Report

Metal-organic framework nanosheets as building blocks for molecular sieving membranes

See allHide authors and affiliations

Science  12 Dec 2014:
Vol. 346, Issue 6215, pp. 1356-1359
DOI: 10.1126/science.1254227

Abstract

Layered metal-organic frameworks would be a diverse source of crystalline sheets with nanometer thickness for molecular sieving if they could be exfoliated, but there is a challenge in retaining the morphological and structural integrity. We report the preparation of 1-nanometer-thick sheets with large lateral area and high crystallinity from layered MOFs. They are used as building blocks for ultrathin molecular sieve membranes, which achieve hydrogen gas (H2) permeance of up to several thousand gas permeation units (GPUs) with H2/CO2 selectivity greater than 200. We found an unusual proportional relationship between H2 permeance and H2 selectivity for the membranes, and achieved a simultaneous increase in both permeance and selectivity by suppressing lamellar stacking of the nanosheets.

Metal-organic framework material membranes

There continues to be a lot of interest in developing membranes for gas separations that go beyond the current polymer membranes used commercially for this purpose. Peng et al. took a porous metal-organic framework material with a layered structure and exfoliated it to give nanometer-thick molecular sieves. The membranes were exceptionally good at separating hydrogen gas from carbon dioxide both in terms of permeance and selectivity.

Science, this issue p. 1356

Gas separation with membranes is an energy-efficient and environmentally friendly alternative to cryogenic and adsorptive or absorptive gas separation processes. Polymer membranes are subject to a trade-off between productivity (permeability) and efficiency (selectivity), known as Robeson’s upper bound (1, 2). Membranes based on molecular sieve materials are expected to overcome this limitation by relying on their ability to distinguish molecules based on size and shape (38). Molecular sieve nanosheets (MSNs) with large lateral area and small thickness are the most appropriate building blocks for ultrathin, and thus ultrapermeable, molecular sieve membranes (9). The permeance of such membranes is measured in gas permeation units (1 GPU = 10−6 cm3 cm–2 s–1 cmHg–1 at STP). Tsapatsis et al. demonstrated the fabrication of molecular sieve membranes based on exfoliated zeolite nanosheets with thickness at the unit cell level (~3 nm) (10), whereas the types of zeolites that can be easily exfoliated are rather limited (11). Recently, graphene oxide (GO) nanosheets with selective defects were used to produce ultrathin membranes with thickness of as low as 1.8 nm (12). However, the measured H2 permeance of these extremely thin GO membranes, ~300 GPUs, was still at the same level as conventional microporous membranes (13). This can be attributed to the low density of selective defects and their random distribution in the GO nanosheets.

In recent years, metal-organic frameworks (MOFs) have emerged as a family of nanoporous molecular sieves. A large number of layered MOFs have been reported (14), and they could serve as a diverse source of MSNs if they could be exfoliated to nanometer-scale thickness (1517). However, structural deterioration and morphological damage have hampered success in obtaining high-quality MOF nanosheets, which has hindered the application of the nanosheets as building blocks for molecular sieve membranes. We report the preparation of 1-nm-thick MSNs with large lateral area and high crystallinity from layered MOF precursors and demonstrate their use in fabricating ultrapermeable membranes that have excellent molecular sieving properties for H2/CO2 separation. H2/CO2 membrane separation is considered one of the key technologies for zero-emission fossil fuel power generation and hydrogen production (18).

The layered MOF precursor exemplified here is poly[Zn2(benzimidazole)4], denoted Zn2(bim)4 (19). The corresponding scanning electron microscopy (SEM) images are shown in Fig. 1A. In this structure, two-dimensional (2D) layers are oriented normal to the c axis, connected by weak van der Waals interactions (Fig. 1A, inset). In the layers, each Zn ion is coordinated by four benzimidazole (bim) ligands in a distorted tetrahedral geometry, and each bim ligand bridges two Zn atoms via a bis-monodentate linkage (Fig. 1B). Powder x-ray diffraction (XRD) measurements confirm that our product is isostructural to a previously determined MOF (Fig. 1C) [ref. no. 675375, Cambridge Crystallographic Data Centre (CCDC)] (20). Zn2(bim)4 can also be obtained in high yield via hydrothermal transformation of the well-known three-dimensional zeolitic imidazolate framework ZIF-7 (19, 21) (fig. S1), which implies that Zn2(bim)4 possesses excellent thermodynamic stability (fig. S2).

Fig. 1 Top-down fabrication of molecular sieve nanosheets.

