Digital Microfluidics

A Digital-Microfluidic
Approach to Chip Cooling
Philip Y. Paik
ICx Biosystems

Krishnendu Chakrabarty
Duke University

Vamsee K. Pamula
Advanced Liquid Logic
figure flows, and therefore cooling
rates, to match the chips thermal
profile. We earlier presented an implementation of this approach that uses
aqueous-based droplets to collect, carry, and dissipate heat away from the
substrate.2 Here, we present a different
approach, which uses liquid-metal
droplets to create an adaptive thermal-interface
material (TIM).

Thermal management is an increasingly important aspect of IC design. After

reviewing several chip-cooling techniques proposed in the literature, this
article presents an alternative approach based on a recently invented digitalmicrofluidic platform. Originally developed for a biological and chemical lab
on a chip, this technology can be adapted for use as a fully reconfigurable,
adaptive cooling platform.

&OVERHEATING IS A major cause of IC failure during

system operation. Although the semiconductor industry continues to double transistor density every two
years (as Moores law predicted), the power dissipation associated with this performance gain is also
increasing. This is not a new or surprising trend, but ICs
are quickly reaching temperatures that current cooling
techniques are hard-pressed to keep up with. The 2006
International Technology Roadmap for Semiconductors
predicts that the peak power consumption of highperformance desktops will jump by 18% (167 W to
198 W) in 2013, and by 51% (91 W to 137 W) in lowerend desktops in 2013. If cooling methods dont
improve, elevated die temperatures in tomorrows
ICs could result in catastrophic failures.1 Furthermore,
as uneven thermal distributions across the substrate
become more pronounced, failures due to physical
stress will increase. Researchers have proposed chipcooling techniques to improve heat transfer, but most
of these solutions fail to address the problems caused
by a nonuniform thermal profile.
In this article, we provide an overview of todays
chip-cooling solutions. We then present an alternative
approach based on the recently invented digitalmicrofluidic platform. This new fluid-handling platform enables a vast array of microliter- to nanoliter-size
droplets to move independently along a substrate.
Unprecedented flexibility allows the system to recon-

372

0740-7475/08/$25.00

2008 IEEE

IC-cooling techniques
There has been considerable developmental effort
to improve chip cooling, particularly in recent years.
Researchers have proposed a wide range of approaches, many leaning toward the microfluidic platform.
Although these techniques have relied on fundamentally different technologies, they share several characteristics. We have classified these common characteristics into two categories, each with its own
advantages in performance, complexity, and cost.

Passive versus active cooling

Cooling methods are either passive or active.
Passive methods include thermal conduction (pastes,
metal lines, and vias), natural convection (finned heat
sinks and ventilation slots), and radiation (coatings
and paints). Heat pipes and thermosyphons are also
passive methods, but they offer higher performance.
Passive devices are generally inexpensive to implement because they have no active component, and
they are relatively simple in design. Passive-cooling
devices typically perform worse than active-cooling
devices.
Active cooling requires input power. Thus, active
methods require external components such as forced-

Copublished by the IEEE CS and the IEEE CASS

IEEE Design & Test of Computers

convection devices (fans and nozzles), pumped loops

(heat exchangers and cold plates), and refrigerators
(Peltier thermoelectric and vapor-compressionbased).

Adaptive versus nonadaptive cooling

We further categorize active-cooling devices with
the notion of adaptability to varying temperatures. An
adaptive cooling system incorporates closed-loop
feedback of the chips temperature (from either direct
or indirect measurement). The temperature-aware
system can dynamically cool different chip areas.
The ability to selectively cool different areas is
important because thermal nonuniformity can cause
detrimental stresses on the IC substrate. Thermal
nonuniformities, or hot spots, arise from varying power
distributions on the IC substrate and can reach power
densities of 300 W/cm2 or more. A nonadaptive
cooling system lacks a temperature feedback mechanism and thus cannot respond to hot spots.
Underlying technologies
Commercial and academic cooling systems use
five general technologies: heat-sink fans, macrofluidics, microelectromechanical systems (MEMS), refrigeration, and microfluidics.
Fan-based cooling. The most widespread technique
for cooling chips uses a heat spreader and forced
convection. A metal-finned heat sink is typically
mounted on the chips packaging, and a fan coupled
to the heat sink circulates air through the fins to
dissipate the heat. Air is usually directly available, and
the cooling device doesnt require any additional
complex or expensive packaging. Although small
improvements in heat-sink design and in chip
interfaces have been made recently, this technology
has remained largely unchanged for many years.
Macrofluidics-based cooling. A second class of
cooling technologies is based on macrofluidics. These
liquid-cooling methods work at the macroscale
(greater than microliters) and fall into two
categories: direct and indirect. Direct methods
involve the immersion of electronic chips in a pool
of inert dielectric liquid. Indirect methods usually use
a two-phase flow. Examples of indirect-cooling devices
include thermosyphons and heat pipes. In
thermosyphons, a liquid evaporates with applied
heat and condenses, dissipating the heat elsewhere

