Fig. 2) with nanometre-scale thickness. Moreover, because the dC5 MFI nanosheets are made by direct synthesis, they do not suffer from the fragmentation caused during exfoliation of the lamellar precursors. For these reasons, they can effectively cover porous and non-porous surfaces to create compact and oriented seed layers. Figure 3a is a top-view scanning electron microscopy (SEM) image of an MFI membrane made by gel-free secondary growth27 of an oriented nanosheet deposit. It shows highly intergrown flat grains that are several micrometres in lateral dimensions. These large lateral dimensions cannot be obtained by using nanocrystalline seeds29 or exfoliated nanosheets28 (Extended Data Fig. 8), and they are advantageous in forming films with lower density of grain boundaries, which are known to contribute to non-selective transport pathways that bypass the selective MFI micropores. Figure 3b shows an SEM cross-section establishing a dense layer after gel-free secondary growth with thickness ranging between 250 nm and 1 μm. TEM and electron diffraction investigation of the cross-sections prepared by focused ion beam reveals the presence of oriented, intergrown, and thickened nanosheets with large lateral dimensions (Extended Data Fig. 9). The few 300-nm near-rectangular crystals observed at the centres of flat grains in Fig. 3a are seed nanocrystals enlarged mostly along their c axis during secondary growth. They are oriented with their a axes perpendicular to the membrane surface. Since they occupy only a small fraction of the surface, XRD confirms that the membrane is preferentially b-axis out-of-plane oriented (Fig. 3c). Considering the microstructure revealed by the SEM, TEM, electron diffraction and XRD analysis presented above, we expect that molecular transport through these membranes will be dominated by the b-axis-oriented MFI grains formed from the nanosheets.

Figure 3: Xylene separation performance for MFI membrane.
figure 3

a, b, Top view (a) and cross-sectional view (b) SEM images for an MFI membrane prepared by gel-free intergrowth of MFI nanosheets; scale bars represent 2 μm. c, XRD pattern for the MFI membrane in a and b (a.u., arbitrary units). d, p-xylene permeance and selectivity for MFI membrane: p-xylene permeances from single-component permeation test measured with decreasing temperature (black circles) and increasing temperature (white circles), and p-xylene permeance (red squares) and separation factor (white squares) measured for a 1:1 p-xylene and o-xylene mixture. The black arrows indicate the directions of the temperature sweep.

PowerPoint slide

Full size image

Indeed, the separation performance of membranes made on porous silica supports using the directly synthesized nanosheets compares favourably with membranes made by exfoliated nanosheets and also with other MFI membranes. The membrane molecular sieving properties were benchmarked using xylene isomer separation performance, which is of both practical and fundamental importance30. Figure 3d shows para-xylene permeances and separation factors (from ortho-xylene) in the range 40–160 °C. Single-component permeances for p-xylene measured with decreasing and increasing temperature over a period of two weeks confirmed stable and reversible membrane performance. The membrane exhibited similar high p-xylene permeances for single-component and binary mixtures (for example, about 0.56 × 10−6 mol Pa−1 m−2 s−1 at 150 °C) and very high mixture separation factors of about 2,500 at 125 °C and 2,000 at 150 °C (the next highest is about 1,000 in ref. 27). The permeance and separation factors obtained are considerably higher than those of membranes made from exfoliated nanosheets28. Tests from additional membranes (Supplementary Table 1) showed that even higher separation factors (approximately 8,000) can be achieved. Although such high separation factors have not been reported before, they are anticipated by simulations31 (see Extended Data Fig. 10a).

Moreover, membrane performance stability was examined. Over a period of five weeks, during which the temperature of the membrane was cycled twice between 150 °C and 50 °C, the separation factor decreased from the initial value of around 2,000 to around 1,200 and the p-xylene permeance changed from 5.6 × 10−7 mol Pa−1 m−2 s−1 to 3.5 × 10−7 mol Pa−1 m−2 s−1 (Extended Data Fig. 10c). It is also important to note that a high separation factor was consistently observed even at low temperatures (about 300 at 75 °C and about 30 at 50 °C). At the feed conditions tested, the p-xylene and o-xylene activities at 150 °C are low, whereas at 50 °C, they are much higher (Extended Data Fig. 10c). The corresponding single-component p-xylene saturation loadings for free-standing MFI crystals at 150 °C and 50 °C are approximately 1 and 8 molecules per unit cell32. The increased loading is partly responsible for the large permeance drop. However, it is remarkable that high separation factors (around 30) can still be obtained under such high nominal loading conditions.

