High performance mixed matrix membranes (MMMs) composed of ZIF-94 filler and 6FDA-DAM polymer

Carbon capture and storage (CCS) using membranes for the separation of CO 2 holds great promise for the reduction of atmospheric CO 2 emissions from fuel combustion and industrial processes. Among the different process outlines, post-combustion CO 2 capture could be easily implemented in existing power plants. However, for this technology to become viable, new membrane materials have to be developed. In this article we present the development of high performance mixed matrix membranes (MMMs) composed of ZIF-94 filler and 6FDA-DAM polymer matrix. The CO 2 /N 2 separation performance was evaluated by mixed gas tests (15CO 2 :85N 2 ) at 25ºC and 1 to 4 bar transmembrane pressure difference. The CO 2 membrane permeability was increased by the addition of the ZIF-94


Introduction
Carbon dioxide concentration in the atmosphere has been increasing significantly over the past century. Fuel combustion for electricity and heat generation represented by far the largest source in 2014, more than 40% of global CO 2 emissions [1]. These overwhelming contribution suggests that, in addition to the development of energy generation processes that rely on renewable resources, carbon capture and storage (CCS) should be implemented in currently running energy generation plants [2,3]. Three major ways have been considered to reduce CO 2 emissions in combustion processes: pre-combustion CO 2 capture (after coal gasification), post-combustion CO 2 capture from power plant flue gas, and oxyfuel combustion [4].
Since the serial production of commercial polymeric membranes was implemented in 1980 by Henis and Tripodi, membrane gas separation has rapidly become a competitive separation technology. Membrane gas separation offers several benefits over conventional gas separation technologies [5]: lower energy cost, a relatively small footprint, low mechanical complexity and operation under continuous, steady-state conditions.
To date only polymeric membranes have been implemented for gas separation on a large scale in industry, mainly due to their easy processing and mechanical strength [6]. However their performance is limited by the trade-off relationship between permeability and selectivity, represented by the 'Robeson upper bound' [7,8]. Low chemical and thermal stability and plasticization at high pressures in the presence of strong adsorbing penetrants such as CO 2 are among the main disadvantages of this type of membranes. On the other hand, although inorganic membranes based on ceramics [9], carbon [10], zeolite [11], oxides [12], metal organic frameworks (MOF) [13] or metals [14] present an excellent thermal and chemical stability, good erosion resistance and high gas flux and selectivity for 5 gas separation, their implementation at industrial scale has been hampered due to the low mechanical resistance, modest reproducibility, scale-up problems and the high fabrication cost of this type of membranes [13,15].
Mixed matrix membranes (MMMs) were presented as an alternative to overcome limitation of both polymeric and inorganic membranes. In a MMM, filler particles are dispersed in a polymer matrix that should improve the properties of the composite relative to the pure polymer [6,16]. Recently metal organic frameworks (MOFs) have been identified as promising filler materials for the preparation of MMMs [17]. They have high surface area and pore volume and their porosity is in general higher than that of their earlier considered inorganic counterpart, zeolites. Moreover, in contrast with zeolites, due to their partially organic nature, MOFs usually display better polymer-filler compatibility. This prevents formation of non-selective voids at the polymer-filler interface and consequently defect free membranes can be made [18]. One of the first reports of a MOF used in a MMM concerned the additions of copper biphenyl dicarboxylate-triethylene diamine to poly (3-acetoxyethylthiophene (PAET) [19]. Since then quite a few MOF/polymer pairs have been studied in literature [20,21]. Zeolitic imidazolate frameworks (ZIFs) are a subclass of MOFs whit a similar structure of zeolites. Several ZIF/polymer pairs have been studied in literature as MMMs for CO 2 /N 2 separation. ZIF-8 was used to improve the permeability of 6FDA-DAM:DABA(4:1) films by Lively et al. [22]. At 20 wt% loading, the membrane permeability increased by 2.5 times over the neat polymer membrane, with only a modest 9.4% loss in CO 2 /N 2 selectivity. ZIF-8 was also used as filler by Nafisi et al. [23] and Wijenayake et al. [24] for the preparation of 6FDA-durene MMMs. In both cases an increase on CO 2 permeability was observed due to polymer chain interruption and increase in fractional free volume caused by the filler, 1.5 times higher CO 2 permeability for 30 wt% 6 ZIF-8 loaded membrane and 3.3 times higher for 33.3 wt% loaded respectively. However a slight decrease in CO 2 /N 2 selectivity was observed in both cases, attributed to the relatively higher increase in permeability for N 2 . ZIF-71 nanoparticles were incorporated to the same polymer by Japip et al. [25]. With a 20 wt% ZIF-71 addition, the pure CO 2 permeability of the MMM was increased by 3-fold, while the ideal CO 2 /N 2 selectivity was reduced from 14.7 to 12.9.
