You are on page 1of 12

Int. Comm. Heat Mass Transfer, Vol. 27, No. 2, pp.

229-240, 2000
Copyright 2000 Elsevier Science Ltd
Pergamon Printed in the USA. All rights reserved
0735-1933/(D/S-see front matter

PH S0735.1933(00)00104-4

DESIGN OF AN o F r I M A L PIN-FIN HEAT SINK WITH AIR IMPINGEMENT


COOLING

J.G. Maveety and H.H. Jung


IA-64 Processor Division
Intel Corporation
2200 Mission College Blvd.
M/S: SC12-201
Santa Clara, California 95052, USA

(Communicated by J.P. Hartnett and W.J. Minkowycz)

ABSTRACT
A comparative investigation between experiments and nmnefical simulation is
presented for turbulent air impingement flow on a square pin-fin heat sink.
Experiments were conducted using an aluminum heat sink subjected to a uniform heat
flux by using a silicon test chip. The numerical method incorporates the k - z
turbulence model. The numerical results illustrate the complex pressure gradients
within the fin array. The predicted thermal resistance of the heat sink agrees well with
those obtained from experiments. This study illustrates the utility of numerical
experiments in the design and optimization of heat sinks. 20ooElsevier Science Ltd

Introduction

Advances in microprocessor physics along with continued efforts to reduce operational


execution time has resulted in length scale reductions of semiconductor devices. However, as the
physical size of the semiconductor devices has decreased, the heat flux levels within these devices
has increased. An important trend that has exacerbated the thermal management problem involves
the addition o f circuit layers and gates on the silicon substrate. With device sizes continuing to
decrease the thermal management problems associated with maintaining a safe device operating
temperature become more prevalent. Such increases in the microprocessor designs has elevated
the importance of heat dissipation and thermal designs [1 ].
229
230 J.G. Maveety and H.H. Jung Vol. 27, No. 2

Designers are attacking thermal performance problems using both passive and active thermal
control systems [2]. Advancements in material technology [3,4], heat sink optimization techniques
[5] and the use of heat pipes have led to better thermal control. Active techniques such as the use
of thermoelectric coolers [6] and closed-loop cooling systems [1] has provided integrated
solutions for numerous applications.

The aim of this paper is to present a computational analysis illustrating pressure and heat
transfer characteristics in an optimized heat sink. In the discussion that follows both the details of
the optimization technique and numerical analysis are presented. These include the governing
equations, computational method and results. A comparison between numerical and experimental
values of temperature are represented in terms of sink-to-ambient thermal resistance.

Numerical Method

The k - 6 turbulence model is used to describe the characteristics of impinging air flow onto a
heat sink. The assumptions invoked here are: incompressible flow (Ma<0.3), buoyancy and
radiation heat transfer effects negligible and constant thermodynamic properties. Assuming that
the flow variables can be expanded in the form f = f + f ' where f is the mean value and f ' is a

fluctuation about the mean, then the continuity and Navier-Stokes equations can be written as

~' = 0 (1)

Ou, +~- tTffi - 1 -off


- +-/'tV2~ - - -3R~ (2)
gt ~; p a3ci p 3x,
where R,) = u'u'j is the Reynolds stress term. A turbulence model which is less complicated than

the full equations above can be performed by substituting a simper representation for the
Reynolds stress term. Models which have been used with wide success for engineering analysis
are the two-equation boundary-layer models [7]. These models postulate a relationship for the
Reynolds stress in terms o f the mean flow such as

-R = kt' ( cgi' + vuWij)-2r~,jk (3)


p/,3xj ~,
Vol. 27, No. 2 DESIGN OF AN OPTIMAL PIN-FIN HEAT SINK 231

where/a, is the turbulent viscosity andk = (u:uj)/2 is the turbulent kinetic energy. Equation (3)

introduces two unknowns ( ~ , and k ) which require two equations for closure. Using dimensional

analysis, it can be shown for high Reynolds number flows that the turbulent viscosity is
proportional to the energy o f dissipation rate. This can be represented as
k2
/z, _ C~ (4)
p c
where C 1 = C1 (Re) and e is the dissipation rate o f energy. The closure equations used to evaluate
the turbulent viscosity and energy dissipation rate follow an analogy with classical energy
transport equations and take the form

