Embryonic zeolite-mediated suture synthesis of thin and scalable zeolite membranes for tailored gas separation

Abstract

Zeolite membranes exhibit considerable potential for gas separation; however, two critical challenges (low permeation flux and scaling-up fabrication) continue to hinder their practical implementation. Here, we propose an embryonic zeolite-mediated suture (EZMS) strategy to synthesize large-area zeolite membranes with stable gas separation performance. The membrane thickness is equivalent to that of the initial seed layer—a feature validated across three distinct zeolite frameworks (STT, CHA, and MFI). For high-silica CHA (also known as SSZ-13) zeolite membranes, the EZMS strategy enables a 5-fold thickness reduction, yielding a CO₂ permeance of 1.02 × 10^-6mol·m^-2·s^-1·Pa^-1 (3000 GPU) and a CO₂/CH₄ selectivity of 158 at 0.2 MPa. Scalability is validated by the successful synthesis of SSZ-13 zeolite membrane bundles with an individual area of 0.5 m² (40 cm in length). The membranes exhibited excellent high-pressure resistance (>4 MPa) and long-term stability (>220 d) for humid CO₂/CH₄ separation, representing a high-performance benchmark for biogas upgrading. The reliable synthesis protocol and improved performance highlight the industrial application potential of zeolite membranes.

Here authors present an embryonic zeolite-mediated suture synthesis, which enables the scalable fabrication of high-flux SSZ-13 zeolite membrane bundles, showing >4 MPa pressure resistance and >220 days stability for humid CO₂/CH₄ separation, highlighting high industrial potential for biogas upgrading.

Introduction

Membrane gas separation is a compelling energy-efficient alternative to amine absorption and cryogenic distillation for natural gas upgrading. Polymeric membranes dominate current commercial applications; however, their susceptibility to plasticization under high partial pressures of CO₂ and H₂S severely limits their lifetime and separation efficiency^1,2. In contrast, polycrystalline membranes—such as metal-organic frameworks (MOFs)^3,4, covalent organic frameworks (COFs)⁵ and zeolites⁶—are promising candidates to overcome these limitations, owing to their inherent chemical and mechanical robustness. Moreover, polycrystalline membranes show great potential for helium purification and hydrocarbon isomer separation⁷. Nevertheless, their practical application is hindered by low permeation flux and scalable manufacturing challenges¹.

The permeation flux is inversely proportional to thickness according to the generalized Maxwell-Stefan model⁸. To prepare ultrathin polycrystalline membranes, novel interfacial synthesis methods and two-dimensional (2D) nanosheet building blocks have been explored^9–11. For instance, the Zr-fum₆₇-mes₃₃-fcu-MOF membrane thickness was reduced to 30 nm via electrochemical synthesis, leading to an improved N₂ permeance of 3057 GPU (Gas Permeation Unit, 1 GPU = 3.3928 × 10⁻¹⁰mol·m⁻²·s⁻¹·Pa⁻¹)⁹. Similarly, the thickness of 2D CuBDC membranes was reduced to <10 nm by precisely regulating the stacking of MOF nanosheets^10,11. However, the fabrication of these ultrathin polycrystalline membranes is extremely complicated, as conductive Pt coating or nanosheet exfoliation is essential⁹. Furthermore, suitable 2D zeolite nanosheets for gas separation membranes are significantly scarce. Only ten-membered ring (10MR) MFI¹² and six-membered ring (6MR) RUB-15¹³ zeolite nanosheets have been utilized as building blocks; meanwhile, they are retransformed into a three-dimensional structure following secondary growth or high-temperature calcination. Alternatively, nano-seeds are more suitable for synthesizing thin zeolite membranes via dry gel conversion¹⁴, gel-less¹⁵ or hot-dipping methods¹⁶. The final membrane thickness is generally tenfold higher than that of the seed layer due to outward intergrowth from the substrates following the classical crystallization pathways¹⁷. Therefore, preparing large-area gas separation zeolite membranes with a thickness of <1 μm remains extremely challenging^2,18.

