Introduction

Membranes are among the key technologies to unlock sustainable, resource and energy-efficient modern industrial separation processes, reducing the energy required for separations by up to 80 % compared to conventional separation processes1. A membrane acts as a physical barrier for molecules, with its separation efficiency relying only on the material's parameters, offering an energy-efficient and sustainable technology to disruptively substitute conventional cryogenic distillation for the continuous separation of gases2. For porous membranes, narrow pore openings are favourable, as they offer a specific size selective diffusion limitation via the molecular sieving effect3, offering high selectivity and flux for small molecules, while excluding larger molecules4. Metal-Organic Frameworks (MOFs) – reticular inorganic-organic hybrid materials consisting of metal nodes and organic bridging ligands – are promising candidates for membranes, as they offer well-defined crystalline, porous structures and an almost unlimited number of chemical design possibilities.

Recently, melt-quenched amorphous framework glasses have become a special area of interest for the separation processes with several potential benefits over their crystalline counterparts5. Nowadays, zeolitic imidazolate frameworks (ZIFs) as well as carboxylate-based MOFs6 have been found to be meltable. First, framework glasses can be shaped and processed from powders in their liquid state, sometimes with the help of network formers or modifiers7,8, to form a free-standing layer, and therefore do not require complex thin-film growth techniques on expensive ceramic substrates, commonly used for MOFs and COFs9,10. Second, the universal processability of glasses allows for liquid-state shaping and further polishing, cutting, etc., to achieve the desired size and shape11. Further, most importantly for membranes, the ability of glass to form a monolithic layer, which can be hierarchically structured12 and hinders grain boundary diffusion, and therefore higher selectivity can be achieved9,13. Finally, unique for MOF-derived glasses, their porosity can be tuned by varying the processing parameters to obtain the targeted size of the molecular-sieving windows14.

The separation performance of the MOF glass membranes varies substantially between literature references9,10,13,15,16,17,18,19,20,21. Several neat, supported, and composited MOF glass membranes have been reported, showing overall high selectivity for various gas mixtures. Thus, for binary CO2/N2 mixtures, reported values of separation factor α reached α(CO2/N2) = 32 for pristine MOF glasses (agZIF-UC-419) and 34.6 for MOF crystal-glass composites (ZIF-8/agZIF-6221). Helium separation has been investigated on supported MOF glass membranes, allowing for its effective separation: pristine, supported agZIF-62 membrane showed selectivity values of 13.9 and 17.4 for He/CH4 and He/N2 mixtures, respectively13. Other groups tackle the challenging CH4/N2 separation (close kinetic diameters of species dkin(N2) = 3.64 Å and dkin(CH4) = 3.8 Å) for natural gas valorisation with MOF-glasses. A few attempts to employ MOF glass membranes have been made recently, with CH4/N2 selectivity values reaching 6.0 (agZIF-62 foam membrane20) and 6.2 (agZIF-UC-419). Some MOF glass composites with large-pore crystalline MOFs21 and zeolites22 also demonstrated a great promise for hydrocarbon separations, such as propane/propylene (α(propane/propylene) = 18.7) and 1,3-butadiene from different C4 alkanes and olefins, respectively. Comparative data on CO2/N2 separation data has been collected Table S3 and a Robeson-plot23 in Fig. 5S.

Despite the great promise, obtaining neat and free-standing MOF glass membranes of a larger area remains challenging. Practical limitations of melt-quenched MOF glasses are to choose parameters, after which they offer remaining accessible porosity11,14 and, as all glasses in practice, they are prone to cracking24 which complicates the fabrication of the large pieces and commonly necessitates the use of ceramic supports. Several approaches to achieve sufficient sizes of the bubble-free samples were reported, such as hot pressing25 or ball milling with further remelting26. Such glasses show overall good optical quality, but are not suitable for gas separation applications, as any excessive processing leads to significant losses in porosity and inaccessible pore channels. Many groups reported drastic MOF-glass densification caused by the proposed processing techniques27,28. As a result, achieving at least a typical membrane area of the polycrystalline MOF membranes – which is by itself commonly limited to the centimetre scale – while preserving porosity and therefore performance characteristics in MOF glasses remains highly problematic. Moreover, in the absence of such established procedures, further scaling considerations – potentially unlocking industrially-relevant meter-scale membrane areas – remain unexplored.

