Selective Gas Transport Through Few-Layered Graphene and Graphene Oxide Membranes

See allHide authors and affiliations

Science  04 Oct 2013:
Vol. 342, Issue 6154, pp. 91-95
DOI: 10.1126/science.1236098

Gas Separations

When gas separation membranes are made thinner, they usually allow permeating gases to pass through faster. However, a thinner membrane may be poorer at separating between gas species. Kim et al. (p. 91) examined the permeability and selectivity of layered graphene and graphene oxide membranes. Gas molecules diffuse through defective pores and channels that form between the layers. Controlling these structures tuned the properties of the membranes to allow the extraction of carbon dioxide from other gases. Li et al. (p. 95) describe membranes as thin as 1.8 nanometers made from only two to three layers of graphene oxide. Small defects within the layers allowed hydrogen to pass through, separating it from carbon dioxide and nitrogen.


Graphene is a distinct two-dimensional material that offers a wide range of opportunities for membrane applications because of ultimate thinness, flexibility, chemical stability, and mechanical strength. We demonstrate that few- and several-layered graphene and graphene oxide (GO) sheets can be engineered to exhibit the desired gas separation characteristics. Selective gas diffusion can be achieved by controlling gas flow channels and pores via different stacking methods. For layered (3- to 10-nanometer) GO membranes, tunable gas transport behavior was strongly dependent on the degree of interlocking within the GO stacking structure. High carbon dioxide/nitrogen selectivity was achieved by well-interlocked GO membranes in high relative humidity, which is most suitable for postcombustion carbon dioxide capture processes, including a humidified feed stream.

Compared with traditional systems, membrane-based gas separations have allowed much simpler, more efficient separations (1). One of the challenges of developing membrane materials is to achieve high flux as well as high selectivity. Because a membrane flux is inversely proportional to the thickness of selective layer (from tens of nanometers to several micrometers), a graphene sheet with one-atom thickness has been considered as the ultimate membrane platform. In its defect-free form, however, graphene exhibits perfect barrier properties to all molecules and ions (2, 3). Synthesis of large-area monolayer graphene by chemical vapor deposition (CVD) has recently been reported (4), and the resultant high-quality and uniform graphene sheet can be transferred to various substrates. We hypothesized that deposition of defect-free graphene sheets on gas-permeable polymer films would allow for preparation of flexible plastic barrier films to prevent gas and water vapor penetration. Therefore, we prepared polymer-supported thin films of graphene with varying numbers of deposited graphene sheets (Fig. 1A). Poly(1-methylsilyl-1-propyne) (PTMSP) was selected as the substrate because it is one of the most permeable, rigid, and glassy polymers available (5). The gas permeability of the resultant barrier films decreased as the number of graphene layers deposited increased. However, a single graphene-layered PTMSP film (G1/PTMSP) showed a relatively small reduction in gas permeability compared with the original PTMSP (Fig. 1B), implying the presence of defects generated during the CVD growth and transfer process (Fig. 1C). These defects also resulted in a large electrical sheet resistance (>2500 ohms per square) (Fig. 1B), similar to values reported in the literature (6). This sheet resistance was considerably reduced (<1000 ohms per square) as more graphene layers were stacked. The reduction in sheet resistance correlated well with the gas permeability of the film (Fig. 1B).

Fig. 1 Few-layered graphene-coated polymer membranes.

(A) Scanning electron microscopy (SEM) images of graphene/PTMSP membrane surfaces. (B) Changes in gas permeability and sheet resistance of graphene/PTMSP membranes as a function of the number of graphene layers. Error bars indicate the SD of all raw data. (C) Atomic force microscopy (AFM) images of a graphene/PTMSP membrane surface. (D) Relation between O2 permeability and O2/N2 selectivity of graphene/PTMSP membranes. The upper bound is from (10), as well as gas separation performance data of CMS membranes reported in the literature (11). Error bars indicate the SD in the current measured from five parallel experiments.

