International Journal of Heat and Mass Transfer 45 (2002) 2549–2565

www.elsevier.com/locate/ijhmt

Experimental and numerical study of pressure drop and

heat transfer in a single-phase micro-channel heat sink
Weilin Qu, Issam Mudawar *

Boiling and Two-phase Flow Laboratory, School of Mechanical Engineering, 1288 Mechanical Engineering Building,
Purdue University, West Lafayette, IN 47907-1288, USA
Received 6 July 2001; received in revised form 26 October 2001

Abstract
The pressure drop and heat transfer characteristics of a single-phase micro-channel heat sink were investigated both
experimentally and numerically. The heat sink was fabricated from oxygen-free copper and fitted with a polycarbonate
plastic cover plate. The heat sink consisted of an array of rectangular micro-channels 231 lm wide and 713 lm deep.
Deionized water was employed as the cooling liquid and two heat flux levels, q00eff ¼ 100 W=cm2 and q00eff ¼ 200 W=cm2 ,
defined relative to the planform area of the heat sink, were tested. The Reynolds number ranged from 139 to 1672 for
q00eff ¼ 100 W =cm2 , and 385 to 1289 for q00eff ¼ 200 W =cm2 . The three-dimensional heat transfer characteristics of the
heat sink were analyzed numerically by solving the conjugate heat transfer problem involving simultaneous determi-
nation of the temperature field in both the solid and liquid regions. Also presented and discussed is a detailed de-
scription of the local and average heat transfer characteristics of the heat sink. The measured pressure drop and
temperature distributions show good agreement with the corresponding numerical predictions. These findings dem-
onstrate that the conventional Navier–Stokes and energy equations can adequately predict the fluid flow and heat
transfer characteristics of micro-channel heat sinks. Ó 2002 Elsevier Science Ltd. All rights reserved.

1. Introduction sors, laser diode arrays, radars, and high-energy-laser

mirrors.
Micro-channel heat sinks constitute an innovative The micro-channel heat sink cooling concept was
cooling technology for the removal of a large amount of first introduced by Tuckerman and Pease in the early
heat from a small area. The heat sink is usually made 1980s [1]. They fabricated a rectangular micro-channel
from a high thermal conductivity solid such as silicon or heat sink in a 1  1 cm2 silicon wafer. The channels had
copper with the micro-channels fabricated into its sur- a width of 50 lm and a depth of 302 lm, and were
face by either precision machining or micro-fabrication separated by 50 lm thick walls. Using water as cooling
technology. These micro-channels have characteristic fluid, the micro-channel heat sink was capable of dissi-
dimensions ranging from 10 to 1000 lm, and serve as pating 790 W=cm2 with a maximum substrate temper-
flow passages for the cooling liquid. Micro-channel heat ature raise of 71 °C above the water inlet temperature
sinks combine the attributes of very high surface area to and a pressure drop of 2.2 bar (31 psi).
volume ratio, large convective heat transfer coefficient, Due to their inherent advantages, micro-channel heat
small mass and volume, and small coolant inventory. sinks have received considerable attention since Tuck-
These attributes render these heat sinks very suitable for erman and Pease’s pioneering study. Several studies
cooling such devices as high-performance microproces- have since been published which can be grouped as an-
alytical [2–8], numerical [9,10], or experimental [11–20].
The primary objective of most of the analytical
studies is to develop design schemes which can be used
*
Corresponding author. Tel.: +1-765-494-5705; fax: +1-765- to optimize the dimensions of the heat sink. This is
494-0539. typically accomplished by minimizing the overall ther-
E-mail address: mudawar@ecn.purdue.edu (I. Mudawar). mal resistance of the heat sink for a given pressure drop.

0017-9310/02/$ - see front matter Ó 2002 Elsevier Science Ltd. All rights reserved.
PII: S 0 0 1 7 - 9 3 1 0 ( 0 1 ) 0 0 3 3 7 - 4
2550 W. Qu, I. Mudawar / International Journal of Heat and Mass Transfer 45 (2002) 2549–2565

Nomenclature q00bot heat flux at unit cell bottom wall

occupied by copper
Abot area of unit cell bottom wall occupied q00eff effective heat flux based on planform
by copper area of heat sink top surface
Ac cross-sectional area of channel Re Reynolds number
At area of overall heat sink top surface T temperature
cp specific heat at constant pressure Tin fluid inlet temperature
dh hydraulic diameter of channel, four Tm fluid bulk temperature
times ratio of cross-sectional area to T1 ambient temperature
wetted perimeter u velocity component in x direction
hconv natural convection heat transfer uin fluid inlet velocity
coefficient uout fluid outlet velocity
H height of unit cell v velocity component in y direction
Hch height of channel V~ velocity vector
Hw1 thickness of cover plate V_ volume flow rate
Hw2 thickness from unit cell bottom wall to w velocity component in z direction
channel bottom wall W width of unit cell
i number of iteration Wch width of channel
k thermal conductivity Ww half-width of wall separating channels
Kc1 ; Kc2 contraction loss coefficient x Cartesian coordinate
Ke1 ; Ke2 expansion loss coefficient y Cartesian coordinate
L length of unit cell z Cartesian coordinate
L1 ; L2 ; L3 ; L4 distance between thermocouple holes
Greek symbols
n outer normal coordinate at interface
l dynamic viscosity
between solid and liquid
q density
N number of channels in heat sink
Nu Nusselt number Subscripts
P pressure av average
Pw power supplied to micro-channel heat f fluid
sink in inlet
DP total pressure drop out outlet
DPch pressure drop along micro-channels p1 deep plenum
DPc1 ; DPc2 contraction pressure loss p2 shallow plenum
DPe1 ; DPe2 expansion pressure loss s solid
q00 heat flux C interface between solid and liquid

