Flow sensors

Must be able to quantify flow rates on the fly.

This is challenging because of the small amounts of fluid
being pumped.
Consider a moderately extreme example:
Flow channel 10 m by 10 m, with an averaged flow
speed of 100 m/sec.
Volume flow rate: 0.6 nL/min (!).
Things only get more challenging as devices shrink.
Main approaches:
Differential pressure
Drag force
Thermal flow

Differential pressure flow measurement

Many microfluidic flows are laminar and pressure driven.
This means that its possible to measure the differential
pressure across a channel and directly infer the flow rate.
For laminar flow,
D = hydraulic diameter
P
=u Re f = 4 x Area / perimeter
L 2D 2 f = friction coefficient
Re f = 64 for circular channels,
50-60 for rectangular.
For the example on the previous slide, D = 10 m; for water,
= 1 x 10-3 kg/m-s, Re f ~ 50.
So, P/L = 500 Pa/m, and a 100 m long channel would mean
a pressure difference of 0.05 Pa = 5 x 10-7 atm.
This is pretty tough to do, but it can be done.

1
Differential pressure flow measurement
One approach (with some useful formula) is a capacitive measurement.
Thin disk-shaped transducer of
radius R. For Youngs modulus E
and Poisson ratio , deflection vs. r r
given by 3(1 2 ) 2
y (r ) (R r ) p
2 2

16Et 3
Integrating up to get change in capacitance for initial spacing d,

C (1 2 ) 4 So, for d = 100 nm, t = 100 nm, R = 100 m,

R p ~ 0.25, E = 170 GPa, p = 0.05 Pa, we get
C 16 Edt 3
around a 1.7 % change in C.
Certainly measurable, though not trivially.
Alternative: piezoresistive measurement.

Nguyen and Wereley

Drag force
Because of viscosity, laminar flow around an object (cantilever)
(through a gap of area Ag) exerts a drag force on the object that
has two components:
C L C
Fd = F1 + F2 = 1 Q& + 2 Q& 2
Ag A0 Ag

pure viscous effect pressure drop effect

Low flow rates: deflection ~ linear in volume flow rate.

High flow rates: deflection ~ quadratic in volume flow rate.

Basic idea is to construct a flexible obstacle, and use some means

(e.g. piezoresistive) to measure its deflection.
Clearly, low flow rates require a very flexible transducer
(nanoscale thicknesses).

2
Drag force
Nguyen and Wereley

Several examples of drag sensor geometries.

Thermal approaches
Idea here is to use thermal transport to sense mass transport.
Common approaches:
Time-of-flight (heater + remote thermometers)
Heater temperature at constant heater power
Heater power at constant heater temperature
Latter two give electrical signals, since heater temperature is
sensed by looking at the resistance.
h = heat transfer coefficient (Power per unit area per
temperature difference = W/m2-K)

3
Thermal approaches

Relevant dimensionless quantities:

cp
Prandtl number: Pr momentum diffusion / thermal diff.
T
hd
Nusselt number: Nu total heat transfer / conductive part
T

So, for heaters, if a heater power W is being applied, then the

temperature offset of the heater from the nearby fluid is given by
Nu T AT
W=
L
Can back out flow speed because, for laminar flow over flat
plates and wires,
Nu 0.664 Re 3 Pr

Mixers
Weve said before that one major issue for using microfluidic
systems to do useful reactions, assays, etc. is the slowness of fluid
mixing when flows are laminar.
In strict laminar flow, mixing at the molecular scale is governed
by diffusion.

Diffusion constant for

most liquids ~ 10-9 m2/s. Kenis et al., Science 285, 83 (1999)

So, typical distance over which mixing is accomplished in a

time goes like d ~ 2 D

How good is diffusive mixing if we have two parallel streams

5 microns wide in our original 100 micron long, 100
micron/sec flow channel? Pretty good - d for 1 sec is ~ 50
microns, many times the width of the flow channel. Often,
though, were not so lucky.

4
Mixers
So, one sensible approach is to divide the flow stream into as
many narrow parallel streams as possible!
Then one only needs diffusion over ~ the width of one stream.
Ordered way to do this: hydrodynamic focusing.

Knight et al., PRL 80, 3863 (1998).

