N-Aryl–linked spirocyclic polymers for membrane separations of complex hydrocarbon mixtures

See allHide authors and affiliations

Science  17 Jul 2020:
Vol. 369, Issue 6501, pp. 310-315
DOI: 10.1126/science.aba9806

Separating organics without distillation

Hydrocarbon distillation is a widespread and energy-intensive process. Membranes might offer an alternative approach, but few can survive immersion in organic solvents nor are they able to extract relatively small molecules. Thompson et al. developed a series of polymers of intrinsic microporosity that they used for membrane-based separations of organic compounds in an organic solvent (see the Perspective by Brennecke and Freeman). The new membrane has a molecular weight cutoff of 253 daltons, far lower than existing ones closer to 600 daltons. The polymers were used to separate light shale crude oil and succeeded in fractionation of molecular weights of about 170 daltons.

Science this issue p. 310; see also p. 254


The fractionation of crude-oil mixtures through distillation is a large-scale, energy-intensive process. Membrane materials can avoid phase changes in such mixtures and thereby reduce the energy intensity of these thermal separations. With this application in mind, we created spirocyclic polymers with N-aryl bonds that demonstrated noninterconnected microporosity in the absence of ladder linkages. The resulting glassy polymer membranes demonstrated nonthermal membrane fractionation of light crude oil through a combination of class- and size-based “sorting” of molecules. We observed an enrichment of molecules lighter than 170 daltons corresponding to a carbon number of 12 or a boiling point less than 200°C in the permeate. Such scalable, selective membranes offer potential for the hybridization of energy-efficient technology with conventional processes such as distillation.

Each day, ~100 million barrels of crude oil are fractionated in refineries around the globe, primarily by thermal distillation. Though distillation has the potential for high recoveries of molecules in discrete boiling-point ranges, it is energy intensive, requiring more than 1100 terawatt-hours per year—nearly 1% of global energy use (1). Distillation energy and carbon efficiency could be improved if it were paired with low-energy membrane-based separations that fractionate complex mixtures containing thousands of compounds into smaller groups of molecules. Membrane separations based on molecular differences in size, shape, and membrane-penetrant interactions have the potential for a 10-fold increase in energy efficiency over thermal processes (2). Seawater reverse-osmosis systems that currently process greater liquid flows than the largest refineries highlight the potential for membrane technology to match the scales required by the hydrocarbon-processing industry. Whereas membrane technologies are established for certain aqueous and gas separations, they have not been developed for the fractionation of organic solvent mixtures, because of the relative scarcity of effective materials that are easily processable, inexpensive, and stable (3). Polymeric membrane materials provide the best combination of these three metrics but to date have not been sufficiently effective for separations of small liquid organic molecules.

Polymers of intrinsic microporosity (PIMs), with their contorted ladder-like structures, have high guest permeability and have therefore been widely studied for gas separations and, to some extent, for organic solvent separations (46). For the prototypical polymer PIM-1, there is a substantial disparity between the apparent pore-size distributions in the “dry” and organic solvent “wet” states. Cryogenic physisorption experiments in solid PIM-1 suggest an average pore size between 3.5 and 5.0 Å (7). By contrast, molecular weight cutoff measurements of styrene oligomers in toluene by PIM-1 membranes were reported (and were confirmed in this study) to be greater than 600 g mol−1 (8), which corresponds to a pore size of 14 Å in a rigid membrane material such as a ceramic (9). This discrepancy derives from motion-enabled zones of activation in the semirigid polymer network that undergo swelling and plasticization in organic solvents (5). A PIM-like polymer that does not exhibit these phenomena should enable separations of small molecules in the organic solvent reverse osmosis (<200 g mol−1) and “tight” organic solvent nanofiltration (200 to 300 g mol−1) regimes. Although postpolymerization modifications can tune the properties of PIM-1–like materials (10), structures capable of performing difficult separations without postpolymerization or postfabrication modification would provide more chances to create solutions to these impactful, large-scale challenges.