Among others, Knight et al. [5,6] presented an optimi- as thin fins, and simplifies the heat transfer process by
zation scheme that included both laminar and turbulent introducing such major approximations as one-dimen-
flow. Their results indicated that when the pressure drop sional heat transfer, constant convective heat transfer
is small, laminar flow prevails, yielding low thermal re- coefficient, and uniform fluid temperature. While the
sistance. Conversely, when the pressure drop is large, the classical fin analysis method provides a simple method
optimal thermal resistance is found in the turbulent re- to describing the heat transfer performance of a micro-
gion. channel heat sink, its accuracy can be greatly compro-
To develop such an optimization design scheme, it is mised by its simplifying assumptions.
essential to have an analytical description of the trans- A more accurate description of the fluid flow and
port processes in the heat sink. Heat transfer in the heat transfer characteristics of a micro-channel heat sink
micro-channel is a conjugate one, combining heat con- necessitates the use of direct numerical simulation
duction in the solid and convection to the cooling fluid. methods. Wesberg et al. [9] performed a two-dimen-
Due to the complicated nature of this flow, it is impos- sional numerical analysis by assuming both hydrauli-
sible to develop a comprehensive analytical solution for cally and thermally fully developed flow within the
the governing differential equations. Therefore, most micro-channels. Fedorov and Viskanta [10] developed a
analytical studies adopt the classical fin analysis method, three-dimensional model by eliminating the approxi-
which models the solid walls separating micro-channels mation of fully developed flow, accounting for devel-
 W. Qu, I. Mudawar / International Journal of Heat and Mass Transfer 45 (2002) 2549–2565 2551

opment of both the velocity and temperature fields. No of 1500 marked transition from laminar to turbulent
detailed comparison between the experimental results flow, which was attributed to a sharp inlet, relatively
and numerical predictions was provided in these studies long entrance region, and channel surface roughness.
to validate the numerical analysis. They concluded the classical relation for local Nusselt
In addition to the above-mentioned analytical and number was fairly accurate for modeling micro-channel
numerical studies, several experimental studies can be flows.
found in the literature. Micro-channel heat sinks with Although many studies have been conducted on mi-
different materials, dimensions, and cooling fluids were cro-channel heat sinks, findings remain inconclusive,
fabricated and tested. An excellent review of the exper- especially when comparing experimental results with
imental studies in the 1980s was provided by Phillips [4]. theoretical predictions. It is evident that an effective
More recent experimental investigations were conducted design of micro-channel heat sinks requires a funda-
by Ravigururajan et al. [16], Harms et al. [17,19], Rah- mental understanding of the fluid flow and heat transfer
man and Gui [14,15], Rahman [20], and Kawamo et al. characteristics in micro-channels. Detailed experimental
[18]. Ravigururajan et al. [16] tested a copper micro- measurements and theoretical predictions of the local
channel heat sink with 270 lm wide and 1000 lm deep and average heat sink temperature, heat flux, and Nus-
micro-channels. The overall heat sink was 2.25 cm long selt number are essential to acquiring this fundamental
and 2.42 cm wide. Results for Refrigerant-124 showed understanding.
the heat transfer coefficient increased between 300% and In this study, the pressure drop and heat transfer
900% compared with the theoretical value for laminar characteristics of a single-phase micro-channel heat sink
flow based on a Nusselt number of 4.36. They attributed are investigated both experimentally and numerically.
this enhancement to thinning of the boundary layer in The experimental results are compared with the nu-
the narrow channels. Rahman and Gui [14] tested three merical predictions to assess the suitability of macro
micro-channel heat sinks fabricated in silicon wafers. transport models in depicting the transport characteris-
The micro-channels were 1.0 mm wide and 221, 254, tics of micro-channel heat sinks. A detailed description
278 lm in depth, and the overall heat sink was 4.8 cm of the local and average heat transfer characteristics is
long and 4.6 cm wide. Measured values of the average presented and discussed. These results provide new,
Nusselt number for water were larger than predicted fundamental insight into the complex three-dimensional
analytically for developing laminar flow. This enhance- characteristics of these heat sink.
ment was attributed to breakage of the velocity
boundary layer by micro-surface roughness. The tran-
sition from laminar to turbulent flow was somewhat 2. Experimental apparatus
gradual because the small channel dimension was of the
same order of magnitude as the length scale of turbulent 2.1. Flow loop
eddies. Harms et al. [17] tested a 2.5 cm long, 2.5 cm
wide silicon heat sink having 251 lm wide and 1030 lm Fig. 1 shows the flow loop that was constructed to
deep micro-channels. A relatively low Reynolds number supply deionized water to the heat sink at the desired

Fig. 1. Schematic of flow loop.

2552 W. Qu, I. Mudawar / International Journal of Heat and Mass Transfer 45 (2002) 2549–2565

pressure, temperature, and flow rate. The water was block of oxygen-free copper. The planform (top) surface
pumped from a liquid reservoir and circulated through of the heat sink measured 1.0 cm wide and 4.48 cm long.
the flow loop by a gear pump. Upon exiting the pump, Twenty-one rectangular micro-slots were machined into
a portion of the flow, controlled by a by-pass valve, the heat sink top surface by a precision sawing tech-
entered the test loop containing the heat sink, while the nique. The micro-slots were equidistantly spaced within
remaining portion returned to the reservoir though a the 1-cm heat sink width and had the cross-sectional
by-pass loop. The test loop water first passed through a dimensions of 231 lm wide and 712 lm deep. Below the
heat exchanger where the water was cooled to the de- heat sink top surface were four 0.36 mm diameter holes
sired inlet temperature. The water then passed through which were drilled into the side wall of the heat sink up
a 15 lm filter to prevent any solid particles from to the center plane. Four Chromel–Alumel (Type-K)
blocking the heat sink micro-channels. After exiting the thermocouples with a 0.33 mm bead diameter were in-
filter, the water was routed to one of two rotameters serted into these holes to measure the temperature dis-
for volume flow rate measurement. The water then tribution inside the heat sink. Below the thermocouple
entered the micro-channel heat sink test module where holes, a small protruding platform was machined
the electric power supplied to the heat sink was re- around the periphery of the heat sink to both facilitate
moved by the water. Leaving the test module, the water accurate positioning of the heat sink in the housing and
returned to the reservoir where it mixed with the by- ensure adequate surface area for sealing. This platform
passed flow. effectively divided the heat sink into an upper portion
and a lower portion that will be discussed later in con-
2.2. Test module junction with the heat sink’s numerical model. Twelve
6.35 mm diameter holes were drilled into the lower
The test module consisted of a micro-channel heat portion of the heat sink to accommodate the cartridge
sink, housing, cover plate, and 12 cartridge heaters as heaters. These cartridge heaters were connected in par-
illustrated in Fig. 2(a). A schematic of the micro-channel allel and powered by a single 0-110 VAC variac. Power
heat sink with key dimensions is shown in Fig. 2(b). The dissipation by the cartridge heaters was measured by a
micro-channel heat sink was fabricated from a single wattmeter. As shown in Fig. 2(b), three 1.6 mm slots

