Melt-sintering of MOF glass–2D nanosheet composites for enhanced processability and gas separation performance

Abstract

Amorphous metal–organic framework (MOF) glasses offer processible microporous materials for separation, sensing and catalysis, but their practical implementation has been limited by poor mechanical stability and ill-defined microporous structures. Here, we report a strategy to integrate g-C₃N₄ nanosheets into MOF glass via high-temperature sintering. Interfacial coupling, including π–π stacking and coordination-assisted bonding, effectively guides the packing of the nanosheets, improves structural integrity, and mitigates processing-induced stress during MOF glass vitrification. By tuning the sintering temperature, the interlayer spacing of the g-C₃N₄ can be adjusted and stabilized, generating more regulated transport channels. The resulting hybrid glass can be processed into membranes, showing a two-order-of-magnitude increase in hydrogen permeance compared to pure MOF glass, with the selectivity for difference gas pairs also significantly improved. This contribution establishes a generalizable method for interlayer spacing engineering within hybrid glasses, demonstrating that the incorporation of 2D materials can enhance the processability and separation performance of the composite glass.

---

Introduction

The ever growing demand for energy-efficient separation technologies has driven significant efforts towards developing advanced materials capable of selective gas transport.¹ Membrane-based separations, in particular, offer an easy-to-deploy, continuous, low-energy alternative to conventional thermally driven separation processes such as distillation, absorption and adsorption.² Among different membrane material candidates, metal–organic frameworks (MOFs) have emerged as a versatile class of functional materials featuring tuneable pore structures and chemistry, enabling their exceptional molecular separation capability.³ However, despite these advantages, the translation of MOF materials (most are crystalline phases) into macroscopic, processable, and mechanically robust membranes remains challenging.⁴ The intrinsic brittleness of MOF crystals and the difficulty of forming continuous, pinhole-free layers often result in poor scalability and limited long-term stability.^5,6 Recently, the discovery of amorphous MOF glasses has opened new possibilities for overcoming these limitations.^7,8 These materials preserve the local coordination network of MOFs while exhibiting isotropic, continuous structures and thermoplastic processability,^9–11 making them promising candidates for next-generation separation membranes.

Despite their promising processability, MOF glass materials still face intrinsic challenges in microstructure control¹² and mechanical reliability.¹³ The loss of long-range order during the vitrification usually leads to broad pore size distribution with poorly defined gas transport pathways.¹⁴ In addition, the amorphous framework normally possesses high viscosity near its glass transition temperature,^15–17 which restricts the relaxation of internal stress during cooling. The accumulated stress can generate cracks as the high temperature melt solidifies, significantly compromising the mechanical integrity of the MOF glass based thin films and membranes.^18,19 These defects further generate non-selective pathways that degrade gas separation performance. Therefore, although MOF glass is a promising and attractive candidate for defect fabrication given its balanced rigidity and processability, its limited mechanical resilience and uncontrolled microstructure remain key challenges to its implementation in molecular separation membranes.^20,21

Incorporating two-dimensional (2D) materials into glass substrates can be an effective method to tailor both mechanical and transport properties at the nanoscale.^3,22 The intrinsic mechanical robustness, chemical stability and tuneable interlayer spacing of 2D materials offer unique opportunities for reinforcing hybrid glasses and controlling their internal pore structures.²³ Among various candidates, graphitic carbon nitride (g-C₃N₄) has attracted growing attention due to its conjugated structure, rich functional groups and satisfactory chemical and thermal stability.²⁴ When incorporated into a substrate, g-C₃N₄ can act as a stable scaffold that constrains local structural variation of the composite.²⁵ In addition, its periodic nanosheet structure allows regulation of the interfacial interactions and free space within the composite, potentially providing molecular transport pathways for fine-tuning gas separation.²⁶ However, controlling and stabilising the interlayer spacing of g-C₃N₄ remains challenging, as the layers are held together mainly by weak van der Waals forces that tend to fail or reorganise under thermal, solvent or mechanical conditions.^27,28

In this work, we introduce a strategy to address the intrinsic mechanical fragility and poorly defined pore structure of MOF glass by integrating layered 2D g-C₃N₄ into the glass matrix. Through a combination of structural, thermal and chemical structure analysis, we revealed that g-C₃N₄ promotes local interfacial coupling with the MOF glass matrix, which helps regulate glass densification and partially relieve processing-induced stress during vitrification. In addition, the interlayer spacing of g-C₃N₄ within the composite can be readily tuned by adjusting the sintering temperature and its loading, enabling fine tuning of the microstructure and gas transport channels. This composite modulation leads to an enhancement of the composite's processibility, mechanical structure and separation capability. As demonstrated in our gas separation tests, the composite membrane demonstrated an extraordinary two-order-of-magnitude increase in hydrogen permeance compared to the pure MOF glass, along with a substantial improvement in H₂/CO₂ selectivity from 7 to 21. These discoveries underscore the potential of this composite approach for developing robust and high-performing membranes and thin films.

Results and discussion

ZIF-62 crystals were synthesized via a solvothermal method by dissolving Zn(NO₃)₂ and mixed ligands (benzimidazole, bIm, and imidazole, Im) in DMF, followed by a 120 °C treatment. The crystalline structure was confirmed by powder X-ray diffraction (PXRD) (Fig. S1). The molar ratio of the linkers (bIm/Im = 1.75/0.25) in the crystalline phase was determined using ¹H liquid nuclear magnetic resonance spectroscopy (NMR) (Fig. S2). g-C₃N₄ was prepared via a convenient top-down approach, wherein melamine was thermally sintered slowly in air to facilitate self-assembly into a two-dimensional lamellar structure shown by SEM (Fig. S3). The obtained ZIF-62 crystals and g-C₃N₄ powders were mixed in isopropanol according to a predetermined mass ratio, ground into a homogeneous viscous slurry (denoted as ZIF-62_CN-x, where x represents the mass fraction of g-C₃N₄), and then pressed into pellets. The pellets were then thermally treated under an argon atmosphere and held at the target temperature for 30 min, followed by melt-quenching to room temperature to obtain self-supporting, in situ-formed glassy composites.

