Molecular-Scale Hybrid Membranes Derived from Metal-Organic Polyhedra for Gas Separation.

The preparation and the performance of mixed matrix membranes based on metal-organic polyhedra (MOPs) are reported. MOP fillers can be dispersed as discrete molecular units (average 9 nm in diameter) when low filler cargos are used. In spite of the low doping amount (1.6 wt %), a large performance enhancement in permeability, aging resistance, and selectivity can be achieved. We rationalize this effect on the basis of the large surface to volume ratio of the filler, which leads to excellent dispersion at low concentrations and thus alters polymer packing. Although membranes based only on the polymer component age quickly with time, the performance of the resulting MOP-containing membranes meets the commercial target for postcombustion CO2 capture for more than 100 days.

Introduction

The use of composites in membrane separation is attracting a great deal of attention.[1−3] By using two components with different transport properties in one single membrane, such composites have the potential to offer the easy processability of polymers and the superior performance of the filler.[4] Various additives, e.g., silica,[3] zeolites,[5] metal-organic frameworks (MOFs),[6−9] and porous organic frameworks,[10−12] have been employed in combination with a large variety of polymers. MOFs are promising fillers to construct composite membranes in light of their rich chemistry (expected compatibility) and transport properties.[2,13] However, the final performance of the membrane is determined by several additional factors of the fillers, e.g., surface functional groups,[14,15] aspect ratio,[16] and particle size.[7,17−19] The particle size of the fillers is one of these critical properties. For instance, separation performance of mixed matrix membranes (MMMs) based on MOF nanoparticles is usually superior to that of membranes prepared with bigger particles of the same filler.[7,17−19] This is because smaller particles inherently expose larger external surface areas to interact with the polymer, improving in this way compatibility.[7,17−19]

Metal-organic polyhedra (MOPs)[20−23] are considered as discrete porous cage-like MOF analogues.[24] The particle size of individual MOP cages is normally in the range of 2.5–5.0 nm. Moreover, most MOPs are soluble (or dispersible) in a wide variety of solvents. These properties provide important advantages to construct hybrid membranes. In the past few years, a few MOP composite membranes have been reported,[25−28] along with analogous membranes with porous organic cages (POCs) as fillers.[29,30] Surprisingly, in these cases, a large cargo of filler was used (normally more than 10 wt % of MOP or POC), leading to possible agglomeration and hiding the effect of fillers.

Here, we take advantage of the small size of individual MOP units and demonstrate that the best membrane performance is achieved at a very low MOP loading (1.6 wt %). Under these conditions, well-dispersed MOPs are incorporated in the membrane rather than agglomerates, resulting in large separation performance enhancements, including improvements in permeability, selectivity, and aging resistance while maintaining the unmatched processability of the polymer phase.

Experimental Section

Preparation of MOP-15

MOP-15 was synthesized according to the recipe reported by Yaghi et al.[31] Fresh glycine tert-butyl ester hydrochloride (0.242 g) (Sigma-Aldrich, ≥99.0%) was dissolved in 6.0 mL of N,N-dimethylformamide (DMF, Acros, 99.8%, Extra Dry). Then, 0.22 mL of triethylamine (Sigma-Aldrich, ≥99%) was added to the solution, and white precipitate was formed and removed by filtration. The filtrate was mixed with 0.144 g of copper acetate monohydrate (Cu(OAc)2·H2O, Sigma-Aldrich, 99.99%) in DMF (6.0 mL) and labeled as solution I. 5-Aminoisophthalic acid (0.0018 g) (H25-NH2-mBDC, Merck, ≥98.0%) was dissolved in a DMF (2.6 mL)/EtOH (0.4 mL, Acros, 99.5%, Extra Dry) solution and labeled as solution II. Solution II was mixed with 3.0 mL of solution I in a capped vial and maintained at ambient condition for 5 days. Green truncated-octahedral crystals were harvested. After rinsing with DMF (3 × 10 mL), the crystals were stored in DMF for later use.