July/August 2008

in a closed system. Typically, gravity returns the

condensed liquid to the hot area. A heat pipe is a
sealed and evacuated vacuum-tight container partially
filled with a fluid. When heat is applied locally, the
fluid in that part of the pipe vaporizes, travels to the
low-pressure areas, and condenses. The condensate
reaches the hot area through wick structures lining the
heat pipe, ensuring uniform heat distribution. Heat
pipes are passive-cooling devices and are used in
many notebook and quiet desktop computers such as
Apples Power Mac G5.
MEMS-based cooling. A third class of cooling
technologies is MEMS-based devices. Using todays
microfabrication techniques, manufacturers can
develop complex structures on a substrate to
promote heat conduction and dissipation. These
devices replace traditional plain-wall heat sinks, and
although they can be designed to work with either air
or liquid, air is usually preferred for cost-effective
implementation and simple assembly. Examples of
proposed MEMS-based air-cooling methods include
air-impinging jet heat exchangers and arrays of
microfabricated membranes actuated electromagnetically to produce air microjets.3 These methods
work on the principle of injecting small streams of air
through the MEMS device onto the IC substrates
surface to create flow patterns that help redistribute
and therefore dissipate heat. Another example of a
MEMS-based device is a bulk-micromachined microfin
array heat sink, which enhances hydrodynamic
mixing. The array consists of warped bimorph
microcantilevers that vibrate when an air jet
impinges on them, creating vortices and dissipating
heat faster than a plain-wall heat sink.
Refrigeration-based cooling. Another type of ICcooling technology is refrigeration systems using vapor
compression, gas compression, or thermoelectric
devices. These systems can generate a subzero
effective thermal resistance, and require only that the
evaporator attached to the chip package have a
contact temperature less than the cooling-air
temperature. The result is a substantial increase in
heat dissipation from the chip.
Traditional refrigeration units require large amounts
of input power to operate, and their compressors are
large and bulky. Engineers have developed compact
vapor compression refrigeration units, but focus has
shifted toward solid-state refrigeration. Solid-state

373

Digital Microfluidics

refrigeration uses thermoelectric coolers (TECs) to

provide thermal resistance at or below 0uC/W. TECs
depend on the Peltier effect, by which DC current
applied across two dissimilar materials causes a
temperature differential. In a typical system, the TECs
lie between the heat source (the IC die or packaging)
and the heat sink.4 Because one side of the TEC is hot,
the heat sink must dissipate power on the hot side. This
extra energy can cause higher ambient temperatures
at the heat sink, and if the heat sink cannot dissipate
the heat, the cold side of the TEC might heat up as
well. As a result, researchers are looking for ways to
increase the efficiency of TEC devices.
Microfluidics-based cooling. Originally proposed
in the early 1980s, microfluidic cooling devices push
small volumes of liquids (less than microliters) across
an ICs surface to conduct and dissipate heat
downstream. Owing to limitations in microfluidichandling technologies, this method has yet to be
realized in a practical commercial system. However,
as chip power consumption continues to rise and
microfluidic pumping technology matures, microfluidics-based cooling is receiving renewed attention.
The first microfluidic cooling device was the
microchannel heat sink.5 Microchannel heat sinks
are minute flow channels fabricated on the back of a
thin silicon chip substrate. With hydraulic diameters
ranging from 10 to 30 microns, these heat sinks have
an extremely high surface area per unit volume of
working fluid, and a low thermal resistance. However,
because they required a large external supply of
pressurized cooling liquid, commercialization of these
devices was not economically or practically feasible.
Furthermore, an external pressure source for a liquid
supply is inherently unsuitable for compact, embedded systems. Thus, there remained a technology void
and a pressing need for IC cooling in compact
packages.
Over the past decade, microchannel devices have
been developed to miniaturize cooling hardware.
Microchannel cooling loops now have high cooling
performance, large heat-transfer coefficients, small
channel volumes, and a small cooling inventory. Heat
transfer in these systems depends on heat convection
between a solid and a moving liquid. Given a solid
surface at temperature Tb and an adjacent moving
liquid at temperature Tf, where Tb . Tf, the rate of
energy transfer from the surface of the solid to the
liquid is quantified as dQc/dt 5 hA(Tb 2 Tf). This is