The membranes also exhibited stable performance (Extended Data Fig. 10c) and excellent p-xylene selectivity for multi-component mixtures of aromatics and good performance for n-/i-butane (with a separation factor of 50 at 25 °C) and alcohol/water mixtures (with a separation factor of 35 at 60 °C), as shown in Extended Data Fig. 10d and Supplementary Table 2, respectively. Although similar membrane thicknesses (<1 μm) and p-xylene permeance values have been reported before (Extended Data Fig. 10b), this unprecedented combination of high separation factor and high permeance for a wide range of conditions and feed mixtures can be attributed to the improved MFI membrane microstructure (thin, wide and well intergrown b-axis-oriented MFI grains with reduced number of grain boundary defects) enabled by the high-aspect-ratio 5-nm-thick nanosheets.

The gel-free secondary growth method27,28 relies on the consumption of a top sacrificial layer of 50-nm Stöber silica nanoparticles as the silica source for the growth of MFI seed layers. It is not applicable to non-silica supports like the widely used α-alumina supports. For the use of other porous supports, secondary growth methods that rely on the presence of a silica sol or gel should be employed. Recently, we reported on the use of tetraethylammonium-hydroxide silica sols to achieve homoepitaxial growth on exfoliated MFI nanosheets33. Preliminary findings indicate that the method is applicable to dC5-MFI nanosheet seed layers (Supplementary Fig. 5) and may enable the use of MFI nanosheets for the preparation of high-quality membranes on other porous supports.

To further improve membrane performance, future efforts should be directed towards the production of high-aspect-ratio nanosheets with uniform thickness (that is, without the presence of the seed at the centre of the nanosheets). Although the ultimate approach would be based on non-seeded, non-exfoliation-based bottom-up synthesis, one way to achieve uniform thickness using the approach reported here is to attempt to remove seed crystals by chemical or mechanical means. Figure 4a and b shows SEM and AFM height images for MFI nanosheets after seed-crystal removal by rubbing. The SEM image confirms that the seed crystals were effectively removed, resulting in holes of about 50 nm across at the centre of the nanosheets. Figure 4c and d shows tilted-view AFM three-dimensional images of MFI nanosheets before and after rubbing, respectively, which confirm that the flatness of the nanosheets is improved after rubbing. Further work in this direction will enable the large-scale implementation of ultra-thin zeolite materials in membrane separations contributing to energy-efficient separation processes.

Figure 4: Post-synthesis seed removal.
figure 4

a, b, SEM (a) and AFM (b) height images for an MFI nanosheet after seed removal by mechanical rubbing. c, d, AFM three-dimensional height images for an MFI nanosheet before (c) and after (d) seed removal. Both scale bars represent 1 μm.

PowerPoint slide

Full size image

Methods

Chemicals

1,5-diaminopentane (>97%), 1-iodopropane (98%), 2-butanone (≥99.0%), potassium carbonate (anhydrous), tetrapropylammonium hydroxide (1.0 M, aqueous), sodium hydroxide (97%), potassium hydroxide (85%), silicic acid (99.9%, 20 μm), and tetraethyl orthosilicate (98%) were purchased from Sigma-Aldrich. Ethyl acetate (99.9%) and ethyl alcohol (200 proof) were purchased from Fischer Scientific. All chemicals were used as received without any further purification.

Synthesis of bis-1,5(tripropyl ammonium) pentamethylene diiodide (dC5)

dC5 was synthesized via exhaustive alkylation of 1,5-diaminopentane with 1-iodopropane, as reported previously6. In brief, 18.90 g of 1,5-diaminopentane and 82.35 g of anhydrous potassium carbonate were added to 450 ml of 2-butanone in a three-neck round-bottom flask. With a vigorous stirring, the reactor was slowly heated to 80 °C under argon atmosphere. The reactor was wrapped with aluminium foil to avoid iodide oxidation, and 108 ml of 1-iodopropane was added dropwise to the reactor. After 10 h of reaction under reflux, the reaction mixture was cooled and filtered to remove the potassium salts, and the solvent (2-butanone) in the filtrate was removed by rotary evaporation.

