Multi-Physics Analysis of Microfluidic Devices with

Hydrodynamic Focusing and Dielectrophoresis

M. Kathryn Thompson
Mechanical Engineering Dept, MIT
John M. Thompson, PhD
Consulting Engineer
Abstract
Among the applications for MEMS devices is that of sorting biological specimens. Closed form solutions
fail to provide the necessary means for evaluating these devices, particularly if more than one Physics
environment is involve. This paper provides an overview of some of the lessons learned in the fluidic,
electric and thermal domains.

Introduction
As microfluidic devices become smaller and more complex, fewer options are available for modeling the
behavior these devices. Analytic solutions that provided scoping results when the devices were simple fail
to provide adequate assurance when the devices become complex. And if more than one physics
environment is involved, complexity will be present for the degrees of freedom as well as the geometry.
As a result, the only viable alternative is to use finite element simulation. One of the strengths of the
ANSYS finite element program is its ability to simulate multi-physics systems in both the macro and micro
scale domains. This paper presents some preliminary models for a microfluidic device to demonstrate some
of these abilities for a micromachined dielectrophoretic flow cytometer.
The behavior of the cells as they flow through the system undergoing hydrodynamic focusing can be
determined using Flotran for a fluidic model of the system. The dielectrophoretic behavior of cells present
in the fluid can be found as a result the electric fields generated between the submerged electrodes in the
flow. The Joule heating created by those electrodes and the subsequent temperature rise can be determined
by modeling the thermal behavior of the system. Optical detection methods, device fabrication and
packaging, and the performance of the electrodes and of the device as a whole will not be discussed.

The Phenomena

Flow Cytometry
Flow cytometry is defined as “a technique for identifying and sorting cells and their components (as DNA)
by staining with a fluorescent dye and detecting the fluorescence usually by laser beam illumination.”1 In
the simplest type of device, cells flow past an optical detector in a single file line and a count of cells is
made. If the device is a cell sorter, the output signal from the detector may be used to decide what direction
the cell should take. A secondary signal is then be sent to the sorting mechanism to change the path of the
cell, if necessary. Finally, cells flow out of the device through one of the two outlet channels. In this
example, hydrodynamic focusing uses the addition of a sheath fluid as the pipe narrows to focus the sample
flow into a single file line of cells before they enter the detection region.

Dielectrophoresis
Dielectrophoresis is the movement of dipole particles as a result of the “mechanical forces exerted by a
field on a suspended particle, through interaction with its induced dipole moment.”2 This can be useful for
manipulating polymer particles like latex spheres, as well as with living cells. Cells can experience positive
or negative dielectrophoresis depending on the properties of the cells and their surrounding medium. Those
experiencing negative electrophoresis migrate towards regions of low field strength. Depending on the
strength of the field, the cells are then either trapped at the field minima or they are deflected slightly from
their original course of travel.
In flow cytometers, electrodes can be placed near or in a fluid stream to exert a force on the cells as they
pass the electrodes. If the dielectrophoretic force is sufficient to overcome the drag force on the cell, the
cell will be pushed away from the electrodes. In this way, cells can be moved from the center of the
channel to one side or the other, effectively sorting them into separate output channels (figure 1).

Figure 1. Device Schematic

Cytological Geometric Constraints and Assumptions

It is assumed that the device is made from a moldable polymer like PDMS or SU-8 which requires that
device geometry be approximately a two dimensional extrusion instead of being fully three dimensional.
The manufacturing technology requires that the channels must also be square instead of cylindrical and
concentric fluid flows cannot be achieved. The velocity of the cells flowing through the detection region
was given to be 1 m/s. The scale for the electrode geometry was given and the electrodes are actuated at +/-
5V.

Cytological Constraints
It is important that the living cells in the fluid sample are not permanently damaged or killed in the process
of being manipulated. Mammalian cells should be subjected to no more than 1 Pa shear stress, no more
than 70 mV across the cell membrane and a maximum temperature of no more than 39 degrees C. (Body
temperature is 37 C.) It is also preferable to keep the cells in a biologically compatible medium like a
physiological saline solution or a well regulated sucrose solution.