374

Newtons law of cooling. A is the surface area and h is

a proportionality factor called the heat-transfer coefficient. The heat-transfer coefficient is an empirical
parameter that incorporates the flow patterns nature
near the surface, the fluid properties, and the geometry
into the heat-transfer relationship. For example, in a
forced-convection system using fans blowing across
the chip, the heat-transfer coefficients value is far
higher than in a free-convection system, because new
flow patterns are introduced to the otherwise slow,
buoyancy-induced air motion.
Researchers have developed a pumpless loop with
microchannel surfaces for cooling.6 It relies on fluid
density differences between two vertical, parallel
tubes to induce fluid motion. This method requires
no external pumps, and as the temperature increases,
the flow becomes faster. It is a passive-cooling method
with no external power requirements. However, it
cannot be used in applications that need specific hotspot cooling.
Recently, much attention has focused on developing reliable micropumps that can pump liquids
through microchannels. To increase the heat flux
from a microchannel with single-phase cooling, it is
necessary to increase the heat-transfer coefficient
either by increasing the liquid flow rate or by
decreasing the hydraulic diameter. In two-phase
regimes, the evolution of the phase change from
liquid to vapor in microsystems is different than in
macrosystems. Stanford researchers developed a
closed-loop, two-phase microchannel cooling system
based on electro-osmotic pumping of liquids.7 The
electro-osmotic pump is connected to a microchannel heat exchanger, which in turn is connected to a
heat rejecter. This pump has removed 38 W with a
pump power of 2 W. This was the first demonstration
of a hermetically sealed electro-osmotic cooling
system.
A MEMS-based microcapillary pumped loop
(micro-CPL) has been fabricated on a silicon wafer,
integrating an evaporator, a condenser, a reservoir,
and liquid lines.8 Micro-CPLs offer greater geometric
freedom than heat pipes. They also carry greater heat
loads because of the simultaneous flow of vapor and
liquid, in contrast to the countercurrent flows in
conventional heat pipes.
Georgia Tech researchers have developed a
cooling module based on piezoelectric droplet
generation.9 In this structure, secondary droplets form
from a primary drop and then impinge on the hot

IEEE Design & Test of Computers

surface. Unlike a heat pipe, the structure doesnt need

wicks to continuously supply liquid to the chips hot
area. Upon impinging on the hot surface, the droplets
evaporate and dont form an insulating vapor blanket
as occurs with pool boiling. The atomized droplets
have sufficient momentum to penetrate through the
vapor layer and spread into a thin film on the hot
surface.

Digital-microfluidic cooling platforms

Although many of the newer cooling techniques
just reviewed can transfer heat at a significantly higher
rate than current practices do, they lack the flexibility
to accommodate dynamic cooling of hot spots that
cause nonuniform thermal profiles. This is because
these techniques cannot alter cooling rates, either
spatially or temporally, and thus are inadequate for
adaptive cooling. Hence, designers address, in advance, any thermal variations in the IC by aggressively
designing for higher cooling rates near the regions of
high heat flux.
An alternative technique, which enables adaptive
cooling, uses digital microfludics technology. Duke
University researchers have developed a micropump
that uses electrowetting to form and manipulate
discrete droplets electrostatically.10 Dividing liquids
into independently controlled liquid packets for
manipulation provides several important advantages
over continuous-flow or mist-based systems:
&

&

&

&

Microfluidic operations can be reduced to a set

of basic discrete operations (moving one unit of
liquid one unit step), allowing a hierarchical,
cell-based design approach.
The absence of permanently etched structures
allows a completely reconfigurable system. The
only active component, an actuating electrode,
is embedded in a planar surface of the device,
requiring no additional components or structures
such as pumps and valves.
Given a 2D array of actuating electrodes, the device
can manipulate liquid droplets laterally in any
arbitrary path without external pumps and valves.
Without the fixed structures found in continuousflow devices, the digital-microfluidic system can be
completely virtual and reconfigurable.
Liquid flow inherently increases with increasing
temperature. Thus, local hot spots on a chip can
potentially have an inherently increased cooling
rate without requiring external sensors.

