You are on page 1of 6

LETTERS

PUBLISHED ONLINE: 13 NOVEMBER 2017 | DOI: 10.1038/NMAT5025

Ultrathin graphene-based membrane with


precise molecular sieving and ultrafast
solvent permeation
Q. Yang1,2,3,4, Y. Su1,2*, C. Chi1,2, C. T. Cherian1,2, K. Huang1,2, V. G. Kravets3, F. C. Wang5,
J. C. Zhang6, A. Pratt6, A. N. Grigorenko3, F. Guinea3,7, A. K. Geim3 and R. R. Nair1,2*

Graphene oxide (GO) membranes continue to attract intense development of novel inorganic membranes for OSN is particu-
interest due to their unique molecular sieving properties larly vital because of the known instability of many polymer-based
combined with fast permeation19 . However, their use is membranes in organic solvents. The possible lack of motivation
limited to aqueous solutions because GO membranes appear for exploiting graphene-based membranes for OSN could have
impermeable to organic solvents1 , a phenomenon not yet fully come from previous reports on the impermeability of organic sol-
understood. Here, we report efficient and fast filtration of vents through sub-micrometre-thick GO membranes that remained
organic solutions through GO laminates containing smooth highly permeable for water1,2,20 . Although some later studies report
two-dimensional (2D) capillaries made from large (1020 m) the swelling of GO membranes in organic solvents and, accord-
flakes. Without modification of sieving characteristics, these ingly, indicate the permeability of organic molecules even through
membranes can be made exceptionally thin, down to 10 nm, thick GO membranes21,22 , this seems inconsistent with the previous
which translates into fast water and organic solvent perme- reports1,2,20 and could be explained by the presence of extra defects
ation. We attribute organic solvent permeation and sieving that produce a molecular pathway. In another work23 OSN was
properties to randomly distributed pinholes interconnected performed using a solvated reduced GO-polymer composite mem-
by short graphene channels with a width of 1 nm. With brane and only achieved a molecular sieve size of approximately
increasing membrane thickness, organic solvent permeation 3.5 nm due to the nanochannels in the membrane being larger
rates decay exponentially but water continues to permeate than that of pristine GO membranes1,2,5 . Molecular rejection for the
quickly, in agreement with previous reports14 . The potential above membranes involves charge-specific separation rather than a
of ultrathin GO laminates for organic solvent nanofiltration physical size cut-off. Membranes with -size precise sieving along
is demonstrated by showing >99.9% rejection of small with high organic solvent permeance are of great interest for OSN
molecular weight organic dyes dissolved in methanol. Our technology; however, a demonstration of such is still lacking. In
work significantly expands possibilities for the use of GO this report, we investigate permeability and sieving properties of
membranes in purification and filtration technologies. ultrathin GO membranes with respect to organic solutions using
Membrane-based technologies enable efficient and energy- an improved laminar structure, and demonstrate the membranes
saving separation processes which could play an important role potential for OSN.
in society with applications in purifying water or harvesting green Figure 1 shows scanning electron microscope (SEM) and atomic
energy10,11 . Recently, it was shown that molecular separation pro- force microscope (AFM) images and X-ray diffraction (XRD) of the
cesses could benefit from the development of graphene-based mem- GO membranes (Methods). Short-duration ultrasonic exfoliation
branes24 that show tunability in pore size8,1214 and ultimate per- and stepwise separation (Methods) were used to obtain large GO
meance14 defined by their thinness. In particular, GO-based mem- flakes (lateral size D of 1020 m) with a relatively narrow size
branes are considered to be extremely promising for molecular sepa- distribution (Supplementary Fig. 1). The membranes prepared from
ration and filtration applications due to their mechanical robustness these large GO flakes are referred to as highly laminated GO
and realistic prospects for industrial-scale production24,7,9 . Con- (HLGO) membranes due to their superior laminar structure. They
siderable progress in nanofiltration through GO membranes24,15 show a narrow XRD peak (full-width at half-maximum of 0.4 ) as
was achieved mainly for water (due to its ultrafast permeation14 ), compared to 1.6 for the standard GO membranes prepared from
whereas organic solvent permeation has received limited attention. smaller flakes (D 0.10.6 m). Below the latter are referred to
This disparity is rather surprising as organic solvent nanofiltration as conventional GO (CGO) membrane. The narrow X-ray peak
(OSN) attracts tremendous interest due to its prospective appli- for HLGO laminates suggests the importance of the GO flake size
cations in the chemical and pharmaceutical industries11,1619 . The for the interlayer alignment process, which can be attributed to

