Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences


    Separation and flux performance were compared in graphene-based membranes that differed only in the method of deposition of reduced graphene oxide platelets. Membranes with higher degree of order were produced by evaporation-induced capillary-force self-assembly, which showed higher steric rejection properties while simultaneously accentuating water permeance compared to membranes produced by the traditional vacuum filtration technique. These studies attempt to establish structure–property correlations in graphene-based membranes.

    1. Introduction

    Despite an array of established commercial membrane fabrication techniques (phase inversion, interfacial polymerization, track etching, electro-spinning, etc.) and materials compatible with these methods [1], novel materials and fabrication technologies could lead to membranes with improved permeance and rejection properties. Membranes derived from graphene oxide (GO) and reduced graphene oxide (RGO) have recently received considerable attention [2,3]. Successful preparation of colloidal suspensions of GO [4] and RGO [5] have enabled fluid-phase processing of filtration membranes and aided the ability to deposit thin active layers of GO and RGO on porous polymer supports from colloidal suspensions. To date, vacuum filtration [614] has predominantly been used to cast thin GO/RGO membranes because it suits laboratory-scale operations, but other techniques have recently been implemented, such as layer-by-layer assembly [1517], spray coating [18] and spin coating or casting [18,19] for gas and liquid separations.

    GO films and membranes are unstable in an aqueous environment because of the oxygen functional groups that litter its surface and cause GO to dissolve. Consequently, RGO-derived membranes are more suitable in pressure-driven water filtration applications on account of its water stability. Unlike the pores that exist in polymeric membranes, the unique laminate structure [4] of RGO assemblies imparts nano-filtration characteristics, whereby the interlayer spacing between RGO sheets form nano-channels that can allow the passage of solvent but may withhold solute molecules. Methods to control the microstructure of RGO-derived assemblies to produce membranes with maximal flux and separation efficiencies have been explored by techniques that include thermal treatment [10], chemical treatment (reduction [6,10] and oxidation [12]), varying the experimental conditions (pH/salt concentration/pressure) [7,15], cross-linking of GO sheets [15,17], nano-strand channelling [11] and carbon nanotube (CNT) compositing [13] but not without drawbacks. For example, vacuum filtration cannot easily achieve large membrane areas, whereas altering the RGO synthesis temperature resulted in changes to the corrugation of RGO sheets deposited onto a polymer substrate by vacuum filtration, with improved water permeance response (50–250 l m−2 h−1 bar−1) but at the expense of its dye rejection capabilities (67% for Direct Yellow 50) [10].

    Tailoring membrane microstructural properties by vacuum filtration can also be achieved by intercalating species between GO/RGO sheets, although these techniques involve the relatively arduous synthesis of intercalants. Despite this, the intercalation of copper nano-strands into the RGO membrane structure and their subsequent etching [11] resulted in large water permeance (approx. 600 l m−2 h−1 bar−1) and high rejection (100%) of 5 nm diameter gold nanoparticles. Similarly, incorporation of CNT intercalants [13] likewise altered the micro-scale structure, with a water permeance of 11.3 l m−2 h−1 bar−1 and high salt rejection (83.5% for Na2SO4). Besides vacuum filtration, Hu & Mi [17] showed that chemical cross-linking via layer-by-layer assembly enabled promising dye and salt rejection (93–95% of the bulky dye Rhodamine-WT and 26–46% for Na2SO4, respectively) and reasonable water permeance (approx. 15 l m−2 h−1 bar−1). Kim et al. [19] compared the microstructural characteristics of GO first deposited onto polyethersulfone (PES) substrates by either drop or dip casting and then followed by spin casting. Using gas permeation tests, they found that, by first depositing GO by drop casting, more densely stacked assemblies were produced compared to those first deposited by dip casting. One can easily infer that the deposition technique plays a major role; however, it is not clear whether there is a direct structure–property correlation in these thin-film membrane structures.