July/August 2008

Electrowetting-based droplet actuation

A digital-microfluidic device typically consists of a
substrate patterned with an array of electrodes. There
are several ways to pattern the substrate, ranging from
microfabrication vacuum processes to less expensive
printed circuit board (PCB) processes. An insulator
separates the electrodes from the conductive droplets.
The insulator is coated with a hydrophobic material
that prevents the droplet from wetting the surface. An
upper plate provides a channel in which the droplets
can move. The upper plate is coated with a conductive
material, usually indium tin oxide, to keep the droplet
at a ground potential. A spacer material maintains a
gap between the lower substrate and the upper plate,
and the entire channel contains an immiscible fluid
(usually silicone oil) to facilitate droplet movement.
Figure 1a shows a schematic of the assembled glassbased chip.
The digital-microfluidic device achieves droplet
motion through the electrowetting effect. Applying a
potential to an underlying electrode controls a
droplets ability to wet the surface. Applying this
electric field increases the droplets surface energy,
and therefore the droplet wets the surface. This results
in a decrease in the contact angle, causing the droplet
to spread onto the electrode. Placing a droplet on top
of an array of electrodes initiates lateral motion.
Figure 1b illustrates the method. Pollack, Shenderov,
and Fair provide experimental details and protocols.10
Unlike continuous-flow microfluidic devices, which
allow only unidirectional flow induced by an external
pump, a digital-microfluidic device under software
control can drive droplets in any arbitrary direction.
The user can programmatically manipulate many
droplets using a set of fundamental fluidic operations:
transport, dispense, merge, and split.11
Flow-through adaptive cooling
Although digital-microfluidic devices have traditionally functioned in biological or chemical labs on
chips, we propose implementing the same architecture and droplet-based operations for IC cooling
applications. One method, flow-through adaptive
cooling, would use aqueous droplets continuously
dispensed from an actively cooled reservoir and
transported over an ICs surface. The droplets would
transfer, store, and carry heat away from the IC
substrate back to the cooling reservoir. Many droplets
could therefore move in reconfigurable flow paths and
flow rates to eliminate thermal nonuniformities. The

375

Digital Microfluidics

to temperature variation within the die.

To illustrate programmable thermal
switching, Figure 3 shows a digital-microfluidic chip that contains an array of
cells covering the entire area of a target
IC chip. Each cell contains a single
electrode with a droplet of liquid metal
on top. Above the microfluidic chip is
the IC substrate. Beneath each droplet is
a metal via that travels to the chips back
side. The back side is attached to an
existing (thermoelectric) cooling device.
The liquid-metal droplet acts as a
conduit between the IC substrate and
the heat sink. By default, the device turns
all the electrodes on, detaching the
droplet from the top surface of the IC
substrate. The thermal resistance between the IC substrate and the microfluidic chip now matches that of the oil.
When the device detects a hot spot over
Figure 1. An assembled digital-microfluidic chip during droplet actuation:
a cell, the electrode in the cell is turned
side view (a), and top view (b). Applying a voltage to electrode 2 beneath the
off, causing the droplet to relax and
droplet charges the droplet, and increased surface energy causes the
subsequently come into contact with the
droplet to change its interfacial tension with the surrounding oil (if present),
top plate. This reduces thermal resisthus wetting the electrodes surface. Turning off the voltage under electrode
tance between the hot spot and the heat
2 and applying it to adjacent electrode 3 creates an interfacial tension
sink by several orders of magnitude, and
gradient. This causes the droplet to move until the interfacial tension
heat can then pass more efficiently from
equilibrates when the droplet moves completely onto the activated
the hot spot to the heat sink.
electrode. This method is the basis of all microfluidic operations.
Next, we present several studies
microfluidic device would interface with the target IC conducted to establish the feasibility of the programat either the package level or the substrate level. mable thermal-switching method. The liquid metal
Figure 2 illustrates this technique. (We present further chosen for these studies was mercury because of its
established record as an electrowettable material.
details elsewhere.2)