Fig. 2. (a) Test module. (b) Micro-channel heat sink.

 W. Qu, I. Mudawar / International Journal of Heat and Mass Transfer 45 (2002) 2549–2565 2553

were cut from the bottom surface up through most of uin is the inlet velocity:
the heat sink’s height including part of the upper por-
V_
tion. These slots aided in reducing heat spread within the uin ¼ ; ð3Þ
lower portion of the heat sink and providing a more NAch
uniform heat flux distribution. and lf;in is evaluated using the measured inlet tempera-
The heat sink housing was made from G-7 fiber- ture. Using the heat exchanger located in the flow loop,
glass plastic. The central part of the housing was re- the water inlet temperature was set to 15 °C. Once de-
moved where the heat sink was inserted as illustrated sired experimental conditions were reached, the heat
in Fig. 2(a). The protruding portion of the heat sink sink was allowed to reach steady-state conditions, typi-
ensured the top surface of the heat sink was flush with cally within 30–60 min. Once at steady state, three sets of
the top surface of the housing. RTV silicone rubber readings from the rotameters, the thermocouples, pres-
was applied along the interface between the housing sure transducers, and wattmeter were recorded at 15 min
and the heat sink to prevent leakage. The housing intervals.
contained plenums both upstream and downstream of
the micro-channels. Each plenum had a deep portion 2.4. Measurement uncertainty
leading to a shallow portion to help ensure even dis-
tribution of the flow between micro-channels as well as Heat loss to the ambient was estimated at less than 1%
even exit mixing. Two absolute pressure transducers of the total heat input. Therefore, heat losses were as-
were connected to the deep portions of inlet and outlet sumed negligible, and cartridge heater power measured
plenums via pressure taps to measure the inlet and exit by the wattmeter was used for all heat flux calculations.
pressures, respectively. Also located in the inlet and Excellent agreement between electrical power input and
outlet plenums were two Type-K thermocouples to measured enthalpy change of the cooling water further
measure the inlet and exit temperatures of the water, verified this assumption. This issue will be addressed in
respectively. more details later in this paper. Measurement uncer-
A cover plate made from polycarbonate plastic tainties associated with the wattmeter, rotameters, and
(Lexan) was bolted atop the housing to hold the heat pressure transducers were less than 0.5%, 4% and 3.5%,
sink securely in place. The cover plate and micro-slots respectively. Error associated with the thermocouple
in the heat sink top surface formed closed micro- measurements was estimated at less than 0.3 °C.
channels. An O-ring in the housing maintained a leak-
proof seal.
3. Results and discussion
2.3. Data acquisition
In this section, the three-dimensional fluid flow and
During tests, the power dissipated by the cartridge heat transfer characteristics of the heat sink are first
heaters was adjusted manually by the variac. Two input analyzed numerically. Then, the measured performance
power levels were tested. They were defined by an ef- presented and compared with the numerical predictions.
fective heat flux q00eff , based on the top planform area of Finally, detailed descriptions of the local and average
the heat sink, At ¼ 1:0  4:48 cm2 : heat transfer characteristics of the heat sink are dis-
cussed based on the numerical results.
PW
q00eff ¼ ; ð1Þ
At 3.1. Numerical analysis

where PW is the total electric power input measured us- A unit cell containing a single micro-channel and
ing the wattmeter. The two heat flux levels tested were surrounding solid is chosen to perform the numerical
q00eff ¼ 100 W=cm2 and q00eff ¼ 200 W=cm2 . The pump analysis. Symmetry allows these results to be easily ex-
and by-pass valve were set to produce a constant water tended to the entire heat sink. Fig. 3 illustrates the unit
flow rate during a given test. A broad range of flow rates cell and corresponding coordinate system and key no-
was examined, from 93.5 to 1136.4 ml/min for tations. Dimensions of the unit cell are given in Table 1.
q00eff ¼ 100 W=cm2 , and 259.1 to 859.5 ml/min for An alternative unit cell configuration which was not
q00eff ¼ 200 W=cm2 . These flow rates correspond to Rey- used in the present study is one utilizing the channel
nolds numbers from 139 to 1672 for q00eff ¼ 100 W=cm2 mid-plane (0 6 x 6 L, y ¼ W =2, and 0 6 z 6 H ) as the
and 385 to 1289 for q00eff ¼ 200 W=cm2 , where the Rey- unit cell boundary. This arrangement is computationally
nolds number is based on inlet parameters: more efficient since fewer nodes are needed. However,
because the velocity and temperature gradients within
qf uin dh the flow channel are much larger than those in the sur-
Re ¼ ; ð2Þ
lf;in rounding solid, a large numerical error may be incurred
2554 W. Qu, I. Mudawar / International Journal of Heat and Mass Transfer 45 (2002) 2549–2565