Modeling in the Fluid Domain

Hydrodynamic Focusing
The use of a sheath flow for hydrodynamic focusing is described in detail by Lee et al.3 Since the geometry
of the focusing region was not given, it was chosen using the velocity and shear stress requirements.
The behavior of the fluid in the focusing region of the microfluidic device is dependent on shear stress,
velocity profile and conversation of mass. The velocity profile at a radius r for a single fluid flowing
through a cylindrical pipe is given by:
 r2
v(r ) = vmax (1 − ) Eq 1
R2
where r is the radial distance from the centerline of a pipe of inner radius R with a maximum fluid velocity
of vmax found at the center of the pipe. The shear stress on a cell in the flow is given by:
du
τ =µ Eq 2
dy
where τ is the shear stress, µ is the viscosity of the liquid, and du is the change in fluid velocity over the
distance dy. The allowable shear stress over the cell, cell diameter and the velocity of the cells are givens,
so equation (2) is solved for du which is then substituted into equation (1) to solve for the minimum inner
radius of the pipe in the detection region.
The geometry of the sheath flow region is entirely dictated by the principle of conservation of mass which
says that the total mass flow rate of all fluid streams entering a control volume must equal the total mass
flow rate of all fluid streams exiting the control volume. Since concentric sheath flow cannot be achieved,
two dimension sheath flow is used. In two dimensional sheath flow, the sample fluid enters the system with
a channel of sheath fluid on either side. These three streams converge and the total width of the channel is
decreased. It is assumed that the same incompressible fluid will be used for the sheath and sample fluids.
Using conservation of mass, the following relationship can be found between the required diameter of the
focused core (which must be the same as the diameter of the cells) and the geometry and initial velocities of
the core and sheath fluids as they enter the system4:
v2 Da
d= Ds = Eq. 3
vcore vD vD
1.5( 1 1 + 3 3 )
v2 D2 v2 D2
where d is the diameter of the focused core, vc is the velocity of the focused core (and also the maximum
velocity of the focused stream), Da is the diameter of the focused stream, v1 is the initial velocity of the top
sheath fluid, v2 is the initial velocity of the sample fluid, v3 is the initial velocity of the bottom sheath fluid
and D1, D2 and D3 are their respective diameters (figure 2). This method of focusing works because the
flow is in the laminar regime (Reynold’s number ~ 45). The streams will converge but will not mix since
diffusion across those boundaries is very slow. Therefore, the cells in the focused stream will stay in their
separate streamlines through the device.

Figure 2. Hydrodynamic Focusing Schematic

For simplicity, the diameters and velocities of the sheath fluids were set equal and the diameters of all of
the inlet fluid channels were set equal to the diameter of the focused stream. This allowed the initial
velocities of the system to be solved analytically. Since the channels are square instead of cylindrical, the
minimum inner diameter of the pipe set to the width and depth of the flow channels.

Horizontal Focusing
A fully parameterized two dimensional finite element model was created using Flotran to demonstrate the
fluid focusing in the horizontal direction. Units for all models in this paper are in the micro MKS system
and the fluid is assumed to be physiological saline. The L2TAN command was used to create the narrowing
region between the inlet and outlet channels of the horizontal focuser; rectangles formed the other areas of
the focuser. FLUID141 elements were used with no slip boundary conditions specified on all fluid channel
walls. The velocities of all three fluid inlets were set as initial conditions and the pressure was set equal to
zero at the outlet. Meshing was done with the automatic meshing tool using a SMRTSIZE of 1.
A velocity vector plot of the horizontal focusing region is shown in Figure 3. The flow becomes fully
developed very quickly after the streams converge and a parabolic velocity profile can be seen at the outlet.
The maximum velocity is approximately 0.06 m/s. This is lower than the expected maximum velocity of
the system because the flow requires one more focusing step (vertical focusing) to achieve a maximum
velocity of 0.1 m/s before entering the detection region.

Figure 3. Vector Plot of Two-Dimensional Horizontal Focusing

Figure 4 shows a plot of the particle trace of the sample fluid flow. Both the initial sample flow channel and
the final flow channel in this model are 35 microns x 35 microns. The width of the focused sample flow is
10 microns as desired.

Figure 4. Particle Trace of Two-Dimensional Horizontal Focusing

Although two dimensional modeling gives a good approximation of the behavior of the fluid in the system,
three dimensional models were created for the system to bring the models closer to reality. The horizontal
focuser model was created by extruding the 2D geometry with the VEXT command. FLUID142 elements
were used with the same initial and boundary conditions. Figure 5 shows the velocity vector plot for the 3D
horizontal focuser. The three dimensional model predicts a maximum velocity of 0.079 m/s which is
significantly higher than the value predicted by the 2D model justifying the decision to create the 3D
models. Figure 6 shows the particle trace for the 3D horizontal focuser. Careful examination of the particle
trace near the outlet shows that the sample flow has been successfully focused in the horizontal direction
but the sample flow still extends over almost the entire width in the vertical direction. The zero fluid
velocity at the walls prevents the sample fluid from extending all the way to the walls.