Various zeolite membrane geometries have also been reported to enhance permeation flux, such as hollow fibers (HFs)¹⁹, single tubes²⁰, and multi-channel monoliths^21–23. However, all these membranes have a limited effective area for each batch synthesis. Lee et al. fabricated six 12-cm-long SAPO-34 zeolite membranes in one autoclave, with a reproducibility error of 10%²⁴. Recently, we successfully synthesized nine 40-cm-long hollow fiber SSZ-13 zeolite membranes in one autoclave using interzeolite-converted seeds. However, CO₂ permeance was only 1200 GPU due to the relatively thick zeolite membrane (2.8 μm)¹⁸. Nair et al. prepared twenty 6.5-cm-long hollow fiber SSZ-13 zeolite membranes in one batch, but no reproducibility in gas separation performance was reported¹⁹. These limitations originate from the classical layer-by-layer mechanism intrinsic to the conventional secondary growth method²⁵, wherein deposited seeds grow via the conversion of amorphous precursors—an inherently restrictive process creating an unavoidable trade-off between thickness control and defect elimination²⁶. Therefore, the lack of a reliable synthesis protocol for large-area gas separation zeolite membranes has become a critical barrier to their successful practical implementation.

Herein, we present an embryonic zeolite-mediated suture strategy (Fig. 1), enabling the scalable fabrication of thin and highly selective zeolite membranes for tailored gas separation. The amorphous precursors are first converted into short-range ordered embryonic zeolites, and these highly reactive intermediates facilitate the seamless suturing of ball-milled seed particles via rapid condensation reactions. This method deliberately avoids classical crystallization pathways, which typically result in unregulated membrane thickening and defect formation in zeolite membranes. This alternative strategy not only maintains the geometric integrity of the seed layer to ensure precise thickness control but also minimizes defect generation, thereby simultaneously enhancing permeation flux and selectivity. We demonstrate the universality of the EZMS strategy across three distinct zeolite topological frameworks (7MR and 9MR STT, 8MR CHA, and 10MR MFI) for He/CH₄, CO₂/CH₄, and n-/i-butane mixture separation, respectively. The scalability is further validated through the batch synthesis of large-area SSZ-13 zeolite membrane bundles. We also evaluated the separation performance and long-term stability for the integrated removal of CO₂, H₂S, and moisture in biogas upgrading.

Fig. 1. Illustration of embryonic zeolite-mediated suture strategy.

In pre-assembly period, silicon (Si), aluminum (Al) precursors and organic structure directing agent (OSDA) were initially assembled into embryonic zeolites, then participated in the condensation reactions with the pre-deposited ball-milled seed particles, ultimately stitching the discrete into continuous and highly selective zeolite membranes.

Results and discussion

Zeolite membrane synthesis and characterization

Three distinct zeolite membranes (STT, CHA, and MFI) were synthesized using the EZMS strategy. Ball-milled seeds were uniformly coated onto four-channel hollow fibers through a single-step dip-coating method (Fig. S1). After secondary growth, all membranes exhibited a well-intergrown surface (Fig. 2a, c, e), a typical morphology of high-selective zeolite membranes¹⁸. Unlike conventional zeolite membranes that typically exhibit a ~10-fold thickness amplification during hydrothermal synthesis^2,17, the final thickness of our membranes remained nearly identical to their initial seed layers (Fig. 2b, d, f). To further confirm EZMS strategy’s thickness control capability, a total of five SSZ-13 zeolite membranes were synthesized using seed layers with thicknesses ranging from 540 ± 30 nm to 1600 ± 100 nm (Table S1 and Fig. S2). All these membranes exhibited a final thickness consistent with their initial seed layers. This result confirms EZMS strategy is effective in controlling membrane thickness over a broad range.

Fig. 2. Characterization of three distinct zeolite membranes synthesized via EZMS strategy.

SEM images of STT (a, b), SSZ-13 (c, d), and MFI (e, f) zeolite membranes; insert: topological framework of the corresponding zeolite. g Gas separation performance towards tailored gas pairs of He/CH₄ for STT, CO₂/CH₄ for SSZ-13, and n-/i-C₄H₁₀ for MFI zeolite membranes. Synthesis conditions: STT and SSZ-13 zeolite membranes were synthesized at 180 °C for 24 h; MFI zeolite membranes were synthesized at 160 °C for 14 h. Gas separation test conditions: Feed gas was a 200 mL·min⁻¹ equimolar binary mixture at 298 K and 0.2 MPa for all membranes; a 50 mL·min⁻¹ He sweep gas was applied at atmospheric pressure for STT and SSZ-13 zeolite membranes, and no sweep gas was used for MFI zeolite membranes. h Effect of synthesis period on CO₂/CH₄ separation performance of SSZ-13 zeolite membranes synthesized at 180 °C. Error bars represent the standard deviation from three independent measurements.