Therefore, variable temperature-dependent water vis-

cosity based on mean temperature is employed. Natural
convection of air within the deep slots is neglected (as-
sumption (6)) because the small, deep slots suppress
natural circulation within the slots. Therefore, the
plastic cover plate, copper block and air within the deep
slots are treated as a unitary ‘‘solid’’ region with a dif-
ferent thermal conductivity value assigned to each re-
gion. In the following section, subscript s is used to
denote these ‘‘solid’’ regions.
Based on the above assumptions, the governing dif-
ferential equations used to describe the fluid flow and
heat transfer in the unit cell are expressed as follows. For
the cooling water, the continuity, momentum, and en-
Fig. 3. Schematic of heat sink unit cell for numerical simula- ergy equations are expressed, respectively, as
tion. rV~ ¼ 0; ð6Þ
   
qf V~ rV~ ¼ rP þ r lf rV
~ ; ð7Þ
when discretizing the channel mid-plane boundary. This
may compromise the accuracy of the numerical scheme.  
qf cp;f V~ rT ¼ kf r2 T : ð8Þ
Heat transfer in the unit cell is a conjugate one com-
bining heat conduction in the solid and convection to the
cooling fluid. The two heat transfer modes are coupled by For the solid regions, the continuity and momentum
continuities of temperature and heat flux at the interface equations are simply
between the solid and fluid, which are expressed as V~ ¼ 0; ð9Þ
Ts;C ¼ Tf;C ; ð4Þ and the energy equation is



oTs

oTf

ks r2 T ¼ 0: ð10Þ
ks ¼ k : ð5Þ
on
C on
C
f

A numerical method to solve this conjugate heat

Several simplifying assumptions are incorporated transfer problem is to treat the solid and fluid as a unitary
before establishing governing equations for the fluid computational domain and solve the above governing
flow and heat transfer in the unit cell: equations simultaneously [21,22]. Using this method,
(1) steady fluid flow and heat transfer; which is described later, only boundary conditions for
(2) incompressible fluid; the unit cell are needed, which are specified as follows.
(3) laminar flow; First, consider the hydraulic boundary conditions. A
(4) negligible radiation heat transfer; uniform velocity is applied at the channel inlet:
(5) constant solid and fluid properties except water
u ¼ uin ; v ¼ 0; w ¼ 0 for x ¼ 0;
viscosity;
(6) negligible natural convection of air trapped in Ww 6 y 6 Ww þ Wch and Hw2 6 z 6 Hw2 þ Hch : ð11Þ
the deep slots of the heat sink.
The flow is assumed fully developed at the channel
Further explanations are needed for assumption (5)
outlet:
and (6). In assumption (5), the solid and liquid proper-
ties are assumed constant because variations of these ou ov ow
¼ 0; ¼ 0; ¼ 0 for x ¼ L;
properties are small within the temperature range tested. ox ox ox
However, the variation of water viscosity is significant. Ww 6 y 6 Ww þ Wch and Hw2 6 z 6 Hw2 þ Hch : ð12Þ

Table 1
Dimensions of micro-channel heat sink unit cell
Ww (lm) Wch (lm) Hw1 (lm) Hch (lm) Hw2 (lm) Hth (lm)
118 231 12 700 713 5637 3175
L1 (mm) L2 (mm) L3 (mm) L4 (mm) L (mm)
5 11.588 11.588 11.588 44.764
 W. Qu, I. Mudawar / International Journal of Heat and Mass Transfer 45 (2002) 2549–2565 2555

The velocity is zero along all other solid boundaries. As mentioned before, the entire unit cell can be
Thermal boundary conditions are given as follows. treated as a unitary computational domain. The gov-
For the solid region, a constant heat flux is applied at erning differential equations, Eqs. (6)–(10), together with
the unit cell bottom, copper wall (Wall A): the boundary conditions, are discretized along the three
oT spatial coordinates using the finite difference method,
ks ¼ q00bot for 0 6 x 6 L; 0 6 y 6 W and z ¼ 0; which results in a system of algebraic equations. The
oz
SIMPLE algorithm [21] is then applied to solve the re-
ð13Þ
sulting system of equations in primitive variables,
where q00bot is calculated from namely u, v, w, P, and T. Appropriate values are as-
PW signed for the thermophysical properties in the solid and
q00bot ¼ : ð14Þ fluid regions. When solving the momentum equations, a
NAbot
very large value is assigned to the viscosity in the solid.
At the unit cell top wall (Wall C), natural convection is Since velocity at all the liquid–solid boundaries is spec-
assumed: ified as zero, a very large solid viscosity will result in zero
oT velocity throughout the solid. The Gauss–Seidal itera-
 ks ¼ hconv ðT  T1 Þ tive technique is employed during the solution process,
oz
with successive over-relaxation to improve convergence
for 0 6 x 6 L; 0 6 y 6 W and z ¼ H ; ð15Þ
time. The solution is regarded convergent when the
where hconv is the convective heat transfer coefficient criterion of max jð/iþ1  /i Þ=/iþ1 j 6 106 is satisfied,
which is estimated at 10 W/°C m2 . T1 is the ambient where / represents any dependent variable, namely u, v,
temperature of 25 °C. Adiabatic boundary conditions w, or T, and i is the iteration number.
are applied to all other boundaries of the solid region. The grid system employed in the present numerical
The inlet temperature of the cooling water is set equal analysis has 102, 22, and 43 nodes in the x, y and z-
to the water temperature measured at the channel inlet: directions, respectively. A non-uniform grid arrange-
ment in the x-direction, with a larger number of grid
T ¼ Tin for x ¼ 0; Ww 6 y 6 Ww þ Wch points near the channel inlet, is used to resolve the de-
and Hw2 6 z 6 Hw2 þ Hch : ð16Þ veloping region. In order to check the sensitivity of the
numerical results to mesh size, three different grid sys-
The flow is assumed thermally fully developed at the tems were tested. They consisted of 62  12  23,
channel outlet: 102  22  43, and 122  42  63 nodes in the x, y, and
o2 T z directions, respectively. The results from the last two
¼ 0 for x ¼ L; Ww 6 y 6 Ww þ Wch grid systems were very close to each other and local
ox2
and Hw2 6 z 6 Hw2 þ Hch : ð17Þ temperature differences were less than 0.1%. Since less
computational time and computer memory were needed
It should be noted here that the temperature field might for the second grid system, it was employed in the final
not be fully developed if the thermal entrance length is simulation.
longer than the channel itself. However, the change of
temperature gradient along the flow direction at the 3.2. Experimental results and comparison with numerical
channel exit is quite small even for very large Reynolds predictions
numbers. Therefore, using Eq. (17) as an exit thermal
boundary condition will not influence the accuracy of Fig. 4(a) shows the measured pressure drop in-
the numerical results. creasing with increasing Reynolds number. The largest