Figure 5. Vector Plot of Three-Dimensional Vertical Focusing

Figure 6. Particle Trace of Three-Dimensional Vertical Focusing

 Vertical Focusing
A fully parameterized two dimensional finite element model was created using Flotran to demonstrate the
fluid focusing in the vertical direction. In this case, the sheath flow enters the channel through square holes
in the top and bottom of the channel. Instead of converging with the sample flow and entering a narrowed
channel, the sheath flow is entrained by the horizontal flow. The added mass flow rate is sufficient to cause
focusing without the need to reduce the total volume of the system.
The construction of the vertical focusing model was almost identical to the horizontal focusing model.
FLUID141 elements were again used with the no slip boundary condition. The average velocity from the
previous model was used as the inlet velocity of the horizontal stream. The inlet velocity of the vertical
sheath flow was applied to the inlet lines in the model and the pressure was again set to zero at the outlet.
Figure 7 shows a velocity vector plot for the two dimensional vertical focuser. Again, the flow becomes
fully developed very quickly and predicts a maximum velocity of 0.095 m/s. Figure 8 shows a plot of the
particle trace of the sample fluid flow. Again, the sample flow is reduced to a height of 10 microns centered
in the channel as expected.

Figure 7. Vector Plot of Three-Dimensional Horizontal Focusing

Figure 8. Particle Trace of Three-Dimensional Vertical Focusing

The 3D focuser model was created by extending the 3D horizontal focuser model to include the vertical
focusing features. By putting the two models in series, the error of averaging and rounding the velocity of
the sample flow as it leaves the horizontal focuser is eliminated. Figure 9 shows the velocity vector plot of
the 3D focuser. The maximum predicted velocity is 0.12 m/s or just above the desired maximum. Figure 10
shows the particle trace for the 3D focuser. The sample flow is focused in two dimensions and can be seen
in the middle of the channel.

Figure 9. Vector Plot of Three-Dimensional Vertical Focusing

Figure 10. Particle Trace of Three-Dimensional Vertical Focusing

Fluid Flow around the Electrodes

If the analysis is to include dielectrophoresis, electrode must be added to the model. A three dimensional
model was created to visualize the flow past the electrodes. In this model, two pairs of electrodes (one on
each side of the channel) were set inside a rectangular flow channel. The no-slip boundary condition was
applied to the electrode surfaces as well as the channel walls. The inlet velocity was determined by the
outlet velocity of the 3D focuser. Figure 11 shows the velocity vector plot of the flow past the electrodes
and figure 12 shows the particle trace past the electrodes. This clearly shows that the cells in the focuser
flow will no longer be perfectly centered in the channel. The cells move 2 to 3.5 microns away from the
electrodes depending on their initial position in the stream although they return to their initial position
down stream of the electrodes. Since the force that the electrodes exert on the cells is a function of the
distance between the cells and the electrodes, this will affect the electrode design and performance.

Figure 11. Vector Plot Around Electrodes in Flow

Figure 12. Particle Trace Around Electrodes in Flow

Full Three-Dimensional Model

Finally, a three dimensional fluid model was created that combined all of the elements of the cell sorter
including the two outlet channels. With all of the factors affecting the flow combined in one model, the
inlet velocities of the sample flow and the sheath flows could be fine tuned to produce the desired outlet
cell velocity. Figure 13 shows a velocity vector plot of the entire sorting system which predicts the target
maximum velocity of 0.1 m/s. Figure 14 shows a particle trace of the sample fluid as it travels through the
sorting system.