Different ZIF fillers have been added to Pebax polymer matrix. ZIF-8 filler and Pebax 2533 polymer matrix was used by Nafisi et al. [26] to prepare self-supported dual layer mixed matrix membranes. CO 2 permeability was increased by 3.6 times by the addition of 35 wt% ZIF-8, while a slight decrease on CO 2 /N 2 selectivity was observed. In other study, an asymmetric membrane was prepared by Li et al. [27] by depositing a thin mixed matrix layer of < 1mm of Pebax 1657 and ZIF-7 on a porous polyacrylonitrile support. An intermediate gutter layer of PTMSP was applied to serve as a flat and smooth surface for coating to avoid polymer penetration into the porous support. CO 2 permeability was increased by 1.5 times and CO 2 /N 2 selectivity was tripled by the addition of 22 wt% ZIF-7 filler. The enhanced performance was attributed to the combination of molecular sieving effect from ZIF-7 filler and the high solubility of CO 2 in Pebax.
In the present work ZIF-94 particles have been prepared and incorporated into 6FDA-DAM to form MMMs with the aim of achieving membrane properties similar to those recommended by Merkel et al. for the post-combustion CO 2 capture (the focus of this paper). 6FDA-based polyimides possess impressive gas separation performance, pairing high permeability with good permselectivity. Their rigid primary structure contains bulky CF 3 groups through which the efficient packing of polymeric chains is inhibited and local segment mobility is reduced [28]. Many other desirable properties such as spinnability, thermal and chemical stability and mechanical strength as compared with nonfluoropolyimides make this polymer family suitable for gas separation applications [29][30][31][32][33]. In our case, a commercially available high flux 6FDA-DAM polyimide was selected for membrane preparation. The preparation of 6FDA-DAM MMMs by the addition of several fillers such us NH 2 -MIL-53(Al) [34], ZIF-11 [35], CPO-27(Mg) [36], ZIF-90 [37] and ZIF-8 [38] has been reported in literature. Membrane properties for gas separation are shown and compared with our MMM in the results and discussion section of this paper (Table 1). The selection of MOF filler was first based on CO 2 adsorption capacity and selectivity over N 2 . ZIF-94 (also known as SIM-1, Substituted Imidazolate Material-1) has the sod topology and it is constructed by Zn atoms and 4-methyl-5 imidazolecarboxaldehyde (aImeIm) linkers. It has a high CO 2 uptake of 2.4 mmol g -1 at 1 bar, higher than its topological counterpart ZIF-93 with the rho topology (1.7 mmol g -1 , 17.9 Å pore diameter) or other MOFs such as ZIF-7 (1.6 mmol g -1 , 7.5 Å pore diameter) and ZIF-11 (0.8 mmol g -1 , 14.9 Å pore diameter). The higher CO 2 uptake is attributed to the smaller pore diameter of ZIF-94 (9.1 Å) compare to other ZIFs. As it was reported, small pores are advantageous when considering CO 2 adsorption in the low-pressure regime [39]. ZIF-94/SIM-1 has already been used as membrane material in some publications. Marti et al. [40] reported the fabrication of SIM-1 membranes by post synthetic modification of ZIF 8 particles for the separation of water from water/ethanol mixtures. The membrane fabricated using nano SIM-1 crystals separated water completely from the mixture. SIM-1 membrane for CO 2 /N 2 separation has been crystallized in situ on a tubular asymmetric alumina support by Aguado et al. [41]. In a recent study, layered ZIF/polymer hollow fiber membranes for H 2 /CH 4 and CO 2 /CH 4 separation were prepared by Cacho-Bailo et al. growing a continuous ZIF-94 layer on the bore side of a porous P84 polyimide hollow fiber [42].