We took note of previous gas separation work by McKeown and colleagues in which a spirobifluorene monomer was found to frustrate chain packing and rigidify ladder-type polymer structures beyond equivalent materials made by using a spirobisindane monomer (6), resulting in higher polymer free volume and permeability relative to PIM-1. Moreover, swelling and plasticization were reduced, resulting in no apparent loss of separation selectivity compared with PIM-1. We therefore employed a similar spirobifluorene building block but realized that this modification alone would likely be unable to address the challenge posed by swelling in organic solvent separation compared with gas-phase separations. Thus, we sought to enhance packing efficiency in certain segments of the backbone by using a modestly flexible linkage, choosing the aryl–N-aryl connection.

This design allowed the use of catalytic C–N bond formation as the polymerization process, rather than the nucleophilic aromatic substitution reaction that gives rise to the dibenzodioxin linkage of PIM-1. The palladium-catalyzed Buchwald-Hartwig amination reaction is a robust method for the coupling of aryl halides and amines that has been used extensively in medicinal, materials, and natural products chemistry (11); its application to polymer synthesis has been more limited. We chose a single spirobifluorene dibromide (1) (Fig. 1 and fig. S1) as the A–A monomer and a range of commercially available aromatic diamines as the B–B component (compounds I to IV in Fig. 1). Polymerization using the XantPhos Pd G4 palladacycle catalyst (12) was optimized (table S1) with respect to temperature and concentration, providing the spirobifluorene aryl diamine (SBAD) series (Fig. 1).

Fig. 1 Modular reaction scheme, computational simulation of porosity, powder physisorption, liquid sorption, and spectroscopic characterization.

(A) Polymerization of 7,7′-dibromo-2,2′,3,3′-tetramethoxy-9,9′-spirobifluorene, 1, with diamines I to IV giving the SBAD class of materials: SBAD-1 to SBAD-4. Below the diamine structures are representative results of computational modeling of the pore surfaces for the SBAD series compared with PIM-1 applying a 2.2 Å probe diameter, the largest probe that can pass through the SBAD materials; teal indicates accessible (interconnected) surface area, and magenta indicates nonaccessible surface area. (B) CO2 uptake (at 273 K) by SBAD polymers compared with that by PIM-1. (C) Uptake of liquid toluene by dense films of SBAD-1, PIM-1, and Torlon (22), (N = 2) 22°C, 1 atm. Data are the mean of 2 films ± the range. (D) SEM image of SBAD-1 thin-film composite showing an approximate barrier-layer thickness of 200 nm. tBu, tert-butyl; Me, methyl; STP, standard temperature and pressure.

All SBAD polymers were readily soluble in volatile solvents (tetrahydrofuran, chloroform, CH2Cl2) (fig. S2) and were easily cast into films. Gel permeation chromatography analysis revealed a component of high apparent molecular weight, thought to correspond to a small amount of branched material derived from a second N–H insertion event at one or more amine sites along the polymer chain (fig. S3). Even a small amount of branching at secondary amine centers can substantially inflate observed weight-average molecular weights and dispersities (13). The SBAD polymers demonstrated high thermal stabilities, with no observable glass transition temperatures (figs. S4 and S5), as has often been reported for PIM systems (14). These materials also exhibited low uptakes of N2 at 77 K (fig. S6A) but had only moderately lower CO2 sorption than PIM-1 at 273 K (Fig. 1B), suggesting the existence of isolated microporous free-volume elements.

To further probe the pore structure of these materials, computational models for the polymers in the dry state were generated by using the polymerization algorithm Polymatic (15). Three models were generated for each system to sample different structural arrangements and ensure the models were representative of the bulk material. The properties of the models were then characterized (figs. S7 and S8 and table S2) (16). PIM-1 and the SBAD series models showed similar predicted densities (table S3) but notable differences in polymer voids, as seen in Fig. 1A, fig. S9, and table S3. The diameter of the largest probe that can pass through an interconnected pore network in the PIM-1 models is 3.2 ± 0.3 Å, whereas the equivalent value for the SBAD series is 2.3 ± 0.1 Å. Only a small amount of flexible breathing motions or polymer swelling would be required for nitrogen [diameter, 3.64 Å] to diffuse through PIM-1, but much larger dilations would be required with the SBAD series (figs. S10 and S11). This is consistent with the observed nitrogen physisorption isotherms (fig. S6) showing high uptakes for PIM-1 compared with the SBAD series at low relative pressures, which is the range that is typically used to calculate micropore distributions. By defining “interconnected pores” as those that are accessible by N2 at 77 K within a reasonable time frame (2 days), it is apparent that “interconnected microporosity” is a characteristic feature of PIM-1 that is suppressed in the SBAD polymers. This effect is likely due to the absence of the dibenzodioxin linkage and ladder-like morphology of PIM-1 (figs. S12 and S13), replaced in the SBAD series with more-flexible single-bond C–N linkages, allowing for the formation of π–π stacking interactions between and within chains (fig. S14) and more-efficient chain packing.