Figure 13. Vector Plot of Full Three-Dimensional Device

Figure 14. Particle Trace of Full Three-Dimensional Device

Modeling in the Electrical Domain

Sorting in a microfluidic device can be done in a variety of ways, but dielectrophoresis (DEP) is one of the
more promising methods available. Dielectric particles inside an inhomogeneous electric field will move
towards regions of high field strength (positive DEP) or low strength (negative DEP) depending on the
properties of the particles, liquid and electrode frequency.5 Mammalian cells can be approximated as
spherical dipoles that move towards areas of low field strength or away from the source electrodes.
The time averaged dielectrophoretic force on a spherical particle of radius r in a field E is given by the
equation6:

F = 2πε m r 3∇ E 2 Re(u * ) Eq 4
where u* is the Clausius-Mossotti (CM) ratio:

(ε *particle − ε medium
*
)
u* = Eq 5
(ε *particle + 2ε medium
*
)
and the complex permittivity is defined as ε = ε − jσ / w . Since the CM ratio for the example system is
*

negative, the cells are pushed away from fields of high concentration.
A two dimensional model of the electrodes in the fluid flow was created using PLANE67 (2D Thermal
Electric Solid) elements to model the behavior of the system in the electrical domain. The electrodes were
assumed to be gold rectangular boxes 3 microns wide and 15 microns deep with a 6 micron gap between
the top and bottom electrode (figure 15). A voltage of +5 V was applied to the nodes of the top line of the
top electrode. A voltage of -5 V was applied to the nodes of the bottom line of the bottom electrode. A
convection to bulk temperature was also applied at the right side of the model to allow the solution to
converge. (In the actual system, the dominant thermal initial condition is the temperature of the fluid as it
enters the electrode region while the thermal resistances of the walls and the device packaging will have
almost no impact. For this reason, the two dimensional model cannot be used for calculating temperature
increases and is presented only to demonstrate the shape of the electric fields generated.) In this model, DC
current was used for simplicity. Either AC or DC may used, but DC has the disadvantage of generating
electrolysis, which may be a concern for applications with biological specimens.

Figure 15. Material Plot of Two-Dimensional Electrode Geometry

Figure 16 shows the distribution of voltage through the electrodes and the surrounding fluid. Since gold is
an excellent electrical conductor with low resistance, both electrodes appear to have constant voltages
while the fluid shows significant gradients. Figure 17 shows the sum of the electric fields in the system.
This electric field can be exported and used to calculate the dielectrophoretic force applied to the cells and
evaluate the performance of the sorter, however it cannot be calculated directly in ANSYS.
 Figure 16. Voltage Distribution of Two-Dimensional Electrode Geometry

Figure 17. Electric Field Sum for Two-Dimensional Electrode Geometry

Equation 4 can be reduced to a constant multiplied by the gradient of the electric field squared, meaning
that the dielectrophoretic forces may be calculated once the electric field is calculated. However, ANSYS
does not calculate the gradient of the electric field (or the electric field squared) and user calculation of the
gradient of the electric field is not convenient at the current release of ANSYS (Release 7.1). This presents
several logistic problems.

1. The calculation of the gradient requires that the nodes be sorted by coordinate in insure correct
calculation. (To understand this, imagine an ANSYS area meshed in the standard manner (lines
first and then the interior of the area). If the electric field distribution is cylindrical with a variation
in the X direction and no variation in the Y direction, calculation in the ANSYS vector along an
end line would result in a zero value when the actual value would be very steep.)
2. ANSYS does not permit the sorting of one vector based upon a second vector.
 3. ANSYS does not permit calculating the DER1 for vector operations when the denominator vector
is not monotonically increasing. For 2 and 3 dimensional models, the denominator values are
likely to loop from minimum to maximum repeatedly. This will require a check on the
denominator values.
4. User calculation of the gradient will require *DO and *IF commands. The values for electric field
and for the node coordinates can be loaded into an ANSYS vector and the necessary calculations
performed. This calculation will be time consuming for large models and it may be more
convenient to export the data to another program, perform the calculations and return the force
values as loads for the subsequent calculation.
The best options to directly calculate the DEP forces on the cells in the fluid flow would be to export the
electric field and do the calculations in a separate program that was intended to do large scale linear algebra
or to write a custom macro that would calculate the gradient of the element electric energy (ETABLE,
SENE) distribution.

Joule Heating
One of the outputs of the electric domain analysis is the Joule heating that occurs as a result of the electric
current. The fluid between the electrodes acts as an electrical resistor. Some of the energy carried by the
current is passed from one electrode to the other, and some of it is dissipated as heat. The joule heating can
be combined with the fluid model to calculate the temperature rise and distribution in the fluid flow to
ensure that the cells will not go into thermal shock or be killed by the sorting process.
Since electrical conductivity (the reciprocal of resistivity) is a function of frequency, the joule heating is
also a function of frequency. Electrical conductivity is given by

σ = σ 0 + jwε
where σo is the base conductivity, j is equal to the square root of -1, w is electrode frequency, and ε is the
permittivity of the medium. The frequency dependence of the conductivity of saline at 1 MHz was found to
be at least two orders of magnitude smaller than the base conductivity, so for this example, the joule
heating will be approximately the same for AC and DC currents.
A three dimensional model of two electrode pairs in the fluid was created to display the Joule heating
between the electrodes. The Joule heating model uses SOLID69 elements and the same geometry and
initial conditions as the electric field model. Figure 18 shows a cut away view of the model. The area of
highest heat dissipation is in the area between the electrodes as expected but has a very limited range away
from the electrodes.