(A) Scanning electron microscopy (SEM) image of as-synthesized Zn2(bim)4 crystals. The inset image shows the typical flake-like morphology of Zn2(bim)4 crystals. (B) Architecture of the layered MOF precursor. The ab planes are highlighted in purple to better illustrate the layered structure. (C) Powder XRD patterns of Zn2(bim)4. The top trace is the experimental pattern, whereas the bottom trace is the pattern simulated based on the single-crystal data (CCDC-675375). The asymmetric unit of Zn2(bim)4 is also presented to illustrate the coordination environment of Zn atoms. (D) Transmission electron microscopy (TEM) image of Zn2(bim)4 MSNs. The inset shows the Tyndall effect of a colloidal suspension. (E) Illustration of the grid-like structure of the Zn2(bim)4 MSN. The Zn coordination polyhedra are depicted in blue, whereas the bim links are represented by sticks. H atoms are omitted for clarity. Symmetry code: A 2–x, y, 1.5–z. (F) Space-filling representation of a four-membered ring of the Zn2(bim)4 MSN.

Conventional physical exfoliation may damage the in-plane structure of MOF nanosheets. Soft-chemical exfoliation is a potential alternative (22), but requires chemicals that may negatively affect application of the nanosheets. To address these problems, we developed a soft-physical process: Pristine Zn2(bim)4 crystals were first wet ball-milled at very low speed (60 rpm), followed by exfoliation in volatile solvent with the aid of ultrasonication. We found that a mixture of methanol and propanol is the most appropriate for this process (fig. S3), based on our analysis of a number of solvents. We hypothesize that wet ball-milling facilitates the penetration of small methanol molecules into the galleries of the layered Zn2(bim)4, and propanol helps to stabilize the exfoliated nanosheets by adsorbing on the sheets with its hydrophobic alkane tails. The colloid suspension of Zn2(bim)4 MSNs is clear but exhibits a Tyndall effect due to light scattering by the nanosheets in the colloid (Fig. 1D, inset).

Figure 1E illustrates the network structure of a single-layered Zn2(bim)4 MSN. The square Zn4(bim)4 unit, as shown in the red circle, can be considered to be the subunit of the 2D layer. The aperture size of the Zn4(bim)4 unit, a four-membered ring, is ~0.21 nm as estimated from crystallographic data (Fig. 1F). The effective pore size should be slightly larger, considering the structural flexibility of the sheet. Thus, we can expect that membranes based on Zn2(bim)4 MSNs will achieve a high selectivity of H2 (0.29 nm) over CO2 (0.33 nm) through molecular sieving. Moreover, unlike zeolite nanosheets, which consist of cages (or half cages) or zigzag channels (10), the pores of Zn2(bim)4 MSNs are constructed with four flat bim molecules, ensuring the rapid transport of H2 molecules.

A colloidal dispersion containing ~15 mg/liter Zn2(bim)4 MSNs was obtained by removing the larger unexfoliated particles after being left to stand for 2 weeks. The colloid remained stable at room temperature for several months. Zn2(bim)4 MSNs with wrinkles were frequently observed under transmission electron microscopy (TEM), indicating the flexibility of the nanosheets (fig. S4), which is beneficial for achieving a conformal layer on macroporous supports with rough surfaces (10). An isolated Zn2(bim)4 MSN with side length of ~600 nm is shown in Fig. 2A, where folds and curled edges are observed. The nanosheet is very thin, as determined from the contrast in the TEM image. As is common for MOF materials characterized using high-resolution TEM (HRTEM), the Zn2(bim)4 nanosheets visibly and rapidly degraded during imaging, as a result of electron beam irradiation. Thus, it is difficult to visualize the lattice fringes of the (hk0) planes of the single-layered nanosheet by HRTEM. The inset in Fig. 2A shows the cross-sectional HRTEM image of a five-layered Zn2(bim)4 nanosheet. The distance between the adjacent layers is ~1.18 nm, which is close to the d002 of the pristine crystals. The selected-area electron diffraction (SAED) pattern of a few-layered nanosheet collected along the c axis is shown in Fig. 2B, which corresponds to the diffraction pattern of the (hk0) planes of the nanosheet. The simulated SAED pattern is in good agreement with the experimental result, confirming that the exfoliated nanosheets are highly crystalline and retain the same in-plane crystal structure of pristine bulk materials. In Fig. 2C, a tapping-mode atomic force microscopy (AFM) image shows a Zn2(bim)4 MSN with large lateral dimension up to 1.5 μm. The height profile reveals that the nanosheet has a fairly flat, smooth terrace with a uniform thickness of 1.12 nm. Noting that 2D materials are often raised by a few angstroms above the supporting surface (23), the nanosheets in Fig. 2C should be monolayers of Zn2(bim)4. Additionally, it should be noted that Zn2(bim)4 MSNs with lateral size up to several microns were occasionally observed by TEM (fig. S4) and AFM (fig. S5). The exfoliation of layered Zn2(bim)4 precursors into nanosheets leads to a significant increase in the Brunauer-Emmett-Teller (BET) surface area, from 19.9 to 112.4 m2/g (Fig. 2D). According to the classification by IUPAC (24), the isotherm of Zn2(bim)4 nanosheets can be identified as type II with an identifiable H4-type hysteresis loop, which is associated with the slit-like pores formed by aggregation of nanosheets during freeze drying. The Fourier transform infrared (FTIR) spectrum of Zn2(bim)4 nanosheets is identical to that of pristine Zn2(bim)4, indicating that the soft-physical exfoliation process has little effect on the in-plane structure of the Zn2(bim)4 nanosheets (fig. S6). Thermal analysis revealed that the Zn2(bim)4 nanosheets remain stable up to 200°C despite their nanometer-scale thickness (fig. S7).