    Another technique to produce thin films, evaporation-induced self-assembly, has been successfully used to prepare nanoparticle assemblies [20] and, in particular, induce macroscopic ordering [21,22] while having potential as an industrial-scale thin-film coating technique [20,23]. We recently reported the utilization of evaporation-induced self-assembly for the production of highly ordered RGO films on glass [24]. The formation mechanism of these RGO films was found to be governed by capillary-force-assisted self-assembly (CAS) [24] and demonstrated significant conductivity enhancement compared to films produced by vacuum filtration, on account of their densely packed and well-ordered structure. We have also demonstrated the use of orientation-mapped polarized light imaging to quantify the molecular order and fine structure of graphene-based liquid crystalline assemblies and films [24,25]. Herein, we show that this structure–performance correlation can be translated to nano-filtration membranes, whereby polymer-supported, ordered CAS RGO membranes demonstrate superior rejection characteristics for dye probes and superior permeance compared to less-ordered vacuum filtration membranes.

    2. Materials and methods

    GO was synthesized by the modified Hummers method [26,27]. Briefly, pre-treatment of 10 g Bay Carbon graphite (SP1 grade) was accomplished by addition to 50 ml H2SO4 heated to 90°C and dissolved with K2S2O8 (5 g) and P2O5 (5 g) under stirring. The pre-treatment mixture was stirred for 4.5 h at 80°C, after which time it was diluted with deionized (DI) water and left overnight. The pre-treated graphite was washed by vacuum filtration (nylon 0.2 μm Millipore membranes) and then left overnight to dry. H2SO4 (230 ml) was cooled under stirring to 0°C by ice bath and the pre-treated graphite added. KMnO4 (30 g) was slowly added until dissolved, with the temperature kept below 10°C, and then the slurry was left for 2 h to react at 35°C. The slurry was again stirred for 2 h after slow addition of DI water to the oxidized mixture. After further dilution, H2O2 (25 ml of 30% solution) was added and kept undisturbed for at least 24 h. The supernatant was then expelled and the remaining oxidized material was washed with HCl and water until the pH became neutral. Following mild sonication, the resulting material was designated GO.

    RGO colloidal suspensions were synthesized from GO according to the recipe of Li et al. [5]: 50 μl of N2H4 (35 wt%) and 350 μl ammonia was added to a 100 ml, 0.025 wt% GO suspension. The solution was shaken and placed in a 95°C oil bath for 1 h.

    CAS membranes were produced by the vertical immersion of a polymer support into a dilute, aqueous RGO colloid, which, upon evaporation, deposits RGO on the exposed surface of the polymer (figure 1). CAS films were prepared on hydrophilic nylon substrates (nylon 66, Supelco). Nylon membranes (47 mm diameter) were cut in half and each half sandwiched along its long edge between cut glass rectangular pieces of sufficient length to span a 200 ml beaker. A suspension of 5 μg ml−1 RGO was prepared from the RGO stock suspension by appropriate dilution with DI water. The nylon was mounted vertically inside the filled beaker and covered with a parafilm net with sufficient porosity to prevent dust, etc., from collecting inside the beaker and to slow the evaporation rate of the suspension so that a continuous layer of RGO could be formed. Once the membrane was fully exposed to air, it was removed for further testing and characterization. The evaporation rate was 6.4 ml d−1 (0.26 ml h−1). The mass loading of RGO onto nylon by CAS was too small to be determined by a mass balance. Therefore, to assess the mass loading of RGO onto nylon by CAS deposition, ultraviolet–visible (UV-vis) spectrometry was used. A calibration curve for RGO in water (electronic supplementary material, figure S1) was used to ascertain the initial absorbance of the RGO suspension of known volume. Once the CAS film was formed on nylon and dried, the absorbance of the final solution and measurement of the remaining volume are used to determine the mass loss from the RGO suspension and deposited onto the nylon. In calculating the mass loading per unit area, deposits on the interior of the beaker were considered negligible and the deposition area was determined with consideration that RGO is coated onto both sides of the nylon. Once the mass loading of the CAS films was known, films could be prepared by vacuum filtration for comparison with the mass loading controlled by different volumes of RGO suspension of 5 μg ml−1. Those tested here have a mass loading of 61 mg m−2.