1
National Graphene Institute, University of Manchester, Manchester M13 9PL, UK. 2 School of Chemical Engineering and Analytical Science, University of
Manchester, Manchester M13 9PL, UK. 3 School of Physics and Astronomy, University of Manchester, Manchester M13 9PL, UK. 4 Key Laboratory of
Advanced Technologies of Materials (Ministry of Education), School of Materials Science and Engineering, Southwest Jiaotong University, Chengdu,
Sichuan 610031, China. 5 Chinese Academy of Sciences Key Laboratory of Mechanical Behavior and Design of Materials, Department of Modern
Mechanics, University of Science and Technology of China, Hefei, Anhui 230027, China. 6 Department of Physics, University of York, York YO10 5DD, UK.
7
Imdea Nanociencia, Faraday 9, 28015 Madrid, Spain. These authors contributed equally to this work. *e-mail: yang.su@manchester.ac.uk;
rahul@manchester.ac.uk

NATURE MATERIALS | ADVANCE ONLINE PUBLICATION | www.nature.com/naturematerials 1


2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
LETTERS NATURE MATERIALS DOI: 10.1038/NMAT5025

a b Interlayer distance ()
14.7 9.8 7.4 5.9 4.9
10

Height (nm)
5 8 nm

Intensity (a.u.)
0 250 500
Distance (nm)

CGO

HLGO

6 9 12 15 18
2 ()

Figure 1 | Ultrathin HLGO membrane. a, SEM image of an 8-nm-thick HLGO membrane on an Anodisc alumina support. Scale bar, 1 m. Inset: SEM image
of bare alumina support. Scale bar, 500 nm. b, X-ray diffraction for HLGO and CGO membranes. Inset (left): AFM image of HLGO membrane transferred
from an alumina substrate to a silicon wafer. Scale bar, 500 nm. Inset (right): the height profiles along the dotted rectangle.

stronger interlayer interactions between larger overlapping areas24 . The high permeance of organic solvents combined with accurate
The stronger interactions could further assist in eliminating the molecular sieving makes ultrathin HLGO membranes attractive
occasional wrinkles and corrugation found in CGO membranes2,3 , for OSN16,17 . To evaluate this potential for applications, we
and this could lead to achieving smoother 2D capillaries in have performed filtration experiments with methanol solutions
HLGO membranes. of several dye molecules. The dye molecule rejections for an
To probe the molecular sieving properties of HLGO membranes 8-nm-thick HLGO membrane are presented in Fig. 2c. Although
(Fig. 2a), we first performed vacuum filtration of aqueous solutions the permeance was reduced by only 1030% compared to
of several salts and large molecules through HLGO membranes the pure solvent (which is not unusual for nanofiltration15 ), no
(Methods). Similar to micrometre-thick GO membranes5 , HLGO dye molecules could be detected down to 0.1% (our detection
membranes also block all ions with hydrated radii larger than limit) of the feed concentration at the permeate side (Fig. 2c).
4.5 . We emphasize that no molecular sieving was observed in The observed 100% dye rejection and fast solvent permeation
similar experiments but using CGO membranes with a thickness of makes our ultrathin HLGO membranes superior to state-of-the-
850 nm (Fig. 2a inset). Hence, an ultra-sharp sieving cut-off can art polymeric membranes for OSN technology16,18 . A comparison
be achieved in HLGO membranes that are more than two orders of of OSN performance of HLGO membranes and previously reported
magnitude thinner than conventional membranes showing the same OSN membranes is shown in Supplementary Section 4 and indicates
sieving properties5 . This drastic improvement can be attributed to that HLGO membranes could be an outstanding candidate for
the highly laminated nature of HLGO membranes. We failed to OSN technology. With the view of practical applications, we have
observe a cut-off in sieving only for membranes thinner than 8 nm, also performed OSN experiments with polymer (nylon) supported
which sets a minimum thickness for the HLGO membranes used in HLGO membranes, and they showed nearly the same performance
this study. as those on alumina supports (Supplementary Section 5 and
Ultrahigh permeance to fluids may occur in ultrathin mem- Supplementary Fig. 6). We have also studied the influences of ageing
branes due to a decreased molecular permeation length6,14 . To fur- and solvent stability of HLGO membranes on their membrane
ther evaluate the liquid permeance of HLGO membranes, we have performance, and found that HLGO membranes are stable in air
performed vacuum filtration and dead-end pressure filtration (Sup- for more than a year and also stable in different solvents (see
plementary Section 2) with water and various organic solvents using Supplementary Section 6 and Supplementary Movie 1).
8-nm-thick membranes. The flux is found to be linearly propor- To elucidate the mechanism of organic solvent permeation
tional to the differential pressure (1P) across the HLGO membrane and sieving properties of ultrathin HLGO membranes, we have
(Fig. 2b inset). The permeance for the solvents as a function of their conducted two sets of additional experiments. First, we have
inverse viscosity (1/) is shown in Fig. 2b. In contrast to much performed XRD for HLGO membranes immersed in different
thicker GO membranes that exhibit ultrafast water permeation and organic solvents (see Fig. 3a). The results indicate that several
impermeability for organic solvents1 , our HLGO membranes are of the organic solvents, especially polar ones, intercalate between
highly permeable to all tested solvents. The highest permeance is GO layers and increase the interlayer distance, d. However, non-
observed for solvents with the lowest viscosity. For example, hexane polar solvents, such as hexane, the fastest permeating molecule
shows a permeance of 18 l m2 h1 bar1 , that is, a permeability of (Fig. 2b), did not produce any increase in d. This suggests that
144 nm l m2 h1 bar1 , despite its kinetic diameter being almost permeation through ultrathin HLGO membranes is not dominated
twice as large as that of water25 . In contrast, 1-butanol with a kinetic by molecular transport through interlayer capillaries1 . Second, we
diameter similar to that of hexane25 , but much higher viscosity, performed water and organic solvent permeation experiments using
exhibits the lowest permeance of 2.5 l m2 h1 bar1 . The linear HLGO membranes of different thicknesses, h. Figure 3b shows the
dependence of permeance on 1/ (see Fig. 2b) indicates that the exponential decay for methanol and hexane permeance as a function
solvent viscosity dictates its permeability and proves the viscous of h. (Exponential decay is also observed with gas and vapour
nature of the solvents flow through HLGO membranes. permeation, see Supplementary Section 7). HLGO membranes