Fig. 4. (a) Variation of measured pressure drop with Reynolds number. (b) Comparison of measured and predicted values of pressure
drop.
2556 W. Qu, I. Mudawar / International Journal of Heat and Mass Transfer 45 (2002) 2549–2565

pressure drop of 0.86 bar was measured at q00eff ¼

100 W=cm2 and Re ¼ 1672. For the same Reynolds
number, the pressure drop is lower for q00eff ¼ 200
W=cm2 because the higher water temperature at this
condition decreases water viscosity.
A comparison between measured and predicted
pressure drops for the two heat flux levels and different
Reynolds numbers is shown in Fig. 4(b). The predicted
pressure drop accounts for the pressure drop along the
micro-channels, as well as the pressure losses associated
with the abrupt contraction and expansion at the micro- Fig. 5. Comparison of experimental data and energy balance
channel inlet and outlet, respectively. A complete ex- predictions for water temperature rise from heat sink inlet to
pression for the total pressure drop is outlet: (a) q00eff ¼ 100 W=cm2 , (b) q00eff ¼ 200 W=cm2 .

DP ¼ DPc1 þ DPc2 þ DPch þ DPe2 þ DPe1 ; ð18Þ

outlet water temperature should decrease with in-
where the individual components are as follows. DPch is creasing Reynolds number as shown in Fig. 5. This in
the pressure drop across the micro-channels, which can turn increases water viscosity, resulting in a larger
be obtained from the numerical simulation. DPc1 and pressure drop. Second, the inlet and outlet pressure
DPc2 are the contraction pressure losses from the deep losses are proportional to the square of velocity as
plenum to the shallow plenum, and from the shallow indicated in Eqs. (19)–(22). Therefore, increasing Rey-
plenum to the micro-channels, respectively. They are nolds number produces a more pronounced increase in
expressed as [23] the inlet and outlet pressure losses. Another interesting
observation from the experimental pressure drop re-
1   K
c1
DPc1 ¼ qf u2p2;in  u2p1;in þ q u2 ; ð19Þ sults is the absence of an early transition from laminar
2 2 f p2;in
to turbulent flow within the Reynolds number range
and tested.
Fig. 5 compares the measured water temperature rise
1   K
c2 between the channel inlet and outlet, and theoretical
DPc2 ¼ qf u2in  u2p2;in þ q u2 ; ð20Þ
2 2 f in values predicted from the simple energy balance
where subscripts p1 and p2 denote the deep plenum and qf V_ cp;f ðTout  Tin Þ ¼ PW : ð23Þ
shallow plenum, respectively, and Kc1 and Kc2 are the
loss coefficients for the abrupt contractions. Similarly, The excellent agreement between the two proves virtu-
DPe2 and DPe2 are the expansion pressure losses from the ally all the electrical power supplied by the cartridge
micro-channels to the shallow plenum, and from the heaters was removed by the water, and heat losses are
shallow plenum to the deep plenum, respectively, which indeed negligible.
are expressed as Figs. 6(a) and (b) show readings from the four ther-
mocouples located inside the heat sink for q00eff ¼
1   K
e2
DPe2 ¼ qf u2p2;out  u2out þ q u2 ; ð21Þ 100 W=cm2 and q00eff ¼ 200 W=cm2 , respectively. The
2 2 f out
locations of the thermocouples are illustrated in Figs.
and 2(b) and 3. For each location, Figs. 6(a) and (b), show
1   K temperature decreases with increasing Reynolds num-
e1
DPe1 ¼ qf u2p1;out  u2p2;out þ q u2 ; ð22Þ ber, which is similar to the water temperature rise data
2 2 f p2;out
in Fig. 5. For the same Reynolds number, the temper-
where Ke1 and Ke2 are the loss coefficients associated ature increases along the flow direction. Also for the
with the abrupt expansion. For the present heat sink test same Reynolds number, the temperature for a given
module geometry, the values of Kc1 , Kc2 , Ke1 and Ke2 are location is higher for q00eff ¼ 200 W=cm2 than for q00eff ¼
close to unity. 100 W=cm2 .
Fig. 4(b) shows good agreement between the mea- Figs. 7(a) and (b) compare the temperatures mea-
sured and predicted pressure drops. For a fluid with sured along the thermocouple line with numerical pre-
constant properties flowing through a rectangular dictions for the two heat fluxes and different Reynolds
channel, one would expect a linear relationship be- numbers. As expected, the heat sink temperature in-
tween pressure drop and Reynolds number. There are creases along the flow direction. The shallow dips in the
several reasons for the slope change in the pressure predicted distributions correspond to locations of the
drop characteristics in Figs. 4(a) and (b). First, for deep slots where air is trapped. The good agreement
constant power input and water inlet temperature, the between the measured and predicted temperatures
 W. Qu, I. Mudawar / International Journal of Heat and Mass Transfer 45 (2002) 2549–2565 2557

Fig. 6. Thermocouple readings inside micro-channel heat sink versus Reynolds number for (a) q00eff ¼ 100 W=cm2 and (b)
q00eff ¼ 200 W=cm2 .