Purification consisted of two processes. The product was first dissolved in 250 ml of 2-butanone, and after 1 h stirring, an equal amount of ethyl acetate was added. After overnight stirring, solid powder was recovered by filtration. Additional purification was conducted with ethanol to remove KI. The recovered solid was dissolved in minimum amount of 200 proof ethanol, and then KI was removed by filtration. The product was recovered from the filtrate by using rotary evaporation. This process was repeated 4 times. The solid product was further purified with 2-butanone and ethyl acetate, followed by 200 proof ethanol, as described above. Purity was confirmed by 13C nuclear magnetic resonance (NMR).

Preparation of MFI seed crystals

MFI seed crystals were synthesized based on a two-stage varying-temperature synthesis34 with a sol composition of 10SiO2:2.4TPAOH:0.87NaOH:114H2O35. Typically, 8.93 g of 1.0 M TPAOH solution was mixed with 0.16 g of deionized water and 0.127 g of sodium hydroxide. After complete dissolution of sodium hydroxide, 2.5 g of silicic acid was added as the silica source. The mixture was stirred overnight at room temperature and then heated at 50 °C in an oil bath under static condition. After 6 days, the solution was filtered with 0.45-μm GHP (polypropylene) syringe filter, and the filtrate was heated without stirring at 100 °C in an oil bath for 3 days.

MFI seed crystals were washed with deionized water before the MFI nanosheet synthesis. The MFI crystals were collected by centrifugation at 14,500 RCF (relative centrifugal force) for 1 h followed by decantation. The solid was re-dispersed in deionized water by using ultra-sonication. After repeating the washing process twice, the solid content of the dispersion was determined based on weight after drying an aliquot of the dispersion.

MFI nanosheet synthesis

MFI nanosheets were synthesized based on the seeded growth with dC5 as SDA. Typically, a precursor sol with a composition of 80TEOS:3.75dC5:20KOH:9500H2O was hydrolysed at room temperature under air purging (50 ml min−1) to reduce ethanol content and therefore to promote MFI crystallization36. After 16 h, the precursor sol was filtered with 0.45-μm GHP (polypropylene) syringe filter and then mixed with the MFI seed crystal dispersion. The silica molar ratio of the seed suspension to the dC5 precursor sol was typically 1:200 and varied between 50 and 1000. The mixture was transferred into a Teflon-lined stainless-steel autoclave and then hydrothermally treated at 140 °C under static condition. The hydrothermal treatment time was varied from 36 h to 4 d.

MFI nanosheets shown in Figs 2 and 4 were prepared using a 36 h reaction time and possessed a non-negligible amount of amorphous silica, which was removed by KOH treatment. In brief, 1 ml of as-synthesized MFI nanosheet dispersion was mixed with 1 ml of 0.1 M KOH solution and then centrifuged at 10,000 RCF for 30 s in order to remove nanosheet aggregates. Then, 1 ml of the top solution was transferred to a new centrifuge tube and diluted to 2 ml with 0.1 M KOH solution. MFI nanosheets were collected by centrifugation at 14,500 RCF for 3 min. The acquired white slurry was re-dispersed in 2 ml of 0.1 M KOH/2 M KCl solution and kept at room temperature for 8 h to dissolve amorphous silica. MFI nanosheets were then recovered by centrifugation at 14,500 RCF for 3 min and decanted. The acquired material was sequentially rinsed with 0.1 M KOH/2 M KCl solution and then 0.1 M HNO3 /2 M KCl by re-dispersion and centrifugation, as described above. The rinsing process was repeated twice with deionized water.

MFI nanosheets synthesized using a 4-day reaction time do not contain amorphous silica and were used without KOH treatment. 1 ml of as-synthesized MFI nanosheet dispersion was diluted to 2 ml with deionized water and centrifuged at 10,000 RCF for 30 s. 1 ml of the top solution was transferred to a new micro-centrifuge tube and diluted to 2 ml with deionized water. MFI nanosheets were recovered by centrifugation at 14,500 RCF for 1 min and decantation. This rinsing process with deionized water was repeated three times. The collected MFI nanosheet slurry was then dispersed in 2 ml deionized water for characterization or 50 ml deionized water for filtration coating.