Preparation of Membranes

MOP-15 (0.016 g) was dissolved in 10 mL of dimethyl sulfoxide (DMSO, Sigma-Aldrich, ≥99.0%), and transparent green solution was formed. 4,4′-(Hexafluoroisopropylidene)diphthalic anhydride–diaminomesitylene (6FDA–DAM) (0.10 g) (Mw ∼ 272 000 Da, Akron) dissolved in 9.0 mL of DMSO was mixed with 1.0 mL of MOP-15 solution via stirring overnight. The homogeneous light green solution was transferred to a glass Petri-dish and dried at 373 K for 12 h (in a solvent saturated atmosphere) to slowly evaporate the solvent. Finally, the free-standing film was peeled off and dried at 353 K (24 h) and 423 K (20 h) under vacuum. Neat 6FDA–DAM and MOP-15/6FDA–DAM membranes with various filler loadings were prepared by the identical approach via modulating the volume ratio of the served MOP-15 and 6FDA–DAM solutions. Polymer 6FDA–DAM was degassed overnight at 423 K under vacuum before use. The thickness of all of the membranes is around 20–30 μm, according to the values measured with a digital micrometer (Mitutoyo) at different locations within each membrane and then averaged.

Characterizations

Microscope image was captured from a Microscope BRESSER under ambient condition. The sample was prepared by dropping MOP-15 DMF solution on a glass slide without drying. A Bruker-D8 Advance diffractometer (using Co Kα radiation, λ = 0.179 nm at 35 kV and 40 mA) was employed to analyze the crystalline structure of powders and membranes. UV–vis spectra of the MOP-15/DMSO solution were collected on the UNICAM UV 500 spectrometer in the wavelength range of 190–900 nm. CO2 (298 K) and N2 (77 k) adsorption isotherms of the samples were performed in a Tristar II 3020 (Micromeritics) setup. Prior to the measurements, the samples were degassed at 353 K under vacuum for 16 h.

Atomic force microscopy (AFM) micrographs were collected in a noncontact tapping mode using a Solver NEXT AFM instrument from NT-MDT. Gold-coated cantilevers (NSG 03, from NT-MDT) with spring constants ranging from 0.4 to 2.7 N m–1 (resonant frequency of 90 kHz) were used and calibrated by the thermal noise method. The tip diameter of probe is around 20 nm. Nova Px 3.2.5 software was used for all of the data acquisition and analysis. To capture the images, a sample of MOP-15 was prepared by spin coating a dilute MOP-15/DMSO solution (10 μg mL–1) on a silicon wafer and dried under vacuum at room temperature. The silicon wafer was pretreated with acetone and oxygen plasma (at a pressure of 2.1 mbar for 1 min, using a Harrick plasma cleaner, from Anadis Instruments) for cleaning.

Scanning electron microscopy (SEM) images of the membranes were acquired using a JEOL 6010 microscope. The specimen was prepared by cryo-fracturing in liquid N2 and coated with gold. Focused ion beam scanning electron microscopy (FIB-SEM) experiments were performed in an FEI Helios G4 CX microscope. A conductive thin layer of Au (0.1 μm thickness) and a protective thin layer of Pt (0.3 μm thickness) were deposited on the surface of specimen using the sputter coater and the gas injection system, respectively. Slices with a nominal thickness of 2 μm were milled away by the FIB, operating at 30 kV and 80 pA. Two individual SEM micrographs of the membrane cross section exposed on each milling were recorded, with an in-lens secondary electron detector operated at 10 kV. To calculate the average filler size using SEM images, around 70 filler particles were identified and measured by ImageJ software.

Diffuse reflectance infrared Fourier transform (DRIFT) spectra were acquired in a Nicolet 8700 FT-IR (Thermo Scientific) spectrometer. The samples were dried overnight at 423 K under vacuum before recording data. Thermogravimetric analysis (TGA) of MOP-15 was performed on a Mettler Toledo TGA/SDTA851e apparatus under N2 flow (100 mL min–1) from 303 to 1073 K with a ramp speed of 5 K min–1.