2 NATURE MATERIALS | ADVANCE ONLINE PUBLICATION | www.nature.com/naturematerials

2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
NATURE MATERIALS DOI: 10.1038/NMAT5025 LETTERS
a b c
K3[Fe(CN)6] RB

Flux (l m2 h1)
100 20 9 100 9
Na4PTS /MB BB 10 RB

Permeance (l m2 h1 bar1)
Permeance (l m2 h1 bar1)
6 CG DR
100 CV BB
8 3 8

Rejection (%)
Rejection (%)

9 95 6

(l m2 h1 bar1)
MB
Rejection (%)
0.4 0.6 0.8 1.0

Permeance
100
10-hexane 20

(l m2 h1 bar1)
50 P (bar)

Rejection (%)
50 4 7

Permeance
10 9-acetone
6 8-acetonitrile 50 10
7-Tetrahydrofuran 3
NaCl 0 0 5 6-methanol 90
15 35 50 3 0 0
CGO membrane 2 5-butylacetate 15 35 50
4 4-water
thickness (nm) 1 CGO membrane
0 MgCl2 3-ethanol thickness (nm)
1-butanol 2-isopropanol
0 85 0
4 6 8 0 1 2 3 4 300 600 900
Hydrated radius () 1/ (mPa s)1 Molecular weight (g mol1)

Figure 2 | Molecular sieving and organic solvent nanofiltration through HLGO membranes. a, Experiments for salt rejection as a function of the ions
hydrated radius (largest ions within the aqueous solutions are plotted). The HLGO membranes are 8 nm thick. The hydrated radii are taken from refs 5,7.
MB, methylene blue; RB, rose Bengal; BB, brilliants blue. Inset: MB rejection and water permeance exhibited by the standard GO membrane with different
thicknesses (colour coded axes). b, Permeance of pure organic solvents through an 8 nm HLGO membrane as a function of their inverse viscosity. The used
solvents are numbered and named on the right. Inset (top): methanol permeance as a function of pressure gradient (1P). Dotted lines: best linear fits. The
1P dependence has been examined by both vacuum filtration and dead-end filtration and is found to be the same (Supplementary Section 2). c, Rejection
and permeance of several dyes in methanol versus their molecular weight (colour coded axes). The dyes used: chrysoidine G (CG), disperse red (DR), MB,
crystal violet (CV), BB and RB. Left inset: photographs of dyes dissolved in methanol before and after filtration through 8 nm HLGO membranes. Right
inset: MB rejection and methanol permeance of CGO membrane with different thicknesses (colour coded axes). Note that even though the dye rejection
increases and approaches 90% with increasing CGO membrane thickness their permeance is significantly lower than 8 nm HLGO membranes. All the
error bars are standard deviations using at least three different measurements from different samples. Points within the grey bar in a,c show the rejection
estimated from the detection limit (Supplementary Fig. 4 and Methods).