Fig. 2 Characterizations of molecular sieve nanosheets.

(A) Low-magnification TEM image of a piece of Zn2(bim)4 nanosheets. The inset image shows a high-resolution TEM image of a five-layered Zn2(bim)4 nanosheet. (B) SAED pattern (white circle) shows the diffraction from (hk0) planes within a few-layered nanosheet. A simulated SAED pattern of Zn2(bim)4 nanosheet down the c axis is also shown. (C) Tapping-mode AFM topographical image of Zn2(bim)4 nanosheets on silicon wafer. The height profile of the nanosheets along the black lines was marked in the image. (D) N2 adsorption-desorption isotherms (77 K) on pristine Zn2(bim)4 and Zn2(bim)4 nanosheets. The inset presents photographs comparing the pristine Zn2(bim)4 and Zn2(bim)4 nanosheets obtained after exfoliation.

Filtration of nanosheet suspensions through porous supports has been successfully applied to fabricate separating membranes in previous studies (10, 12). This method, however, could not be applied in the current case. We attribute this to the fast restacking of nanosheets back to ordered pristine structures at elevated concentration when the solvent is filtered out (fig. S8). Ordered restacking of the MSNs will result in partial or total blockage of the molecular sieve pores. A hot-drop coating process addressed this problem, with the aim of achieving a disordered stacking of the nanosheets in the membrane layer. In a typical preparation, 3 ml of an MSN suspension was diluted five times with methanol and then deposited dropwise onto the surface of an α-Al2O3 disk (Inocermic GmbH, pore size 70 nm, Fig. 3A), which was heated at 120°C on a heating plate (19) (fig. S9). A uniform surface coverage of Zn2(bim)4 MSNs on α-Al2O3 support was achieved (fig. S10). Figure 3B presents a top-view SEM image of a Zn2(bim)4 MSN membrane at high magnification. Figure 3C shows the cross-sectional view of a membrane, where an unsupported portion of the nanosheet layer folds around the corner of the fractured support, clearly illustrating the high flexibility of Zn2(bim)4 nanosheets. In Fig. 3, B and C, the texture of the underlying α-Al2O3 support is distinguishable, indicating a very thin layer of Zn2(bim)4 MSNs on the support. In addition, a substantial aluminum signal from the underlying α-Al2O3 support was detected by x-ray photoelectron spectroscopy (XPS) for a Zn2(bim)4 MSN membrane, indicating that the MSN layer is only several nanometers thick (fig. S11).

Fig. 3 Morphology and performance of membranes derived from molecular sieve nanosheets.

(A) SEM image of a bare porous α-Al2O3 support. (B) SEM top view and (C) cross-sectional view of a Zn2(bim)4 nanosheet layer on α-Al2O3 support. (D) Scatterplot of H2/CO2 selectivities measured from 15 membranes. The red line with symbols shows the average selectivity and dispersion of selectivity of the membranes prepared at different coating temperatures. (E) Anomalous relationship between selectivity and permeance measured from 15 membranes. (F) Powder XRD patterns of four membranes with different separation properties. The cartoons at the left schematically illustrate the microstructural features of the nanosheet layers. The yellow and green portions correspond to the low-angle humps and the (002) peaks in the XRD patterns, respectively. All the membranes were measured for equimolar mixtures at room temperature and 1 atm.