To investigate the kinetics of EZMS-derived zeolite membranes formation, the evolution of a 950 nm-thick SSZ-13 zeolite seed layer was monitored via time-resolved SEM and elemental analysis of Si/Al ratio (SAR). During the early synthesis stage (0.5–2.5 h), the substrate was covered with irregular aggregates (thickness: 1.0 ± 0.2 μm, Fig. S3a–f). The SAR was lower than both the initial seed SAR (50, Fig. S4) and the precursor SAR (105, Fig. S5), which is attributed to gaps between discrete seed particles. These gaps exposed the Al-rich substrate, leading to the inclusion of extra Al signals in the analysis and thereby reducing the measured SAR. When the synthesis duration exceeded 4 h, irregular aggregates gradually transformed into well-defined cubic crystals (Fig. S3g, h). This crystallization filled the interparticle gaps, eliminating the substrate-derived Al signal. Consequently, the membrane SAR finally stabilized at ~40, consistent with the typical high-silica CHA zeolites²⁷. Intriguingly, in the subsequent crystallization stage (4–24 h), both surface crystal size and membrane thickness remained constant (Figs. S3i–l and Fig. S6). This observation further proves the EZMS strategy suppresses uncontrolled overgrowth and achieves precise thickness control.

Membrane quality was further evaluated through binary gas mixture tests tailored to the topological feature of each zeolite membrane. The STT, SSZ-13, and MFI zeolite membranes exhibited selectivities of 275 for He/CH₄, 158 for CO₂/CH₄, and 56 for n-/i-C₄H₁₀ binary mixture separation, respectively (Fig. 2g). These values are significantly higher than the corresponding Knudsen selectivities, confirming that the interzeolite gaps have been effectively eliminated and the gas separation is primarily governed by the intrinsic zeolitic channels. Subsequently, an equimolar CO₂/CH₄ mixture was used to probe the SSZ-13 zeolite membrane evolution with synthesis time (Fig. 2h) and thickness. Initially, the CO₂/CH₄ selectivity was relatively low, which is another evidence of gaps between seed particles and precursor species. The selectivity exceeded 100 and plateaued once the synthesis time extended to 4 h. This result proves a critical period of 4 h for SSZ-13 zeolite membrane synthesis, which is reduced by 83% compared to the conventional synthesis protocol¹⁸. CO₂ permeance increased significantly with decreasing membrane thickness (Table S2). The 560-nm-thick membrane achieved a CO₂ permeance of 1.02 × 10⁻⁶mol·m⁻²·s⁻¹·Pa⁻¹ (3000 GPU) and a CO₂/CH₄ selectivity of 158. The performance not only far exceeds Robeson’s upper bound but also ranks among the state-of-the-art for CO₂/CH₄ separation membranes (Figs. S7 and S8)^28,29.

Zeolite membrane formation mechanism

To elucidate the thickness control mechanism of the EZMS strategy, the precursor’s pre-assembly period (t₁ = 0, 12, 84 h) was significantly investigated. A Q⁴Si(1Al) signal at −104.7 ppm was instantaneously observed in ²⁹Si MAS NMR spectra (Fig. 3a), indicating a rapid condensation reaction between silicon sources and aluminum species, which is the initial step of embryonic zeolite formation³⁰. With an extended pre-assembly period to 84 h, the peak area of Q⁴Si(2Al) signal at −99.2 ppm increased monotonously and reached 25.4% (Fig. S9). This result confirms that more aluminum atoms are incorporated into 4MRs following Lowenstein’s rule³¹. The formation of quaternary ring structures was further verified through electrospray-ionization mass spectrometry (ESI-MS, Fig. S10)^32,33. In the ²⁷Al MAS NMR spectra (Fig. 3b), the Al signal shifted from 61.1 ppm to 53.3 ppm. The result confirms an increased Al-O-Si bond angle in the precursor, which is favorable for the formation of embryonic zeolite³⁰. Only the tetrahedral-coordinated Al signal (no octahedral Al at ~0 ppm) is detected, confirming that Al species are uniformly integrated into the zeolite framework. Additionally, the narrowed full width at half-maximum (FWHM) of both the Q⁴Si(2Al) and tetrahedral Al peaks indicated the formation of more ordered precursor species³⁴.

Fig. 3. Effect of pre-assembly period (t₁) on precursor evolution.

a, b ²⁹Si and ²⁷Al MAS NMR spectra of the gel. c ATR-FTIR spectra of the gel, the yellow region corresponds to the 4MR vibration, while the red region indicates the d6r vibration. d TEM and HR-TEM of the gel with pre-assembly period of t₁ = 84 h, e ¹³C CP MAS NMR spectra of the gel. Inset: molecular structure of TMAdaOH, with carbon atoms at distinct chemical environments labeled by numbers. f OSDA/SiO₂ of the gel calculated by TGA. Inset: schematic illustration of the local environment in the gel, where yellow spheres with tails represent TMAdaOH molecules, blue triangles represent Si atoms, yellow triangles denote Al atoms, and red spheres indicate O atoms.