The pure ZIF-62 exhibits representative MOF glass formation behaviour. The transformation from crystalline ZIF-62 to its amorphous glass form (denoted as a_gZIF-62, where “a_g” stands for amorphous glass) was achieved by heating the crystals above their melting point (T_m = 440 °C), as determined from the results of the first differential scanning calorimetry (DSC) upscan in Fig. S4, followed by quenching to room temperature. After quenching, the Bragg diffraction peaks of crystalline ZIF-62 disappeared in XRD, replaced by a diffuse scattering at ca. 16°, confirming the formation of an a_gZIF-62 glassy phase. A comparable tendency can be seen in Fig. 1b, in which the melting point of the ZIF-62_CN-25 composite occurs at 448 °C during the first upscan, and the emergence of glass transition behaviour appears at 332 °C in the second DSC upscan. The thermal behavior of ZIF-62_CN-10 and ZIF-62_CN-40 composites during melting was further studied through combined mass reduction in thermogravimetric (TG) and DSC analyses. The endothermic peak observed between 150 °C and 250 °C in the first DSC upscan corresponds to the mass loss of residual solvent, i.e. DMF, retained in the ZIF-62 pores (Fig. S4), as corroborated by the mass loss observed in the TG curve at ca. 150–250 °C (Fig. S4d). TG confirmed that both g-C₃N₄ and ZIF-62_CN were thermally stable up to 500 °C without noticeable decomposition (Fig. S4). A key observation from the first DSC upscan is that the melting temperature (T_m), identified as the second endothermic peak of the ZIF-62_CN composites, initially increased and then decreased with increasing g-C₃N₄ content, accompanied by progressive peak broadening. During the second heating ramp, ZIF-62_CN composites exhibited a glass transition temperature (T_g), a defining characteristic of amorphous glassy materials, confirming the successful transformation of ZIF-62 into the glassy state. This confirms that the incorporation of g-C₃N₄ does not compromise the thermal stability of ZIF-62 nor hinder its glass-forming ability. Furthermore, T_g of the ZIF-62_CN-x composites was found to be elevated compared to that of pure ZIF-62 (Fig. 1b and c), likely due to interfacial interactions that enhance the rigidity and connectivity of the hybrid network.²⁹ High-resolution synchrotron in situ temperature variation X-ray powder diffraction (XRPD) further reveals the structural transformation during thermal processing of the ZIF-62_CN-25 composite (Fig. 1d). Upon heating, the diffraction peaks of crystalline ZIF-62 gradually diminish and ultimately vanish as it melts into an amorphous glass. Meanwhile, g-C₃N₄ exhibits the characteristic layered conjugated polymeric structure, where a prominent peak at ca. 27° is attributed to the (002) plane associated with interlayer π-stacking of the tri-s-triazine aromatic frameworks.³⁰ This peak initially downshifts during the heating stage (400–450 °C), followed by an upshift during quenching, ultimately stabilizing at ca. 27.8°.

Fig. 1 Transformation of crystalline structure and thermal behavior of the synthesized glass composite: (a) fabrication of glass composite pellet. (b) DSC analysis of the a_gZIF-62_CN-25 composite. (c) Glass transition temperatures of various glass composites. (d) Temperature variation synchrotron radiation XRPD patterns of selected ZIF-62_CN-25 derivatives before and after heating to 450 °C (λ = 0.45920 Å). (e) XRD patterns of the a_gZIF-62_CN-25 composite at 400 °C, 420 °C, 440 °C, and 460 °C.

According to the in situ powder diffraction results, the interlayer spacing of g-C₃N₄ exhibits a strong temperature dependence. The glass composites, including samples with different g-C₃N₄ loadings and treatment temperatures, were further examined by ex situ PXRD to probe the variation of the interlayer distance. In this work, the composite glassy samples were denoted as a_gZIF-62_CN-x. Taking the a_gZIF-62_CN-25 composite as an example, ex situ PXRD patterns in Fig. 1e exhibit a trend of upward shifting by ca. 0.2° with higher sintering temperature, indicating that the glassy composite can stabilize the different interlayer distances even when being cooled back to ambient temperature.

We further identified that the interlayer distance can be readily tuned through different g-C₃N₄ loadings (Fig. 2a). Similar to pure a_gZIF-62, all composite samples exhibit a broad diffuse scattering peak, indicating the loss of long-range order and the formation of an amorphous phase. Compared with pristine g-C₃N₄, the diffraction peaks of g-C₃N₄ in a_gZIF-62_CN-x composites are preserved after melt-quenching at 440 °C, confirming the preservation of the g-C₃N₄ stacking structure. Notably, a constant up shift is observed in the a_gZIF-62_CN-10, a_gZIF-62_CN-25, and a_gZIF-62_CN-40 composites, indicating a gradual contraction of the g-C₃N₄ materials. However, when the g-C₃N₄ loading reaches 50 wt%, this peak shift is no longer apparent when compared with that of the pristine g-C₃N₄. The observed shift of the g-C₃N₄ (002) reflection is closely linked to interfacial interactions within the composite. At moderate g-C₃N₄ loadings, the molten a_gZIF-62 phase effectively wets and confines g-C₃N₄ stacks during melt-quenching, producing strong interfacial confinement that contracts the interlayer spacing and shifts the (002) peak to higher angles. By contrast, at excessive g-C₃N₄ loadings, a larger fraction of g-C₃N₄ remains unconfined or undergoes aggregation/phase separation; the diffraction signal thus becomes dominated by these bulk-like domains and the net peak shift is diminished. Hence, the magnitude of the (002) peak shift reflects the extent of interfacial confinement afforded by the glass phase rather than the nominal g-C₃N₄ content alone.

Fig. 2 Morphology and interlayer structure of glass composites: (a) XRD patterns of a_gZIF-62_CN-x composites and pure g-C₃N₄. (b1–b3) High-resolution TEM image of the a_gZIF-62_CN-25 composite and its line profile analysis extracted from the HRTEM image, illustrating the periodic contrast with apparent spacing. (c) SEM surface and cross-sectional images of the a_gZIF-62_CN-25 glass composite pellet with the Zn and N elemental distribution.