In this work, we demonstrate a method to achieve larger free-standing MOF glasses from crystalline MOFs (c.f. Fig. 1) on the example of meltable zeolitic imidazolate framework (ZIF) ZIF-62. Specifically, first we were able to produce centimetre-scale, free-standing MOF glass membranes made directly of melt-quenched ZIF-62 (agZIF-62), consisting of zinc metal sites coordinated by imidazolate and benzimidazolate linkers. Mechanical stress, which usually results in glass cracking29, is prevented by two primary changes in the melting procedure. (First, we empirically studied the influence of thermo-mechanical parameters and selected, based on the results, suitable substrates for melting and liquid-state processing of agZIF-62 to ensure optimal interfacial interaction. Second, we optimized the glass annealing procedure for agZIF-62, which is performed directly after melting at a temperature below Tg to relieve the internal stress caused by the high viscosity of the melt. This approach did not only provide us with large, crack-free, and utilizable thin layers, but also allowed us to avoid unnecessary processing (such as hot-pressing or remelting) and therefore preserve the most possible accessible porosity in agZIF-62, in accordance with our earlier work14. Further, we designed a procedure to fixate the MOF-glass as a free-standing membrane film, enabling long-lasting MOF-glass membranes for gas permeation testing, with a large active glass area. Our prototype agZIF-62 membrane shows excellent performance in gas-separation experiments, especially in the exclusion of CH4, with an approximate 100 % retention. The permeability of CH4 was so low that it became undetectable by our gas chromatography setup in single-gas and mixed-gas experiments, yet it allowed for the highly selective permeation of other gas species, even those with kinetic diameters just slightly higher than that of CH4. This is a step towards perfect homogeneity, crack-free, and defect-free agZIF-62 membranes without any evidence of grain boundary diffusion or leakage. All applied optimizations – from melting and annealing procedures to the prototype membrane fabrication – did not require any energy-intensive modifications and anything but readily available, cheap materials, enhancing the technology transfer potential. As an outlook, we suggest that this or similar – but always systematic, taking both reticular and glassy nature of such systems into account – approaches can be transferred to other glass formers within the MOF family. Further, we provide strategies for the further development of MOF-glass membranes towards real-world application, including a perspective approach for their upscaling to the industrially relevant membrane areas.

Fig. 1: Flow chart and overview of the processes developed within this work.
Fig. 1: Flow chart and overview of the processes developed within this work.
Full size image

a Synthesized crystal of ZIF-62. b Melt-pressed agZIF-62 on Al-foil, which was made following c the temperature-time diagram for glass preparation. d Thin, large area annealed agZIF-62 glass sheets can be produced reproducible in larger quantities. e Manufactured agZIF-62 membranes with a thickness of 330 µm. f The membranes stabilized by silica-lime glass rings and epoxy resin for handling. g These free-standing agZIF-62 films were installed into our permeation cell and measured. h Schematic image of the gas permeation is highly selective for He, H2 and impenetrable by methane.

The benefits of our advancements in MOF-glass processing demonstrated in this work, are shown in Fig. 1a–h, which starts from melting ZIF-62 crystals into scalable films, annealing them into crack-free, mechanically robust and utilizable agZIF-62 films, which can then be manufactured into stabilized membranes, enabling handling in our technical gas permeation devices.

Results

Overcoming Stress and Strain in Larger agZIF-62 Glasses

Meltable porous ZIF-62 (Fig. 2a), synthesized according to our previously published procedure11,14, was used to fabricate the MOF-glass membrane. Both synthesized material and glass were studied by XRD and the full loss of the crystalline structure after melt-quenching was observed (Fig. 2b).

Fig. 2: Preparation of MOF and MOF-glass.
Fig. 2: Preparation of MOF and MOF-glass.
Full size image

a Crystal structure of ZIF-62(Zn). Color code: C = green, N = blue, H = orange, polyhedra (Zn central atom) = gray. b XRD of ZIF-62 (red), calculated PXRD of ZIF-6241 (dark-gray) for comparison and the XRD pattern of the amorphous agZIF-62(Zn) (blue). c Results of ZIF-62 melting while pressed between soda-lime (SL) silica glass, platinum, and aluminium substrates and samples prepared for the melting (inset). d Large agZIF-62 layer on aluminium foil without an annealing step, after melting e the same sample after a few days, with a crack due to missing annealing program.