We anticipated that the porosity caused by the graphene sheet defects could be mitigated by depositing more graphene layers. Although the gas permeability of the films decreased gradually with increasing number of graphene layers, we were not able to obtain excellent barrier films. Surprisingly, the O2/N2 selectivity of G#/PTMSP films (where # represents the number of graphene layers) improved (Fig. 1D) with increasing graphene stacking, which suggests that gases diffused not only through defective pores on the graphene sheet but also between the graphene interlayers (79), leading to higher selectivity than the original PTMSP film. The O2/N2 selectivity increased from 1.5 (for PTMSP) to 6 (for G5/PTMSP), whereas the O2 permeability decreased from 730 to 29 barrer [1 barrer = 1 × 10−10 cm3 cm/cm2·sec·cmHg at standard temperature and pressure (STP)], surpassing the O2/N2 separation limitation of polymeric membranes (10) with a selectivity similar to that of carbon molecular sieve (CMS) membranes (11).

A possible explanation for this phenomenon is that the defective graphene sheets were irregularly aligned, resulting in the formation of gas-permeable, slit-like interlayer spacings. Pristine graphene synthesized via CVD has a polycrystalline structure with grain boundaries linked to pentagon-heptagon pairs (12). Mechanical cleavage or oxidation at these sites can lead to sheet defects, such as tears or holes (12, 13), which allow ingress of gas molecules. Furthermore, because of the irregular stacking caused by corrugations, wrinkles, and ripples, randomly stacked graphene sheets had an interplanar spacing larger than that of closely packed graphitic layers (0.335 nm). The average interlayer spacing of randomly stacked graphene layers on a silicon wafer was 0.355 nm, albeit with a broad spacing distribution, as measured by two-dimensional (2D) grazing-incidence x-ray diffraction (fig. S1). The interlayer spacing of the graphene layers on PTMSP would be larger because the surface of this polymer film is not as flat as that of mica or silicon wafers (14). The increase in selectivity of the G#/PTMSP films, with the number of graphene layers deposited, indicates the formation of slit-like interlayer spacings of similar size to those present in CMS membranes.

We investigated the gas permeation behavior of thin graphene oxide (GO) membranes. GO’s interlayer spacing is larger than that for graphene—namely, 0.6 to 1.0 nm—depending on the preparation method and the presence of water intercalated in the GO layers (15, 16). Thick (>10-nm) GO paper and thin (3- to 7-nm) GO membranes can be easily prepared by various solution methods, such as vacuum filtration, direct evaporation, and spray and spin coatings (16).

GO films in the dry state are not permeable to small gases (e.g., He), but water molecules permeate through hydrated GO films faster than expected (17). The potential energy barrier for gas entrance, tortuosity determined by the GO sheet size and thickness, layer structure, residual water content, and deformation of thin-layered structures during dewetting can all affect molecular transport in GO films. In contrast to graphene sheets, GO is not completely 2D (18), which suggests that GO should stack nonuniformly. For the above reasons, thin GO membranes should be permeable to small gases if sufficient transmembrane pressure is applied to overcome the energy barriers to pore entry and diffusion within pores or channels. In fact, we found that gas can permeate through even thick GO membranes at elevated transmembrane pressures and that the gas permeability can be tuned by changing the average GO sheet size (fig. S2). Furthermore, detectable gas permeance through the ultrathin GO membranes is strongly dependent on applied transmembrane pressure for a given average GO sheet size (fig. S3).