To further validate the nanoscale structural change at the microscopic level, high-resolution transmission electron microscopy (HRTEM) was employed. As shown in Fig. S5 and 2b, the pristine g-C₃N₄ exhibits well-defined lattice fringes with an interlayer spacing of ca. 0.329 nm, corresponding to the (002) graphitic plane. In contrast, in the a_gZIF-62_CN-25 composite, the layered domains preserved within the amorphous glass matrix display a slightly contracted spacing of ca. 0.32 nm. The magnified view of the amorphous glass area (Fig. 2b1) confirms the absence of lattice ordering,³¹ while the confined g-C₃N₄ stacking regions (Fig. 2b2) and their corresponding FFT-assisted fringe fitting (Fig. 2b3) clearly reveal periodic fringes with reduced spacing. This nanoscale densification of g-C₃N₄ layers aligns with the PXRD peak upshift, collectively indicating that the ZIF-62 glass matrix imposes interlayer confinement and enhances g-C₃N₄ stacking compactness during vitrification. The microstructural observations further corroborate this interfacial confinement, as the TEM images reveal intimate contact between the glass and g-C₃N₄ phases without discernible interfacial voids or defects.

To evaluate the microstructural integrity of the composites, scanning electron microscopy (SEM) was employed to examine the surface and cross-sectional morphologies before and after melting (Fig. 2c, S6 and S7). The thickness of the self-supporting a_gZIF-62_CN-25 membrane was determined from the cross-sectional SEM image to be approximately 367 µm (Fig. S7). The loose, two-dimensional g-C₃N₄ can be readily embedded and dispersed within the molten ZIF-62 phase, in contrast to crystalline fillers with distinct grains. The cross-section also reveals a dense, void-free structure compared with the pre-melted samples, confirming that the two distinct phases are intimately integrated. In addition, satisfactory interfacial compatibility ensures structural uniformity and homogeneity, as further verified by the SEM-EDS elemental mapping.

To further elucidate the chemical environment changes and interfacial interactions, synchrotron X-ray absorption spectroscopy was employed to probe the local environment of Zn atoms for the composite. The X-ray absorption near-edge structure (XANES) spectra revealed distinct intensity differences between the ZIF-62_CN-25 crystalline composite and the a_gZIF-62_CN-25 glass composite (Fig. 3a). In particular, the feature at ∼9673 eV decreased in intensity, while the peak at ∼9667 eV became slightly more pronounced in the glass composite. These spectral variations indicate modifications in the local electronic environment and coordination symmetry of Zn,³² consistent with a slight reduction in the average coordination number. To further resolve the coordination shells, extended X-ray absorption fine structure (EXAFS) analysis (Fig. 3b) combined with wavelet transform (WT) mapping was performed (Fig. 3c, d and S8). Both methods confirmed differences in bond distances and coordination numbers (Table S1), with the glass composite showing an additional weak contribution from Zn–Zn atom pairs. The weak Zn–Zn contribution at ca. 2.5 Å with a low coordination number indicates that a small fraction of Zn sites undergo local rearrangement during melt-quenching, although minor thermally induced local decomposition cannot be fully excluded. This interpretation is consistent with the additional DFT cluster models, which suggest a locally off-centered and distorted Zn coordination environment near the g-C₃N₄ interface. These results demonstrate that vitrification preserves the Zn–N tetrahedral framework while introducing weak but detectable Zn–Zn correlations, which likely reflect partial inter-tetrahedral proximity induced by structural rearrangements and interfacial confinement during melt-quenching.³³

Fig. 3 Characterization of the chemical environment of the composites. (a) Normalized Zn K-edge XANES spectra of ZIF-62_CN-25 and a_gZIF-62_CN-25. (b) X-ray absorption fine structure (EXAFS) Fourier transform magnitude spectrum of ZIF-62_CN-25 and a_gZIF-62_CN-25. (c and d) Full-range wavelet transform (WT) representation of the Zn K edge of (c) the ZIF-62_CN-25 composite and (d) a_gZIF-62_CN-25 glass composite. (e) Second derivative spectra corresponding to 270–320 cm⁻¹ and 610–650 cm⁻¹ ex situ THz/FarIR of a_gZIF-62 and a_gZIF-62_CN-x composites. (f) DFT calculated IM₃Zn(IM)Zn-CN tetrahedron spectra in the wavenumber range of 1000 cm⁻¹, with the schematic diagram of the corresponding vibrational movements for DFT calculations at 298 K. Zn, blue; C, green; N, orange; and H, white. (g) In situ THz/FarIR spectra for ZIF-62_CN-25 up to 573 K. (h) Second derivative spectra corresponding to subtle changes of (g) in situ THz/FarIR spectra.

The collective framework motion of localized environments in glass composite materials is exposed by terahertz (THz)/far-infrared (FarIR) spectroscopy. Compared to a_gZIF-62 and pristine g-C₃N₄ (Fig. S9), glass composite materials with different loadings exhibit two key structural changes in Fig. S10. Firstly, in the low-frequency region, a bond around 300 cm⁻¹ shifts toward higher wavenumbers in the glass composite relative to pure a_gZIF-62 (Fig. 3e and S10), which is attributed to Zn-centered framework deformation within Zn tetrahedra. Secondly, a small hump appears near 630 cm⁻¹ in the composite, more clearly visible in the second derivative analysis shown in Fig. 3(e), which is consistent with the deformation behavior of the aromatic ring within the structure.