Seed removal of MFI nanosheets

MFI nanosheets were coated on Si wafer using the Langmuir–Schaefer deposition method, as reported previously37. A small amount of ethanol (5 vol%) was added to freshly-prepared MFI nanosheet dispersion (2 ml, as described above), and 300 μl of the dispersion was deposited on the surface of water in polystyrene Petri dish (35-mm diameter). Si wafer was slowly lowered and contacted with the water surface to transfer the MFI nanosheets to the Si wafer. Then, the Si wafer was tilted and lifted upward, and water remaining on the Si wafer was removed by air blow. Prepared MFI nanosheet coating was then calcined at 400 °C for 6 h at a ramp rate of 1 °C min−1. MFI nanosheets on Si wafer were then rubbed by cotton fabric to detach the seeds, followed by additional calcination under identical conditions to remove any organic contamination.

MFI membrane fabrication

MFI membrane was fabricated on porous silica support based on inter-growth of MFI nanosheets. Sintered silica fibre (SSF) supports were prepared from quartz fibre and Stober silica, as previously reported28. 1 g of freshly prepared (50 ml, as described above) MFI nanosheet dispersion was then coated on SSF supports with the vacuum-assisted filtration method15,28. During the filtration coating, vacuum was maintained above 10 psi to keep the filtration slow. The filtration was typically finished within 2 h and kept under vacuum overnight for complete drying. The MFI nanosheet coatings on SSF supports were calcined at 400 °C for 6 h at a ramp rate of 1 °C min−1 under 100 ml min−1 of air flow after each coating step. Filtration coating was repeated until no vacancy was observed by SEM analysis.

Continuous MFI membranes were prepared by gel-free growth of MFI nanosheets on SSF supports27. In brief, nanosheet-coated SSF supports were impregnated with 0.025 M TPAOH solution and placed in a Teflon-lined stainless-steel autoclave after excess solution on the side and bottom of the support was removed by Kimwipes. The autoclave was then heated at 180 °C for 2 d. The resultant membrane was rinsed with deionized water and dried at 70 °C. Before xylene isomer vapour permeance measurements, the membrane was calcined at 450 °C for 8 h at a ramp rate of 1 °C min−1.

Characterization

TGA was conducted using a Shimadzu TGA-50 analyser. Samples were dried at 120 °C for 8 h under nitrogen atmosphere, and TGA profiles were recorded in air flow (100 ml min−1) with a ramp rate of 1 °C min−1. Argon porosimetry was performed at 87 K by using Quantachrome AutosorbiQ MP. Prior to measurements, the samples were outgassed and heated at 573 K for 16 h under vacuum. The samples were analysed for adsorption and desorption ranged from 10−7 to 1 of P/P0. SEM images were recorded by using a Hitachi SU 8230 or Hitachi S-4700 operated at 1.5 kV. High-resolution SEM images were acquired in a deceleration mode of Hitachi SU 8230 with 0.8-kV landing voltage. Height profiles of calcined MFI nanosheets on Si wafers were acquired in a tapping mode under ambient condition by using Bruker Nanoscope V Multimode 8. Focused-ion-beam milling was performed with a FEI Quanta 200 3D, in order to make a trench on the membrane for cross-sectional measurements. XRD patterns of MFI membranes were recorded by using a PANalytical X’Pert Pro diffractometer with Cu Kα radiation. High-resolution powder diffraction data were collected at beamline 17-BM of the Advanced Photon Source with a monochromatic beam of 0.72768 Å at Argonne National Laboratory. Powder samples were prepared by freeze-drying with the Labconco FreeZone 4.5 l Benchtop Freeze-Dry System. Collected data were processed with GSAS II38 and converted to 2θ values corresponding to Cu Kα radiation. Pawley fitting was employed to extract lattice parameters using GSAS II. The in-plane XRD measurements were performed at beamline 33-BM-C at the Advanced Photon Source (0.8267 Å), Argonne National Laboratory. The instrumentation consists of a bending magnet source with a Si(111) monochromator with a 0.9 × 0.5 mm beam spot. The sample was placed on a Huber 4-circle stage with the sample held almost parallel to the incident beam, and the detector was moved in the plane of the sample. 2θ values were then converted to those corresponding to Cu Kα radiation.