Dk 1 0 I~t ( c~, + c~,


D~- - p dr k C2/~t + - ~ (5)

De 1 d (C3/6 -~-k) + C /6 e ( c ~ , + c~-,) c~- e2 (6)


Dt p ~xk 4 --P--k'~'-~k-k ~ICI ~k-k -65 k
For the I - 6 model, the (7. terms take the values

C~ = 0.09 C 2 = 1.0 C 3 = 0.769 C 4 = 1.44 Q = 1.92 (7)


when modeling plane jets and mixing layers [7]. The energy equation that is solved for the
conduction heat transfer within the heat sink is

PC~51"~-dt ~,d(k'-~,,)+( 1 (8)

where q is the heat generated per unit volume within the test chip, k s is the heat sink thermal

conductivity and T, is the temperature within the heat sink. The energy equation that is solved for
the fluid flow is

pC--z-
ctC~+ u' CgF, - o3:
d , ( k ~ - T'u' ) (9)

The boundary conditions invoked on the walls o f the heat sink are: all velocity components are
zero (no slip), conservation o f energy between conduction and convection, natural boundary
conditions on pressure and wall function for the turbulence model variables. For the inflow:
specified velocity which has been matched to experiments, natural boundary conditions on
pressure, specified turbulence model variables and specified ambient temperature and pressure.
232 J.G. Maveety and H.H. Jung Vol. 27, No. 2

For the outflow: specified pressure, temperature and natural botmdary conditions on all other
variables. Symmetry was invoked along the centerlines of the heat sink. The natural boundary
conditions are the flux terms of the dependent variables These terms are automatically calculated
by Flotherm.

Numerical experiments were performed using the Flotherm CFD package produced by
Flomerics. A grid structure fine enough to give grid independent results was determined. A
variable grid structure was used in order to capture the resolution necessary to resolve steep
pressure, temperature and velocity gradients. The high resolution grid was only necessary in the
impingement area of the heat sink.

Experimental Method

The heat sink studied here was constructed from aluminum. The heat sink had square pin-fins
(see Fig. 1) and a 7 x 7pin-fin army. As shown in Fig. l(a), the air supplied from the air delivery
system passed through a filter, pressure regulator and flow meter before entering the nozzle. All
of the experiments were conducted using a round nozzle of diameter D = 6.4ram. The nozzle-to-
sink spacing used throughout this work was z / D = 10. As discussed in [4], the best cooling
performance occurs when the nozzle-to-sink spacing is within the range 8 < z / D < 12.

Figure l(b) shows more heat sink details. The heat sink had a square base
measuring 50.8ram x 50.8ram. The fins were square (W x gO and had height H. A silicon test chip

having dimensions of 12 x 12 x 3ram and capable of generating 30W was used. All measurements
were collected using a NetDaq data-acquisition system. For each data point the nozzle height was
set, and the air flow rate and chip-power adjusted. The test setup was allowed to reach steady-
state prior to taking any data. A measurement capability analysis (MCA) study was performed to
ensure the testing conditions would produce repeatable and reproducible results. Figure 1 (e) is
the plan view showing the channel spacing numbering scheme. The numbering scheme
supplements our pressure and flow discussion presented below.
Vol. 27, No. 2 DESIGN OF AN OPTIMAL PIN-FIN HEAT SINK 233

Flow

Filter ---.
z~
!
-- To, Po

jet

Processor B e a r d ~

w
Y~ 3 2 1
~TJ[J L J U*t t_ 1
~[~
2
3
[3 E3 E3 ~ x
Y I (c)
H
I~ E3 E3 ~
9Fql~ Vq ~ Fq r-

FIG. 1
(a) Schematic illustrating the experimental setup, (b) heat sink geometry and (c) top view showing
channel numbering and coordinate system.