The computational work also suggests the existence of narrow distributions of ultramicropores between 2 and 8 Å for SBAD-1 and -3 and 2 and 10 Å for SBAD-2 and -4, whereas PIM-1 has an apparently broader distribution between 2 and 12 Å (figs. S15 and S16). Even tighter distributions of ultramicropores are seen in traditional glassy polymers (17), but the pore sizes are often too small for meaningful organic solvent separations. Thus, the presence of intermediate noninterconnected pores exhibited by the SBAD materials could offer penetrant-penetrant differentiation of small-molecule organic solvents at reasonable fluxes if diffusion from one pore to the next were reasonably fast. In the wet state, separating small organic molecules requires that the membrane material exhibit resistance to solvent-induced swelling (e.g., Torlon, a highly solvent-stable polyamide-imide) (Fig. 1C). Solvent-induced dilation experiments showed that SBAD-1 experienced substantially less mass change than PIM-1 at unit activity of toluene (Fig. 1C), suggesting that the SBAD materials retain their narrow distribution of ultramicropores even when exposed to solvents.

Mechanically robust thin-film composite membranes were fabricated on cross-linked polyetherimide supports with no observable interlayer delamination (Fig. 1D and fig. S17). The molecular separation performance of these materials was surveyed with binary mixture separations of 1,3,5-triisopropylbenzene (TIPB) in toluene (Fig. 2A). TIPB serves as a surrogate solute molecule for organic solvent separations, with a molecular weight (204.35 Da) at the lower end of the typical nanofiltration range (200 to 1000 g mol−1) in which PIMs have often been tested (9). Membrane performance was evaluated through cross-flow filtration (fig. S18) after more than 48 hours of continuous permeation of the mixture.

Fig. 2 Dilute hydrocarbon separation performance using polymer thin-film composites.

(A) TIPB rejection versus toluene permeance for a 1/99 mol % feed of TIPB/toluene (N = 3), 22°C, 15 bar. (B) Molecular weight cutoff curves of SBAD-1 (N = 3) at 30 bar (threshold = 335 g mol−1), PIM-1 at 5 bar (1220 g mol−1), and PIM-1 data reported at 30 bar (786 g mol−1) (8) for polystyrene standards in toluene at 22°C. The high flux in PIM-1 in our work limited the applied pressure difference to 5 bar to maintain <5% stage cuts. Data reported by Cook et al. at the same conditions as SBAD-1 was shown for direct comparison. (C) Hydrocarbon molecular weight cutoff curve (threshold = 253 g mol−1, fig. S21) for SBAD-1 (N = 2) and Puramem 280 (N = 3) from solute mixtures listed in table S4. Data are the mean of N membranes ± standard deviation. i-Pr, isopropyl; MWCO, molecular weight cutoff.

As shown in Fig. 2A, all of our polymers exhibited significantly greater rejections at steady state than PIM-1, with permeances in the range of 0.1 to 0.7 liters m−2 hour−1 bar−1. SBAD-1 showed the highest TIPB rejection of 80% throughout the duration of the experiment, and SBAD-3 was the least effective, despite having similar predicted micropore size distributions as the other materials. The most relevant structural difference appeared to be the larger biphenyl bridge structure used in SBAD-1, 2, and 4 versus the smaller phenyl bridge in SBAD-3. The improvement in TIPB rejection with the larger bridge suggests that aromatic chain-chain interactions, more available with biphenyl linkages (fig. S14), are advantageous. The rejection properties of SBAD-2 and SBAD-4 increased over 48 hours, reaching steady-state values only slightly lower than those of SBAD-1 (fig. S19). Such time-dependent behavior is consistent with solvent-induced polymer relaxation toward a less swollen state (18). Only a handful of commercial organic solvent nanofiltration membranes and other polymeric membranes with complex fabrication procedures are capable of high rejections of ~200 g mol−1 solutes and have toluene permeances in the same range (9). As SBAD-1 demonstrated the highest rejections, it was chosen for scale-up and further evaluation with more-complex mixtures.