    Figure 1.

    Figure 1. Schematic of the techniques used to prepare graphene membranes on nylon supports in this work: (a) evaporative self-assembly (capillary-force-assisted self-assembly or CAS); (b) vacuum filtration, which has been traditionally used for the production of graphene-based thin films and membranes. The dotted arrows show the direction of mass transport for each technique. (Online version in colour.)

    Films were transferred from their polymer (nylon) support onto glass for polarized light imaging by the Abrio LC-PolScope (LPS) system (operating on an inverted Leica DMIRB base) by placing an ethanol and acetone cleaned, cut square piece of glass slide at the bottom of a Petri dish filled with sulfuric acid. A small (approx. 0.5 mm square) cut piece of the nylon-coated films was added nylon-side down and soaked for 4 h, during which time the polymer dissolved, revealing a thin floating layer of RGO. The sulfuric acid was slowly pipetted out and the floating RGO film carefully manoeuvred over the top of the glass slide, after which the residual sulfuric acid was carefully diluted with DI water. Care was taken not to drop water directly on top of the floating RGO film because water dropped directly atop its surface caused the as-yet non-adhered film to wrinkle and fold undesirably. Because the film was observed to still be floating on the glass surface, it was placed on a hot-plate at 50°C for an hour to remove excess water, after which time the adhered film could be cleaned with DI water and ethanol.

    Membrane filtration characteristics were tested in a dead-end pressure cell (electronic supplementary material, figure S2) at 3 bar with three probe dye molecules: small cationic Methylene Blue (MB), small anionic Orange G (OG) and large anionic Direct Yellow 50 (DY) (table 1). Blue-Tac was used to remove one side of the RGO deposited on the membrane so that the rejection and permeance characteristics were those of a single side of deposited RGO only. Permeance was calculated by measuring the volume of permeate, V , over a time, t (hours):

    Display Formula
    where A is the membrane active area and ΔP was the applied pressure (3 bar). To ensure that any adsorption was exhausted and because the permeate volumes were usually very small, the system was left for more than 10 h and the final steady-state permeance and rejection were found after that time. The rejection was calculated by
    Display Formula
    where the initial concentration C0 of the feed and the permeate concentration C of the dye probes was measured using an Ocean Optics USB4000 UV-vis spectrometer.

    Table 1.Details of the dye probes used for testing the rejection of RGO membranes.

    dye molecular structure Inline Formula (nm) MW (g mol−1) size (Å) charge
    Methylene Blue (MB) Inline Graphic 665 373.9 4.0×7.9×16.3 [28] cationic
    Orange G (OG) Inline Graphic 485 452.4 5.4×10.1×15.6 [29] anionic
    Direct Yellow 50 (DY) Inline Graphic 395 956.8 4.0×10.7×23.8 [29] anionic

    3. Results and discussion

    The presence of RGO loaded onto polymer supports is evidenced by their grey colour, while scanning electron microscopy (SEM) shows that CAS and vacuum filtration membranes at equivalent mass loads both have thickness approximately 100 nm (figure 2). SEM also shows that the top surfaces of the CAS membranes appear less rough than the vacuum filtration membranes (see also electronic supplementary material, figure S3, for cross-sections). Atomic force microscopy (AFM) of the top surface of each membrane confirms this observation (electronic supplementary material, figure S4). Fourier transform infrared spectroscopy (FTIR) (see electronic supplementary material, figure S5) of the deposited layer shows peaks associated with residual oxygen functional groups characteristic of RGOs [5]. CAS and vacuum filtration membranes share an identical chemical character; therefore, differences in their membrane performance can be attributed solely to physical differences as a consequence of the deposition technique.

    Figure 2.