8 ZIF-94 also meets several highly important requirements for product development: (i) Preparation as nanoparticles for inclusion in thin membranes (< 1 µm as target), (ii) scale up production via green synthesis, using non-or less toxic solvents such as water, THF or DMSO, (iii) low cost of metals and linkers and (iv) stability in water vapor. Prior to up scaling, the synthesis of the ZIF-94 MOF was optimized at the lab scale to yield particles in accordance with membrane fillers requirements.
Here we report the preparation, characterization and performance of unique mixed matrix membranes made of highly engineered materials ZIF-94 and 6FDA-DAM. The membranes have remarkable gas separation properties tested under process conditions relevant for CO 2 capture in post-combustion applications.

Synthesis of ZIF-94 crystals
The synthesis of the ZIF-94 particles was first optimized at the lab scale and then scale up.
Lab scale synthesis of ZIF-94 involved dissolving 0.4392 g of Zn(CH 3 COO) 2 ·2H 2 O (2 mmol) in 20 mL of methanol and 0.4404 g of 4-methyl-5-imidazolecarboxaldehyde (aImeIm, 4 mmol) in 50 mL of THF. For up scaling, 3.52 g of Zn(CH 3 COO) 2 ·2H 2 O (160 mmol) were dissolved in 160 mL of methanol and 3.52 g of 4-methyl-5imidazolecarboxaldehyde (aImeIm, 31 mmol) in 400 mL of THF. After the solids were completely dissolved, Zn(CH 3 COO) 2 ·2H 2 O-methanol solution was poured slowly into the aImeIm-THF solution. The mixture was continuously stirred for 60 mins at room temperature (30 mins for up scaling). The product was collected by centrifugation and washed with methanol three times before drying at room temperature (at 105ºC for up scaling).

ZIF-94 characterization
Scanning electron micrographs were obtained from a JEOL JSM-6700F FE-SEM. Samples were sputter coated three times with gold in a Quorum Q150R ES (10 mA, 30 s and 2.3 tooling factor).
Powder X-ray diffraction (PXRD) data of lab scale sample was collected in Debye-Scherrer (capillary) geometry from STOE STAD i/p diffractometers with primary monochromation (Cu K α1 , λ = 1.54056 Å). Prior to analysis, samples were ground to a fine powder and introduced to a 0.7 mm glass capillary. PXRD pattern of up scaled sample was collected using a Bruker AXS D8 diffractometer using Cu Kα radiation (λ = 1.5406 and 1.54439 Å) over the 2θ range of 3-130° in 0.02° steps. Powder was places on a PTFE sample holder and analyzed in Bragg-Brentano reflection geometry. Le Bail refinement was performed using Topas with reflection profiles modelled using a fundamental parameters approach [43] with reference data collected from NIST660 LaB 6 .

Membrane preparation
6FDA-DAM/ZIF-94 MMMs were prepared at different MOF loadings (10, 20, 30 and 40 wt%). Up-scaled ZIF-94 was used for membrane preparation. For comparison purposes, the pure polymer membrane was also prepared. Membranes were prepared by a casting method. Polymer and MOF were dried in a vacuum oven at 100 ºC overnight before casting solution preparation. A polymeric pre-dope composed of 13 wt% 6FDA-DAM in THF was prepared. ZIF-94 was dispersed in tetrahydrofuran in an ultrasonic bath for 1 h. The polymeric pre-dope was added to the ZIF-94/THF suspension and was stirred overnight at room temperature. The solvent/filler-polymer ratio of the final solution was of 91/9. Solution was cast with a dr. Blade over a glass plate (casting thickness of 80 µm) and solvent evaporated at room temperature for 24 h in a solvent rich environment. Membranes were heat treated in a vacuum oven at 160 ºC overnight to eliminate residual solvent.