To further characterize the ordered structure in the precursor, ATR-FTIR (Fig. 3c) and FTIR (Fig. S11) analysis were performed. The peak at 1060 cm⁻¹ is assigned to the asymmetric vibration of T-O bonds, which is well explained by the incorporation of Al atoms into the frameworks^35,36. The peaks at 652 and 629 cm⁻¹ are assigned to the 4MR-related structures, which are consistent with the ²⁹Si MAS NMR and ESI-MS results. With an 84 h pre-assembly period, the increase in peak intensity at 543 cm^-1 revealed the assembly of 4MRs into double six-membered rings (d6r), which is a key intermediate structure for CHA zeolite framework formation³⁵. Since the UV Raman signals at 472 and 319 cm⁻¹ overlapped with those derived from TMAda⁺ cations (Fig. S12)³⁶, the increased intensity was assigned to newly formed 4MRs and 6MRs. Additionally, the enhanced signal at 197 cm⁻¹ further revealed the formation of cha cages³⁷. In the high-resolution transmission electron microscopy (HR-TEM) and selected area electron diffraction (SAED, Fig. 3d and Figs. S13 and S14), an electron diffraction ring associated with (3 3 3) crystal plane of CHA zeolite was observed. However, the precursor remained amorphous in powder X-ray diffraction (PXRD) analysis (Fig. S15). The result confirms the absence of long-range crystalline order, which is a defining structural characteristic of embryonic zeolites as reported³⁸.

To probe the role of TMAda⁺ in cha cage formation, the precursor was analyzed by thermogravimetric analysis (TGA) and solid-state ¹³C CP MAS NMR spectroscopy. The weight loss of crystalline SSZ-13 zeolite is primarily triggered at temperatures above 350 °C (Fig. S16). While the precursor exhibited predominant weight loss between 100 and 250 °C, which is due to evaporation and decomposition of water and TMAda⁺. Intriguingly, a secondary weight loss was observed between 300 and 550 °C, which is due to the removal of TMAda⁺ trapped in the ordered structures (including cha cages)³⁹. The solid-state ¹³C CP MAS NMR spectra demonstrated a shift to a lower magnetic field, confirming the TMAda⁺ cations were more tightly enclosed in the ordered structures (Fig. 3e)³⁶. The amount of trapped TMAda⁺ was quantified by calculating the molar ratio of trapped OSDA to SiO₂ (Fig. 3f), reaching 2.6% after an 84 h pre-assembly period. This value is 3.4 times higher than that in the initial precursor (without pre-assembly) but far lower than that in fully crystalline SSZ-13 zeolite (8.3%)⁴⁰. This result indicates that the pre-assembly promotes TMAda⁺ encapsulation in the ordered structures, though the precursor has not yet formed a well-crystallized zeolite.

The interaction between the seeds and the precursor was further investigated to elaborate on the zeolite membrane formation mechanism. Well-intergrown and highly selective membranes were successfully prepared via the aforementioned EZMS strategy, whereas only sparse crystals were observed when using the conventional dry gel method (Fig. S17). The short pre-assembly period of t₁ = 0 h and t₁ = 12 h caused worse intergrowth (Figs. S18 and S19) and significantly reduced CO₂/CH₄ selectivity (Fig. S20 and Table S3). Both results confirmed that sufficient pre-assembly is essential to generate ordered structures (e.g., 4MR, d6r, CHA-type cages), which are required for effective integration of the ball-milled seed particles. This explains why conventional hydrothermal synthesis typically requires a long duration of 96 h⁴¹. The CO₂/CH₄ selectivity of membranes prepared using hydrothermally synthesized seeds was merely 24 (Fig. S21), which is five times lower than that of membranes using ball-milled seeds. Furthermore, membranes synthesized via the in-situ method (Table S4) were even non-selective towards CO₂/CH₄ separation, which is attributed to poor intergrowth (Fig. S22). These membranes also exhibited slightly lower permeance, likely due to insufficient crystallization that left amorphous substance blocking the zeolitic channels³⁰. Following this line, we propose an integration mechanism for EZMS-derived membranes, wherein the ball-milled seed particles featuring abundant Si–OH groups⁴² are firstly coated onto porous substrates, serving as anchoring sites for the ordered structures (e.g., cha cages, 4MR, 6MR, and d6r) in the pre-assembled precursor. The ordered precursor structures then bridge the fractured surfaces of adjacent seed particles via condensation reactions between their silanol groups and those on the seed surface, forming a continuous and defect-minimized zeolite membrane.