Fig. 7. Comparison of experimental data and numerical predictions for temperature distribution along thermocouple line: (a)
q00eff ¼ 100 W=cm2 , (b) q00eff ¼ 200 W=cm2 .

proves the conventional Navier–Stokes and energy the deep slots. These dips are virtually eliminated when
equations can adequately predict the pressure drop and moving to the channel bottom wall (Wall a), Fig. 8(b),
heat transfer characteristics of the micro-channel heat because of the excellent heat spread in the copper region
sink. Based on the present results as well as those re- between the two planes. There are very slight changes in
ported in several recent studies (e.g., [17,19]), it is ex- the temperature gradient in the x-direction at the
pected that fluid flow and heat transfer within channels channel bottom wall (Wall a), Fig. 8(b), and channel top
with characteristic dimensions greater than 100 lm and wall (Wall c), Fig. 8(c). In fact, a linear temperature rise
Reynolds numbers below 1700 should follow the con- can be regarded as a good approximation for both
ventional Navier–Stokes equations. planes. The temperature along the transverse y-direction
Having validated the numerical method, a more de- is virtually constant for all the x–y planes just discussed.
tailed depiction of the heat transfer characteristics of the As expected, the temperature decreases from the unit cell
heat sinks is now discussed. bottom wall to the unit cell top wall.
Notice that the channel side wall (Wall b) is an x–z
3.3. Local and average heat transfer characteristics plane. Fig. 8(e) shows higher temperatures for small z
values close to the channel bottom wall (Wall a).
The temperature distributions at several key planes Figs. 9(a)–(c) illustrate the heat flux distribution
of the unit cell are illustrated in Figs. 8(a)–(e) for along the channel walls for q00eff ¼ 100 W=cm2 and
q00eff ¼ 100 W=cm2 and Re ¼ 890. Note that the unit cell Re ¼ 890. The local heat flux, q00 , is evaluated from
bottom wall (Wall A), channel bottom wall (Wall a),

channel top wall (Wall c), and unit cell top wall (Wall C) 00 oTs


q ¼ ks : ð24Þ
are all x–y planes. Some interesting features are readily on
C
observed. The temperature increases along the longitu-
dinal x-direction in all key planes except at the unit cell For all the channel walls, higher heat fluxes are en-
top wall (Wall C), where highest temperature is found countered near the channel inlet. This is attributed to the
about midway due to natural convection effects. In Fig. thin thermal boundary layer in the developing region.
8(a), the temperature distribution for the unit cell bot- The heat fluxes vary around the channel periphery, ap-
tom wall (Wall A) shows shallow dips at the locations of proaching zero in the corners where the flow is weak for
2558 W. Qu, I. Mudawar / International Journal of Heat and Mass Transfer 45 (2002) 2549–2565

Fig. 8. Numerical predictions of local temperature distribution for q00eff ¼ 100 W=cm2 and Re ¼ 890: (a) unit cell bottom wall (Wall A),
(b) channel bottom wall (Wall a), (c) channel top wall (Wall c), (d) unit cell top wall (Wall C), (e) channel side wall (Wall b).

a rectangular channel. The same conclusion concerning weak. This is due to the very low thermal conductivity of
the vanishing heat flux in the corners is well documented the plastic cover plate (0.2 W/m °C) comparing with that
in the heat transfer literature [24]. Fig. 9(b) shows heat of the copper block (401 W/m °C). Fig. 9(c) shows the
transfer at the channel top wall (Wall c) is also very heat flux along the channel side wall (Wall b) is higher
 W. Qu, I. Mudawar / International Journal of Heat and Mass Transfer 45 (2002) 2549–2565 2559

Fig. 8. (continued).

than along the channel bottom wall (Wall a) due to the The fluid bulk temperature and average wall temper-
short distance between the channel side walls and the atures at the unit cell bottom wall (Wall A), channel
large velocity gradient present. bottom wall (Wall a), channel top wall (Wall c), unit cell
The heat flux distribution can be better understood top wall (Wall C), and thermocouple plane are plotted in
from the local Nusselt number distribution along the Fig. 11(a) with respect to longitudinal distance x for
three channel walls, which is illustrated in Figs. 10(a)– q00eff ¼ 100 W=cm2 and Re ¼ 890. The fluid bulk temper-
(c). The local Nusselt number, Nu, is expressed as ature Tm is given by Eq. (26), and the average wall tem-
peratures at the unit cell bottom wall (Wall A), unit cell
q00 dh top wall (Wall C), and thermocouple plane are defined as
Nu ¼ ; ð25Þ
kf ðTs;C  Tm Þ Z W
1
Tav ðxÞ ¼ Ts;C dy for z ¼ 0; z ¼ H and z ¼ Hth :
where Tm is the fluid bulk temperature defined as W 0
R ð27Þ
uT dAc
Tm ¼ RAc : ð26Þ
Ac
u dAc The average wall temperatures at the channel bottom
wall (Wall a) and channel top wall (Wall c) are defined as
General features of the Nusselt number distribution are Z Ww þWch
1
similar to those in the heat flux distribution, such as high Tav ðxÞ ¼ Ts;C dy
value in the entrance region and low near the channel Wch Ww
corners. for z ¼ Hw2 and z ¼ Hw2 þ Hch : ð28Þ
2560 W. Qu, I. Mudawar / International Journal of Heat and Mass Transfer 45 (2002) 2549–2565

Fig. 9. Numerical predictions of local heat flux distribution for q00eff ¼ 100 W=cm2 and Re ¼ 890: (a) channel bottom wall (Wall a), (b)
channel top wall (Wall c), (c) channel side wall (Wall b).

