Professional Documents
Culture Documents
Fluid Flow in Carbon Nanotubes and Nanopipes
Fluid Flow in Carbon Nanotubes and Nanopipes
Fluid Flow in Carbon Nanotubes and Nanopipes
M. WHITBY AND N. QUIRKE arrays when assembled as nanoporous membranes, are discussed
in later sections.
Chemistry Department, Imperial College, South Kensington, Our understanding of the static properties of liquids in
London SW7 2AZ, UK. macroscopic capillaries is based on the nineteenth century work
e-mail: n.quirke@imperial.ac.uk of Laplace, Poisson and Young2. Steady-state flow of simple
incompressible fluids in a channel width 2h, driven for example
Understanding and controlling the flow of liquids at the nanoscale by gravity ρg or a pressure gradient dP/dy, can be described by
is currently a subject of great interest. More than 100 papers the Navier Stokes equation3. The solution for the velocity in the
describing work in nanofluidics have been published in the past direction of flow, y, as a function of the distance from the wall,
five years. Recent experimental measurement of fluid transport z, has a parabolic profile (Fig. 2) given by:
through channels with widths of a few nanometres has confirmed
predictions derived from computer simulations that such flow Uy(z) = (ρg/2η)[(δ + h)2 – z2]
occurs at a much greater rate than expected. This discovery has
significant implications, both for our understanding of how fluids where η is the viscosity and δ, the slip length, is the distance into
behave at very small length scales and for the design of nanofluidic the wall at which the velocity extrapolates to zero. Conventionally,
devices. Here we review selected papers with an emphasis on fluid the slip length is assumed to be zero. However, if it is not then
flow through the central pores of carbon tubes. we are said to have slip boundary conditions at the wall, which
The practical investigation of fluid flow in nanoscale channels has for very small channels can significantly enhance the fluid flow.
been facilitated by the availability of tubular carbon structures with Integrating the velocity profile over the cross-sectional area of the
open central pore diameters in the range of one to several hundred pipe gives the Hagen-Poiseuille law for the flux through a pipe.
nanometres. Carbon offers a number of attractive features for the This law states that the flow rate of a fluid passing through a tube is
fabrication of nanofluidic devices including: the existence of several directly proportional to the pressure difference between the tube
well studied processes for self-assembly of tubular structures; useful ends and to the fourth power of the tube’s internal radius. The
electronic properties for sensing and signalling; biocompatibility, flow rate is inversely proportional to the tube length and to the
ready chemical modification to allow functionalization; and low- viscosity of the fluid.
friction transport thanks to graphitic surfaces. The dynamics of capillary filling were elucidated by Washburn4,5
Two types of nanoscale carbon tubes have been used using the Hagen-Poiseuille law with the driving force for flow given
experimentally, each with distinct properties and methods by the Laplace equation for the pressure difference across the
of synthesis. The first are carbon nanotubes, both single and invading liquid meniscus. Thus the penetration length L at time t of
multiwalled, with the fullerene molecular structure reported by a fluid in a capillary of radius R is given by:
Ijima in 1991 (ref. 1). The second are carbon nanopipes produced
by chemical vapour deposition (CVD) of amorphous carbon in L2 = (Rγ/2η)t
alumina templates with a honeycomb pore morphology (Fig. 1).
Throughout this review we reserve the term ‘nanotube’ to refer with γ being the liquid/vapour surface tension and η the shear
to true molecular carbon structures. The term ‘nanopipe’ will viscosity of the invading liquid. In deriving the Washburn equation
be used to describe carbon pipes produced in templates using it is assumed that there is no fluid motion at the wall (that is, stick
CVD. The latter are generally larger and composed mainly of boundary conditions). The Washburn equation is very successful
amorphous rather than well ordered graphitic carbon. The on time scales sufficient to establish viscous flow but breaks
properties and fabrication of both types of carbon tube, and their down for short times. A more general solution can be found for
b a
20 μm
1 μm
2 μm
500 nm
Figure 1 Scanning electron microscope images of carbon nanopipes produced by the authors using standard chemical vapour deposition. a, Nanopipes partially
released from an anodic aluminium oxide template following sonication in NaOH. b, Cross section of intact carbon coated membrane. c, Higher magnification view of
individual aligned carbon pipes. d, Surface of carbon membrane showing open pores (diameter ~160 nm). The growth procedures are described in the section on “Flow
through nanoscale channels including carbon nanopipes”.
non-steady viscous incompressible flow (the Bosanquet equation6), that the smaller the value of α, the greater the slip length, a direct
which approaches the Washburn equation at long times. relationship between the two only exists at low density10.