a Interlayer distance () b
14.7 9.8 7.4

0.4
Acetontrile
Methanol

Hexane
Isopropanol

Acetone
Ethanol

Dry GO
Hexane
Butanol

(l m2 h1 bar1)
Water

10 Permeance 0.3
THF

Permeance (l m2 h1 bar1)

0.2

0.1
Intensity (a.u.)

3 6 9
1 1/thickness
(103 nm1)

Water
Methanol
0.1

6 9 12 0 50 100 150
2 () Thickness (nm)

Figure 3 | Probing molecular permeation through HLGO membranes. a, X-ray diffraction for 70-nm-thick HLGO membranes immersed in various organic
solvents (colour coded). b, Thickness dependence of permeance for methanol, hexane, and water through HLGO membranes (colour coded). Red and blue
dotted lines are the best exponential fits. The black dotted curve is a guide to the eye. Inset: water permeance as a function of inverse thickness for
HLGO membranes with thicknesses 100 nm. Dotted line: best linear fit. The slope of linear fit provides the water permeability as approximately
32 nm l m2 h1 bar1 . The solid line in the main figure shows the detection limit for methanol and hexane in our experiment. All the error bars are
standard deviations using at least three different measurements from different samples.

with h > 70 nm show no detectable solvent permeation, consistent pathways for permeation through HLGO membranes. The first
with the impermeability reported for sub-micrometre-thick GO involves permeation through pinholes (pathway 1) and the second
membranes1 . In contrast, water permeance initially also decayed one is through the previously suggested model of a network of
exponentially, but for h > 70 nm it followed a linear dependence on graphene capillaries1,5 (pathway 2).
1/h (Fig. 3b inset). Pinholes in GO membranes originate from random stacking of
The exponential decrease of organic solvent permeance with h is individual GO flakes and can also involve nanometre-size holes2
surprising and seemingly contradicts the viscous flow inferred from within flakes. At thickesses of a few nanometres, GO laminates
the observed 1/ dependence. Indeed, the viscous flow suggests contain many pinholes (Supplementary Section 8) that pierce
that the permeance should be proportional to the pressure gradient through the entire film. Such thin GO films allow relatively
(1P)/L, where 1P is the driving pressure gradient and L is the easy permeation through pinholes without the atomic-size cut-
permeation length (proportional h) (refs 26,27). To explain these off observed for thicker laminates. At a certain critical thickness
two functional dependences, we propose two different molecular hc , GO films become continuous with all pinholes blocked, as

NATURE MATERIALS | ADVANCE ONLINE PUBLICATION | www.nature.com/naturematerials 3


2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
LETTERS NATURE MATERIALS DOI: 10.1038/NMAT5025