As expected, Fig. 11(a) shows the average temperature channel bottom wall (Wall a), channel top wall (Wall c),
decreases from the unit cell bottom wall (Wall A) to the and channel side wall (Wall b) for q00eff ¼ 100 W=cm2 and
unit cell top wall (Wall C). The liquid bulk temperature Re ¼ 890. The definitions of q00av and Nuav at the channel
is lower than all five solid wall planes. Slope changes of top wall (Wall a) and channel bottom wall (Wall c) are
the water bulk temperature and average temperatures at similar to that of the average wall temperature given in
the channel bottom wall (Wall a) and top wall (Wall c) Eq. (28). q00av and Nuav at the channel side wall (Wall b)
are small, proving a linear temperature rise is a good are defined, respectively, as
approximation for these temperature distributions. The Z Hw2 þHch
highest temperature point is located at the unit cell 1
q00av ðxÞ ¼ q00 dz; ð29Þ
bottom wall (Wall A) immediately below the channel Hch Hw2
outlet. Z Hw2 þHch
Figs. 11(b) and (c) illustrate the average heat flux q00av 1
Nuav ðxÞ ¼ Nu dz ð30Þ
and average Nusselt number Nuav , respectively, at the Hch Hw2
 W. Qu, I. Mudawar / International Journal of Heat and Mass Transfer 45 (2002) 2549–2565 2561

Fig. 10. Numerical predictions of local Nusselt number distribution for q00eff ¼ 100 W=cm2 and Re ¼ 890 : (a) channel bottom wall
(Wall a), (b) channel top wall (Wall c), (c) channel side wall (Wall b).

for y ¼ Ww . Figs. 11(b) and (c) show the average heat Lth ¼ 0:05Re Pr d; ð32Þ
flux and average Nusselt number are very small at the
channel top wall (Wall c) compared to those at the other respectively. For the rectangular channels of the present
two walls. In fact, it is safe to assume the channel top heat sink, the hydrodynamic and thermal entry lengths
wall (Wall c) is adiabatic. For the channel bottom wall can be estimated by replacing the tube diameter d in
(Wall a) and channel side wall (Wall b), the average heat Eqs. (31) and (32) with the hydraulic diameter, dh . For
flux and average Nusselt number are greatest near the the experimental conditions indicated in Fig. 11(c), the
channel inlet and decrease rapidly to nearly constant hydrodynamic and thermal entry lengths are 0.016 m
values. As indicated by Incropera and DeWitt [24], the and 0.081 m, respectively, assuming Pr ¼ 5:2. From
hydrodynamic and thermal entry lengths of a circular Fig. 11(c), it can be seen that flow is fairly well devel-
tube may be estimated from oped near channel exit, indicating Eq. (32) may not be
accurate for rectangular channels. For most of the
Lh ¼ 0:05Re d; ð31Þ channel length, the average heat flux and average
2562 W. Qu, I. Mudawar / International Journal of Heat and Mass Transfer 45 (2002) 2549–2565

Fig. 11. Numerical predictions of average heat transfer characteristics for q00eff ¼ 100 W=cm2 and Re ¼ 890: (a) average temperature,
(b) average heat flux, (c) average Nusselt number.

Nusselt number at the channel bottom wall (Wall a) are atures were measured at six points, which include the
smaller than the corresponding values at the channel channel inlet and outlet as well as four locations inside
side wall (Wall b). the heat sink. Due to the limited temperature data and
The theoretical values of Nusselt number for rect- the complex nature of the conjugate heat transfer
angular channels subjected to different thermal bound- problem, the local and average Nu and q00 cannot be
ary conditions were summarized by Shah and London determined without the numerical solution.
[25]. It would be ideal to compare the experimental re- Figs. 12(a)–(c) show heat transfer characteristics for
sults with either those theoretical Nusselt numbers or the the conditions of q00eff ¼ 200 W=cm2 and Re ¼ 864. The
present numerical predictions. Unfortunately, the pre- average temperatures and average heat fluxes in
sent experiments alone cannot provide sufficient infor- Figs. 12(a) and (b), respectively, have the same trends
mation to fully characterize the local and average as those presented in Figs. 11(a) and (b), but higher
Nusselt number or heat flux. During the tests, temper- values. The average Nusselt number in Fig. 12(c) is
 W. Qu, I. Mudawar / International Journal of Heat and Mass Transfer 45 (2002) 2549–2565 2563

Fig. 12. Numerical predictions of average heat transfer characteristics for q00eff ¼ 200 W=cm2 and Re ¼ 864: (a) average temperature,
(b) average heat flux, (c) average Nusselt number.

nearly identical to the one presented in Fig. 11(c). This heat flux, and Nusselt number. Key findings from the
is because the Nusselt number for laminar flow is de- study are as follows:
termined solely by the channel geometry and the local 1. The measured pressure drop is in good agreement
flow conditions. with the predicted values. A gradual slope change
in the pressure drop variation with Reynolds number
is attributed to the temperature dependence of water
4. Conclusions viscosity, and the increasing contraction and expan-
sion pressure losses at the channel inlet and outlet, re-
The fluid flow and heat transfer characteristics of a spectively, with increasing Reynolds number.
micro-channel heat sink were investigated both experi- 2. The early transition from laminar to turbulent flow
mentally and numerically. Also presented and discussed cited in previous micro-channel studies was never ob-
was a detailed description of the local and average heat served in the present study within the tested Reynolds
transfer characteristics of the heat sink, i.e. temperature, number range 139–1672.
2564 W. Qu, I. Mudawar / International Journal of Heat and Mass Transfer 45 (2002) 2549–2565