Uncertainty estimates in the experimental quantities were determined using the ANSI/ASME
Standard [8] which accounts for both bias and precision errors. All bias and precision
components were estimated at 95% confidence. Experimental uncertainty estimates were

determined to be; 0.2C in heat sink temperatures, 0.1 C in silicon junction temperature, 3.2% in
device power and 1.67 x 10-sin 3 / sin air flow rate. The Reynolds number is based on nozzle
diameter. The uncertainty in the Reynolds number decreases with increased flow rate and the
uncertainty in Rio is consistently between 3 and 3.5%.

The heat transferred t~om a heat-dissipating chip to the ambient is a complicated conjugate
process depending on flow conditions, geometry and thermophysical properties o f the working
234 J.G. Maveety and H.H. Jung Vol. 27, No. 2

fluid. However, for practical reasons, it is common in the thermal characterization of steady-state
conditions to quantify the die-to-ambient thermal resistance as

8j.= Q- T (10)

where ~ is the die junction ternperature, TOis the ambient temperature and Qis the thermal power

dissipated from the die.

The optimization study developed by Ledzma et al.[5] was shown to be successful for low
Reynolds number flows. Maveety and Hendricks [4] used the technique and showed it could be
extended to higher Reynolds number flows. In this work, we investigate the sensitivity of fin
geometry on heat dissipation. In addition we illustrate the effects fin geometry has on the
pressure field. Referring to Fig. 1(c), the flags 1, 2 and 3 represent the heat sink channel locations
in which the pressure distribution was investigated. Here we represent the pressure in
dimensionless form as

P-P (11)
Cp ~ 2
poUo / 2

where Po and u o are the pressure and mean nozzle velocity, respectively.

Results and Discussion

Figure 2 shows the comparison between numerical and experimental values of Oj~ for

the7 x 7heat sink geometry over the range7,800 < Re D < 19,700. Figure 2 shows poor
agreement between numerical and experimental values for low Reynolds number flow. The over
prediction in thermal performance indicates that the C, values need to be adjusted from those
listed in equation (7) for lower Reynolds number flows. However, the data is in excellent
agreement forRe o > 11,000. An important feature illustrated in Fig. 2 is that there is little gain in
cooling performance for flows in excess of 15,000. This is important to the computer industry
because it allows for a reduction in pumping power and noise level without any loss in cooling
performance.
Vol. 27, No. 2 DESIGN OF AN OPTIMAL PIN-FIN HEAT SINK 235

2.5
--e Numerical
A
2 I " Experimentai
\
e= 1.5
IT' \

o, - 1
I-

0.5

0
5000 10000 15000 20000 25000
Rea

FIG. 2
Comparison between numerical and experimental values of sink-to-ambient thermal resistance 0j~.

Numerical experiments were also performed to assess the effects of fin height on cooling
performance. Guided by the work of Ledezma et al.[5] the optimal fin height is obtained when
H~ L = 053. This corresponds to a fin height of 15ram when the hase-plate is 12ram thick. In
the computational study we increased and decreased the fin height by 6ram in order to quantify
the impact of fin height on cooling performance. The data is represented in Fig. 3. Clearly the
shorter fin does not provide an optimal cooling condition. However Fig. 3 shows that the cooling
capability is extremely sensitive as the fin length is increased from 9ram to 15ram. The cooling
performance is increased by 13.8% over this range. The longer fin has slightly better performance
(2.13%), over the 15ram fin, but the gain could be offset by material and manufacturing costs.

1.1

I !
I
I

|
el

0.9 L . . . . . . . .
0.4 0.5 0.6 0.7
H/L
FIG. 3
Numerical experiments showing the effects of fin height on sink-to-ambient thermal resistance 0i~.
236 J.G. Maveety and H.H. Jung Vol. 27, No. 2

The next objective o f the optimization study was to assess the effect o f fin width (i.e., fin
cross-sectional area) on cooling. Using the technique discussed in [5], the optimal fin width was
determined to be W / L = 0.1. In this study we increased and decreased the fin width by lmm.
The cooling characteristics are shown in Fig. 4. As the cross-sectional area o f the fins are
reduced, the thermal mass o f the heat sink is reduced. This in turn reduces the heat sinks capacity
to dissipate heat. As the fin cross-sectional area is increased over the range 0.08 _<W / L < 0.1
there is a 95% increase in cooling capability. As the fins are allowed to increase
beyond W / L = 0.1 the cooling capability starts to degrade again. The degradation in cooling over
this range is attributed to the narrow channel spacing between the fins. As the channel width is
reduced (fin cross-sectional area is increased) less air flows between the fins and more air is
deflected in the z-direction away from the heat sink into the ambient surroundings. The increased
resistance to air flow increases the overall heat sink temperature.