We used two different coating methods to prepare GO membranes on microporous polymeric membranes. The microporous polymeric support was selected to optimize both surface roughness and wettability to achieve a homogeneous coating. Among the commercially available supports, polyethersulfone (PES) membranes yielded the most uniform coating (fig. S4). We prepared several-layer GO membranes by contacting the support membrane surface to the air-liquid interface of a GO solution, followed by spin-coating (method one). We also prepared thin GO membranes by spin-casting of a GO solution on the membrane surface (method two) (fig. S5). Macroscopically, after several coatings, both membrane surfaces appeared to be homogeneously coated by a GO thin film without detectable pinholes or cracks under an optical microscope. The color of the GO-coated membranes (light yellow) differed from that of the pristine PES membrane surface (white) (fig. S6). Transmission electron microscopy (TEM) revealed that the coated GO thicknesses were ~3 to 7 nm (Fig. 2, A and B), depending on the number of coating iterations. We selected these GO coating methods to test the relative importance of electrostatic repulsion and immersion capillary force on the GO stacking structure. When considering GO deposition by method one, initial GO solution-membrane contact is performed without spin casting so that GO deposition is governed by electrostatic and hydrophilic interactions between the GO sheets and the polymer membrane. Generally, in an alkaline aqueous solution, GO sheet edges are negatively charged due to pendant ionized carboxylic acid groups. Therefore, GO sheet edge-to-edge interactions are repulsive, leading to an island-like assembly of GO sheets on the membrane surface. After this initial GO deposition on the membrane surface, water between deposited GO sheets and GO sheets in solution is removed through spin casting, allowing strong attractive capillary forces to further deposit GO sheets from solution. A second iteration of the dipping procedure in method one leads to further GO deposition through face-to-face attractive interactions (for instance, van der Waals interactions and hydrogen bonding) between the size-mismatched GO sheets (19). We found that this method leads to a relatively heterogeneous GO deposition (Fig. 2, C and E), in which the GO stacking structure resembles islands. In contrast, method two, in which the GO solution-membrane contact occurs only during spin casting, leads to a considerably more dense GO deposition. Presumably, this dense stacking occurs because the face-to-face attractive capillary forces created by the spin coating overcome the repulsive edge-to-edge GO sheet interaction. Therefore, the initial deposition is governed more by capillary interactions between the GO sheet faces and not the electrostatic interaction between the GO edges (Fig. 2, D and F). Cross-sectional TEM images show different stacking structures using each coating method (Fig. 2, A and B).

Fig. 2 Ultrathin GO membranes.

Cross-sectional TEM images of ultrathin GO membranes. (A) Method one; (B) method two. SEM images of GO films coated on an SiOx wafer (here, x indicates the number of O atoms). (C) Method one; (D) method two. AFM images of GO membrane surfaces and inert figures showing depth profiles of GO membrane surfaces. (E) Method one [root mean square roughness (Rq) = 0.8 nm, average roughness (Ra) = 0.614 nm]; (F) method two (Rq = 0.608 nm, Ra = 0.467 nm).

Gas permeance [in gas permeation units (GPUs), 1 GPU = 1 × 10−6 cm3/cm2·sec·cmHg at STP was measured at a feed pressure of 1 bar, using the common constant-pressure, variable-volume method. We found that the gas transport behavior of membranes prepared by methods one and two was very different. The GO membranes prepared by method one showed typical gas permeation behavior explained by Knudsen transport of gases in nanoporous membranes (Fig. 3A). Here, gas transport is determined by the pore size of the membrane and the free path length of the molecules in a gas mixture, and Knudsen diffusion leads to separation of gases with large differences in their molecular weights (20). As expected, gas permeance decreased according to the molecular weight of the gases. The selectivity of membranes prepared by method one to He was in good agreement with theoretical Knudsen selectivity, but this did not hold true for their selectivity to CO2 (Fig. 3A). During the permeation test at a feed pressure of 1 bar, CO2 permeance decreased sharply with time and stabilized after 1 hour, whereas other gases did not show this reduction (Fig. 3B). Consequently, the small gas/CO2 selectivity of the membranes exceeded the Knudsen selectivity to a large extent; for example, the H2/CO2 selectivity was as high as 30, whereas the theoretical Knudsen selectivity of H2/CO2 is 4.67.