To reveal the mechanism of structural deformation and interfacial interactions of glass composite materials, we carried out simulations using density functional theory (DFT). The coordination environments between Zn and Lewis basic functions of the g-C₃N₄ structure give rise to the bond variation for glass composite materials. Specifically, when Zn interacts with the N sites of g-C₃N₄, partial interactions between ZIF and g-C₃N₄ (IM₃Zn(IM)ZnIM–CN) happen (Fig. S11), leading to a slight shift of the band to ∼260 cm⁻¹, whereas complete accommodation of Zn in the in-plane intrinsic g-C₃N₄ pores (IM₃Zn(IM)Zn–CN) resulted in a steric deformation with a result of a more pronounced shift to ∼280 cm⁻¹ and the appearance of a new feature, which is in qualitative agreement with the experimental observations (Fig. 3f). Additional DFT cluster models (Fig. S22) with increased structural size suggest that Zn becomes locally off-centered near the g-C₃N₄ interface, giving rise to a broader distribution of Zn–N distances (Table. S3). Mode analysis indicates that stronger interfacial coupling enhances Zn-centered framework deformations and mixes them with low-frequency motions of the g-C₃N₄ scaffold. It should be noted that these finite DFT cluster models are used to probe local Zn coordination geometry and short-range vibrational features, rather than to fully reproduce long-wavelength collective phonon modes in the bulk glass. Temperature-resolved synchrotron terahertz (THz) far-infrared (FarIR) vibrational spectroscopy was further employed to monitor the bonding evolution during the formation of a_gZIF-62_CN-25 (Fig. 3g and h). Upon heating from 25 to 293 °C, Zn-related vibration remains largely stable in the raw spectra, suggesting the relative persistence of Zn-centered local environments during thermal treatment. The 400–640 cm⁻¹ region remains largely invariant upon heating, with only minor changes near ∼630 cm⁻¹. But there is still the same phenomenon as the ex situ THz/FarIR result, showing a peak splitting around 270 cm⁻¹ and peak broadening around 630 cm⁻¹. A detailed analysis with second-derivative spectra (Fig. 3h and S12) gives more information: in the 200–400 cm⁻¹ region, the emergence of a new mode at ca. 300 cm⁻¹ clearly visible upon heating to 300 °C is consistent with the new Zn–N bond, accompanied by a slight frequency shift and intensity redistribution. The correlated shifts in Zn-centered collective modes reveal that strong interfacial interactions are established during the thermally induced reorganization process. In the 600–700 cm⁻¹ region, the processed spectra (Fig. S12) indicate that the corresponding molecular structure undergoes strong intense frequency vibrations during the re-coordination process. These vibrations can be attributed to in-plane and out-of-plane aromatic ring deformation modes of imidazolate,³⁴ which are further associated with π–π interactions between the imidazolate-based linkers and the heptazine units of g-C₃N₄. These results suggest that local Zn–N re-coordination between Zn centers and Lewis-basic nitrogen sites in g-C₃N₄ modifies the Zn-centered local geometry and induces interfacial structural distortion, rather than forming a perfectly pristine coordination environment.

Fourier transform infrared spectroscopy (FTIR) was employed to investigate the stretching vibrations of chemical bonds and functional groups in a_gZIF-62_CN-x composites before and after sintering (Fig. 4a and S13). The spectra of a_gZIF-62_CN-x composites exhibit characteristic vibrational features from both g-C₃N₄ and a_gZIF-62. Specifically, the C–N stretching vibration from ZIF-62 appears at ∼1100 cm⁻¹, which can also be observed in a_gZIF-62 and a_gZIF-62_CN-x composites. In addition, from composites' spectra, the broad absorption band in the range of 3250–2970 cm⁻¹ can be assigned to the N–H stretching vibrations of terminal –NH₂/–NH groups in g-C₃N₄,³⁵ consistent with the spectrum of pristine g-C₃N₄ (Fig. S13). These results confirm the coexistence of both structural motifs within the composites. In the FTIR spectra of the a_gZIF-62_CN-25 and a_gZIF-62_CN-40 glass composites, a remarkable absorption band appears at ∼1033 cm⁻¹ which is absent in pristine a_gZIF-62. This new band is assigned to a C–N stretching vibration^36,37 associated with interfacial interactions between ZIF-62 and g-C₃N₄. The emergence of this new vibrational peak indicates that the composite formation process induces local structural rearrangement, changing the C–N bonding environment and potentially forming new coordination states. To further corroborate these findings, X-ray photoelectron spectroscopy (XPS) was performed on ZIF-62, a_gZIF-62, and a_gZIF-62_CN-25 (Fig. S14). All samples appear to have corresponding Zn peaks of 2p_3/2 and 2p_1/2, respectively. However, the Zn 2p binding energies in the sintered samples (1045.24 eV and 1022.14 eV) are shifted towards higher bonding energies relative to those in the crystalline state (1044.9 eV and 1021.8 eV), indicating decreased electron density around Zn and the formation of stronger Zn–N interfacial coordination bonds.³⁸