the found onset of atomic-scale sieving indicates. The experiment enhance the nanofiltration properties of GO membranes by cation-
shows that for HLGO membranes, hc is 8 nm. After hc , molecular crosslinking is enticing, but further research is needed to optimize
transport is expected to occur in two steps. Liquids continue to the performance.
rapidly fill the same pinholes, but this is not a limiting process.
Instead, molecular transport is limited by the necessity to travel from Methods
one pinhole to another, which involves in-plane diffusion between Methods, including statements of data availability and any
the GO sheets. This bottleneck involves interlayer diffusion by a associated accession codes and references, are available in the
distance on the order of the size of the GO sheets, which will online version of this paper.
provide atomic-scale sieving. Assuming that the probability for a
molecule to find a pathway through the continuous membrane with Received 1 March 2017; accepted 5 October 2017;
thickness hc is p, we can write the probability of transport through published online 13 November 2017
a thicker membrane of thickness h as P = pN , where N = h/hc .
This can be re-written as P = exp[ln(p)h/hc ] and yields the flux
Q exp(h/a), with a = hc / ln(1/p). By definition p should be References
1. Nair, R. R., Wu, H. A., Jayaram, P. N., Grigorieva, I. V. & Geim, A. K.
on the order of 1/2 because we define it at the threshold, which Unimpeded permeation of water through helium-leak-tight graphene-based
means that P 1 for h < hc and P 0 for thicker layers. Therefore, membranes. Science 335, 442444 (2012).
a = hc / ln(1/p) hc , in agreement with the exponential fit in Fig. 3b. 2. Sun, P., Wang, K. & Zhu, H. Recent developments in graphene-based
This proposed model could also explain why the molecular sieve size membranes: structure, mass-transport mechanism and potential applications.
of approximately 4.5 is not preserved in thin CGO membranes, Adv. Mater. 28, 22872310 (2016).
where the smaller flake size increased the critical thickness and 3. Liu, G., Jin, W. & Xu, N. Graphene-based membranes. Chem. Soc. Rev. 44,
50165030 (2015).
weakened the interlayer alignment. 4. Fathizadeh, M., Xu, W. L., Zhou, F., Yoon, Y. & Yu, M. Graphene oxide: a novel
The deviation of water permeance from the exponential decay 2-dimensional material in membrane separation for water purification.
and its faster transport at large h can be understood by considering Adv. Mater. Interfaces 4, 1600918 (2017).
the molecular pathway 2, where the permeation occurs through 5. Joshi, R. K. et al. Precise and ultrafast molecular sieving through graphene
the graphene capillaries1,5 between GO sheets (typically, an area of oxide membranes. Science 343, 752754 (2014).
4060% remains free from functionalization28,29 ). The permeation 6. Li, H. et al. Ultrathin, molecular-sieving graphene oxide membranes for
selective hydrogen separation. Science 342, 9598 (2013).
through pathway 2 is primarily restricted by the hydraulic resistance 7. Akbari, A. et al. Large-area graphene-based nanofiltration membranes by shear
due to a large L((D/d) h) (ref. 1). However, water permeation alignment of discotic nematic liquid crystals of graphene oxide. Nat. Commun.
through these capillaries experiences three orders of magnitude 7, 10891 (2016).
enhancement due to the large slip length1,8,26,27 , and therefore 8. Abraham, J. et al. Tuneable sieving of ions using graphene oxide membranes.
reduces the flow resistance. This suggests that, with increasing h, Nat. Nanotech. 12, 546550 (2017).
the exponentially growing flow resistance for water in pathway 9. Hong, S. et al. Scalable graphene-based membranes for ionic sieving with
ultrahigh charge selectivity. Nano Lett. 17, 728732 (2017).
1 could be overcome by the lower flow resistance in pathway 2, 10. Mulder, J. Basic Principles of Membrane Technology (Springer Science &
consistent with the deviation from the exponential decay observed Business Media, 2012).
above 50 nm in Fig. 3b, where, instead, a linear dependence 11. Koros, W. J. & Zhang, C. Materials for next-generation molecularly selective
is observed (Fig. 3b inset). In contrast, for organic solvents, the synthetic membranes. Nat. Mater. 16, 289297 (2017).
experimentally undetectable permeation for h > 70 nm indicates 12. Wang, L. et al. Molecular valves for controlling gas phase transport made
the permeation through pathway 2 is negligible and suggests a from discrete ngstrm-sized pores in graphene. Nat. Nanotech. 10,
785790 (2015).
non-slip flow. This is not surprising because graphitic surfaces 13. Jain, T. et al. Heterogeneous sub-continuum ionic transport in statistically
are known for their lipophilicitythat is, they interact strongly isolated graphene nanopores. Nat. Nanotech. 10, 10531057 (2015).
with hydrocarbons. This is consistent with the recent calculation 14. Celebi, K. et al. Ultimate permeation across atomically thin porous graphene.
of larger interfacial friction for ethanol in graphene capillaries Science 344, 289292 (2014).
compared to water30 . The non-slip behaviour of organic solvents 15. Han, Y., Xu, Z. & Gao, C. Ultrathin graphene nanofiltration membrane for
also explains why certain organic molecules (polar solvents) can water purification. Adv. Funct. Mater. 23, 36933700 (2013).
16. Marchetti, P., Jimenez Solomon, M. F., Szekely, G. & Livingston, A. G.
uniformly intercalate between GO layers, similar to water, but show Molecular separation with organic solvent nanofiltration: a critical review.
little permeability through thicker membranes. To confirm this Chem. Rev. 114, 1073510806 (2014).
model further, we have also ruled out the influence of polarity, 17. Vandezande, P., Gevers, L. E. & Vankelecom, I. F. Solvent resistant
dynamic diameter, and solubility parameters of organic solvents on nanofiltration: separating on a molecular level. Chem. Soc. Rev. 37,
their permeance (Supplementary Section 9). 365405 (2008).
Based on the understanding of organic solvent permeation 18. Jimenez-Solomon, M. F., Song, Q., Jelfs, K. E., Munoz-Ibanez, M. &
Livingston, A. G. Polymer nanofilms with enhanced microporosity by
through GO membranes, we further increased the permeance interfacial polymerization. Nat. Mater. 15, 760767 (2016).
(50%) without substantially reducing the organic solute rejection 19. Karan, S., Jiang, Z. & Livingston, A. G. Sub10 nm polyamide nanofilms with
by using partially reduced Mg2+ -crosslinked GO membrane ultrafast solvent transport for molecular separation. Science 348,
(Supplementary Section 10). We believe that the performance 13471351 (2015).
of such ion-modified membranes could be further improved by 20. Huang, K. et al. A graphene oxide membrane with highly selective molecular
optimizing the crosslinking and reduction process. separation of aqueous organic solution. Angew. Chem. Int. Ed. 53,
69296932 (2014).
In conclusion, we show that HLGO membranes of only 21. Huang, L., Li, Y., Zhou, Q., Yuan, W. & Shi, G. Graphene oxide membranes
several layers in thickness exhibit outstanding sieving properties with tunable semipermeability in organic solvents. Adv. Mater. 27,
accompanied by ultrafast solvent permeation. The proposed model 37973802 (2015).
based on non-slip permeation of organic solvents and slip- 22. Aba, N. F. D., Chong, J. Y., Wang, B., Mattevi, C. & Li, K. Graphene oxide
enhanced water permeation offers a long-sought explanation for membranes on ceramic hollow fibersmicrostructural stability and
the ultralow permeability of sub-micrometre-thick GO membranes nanofiltration performance. J. Membr. Sci. 484, 8794 (2015).
23. Huang, L. et al. Reduced graphene oxide membranes for ultrafast organic
for organic solvents. Taking into account the excellent chemical solvent nanofiltration. Adv. Mater. 28, 86698674 (2016).
stability of GO, the reported membrane can be used for organic 24. Lin, X. et al. Fabrication of highly-aligned, conductive, and strong graphene
solvent nanofiltration, with pharmaceutical and petrochemical papers using ultralarge graphene oxide sheets. ACS Nano 6,
industries being potential beneficiaries. The proposed strategy to 1070810719 (2012).