Large-area zeolite membrane synthesis optimization

To validate the feasibility of scaling up, we successfully synthesized 40-cm-long hollow fiber membranes (Fig. 4a). Our method led to a uniform layer thickness of 2.5 ± 0.6 μm along the axial direction (Fig. S23), which is also identical to that of the seed layer (Fig. S24). This contrasts sharply with the conventional method⁴¹, which produced a significant thickness gradient from 3 μm at the top to 100 μm at the bottom (Fig. S25). The membrane quality was evaluated using an equimolar CO₂/CH₄ mixture at 0.3 MPa. The conventionally synthesized membranes exhibited a significant decline in selectivity from 149 ± 84 at the top to 18.5 ± 16.5 at the bottom along the axial direction (M11–M14, Table S5). However, the entire membrane synthesized using our EZMS strategy showed exceptional uniformity of CO₂/CH₄ selectivities exceeding 150 along the entire axial direction (M15–M18, Table S5). Subsequently, the scaled-up synthesis was extended from a single HF membrane to membrane bundles, each composed of 17, 33, and 102 HF membranes with effective areas of 0.08, 0.15, and 0.5 m², respectively. To the best of our knowledge, this latter is the largest SSZ-13 zeolite membrane for gas separation reported to date, which is approximately twentyfold larger than a single tube (0.019 m²)⁴³ or multi-channel geometry membranes (0.024 m²)²³. The integration of the 0.5 m² membrane element into stainless-steel housing is demonstrated in Supplementary Video 1. All the membranes and membrane bundles exhibited a CO₂/CH₄ selectivity above 150 (Fig. 4b and Table S6). Furthermore, more than a dozen SSZ-13 zeolite membrane bundles, each with an effective area of 0.5 m², were fabricated to confirm superior reproducibility and scalability (Fig. S26).

Fig. 4. Large-area zeolite membranes and zeolite membrane bundles synthesized via the EZMS strategy.

a Photo of single membrane (Number 1, 2) and the membrane bundles composed of 17 (Number 3), 33 (Number 4), and 102 (Number 5) single membranes. b Binary equimolar CO₂/CH₄ mixture separation performance at room temperature and 0.3 MPa (no sweep gas used). Error bars represent the standard deviation of four independent measurements.

Biogas upgrading performance evaluation

The membrane bundle composed of 33 zeolite membranes was tested for single gas permeation of He, H₂, CO₂, N₂, CH₄, and SF₆ at 0.3 MPa. The CO₂ permeance was significantly higher than that of the other gases and decreased with increasing kinetic diameter of the gas molecules (Fig. S27). The CH₄ permeance remained constant even when the pressure was increased to 1.1 MPa. Meanwhile, the bulky SF₆ molecules (kinetic diameter: ~0.55 nm) primarily permeate through the defective pores, exhibiting a low permeance of 1.2 × 10⁻¹⁰mol·m⁻²·s⁻¹·Pa⁻¹ (0.36 GPU). Both results confirm that zeolitic pores dominate the gas transport pathway⁴⁴. This leads to a superior high selectivity of 16 for CO₂/N₂, 170 for CO₂/CH₄, and 1380 for CO₂/SF₆, which is comparable to those of a single HF membrane⁴¹. This consistency proves the membrane bundle has minimal defects, maintaining the superior separation performance even after scaling up.

The separation performance was evaluated via stepwise pressurization of a binary CO₂/CH₄ mixture up to 4.0 MPa. Both CO₂ permeance and CO₂/CH₄ selectivity decreased with increasing pressure (Fig. 5a), which is a typical phenomenon of the adsorption-diffusion mechanism for zeolite membranes⁴⁵. Importantly, the membrane exhibits fully reversible performance when depressurization back to ambient pressure, confirming the mechanical robustness. A comparison of different feeding directions shows that feeding from the substrate side (tail) results in lower CO₂ permeance and CO₂/CH₄ selectivity (Fig. S28a and Fig. 5b). This is attributed to severe concentration polarization in the support pores, which hinders efficient mass transfer to the selective membrane layer. Thus, feeding to the selective membrane layer side (head) is more favorable for the practical natural gas upgrading processes⁴⁶.