    Figure 2. SEM images of RGO membranes. The top surface of (a) vacuum filtration membranes exhibit a more rough morphology compared with the top surface of (c) CAS. Vacuum filtration and CAS membranes with an equivalent mass loading also have approximately the same thickness as shown in (b) and (d), respectively. (Online version in colour.)

    Films transferred onto glass after etching away the polymer support were characterized by polarized light microscopy (PLM). In contrast to typical PLM, we use the Abrio LPS system [3032], as was recently used for RGO liquid crystal characterizations [24,25] (see also the electronic supplementary material). The LPS system obtains accurate PLM images of every point within the view-space simultaneously, thereby enabling efficient characterization of the optical character. In particular, we investigate the azimuth, a characteristic parameter of optically anisotropic materials that describes the angle of maximum transmittance. The hue represents the azimuth angle in the central panes of figure 3, which shows the orientation mapped PLM image from the LPS. The CAS membrane exhibits a more uniform hue compared to the vacuum filtration membranes, indicative of molecular-scale order in the former and relatively less order in the latter. This is reflected by the polar depiction of the azimuth orientation vectors, quantitatively represented by the order parameter (see the electronic supplementary material). An order parameter S=1.00 implies perfect molecular-scale order, whereas S=0.00 implies no order at all. Therefore, CAS membranes with S=0.88 exhibit distinctive molecular-scale order compared to vacuum filtration films with S=0.41. The film transfer process and deposition on the surface of rough membranes may have further reduced the degree of order in both CAS and vacuum films. However, optical images comparing the surface of the membranes prior to transfer onto glass with those transferred onto glass (see electronic supplementary material, figure S6) show no obvious morphological differences apart from large-scale wrinkles that form upon drying of the films after transfer (such a wrinkle is also observed in figure 3).

    Figure 3.

    Figure 3. Bright-field transmitted light images (a,d), the corresponding false-colour LPS processed images (b,e) and their associated polar histograms of the azimuth angles and in-plane order parameter S (c,f) for vacuum filtration (top) and CAS (bottom) films. Films were transferred from their polymer support onto glass for imaging. (Online version in colour.)

    The formation mechanism of assembled films of GO/RGO by vacuum filtration has been described as a process of semi-ordered accumulation [33], whereby GO/RGO sheets are randomly oriented within a colloid and form a loosely aggregated assembly once deposited onto the substrate by vacuum. As excess solvent is removed in the final stages of filtration, the loosely aggregated mass is compressed perpendicular to the net direction of solvent flow; however, this process does not completely align the resultant film. Consequently, vacuum filtration produces irregularly stacked RGO assemblies on a polymer support [4,33]. In contrast, thin and ordered CAS films form as a consequence of diffusive mass transport to the triple-phase (air, substrate and colloid) contact line, with self-assembly occurring via alignment by capillary forces at the liquid meniscus and subsequent assembly by van der Waals interactions between adjacent RGO sheets [24] as the liquid meniscus recedes.

    Pure water permeance was determined from the variation of flux versus pressure (figure 4a) for both membranes. A linear variation in flux with pressure is observed over the range of pressure tested, with a permeance of 1.7 l m−2 h−1 bar−1 for vacuum filtration membranes and 7.9 l m−2 h−1 bar−1 for CAS membranes determined from the gradient. Nair et al. [18] compared their experimental results of water vapour flux through thin GO films to theoretical flux estimations from the Hagen–Poiseuille equation. In doing so, they calculated the so-called flow enhancement factor, which is defined as the ratio of the measured flux to the theoretical flux calculated by the Hagen–Poiseuille equation. The flow enhancement factor has previously been used to quantify flow enhancement in CNT membrane pores resulting from slip flow of water at the hydrophobic CNT surface [34] and, more recently, has been adopted to describe anomalously fast permeance by graphene-based membranes [6,11,18]. Acknowledging the fact that fluid flow in these nano-capillaries may not conform to continuum mechanics, we similarly compare our experimental results to the theoretical value. Following the procedure first employed by Nair et al. [18] for GO/RGO membranes, the theoretical water permeance through RGO nano-channels can be estimated by using the classical solution to the Navier–Stokes equation for viscous, laminar volumetric flux through a channel comprising parallel plates (i.e. RGO sheets):

    Display Formula
    Figure 4.