The improved processing integrity of the composite is first evidenced at the macroscopic level. As shown in Fig. S18, pure a_gZIF-62 pellets consistently develop visible cracks upon cooling from the melt, whereas a_gZIF-62_CN-25 pellets prepared under identical conditions remain macroscopically intact. In addition, to demonstrate the feasibility of employing the composite materials as self-supporting devices, their mechanical properties were further evaluated. Nanoindentation was employed to compare the local mechanical response of the pellets, including elastic stiffness and resistance to plastic deformation (Fig. 4b). Measurements were performed using a Berkovich diamond tip under load-control mode (maximum load 1 mN; loading-holding-unloading sequence of 10 s, 5 s, and 15 s, respectively). As shown in Fig. 4b, both the hardness (H) and elastic modulus (E) of the a_gZIF-62_CN-x composites depend strongly on the g-C₃N₄ loading. Pristine a_gZIF-62 exhibited an E of 6.29 ± 0.26 GPa and H of 0.76 ± 0.06 GPa, in line with the reported values.^39,40 The incorporation of 10 wt% g-C₃N₄ produced a modest increase in both parameters, while 25 wt% g-C₃N₄ resulted in the highest E (8.18 ± 0.33 GPa) and H (1.1 ± 0.03 GPa), demonstrating an optimal reinforcement effect. Carbon nitride is an exceptionally stiff and hard material, with its bulk modulus and hardness reported to approach those of diamond.⁴¹ Thus, integrating g-C₃N₄ into a_gZIF-62 is expected to enhance both E and H. Moreover, π–π stacking interactions between the conjugated heptazine units of g-C₃N₄ and the imidazolate-based linkers of ZIF-62 likely enhance interfacial load transfer and improve membrane formability, contributing to the effective reinforcement. Additionally, the relatively processable and loosely stacked layered structure of g-C₃N₄ acts as a stress-buffering phase, improving processing tolerance during materials handling and shaping.⁴² However, excessive g-C₃N₄ addition leads to a marked decrease in both E and H at 40 wt% loading. This deterioration is unlikely to arise from nanosheet agglomeration alone but may also reflect a loading-dependent wetting/compatibility limitation during melt processing. g-C₃N₄ contains abundant nitrogen sites and terminal –NH/–NH₂ groups, giving it a relatively polar surface, whereas the imidazolate-rich ZIF-62 glass matrix is comparatively less polar. At moderate loading, interfacial interactions between g-C₃N₄ and ZIF-62 can promote effective reinforcement. However, when the g-C₃N₄ content becomes excessive, the molten ZIF-62 phase may be insufficient to fully wet and confine the increasing nanosheet surface area, resulting in nanosheet self-aggregation, structural heterogeneity, and inefficient load transfer. These effects collectively account for the reduced E and H observed at 40 wt% loading. To further probe the effect of g-C₃N₄ incorporation on the melt and glass-forming dynamics,¹⁶ DSC-based fragility analysis was additionally performed (Fig. S23). The fragility index (m) decreases slightly from 23.18 for pristine a_gZIF-62 to 22.01 for a_gZIF-62_CN-25, indicating moderately suppressed structural relaxation dynamics and slightly stronger glass-forming behaviour after g-C₃N₄ incorporation.

Fig. 4 Physicochemical properties of glass composites. (a) FTIR spectra of g-C₃N₄, a_gZIF-62, a_gZIF-62_CN-10, a_gZIF-62_CN-25, and a_gZIF-62_CN-40. (b) Hardness–modulus correlation of glass composite pellets. (c) CO₂ adsorption isotherm curves and (d) pore size distribution at 273 K of a_gZIF-62, a_gZIF-62_CN-10, a_gZIF-62_CN-25, and a_gZIF-62_CN-40.

To evaluate the evolution of microstructures and accessible microporosity, gas adsorption–desorption isotherm tests were carried out on the composites before and after sintering. Previous gas adsorption studies on ZIF glass materials demonstrated that when N₂ was used as a probe gas, it could not access the micropores of a_gZIF-62 due to kinetic limitations.¹⁴ Therefore, CO₂ was employed as the probe gas to accurately determine the Burnauer–Emmett–Teller (BET) surface areas for these samples (Fig. S15), where CO₂ adsorption at 195 K occurs under sub-critical conditions, enabling efficient micropore filling which can be reliably measured up to P/P₀ = 1 with more accurate estimation of micropore volumes. The BET surface area of a_gZIF-62_CN-25 is 115.2 m² g⁻¹, compared with 134.4 m² g⁻¹ for a_gZIF-62. Notably, the composite shows a narrower hysteresis, indicating lower desorption barriers or a more readily recoverable pore network.⁴³ In addition, CO₂ adsorption at 273 K was applied to analyze addition isotherms and pore size distributions with an NLDFT model (Fig. 4c and d) under conditions more relevant to practical gas separation.²¹ Although both isotherms' datasets are discussed in terms of relative pressure, the absolute saturation pressure P₀ differs: at 195 K, P₀ ≈ 1 bar, whereas at 273 K, P₀ ≈ 35 bar, leading to different thermodynamic driving forces and transport behaviors. At a relative pressure of P/P₀ = 0.035, a_gZIF-62_CN-10 shows the highest CO₂ adsorption capacity of 18.12 cm³ g⁻¹ at STP. With further increasing of g-C₃N₄ (25 and 40 wt%), the gas uptake capacity decreases, approaching the value of pristine a_gZIF-62. This decline is attributed primarily to a compositional dilution of the pore-forming glassy phase rather than pore blocking, because g-C₃N₄ itself exhibits very low CO₂ uptake at 273 K (Fig. S16) and contributes minimally to accessible microporosity. Based on NLDFT (CO₂ 273 K), the pore-size distributions (PSD) show a pronounced population of ultramicropores (∼3.5–6 Å) in a_gZIF-62_CN-10 and a_gZIF-62_CN-25, consistent with its higher CO₂ uptake. With increasing g-C₃N₄ loading, the overall PSD intensity decreases and shifts slightly toward larger pores (Fig. 4d). Pure a_gZIF-62 displays only a weak contribution in the same range which was caused by pores collapsing during the melting quenching process. The isosteric heat of CO₂ adsorption (Q_st), obtained from dual-temperature isotherms under 273 and 298 K, via the Clausius–Clapeyron approach (Fig. S17),⁴⁴ further corroborates this result. a_gZIF-62_CN-25 exhibits a higher Q_st at low coverage than a_gZIF-62, evidencing stronger CO₂ surface interactions in the composite. With increasing uptake, Q_st for both samples converges to 30 kJ mol⁻¹ at 7.45 mg g⁻¹, consistent with saturation of the strongest interfacial sites followed by filling of more uniform and weaker sites.^45–48 Therefore, we found that at low temperature, the pure glass engages in stronger site-specific interactions under diffusion-limited conditions, whereas near ambient temperature the interfacial sites and improved accessibility in a_gZIF-62_CN-25 dominate.