Fig. 5. Biogas upgrading performance of large-area SSZ-13 zeolite membrane bundles.

a High-pressure CO₂/CH₄ separation performance at a feed flowrate of 6 L·min⁻¹. b CO₂ permeance and CO₂/CH₄ mixture selectivity at 2.0 MPa feed to the zeolite membrane side (head, blue area) and to the porous support side (tail, red area). c Photo of a biogas upgrading system composed of ten zeolite membrane bundles. d Long-term stability for dry (yellow area) and humid (blue area) biogas upgrading. e Desulfurization of the biogas containing 2500 ppm H₂S. f Pressurization−depressurization cycle stability between 0.1 and 2.0 MPa. The biogas containing 43% CO₂: 55% CH₄ was provided by Xuzhou Huanneng Environmental Energy & Ecology Technology Co., Ltd., wherein the H₂S concentration ranged between 1000 ppm and 3000 ppm.

To further validate practical applicability, a pilot-scale biogas upgrading test was conducted using ten zeolite membrane bundles (Fig. 5c). Biogas containing 43% CO₂ from the fermentation tank was pressurized to 2.0 MPa and fed to the selective membrane side (head). For dry biogas feed, the average CO₂ permeance stabilized at ~4 × 10⁻⁸ mol·m⁻²·s⁻¹·Pa⁻¹ (~120 GPU) (Fig. 5d), which is only half of the value observed for a single bundle in laboratory tests. This is attributed to a high stage-cut of 40%, over 80% CO₂ permeated through the membranes, resulting in strong competitive CH₄ adsorption⁴⁷. The IAST calculation further revealed that up to 55% of the transport pathways in zeolite membranes near the retentate outlet were occupied by CH₄ molecules (Fig. S29), directly inhibiting CO₂ adsorption and diffusion⁴⁸. When the biogas was saturated with water vapor (partial pressure: 5.7 kPa, relative humidity = 100%), the CO₂ permeance further decreased to ~2 × 10⁻⁸mol·m⁻²·s⁻¹·Pa⁻¹ (~60 GPU), indicating an extra drying procedure is required to keep the maximum permeance. The membrane simultaneously achieved water vapor removal with a permeance of 2.3 × 10⁻⁹mol·m⁻²·s⁻¹·Pa⁻¹ (6.87 GPU), as evidenced by the reduced water partial pressure (1.9 kPa, relative humidity = 33%) in the retentate. The membrane maintained stable performance over a long-term test of >220 days, demonstrating superior resistance to water vapor, a high partial pressure of CO₂ and H₂S. The H₂S concentration in the retentate decreased from 2500 ppm to 300 ppm (Fig. 5e). These results validate the triple functions of zeolite membranes, which have high potential to eliminate the complicated pretreatment process (e.g., dehydration, desulfurization) required for commercial polymeric membranes⁴⁹. Furthermore, the CO₂/CH₄ selectivity remained stable at ~22 throughout 43 pressurization-depressurization cycles (Fig. 5f), confirming the mechanical robustness under industrially relevant biogas upgrading conditions.

In summary, we report an embryonic zeolite-mediated suture strategy for the thickness-controlled synthesis of highly selective zeolite membranes for tailored gas separation. This approach leverages the intrinsic reactive suturing capacity of embryonic zeolites to seamlessly integrate ball-milled seed particles into continuous and defect-minimized zeolite membranes. The membrane thickness is precisely controlled by the initial seed layer, eliminating the uncontrolled overgrowth that generally occurs in the conventional hydrothermal synthesis. This approach enabled the fabrication of thin (0.56 μm) SSZ-13 zeolite membranes that simultaneously achieve high CO₂ permeance (3000 GPU) and CO₂/CH₄ selectivity (158), which represents one of the state-of-the-art benchmarks for CO₂/CH₄ separation. Beyond lab-scale investigation, the EZMS strategy is applied to successfully synthesize zeolite membrane bundles composed of over 100 membranes (total area >0.5 m²), all exhibiting consistent separation performance and exceptional long-term stability (>220 days) under industrially relevant biogas upgrading. SSZ-13 zeolite membranes realized the simultaneous removal of CO₂, H₂S, and H₂O, eliminating the pretreatment steps mandatory for polymeric membranes. This reliable and scalable synthesis protocol, coupled with stable separation performance, contributes to the industrial deployment of zeolite membranes for gas separation.