    Figure 4. (a) Pure water flux variation with pressure and (b) the rejection character for three organic dyes for CAS and vacuum filtration membranes. (Online version in colour.)

    Here h is the interlayer (channel) spacing between graphene sheets, L is the lateral size of a GO/RGO sheet, a is the effective length of the channel, η is the solvent viscosity (0.001 Pa s for water) and ΔP is the driving pressure between the ends of the length of the pore/channel. The length of the channel in terms of the thickness of the RGO membrane, Δx, can be estimated as

    Display Formula
    which corresponds to flow as depicted in electronic supplementary material, figure S7, in which water would traverse the lateral length of each individual RGO sheet within a stack that has a total thickness of the active RGO layer. Assuming that the area density of channels can be approximated as 1/L2 (i.e. the density when one channel can be described as having lateral area L2 and comprises a single channel unit through the length of the membranes thickness, per unit lateral area of the membrane, as shown in electronic supplementary material, figure S7), the following expression for flux Q is derived:
    Display Formula
    Assuming 1 μm lateral size of RGO sheets, the calculated permeance is approximately 10−3 l m−2 h−1 bar−1, which is three orders of magnitude smaller than our experimental results. Defects in the form of tears or holes in the membrane may be discounted based on inspection of the SEM images, on account of the high rejections observed, especially for CAS membranes for which no DY was found in the filtrate, and for the uncompromised permeance of dye solutions irrespective of their rejection (see electronic supplementary material, figure S8). Large water permeance observed in graphene assemblies has been ascribed to either high porosity of the membrane [11] or boundary slip within the hydrophobic, frictionless graphene channels similar to that observed within CNTs [34,35]. Nair et al. [18] and subsequent authors who have since adopted these theoretical approximations for viscous, laminar flow have noted that equation (3.2) does not describe the likely passage of water between stacked RGO sheets because inherent defects and anisotropy in their lateral lengths would increase the porosity of the membrane, which would necessarily decrease the effective channel length and therefore be responsible for the large water permeance. To discount this possibility, the smallest possible channel length is considered, that is, a path directly through the length of the membrane:
    Display Formula
    This flow path is also an unlikely one, even for well-ordered structures such as CAS membranes. However, its consideration provides a useful upper limit in contrast to the lower limit described by equation (3.2). Using equation (3.4), the theoretical permeance is estimated as approximately 10−1 l m−2 h−1 bar−1, still an order of magnitude smaller than the experimental results of either CAS or vacuum filtration membranes. Wei et al. [36] calculated a flow enhancement factor of 227.5 for water flow between graphene sheets separated by 1.7 nm. Based on their simulation data, we could reasonably expect a flow enhancement factor of this order in our RGO membranes with maximum channel size approximately 2 nm (as determined in the discussion of rejection properties below). However, the result reported by Wei et al. is considerably smaller than the flow enhancements observed for our RGO membranes and also for RGO membranes investigated by Han et al. [6]. Even within the CNT literature, large discrepancies have been reported between experimental and theoretical investigations of slip flow [37,38] which are not yet accounted for. Further, there is still considerable debate regarding the prevalence of fast water transport in GO/RGO membranes, whereby the presence of residual oxygen functional groups might hinder the flow of water [36]. Certainly, the degree and type of reduction would play a role in the extent of flow enhancement. Irrespective of these debates, anomalously fast water flux has been attributed to slip-flow enhancement in the case of GO laminates through which unimpeded flow of water was observed [18] and for thin GO and RGO vacuum filtration membranes [6] with results equivalent to ours. Consequently, slip flow appears to be a dominant mass transport mechanism within our RGO membranes analogous to confined, frictionless, hydrophobic channels of CNTs.