Conventional TEM was performed on a FEI Tecnai G2 F30 (S)TEM with TWIN pole piece, a Schottky field-emission electron gun operating at 300 kV and equipped with a Gatan 4k × 4k Ultrascan CCD. Imaging and diffraction data collection were performed under low electron dose to minimize electron beam damage of the zeolite sample. HAADF-STEM images were acquired in an aberration-corrected FEI Titan 60-300 (S)TEM, operating at 200 kV and having a STEM incident probe convergence angle of 24 mrad with 20 pA screen current and about 50 mrad HAADF detector inner angle. MFI nanosheet samples were prepared for TEM measurement by drop-casting an aqueous suspension of the MFI nanosheets on TEM grids (ultrathin carbon film on holey carbon support film, 400 mesh Cu, Ted Pella). The grid was dried at room temperature before imaging in the TEM. An electron transparent TEM sample of the membrane cross-section was prepared by focused-ion-beam milling using a FEI Quanta 200 3D instrument. The sample was first coated with a platinum layer about 5 μm thick, and polishing of the cross-section was performed using a 10-pA focused ion beam to minimize damage to the membrane by the ion beam.

Theoretical calculations

The model membrane system consists of 1 × 3 × 1 unit cells of the MFI structure, with periodic replication in the a axis and c axis directions and the two {010} surfaces truncated, and 1–6 dC5 cations. The exposed oxygen atoms were terminated as surface silanol groups, and 30 Å vacuum space was added to the b-axis direction. To maintain charge neutrality of the system, an appropriate number of deprotonated silanol defects (SiO−) was created to compensate for the SDA cations. Periodic Kohn–Sham density functional theory calculations were performed using the Vienna Ab initio Simulation Package, version 5.4.139, with the PBE exchange-correlation functional40, Grimme D2 dispersion corrections41, the valence electron density expanded in a plane-wave basis set using a kinetic energy cutoff of 400 eV, and the core electrons described by the projected-augmented wave method39. Sampling of the unit cell was carried out at the Γ-point. During the geometry optimizations, the cell shape was allowed to change but the total volume was kept constant. The vacuum space present in the supercell permits the MFI membrane to swell or contract freely.

To estimate the ratio of self-diffusion coefficients along the straight channel for the p-xylene/o-xylene pair, first principles (FP) and force-field-based (FF) molecular dynamics simulations (using CP2K, version 3.042, and GROMACS, version 2016.143) in the canonical ensemble in conjunction with umbrella sampling and weighted histogram analysis methods44 were carried out. The FP molecular dynamics simulations considered a 3-nm MFI nanosheet (1.5 unit cells terminated by silanol groups) and used the PBE exchange-correlation functional40 and Grimme D3 dispersion correction45. The FF molecular dynamics simulations probed a bulk MFI sample (represented by a periodic simulation box consisting of 2 × 2 × 3 unit cells), and the TraPPE force field was used to describe xylene–xylene and xylene–zeolite interactions with harmonic potentials allowing for flexibility of sorbate and framework46.

Membrane testing

Xylene isomer vapour permeance measurements were conducted in the Wicke–Kallenbach mode, in which total pressures of feed and permeate are maintained at atmospheric pressure47. Feed comprises approximately equimolar mixture of p-xylene (about 0.5 kPa) and o-xylene (about 0.5 kPa), and helium was used as a carrier and sweep gas. The compositions of feed and permeate were assessed with a gas chromatograph (Agilent, 7890B) equipped with a flame ionization detector and a capillary column (DB-WAXetr, Chrometech). The permeance is defined as the flux divided by the partial pressure gradient. The separation factor is defined as the molar ratio of isomers in the permeate divided by the molar ratio of isomers in the feed. The separation performances of identically prepared membranes were also evaluated for the n-/i-butane mixture28 and the ethanol/water mixture48, as reported previously.

Data availability

The data sets generated during and/or analysed during the current study are available from the corresponding author on reasonable request.