Free-standing layers of pristine agZIF-62 were prepared through the melt-quenching of slightly hand-ground crystalline ZIF-62 between two identical substrates. Formation of a bulk agZIF-62 sheet without defects and grain boundaries requires the application of some external load due to the very high viscosity of the melt11,30. On the lab scale, simple metal paper clamps simultaneously fix the substrate-powder-substrate system and ensure moderate loading on the material. At the moment the melting temperature (for ZIF-62 in this case, Tm = 450 °C) is reached, the paper clamp (see inset in Fig. 2c) has already irreversibly lost its force (spring force drops significantly beyond 80 °C) and does not have a major impact on the system during the cooling stage. Maintaining only a low load at this point is crucial, as it prevents excessive sample densification through pore collapse, preserving the accessibility of the pore channels for the guest molecules14. Moreover, in this way, large differences in volume change stemming from the elastic mismatch between the soft agZIF-62 (E < 10 GPa31) and the stiffer substrates (E ≥ 70 GPa32) are also avoided.

What must be considered is the interaction between the glass melt and a substrate during cooling, as incompatibility at this interface can lead to the build-up of additional internal stress in the resulting glass. The same problem is very likely to occur while melting the grown crystalline MOF layer on alumina to form a supported glass membrane19,33. To tackle this, various inert substrates have been initially selected for the melting of ZIF-62 to form agZIF-62 based on their coefficient of thermal expansion (CTE) and either brittle or ductile nature: soda-lime silica glass (SL; low CTE, brittle), aluminium (high CTE, ductile), and platinum (low CTE, ductile) (Fig. 2c). Among the employed substrates, our empirical finding is that only aluminium yielded bulk and crack-free layers (Fig. 2d), whilst contact with soda-lime silica glass and platinum resulted in significant cracking. This can be explained by both agZIF-62 and aluminium exhibiting relatively large CTE (32.1 × 10-6 K-134 and 24.0 × 10-6 K-135, respectively). This stands in contrast with platinum and soda-lime silica glass, which have low CTEs (9.0 × 10-6 K-135, and 9.2 × 10-6 K-136, respectively): while tensile stresses are induced in agZIF-62 during cooling, compressive stresses develop in the surrounding materials. Additional context and quantitative estimations to support this explanation are provided in Table S 1, Supplementary Note 1 and Table S 2.

As ductile platinum showed a result similar to the soda-lime silica glass, it can be concluded that the ability of the substrate to deform plastically cannot compensate for the difference in their thermal behaviour (c.f. Fig. 2c). As ductility by itself is not a limiting parameter, it is expected that the last untested combination – brittle material with high CTE – could also potentially yield a crack-free agZIF-62; however, designing such an experiment is nearly impossible due to the absence of inert and temperature-stable materials with this set of characteristics (for instance, matching for the role, phosphate glasses with high CTE would normally undergo a glass transition Tg way below the melting point of ZIF-62)32.

Testing other inert substrates such as borosilicate glass BOROFLOAT® 33, referred to as BF33 (brittle, very low CTE of 3.25×10-6 K-1) and gold (ductile, relatively low CTE of 14.2 × 10-6 K-1) also resulted in significant cracking (Figs. S1 and S2). Worth noting is that the visual “turbidity” of the glass layer (Fig. 2b, c) is only caused by the transferred surface pattern of the aluminium foil used for melting (Fig. S3), and the material remains transparent in its volume.

Although the selection of a compatible substrate allows us to make larger layers, the fabrication procedure is not comprehensive without a sufficiently designed annealing step. If melt-quenched and cooled to room temperature directly after the melting, agZIF-62 films can only keep their structural integrity for hours, or at most up to a few days, as at some point the residual stress causes spontaneous cracking (Fig. 2d, e). To avoid that, MOF glass was annealed slightly below the Tg of agZIF-62 at the temperature of 300 °C (Tg (ZIF-62) = 322 °C) for 2 h before cooling down to room temperature. This temperature choice appears close to the optimum, as it allows for the balance between not densifying the glass in its liquid state and keeping the matrix mobile enough at the local scale to allow atomic rearrangements, releasing the residual inner stress.