    The rejection for dye probes is shown in figure 4b for both membranes (typical UV-vis spectra of the feed and permeate are shown in electronic supplementary material, figure S9). Both CAS and vacuum filtration membranes showed similar trends in their rejection characteristics; DY and MB rejections were large while OG rejection was small. We will first compare OG versus DY, noting that OG is a smaller probe compared to DY, although both are anionic and both are azo dyes having a similar chemical nature (table 1). CAS membranes reject 100% of DY and less than 4% of OG. OG and DY are unlikely to adsorb appreciably within RGO nano-channels because OG and DY are azo dyes, which have been shown to have poor interactions with the graphene surface [39]. OG and DY are also anionic dyes and could also experience electrostatic repulsion at the surface of RGO. Furthermore, the experiments were conducted over a time period sufficient for steady-state conditions to be established, during which time any adsorption would have exhausted. Consequently, the data indicate that size exclusion is a dominant mechanism. The largest dimension of DY is approximately 2.4 nm and for OG is approximately 1.6 nm (table 1). Therefore, it can be estimated that CAS membranes have a maximum channel dimension of approximately 2 nm and that the channel size distribution is relatively narrow. In contrast to the CAS membrane, it is evident that vacuum filtration membranes, which reject approximately 80% DY (i.e. some channels are larger than DY, allowing it to pass through the membrane) and approximately 12% of OG (i.e. some channels are sufficiently small to filter OG), contain a wider distribution of channel sizes.

    Testing with MB shows 100% rejection for the CAS membrane and approximately 63% for vacuum filtration. If size selectivity was the only mechanism for MB, then MB would be expected to pass un-filtered through CAS membranes, since its molecular size is approximately the same as OG. Adsorption of cationic dyes has been reported for GO and RGO materials [40,41] and could occur at oxygen functional groups on RGO by electrostatic interactions or on aromatic regions by van der Waals interactions as a consequence of the planar aromatic structure of MB. Since MB dye is cationic, then the high degree of removal can be attributed to physical adsorption on the graphene surface. Since the mass loading and the RGO stock solution used to prepare all membranes are identical, there should be negligible differences in the amount of adsorption observed; however, there is a significant difference in the rejection capacity between CAS and vacuum filtration. As MB adsorbs within the channel structure of the membranes, their presence could decrease the effective channel size [42,43]. Since CAS membranes exhibit molecular-scale order with well-defined channel structure, the decrease in the effective channel size as a result of MB adsorption would be more obvious, resulting in higher separation efficiency. Finally, we tested the rejection of salts (see the electronic supplementary material, figures S10 and S11). Given that CAS and vacuum filtration membranes are unable to reject the relatively small OG probe molecule, it is not surprising that their rejection of salts is very small.

    We now discuss the observation that the permeance of CAS membranes is larger than for vacuum filtration membranes. It would be expected (such is the case for polymeric membranes) that water would preferentially flow through the larger channels within vacuum filtration membranes, resulting in an overall larger flux compared to CAS membranes. Contrary to this, CAS membranes exhibit an almost fivefold improvement in water permeance compared to vacuum filtration membranes (figure 4a). The previous discussion on theoretical flux determination from the Hagen–Poiseuille equation discounts the possibility of high porosity being responsible for the enhanced flow of water observed in these RGO membranes. Irregular stacking within vacuum filtration membranes could present a more tortuous path to the flow of water, which could decrease the flow compared to CAS membranes. However, this is not likely since both membranes are approximately the same thickness and we can infer from the high rejections of DY by vacuum filtration membranes that their channel size, while certainly larger than the approximate 2 nm channel size of CAS membranes, is only marginally larger. Therefore, tortuosity would not differ significantly between the two membranes.