Owing to its good processability and mechanical robustness, the absence of significant microscopic defects, and the presence of accessible micropores, this composite was employed to fabricate separation membranes via a hot-pressing approach. This platform enables us to systematically investigate how modifications in the stacking of the two-dimensional g-C₃N₄ sheets influence the gas transport and separation performance of the resulting composite membranes. The single gas permeation performance was systematically evaluated at 30 °C (Fig. 5 and Table S2). Because the present membrane is a thick self-supporting MOF glass–2D nanosheet composite rather than a homogeneous dense film, macroscopic thickness-normalized permeability values should be interpreted with caution. The gas separation performance is therefore discussed mainly in terms of measured permeance and selectivity, together with comparison to related self-supported MOF glass and porous framework-based membranes. For pristine a_gZIF-62, the H₂ permeance was limited to 14.9 GPU with a low ideal H₂/CO₂ selectivity, reflecting the intrinsic transport restriction of the dense and disordered glassy network (Fig. 5a). Upon incorporation of g-C₃N₄ nanosheets, the permeance of all tested gases increased significantly, with H₂ showing the most pronounced enhancement. Among all compositions in Fig. 5b, a_gZIF-62_CN-25 shows the best performance, achieving both high H₂ permeance (1716 GPU) and H₂/CO₂ selectivity (21). Higher loading (40 wt%) increases H₂ permeance to 2133.5 GPU but sacrifices H₂/CO₂ selectivity (12.5) due to partial nanosheet aggregation and the formation of non-selective pathways. Thus, a_gZIF-62_CN-25 represents an optimal balance between interfacial compatibility, nanosheet dispersion, and molecular sieving. The single-gas permeance trend reflects a strong size-sieving effect (Fig. 5c), where H₂ diffuses quickly whereas CO₂ shows a sharp reduction in flux, followed by a plateau for N₂ and CH₄. The tightened interlayer spacing (ca. 0.32 nm) falls between the kinetic diameters of H₂ (0.289 nm) and CO₂ (0.33 nm), enabling facile H₂ diffusion while imposing an orientation-dependent barrier for CO₂.^30,49 It should be noted that this does not imply that gas transport occurs exclusively through the g-C₃N₄ interlayers. The CO₂ adsorption-derived pore-size distribution reflects accessible pore cavities and free-volume elements, whereas membrane sieving is governed by the effective aperture and connectivity of transport channels. Correspondingly, the measured a_gZIF-62_CN-25 ideal selectivity for H₂/CO₂ (21), H₂/N₂ (14.6), and H₂/CH₄ (10.9) substantially exceeds their respective Knudsen values (4.7, 3.7, and 2.8). Fig. S21 compares the H₂/CO₂ separation performance of the a_gZIF-62_CN-x membranes with MOF, COF, 2D material and porous framework-based membranes reported in the literature at room temperature. Despite being a self-supporting membrane with a thickness of several hundred micrometres, a_gZIF-62_CN-25 exhibits high H₂ permeance while maintaining considerable H₂/CO₂ selectivity, placing its overall performance within the upper range of emerging porous framework-derived functional membranes.

Fig. 5 Gas permeation studies. (a) Schematic illustration of interlayer spacing manipulation and gas separation mechanism of a_gZIF-62_CN-x composites. Single gas permeance test at 30 °C for (b) a_gZIF-62_CN-x CGC treated at 440 °C, (c) a_gZIF-62_CN-25 CGC obtained at 440 °C, and (d) a_gZIF-62_CN-25 CGC under different temperature annealing.

To further distinguish diffusion-related transport trends across the membrane series, apparent diffusivity values were obtained from pure-gas time-lag measurements and summarized in Table S4. It should be noted that these values are used only for internal comparison among membranes tested under the same conditions, rather than as intrinsic material constants, because the thick self-supporting composite pellets may not behave as homogeneous dense membrane phases. Despite this limitation, H₂ exhibits a higher apparent diffusivity than CO₂ across the membrane series, supporting preferential H₂ diffusion through confined microporous or interfacial pathways. Therefore, the enhanced H₂/CO₂ separation performance is reasonably attributed to the cooperative effects of the introduced interlayer spacing, improved pore and interlayer connectivity, and faster H₂ diffusion through the composite transport network.

To assess the separation performance under more practical conditions, mixed-gas H₂/CO₂ permeation tests were further conducted using a Wicke–Kallenbach apparatus under steady-state conditions in Fig. S19. The mixed-gas separation factor remained close to the ideal selectivity with only a slight decrease, indicating that competitive CO₂ adsorption does not significantly disrupt the H₂ selective transport pathway within the rigid MOF glass composite structure. In addition, long-term permeation measurements over ∼90 h (Fig. S20) further demonstrate that the a_gZIF-62_CN-25 membrane maintains stable H₂ permeance and H₂/CO₂ selectivity with only minor fluctuations, suggesting that the incorporation of g-C₃N₄ helps suppress rapid structural relaxation of the glass matrix and stabilizes the composite transport network during operation.

To further probe the role of thermal treatment and change of interlayer space, a_gZIF-62_CN-25 membranes were annealed at different temperatures between 400 and 460 °C (Fig. 5d). H₂ permeance remained above 1000 GPU across all samples, but showed a slight decrease compared with the increase of treatment temperature, aligning well with the change of interlayer space. In contrast, CO₂ permeance decreased more noticeably with increasing treatment temperature, leading to a modest but consistent improvement in H₂/CO₂ selectivity. The highest H₂/CO₂ selectivity of 23.4 was achieved after annealing at 460 °C, indicating that high-temperature treatment reinforces Zn–N coordination and π–π interactions. These synergistic interfacial interactions between g-C₃N₄ and the ZIF-62 glass matrix enhance interlayer confinement and promote the formation of continuous, well-connected transport channels, facilitating rapid H₂ diffusion and superior molecular sieving. Overall, these results demonstrate that nanoscale incorporation combined with thermal treatment can precisely tune interfacial coordination and interlayer spacing, enabling hybrid glass membranes with both high H₂ permeance and elevated H₂/CO₂ selectivity. More broadly, this strategy establishes a general structure–performance paradigm for designing multifunctional hybrid materials, where interfacial and nanoscale control can be leveraged to optimise transport, mechanical, and functional properties.