1.9
1.7
\\\ Numerical
~ 1.5 \\
\
1.3 \
m \\
1.1
\
\
0.9

0.7
0.5
0.06 0.08 0.1 0.12 0.14
WIL

FIG. 4
Numerical experiments showing the effects o f fin width on sink-to-ambient thermal resistance Oj=.

Figures 5(a) - (f) show the pressure profiles o f channels 1, 2, and 3 starting from the base o f
the beat sink channel (z = 0) to the respective fin height (z = 9mm, ere). Figures 5(a) - (c)
illustrate the effects o f fin height on channel pressure and Figs. 5(d) - (f) illustrate the effects o f fin
thickness on channel pressure.

The flow under the nozzle in the pin fin array is mostly in the downward direction due to the
strong momentum. The high fluid velocity creates the large pressure gradients due to circulation
and eddies resulting from the flow changing from a vertical direction to a horizontal direction.
Vol. 27, No. 2 DESIGN OF AN OPTIMAL PIN-FIN HEAT SINK 237

The large pressure gradient within the heat sink centerline channels induces mixing o f the air
which produces a more uniform air temperature and higher heat transfer coefficient.

25 H = 9mm, W = 5.08mm (d) H = 15ram, W = 4.08ram


25
20 20
15 ~1~~'~"&~ / 15
10 / ~ I L ~ ' ~ ~A. -~._ .~ 10
5
o : ~ -~-~
~ ~.-,
-5 a
-10 ~coll I -10 ~co11
-is --m-co~l -15 ~co12
-20 -20

H = 15mm, W = 5.08mm H = 15ram, W -- 5.08mm


25 25
20 2O
15 15
10

o, ~ b~_~..~.._ o: ~ ~ t r - ~ , .~__./.
-5 1 \'~ --/ / 5 )
\1
-10 - 0 - - oo11 \/ -10 --e-- call ~
-15 - - n - - col2 -15 ~ B - - cd2
-20 + col3 -20 ~ cd3

H = 21mm, W = 5.08mm 25 W = 6.08ram


25
2O 2O
15 15
10 10

0
-5 5

-10 -10 ~ cdl


- O - - co11
-15 -15 ~ cd2
co43 "~ cd3
-20 -20
Channel Height (mm) Channel Height (ram)

FIG. 5
Cp versus channel z-height; Re D = 15,700.

The largest percentage o f air flows along the heat sink centerline channels (col 3) as indicated
by the large values of Cp. Figure 5(b) does provide insight into why the H = 15mm,W = 5.08ram

fin geometry is the most optimal The pressure gradient in channel 2 is the largest for this
geometry. This implies that more air is directed through channel 2 which enhances the heat
238 J.G. Maveety and H.H. Jung Vol. 27, No. 2

transfer from the heat sink. The H = 21mm, W = 5.08ram heat sink illustrated in Fig. 5(c) also
shows evidence of a high pressure drop along channel 2. Recall from Fig. 3 that this fin
configuration produced the most optimal 0jo results.

Both t h e H = 9mm, W = 5.08ram (short fin) a n d H = 15mm,W = 4.0gram (wide channel) heat
sinks produce large pressure gradients along the centerline channel. These large gradients
produce flow reversals and stagnation points. In addition, the pressure profiles within channels 1
and 2 are small indicating little air is passing through these channels. This information helps
explain the poor thermal performance for these two particular fin geometries. Finally, the small
pressure patterns shown in Fig. 5(0 result from reducing the flow resistance by increasing the fin
spacing.