Membrane fabrication

As membrane performance testing and utilization require proper sealing, normally involving the application of a certain load to O-rings and the membrane, the obtained thin brittle glass layer must be somehow stabilized to avoid cracking and damage. We demonstrate that a pristine free-standing MOF glass membrane can be fabricated by stabilizing with soda-lime glass and sealing with epoxy resin, enabling unproblematic handling. The agZIF-62 thin film was fabricated as described above – by melting the crystals between aluminium foils, which is carried on soda lime glass sheets, fixed by paper clamps (Fig. 3a), and followed by an annealing step prior to cooling to room temperature to release residual stress of the material. After removal of the aluminium foil, the agZIF-62 thin film was manually glued between two identical ring-shaped soda-lime silica glasses with epoxy resin (Fig. 3b, c), allowing for proper contact and stabilizing the active area. Excess agZIF-62 was then removed by breaking off the edges and polishing the outer ring (Fig. 3d, e) to maintain the required circular shape.

Fig. 3: Fabrication of a pristine MOF glass membrane via hot-pressing.
Fig. 3: Fabrication of a pristine MOF glass membrane via hot-pressing.
Full size image

a thin glass sheet is prepared by pressing between glass slides, with a layer of metal foil in-between b Exploded drawing of the MOF-glass membrane sandwich which is then c fixed together; d) area marked for cutting and polishing; e final membrane and f a photograph of the final MOF-glass membrane.

As a binding component, epoxy resin is readily available and easy to handle, while providing sufficient thermal and mechanical stability, as well as being gas-tight, thereby preventing leakages. The resulting agZIF-62 membrane remains undamaged and stable upon handling and is easily transported, sealed, and can thereby run as a prototypical membrane in the gas permeation setup (Fig. 3f). It requires harsh handling to break the obtained membranes after they have been fixed.

Performance

Gas permeation of single gases and binary gas mixtures was evaluated using a Wicke-Kallenbach setup and a gas chromatography system (GC). The gas separation performance of the fabricated agZIF-62 membrane with a thickness of 330 µm was tested under atmospheric conditions at 1 atm and room temperature (RT). A total of 100 ml/min of the feed gases (He, H2, CO2, N2, and CH4) were used, while sweep gas (N2, or CH4) flow rate was kept at 1 ml/min. The single and mixed gases were measured twice, with different carrier gases (He and Ar) for maximum intensity and accuracy of the individual peaks and their area, with a minimum of 99 cycles per experiment. The mounting of the agZIF-62 membrane is demonstrated in the photograph in Fig. 4a, featuring the large, transparent, and free-standing thin film in the middle with ⌀ 18 mm in diameter of which ⌀ 10 mm – and thus about 78.54 mm² – can be considered active membrane area. Screwing the O-rings gas-tight requires high pressure to seal the cell. The proposed design allows MOF glass to withstand this stress due to the epoxy resin and soda-lime glass rings fixing it from both sides, ensuring the mechanical stability of the whole cell. The measured single gas permeability shows a decreasing trend with an increase of the kinetic diameters, dkin, of the tested gas species: He (dkin = 2.6 Å), H2 (dkin = 2.89 Å), CO2 (dkin = 3.3 Å), N2 (dkin = 3.64 Å), and CH4 (dkin = 3.8 Å) (Fig. 4b). Notably, the methane permeability P(CH4) tends to zero with no molecules being detected over the course of 16.5 h (99 cycles) of each measurement (carrier gas He, Ar) – single gas CH4 and mixed gas He/CH4, H2/CH4, CO2/CH4, N2/CH4 are thus leading to an estimated separation factor α (Pi/PMethane) = ∞ in all cases. The minimum concentration of a gas in a sample that can be detected by the gas chromatography system with a thermal conductivity detector is 0.00009%, while it is safe to quantify concentrations at 0.001% ( > 10 ppm) or above. We do not observe a peak in CH4, even after switching carrier gases from He to Ar across all measurements, indicating that agZIF-62 is completely impermeable to CH4. The monolithic nature of glass, not allowing grain boundary diffusion, kinetically excludes larger molecules, starting at the kinetic diameter of methane. This trend can be observed in single-gas permeabilities, as high permeabilities for smaller species, such as He (P(He) = 5500 barrer) and H2 (P(H2) = 5400 barrer), are measured. Further, significant loss of permeability occurs for CO2 with P(CO2) = 500 Barrer, then very low permeability of P(N2) = 17 Barrer for N2 is determined, indicating the drastic molecular sieving cut-off. To the best of our knowledge, such a sharp molecular-sieving cut-off has not been shown for MOF glasses or any other MOF-membranes, yet. This result is supported by our previous preliminary work, where an absolute exclusion of ethane (dkin (C2H6) = 4.4 Å) diffusion was achieved after liquid-state processing of agZIF-62, suggesting that this is likely possible, also for CH4 and other relatively large gas molecules. In particular, the pore channel size distribution in such, but millimetre-scale glasses prepared by following the same initial synthetic and melting protocols have been investigated in detail via HR-TEM: pore channel sizes varying from approximately 1.7 Å to 4 Å with an average pore-limiting diameter of 2.7 ± 0.5 Å have been observed in the normal distribution for the analogous to the current work samples14. The above-demonstrated permeation results are indeed in agreement with this data: a disordered glassy nature results in a relatively broad pore channel size distribution, but the ≥3.8 Å (potentially allowing for CH4 (dkin = 3.8 Å) permeation) fraction is almost negligible compared to the smaller sizes. At the same time, the 3.64 Å - 3.8 Å fraction, enabling selective N2 permeation, is more present and therefore a notable N2 signal is observed. Aside from H2/CH4, He/CH4, and CO2/CH4 gas pairs, where pristine supported9,10,13 and composite37 MOF-glass membranes have already demonstrated great promise, a highly challenging N2/CH4 separation with relatively low reported selectivities19,20 can now be addressed more effectively.