Gas Permeation

The CO2/N2 separation performance was evaluated in a home-made setup described elsewhere.[16] The membranes, with a diameter of 1.8 cm and an effective area of 1.3 cm2, were cut from the as-synthesized films and mounted in a flange between two Viton O-rings. A macroporous stainless steel disc (316L, 20 μm nominal pore size) was used as support. The permeation module was placed inside a convection oven, where the temperature was set to 298 K. A flow of CO2 (15 mol %) and N2 (85 mol %) mixture (133 mL min–1, standard temperature and pressure (STP)) was applied as feed and helium (5 mL min–1, STP) as a sweep gas. The feed pressure was adjusted in the range of 1–4 bar absolute using a back-pressure controller at the retentate side, whereas the permeate side was kept at atmospheric pressure (1 bar absolute) for all measurements. The permeation results of the membranes were recorded after steady state was confirmed using consecutive online gas chromatography (GC) analyses (Interscience Compact GC). Gas separation performance is defined by the gas permeability (P) of the individual components and selectivity (α). The permeability for the component i (P) was calculated as follows (eq )where F denotes the molar flow rate of compound i, l is the thickness of the membrane, Δp is the partial pressure difference of i across the membrane, and A is the membrane area. The unit of P adopts Barrer, where 1 Barrer = 3.35 × 10–16mol m m–2 s–1 Pa–1.

The mixed gas selectivity (α) of CO2 over N2 is defined as the ratio of their permeabilities (eq )The Maxwell model is quite useful for understanding the effective permeability and structure of mixed matrix membranes.[2,3,5] The model was initially developed by Maxwell to estimate the dielectric properties of heterogeneous media.[32] Membrane scientists used this model to predict molecular permeation of mixed matrix membranes (eq )[33] under a pressure-driving force because it is analogous to the conduction of a dielectric in heterogeneous media under an electric potentialwhere Peff is the effective permeability of the mixed matrix membrane, ⌀d is the volume fraction of filler (dispersed phase), Pc and Pd represent the permeability of the continuous phase (polymer) and dispersed phase (MOPs here), respectively. In this study, to maximize the predicted Peff, an ultrahigh permeability of Pd (relative to Pc) is adopted by assuming the transport of the penetrant in MOPs is ultrafast. So, the above equation is transformed to the following one to predict the permeability of hybrid membranes⌀d is calculated based on the corresponding mass loading of fillers and material density. The required particle loading (⌀d′, vol %), which could provide equal external surface area, with varied filler size (r′, diameter, nm), was calculated employing eq The morphology of the filler was assumed as sphere. The optimal MOP particle loading (⌀d0 = 1.8 vol %) and the related average MOP diameter (r0 = 9 nm) were adopted.

Results and Discussion

Metal-Organic Polyhedra

MOP-15[31] with a formula Cu24(NH2-mBDC) (5-amino-1,3-benzenedicarboxylate)24 (solvents omitted) was selected as filler. The amino groups on its framework are expected to engage in hydrogen bonding with the polyimide matrices according to our previous research on MOF mixed matrix membranes.[7,8] This discrete molecular cage is constructed from 12 copper paddle–wheel clusters bridged by 24 NH2-mBDC linkers (Figure S1).[31] The cage is porous with an average size of ca. 2.6 nm, a cavity diameter of ca. 1.5 nm, and aperture diameters of around 0.6 nm (triangular windows) and 0.9 nm (square windows), as estimated from crystallographic data.[31]

On the basis of the synthesis reported by Yaghi et al.,[31] green truncated-octahedral crystals (Figure a) of MOP-15 were successfully prepared as confirmed by powder X-ray diffraction (XRD) analysis (Figure b). The crystals are soluble in DMSO, forming a transparent green solution with an intense absorbance around 720 nm, assigned to the copper paddle–wheel units (Figure c).[23]

Figure 1

(a) Microscope image of the as-synthesized MOP-15 crystals. The inset is a photo of the MOP-15 crystals precipitated in DMF solution; (b) XRD pattern of the as-synthesized MOP-15. The simulated XRD is shown for reference. (c) Absorption spectra of MOP-15 in DMSO solution. The insets are a photo of MOP-15 dissolved in DMSO solution and a unit cage structure of MOP-15; AFM height images ((d) low magnification and (e) high magnification) and corresponding height profiles (f) of MOP-15 particles on top of a silicon wafer along the white lines marked in image (e).