4 NATURE MATERIALS | ADVANCE ONLINE PUBLICATION | www.nature.com/naturematerials

2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
NATURE MATERIALS DOI: 10.1038/NMAT5025 LETTERS
25. Wu, H., Gong, Q., Olson, D. H. & Li, J. Commensurate adsorption of for X-ray measurements, and K. Huang for assisting in setting up the cold trap for
hydrocarbons and alcohols in microporous metal organic frameworks. filtration experiments.
Chem. Rev. 112, 836868 (2012).
26. Secchi, E. et al. Massive radius-dependent flow slippage in carbon nanotubes. Author contributions
Nature 537, 210213 (2016). R.R.N. and Y.S. designed and supervised the project. Q.Y., Y.S. and C.C prepared the
27. Radha, B. et al. Molecular transport through capillaries made with atomic-scale samples, performed the measurements and carried out the analysis with help from R.R.N.
precision. Nature 538, 222225 (2016). C.T.C. and K.H. helped in sample preparation, characterization and data analysis. J.C.Z.
28. Wilson, N. R. et al. Graphene oxide: structural analysis and application as a and A.P. contributed to XPS characterization. V.G.K. and A.N.G. contributed to optical
highly transparent support for electron microscopy. ACS Nano 3, measurements. F.G., F.C.W. and A.K.G. contributed to theoretical modelling. Y.S.,
25472556 (2009). A.N.G., A.K.G. and R.R.N. wrote the manuscript. All authors contributed to discussions.
29. Loh, K. P., Bao, Q., Eda, G. & Chhowalla, M. Graphene oxide as a chemically
tunable platform for optical applications. Nat. Chem. 2, 10151024 (2010). Additional information
30. Dai, H., Liu, S., Zhao, M., Xu, Z. & Yang, X. Interfacial friction of ethanolwater Supplementary information is available in the online version of the paper. Reprints and
mixtures in graphene pores. Microfluid. Nanofluid. 20, 141 (2016). permissions information is available online at www.nature.com/reprints. Publishers note:
Springer Nature remains neutral with regard to jurisdictional claims in published maps
and institutional affiliations. Correspondence and requests for materials should be
Acknowledgements addressed to Y.S or R.R.N.
This work was supported by the Royal Society, Engineering and Physical Sciences
Research Council, UK (EP/K016946/1), Lloyds Register Foundation, and European
Research Council (contract 679689). Q.Y. acknowledges support from the China Competing financial interests
Scholarship Council. We thank P. Bentley for assisting with XPS measurements, J. Waters The authors declare no competing financial interests.