For nanoscale capillaries the flow behaviour is dominated not
by bulk properties (γ, η) but by the interaction of the fluid with SIMULATION AND THEORY OF NANOSCALE FLOW
the capillary walls. In this regard it is important to include the
role of surface friction due to molecular corrugation, especially A central question in nanofluidics concerns the extent to which
for graphitic surfaces where the high surface density of atoms the classical equations describing the way fluids interact with
leads to a very low corrugation. A convenient parameter here is nanomaterials hold at the nanoscale. Molecular simulation is
the Maxwell coefficient7 α, which represents the fraction of the ideally suited to shed light on this problem and has shown that
molecular collisions with the capillary walls that undergo diffuse static properties such as surface tensions and contact angles
scattering and/or trapping–desorption, the remainder being of simple fluids obey classical relations down to almost single
specularly reflected (no energy loss). The Maxwell coefficient nanometre dimensions11. Steady-state Poiseuille flows maintain a
has been shown to provide a useful description of molecular parabolic velocity profile down to widths of several nanometres12.
friction8 for a variety of materials9. However, although it is clear Deviations from this behaviour can typically be linked to density
inhomogeneities in the confining walls.
In performing simulations relevant to experimental work on
+h
flow in nanopores, a major challenge has been to choose realistic
potentials between the molecules in the fluid and in the pore walls.
Although it has proved possible to be reasonably confident about,
Uy(z) say, determining alkane/carbon potentials by fitting to heats of
z adsorption, for fluids such as water considerable uncertainty has
arisen. For example, reports of the measured contact angle of water on
y
a graphitic surface differ by as much as a factor of two13. The choice of
intermolecular potential determines whether a model nanotube can
be filled with water14,15 or not. However recent work has shown how it
may be possible to calibrate solvent–nanotube potentials using Raman
shifts in solution16. The uncertainty in water potentials notwithstanding,
some general features of nanoscale flow have been determined from
molecular dynamics using simple fluids relevant to the experimental
work. For example Sokhan et al.8,17 were the first to show that fluids
–h flowing through (slit) carbon nanopores experience very low surface
friction, that is very small Maxwell coefficients (~0.01) or equivalently
large slip boundary conditions, whereas pores made from other
Figure 2 Parabolic velocity profile Uy(z) for Poiseuille flow in a capillary of width 2h. materials show higher values and stick boundary conditions.
600
V m s–1
500
400
300
200
100
0
0 2 4 6 8 10 12
log t (ps)
Figure 3 Imbibition of decane in a (7,7) carbon nanotube (diameter = 0.951 nm). Left: Simulated imbibition, showing a snapshot of the fluid after 60 ps50. Right: plot of the
imbibition velocity obtained from the analytic model using parameters derived from nanoscale simulation, the upper (blue) line shows results for a carbon nanotube, the
bottom (black) line, a non-carbon tube50.
Subsequent theoretical work shows that in a cylindrical geometry, lithographically. The height of the channels varied from 2.7 μm down
decreasing the radius can reduce the Maxwell coefficients by orders to 40 nm. The results show a marked and progressive departure from
of magnitude compared with a slit (~0.000002 for decane in a Poiseuille flow as the channel height falls below 200 nm. This effect
(7,7) nanotube, diameter 0.75 nm) and that carbon nanotubes are was observed most strongly for hexadecane and is also evident for
predicted to be essentially frictionless pipes18,19. Skoulidas et al.20 decane and for silicone oil. Slip lengths were estimated at 25 to 30 nm,
compared transport diffusivities for carbon nanotubes with zeolites ~9 nm and ~14 nm respectively. The effect was not seen for water at
using molecular dynamics and found exceptionally high transport the length scales investigated.
rates for the nanotubes due to the smooth carbon surface. We Several groups have recently reported results from experiments
should expect therefore that crystalline carbon channels (as opposed involving flow of liquids and gases through membranes with
perhaps to amorphous surfaces) show significantly increased flow. nanoscale pores composed of carbon. Two principal methods have
Another consequence is the extremely rapid imbibition of wetting been used to construct suitable membranes. The first involves CVD
fluids (decane is predicted to fill (7,7) nanotubes, at 800 m s–1; ref. 18) of carbon in an anodic aluminium oxide (AAO) template. The second
observed in non-equilibrium molecular dynamics. Note that the is based on production of polymer composites containing an aligned
same model predicts external wetting to be much slower by factors array of carbon nanotubes spanning the thickness of the membrane.
of roughly 30 for a (13,13) tube with diameter 1.765 nm. This has Both techniques allow liquid flow through the central pores of carbon
implications for nanotube arrays, where flow may occur along either nanotubes and nanopipes to be studied without the complication of
the external or internal surfaces of the tubes. external wetting.