After solvent evaporation, MOP-15 crystallizes into aggregates (Figure d,e). Individual MOP-15 cages with an average height of 2.7 nm were identified based on the corresponding AFM height profiles (Figure f), consistent with the value (2.6 nm, Figure c) calculated from crystallographic data. The particles visualized with an average height of 5.1 nm (Figure e,f) probably result from the stacking of two MOP-15 cages.

The thermogravimetric analysis (Figure a) indicates that MOP-15 has a good thermal stability (up to 515 K). The as-synthesized aggregated MOP-15 particles possess a low surface area (Brunauer–Emmett–Teller 17.5 m2 g–1, Figure b) with a CO2 uptake of 0.7 mmol g–1 at 1.2 bar and 298 K (Figure c).

Figure 2

a) TG profile of MOP-15. (b) N2 (77 K) and (c) CO2 (298 K) adsorption (solid symbols) and desorption (open symbols) isotherms. (d) DRIFT spectra of MOP-15, 6FDA–DAM, and MOP-15 (7.4 wt %)/6FDA–DAM.

Formation and Characterization of Membranes

By dissolving both MOP-15 and 6FDA–DAM polymers in DMSO, stable, transparent (light green) solutions were formed. To fabricate homogeneous hybrid membranes, preliminary experiments were carried out to determine the temperature and drying rate for solvent evaporation. A temperature of 373 K and the controlled drying rate (for 12 h in a solvent saturated atmosphere) were finally selected as optimized conditions (see Experimental Section for details). Lower temperature and slower drying rate resulted in particle agglomeration. This is in essence because the solubility of MOPs increases with temperature[22] and a fast drying rate could possibly fix MOPs among the polymer chains before aggregation happens. However, attempts to further increase the temperature and drying rate lead to the formation of defects in the continuous polymer phase (Figure S2). These observations highlight the importance of well-controlled conditions during membrane drying. The structural integrity of MOP-15 (NH2 stretching vibration (3300 cm–1) from aromatic amines and C=O stretching vibration (1600 cm–1) from carboxylate moieties) and 6FDA–DAM in the hybrid membrane was verified by DRIFT analysis (Figure d).

The morphology of neat and hybrid 6FDA–DAM membranes was characterized by SEM (Figure ). At a low particle loading (1.6 wt %), no visible MOP-15 fillers were observed (Figure c,d), whereas pronounced cage agglomeration appeared at higher concentration of MOP-15 (7.4 wt %) (Figure e,f). Consequently, microcracks between the filler and polymer phases were generated. To further investigate MOP distribution in the hybrid membrane at low particle loading, focused ion beam scanning electron microscopy (FIB-SEM) (Figure a–e2) with relatively high magnification was employed. Hereafter, unless otherwise stated, the default particle loading of MOP-15/6FDA–DAM hybrid membrane is 1.6 wt % (i.e., 1.8 vol %). Very homogeneous membranes with particles embedded in the polymer matrix were found (Figure b1–e2). As anticipated, no detectable gaps between the filler and the matrix could be observed, illustrating an intimate adhesion between both phases. Individual MOP cages with a size around 2–3 nm can be recognized (highlighted in white circles) along with bigger particles resulting from agglomeration. On the basis of particle size distribution analysis (Figure f), the average filler size is around 9 nm.

Figure 3

SEM surface (a, c, e) and cross-sectional (b, d, f) images of neat 6FDA–DAM (a, b), MOP-15 (1.6 wt %)/6FDA–DAM (c, d), and MOP-15 (7.4 wt %)/6FDA–DAM (e, f) membranes.