Fig. 4: Performance data for agZIF-62 free-standing membrane.
Fig. 4: Performance data for agZIF-62 free-standing membrane.
Full size image

a Photograph of the transparent and colourless glass membrane fixed in the measurement cell. b Single-gas permeability data and corresponding kinetic diameter of the gases. c Gas permeability for binary gas mixtures of the target molecules. d Corresponding real selectivity.

Permeances of gases in a binary mixture of H2/CO2, CO2/N2, He/CO2 and H2/CO2 are shown in Fig. 4c, with corresponding real selectivity α in Fig. 4d. Straightforward comparison of this data with other MOF-glass membranes is complicated by the fact that – as demonstrated in the introduction – currently reported membranes vary significantly in their type (pristine or composited with crystalline MOFs or polymers), support nature (free-standing or supported), and the initial glass-former composition. All of these variations are employed to design the membrane exhibiting the molecular-sieving characteristics that specifically target one or another separation, while our membrane in particular works best for the challenging methane separations. In Fig. S5, we present a qualitative overview of the various reported MOF-glass membranes tested for different gas separations. Thus, our experimental value of the real selectivity αreal(CO2/N2) = 21,8 from a mixed-gas CO2/N2 measurement is in good agreement with the ideal selectivity, published for supported membranes of agZIF-62 (αideal(CO2/N2) = 23)9 and also agTIF-4 (αideal(CO2/N2) = 25)10 with the latter MOF glass appearing to be very similar to agZIF-62, structurally and property-wise. In our present work, however, the ideal selectivity we measure reaches a value of αideal(N2/CO2) = 29. Smaller species, such as H2 and He, were also effectively separated from CO2 with our agZIF-62 membrane, calculating an ideal selectivity of αideal(H2/CO2) = 10 and αideal(He/CO2) = 11 and real selectivity of αreal(H2/CO2) = 20,2 and αreal(He/CO2) = 23, respectively. A mixed gas separation factor much larger than an ideal separation factor is not often seen; however, it can be explained by the nature of the agZIF-62 having mainly pore channels, and not much of pore space for diffusion, in which the gases block each other, and the faster permeating species are highly favored to overcome the surface barrier38,39. Although the mixed-gas performance of MOF glass-based membranes was not reported for H2/CO2 and He/CO2 pairs, our estimations of ideal separation factors from other works13,19 give notably lower values, proving the defect-free nature of the present membrane. Table 3S and Fig. 5S puts our data in comparison with data on different MOF-glass membranes.