Programmable thermal switching

An alternative and fundamentally different approach to adaptive cooling uses electrowetting to
create a programmable thermal switch. This method
manipulates a liquid-metal droplet to selectively
switch an area in the cooling device from a high
thermal-conductivity mode to a higher thermal-conductivity mode. The result is a programmable TIM that
can reconfigure itself in response to the nonuniform
heat flux in the target IC substrate. This programmable
TIM interfaces the target IC substrate with an activecooling device, thus providing a method to eliminate
hot spots. Although this method cannot obtain
maximum cooling, it can prevent physical stress due

376

Mercury droplet transportability. We tested

mercurys transportability with our own digital-microfluidic devices. Using PCB-based digital-microfluidic
chips, we demonstrated successful transport of
mercury for a system with and without a filler fluid
(1.5-centistoke silicone oil). The maximum transport
switching frequencies were higher in oil than in air,
and the droplet motion was more repeatable in the
absence of a top plate.
Steady-state heat transfer of mercury. Typically,
a microprocessor operates with a nominal level of
background heat, and it produces localized hot spots
when a functional unit is turned on (for example, a

IEEE Design & Test of Computers

floating-point unit). The functional unit

operates for a given period of time, and
the hot spot ceases to exist when the
functional unit is no longer running.
Thus, it is important to study the steadystate effects of heat transfer when a
mercury droplet is conducting heat
away from a hot spot, because a
functional units turn-on time can vary
depending on the CPUs workload.
In one set of experiments, we placed a
mercury droplet between a hot spot and
a conductive via in a PCB-based microfluidic chip and filled the rest of the chip
with silicone oil. We turned on the hot
spot at a specified power density and
measured its temperature as a function of
time. We performed this experiment for
hot-spot power densities of 100, 333, 500,
and 750 W/cm2. For comparison, we
obtained the hot spots steady-state
temperature in oil and in air. An on-chip
resistance temperature detector (RTD),
intertwined with the heater, recorded the
droplets temperature at a 100-ms sampling interval. We investigated mercurys
steady-state heat-transfer characteristics
for a 1-mm diameter droplet with a gap
height of 100 microns. We obtained
steady-state temperatures for the hot spot
in air, in oil, and in the presence of a
mercury droplet. Figure 4 shows the
results. We observed that for each hot- Figure 2. Proposed digital-microfluidic flow-through cooling architecture:
spot power density, the steady-state As the side view (a) shows, droplets are sandwiched between the IC or IC
temperature for a system filled with oil packaging and the digital-microfluidic chip. Attached to the digitalwas far lower than for one filled with air. microfluidic chip is an active-cooling device (such as a thermoelectric
We observed further cooling when a cooler or a heat-sink fan), which dissipates heat from the system. As the
mercury droplet was placed underneath top view (b) shows, droplets form from reservoirs of cooling liquid and
the hot spot. In these systems, adding oil move in reconfigurable flow paths back to the reservoirs, where they are
and mercury reduced thermal resistance recycled. The remainder of the chip contains oil.
between the hot spot and the bottom
microfluidic chip, which acted as a heat sink because Transient heat transfer of mercury. We performed
an experiment to demonstrate the feasibility of using a
of its large thermal mass.
We also observed that as hot-spot power density liquid-metal switch to selectively control the
increased, each systems steady-state temperature conduction of heat away from a hot spot. A mercury
increased. However, in air and in oil, the rate of droplet in oil, initially removed from a hot spot, was
increase is higher than in mercury, suggesting that the delivered to the hot spot for a period of time and then
system will perform no worse (and will most likely again removed from the hot spot. As a control, we
perform better) if this trend continues for higher power observed a similar system in which droplets consisted
of water rather than mercury. As an additional
densities.

July/August 2008

377

Digital Microfluidics

Figure 3. An array of liquid-metal droplets on a digital-microfluidic chip: top view (a), and side view (b). Droplets
interface with the IC chip or package. The device switches between low and high thermal conduction by switching
an electrode on and off. If an electrode or the device fails, the default position is off, so that maximum heat
transfer can occur.