Membrane thickness was measured with a digital micrometer (Mitutoyo) at different 11 locations of the membrane. The average thickness value of ten measurements was used for permeability calculations.

Membrane characterization
The surface and cross-section morphology of the dense MMMs were characterized by scanning electron microscopy (SEM) (Quanta 250 ESEM) equipped with energy dispersive X-ray spectroscopy (EDX). The samples for cross-section SEM characterization were prepared by freeze-fracturing in liquid nitrogen. The low voltage high contrast backscatter electron detector (vCD) and the large field detector (LFD) were used for the analysis of the membranes.
Fourier transform infrared spectroscopy (FTIR) of pure components was performed on a Vertex 70 instrument (Bruker). Infrared chemical mapping of the MMMs with nanoscale spatial resolution was performed with a scattering-type scanning near-field optical microscope (IR s-SNOM) [44] (neaSNOM, Neaspec GmbH, Germany). It is based on an atomic force microscope (AFM), where the tip is illuminated with monochromatic infrared radiation of frequency ω. Recording of the tip-scattered infrared field with a pseutoheterodyne interferometer yields infrared amplitude and phase images simultaneously with topography [45]. Strong phase contrast reveals areas of strong molecular vibrational absorption [46,47]. We used standard Pt-coated AFM tips for both topography and infrared imaging, and a frequency-tunable quantum cascade laser (QCL) (MIRcat, Daylight Solutions Inc., USA) for tip illumination.
Permeation experiments were performed for pure gases and CO 2 /N 2 gas mixtures in the gas permeation setup described elsewhere [48]. Circular samples of 3.14 cm 2 were cut and placed in the permeation cell over a macroporous stainless steel SS 316L support with 20µm nominal pore size. Gas was fed at 25 ºC and different pressures (1-4 bar transmembrane pressure difference). Transmembrane pressure was adjusted using a backpressure regulator at the retentate side while permeate side of the membrane was kept at atmospheric pressure. A CO 2 /N 2 gas mixture (15:85) was used as feed gas for mixed gas experiments (20 mL min -1 CO 2 and 113 mL min -1 N 2 ) and helium (3 mL  The permeability for gas i was calculated by the following equation: where P i is the gas permeability in Barrer (1 Barrer = 10 -10 cm 3 (STP) cm cm -2 s -1 cmHg -1 ), F i is the volumetric flow rate of component i (cm 3 (STP)/s), l is the thickness of the membrane (cm), ∆p i is the partial pressure difference of component i across the membrane (cmHg) and A is the effective membrane area (cm 2 ).
The separation factor or mixed gas selectivity α was calculated as the ratio of the permeability of more permeable compound i to the permeability of the less permeable compound j:

MOF characterization
ZIF-94 crystals were synthesized in this work by replacing dimethylformamide (DMF), previously used for synthesis of ZIF-94 [39], with a 2:5 ratio mixture of methanol:THF. A reaction yield of 82 % for lab scale synthesis and 99 % for up scaling, with respect to zinc were achieved. Higher reaction yield might be due to the use of a high-speed centrifuge for up-scale synthesis, not available for lab scale synthesis. The SEM image shown in Figure 1 indicates that spherical particles of ZIF-94 were produced with a diameter of 100-500 nm.
The PXRD pattern of this material ( Figure S1, Supporting information) was consistent with that reported by Aguado et al [49] and with the sodalite topology.
CO 2 adsorption capacities at 25 ºC were 0.85 mmol g -1 at 0.10 bar and 2.3 mmol g -1 at 0.9 bar for lab scale synthesis and 1.25 mmol g -1 at 0.10 bar and 2.75 mmol g -1 at 0.9 bar for up scaling (Figure 2). The BET surface area derived from the N 2 isotherm were 424 m² g -1 and 506 m² g -1 respectively ( Figure S2), which is close to what has previously been reported (471 and 480 m 2 g -1 ) [39,50]. TGA analysis of the as-prepared ZIF-94 in air showed a thermal stability up to ~225 ºC with ~20% weight loss due to THF and methanol removal ( Figure S3).