Univ. of Southampton
Mixers

Whitesides, Harvard

5
Mixers
Another way to do this is by chaotic mixing.
Turbulence is effective at mixing because streams break up into
eddies on length scales going down to the very small, for which the
diffusion process weve been discussing is rapid.
Need to mimic this in a laminar system.

Ridges on
bottom of
flow channel
can cause
chaotic
motion in
cross-section.

Stroock et al., Science 295, 647 (2002)

Mixers Stroock et al., Science 295, 647 (2002)

By making less regular patterns of ridges, can produce chaotic flow

patterns laterally.
These sorts of techniques can lead to much more rapid mixing.

6
Integration capabilities
While its early yet for the field of microfluidics (first use of
the term: c. 1983) , rapid progress has already been made in
terms of large scale integration of microfluidic devices.
Key idea: multiplexing Thorsen et al., Science 298, 580 (2002)

Instead of trying to
individually control
every valve, use a clever
binary scheme to direct
fluid.
Valves here work by
pneumatically pressing
down a thin membrane
of PDMS.

Integration capabilities Thorsen et al., Science 298, 580 (2002)

Heres an example
of a fluidic
memory, with
1000 chambers
(each ~ 250 pL)
that are
individually
addressible.
Think about
combinatorial
surveys for
analytical
chemistry or
biology.

7
Integration capabilities
Thorsen et al., Science 298, 580 (2002)

With this approach, can purge individual chambers one at a time with a
buffer solution.
An example of this control is at the right.

Integration capabilities
Thorsen et al., Science 298, 580 (2002)

These capabilities can permit assay strategies that would otherwise

be extremely difficult.

8
Integration capabilities
Thorsen et al., Science 298, 580 (2002)

An example of this:
Left-hand chambers filled with a
dilute solution of E. coli
containing a particular enzyme.
Typical occupancy of left
chambers is ~ 20% with single
cells, rest empty.
Right hand chambers filled with
indicator that reacts with
enzyme to form fluorescent
product.
Can check individual cells, but
in parallel this way.

Applications of microfluidics
Weve already touched on a number of applications for
microfluidic technology (e.g. inkjet printers).
Far and away, the fields with the most interest right now in
these techniques are analytical chemistry and (molecular and
cellular) biology.
Analytical chemists foresee means of automating massive
combinatorial surveys.
The real advantages over bulk techniques:
Can get away with using much less analyte (tiny reaction
volumes)
May be integrated and automated into a small, stand-
alone package much more readily.

9
 Applications of microfluidics
Another possibility is
performing chemical
reactions under precisely
controlled conditions in a
very quiescent environment.
Example: protein crystal
growth
Can use integrated
microfluidics to explore a
huge parameter space of
crystallization conditions
while using a miniscule
amount of protein (which
can be very hard to isolate
and purify).

Hansen et al., PNAS 99, 16531 (2002)

Applications of microfluidics

As for biology, a couple of

examples.
Using the rotary (peristaltic)
pump at right, a solution
containing beads tagged with a
fluorescent analyte were pumped
past binding sites patterned in the
large channels.
Result: could get binding 60x
faster with active pumping than
without.
Again, speeds up surveys and
assays.

Quake and Scherer, Science 290, 1536 (2000)

10
Applications of microfluidics
Takayama et al., Nature 411, 1016 (2001)

Can control fluid flows on length scales smaller than cellular sizes.
Great for tagging certain organelles and watching intracellular
transport.

11
Limits and applications of fluids at the nanoscale

Two particular areas of concern:

Do hydrodynamic assumptions fail at the nanoscale?
What are the consequences?
Does the continuum approximation underlying our
treatment of fluid mechanics fail? What are the
consequences?
Also, how do we prepare and study nanoconfined fluids?
Of what use is nanofluidics?
Do hydrodynamic assumptions break down at the nanoscale?

Yes - weve already discussed the no-slip boundary condition

for fluids at liquid-solid interfaces.
The no-slip condition is an idealization assuming moderately
strong attractive interactions between the fluid particles and
the walls.
Clearly, if we can tune the surface chemistry or otherwise (e.g.
electrowetting) alter the interfacial energy, one has to wonder
about the no-slip situation.