Upon completion of performance testing, the measurement cell was opened and the membrane was cut along the stacking axis to observe the cross-section, which was further polished (Fig. 5a). That the membrane architecture survives mechanical stress is shown by the technical drawings on how the membrane is mounted to become gas tight in Fig. 5b, c. From micrographs and SEM image it is observable that all layers – agZIF-62, epoxy resin binder, and soda-lime silica glass – remained macroscopically (Fig. 5a) and microscopically undamaged after cutting and polishing with no signs of crumbling or detaching (Fig. 5d). In Fig. 5d the SEM image and corresponding energy dispersive x-ray (EDX) mapping of the agZIF-62 membrane’s cross-section are presented. The EDX mapping proves the homogeneous structure of all layers with no signs of voids or detachments – thus, a great binding between epoxy resin and MOF/soda-lime silica glasses – which is perfectly visible through characteristic elemental signals: silicon (green) and oxygen (light blue) for soda-lime substrate, carbon (red) for epoxy resin, zinc (blue), nitrogen (yellow) and carbon, all well-distributed for agZIF-62. Single element maps can be found in the Supplementary Information (Fig. S4).

Fig. 5: Membrane material investigation.
Fig. 5: Membrane material investigation.
Full size image

a Micrograph of the polished agZIF-62 membrane’s cross-section and b the corresponding schematic showing how it seals with O-rings. c A cross-section of the measurement cell from computer-aided design, and the real photograph with the MOF glass membrane fixed in it. d SEM image of the agZIF-62-glass architecture between two silica glass rings, glued with epoxy resin. EDX mapping of the membrane’s cross-section (color code: silicon – green, carbon – red, zinc – blue, nitrogen – yellow, oxygen – cyan). The membrane thickness is measured to be 330 µm.

The MOF glass herein investigated does not possess any defects like bubbles, cracks, or densified areas throughout the volume, which is a result of sufficient cleaning of the MOF particles and high-purity chemicals in their synthesis. The SEM analysis once again confirms the absence of grain boundaries and explains the observed exceptional selectivity for CH4 and smaller gas species. Second, no gaps or defects are observed at the interfaces, demonstrating that this material combination is suitable for stabilizing the active layer while eliminating potential leaks.

Discussion

MOF glass membranes possess tremendous potential for high-precision separation technologies, as they exhibit very sharp molecular-sieving cut-offs. The high selectivity is achievable because the system combines the well-defined porosity in MOFs with the processability of glasses, allowing for the fabrication of monolithic defect-free layers without grain boundaries. As we demonstrated before, the material easily withstands polishing and other types of conventional glass treatment, such as liquid-state shaping, so thinner layers can indeed be fabricated by such processing techniques11,14. Moreover, the initial membrane thickness can be modified within a certain range by controlling the amount of the crystalline powder per substrate area before melting.

Even more possibilities are introduced by the tunable nature of the material’s porosity14, enabling targeted modification of MOF glass for the parameters of a specific separation process. We envision that the systematic fabrication and processing considerations presented in this work can be applied to other MOF-glass-formers beyond ZIF-62 – in such cases, the intrinsic properties and behaviours of these specific systems must always be taken into account. However, such complex materials at the early stages inevitably raise some challenges.

Although the formation of free-standing layers rules out the need for the synthesis of thin polycrystalline films and further possible glass-substrate incompatibilities during melt-quenching, it raises the question of glass brittleness instead. Indeed, handling such thin layers without breaking them is challenging, and simply increasing the glass area on the way to upscaling seems unreasonable. A more advanced solution is applying modular designs to the membranes (Fig. 6a). In perspective, such design does not only fix the membranes to decrease the probability of breakage but allows for changing the modules independently, extending the overall life cycle of the system. If connected by a flexible material, such as rubber or a resin (Fig. 6b), more convenient handling and transportation are possible.

Fig. 6: A proposed, efficient modular designs for free-standing MOF glasses towards technical applicable glass membranes.
Fig. 6: A proposed, efficient modular designs for free-standing MOF glasses towards technical applicable glass membranes.
Full size image

a Non-flexible and b Flexible form.