SBAD-1 and PIM-1 were compared in a standardized measurement of molecular weight cutoffs by using oligostyrene markers in toluene. SBAD-1 was significantly better than PIM-1 in the separation of small solutes (molecular-weight cutoff of 335 versus 1220 g mol−1, respectively), albeit at a much lower permeance (0.2 versus 3.49 liters m−2 hour−1 bar−1). The molecular weight cutoff of PIM-1 increased over the course of 3 days, whereas that of SBAD-1 remained constant (fig. S20).

The separation performance of SBAD-1 using a group of nonpolar hydrocarbon molecules in the organic solvent reverse-osmosis and nanofiltration ranges that are representative of hydrocarbons found in light crude oils is shown in Fig. 2C. A hydrocarbon molecular-weight cutoff of ~253 g mol−1 (fig. S21) was calculated. Below 150 g mol−1, greater aromaticity (ratio of aromatic carbons to aliphatic carbons) of the solute was found to diminish rejection, perhaps because of greater sorption interactions with the aromatic rings present in the polymer backbone. Dual dependence on both size and aromatic interaction was supported by a linear correlation to a combination of aromaticity (sorption) and molecular weight (diffusion) factors, but to neither factor alone (fig. S22). When compared with Puramem 280, a commercial polyimide membrane, SBAD-1 showed higher rejections, although this was accompanied by lower permeances, as shown in table S4 at 22°C. As the thin-film thicknesses were not controlled in this work, the differences in permeance cannot be directly compared, because the SBAD-1 films are considerably thicker.

The naphtha and kerosene fractions of crude oil make up the primary component of gasoline and jet fuel, respectively. These two fractions typically contain a range of alkanes, alkyl aromatics, and polyaromatics with boiling points from 90° to 300°C, representing a good candidate for membrane-based hydrocarbon fractionation. A thin-film composite membrane sheet of SBAD-1 was produced through roll-to-roll coating of the polymer onto a prefabricated Ultem support (fig. S23, A and B). Membrane coupons from the sheet (diameter, 47 mm) were subjected to a complex feed of hydrocarbons that serve as a model naphtha-kerosene fraction. Several large spiral-wound modules (fig. S23C) (containing 1.8 m by 0.2 m of membrane) were also fabricated from the sheet to highlight the scalability of these materials.

Fig. 3A shows the membrane performance (feed composition shown in table S5) as the ratio of permeate to retentate concentrations plotted against the molecular weights of the components. A permeance of 0.022 ± 0.013 liters m−2 hour−1 bar−1 for the coupon and 0.076 ± 0.003 liters m−2 hour−1 bar−1 for the module (fig. S24) was obtained. This permeance is lower than that in the dilute-mixture experiment (Fig. 2) because of the high osmotic pressure of the complex, multicomponent mixture. However, good separation of molecules of different molecular weights, independent of molecular class, as well as for molecules from different classes without a molecular-weight advantage was obtained. Additionally, exposure to temperatures up to 75°C did not negatively affect membrane performance (fig. S25). Some notable separation factors (based on coupon data) are shown in table S6, illustrating the potential of SBAD-1 for membrane-based concentration of solutes of both low molecular weights and high aromaticity. The values highlighted are comparable to separation factors obtained for liquid xylene isomers by using carbon molecular sieve membranes (19). Aromatic-rich permeates streams (observed in this work) are useful for the production of plastics and provide a rich source of industrial organic solvents. Fractionation of crude oil into groups of smaller and larger molecules also has direct value in generating products classified by the number of carbons present such as gasoline (C5-C12 alkanes and cycloalkanes) and jet fuel (C10-C18 alkanes and aromatics). The retentate stream shown in Fig. 3A contains both of these.