Fig. 3 Gas transport behavior through ultrathin GO membranes.

(A) Gas permeances of GO membranes as a function of molecular weight (method one; dashed line represents the ideal Knudsen selectivity) under dry and humidified conditions. amu, atomic mass unit. (B) H2 and CO2 permeances and H2/CO2 selectivity of method one GO membranes as a function of permeation time. (C) Gas permeances of GO membranes as a function of kinetic diameter (method two) under dry and humidified conditions. (D) Relation between CO2 permeability and CO2/N2 selectivity of method two GO membranes under dry and humidified conditions [TR (23); tetrazole PIM (TZPIM) and PIM (25); CMS (26)]. Error bars indicate the SD of all raw data.

We now consider the role of both the porosity and chemical nature of the gas transport channels within the stacked GO structure. Highly interlocked, close-packed layering structures were not obtained, even after repeated coatings (fig. S7); rather, nanopores created by the edges of noninterlocked GO sheets formed within the GO structure. Gas permeation occurred through these nanopores, as confirmed by the Knudsen transport behavior of the GO membranes. Furthermore, a number of carboxylic acid groups line the pore outer walls, because free carboxylic acid groups are distributed at GO edges. Torrisi et al. calculated the isoteric heats of sorption for CO2 in functionalized metal-organic frameworks and predicted that the incorporation of a carboxylic acid group would lead to the highest isoteric heat (21). Although CO2 is a nonpolar gas, the polarity of the individual C-O bonds in the molecule allows for interaction with polar groups in GO. Thus, CO2 can act as a Lewis acid or a Lewis base and can participate in hydrogen bonding. Carboxylic acid groups provide a preferential site for CO2 adsorption, consequently retarding CO2 transport in the nanopores by strongly trapping CO2 molecules. This phenomenon is counterintuitive, because strong affinity between penetrant and nanopore walls often leads to surface diffusion of the penetrant preferentially sorbed on the pore wall, resulting in flux enhancement of highly sorbed or condensable penetrants (22).

Because GO is hydrophilic, we also evaluated the gas transport behavior of GO membranes (method one) by changing the relative humidity of the feed streams (Fig. 3A). In general, all gas permeances decreased as the humidity in the feed increased, implying that condensed water molecules in the pores or between layers hindered the transport of noncondensable small gas molecules. In contrast, CO2 permeance was 50 times higher in the presence of water, leading to an enhancement in the selectivities of gas pairs of interest (for example, CO2/CH4, CO2/H2, and CO2/N2) (fig. S8).

GO membranes prepared by method two formed highly interlocked layer structures and also exhibited extraordinary gas permeation behavior, albeit different than the behavior exhibited by membranes prepared via method one. For these GO membranes, the gas permeance order at 298 K was CO2 > H2 ≥ He > CH4 > O2 > N2 (Fig. 3C). This order has been observed for a few high free-volume microporous glassy polymers, such as PTMSP, that consist of a loosely packed network of polymer chains with continuous channels or pores for diffusion (5, 23). The gas transport behavior of these GO membranes means that they comprised closed-packed, interlocked GO layered structures. These GO membranes were less gas permeable than those prepared by method one but were also more selective, indicating that gases diffused selectively between the GO interlayers and, thus, had a different effective diffusion pathway. Based on the resistance model (24), we calculated the intrinsic gas permeability of the thin GO membranes. The CO2 permeability is ~8500 barrer, and CO2/N2 selectivity is ~20. The separation performance is higher than that reported for CO2 for polymeric membranes, including thermally rearranged (TR) polymers (23), polymers of intrinsic microporosity (PIM) (25), and inorganic membranes, such as carbon, silica, and zeolite (26) (Fig. 3D). The gas permeances decreased gradually as the humidity increased (27), but the CO2 permeance remained constant, irrespective of the relative humidity (fig. S9), resulting in highly CO2-selective, ultrathin (<10-nm) carbon membranes (Fig. 3D and fig. S10).