Another issue to be addressed is the relatively low values of permeability of the agZIF-62 membrane. Although this is somewhat inevitable for a membrane that offers this degree of high precision molecular sieving, certain improvements can be applied to decrease the thickness of the glass sheet. As has been demonstrated earlier11, the material easily withstands polishing and other types of conventional glass treatment, such as liquid-state shaping, so thinner layers can be indeed fabricated by such processing techniques.

Therefore, real-world applicability of the MOF-glass membranes will become even more feasible through such further optimization steps and design solutions, and their introduction will be paid off by the outstanding molecular sieving abilities. A good supporting example are perovskite-based membranes that have been successfully commercialized for O2 separation. They require high temperatures to show operable fluxes, but their 100% separation efficiency makes up for the higher costs that arise from an external energy input40. In the same fashion, producing thinner MOF-glass membranes or applying modular designs, despite potential complications in fabrication and operating, could address challenging separations where precise molecular sieving is of high importance.

Methods

Materials

Zinc nitrate hexahydrate ( ≥ 99%) was purchased from ABCR. Benzimidazole ( ≥ 99%) was supplied by Alfa Aesar, and Imidazole ( ≥ 99.5%) was purchased from Sigma-Aldrich. N,N-Dimethylformamide ( ≥ 99.9%) was supplied by VWR, dichloromethane ( ≥ 99%, stabilized with ethanol) – by Acros Organics. Epoxy resin DIPOXY-2K-4000Geode-ART by Dipoxy-Germany. Au, Pt, Al foil had a purity of 99.99 % as provided by the vendors. BF33 was provided by Schott AG (technical data sheet: https://www.schott.com/en-gb/products/borofloat-p1000314/downloads).

Synthesis of ZIF-62(Zn)

The synthetic procedure for ZIF-62(Zn) was fully reproduced from our previous work11. A total of 12.79 g of benzimidazole and 38.02 g of imidazole were sequentially dissolved in 480 ml of DMF, with stirring for 5 min. Following this, 19.93 g of zinc nitrate hexahydrate was added to the mixture, and stirring was continued until complete dissolution was achieved. The solution was then transferred to a 500 ml glass jar and maintained at 130 °C for 60 hours. The white sediment formed was separated from the mother liquor by centrifugation (10 minutes, 8.528 × g) and washed thoroughly twice with both DMF and DCM. Finally, the ZIF-62 powder was activated in a vacuum furnace (25 mbar, 150 °C) for 72 h.

Melting of agZIF-62

The melting of ZIF-62(Zn) was performed following the procedure outlined in our previous report with some changes. The crystals were carefully ground into a fine powder without applying to much mechanical pressure. The powder was then placed between two silica glass slides (microscope slides). Depending on the experiment, melting was performed directly on soda-lime silica or borosilicate glass, or on aluminium, gold, or platinum foil. The foils were placed between powder and silica glass substrates. The assembly was secured with two metal paper clamps and then heated to 450 °C in a nitrogen atmosphere. The clamps are from an office supply, thus they soften above 80 °C and should apply only a very minimal load to the sample, as they lose their spring force. Heating from room temperature to 450 °C was done using a heating rate of 10 K/min. After maintaining 450 °C for 5 min, the setup was allowed to cool to room temperature with a cooling rate of 2 K/min. In case of adding an annealing step, the system was cooled down from 450 °C to 300 °C first with 2 K/min cooling rate and kept there for 120 min before cooling down to room temperature with 2 K/min cooling rate. The resulting agZIF-62 was carefully removed from the substrate.

Optical microscopy

A digital microscope (VHX-6000, Keyence) equipped with a universal zoom lens (VH-Z100UR and VHX-S650) and a free-angle XYZ observation stage was employed to capture digital images of agZIF-62 membrane and its cross-section.

Powder X-ray diffraction (PXRD)

PXRD data for crystalline and amorphous samples were collected using a Rigaku MiniFlex diffractometer configured with Bragg-Brentano geometry. The instrument was equipped with a 600 W X-ray generator producing Cu Kα radiation at a wavelength of 1.54059 Å. Diffraction patterns were recorded over a 2θ range of 5–50°, with a step size of 0.02°.