SBAD-1 was also challenged with a light shale crude oil to illustrate the potential of membrane materials as an alternative for the fractionation of crude oil. Figure 3B shows the boiling-point distribution of molecules in the feed, permeate, and retentate. Approximately 38% of the mixture boiled below 200°C in the feed tested in this work. After passing through a thin-film composite of SBAD-1, the permeate was found to be appreciably enriched in the lighter molecules present in the feed, such that more than 60% of the permeate boiled below 200°C. Figure 3C shows a picture of the crude oil and the lighter-colored permeate indicating removal of color bodies that are typically high in molecular weight. Figure 3D visualizes the differences in GCxGC-FID chromatograms of the feed and permeate, with the three axes representing boiling point (time scale in minutes), polarity (time scale in seconds), and response factor (z axis). The GCxGC-FID data indicates enrichment of molecules lighter than carbon numbers of 12 or around molecular weight of 170 Da (Fig. 3E). The membrane permeance is low (0.016 liters m−2 hour−1 bar−1), as the thin-film fabrication process has not yet been optimized to create more-productive composites. Overall, the SBAD-1 membrane-based separation demonstrated selectivity for low–molecular-weight saturated molecules in the naphtha and early kerosene boiling range with slight permeation of higher–molecular-weight linear paraffins.

This result is similar to a boiling-point separation of a complex hydrocarbon mixture, but without expending the energy associated with vaporization. A clear “interclass” separation of the crude was observed with a preferential order of permeation: naphthenes, linear paraffins, aromatics, and branched paraffins (fig. S26 and table S7). Moreover, intraclass size-based separation also occurred for each of these individual classes of hydrocarbons, such that the permeate was enriched in the low boiling (or “light”) compounds in the crude oil (fig. S27). Although we did not specifically explore fouling effects from the real crude feed, the permeance and separation factors changed minimally over a period of 2 months during synthetic crude testing at temperatures ranging from 25° to 75°C (fig. S25).

An interesting distinction emerges when comparing traditional binary separation experiments (i.e., toluene and TIPB) and dilute multicomponent hydrocarbon separations with the complex mixture fractionation experiments (i.e., fractionation of lab-made and real crude oil). Whereas molecular weight cutoff and rejection are important figures of merit for dilute separations such as water purification and nanofiltration, the utility of this metric is less clear in the case of complex mixture fractionation, in which there is no discrete solvent or solute: Every molecule in crude oil is a solvent. In this case, each molecule in the mixture experiences highly nonideal coupling effects (e.g., diffusion and sorption) during permeation. Moreover, the osmotic effects in this crude-oil system are complex and certainly contribute toward reducing the fluxes relative to the diluted mixtures shown in Fig. 2. The membrane-based crude-oil separation occurs through a combination of diffusion-based selection processes as well as solvent-membrane (and solvent-solvent) interactions, which further reinforces the importance of testing real mixtures. Beyond testing industrial feeds, it will be important to establish how these materials fit into an energy-efficient membrane-fractionation process (20). To produce distillation-like products through membrane fractionation, a cascade of membrane materials such as SBAD-1, capable of fractionating feeds across a specific range of hydrocarbon sizes, will be necessary (fig. S28).

The results described here highlight that the dibenzodioxin connection between spirocyclic monomers that characterizes PIM-1–like materials is not necessary to enable meaningful small-molecule membrane-based separations. We speculate that the advantageous properties of SBAD-based membranes derive from the nature of the aromatic nitrogen-carbon linkage, which has long been known to demonstrate hindered rotation (21). This connection between spirocyclic and aromatic building blocks gives rise to bulk rigidity and low swelling, isolated and large micropores, and enough segmental mobility to allow migration of molecules between micropores at rates influenced by both molecular size and molecular type.

Together, these properties make the SBAD family promising for the separation of small-sized molecules spanning the organic solvent reverse osmosis and the lower end of the organic solvent nanofiltration range (100 to 350 g mol−1) in complex mixtures such as crude oil. Most other membranes that operate in similar regimes require postpolymerization or postfabrication modifications (e.g., cross-linking) that can be costly and difficult to scale up. By contrast, the polymers used in this study can be tailored with simple changes in composition and performance. The proposed dependence of SBAD performance on both pore structure and dynamic motion enables complex mixture fractionation that may be scalable and translatable to real crude-oil fractions.