Important because sometimes we care about the details of

hydrodynamic flow on the nanometer scale!
The demise of the no-slip condition
You have two papers on your problem set that deal with
experiments claiming to measure deviations from the no-slip
condition on nanometer scales.
Heres another approach:

Take two atomically smooth (mica-coated) crossed cylinders

immersed in fluid, and oscillate their spacing (d) vertically.
A hydrodynamic viscous force will act to damp this motion.
Solving the N-S equations with no-slip condition,

d&
FH = f 6R
* 2

d
Demise of the no-slip condition
d&
FH = f 6R
* 2

d
So, if one works with an apparatus that vibrates at a particular
frequency and measures the peak force, one can define a
parameter
6R 2 d
G =
FH , peak

The experiment: measure the force as a function of cylinder

separation d for a variety of fluid combinations (different
wetting conditions as determined by contact angle
measurements).
No-slip: a plot of G vs. d should extrapolate to 0 as d goes to 0.
Demise of the no-slip condition

Upper case: tetradecane

on mica (wetting =
circles) and on a methyl-
terminated SAM (contact
angle = 44 degrees =
diamonds).
Lower case: water on
SAM-coated mica.
Dashed lines = no-slip
predictions.
Insets: curves supposed
to extrap. to zero.

Zhu et al., PRL 87 096105 (2001)

Demise of the no-slip condition
So, for poorly wetting interfaces, there appears to be slippage.
Can quantify this by a slip length:
u(z)

Slip length = fictitious distance inside the solid where liquid

velocity extrapolates to zero.
With this simple assumption, can then analyze data to get slip
length and see how it depends on, e.g., flow rate.
Demise of the no-slip condition

Two trends are observed in the

data:
The higher the would-be
induced shear stress, the more
slip.
Deviation from no-slip
happens at lower effective
shear stresses in poorly wetting
surfaces than in well-wetting
surfaces.
Still an active topic of
investigation!

Zhu et al., PRL 87 096105 (2001)

When should the continuum approximation break down?

One can define the Knudsen number for a fluid:

l mean free path for particle

Kn
L characteristic dimension of interest

In a gas, a high Knudsen number means that flow is in the

molecular regime - particles are vastly more likely to scatter
off the walls of the container, for example, than each other.
This means that using continuum hydrodynamics to describe
many of the gas properties is not appropriate.
In a liquid, Kn approaching 1 implies that the fluid is confined
on scales comparable to the interparticle separation (since the
mean free path is that short).
How should the continuum approximation break down?

This gets to the heart of our definitions of a homogeneous liquid.

One typical aspect of this definition is the idea that the density-
density correlation function of a liquid has a particularly simple
form:
S
( 0) ( r )
S (r )
( 0) ( 0)

In a liquid, average r
relative positions of other
particles can be quite typical interparticle
variable - no long range separation
order.
In a crystalline solid, relative positions take on very specific values.

Under confinement (or within molecular scales of interfaces),

expect to see signs of ordering.
Example: nontrivial changes in viscosity

Same basic geometry as no-slip experiment described

above, but reduce spacing between cylinders toward the
molecular scale.
NOTE: requirement of charge neutrality + surface charge
effects at interfaces mean that ionic concentration in fluid
when spacing gets small becomes much larger than in bulk
(e.g. 15 M even when starting with ~ deionized water).
At low frequencies, try to oscillate sliding of one cylinder
past the other
Can measure amplitude and phase of resulting oscillations,
and convert into some effective viscoelastic modulus (real part
like shear modulus; imaginary part like viscosity).
Nontrivial changes in viscosity:
Zhu et al., PRL 87 096104 (2001)

Nothing too odd happens

when spacing is larger than
~ 1 nm.
Data at right show expected
drag forces for given
experimental conditions;
dashed lines are in line with
expectations of bulk fluid
behavior.

Note that something happens at very small distances.

A number to recall: typical thickness of a water molecule is 0.25 nm.
Nontrivial changes in viscosity: Zhu et al., PRL 87 096104 (2001)

First indications of
weirdness: at small (< 2 nm)
separations, relative
orientations of mica crystals
lead to impressively varied,
frequency-dependent
response!