Figure 5 shows temperature data obtained at a hot

spot when we delivered a mercury droplet for
20 seconds and then removed it. The hot spot had a
heat flux density of 500 W/cm2. We observed that the
hot spot had an initial steady-state temperature of
89uC, which dropped by 9uC when the mercury
droplet moved to it. The hot spot returned to its initial
steady-state value when the mercury droplet moved
away. We repeated this experiment for
both a mercury droplet and a water
droplet, with and without a via beneath
the droplet. As Figure 6 shows, there was
a significantly greater increase in temperature drop with the mercury droplet
than with the water droplet.
The presence of the via also promoted cooling. Because of its higher thermal
conductivity, mercury was more effective in cooling the hot spot than water,
and the via allowed greater heat conduction through the PCB, and thus
greater cooling. The via in our system
was epoxy filled and copper plated, but
a thermally conductive filled via is
commercially available and would significantly improve cooling.
We also studied the relationship
Figure 4. Steady-state temperature of a hot spot in air, oil, and in the
between the mercury droplets heat
presence of a mercury droplet with respect to hot-spot heat flux density.
variable, we observed heat conduction both when
an electrical via was present and when it was absent beneath the mercury droplet. This let us determine the vias effect on cooling. An on-chip resistive
temperature device (RTD), intertwined with the
heater, recorded the droplets temperature at a
sampling interval of 100 ms. Figures 5 and 6 show
the results.

378

IEEE Design & Test of Computers

transfer and the hot spots power density

(W/cm2). We applied power densities as
high as 500 W/cm2 to a 500-micron 3
500-micron thin-film heater. The resulting steady-state hot-spot temperatures
ranged from 110uC to 557uC in air, 76uC
to 466uC in oil, and 73uC to 373uC for a
mercury droplet of about 1 mm2 in area.
Figure 7 shows the results. At 500 W/cm2,
the hot-spot temperature drop from the
uncooled system (air to oil) to the system
cooled by the mercury droplet (air to
mercury) was as great as 184uC, twice as
large as that observed for the hot spot
cooled by oil only. This is a significant
improvement over the 11% increase in Figure 5. Temperature profile of a hot spot as a mercury droplet passes
temperature drop observed for the over it.
hot spot with a heat flux density of 133
W/cm2.
Table 1 compares heat flux density with relative both improve cooling performance and facilitate the
heat-transfer coefficients, defined by Newtons law of packaging of this technology with current and future
&
cooling, as calculated from the data shown in Figure 7. IC devices.
We observed a linear increase in the heat-transfer
coefficient for increasing hot-spot power densities. Acknowledgments
This suggests that the addition of mercury droplets in
This research was supported by the National
oil results in improved heat transfer, which further Science Foundation under grant CCR-0306349.
improves as the hot spot becomes more intense.

& References
1. S.H. Tsai and S.M. Kang, Cell-Level Placement for
THROUGHOUT THIS ARTICLE, we have demonstrated
Improving Substrate Thermal Distribution, IEEE Trans.
the feasibility of using the concept of digital microfluidics for applications in chip cooling.
The benefits of using aqueous or liquidmetal droplets to rapidly respond to
changing cooling demands are exclusive
to this type of microfluidic platform.
These benefits promise a cost-effective
way to handle potentially devastating
overheating side effects such as hot
spots. Several challenges remain, however, before adaptive hot-spot cooling is
possible in a real system. These include
scaling the platforms physical dimensions to transition from microliter- to
nanoliter-size droplets, exploring alternative liquids to achieve even higher
thermal conductivities, and implementing alternative dielectric materials to
insulate the electrodes from the droplet Figure 6. Temperature drop of a hot spot when a mercury (Hg) droplet in oil
in order to achieve even faster droplet or a water (H2O) droplet in oil passes over it, with and without a via beneath
transport speeds. All three areas aim to the droplet.

July/August 2008

379

Digital Microfluidics

Trans. Components and Packaging

Technologies, vol. 25, no. 3, Sept. 2002,
pp. 347-355.
8. K. Pettigrew et al., Performance of a
MEMS Based Micro Capillary Pumped
Loop for Chip-Level Temperature
Control, Proc. 14th IEEE Intl Conf. Micro
Electro Mechanical Systems (MEMS 01),
IEEE Press, 2001, pp. 427-430.
9. S.N. Heffington, W.Z. Black, and A.
Glezer, Vibration-Induced Droplet
Atomization Heat Transfer Cell for HighHeat Flux Applications, Proc. 8th
Intersociety Conf. Thermal and
Thermomechanical Phenomena in
Electronic Systems (ITherm 02), IEEE
Figure 7. Temperature drop for the hot spot (air-to-oil and air-to-mercury)
versus heat flux density.

Press, 2002, pp. 408-412.