NATURE MATERIALS | ADVANCE ONLINE PUBLICATION | www.nature.com/naturematerials 5


2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
LETTERS NATURE MATERIALS DOI: 10.1038/NMAT5025

Methods Permeation and molecular sieving measurements. To probe molecular sieving


Preparation of GO membranes. Graphene oxide was prepared by the Hummers and solvent permeation through various GO membranes we used a vacuum
method and then dispersed in water by sonication1 , which resulted in stable GO filtration set-up, where the membrane is clamped and sealed with a silicone rubber
solutions. It is worth noting that, to avoid any other possible influence, we O-ring between the feed and permeate side. For each test, at least three membranes
repeatedly washed our GO nanosheets until the pH value of their solutions were used to validate the reproducibility. The permeate side was connected to a
reaches 7. GO membranes were prepared by vacuum filtering aqueous vacuum pump with a controllable pumping speed and a cold trap. The vacuum on
GO solutions through either an Anodisc alumina or a nylon membrane5 the permeate side creates a pressure gradient (1P) which drives the molecular
(47 mm diameter Whatman filters with 200 nm pore size). To obtain a uniform permeation across the membrane. The vacuum level on the permeate side was
membrane, the GO suspension was diluted to less than 0.001 wt% before the controllable from 0.6 to 0.01 bar (VARIO chemistry diaphragm pump,
vacuum filtration. After filtration, the membrane was allowed to dry under vacuum Vacuubrand) and the feed pressure was approximately 1 bar. To study the influence
at room temperature for at least 24 h before the measurements. The membranes of 1P on the permeance, we have performed filtration experiments with different
with different thickness were obtained by filtrating different volumes of GO 1P created using different pumping speeds. The permeance of various solvents
suspension through the alumina or nylon support. It is noteworthy that the was obtained by measuring both the volume and weight of the solvent from the
influence of the support membrane on the reported permeation was minimal due permeate side in a liquid nitrogen cold trap and the liquid leftover in the feed side.
the large porosity. The solvent permeance through both the bare alumina and All the permeance values were recorded after reaching a steady state condition,
nylon support layers was found to be a minimum of 1,000 times larger than the typically achieved within 30 min. The system leakage was examined by replacing
GO membrane on support layers, suggesting hydraulic resistance from the support the membrane with a 100 m polyethylene terephthalate plastic sheet, or a 200 m
layers was negligible and could be ignored. Cu foil, the leakage was found to be <0.1 l m2 h1 bar1 . The solvent permeance
The two types of GO membranes used in this study are HLGO and CGO through GO membrane was also measured by a dead-end pressure filtration system
membranes. The difference between the preparation of HLGO and CGO at room temperature and good consistency was found between the two methods
membranes lies in the ultrasonic exfoliation and centrifugal separation process. For (Supplementary Section 2).
HLGO membranes, the graphene oxide was exfoliated by a 3-minute ultrasonic We have noticed that for water, due to its high surface tension, the HLGO
exfoliation (40 W power) and then subsequently centrifuged twice at 3,000 r.p.m. membrane breaks once the water comes into contact with the membrane.
for 10 min to separate un-exfoliated thick GO flakes. The supernatant GO solution Therefore, we used a small amount of surfactant (0.6 mg ml1 sodium dodecyl
was further centrifuged at 12,000 r.p.m. to separate large and small GO flakes. In benzene sulfonate) to decrease the surface tension of water and thereby avoid
this step, the sediment was collected because the small size, and hence lighter, GO membrane damage during the water permeation experiments.
flakes remain in the supernatant and larger GO flakes sediments. This sediment To probe the molecular sieving property of HLGO and CGO membranes we
was then collected and re-dispersed in water by mild shaking, followed by repeated used aqueous solutions of NaCl, MgCl2 , K3 [Fe(CN)6 ], pyrenetetrasulfonic acid
centrifugation steps at 10,000 and 8,000 r.p.m., respectively. These repeated tetrasodium salt (Na4 PTS), MB, RB, and BB. For MB, RB, and BB the feed
centrifugation cycles with sequentially decreasing centrifugation speeds enable concentrations were 20 mg l1 , and for K3 [Fe(CN)6 ] and Na4 PTS, the
theseparation of medium size GO flakes from the large flakes and allows one to concentrations were 1,000 and 250 mg l1 , respectively. For NaCl and MgCl2 we
obtain the uniform large GO flakes required for the preparation of HLGO used 1M concentration. All the experiments were repeated at least three times. The
membranes. For the preparation of CGO membranes, the graphene oxide in water amount of sodium and magnesium salts permeated were measured by probing the
was sonicated for 24 h and then centrifuged three times at 8,000 r.p.m. The concentration of salt in the permeate side by checking the conductivity of the