The scale-up of the synthesis from the laboratory small scale to the laboratory pilot scale was achieved for ZIF-94. The characteristics of the resulting up-scaled sample match on the whole those of the solid produced at smaller scale, both in terms of crystallinity, purity and particle size and shape.

Pure polymer: pure gases vs mixed gases for 6FDA-DAM membrane
To date, most of the permeation data reported in literature is for pure gases. In most cases transport behavior of gas mixtures through membranes is different from that of pure gases [52,53]. This leads to differences in pure gas and mixed gas permeabilities and selectivities.
Hence, in order to assess membrane properties under real process conditions, it is essential to determine mixed gas permeation performance.
The bare 6FDA-DAM membrane was tested both for pure gases (CO 2 and N 2 ) and CO 2 /N 2 gas mixtures (15:85) at different transmembrane pressure differences (1 to 4 bar). Single and mixed gas permeabilities (CO 2 and N 2 ) and CO 2 /N 2 selectivities are presented as a function of transmembrane pressure in Figure 7. For pure CO 2 , permeability decreased with increasing feed pressure (from 540 to 450 Barrer at 1 and 4 bar, respectively), a behavior predicted by the dual-mode sorption and mobility models for gas permeation of condensable gases such as carbon dioxide in glassy polymers [54]. Meanwhile the permeability of the low adsorbing penetrant N 2 exhibited little or no dependency on pressure [54]. As a result, a slight decrease in the ideal CO 2 /N 2 selectivity was observed with increasing feed pressure for pure gases. It is worth mentioning that another phenomenon that can influence gas permeation through membranes is plasticization, i.e. sorption induced swelling of the polymer matrix, causing an increased polymer chain mobility and consequently increased gas permeability. For 6FDA-DAM polymer, plasticization with CO 2 takes place at pressures higher than 10 bar [55]. Therefore, plasticization is excluded to interfere under the studied conditions. As expected, significant differences between pure and mixed gas permeation are observed.
The mixed gas permeability for N 2 is lower than the pure gas permeability. In the case of CO 2 , at first sight the mixed gas permeability seems higher than the pure gas permeability but, if instead of the total feed pressure, one takes into account the partial CO 2 transmembrane pressure difference (from 0.15 to 0.60 bar) then the values for CO 2 permeability fit the trend of higher permeability at lower pressures. The dual-mode sorption model for gas permeation estimates lower permeability for all mixture components due to competitive sorption between gases for the polymer matrix sorption sites. Nevertheless, CO 2 has a much higher affinity constant and solubility in glassy polymers than N 2 . Therefore polymer matrix sorption sites are saturated with CO 2 , and N 2 permeability is decreased. Similar to pure gases, a decrease in CO 2 permeability was also observed as the transmembrane pressure increased in the mixed gas test (from 768 to 670 Barrer at 1 and 4 20 bar, respectively). Furthermore, also a small increase in N 2 permeability with increasing pressure was observed, resulting in an unchanged CO 2 /N 2 separation factor.