From a Bosanquet-like equation (where we approximate the The first method for creating carbon nanomembranes takes
surface friction with an expression involving the Maxwell coefficient advantage of the self-assembly of highly-ordered nanopores arranged
and the surface tension and the driving force is represented by the in a dense hexagonal pattern in alumina when aluminium foil is
wall-wetting liquid tension), it is possible to obtain a solution for the etched in acid23,24. Using this approach, monodisperse-sized pores with
imbibition velocity and penetration length L at all times18. There is an densities as high as 1011 cm–2 are possible. To line the nanopores, the
L ∝ t dependence for short times tending to t1/2 at long times (as in the AAO template is placed in a furnace and a hydrocarbon gas is flowed
Washburn equation), the timescale for which depends on the value for several hours, whereupon carbon is deposited on the hot surface of
of the Maxwell coefficient. Figure 3 shows how the imbibition speed the alumina25. Carbon pipes with an outside diameter as small as 10 nm
for decane varies as a function of time with ultrafast uptake for t<μs have been formed in this way although larger 100 to 200 nm nanopipes
falling to cm s–1 at longer times. are more usual. The length of these pipes is typically 50 to 100 μm.
The situation is more complex for hydrogen bonding fluids such as It is important to appreciate that, as synthesized, the typically
water. In very narrow tubes the water molecules form a one-dimensional 20-nm-thick walls of carbon pipes produced by the AAO templating
chain with linking hydrogen bonds. At equilibrium, conduction method will be at least partly amorphous. Che et al.26 reported electron
through such tubes occurs in bursts14 and this behaviour is attributed diffraction studies in a transmission electron microscope (TEM),
to density fluctuations at the openings of the tube. which produced a pattern characteristic of non-aligned graphite
fragments. There are reports that high-temperature annealing can
FLOW THROUGH NANOSCALE CHANNELS INCLUDING CARBON NANOPIPES improve the order of the tube walls through graphitization27,28,
which may enhance transport properties. Miller et al.25 confirmed
Initial indications of rapid fluid flow through very small channels electro-osmotic flow through these carbon nanomembranes and
in an experimental system were published by Pfahler et al. in reported a linear relationship between electro-osmotic flow velocity
1990 (ref. 21). Working with rectangular channels etched in a and applied current density.
silicon substrate, they found an n-propanol flux approximately
three times greater than expected in the smallest channels (height FLUID FLOW THROUGH CARBON NANOTUBES
0.8 μm and width 100 μm). Following up on this observation in 2002,
Cheng and Gordanio22 reported measurements for Newtonian Hinds et al.29 published the first account of experiments involving fluid
fluids flowing under pressure through nanoscale channels produced flow through a membrane composed of true molecular nanotubes. This
Table 1 Comparison of observed flow velocity with predictions for pressure-driven flow through aligned MWNT membranes30. These results indicate liquid flow
rates four to five orders of magnitude faster than would be predicted from conventional fluid-flow theory. The slip lengths are calculated from observed data. *
Units: cm3 per cm2 at minimum pressure. †Flow velocities in cm s–1 at 1 bar.
Liquid Initial permeability* Observed flow velocity† Expected flow velocity† Slip length (µm)
Water 0.58 25 0.00057 54
1.01 43.9 0.00057 68
0.72 9.5 0.00015 39
Ethanol 0.35 4.5 0.00014 28
Iso-Propanol 0.088 1.12 0.00077 13
Hexane 0.44 5.6 0.00052 9.5
Decane 0.053 0.67 0.00017 3.4
~15º
~5º Water
Vapour 200 nm
200 nm
c d
200 nm 200 nm
e f
Vapour
Water
200 nm 50 nm
Figure 5 ESEM images of the dynamic behaviour of a water plug in a carbon nanopipe. a–e, The meniscus shape changes when the stage temperature is constant but the
vapour pressure of the water in the chamber is changed: a, 5.5 Torr, b, 5.8 Torr, c, 6.0 Torr, d, 5.8 Torr and e, 5.7 Torr (where the meniscus returns to the shape seen in a). The
asymmetrical shape of the meniscus, especially the complex shape of the meniscus on the right side in a and e is a result of the difference in the vapour pressure caused by
the open left end and closed right end of the tube. Estimated contact angles between the meniscus and the tube wall are indicated at two locations. f, TEM image showing a
similar plug shape in a closed carbon nanotube under pressure. Reproduced with permission from ref. 32. Copyright (2004) ACS.
nanoscale and to find low friction coatings or channel materials and will form a random coil with a radius of some 700 nm in aqueous
that help to reduce blocking42. solution at 20 °C. This is several times greater than the pore diameter
Another problem is more fundamental: the large size of polymers, of even large carbon pipes and two orders of magnitude greater than
including biologically relevant molecules such as DNA, which are the diameter of a single walled carbon nanotube such as might be
often tangled and tightly folded in vivo. As Cao et al.41 point out, a functionalized to detect the presence of specific base sequences43. DNA
typical DNA molecule from a virus has a length of 100 to 200 kilobases molecules will fit into the central channel of even small nanotubes,