supernatant was then collected and used for the membrane preparation. permeate water. Furthermore, we cross-checked the results of our conductivity
It is important to note that the HLGO and CGO membranes are prepared by analysis by weighing the dry material left after evaporation of water in the
identical procedures except for a different exfoliation time. The influence of permeate. The permeation of other salts and dyes through GO membranes was
which on the chemical composition of GO sheets was carefully examined by measured by checking their concentration at the permeate side by
X-ray photoelectron spectroscopy (XPS) and found to give no difference in ultravioletvisible (UVvis) absorption, as detailed below. The salt rejection was
oxygen content between two membranes (Supplementary Section 1 and calculated as (1 CP /CF ), where Cp is the salt concentration at the permeate side
Supplementary Fig. 2). and CF is the salt concentration at the feed side.
The flake size distribution of GO used for the preparation of CGO For organic solvent nanofiltration experiments, CG, MB, DR, CV, BB, and RB
and HLGO membranes was measured by analysing more than 700 flakes with with a concentration of 200 mg l1 were dissolved in methanol. The concentration
scanning electron microscopy (SEM) or optical microscopy. Due to the of the dye at the permeate side was measured by UVvis absorption as detailed
long-duration ultrasonication, all the GO flakes used for the CGO membranes below and the permeance was determined by the same method for the
are found to be smaller than 1 m in nominal size, with more than 75% of measurement of pure solvent as detailed above.
these flakes having a size in the range 0.10.4 m. In comparison, for HLGO
membranes, 75% of the flakes used were found to be larger than 10 m UVvis absorption. To obtain the concentrations of K3 [Fe(CN)6 ], Na4 PTS and
(Supplementary Fig. 1). organic dye molecules in the permeate we used optical absorption spectroscopy.
A UVvisiblenear-infrared grating spectrometer with a xenon lamp source
Membrane characterizations. SEM and AFM techniques were used to measure the (2401,700 nm) was used for this study. For the HLGO membranes, we could not
size of the GO flakes and the thickness of the membranes. A Veeco Dimension detect any absorption features of the above salts or dye in the permeate side
3100AFM in the tapping mode was used for the AFM measurements. To measure (Supplementary Fig. 4). To cross-check this further, we have also measured the
the thickness of the GO membranes, we transferred the membrane from the concentration of the leftover feed solution after the filtration experiment. The
alumina support to a silicon substrate by floating the alumina supported GO leftover concentrated feed solutions (including the salt or dye absorbed on the
membrane in water and subsequently fishing out the GO membrane onto a silicon membrane) were diluted to the same volume as before the filtration experiment,
substrate. GO membrane transferred silicon substrates were completely dried in then the optical absorption features were compared with the pristine original feed
vacuum before the AFM measurements. solution. We could not find any difference in the absorption spectra, suggesting all
X-ray diffraction measurements in the 2 range from 3 to 25 (with a step size the solutes were retained at the feed side. The detection limits in Fig. 2a,c were
of 0.02 and recording rate of 0.2 s) were performed using a Bruker D8 estimated by measuring a reference solution and gradually decreasing its
diffractometer with Cu K radiation ( = 1.5406 ). Due to the weak intensity of concentration until the signature peaks completely disappeared. The penultimate
the X-ray peak from an 8-nm-thick membrane we used 70-nm-thick membranes concentration is set as the corresponding detection limit. For the case of CGO
for our experiments. To collect an XRD spectrum from HLGO membranes membranes and Mg2+ -crosslinked membranes (Fig. 2a,c, and Supplementary
exposed to different organic solvents, the membranes were first aged in a glovebox Fig. 12), the absorbance for the most intense optical absorption peak for various
filled with dry argon gas for more than five days to remove any interlayer water known concentrations of salt and dye molecules were plotted against their
present in the membranes8 , and then immersed in various solvents for more than concentration and yielded a linear fit. From this linear dependence, we estimated
three days inside a glove box. For the XRD measurements, the samples were the concentration of salt and dye at the permeate side.
collected from the solvents and kept inside an airtight XRD sample holder
(Bruker, A100B36/B37) filled with the same organic solvent vapour to avoid any Data availability. The data that support the plots within this paper and other
influence of the environmental humidity and evaporation of solvent from the findings of this study are available from the corresponding author upon
membrane on the measurements. reasonable request.

NATURE MATERIALS | www.nature.com/naturematerials

2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.