Fig. 3 Synthetic and real crude-oil separation performance using SBAD-1 thin-film composites.

(A) Ratio of concentrations in permeate (Cp) versus retentate (Cr) of components in a complex, model crude oil fractionated by a thin-film composite membrane coupon of SBAD-1 (N = 3); 40 bar, 22°C. Molecules above the reference line were more concentrated in the permeate compared with the feed, whereas the molecules below were more concentrated in the retentate. Data are the mean of three membranes ± standard deviation. (B) Boiling-point distribution of feed, permeate, and retentate from SBAD-1 membrane fractionation of shale-based crude oil. (C) Picture of feed (left) and permeate (right) from SBAD-1 membrane fractionation of real shale-based light whole crude oil. (D) GCxGC-Flame-ionized detection (FID) analysis of membrane fractionation of shale-based crude oil at 130°C and 55 bar showing the feed chromatogram subtracted from the permeate chromatogram. Teal elution peaks are concentrated in the permeate, whereas the magenta elution peaks are concentrated in the membrane retentate. General hydrocarbon classifications are highlighted with dashed red lines to guide the eye. (E) Side view of GCxGC-FID [shown in (D)] showing the carbon number partition of real crude oil obtained by SBAD-1.

Supplementary Materials

Materials and Methods

Figs. S1 to S28

Tables S1 to S7

NMR Spectra

References (2339)

References and Notes

Acknowledgments: We thank N. C. Bruno (Georgia Institute of Technology) for Pd-catalyzed cross-coupling guidance, G. Zhu (Georgia Institute of Technology) for assistance with nitrogen and carbon dioxide physisorption experiments, Y. Ma (Georgia Institute of Technology) for assistance with the dense film fabrication, D. A. Bhandari (ExxonMobil Research and Engineering) for suggesting the synthetic kerosene mixture composition, and M. L. Jue (Georgia Institute of Technology) for the synthesis of PIM-1. Funding: This work was supported by ExxonMobil Research and Engineering. K.A.T. acknowledges support from the Department of Education Graduate Assistance in Areas of National Need (GAANN) program at Georgia Institute of Technology (award no. P200A180075). K.E.J. acknowledges a Royal Society University Research Fellowship. A.T. acknowledges a Royal Society Enhancement Grant (2018) and the European Research Council (ERC) under FP7 (CoMMaD, ERC grant no. 758370). Author contributions: Monomer and polymer synthesis, gel permeation chromatography, and elemental characterization were conducted by K.A.T. Polymer model generation and analyses were carried out by I.B., A.T., and K.E.J. Matrimid support fabrication, blade coating, spectroscopic characterization, coupon separation performance, thermal analysis of powders, and physisorption model fitting were done by R.M. Ultem support fabrication and roll-to-roll coating were performed by D.K. Spiral-wound–module fabrication and spiral-wound–module performance analysis were conducted by D.K. and J.K. Whole crude separation was conducted by N.R. and J.R.J. GCxGC analysis was conducted by S.J.H. Complex mixture separation results were confirmed by R.M., D.K., J.K., N.R., and J.R.J. at separate facilities. Temperature resistance of membranes was investigated by N.R., and J.R.J., B.A.M., A.G.L., R.P.L., and M.G.F. conceived the research. Competing interests: A.G.L. serves as a non-Executive Director of Evonik Membrane Extraction Technology UK Limited. Data and materials availability: Detailed descriptions of materials and methods, monomer synthesis, polymer synthesis, polymer model generation and analysis, powder characterization, membrane fabrication, spectroscopic characterization, and separation performance are in the supplementary materials. K.A.T., R.M., B.A.M., R.P.L., and M.G.F. are inventors on patent applications (US20190275469, US20190390126, and 62/640,253) submitted by Georgia Tech Research Corporation and ExxonMobil Research and Engineering that covers SBAD polymers, membrane fabrications, and crude-oil fractionation applications. All data needed to evaluate the conclusions in the paper present in the paper or the supplementary materials.

Stay Connected to Science

Navigate This Article