More permeable, selective graphene-based membranes can be prepared by creating pores on the basal plane by simple heat treatment (Fig. 4, A to C). Thermal treatment of ultrathin GO membranes (prepared with method two) caused decomposition accompanied by the release of, primarily, water and CO2 at a temperature of ~140°C, as confirmed by analysis with a thermogravimetric analyzer–mass spectrometer (Fig. 4D), consistent with the literature (28). At this temperature, CO and CO2 are released from oxygen-containing functional groups on basal planes or at edges, resulting in irreversible defects. After the membrane was mounted in a cell surrounded by a heating chamber, the gas permeation rates of H2 and CO2 at a feed pressure of 1 bar were recorded as a function of cell temperature. Below 130°C, H2 permeance increased gradually, whereas CO2 permeance decreased steadily. At ~130° to 140°C, H2 permeance increased abruptly and CO2 permeance increased slightly (Fig. 4E), implying that thermal annealing made the microstructures of the GO active layers more porous due to irreversible pore formation. Without such structural deformation, the slope of gas permeance versus temperature should be linearly positive (for H2) or negative (for CO2) under thermally activated diffusion conditions. However, two distinct slopes were observed, presenting evidence of a more open porous structure. The gradual reduction in CO2 permeance below 130°C is due to the low enthalpy of sorption. The increase in CO2 permeance above 140°C also indicates more porous microstructures in the GO thin layers. The H2/CO2 permselectivity at 140°C was as high as 40 (Fig. 4F), which is one of the highest values reported for other comparable membranes (10, 26, 29, 30).

Fig. 4 Gas transport behavior through thermally reduced GO membranes.

(A) Schematic illustration of pore generation in GO membranes using a thermal treatment process. AFM images of method two GO membranes (B) before (Rq = 0.553 nm, Ra = 0.450 nm) and (C) after thermal treatment at 180°C for 3 hours (Rq = 1.75 nm, Ra = 1.09 nm). The insets to (B) and (C), respectively, show TEM images before and after thermal treatment at 180°C for 3 hours. (D) He permeance trend and CO2 trace of mass spectroscopy as a function of temperature. (E) H2 and CO2 permeability of thermally treated GO membranes as a function of temperature. Error bars indicate the SD of all raw data. (F) Comparison of gas separation performance between GO membranes and other membranes in the literature [polymers (10); CMS (26, 29); zeolite (26); silica (30)].

Overall, transport in graphene or GO nanosheets has been covered less than that in carbon nanotubes, but there is great potential and the research area is growing. The ability to have adaptive and dynamic layer distances is a distinct attribute of stacked graphene or GO nanosheets. With careful control of the stacking structures, separation factors and flow rates can be further improved. Indeed, several layered GO membranes behave as either nanoporous (method one) or molecular-sieving membranes (method two) by the stacking method. Both membranes show CO2-philic permeation behavior, which is further enhanced by the presence of water. As such, these membranes are promising materials for industrial CO2 separation processes related to petrochemical engineering (CO2 removal from natural gas), the environment (CO2 capture from flue gas), and biomass energy (CO2 recovery from landfill gas).

Supplementary Materials

Materials and Methods

Figs. S1 to S10

References (31, 32)

References and Notes

  1. Acknowledgments: This work was mainly supported by Korea Carbon Capture and Sequestration Research and Development Center (KCRC) (grant no. 2012-0008907) funded by the Korea government (Ministry of Science, Information Communication Technology and Future Planning). We thank the National Synchrotron Light Source at Brookhaven National Laboratory for providing the X9 beam line (grant no. DE-AC02-98CH10886) and B. D. McCloskey for thoughtful reading and commentary on our paper. H.B.P. also acknowledges support by the research fund of Hanyang University (grant no. 200900000001052).

Stay Connected to Science

Navigate This Article