Conclusions

We have demonstrated in this work that the integration of g-C₃N₄ nanosheets into ZIF-62 glass through high temperature sintering enables precise control and stabilization of interlayer spacing, formed via strong and effective interfacial interactions. The incorporation of nanosheets also enhances processability and mechanical integrity. Future work should include direct quantification of fracture toughness using single-edge notched beam testing or micro-cantilever bending to better understand the bulk crack resistance of these composites. A more comprehensive rheological analysis of the composite melt is also identified as an important direction for future study. The resultant hybrid glass membranes show significantly improved membrane hydrogen separation performance. Beyond gas separation, this approach provides a versatile platform for designing hybrid amorphous materials with tunable microstructure and interface properties. It could be further extended to other two-dimensional or layered materials, opening opportunities for nanosheet materials assembly towards applications like selective separation, catalysis, ion transport, and beyond.

Author contributions

J. H. and J. Y. conceived the idea for this research and contributed to drafting and editing the manuscript. S. Z., B. C. and W. L. conducted testing experiments and collected the data. W. H. and M. Y. carried out data analysis. Zhonghao Xu prepared the figures and visualization materials. M. C., S. Y. and M. L. provided training on characterization instruments. W. L., X. Z., V. C., J. H. and R. L. provided instrumentation access and supervised the project with critical feedback. All authors have given approval to the final version of the manuscript.

Conflicts of interest

There are no conflicts to declare.

Data availability

All of the necessary data have been included in the main text and the supplementary information (SI). Supplementary information is available. See DOI: https://doi.org/10.1039/d6sc00077k.

Acknowledgements

The authors acknowledge the funding support from the Australian Research Council (ARC) (FT210100589 and DP230101901). This research was conducted at the ARC Centre of Excellence for Green Electrochemical Transformation of Carbon Dioxide (CE230100017) and funded by the Australian Government. The authors acknowledge the Natural Science Research Project of Anhui Provincial Education Department (2022AH030135). This research was undertaken on the THz, PD and X-ray Absorption beamlines at the Australian Synchrotron, part of ANSTO.