Figures 6 and 7 summarize the centerline pressure distributions for the six geometries
investigated. Representing the abscissa in units of dimensionless channel height, Fig. 6 shows
how the centerline pressure varies as the fin height is changed. Over the range 0 _<z / Zm~ < 0.8,
the pressure increases with decreasing fin height. The increase in pressure resulting from
deceleration of the flow increases the overall heat transfer coefficient. However the convective
thermal resistance is the reciprocal of the hA -product, where A is the heat sink surface area and
h is the area-average heat transfer coefficient. In the range 0.8 _<z / Zm~ < 1 there is evidence of

flow reversal and recirculation occurring within all three geometries.

25
20
15
10
5
0
-5 2

-10
-15
-20
-25
Z~Zmax
FIG. 6
Cpversus dimensionless channel z/z,~ for W = 5.08ram and Re D = 15,700.
Vol. 27, No. 2 DESIGN OF AN OPTIMAL PIN-FIN HEAT SINK 239

Figure 7 illustrates how the centerline pressure varies as the fin diameter is changed. As can be
seen, the optimized fin diameter (W = 5.08ram) generates the largest pressure distribution over
the range 0 -< z / Zm~ < 1. It is through the optimization o f both the flow conditions and heat sink

mass distribution that the lowest overall thermal resistance can be achieved.

20 F
l i T ~ [ 1
15
10
5

d OI
2
-5
-10 t vw-'*.uu i ~ \ ? ~ i
i ~ w-s.o8
-15
-20
-4 ~ W=6"08~ i 1 !1 !

Z/Zmax

FIG. 7
Cpversus dimensionless channel Z/Zmaxfor H = 15mm and Re D = 15,700.

Conclusions

This paper presents comparisons between computational and experimental results for heat
transfer from a heat sink array. The results show excellent agreement between Flotherm and the
experimental results f o r R % > 11,000. The beat transfer results indicate that t h e k - e model is

adequate for predicting Oja. In addition, optimization studies were performed to quantify the

effects o f changing the fin length and fin cross-sectional area on cooling performance. The results
show that the cooling performance can be greatly affected by minor changes in fin dimensions.

Computations were carried out to investigate the flow characteristics within the fin array. The
flow is observed to be very strong along the array centerline channels o f the heat sink, and
diminishes in the channels away from the heat sink centerline. The predicted values o f pressure
drop show complex flow patterns with large pressure gradients that generate vorticity, circulation
and flow reversals.
240 J.G. Maveety and H.H. Jung Vol. 27, No. 2

The present study demonstrates the utility o f computational analysis in both optimizing a heat
sink geometry, and providing physical insight into the flow and heat transfer characteristics o f a
heat sink

Nomenclature

C p = Pressure coefficient, ( P - Po) / (1/2poUo)


D = Nozzle diameter, m
H = Fin height, m
P = Pressure, N / m 2
Q = Heat transfer from fin array, W
ReD = UoD / v , Reynolds number based on nozzle diameter
TO = Ambient air temperature, o c
uo = Air free stream velocity exiting nozzle, m / s
W = Fin width, m
Po = Density o f air, kg / m 3
OjQ = ( Tj - To) / Q , junction-to-ambient thermal resistance, C / W

References

1. M. Malinoski, J. Maveety, S. Knostman and T. Jones, IEEE Proceedings o f the International


Test Conference, 119 (1998)
2. F.P. Ineropera, Bulletin o f the International Center f o r Heat and Mass Transfer, 3, 1261
(1972)
3. M. Montesano, Materials Technology, 3, 87 (1996)
4. J.G. Maveety and J.F. Hendricks, A S M E J. Electronic Packaging, 121, 156 (1999)
5. G. Ledzma, A.M. Morega and A. Bejan, A S M E J . Heat Transfer, 118, 570 (1996)
6. R.L. Webb, M.D. Gilley and V. Zarnescu, .4SME J. Electronic Packaging, 120, 98 (1988)
7. B.E. Launder and D.B. Spalding, Computer Methods in Applied Mechanics and Engineering,
3, 269 (1974)
8. ANSUASME PTC 19.1-1985, Part 1 Measurement Uncertainty - Instruments and
Apparatus, (1986)

Received October 5, 1999

You might also like