Water apparently orders

between the two crystals, and
whether the surfaces are
commensurate or not
significantly modifies the
water ordering and its elastic
response.
Example: nontrivial changes in vibrational properties
Can confine water into nanometer-sized regions using micelles
(ordered membrane of surfactant molecules that closes on itself)

As water is increasingly
confined and begins (?) to
order, one might expect a
change in its vibrational
properties.
Also, since water is polar and
has a reasonable dielectric
response, this could be probed
with some kind of EM
scattering technique.
Univ. of Twente, Netherlands

Note: the vibrational frequencies were talking about end up being

in the terahertz regime.
Nontrivial changes in vibrational properties
Boyd et al., PRL 87, 147401 (2001)

Note that absorption coefficient varies dramatically, and absorption

peak pushes toward higher frequencies with increasing confinement.
Nontrivial changes in vibrational properties
Boyd et al., PRL 87, 147401 (2001)

The water nanopools become vibrationally stiffer with increasing

confinement!
Evolution of elastic properties is due to molecule-molecule
ordering and interactions.
Example: nontrivial changes in density
Hueberger et al., Science 292, 905 (2001)

Use fact that thin mica is

optically transparent to do
another optical study of
nanoconfined fluid.
Cyclohexane: weak charge
effects (nonpolar, few ions).
With this approach, the
optical spectroscopy is
sensitive to ~ attoliter
volumes (!).
Can do correlations across
multiple frequencies to arrive
at fluctuations in density and
mica spacing as load is
varied.
Nontrivial changes in density
Hueberger et al., Science 292, 905 (2001)

When greatly confined, thickness of film fluctuates with time.

Thickness has certain preferred values for given loading.
Index of refraction (and therefore density) also really fluctuates!
Challenges: nanoscale effects on ion concentration
Challenges: nanoscale effects on ion concentration

Can use interplay between

walls and ionic concentrations
to control ionic conduction in
solutions!
Sometimes hydrodynamics works better than expected!
Burton et al., PRL 92, 244505 (2004)

Use electrical resistance to monitor diameter of constriction as

mercury drops pinch off from each other.
Sometimes hydrodynamics works better than expected!
Burton et al., PRL 92, 244505 (2004)

Scaling analysis in limit of surface tension and inertia

dominating: 1/ 3
2
d

Here, is the time remaining until pinch-off.

2 r cot( )
R=
d
Here, r is the electrical resistivity, and it turns out that
approaches 18 degrees universally in this limit (!).
Can pass current through the neck, and use voltage across a
series resistor as an effective ammeter, and can trigger scope
accordingly.
Sometimes hydrodynamics works better than expected!
Burton et al., PRL 92, 244505 (2004)

Result of all this:

Simple hydrodynamic picture works all the way down to neck

diameters ~ 5 nm.
Applications of nanoscale fluid flow

We see that under true nanoscale confinement, liquids may no

longer act much like liquids! Can develop properties that look
much more like those of solids, and exhibit significant dynamic
effects.
Why are people interested in fluid flows at these scales?
Two quick applications:
Sorting of macromolecules (electrophoresis)
Sorting of micro/nanoparticles
Sorting of macromolecules
Sorting macromolecules by length can
be a painful process - is there any way
we can let nanoscale fluid dynamics
make our lives easier?

Yes - viscous flow through long

channels tends to stretch out these
molecules.
Tricky bit: want channels narrow
enough that only one molecule fits per
channel, but also need to find way for
coiled molecule to get itself straight and
in there.
Answer: entropic sieve.
Cao et al., APL 81, 3058 (2002)
 Sorting of macromolecules

Can make nanoscale

enclosed channels by clever
combination of large area
patterning and kinetics of
sputtering.
Scale bars = 500 nm.

Cao et al., APL 81, 174 (2002)

Sorting of macromolecules
By having a gradient of
mesoscale obstacles, molecules
explore more microstates - mor
likely to end up in a configuration
that allows them to enter the
nanoscale channels.

Cao et al., APL 81, 3058 (2002)

Sorting of micro/nanoparticles

Laminar flow around asymmetrically positioned obstacles divides

asymmetrically.
Can take advantage of this to separate particles by size.

Particle too big for lane 1 ends up getting shunted laterally.

Huang et al., Science 304, 987 (2004)

 Sorting of micro/nanoparticles

Can keep lattice spacing and

registry of obstacles constant and
vary the spacing.
Use fluorescently tagged
microspheres of different
diameters to watch the sorting.

Huang et al., Science 304, 987 (2004)

Summary

Microfluidics is a rich and exciting area of research, and is just

really getting going.
Potential chemical and biological applications are huge.
Fundamental physics, physical chemistry and surface science
questions arise when considering fluid really confined on the
nanoscale.