Figure 4

Cross-sectional FIB-SEM images of MOP-15/6FDA–DAM membrane (a–e). Two individual images (numbered as 1 and 2) exposed on each milling (marked in (a)) were recorded. Fillers with a diameter around 2–3 nm are highlighted with dashed circles; (f) particle size distribution of MOP-15 fillers in the hybrid membrane. Around 70 filler particles were identified from (b1) to (e2), and the diameters were measured by ImageJ software. The filler loading is 1.6 wt %.

The microstructure of the membranes was studied by XRD. As depicted in Figure a, the neat 6FDA–DAM membrane exhibits one broad diffraction peak centered at 15.9° (corresponding to a d-spacing of 0.64 nm), together with a relatively weak peak at 6.7° (corresponding to a d-spacing of 1.53 nm). In the case of the hybrid membrane, after the cages rearranged, no diffraction peaks from the original MOP-15 crystals (Figure a) were found. In contrast, the reflections from the parent polymer shift to lower angles (15.3 and 6.4°), indicating the stabilization of polymer chains at higher d-spacing values (0.67 and 1.60 nm). We speculate that this is due to disruption of the polymer chains as a result of addition of such small filler particles. Not surprisingly, when agglomeration of the MOP into bigger units occurs (membranes with a 7.4 wt % MOP loading), no change in the diffractogram is observed (Figure b).

Figure 5

XRD patterns of the fresh and aged (100 days) 6FDA–DAM and MOP-15/6FDA–DAM membranes: (a) fresh 6FDA–DAM and MOP-15 (1.6 wt %)/6FDA–DAM, (b) fresh 6FDA–DAM and MOP-15 (7.4 wt %)/6FDA–DAM, (c) fresh and aged 6FDA–DAM, and (d) fresh and aged MOP-15 (1.6 wt %)/6FDA–DAM.

Gas Separation Performance

After implanting MOP fillers, the CO2 adsorption uptake of the membrane was well maintained (Figure S3). The lower CO2 uptake of MOP-15 is responsible for the slight decline of CO2 adsorption on the hybrid membrane. The membrane performance was evaluated by separating CO2 from N2 at 298 K (Figure ) under conditions relevant to postcombustion CO2 capture (15 mol % CO2 and 85 mol % N2). The neat 6FDA–DAM membranes prepared following exactly the same drying conditions as in the case of the MMMs exhibit a CO2 permeability of 1010 Barrer with a CO2/N2 selectivity of 23.9 under 1 bar absolute feed pressure (Figure a). This permeability was slightly higher than that in our previous publications[34,35] and dropped to a comparable value after aging (vide infra). By doping 1.6 wt % MOP-15, an increase in CO2 permeability (40%, to 1413 Barrer) along with a modest improvement in selectivity (26.7) was observed. With a further increase of particle loading to 3.2 wt %, no improvement in CO2 permeability or selectivity occurred. We attribute this effect to the agglomeration of MOP-15. Indeed, further increase in filler concentration (7.4 wt %) leads to a dramatic increase in CO2 permeability at the expense of selectivity as a consequence of further filler agglomeration and the formation of defects (Figure e,f).

Figure 6

Effect of particle loading ((a, b) 1 bar absolute feed pressure) and feed pressure ((d) 0 and 1.6 wt % filler loading; (e) 0.8, 3.2, and 7.4 wt % filler loading) on the membrane performance: The relative permeability based on the Maxwell model is shown for reference in (b). Open, half open, and solid symbols represent CO2 permeability, selectivity, and N2 permeability, respectively. Circles, triangles, lozenges, pentagons, and hexagons denote the performance of membranes with particle loadings of 0, 0.8, 1.6, 3.2, and 7.4 wt %, respectively. The volume loading in (b) is calculated based on the corresponding mass loading and material density. Error bars in (d) correspond to standard deviation of duplicate membranes. (c) A plot of particle loading required to provide identical external area vs filler diameter based on eq . (f) CO2/N2 separation performance of the fresh (solid symbols) and aged (open symbols) membranes at 1 bar absolute feed pressure. The Robeson upper bound (2008)[36] is shown for reference, as well as the commercial target region for CO2 capture from flue gas from Merkel et al.[37] assuming a membrane thickness of 1 μm. 1 Barrer = 3.35 × 10–16 mol m m–2 s–1 Pa–1. The feed (15 mol % CO2 and 85 mol % N2) temperature was kept constant at 298 K for all of the permeation measurements.