10. M.G. Pollack, A.D. Shenderov, and R.B.

Computer-Aided Design of Integrated Circuits and

Fair, Electrowetting-Based Actuation of Liquid Droplets

Systems, vol. 19, no. 2, Feb. 2000, pp. 253-266.

for Microfluidic Applications, Lab on a Chip, vol. 2, no. 2,

2. P.Y. Paik, V.K. Pamula, and K. Chakrabarty, Adaptive

Cooling of Integrated Circuits Using Digital Microfluidics,

2002, pp. 96-101.

11. S.K. Cho et al., Toward Digital Microfluidic Circuits:

IEEE Trans. Very Large Scale Integration (VLSI)

Creating, Transporting, Cutting and Merging Liquid

Systems, vol. 16, no. 4, Apr. 2008, pp. 432-443.

Droplets by Electrowetting-Based Actuation, Proc.

3. S.J. Campbell Jr. et al., Thermal Management of a

Laptop Computer with Synthetic Air Microjets, Proc. 6th
Intersociety Conf. Thermal and Thermomechanical

15th IEEE Intl Conf. Micro Electro Mechanical

Systems (MEMS 02), IEEE Press, 2002, pp. 32-52.
12. H. Zeng et al., Piston-Motion Micro-Mirror Based on

Phenomena in Electronic Systems (ITherm 98), IEEE

Electrowetting of Liquid Metals, J. Microelectro-

Press, 1998, pp. 43-50.

mechanical Systems, vol. 14, no. 2, 2005, pp. 285-294.

4. I. Sauciuc et al., Thermal Devices Integrated with

Thermoelectric Modules with Applications to CPU
Cooling, Proc. InterPack 05, American Soc. Mechanical
Engineers, 2005, article 73243.
5. D.B. Tuckerman and R.F.W. Pease, High-Performance
Heat Sinking for VLSI, IEEE Electron Device Letters,
vol. 2, no. 5, May 1981, pp. 126-129.
6. S. Mukherjee and I. Mudawar, Smart Pumpless Loop for
Micro-Channel Electronic Cooling Using Flat and
Enhanced Surfaces, IEEE Trans. Components and
Packaging Technologies, vol. 26, no. 1, Mar. 2003, pp.
99-109.
7. L. Jiang et al., Closed-Loop Electroosmotic
Microchannel Cooling System for VLSI Circuits, IEEE

Table 1. Heat-transfer coefficient relative to heat flux density.

Heat flux density (W/cm2)

134.6

197.5

Relative heat-transfer coefficient

1.11

1.40

380

351.4 504.0
1.67

2.02

Philip Y. Paik is a senior microfluidics design engineer at ICx Biosystems in La Jolla, California. He completed the work described in this
article while he was a digital microfluidics design engineer at Advanced Liquid Logic.
His research interests include applying microfluidic
technologies to applications such as chip cooling
and biological and chemical detection systems. He
has a BSE in biomedical and electrical engineering,
and an MS and a PhD in electrical engineering, all
from Duke University.
Vamsee K. Pamula is a cofounder
and co-CTO of Advanced Liquid
Logic, which develops digital microfluidics for lab-on-a-chip applications.
His research interests include development of inexpensive and accessible diagnostics,

IEEE Design & Test of Computers

particularly the application of digital microfluidics to

clinical diagnostics. He has a BE in electronics and
communications engineering from Osmania University, Hyderabad, India, an MS in electrical engineering from the University of Nevada, Las Vegas, and
a PhD in electrical and computer engineering from
Duke University. He is a member of the IEEE, Sigma
Xi, and Tau Beta Pi.
Krishnendu Chakrabarty is
a professor of electrical and computer
engineering at Duke University. His
research interests include design and
test of SoC ICs, design automation of
microfluidic biochips, and wireless sensor networks.

July/August 2008

Chakrabarty has a BTech from the Indian Institute of

Technology, Kharagpur, India, and an MSE and
a PhD from the University of Michigan, Ann Arbor,
all in computer science and engineering. He is
a Fellow of the IEEE, a senior member of the ACM,
and a member of Sigma XI.
& Direct questions and comments about this special
section to Krishnendu Chakrabarty, Dept. of Electrical and Computer Engineering, Duke University,
Durham, NC 27708; [email protected].
For further information on this or any other computing
topic, please visit our Digital Library at http://www.
computer.org/csdl.

381