    Another possibility is that RGO sheets of CAS membranes may not be assembled entirely flat. This is different from corrugation, which has been reported for individual RGO sheets [10], but is a consequence of the self-assembly process whereby folds, as alluded to by Tkacz et al. [24], may arise during the formation of CAS films. Such folds would increase the porosity of the CAS membranes, and thereby increase their permeance, while the mismatch of peaks and troughs between adjacent folds within the membrane would create small pores, which could physically block dye molecules while allowing relatively free passage of solvent. There is no evidence of such folds in PLM and SEM images and it is unlikely that such folds would be arranged in a regular manner so as to produce the sharp size cut-off observed for CAS membrane rejection.

    Another more likely possibility arises from consideration of slip flow within the RGO membranes. It has been reported that larger flow enhancement factors are expected as the graphene interlayer distance decreases [36]. Therefore, disordered vacuum filtration membranes with a wide channel size distribution would exhibit a range of flow enhancement factors, including negligible or non-existent flow enhancements for the largest channels. It is difficult to definitively estimate the channel size distribution in RGO membranes produced by vacuum filtration. However, vacuum filtration membranes show smaller rejection of DY compared to CAS, and therefore it is reasonable to assume that a high proportion of their channels are more than 2 nm, which would decrease the degree of flow enhancement. In contrast, the ordered nano-channels within CAS membranes would exhibit persistent flow enhancement, which would accentuate the permeance. Therefore, our experimental results provide evidence that significant flow enhancement is prevalent within RGO membranes and, further, that a regular ordered interlayer structure accentuates the flow in accordance with the presence of a slip-flow mechanism. Certainly, more experimental and theoretical investigations of flow enhancement of water confined within planar nano-channels are needed to gain further insights.

    4. Conclusion

    We have shown that controlling the stacking of RGO sheets, by using an evaporative assembly technique, results in nano-filtration membranes that have distinct channel size (approx. 2 nm under these conditions) and exhibit large water permeance. By comparison with traditional vacuum filtration, the molecular-scale order of CAS membranes is found to accentuate the flow of water within the graphene channels (figure 5). This work paves the way towards high-performance pressure-driven membranes, whereby the assembly conditions are an important consideration. Novel evaporation-induced techniques could be controlled by various means (RGO colloid concentration, particle size, functionalization, pH and evaporation rate; and for the substrate: the roughness, hydrophilicity, chemistry and withdrawing speed) and can be further optimized. Finally, this work highlights the importance of tailoring the molecular-scale structure to specific applications. Moreover the technique could be extended to other two-dimensional materials that have also seen recent application as nano-filtration membranes [44,45], particularly as the preparation of their colloidal suspensions for materials processing facilitates their use within evaporation-based self-assembled architectures.

    Figure 5.

    Figure 5. Schematic showing the effect of RGO molecular-scale order on the membrane performance. In (a), the large DY molecules (in yellow) cannot be totally blocked by vacuum filtration membranes on account of the disordered stacking, which indicates a broad channel size distribution. In (b), CAS membranes have regular molecular-scale order, which prevents DY from passing through the membrane and is therefore totally rejected. Water molecules (in blue) are able to pass through both membranes, with water permeance governed by slip flow between graphene sheets. (Online version in colour.)

    Authors' contributions

    P.S. and M.M. developed the concept. P.S. designed and performed the experiments, conducted the characterizations and analysed and interpreted the data. P.S. and M.M. wrote the manuscript. M.M. supervised the project.

    Competing interests

    We have no competing interests.


    The authors acknowledge support from the Australian Research Council through the ARC DP 110100082 and ARC LP 110100612 schemes.


    The authors acknowledge the use of facilities within the Monash Centre for Electron Microscopy and the facilities and assistance of Dr Alex Fulcher at Monash Micro Imaging. The authors also thank Dr Rachel Tkacz, Dr Dhanraj Shinde and Mr Abozar Akbari for useful discussions during the preparation of the manuscript.


    One contribution of 11 to a Theo Murphy meeting issue ‘Nanostructured carbon membranes for breakthrough filtration applications: advancing the science, engineering and design’.

    Published by the Royal Society. All rights reserved.