To get further insight into the structure/performance correlation, the CO2 permeability of hybrid membranes was predicted by the Maxwell model (eq ).[1−3] The model is strictly applicable to an ideal combination of filler and matrix phases, i.e., diluted suspensions of spherical fillers at low loadings (≤20 vol %) without interfacial defects, aggregations, pore blockages, and rigidification of surrounding polymer chains.[1−3] As shown in Figure b, the experimental CO2 permeability of the typical hybrid membrane (1.8 vol % or 1.6 wt %) is much higher than the predicted one even when an ultrafast permeability of MOP-15 fillers is adopted in the model (eq ). These results indicate that the behavior of the composite membranes is far from ideal blending and suggest that incorporation of the filler changes dramatically the performance (and most likely configuration), as already anticipated from XRD (vide supra). The ultrafine MOP-15 filler (average 9 nm in diameter) could provide a much higher surface to volume ratio than that of regular MOFs. To achieve an equal external surface area generated from MOP-15 at a filler loading of 1.8 vol %, 36 vol % is required for typical MOFs with a diameter of 180 nm (Figure c). The high external area of MOP-15 fillers could provide more chance to interact with polymer chains. We speculate that the larger d-spacings observed upon doping 6FDA–DAM with small amounts of MOP are largely responsible for this increase in permeability although the MOPs could provide additional pathways, whereas the little increase in selectivity is mostly due to contribution of the filler. Moreover, stabilization of polymer chains should also affect the membrane resistance against aging, as shown below.

Generally, the performance of polymeric membranes is limited by the trade-off defined by the Robeson upper bound.[36,38] Improvements in permeability are always at the expense of selectivity and vice versa. In the present case, both permeability and selectivity are improved upon 6FDA–DAM doping with MOP-15, bringing membrane performance into the commercial target[37] for postcombustion CO2 capture (Figure f). At varying feed pressure, the hybrid membranes still outperform the ones based on the pure polymer (Figure d), with CO2 permeability decreasing gradually with increasing feed pressure (Figure d,e) at an almost constant selectivity. The effect of aging was evaluated after keeping the membranes in a desiccator for 100 and 220 days. The results are given in Figures f, S4, and S5. It is well known that during physical aging, the polymer chains of 6FDA–DAM tend to pack more efficiently,[39] leading to a reduced porosity. The large d-spacing corresponding to the diffraction at low angle in the XRD patterns was absent after aging (Figure c,d). Consequently, a drop in CO2 permeability and a slight increase in selectivity are understandable for the neat and hybrid membranes. Nevertheless, the achieved permeability (934 Barrer) of the aged hybrid membranes (220 days) is still 54% higher, probably explained by the fact that the d-spacing corresponding to the diffraction at high angle was well retained (Figure c,d). The performance of hybrid membrane still meets the commercial target for postcombustion CO2 capture after 100 days of aging.

Conclusions

In comparison with MOF-based mixed matrix membranes (Figure S6), which normally reach their optimal performance at a filler loading higher than 10 wt %, the use of MOPs as fillers allows decreasing cargo by 1 order of magnitude while still delivering positive effects on membrane performance in terms of permeability, selectivity, and stability. This effect has been attributed to the molecular size of the MOP-15 filler (9 nm in diameter) that provides a much higher external surface to volume ratio than other fillers. It can be efficiently encapsulated in between polymer chains, modifying in this way polymer packing and increasing permeability and aging resistance. Considering the rapid developments in MOP chemistry, we believe that these molecular materials will offer unrivalled possibilities for the development of mixed matrix membranes.