The MSN membranes were sealed into Wicke-Kallenbach cells for measuring the permeance and selectivity of both single and binary feed of H2 and CO2. There was no absolute pressure differential across the membrane to avoid deforming the thin MSN layers (19) (fig. S12). We first conducted a control experiment using the α-Al2O3 support. The support showed an H2 permeance of >50,000 GPUs with a low H2/CO2 selectivity (~1.2). Figure 3D shows the H2/CO2 selectivities measured on 15 membranes prepared via hot-drop coating at different support surface temperatures (80° to 200°C; table S1). In our investigation, 120°C optimizes the average selectivity and reproducibility. We hypothesize that this temperature is just high enough to maintain a solvent evaporation rate that avoids ordered restacking of the nanosheets during the coating process, which occurs at lower temperature (fig. S13). At higher temperature, pinholes or gaps might form between nanosheets when the solvent evaporates too quickly. The three membranes prepared at 120°C exhibited H2/CO2 selectivity of 261 ± 39. Nine membranes were prepared at 120°C with different drop volumes (1 to 15 ml), which proved to be a less significant factor than the support surface temperature (fig. S14 and table S2).

A membrane prepared at 120°C was tested for single gas permeation of H2 (0.29 nm), CO2 (0.33 nm), N2 (0.36 nm), CH4 (0.38 nm), and C2H6 (0.44 nm). There was a clear cutoff between H2 and CO2 (fig. S15). H2/CO2 separation measurements at different temperatures indicate activated diffusion for both H2 and CO2 (fig. S16). These results suggest a size exclusion mechanism for separation of H2 from CO2. H2/CO2 separation with different feed compositions indicates that CO2 adsorption has influence on the H2 permeation to some extent (fig. S17).

Normally, researchers prepare thicker membranes to minimize nonselective defects. However, there is an inverse relationship between membrane permeance and selectivity. In contrast, we found an anomalous proportional relationship between permeance and selectivity for the Zn2(bim)4 MSN membranes, as shown in Fig. 3E. The H2 permeance varies from 760 to 3760 GPUs along with the variation of H2/CO2 selectivity from 53 to 291. The linear regression of these data accounts for 90.6% of the variance (table S1). Meanwhile, a small fluctuation in CO2 permeance (12.3 ± 4.5 GPUs) was measured on all 15 membranes (table S1). Theoretically, only H2 molecules can pass through the four-membered rings of the MSNs. The nonzero CO2 permeance could be attributed to the non–size-selective mass transport through imperfect sealing or through boundaries of the nanosheet layer.

To gain insight into the structure-performance relationship of the MSN membranes, four membranes with different performance were characterized by powder XRD (Fig. 3F). The proportional relationship between permeance and selectivity can be observed in the table in Fig. 3F. Membranes with higher permeance are more selective, and vice versa. Membrane performance can be correlated with the order of nanosheet stacking, as indicated by XRD patterns. The low-angle hump suggests regions with expanded stacking of nanosheets. The appearance of (002) reflection coincides with the formation of bulk Zn2(bim)4 via ordered restacking. Lamellar ordering of nanosheets would block the permeation pathway for H2, but have only a slight effect on CO2 leakage, resulting in a reduction in membrane performance. Noting that a tiny hump is still identifiable for the high-performance membrane, we speculate that an ultrapermeable and superselective molecular sieve membrane could be obtained if the membrane consisted of fully disordered MSNs.

The performance of our membranes exceeds the latest Robeson’s upper-bound for H2/CO2 gas pair and is higher than that of the molecular sieve membranes reported to date, including polycrystalline membranes composed of three dimensional MOFs (fig. S18). For practical use, thermal and hydrothermal stability is an important issue for H2 selective membranes (25). A Zn2(bim)4 MSN membrane was tested under different conditions for more than 400 hours in total, including two temperature cycles, showing no degradation in membrane performances. When exposed to an equimolar H2/CO2 feed containing ~4 mol % steam at 150°C, the membrane showed good stability after a 120-hour test (fig. S19).

Supplementary Materials

www.sciencemag.org/content/346/6215/1356/suppl/DC1

Materials and Methods

Figs. S1 to S19

Tables S1 and S2

References (2640)

References and Notes

  1. See supplementary materials on Science Online.
  2. Acknowledgments: We thank the National Natural Science Foundation of China and the Key Research Program of the Chinese Academy of Sciences for funding, L. Liu for experimental assistance in wet ball milling, S. Miao for experimental assistance in TEM and SAED, and H. Duan for experimental assistance in TG analysis. Y.P., Y.L., and W.Y. are inventors on a Chinese patent filed through the Dalian Institute of Chemical Physics (CN102974229A, “Exfoliation and application of two-dimensional layered metal-organic frameworks”).
View Abstract

Stay Connected to Science

Navigate This Article