3. Higher Reynolds numbers are beneficial at reducing [6] R.W. Knight, D.J. Hall, J.S. Goodling, R.C. Jaeger, Heat
both the water outlet temperature and the tempera- sink optimization with application to microchannels, IEEE
tures within the heat sink, alas at the expense of Trans. Components, Hybrids, Manufac. Technol. 15
greater pressure drop. (1992) 832–842.
[7] A. Bejan, A.M. Morega, Optimal arrays of pin fins and
4. Very good agreement was achieved between the ex-
plate fins in laminar forced convection, ASME J. Heat
perimental data and numerical predictions for the Transfer 115 (1993) 75–81.
temperature distribution within the heat sink. [8] D.Y. Lee, K. Vafai, Comparative analysis of jet impinge-
5. At any longitudinal distance x along the micro-chan- ment and microchannel cooling for high heat flux appli-
nel, the bulk liquid constitutes the region of lowest cations, Int. J. Heat Mass Transfer 42 (1999) 1555–1568.
temperature in the unit cell, while the highest temper- [9] A. Weisberg, H.H. Bau, J.N. Zemel, Analysis of micro-
ature is encountered in the bottom wall (Wall A) im- channels for integrated cooling, Int. J. Heat Mass Transfer
mediately below the channel outlet. The water bulk 35 (1992) 2465–2474.
temperature and average temperatures at the channel [10] A.G. Fedorov, R. Viskanta, Three-dimensional conjugate
bottom wall (Wall a) and top wall (Wall c) increase heat transfer in the microchannel heat sink for electronic
packaging, Int. J. Heat Mass Transfer 43 (2000) 399–415.
fairly linearly along the longitudinal direction.
[11] T. Kishimito, T. Ohsaki, VLSI packaging technique using
6. The channel top wall (Wall c) can be assumed adia- liquid-cooled channels, IEEE Trans. Components, Hy-
batic due to the small thermal conductivity of the brids, Manufac. Technol. CHMT-9 (1986) 328–335.
plastic cover plate. For the channel bottom wall [12] S. Sasaki, T. Kishimito, Optimal structure for microgro-
(Wall a) and side wall (Wall b), much higher heat flux oved cooling fin for high-power LSI devices, Electron. Lett.
and Nusselt number values are encountered near the 22 (1986) 1332–1334.
channel inlet. The heat flux and Nusselt number also [13] D. Nayak, L.T. Hwang, I. Turlik, A. Reisman, A high
vary around the channel periphery, approaching zero performance thermal module for computer packaging, J.
near the corners. Electron. Mater. 16 (1987) 357–364.
7. Excellent agreement between both the pressure drop [14] M.M. Rahman, F. Gui, Experimental measurements of
fluid flow and heat transfer in microchannel cooling
and heat sink temperature data and corresponding
passages in a chip substrate, Adv. Electron. Packaging
numerical predictions proves the conventional Na- ASME EEP-4 (1993) 495–506.
vier–Stokes and energy equations can accurately pre- [15] M.M. Rahman, F. Gui, Design, fabrication, and testing of
dict the heat transfer characteristics of micro-channel microchannel heat sinks for aircraft avionics cooling, in:
heat sinks for the channel size employed in the pre- Proceedings of the 28th Intersociety Energy Conversion
sent study and recommended for practical cooling ap- Engineering Conference, vol. 1, 1993, pp. 1–6.
plications. [16] T.S. Ravigururajan, J. Cuta, C.E. McDonald, M.K. Drost,
Single-phase flow thermal performance characteristics of a
parallel micro-channel heat exchanger, in: National Heat
Transfer Conference, vol. 7, ASME HTD-329, 1996, pp.
Acknowledgements 157–166.
[17] T.M. Harms, M.J. Kazmierczak, F.M. Cerner, A. Holke,
The authors are grateful for the support of the Office H.T. Henderson, J. Pilchowski, K. Baker, Experimental
of Basic Energy Sciences of the US Department of En- investigation of heat transfer and pressure drop through
ergy (Award No. DE-FG02-93ER14394 A7). deep microchannels in a (1 0 0) silicon substarte, in:
Proceedings of the ASME Heat Transfer Division, HTD-
351, 1997, pp. 347–357.
[18] K. Kawano, K. Minakami, H. Iwasaki, M. Ishizuka,
References Micro channel heat exchanger for cooling electrical equip-
ment, Appl. Heat Transfer Equip., Syst. Educ. ASME
[1] D.B. Tuckerman, R.F.W. Pease, High-performance heat HTD-361-3/PID-3 (1998) 173–180.
sinking for VLSI, IEEE Electron. Dev. Lett. EDL-2 (1981) [19] T.M. Harms, M.J. Kazmierczak, F.M. Cerner, Developing
126–129. convective heat transfer in deep rectangular microchannels,
[2] R.W. Keyes, Heat transfer in forced convection through Int. J. Heat Fluid Flow 20 (1999) 149–157.
fins, IEEE Trans. Electron Dev. ED-31 (1984) 1218–1221. [20] M.M. Rahman, Measurements of heat transfer in micro-
[3] V.K. Samalam, Convective heat transfer in microchannels, channel heat sinks, Int. Commun. Heat Mass Transfer 27
J. Electron. Mater. 18 (1989) 611–617. (2000) 495–506.
[4] R.J. Phillips, Micro-channel heat sinks, in: A. Bar-Cohen, [21] S.V. Patankar, Numerical Heat Transfer and Fluid Flow,
A.D. Kraus (Eds.), Advances in Thermal Modeling of Hemisphere, Washington, DC, 1980.
Electronic Components, vol. 2, ASME Press, New York, [22] S.V. Patankar, A numerical method for conduction
1990, pp. 109–184. in composite materials, flow in irregular geometries
[5] R.W. Knight, J.S. Goodling, D.J. Hall, Optimal thermal and conjugate heat transfer, in: Proceedings of 6th Inter-
design of forced convection heat sinks – analytical, ASME national Heat Transfer Conference, vol. 3, 1978, pp. 297–
J. Electron. Packaging 113 (1991) 313–321. 302.
 W. Qu, I. Mudawar / International Journal of Heat and Mass Transfer 45 (2002) 2549–2565 2565

[23] N.E. Todreas, M.S. Kazimi, Nuclear Systems I, Hemi- [25] R.K. Shah, A.L. London, Laminar Flow Forced Convec-
sphere, New York, 1990. tion in Ducts: A Source Book for Compact Heat Ex-
[24] F.P. Incropera, D.P. DeWitt, Fundamentals of Heat and changer Analytical Data, Suppl. 1, Academic press, New
Mass Transfer, fourth ed., Wiley, New York, 1996. York, 1978.