Methods

Synthesis of SSZ-13 zeolite seeds

Typically, NaOH (Sinopharm Chemical Reagent Co., Ltd, AR) and Al(OH)₃ (Shanghai Lingfeng Chemical Reagent Co., Ltd., AR) were added into deionized water at room temperature and stirred for 2 h. Then a certain amount of TMAdaOH (25% in water) was added. Finally, colloidal silica was added dropwise into the solution. The final molar composition was 20 TMAdaOH: 16 NaOH: 0.5 Al₂O₃: 100 SiO₂: 4400 H₂O. The solution was transferred to a polytetrafluoroethylene (PTFE) lined autoclave and placed in an oven at 433 K for 96 h. After hydrothermal synthesis, the SSZ-13 crystals were recovered by centrifugation at 12840 g for 10 min and washed three times with deionized water. The powders were dried overnight at 333 K and calcined at 823 K for 24 h to remove the OSDA. The ball-milled seeds were prepared by ball-milling the pristine SSZ-13 zeolite with a planetary ball mill (Retsch Ind. Ltd., PM-100), using zirconium oxide balls as grinding media. The milling treatment was operated for 3 h at a rotating speed of 450 rpm.

Preparation of seed layers with different thicknesses

The four-channel α-Al₂O₃ hollow fibers were provided by Jiangsu Shining HiTech Co., Ltd., have outer/inner diameters (O.D./I.D.) of 3.8 mm/0.9 mm, porosity of ~40% and average pore size of 300 nm. Before seeding, they were ultrasonically cleaned for 2 h in deionized water and dried at 343 K overnight. The external surface was seeded with ball-milled SSZ-13 zeolite crystals by the dip-coating method. The top section of the hollow fibers was capped with Teflon tape, then they were vertically dipped into the seed suspensions as shown in Table S1 to prepare the seed layer with different thicknesses. The hollow fibers were dip-coated in an aqueous 0.5 wt% seed suspension for 15 s twice to get the seed layer with a thickness of ~2.5 μm. To reduce the seed layer thickness, a hollow fiber was vertically dipped into an aqueous 0.5 wt% seed suspension for 15 s to achieve a seed layer thickness of ~900 nm. Similarly, dipping the hollow fiber into the same suspension for 10 s resulted in a seed layer thickness of around 800 nm. When the hollow fiber was dipped into an aqueous 0.4 wt% seed suspension for 10 s, the seed layer thickness was about 700 nm. Finally, dipping the hollow fiber into an aqueous 0.3 wt% seed suspension for 15 s produced a seed layer with a thickness of ~560 nm. The seeded hollow fibers were dried at 343 K overnight before the membrane synthesis.

For the hollow fiber bundles, 17, 33, and 102 pieces of hollow fiber supports were assembled into a bundle unit with a dead-end flow configuration. One end of the unit is sealed with glaze, and the other end is open for gas flow. The effective length of the hollow fiber support is 40 cm.

Synthesis of SSZ-13 zeolite membranes

The precursor for membrane synthesis was prepared following the same procedure as the pristine SSZ-13 zeolite crystals. The molar composition was 20 TMAdaOH: 16 NaOH: 0.5 Al₂O₃: 105 SiO₂: 4400 H₂O. The precursor was aged at room temperature or 368 K in an oven for different periods. Then, the seeded hollow fibers or hollow fiber bundles, as mentioned above, were transferred to the precursor. The hydrothermal synthesis was conducted at 453 K for a certain period. After the crystallization at 453 K, the membranes were washed with DI water and dried overnight at 373 K. The template was removed by calcination in an ozone/oxygen atmosphere at 473 K for 72 h with a heating and cooling rate of 0.5 K·min⁻¹. Ozone was supplied by an electrical discharge generator (Lvbang, NPO50P–S–2). The ozone/oxygen gas mixture was fed at a flowrate of 4 L min⁻¹.

Synthesis of STT zeolite membranes

The STT-type zeolite membrane was prepared from a synthesis solution with a molar composition of 1 SiO₂: 0.2 TMAdaOH: 44 H₂O. The precursor was aged at room temperature or at 368 K in an oven for 72 h. The hydrothermal synthesis was conducted at 453 K for 24 h.

Synthesis of MFI zeolite membranes

The precursor solution for MFI-type zeolite membrane synthesis was prepared with a molar composition of SiO₂: TPAOH: H₂O = 1: 0.24: 178. Deionized water and tetrapropylammonium hydroxide (TPAOH) were first added to a PTFE bottle and stirred until homogeneous. Tetraethyl orthosilicate (TEOS) was then slowly added dropwise under continuous stirring. The precursor was aged at room temperature or 368 K in an oven for 48 h. The hydrothermal synthesis was conducted at 433 K for 14 h.