Figure 18. Joule Heating Distribution in Saline

Modeling in the Thermal Domain
Finally, the results of the Joule heating model must be transferred to the thermal domain to calculate the
temperature increase in the fluid flow. The thermal model uses SOLID70 (Thermal Solid) elements with
the same geometry as the electric models. One of the key options of the SOLID70 elements allows the
specification of a velocity in the solid elements. In this way, the velocity of the fluid flow within the
channels was added to the model. Power graphics for all thermal models was turned off to ensure that the
true maximum temperature of the system would be shown.
First, an approximate model of the temperature distribution was created using an average value of the joule
heating. To apply the boundary conditions, the element table from the electric model was generated,
summed and then averaged over the area of the model which was affected by the Joule heating. This
average Joule heating value was then applied to the same affected area in the thermal model to give a close
approximation of the temperature rise in the fluid. The temperature distribution from this approximation
can be seen in Figure 19.

Figure 19. Approximate Temperature Distribution in Saline

To test the validity of the approximation, a second model was created for the temperature distribution. In
the second case, the exact joule heating distribution was read into the thermal model using the
LDREAD,HGEN command. Figure 20 shows the temperature distribution created by reading the Joule
heating element table directly into the thermal model, thus bypassing the approximation and leading to a
better solution. The exact solution predicts a maximum temperature of 51.911 degrees C while the
approximate solution predicts a maximum temperature of 28 degrees C. Thus, the exact solution is
necessary to calculate secondary effects like electrohydrodynamic and density driven flows. However, in
both cases the temperature distribution is approximately the same and both predict that the temperature of
the sample fluid will rise only a few degrees, indicating that the approximation holds for fluid far away
from the electrodes.
 Figure 20. Exact Temperature Distribution in Saline
The high temperature increase in the electrode region prompted the creation of one last model which
assumed the use of an isotonic sucrose solution with a conductivity of 0.12 S/m based on the work of
Gimsa and Wachner.7 The electrical conductivity of the sucrose solution in much higher than the
conductivity of the physiological saline solution which results is much lower temperature increases in the
fluid. Figure 21 shows the joule heating of the electrodes submerged in a sucrose solution. Figure 22 shows
the temperature distribution in the sucrose solution using the exact joule heating distribution. The
temperature profile looks identical to those generated in the previous models, however the maximum
temperature in this system is only 22.581 C.

Figure 21. Approximate Temperature Distribution in Sucrose

 Figure 22. Exact Temperature Distribution in Sucrose

Conclusion
It has been shown that the behavior of microfluidic systems that operate in a number of physical domains
can be modeled and optimized using finite element methods. However, dielectrophoretic forces cannot be
modeled in ANSYS at this time without writing a custom macro to perform a direct calculation. The
creation of this macro will be the focus of future work.

Acknowledgements
The authors would like to thank the MIT Dept. of Chemical Engineering MicroChemical Systems
Technology Center for its support of this work, Prof. Alexander Slocum, Jason Kralj, Eddy Karat, Elena
Antonova and Dave Looman for their feedback, and the MIT 6.777 class for the Spring, 2003 for their
contributions to a similar project. M. K. Thompson would like to thank the National Science Foundation for
its support of her research through a Graduate Research Fellowship. The authors would also like to thank
ANSYS, Inc. for providing the software to make this analysis possible.

References
1. http://webster.com
2. Fiedler et al. Anal. Chem. 1998, 70, 1909-1915
3. Lee et al. Journal of Fluids Engineering, 2001, (123) 672-679
4. Lee et al. Journal of Fluids Engineering, 2001, (123) 672-679
5. Fiedler et al. Anal. Chem. 1998, 70, 1909-1915
6. Foster et al. Biophys. J. 1992,63,180-190.
7. Gimsa and Wachner. Biophys J, August 1998, 75, 1107-1116.