Notes and references

1. H.-N. Li, Z.-Y. Sun, Z.-J. Yu, K. Man, C. Zhang and Z.-K. Xu, Nat. Commun., 2025, 16, 8199.
2. P. Motamed, R. S. Othman, M. Golriz, M. Naqvi, A. Afshar Taromi and H. A. Khonakdar, ChemistrySelect, 2025, 10, e03697.
3. C. Ji, H. Shao, Y. Pu, H. Zhang, H. Li, W.-H. Li, N. Zhu, W. Zhao, D. Jiang and D. Zhao, Adv. Mater., 2025, e13499.
4. Y. Zhang, B. H. Yin, L. Huang, L. Ding, S. Lei, S. G. Telfer, J. Caro and H. Wang, Prog. Polym. Mater. Sci., 2025, 151, 101432.
5. Y. Feng, W. Yan, Z. Kang, X. Zou, W. Fan, Y. Jiang, L. Fan, R. Wang and D. Sun, Chem. Eng. J., 2023, 465, 142873.
6. G. Liu, Y. Guo, C. Chen, Y. Lu, G. Chen, G. Liu, Y. Han, W. Jin and N. Xu, Nat. Mater., 2023, 22, 769–776.
7. S. Li, C. Ma, J. Hou, S. Yu, A. Chen, J. Du, P. A. Chater, D. S. Keeble, Z. Qiao, C. Zhong, D. A. Keen, Y. Liu and T. D. Bennett, Nat. Commun., 2025, 16, 1622.
8. D. Li, Z. Yang, L. Yang, C. Ma, M. Ye, Y. Sun, Z. Qiao and A. Chen, J. Membr. Sci., 2024, 695, 122492.
9. M. Kim, H.-S. Lee, D.-H. Seo, S. J. Cho, E. Jeon and H. R. Moon, Nat. Commun., 2024, 15, 1174.
10. Z. Chen, Z. Xu, X. Zhong, Y. Chen, S. Yang, Y. Feng, P. Zhang, Y. Li, M. Xue and X. Chen, Adv. Mater., 2025, e12654.
11. J. Li, J. Wang, Q. Li, M. Zhang, J. Li, C. Sun, S. Yuan, X. Feng and B. Wang, Angew. Chem., Int. Ed., 2021, 60, 21304–21309.
12. A. E. Khudozhitkov, N. Ogiwara, M. Donoshita, H. Kobayashi, A. G. Stepanov, D. I. Kolokolov and H. Kitagawa, J. Am. Chem. Soc., 2024, 146, 12950–12957.
13. T. To, S. S. Sørensen, M. Stepniewska, A. Qiao, L. R. Jensen, M. Bauchy, Y. Yue and M. M. Smedskjaer, Nat. Commun., 2020, 11, 2593.
14. L. Frentzel-Beyme, P. Kolodzeiski, J.-B. Weiß, A. Schneemann and S. Henke, Nat. Commun., 2022, 13, 7750.
15. J. C. Mauro, Y. Yue, A. J. Ellison, P. K. Gupta and D. C. Allan, Proc. Natl. Acad. Sci. U.S.A., 2009, 106, 19780–19784.
16. A. Qiao, T. D. Bennett, H. Tao, A. Krajnc, G. Mali, C. M. Doherty, A. W. Thornton, J. C. Mauro, G. N. Greaves and Y. Yue, Sci. Adv., 2018, 4, eaao6827.
17. O. Smirnova, R. Sajzew, S. J. Finkelmeyer, T. Asadov, S. Chattopadhyay, T. Wieduwilt, A. Reupert, M. Presselt, A. Knebel and L. Wondraczek, Nat. Commun., 2024, 15, 5079.
18. O. N. Senkov, Phys. Rev. B: Condens. Matter Mater. Phys., 2007, 76, 104202.
19. T. D. Bennett, J.-C. Tan, Y. Yue, E. Baxter, C. Ducati, N. J. Terrill, H. H.-M. Yeung, Z. Zhou, W. Chen, S. Henke, A. K. Cheetham and G. N. Greaves, Nat. Commun., 2015, 6, 8079.
20. T. D. Bennett, S. Horike, J. C. Mauro, M. M. Smedskjaer and L. Wondraczek, Nat. Chem., 2024, 16, 1755–1766.
21. L. N. McHugh and T. D. Bennett, J. Mater. Chem. A, 2022, 10, 19552–19559.
22. F. Cui, V. García-López, Z. Wang, Z. Luo, D. He, X. Feng, R. Dong and X. Wang, Chem. Rev., 2025, 125, 445–520.
23. Y. Ren and Y. Xu, Chem. Soc. Rev., 2024, 53, 1823–1869.
24. T. Bao, X. Li, S. Li, H. Rao, X. Men, P. She and J. Qin, Nano Mater. Sci., 2025, 7, 145–168.
25. F. K. Kessler, Y. Zheng, D. Schwarz, C. Merschjann, W. Schnick, X. Wang and M. J. Bojdys, Nat. Rev. Mater., 2017, 2, 17030.
26. C.-C. Wang, X.-H. Yi and P. Wang, Appl. Catal., B, 2019, 247, 24–48.
27. Y. Liu, D. Luo, Y. Wang, X. Ma, S. Shen, B. Bai and C. Sun, Nano Lett., 2025, 25, 14582–14590.
28. Y. Wang, T. Lian, N. V. Tarakina, J. Yuan and M. Antonietti, Nat. Commun., 2022, 13, 7339.
29. X. Li, W. Huang, A. Krajnc, Y. Yang, A. Shukla, J. Lee, I. Martens, B. Chan, D. Appadoo, P. Chen, X. Weng, J. A. Steele, Q. Sun, G. Mali, R. Lin, N. Bedford, V. Chen, K. Cheetham, L. H. G. Tizei, S. M. Collins, L. Wang and J. Hou, Nat. Commun., 2023, 14, 7612.
30. Y. Zhou, Y. Wu, H. Wu, J. Xue, L. Ding, R. Wang and H. Wang, Nat. Commun., 2022, 13, 5852.
31. H. Wang, D. Chen, X. An, Y. Zhang, S. Sun, Y. Tian, Z. Zhang, A. Wang, J. Liu, M. Song, S. P. Ringer, T. Zhu and X. Liao, Sci. Adv., 2021, 7, eabe3105.
32. W. Huang, B. Chan, Y. Yang, P. Chen, J. Wang, L. Casey, C. Atzori, T. Schulli, O. Mathon, H. G. Hackbarth, N. M. Bedford, D. Appadoo, X. Li, T. Lin, R. Lin, J. Lee, Z. Wang, V. Chen, A. K. Cheetham, L. Wang and J. Hou, J. Am. Chem. Soc., 2025, 147, 3195–3205.
33. W. Huang, W. Xue, P. Chen, Z. Wang, R. Lin, S. Zhao, S. Yu, Y. Chen, X. Zha, D. Appadoo, T. Tian, V. Chen, S. Henke, A. K. Cheetham, L. Wang and J. Hou, Angew. Chem., 2025, 137, e202506570.
34. J. Hou, M. L. Ríos Gómez, A. Krajnc, A. McCaul, S. Li, A. M. Bumstead, A. F. Sapnik, Z. Deng, R. Lin, P. A. Chater, D. S. Keeble, D. A. Keen, D. Appadoo, B. Chan, V. Chen, G. Mali and T. D. Bennett, J. Am. Chem. Soc., 2020, 142, 3880–3890.
35. H. Liu, B. Yang, G. Liao, B. Huang, J. Li, R. D. Rodriguez and X. Jia, Nat. Commun., 2025, 16, 5909.
36. Z. Öztürk, M. Filez and B. M. Weckhuysen, Chem.–Eur. J., 2017, 23, 10915–10924.
37. Y. Hu, Z. Liu, J. Xu, Y. Huang and Y. Song, J. Am. Chem. Soc., 2013, 135, 9287–9290.
38. R. Lin, J. Hou, M. Li, Z. Wang, L. Ge, S. Li, S. Smart, Z. Zhu, T. D. Bennett and V. Chen, Chem. Commun., 2020, 56, 3609–3612.
39. T. D. Bennett, Y. Yue, P. Li, A. Qiao, H. Tao, N. G. Greaves, T. Richards, G. I. Lampronti, S. A. T. Redfern, F. Blanc, O. K. Farha, J. T. Hupp, A. K. Cheetham and D. A. Keen, J. Am. Chem. Soc., 2016, 138, 3484–3492.
40. R. Lin, M. Chai, Y. Zhou, V. Chen, T. D. Bennett and J. Hou, Chem. Soc. Rev., 2023, 52, 4149–4172.
41. H. Dong, A. R. Oganov, Q. Zhu and G.-R. Qian, Sci. Rep., 2015, 5, 9870.
42. L. Chen and J. Song, Adv. Funct. Mater., 2017, 27, 1702695.
43. S. Yang, X. Lin, W. Lewis, M. Suyetin, E. Bichoutskaia, J. E. Parker, C. C. Tang, D. R. Allan, P. J. Rizkallah, P. Hubberstey, N. R. Champness, K. Mark Thomas, A. J. Blake and M. Schröder, Nat. Mat., 2012, 11, 710–716.
44. M. Yin, L. Wang and S. Tang, ACS Catal., 2023, 13, 13021–13033.
45. J. Seo, R. Matsuda, H. Sakamoto, C. Bonneau and S. Kitagawa, J. Am. Chem. Soc., 2009, 131, 12792–12800.
46. F. Li, X. Yue, Y. Liao, L. Qiao, K. Lv and Q. Xiang, Nat. Commun., 2023, 14, 3901.
47. Z. Zhang, Y. Chen, K. Chai, C. Kang, S. B. Peh, H. Li, J. Ren, X. Shi, X. Han, C. Dejoie, S. J. Day, S. Yang and D. Zhao, Nat. Commun., 2023, 14, 3789.
48. G. Férey and C. Serre, Chem. Soc. Rev., 2009, 38, 1380–1399.
49. R. Wang, J. Qian, X. Chen, Z.-X. Low, Y. Chen, H. Ma, H.-A. Wu, C. M. Doherty, D. Acharya, Z. Xie, M. R. Hill, W. Shen, F. Wang and H. Wang, Nat. Commun., 2023, 14, 2161.

---