Gas separation test

The single gas permeation was tested using a transient permeation apparatus without sweep gas. The feed pressure was modulated by a back-pressure regulator, and the permeate side was operated under atmospheric pressure. A gas chromatography (GC, 8860 Agilent Technologies), equipped with a thermal conductivity detector (TCD) and a ShinCarbon column, was used to analyze gas compositions. The permeance of component i (P_i), ideal selectivity (S_i/j), and mixture selectivity (α_i/j) of component i over component j is defined by the following equation:

$$P_i=\frac{J_i}{\mathrm{\Delta }{p}_i}$$ 1

$$S=\frac{P_i}{P_j}$$ 2

$${\alpha }_{ij}=\frac{y_i/y_j}{x_i/x_j}$$ 3

Where J_i is the permeation flux through the membrane, mol·m⁻²·s⁻¹; x_i (x_j) and y_i (y_j) are the molar fraction of component i (j) in feed and permeate, respectively. Δp_i is the trans-membrane partial pressure difference for component i (Pa). Considering the variation of gas composition along the membrane, we used the logarithmic average to calculate (Δp_i)_ln (Pa)⁴⁵:

$${\left(\mathrm{\Delta }{p}_{\mathrm{i}}\right)}_{\ln }=\frac{\left(p_{\mathrm{i}}^{\mathrm{feed}}-p_{\mathrm{i}}^{\mathrm{perm}}\right)-\left(p_{\mathrm{i}}^{\mathrm{ret}}-p_{\mathrm{i}}^{\mathrm{perm}}\right)}{\ln \left(\frac{p_{\mathrm{i}}^{\mathrm{feed}}-p_{\mathrm{i}}^{\mathrm{perm}}}{p_{\mathrm{i}}^{\mathrm{ret}}-p_{\mathrm{i}}^{\mathrm{perm}}}\right)}$$ 4

The methane loss is defined as:

$$\mathrm{C}{\mathrm{H}}_4\mathrm{loss}=\frac{\mathrm{C}{\mathrm{H}}_4\mathrm{flow}\mathrm{in}\mathrm{permeate}}{\mathrm{C}{\mathrm{H}}_4\mathrm{flow}\mathrm{in}\mathrm{feed}}$$ 5

The ratio of the total permeate over the total feed flowrate is defined as stage-cut:

$$\mathrm{stage}-\mathrm{cut}=\frac{\mathrm{permeate}\mathrm{flow}}{\mathrm{feed}\mathrm{flow}}$$ 6

The relative humidity (RH) is defined as:

$$\mathrm{RH}=\frac{\mathrm{Actual}\mathrm{vapor}\mathrm{pressure}}{\mathrm{Saturation}\mathrm{vapor}\mathrm{pressure}}$$ 7

Characterization

The morphologies of zeolite seeds and membranes were observed by Field Emission Scanning Electron Microscopy (FE-SEM, S-4800, Hitachi). Before characterization, the samples were sputtered with platinum. The phase compositions were determined by powder X-ray diffraction (PXRD, MiniFlex 600, Rigaku) with Cu K_α radiation (λ = 0.154 nm at 40 kV and 15 mA) in the 2θ range of 5–50° with the scan speed of 10° min⁻¹. For the membrane sample, the membranes were cut into small segments and tightly pasted on the sample rack for PXRD characterization. 10 mg zeolite crystals were dispersed in 10 mL deionized water and then ultrasounded for 30 min to make it evenly dispersed for particle size distribution analysis by dynamic light scattering (DLS) method using Malvern ZS90. Fourier transform infrared (FTIR) spectra were recorded on a Nicolet™ Summit™ FTIR Spectrometer with a DTGS KBr detector. ATR-FTIR spectra were recorded on a Bruker alpha ii Compact FTIR spectrometer with Quick Snap^TM Platinum-ATR Convenience sampling module. Solid-state MAS NMR experiments were conducted on a Bruker Avance Neo 400 Solid-State NMR Spectrometer.

Author contributions

X.W., X.G., F.K., and L.Y. conceived the idea and designed the experiments, analyzed the data, and wrote the manuscript. Y.J. and Y.F. synthesized and characterized the materials. Z.Z. performed the long-term stability test. X.P. performed the adsorption calculations. All the authors contributed to the scientific discussions and to the preparation of the manuscript.

Peer review

Peer review information

Nature Communications thanks the anonymous reviewers for their contribution to the peer review of this work. A peer review file is available.

Data availability

Data generated in this study are provided in the manuscript, or supplementary information and Source data are provided with this paper.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-70549-2.