An ideal CO 2 /N 2 selectivity of around 14 was obtained based on pure gas permeation, whereas the CO 2 /N 2 separation factor for the mixed gas test was 24 over the whole total transmembrane pressure difference range from 1 to 4 bar. This emphasizes the importance of performing mixed gas experiments in order to know membrane performance under relevant conditions for commercial applications. Hence only mixed gas performance is reported below for the MMMs prepared in this work. The increased CO 2 permeability with filler loading might be related to (1) disruption of the chain packing of the polymer, (2) porosity introduced by MOF particles [34] and (3) increase in the polymer free volume [56][57][58]. The characterization of the MMM shows that in our case the MOF distribution is quite even throughout the membrane with good adhesion between the MOF filler and the polymer. Therefore, the permeability of MMMs is evaluated by the Maxwell equation [6]. First, the unknown parameter of the Maxwell equation P d (disperse phase permeability) was calculated from experimental 10 wt% membrane permeability, whereas P c (continuous phase permeability) was determined from the pure polymer permeation. P d could in principle be determined experimentally from films made of pure MOF. Preparing such films is extremely challenging if not impossible because they should be self-supported with MOF phase densely packed, with no subnanometer defects between filler particles. MOF phase with such morphology was achieved only by in situ synthesis on top of porous supports [60] and was not the subject of this study. ZIF-94 (SIM-1) membrane has been crystallized in situ on a tubular asymmetric alumina support with pore size of 200 nm by Aguado et al. [41]. They demonstrate that the gas transport obeys the Knudsen diffusion mechanism such as found for microporous membranes. This is translated in higher gas permeability than in polymers and low (Knudsen) gas selectivity. At 35 ºC they have achieved overall single gas CO 2 permeance of 104 GPU. Considering the reported MOF layer thickness was of approximately 25 µm and no resistance to the overall transport by the support, we estimate the pure ZIF-94 phase permeability by multiplying permeance with layer thickness to a value of 2613 Barrer. This value is related to pure gas permeation. Our estimated P d from the 10 wt.% experiment represents the mixed gas permeability and was equal to 4735 Barrer.
Then theoretical MMM permeabilities have been predicted for other loadings. The experimental and predicted permeability are presented in Figure 8. Experimental permeability values of 10-30 wt% loaded membranes follow the trend predicted by the Maxwell relation. However, a higher experimental permeability than predicted is obtained for the 40 wt% loaded membrane. The lack of significant change in selectivity with MOF loading relative to the bare polymer demonstrates that the prepared MMMs are 'defect-free' and suggests that the polymer determines the selectivity, while the MOF introduces faster 22 transport pathways. The larger permeability than predicted according to the Maxwell model indicates that at high loadings the ZIF-94 influence is not captured by this model, which is approximately only valid up to volume fractions of 20%.   Table 1. In terms of CO 2 permeation through pure polymer we can observe different results. Polymers used in literature were synthesized in the laboratory and different molecular weight could lead to differences in gas separation performance. Also, as 6FDA-DAM has a high free volume, the solvent used for the fabrication of the membrane and the membrane history might influence final separation performance of the membrane. Also, it is worth to mention that we compare single with mix gas results. Nevertheless, the common feature is a significant increase in CO 2 permeability observed by the addition of the filler particles in the 6FDA-DAM polymer matrix. In some cases, there is a selectivity increase due to either molecular sieving [37] or to solubility increase ( [36]. Zhang et al. went a step further and ZIF-8/6FDA-DAM mixedmatrix hollow fiber membranes were prepared with ZIF-8 nanoparticle loading up to 30 wt % [61]. The mixed-matrix hollow fibers showed significantly enhanced propylene/propane selectivity that was consistent with mixed-matrix dense films. We extend the benchmarking of our best membrane via a Robeson plot in Figure 9, Barrer CO 2 , ideal CO 2 /N 2 selectivity 19.1, point k in Figure 9) and (c) 30 wt% loaded PIM-1 membrane (6300 Barrer CO 2 , ideal CO 2 /N 2 selectivity 18, point n in Figure 9). Their performance has been only tested for pure gases (ideal selectivity) while our membranes have been tested under relevant process conditions (mixed gas 15/85 CO 2 /N 2 ).  Furthermore, the increase in CO 2 permeance at higher ZIF-94 loading would lead to significant reduction in total membrane area required, and hence in the reduction of investment cost for CO 2 capture, as reported by Lively et al. [22].
The developed MMM have great potential to be spun into a hollow fiber membrane configuration since the bare polymer is spinnable and the ZIF-94 filler has a smaller size than the selective layer (< 500 nm). Therefore, the region of optimal membrane properties for the separation of CO 2 from flue gas identified by Merkel et al. [69] can be amply reached with any of the MMMs prepared in this work (10 to 40 wt% ZIF-94 loading).