Gas Permeation Measurements and Calculations

Gas permeation was measured using a Wicke-Kallenbach setup using sweep gas as a transport medium to the gas chromatography system (GC). Bronkhorst EL-Flow mass flow controllers were used to transport gases upstream and downstream. The gas composition was analyzed using an online Shimadzu Nexis GC-2030, equipped with a TCD detector and switchable carrier gas option (He, Ar, N2, and H2). The GC has a two-column setup, where permanent gases are fractioned on a 13x molecular sieve column, while CH4 and CO2 are fractioned using an SH-Q-Bond (100% divinylbenzene) plot column. Feed gas was fed to the upstream at 100 ml/min, while sweep gas was flown to the downstream at 1 ml/min. No heating was applied, and the feed and sweep sides were unpressurized so that the measurement was performed at room temperature (T = 22°C) and 1atm (atmospheric pressure). Binary gas mixtures were fed with partial pressures of 0.5 atm each. Permeability of gas i (Pi) is the thickness and chemical potential normalized flux through the membrane, calculated from Eq. (1)

$${{{{\rm{P}}}}}_{{{{\rm{i}}}}}=\frac{{{{{\rm{n}}}}}_{{{{\rm{i}}}}}}{{{{\rm{A}}}}\Delta {{{\rm{p}}}}}{{{\rm{d}}}}$$
(1)

where n is the amount of gas molecules, d is the thickness of the membrane, A is the effective area of the membrane, and Δp is the transmembrane pressure (chemical potential gradient, driving force). Dividing the molar ratios of the gases i, j in the permeate by the molar ratio of the components i, j in the retentate gives the separation factor α in Eq. (2). The feed gas is pumped to the membrane in a large excess and thus, the retentate quotient becomes 1. This simplifies the quotient to contain only the components in the permeate.

$${{{\rm{\alpha }}}}({{{\rm{i}}}}/{{{\rm{j}}}})=\frac{{{{{\rm{n}}}}}_{{{{\rm{i}}}},{{{\rm{permeate}}}}}/{{{{\rm{n}}}}}_{{{{\rm{j}}}},{{{\rm{permeate}}}}}}{{{{{\rm{n}}}}}_{{{{\rm{i}}}},{{{\rm{retentate}}}}}/{{{{\rm{n}}}}}_{{{{\rm{j}}}},{{{\rm{retentate}}}}}}=\frac{{{{{\rm{P}}}}}_{{{{\rm{i}}}},{{{\rm{permeate}}}}}}{{{{{\rm{P}}}}}_{{{{\rm{j}}}},{{{\rm{permeate}}}}}}$$
(2)

The permeability is given for better comparison to literature data in Barrer, a non SI-unit, which is equal to 1 Gas Permeability Unit (GPU). In Eq. 1, the transformation into SI units is given:

$$1\,{{{\rm{Barrer}}}}=1\,{{{\rm{GPU}}}}=3.346\cdot {10}^{-16}\frac{{{{\rm{mol}}}}\cdot {{{\rm{s}}}}}{{{{\rm{kg}}}}}$$
(3)

Scanning electron microscopy (SEM) and energy dispersive X-ray (EDX) spectroscopy

First, the membrane stacks were carefully cut using a diamond saw to expose the cross section, manually ground using SIC abrasive paper with progressively increasing grit from P1200 to P4000 and polished using a water-based 1 µm diamond polishing suspension (Struers DiaPro). Further, a piece of the membrane was fixed to a carbon adhesive disc, and a small amount of conductive silver paint was added to ensure good conductivity. Finally, the samples were coated with Pt (8 nm) in a high-vacuum sputter coater (Safematic CCU-O10 HV) to avoid charging effects.

An Oxford EDX system in combination with a Sigma VP Field Emission Scanning Electron Microscope (Carl-Zeiss AG, Germany) was used to analyse the sample. The SEM micrographs were obtained with an acceleration voltage of 6 kV an emission current of 271 pA and using secondary electrons and in-lens detectors. EDX investigations were performed with a 50 mm2 XMax detector with a resolution of 127 eV. Mapping was performed within 20-30 minutes with an acceleration voltage of 20 kV.