Sie sind auf Seite 1von 11

832

IEEE TRANSACTIONS ON COMPONENTS, HYBRIDS, AND MANUFACTURING TECHNOLOGY, VOL. 15, NO. 5, OCTOBER 1992

Heat Sink Optimization with Application to Microchannels


Roy W. Knight, Donald J. Hall, John S. Goodling, and Richard C. Jaeger, Fellow, IEEE
Abstruct- The equations governing the fluid dynamics and combined conduction/convection heat transfer in a heat sink are presented in dimensionless form for both laminar and turbulent flow. A scheme presented for solving these equations e nables the determination of heat sink dimensions that display the lowest thermal resistance between the hottest portion of the heat sink and the incoming fluid. Results from the present method are applied to heat sinks reported by previous investigators mckerman and Pease [l], [2], Goldberg [3], and Phillips [4], [ 5 ] ) to study effects of their restrictions regarding the nature of the flow (laminar or turbulent), the ratio of fin thickness to channel width, or the aspect ratio of the fluid channel. Present results indicate that when the pressure drop through the channels is small, laminar solutions yield lower thermal resistance than turbulent solutions. Conversely, when the pressure drop is large, the optimal thermal resistance is found in the turbulent region. With the relaxation of these constraints, configurations and dimensions found using the present procedure produce significant improvement in thermal resistance over those presented by all three previous studies. These improvements range from 10 to 35% in values for design thermal resistance.

NOMENCLATURE

Total mass flow rate of coolant through channels. Number of cooling channels. Pressure difference number, ( Ap/ L) W3/ (p u 2 ) . Work rate number, tbW/ (pu3). Nusselt number, h D h / k f i u i d . Nusselt number for fully developed flow. Fin perimeter. Pressure drop through the heat sink channels. Prandtl number, u / a . Heat source power. Reynolds number based on hydraulic diameter. Temperature. Largest temperature difference between coolant and source. Improvement in thermal resistance. Mean fluid velocity. Volumetric flow rate. Pumping power. Width of heat sink. width of channel, I?. Width of fin, re.

A B C
CP

C1 D
De,

Dh

G h

k
L
m

Area. Constant defined in (31). Constant defined in (32). Specific heat at constant pressure. Coefficient defined as n r ( n - 1). Depth of heat sink. Equivalent laminar diameter of the fluid flow channel. Hydraulic diameter of the fluid flow channel. Friction factor. Parameter defined in (23). Heat transfer coefficient. Thermal conductivity. Channel width. Length of heat sink in the direction of fluid flow.

Greek Symbols Thermal diffusivity of fluid or aspect ratio. Q Percent of infinite fin performance. P Coefficient defined by (24). 7 1 Ratio of fin thickness to channel width. r rl Fin efficiency. e Thermal resistance, AT/q. 0 Dimensionless thermal resistance,
(AT/Q)/huidW.
U

Kinematic viscosity of fluid. Mass density.

[(hPfin/kfinAc,fin)] 1'2*

Manuscript received August 31, 1991; revised June 25, 1992. This work was supported by the National Science Foundation under Grant CBT-8805607, by the Alabama Microelectronics Science and Technology Center of Aubum University, and by the Alabama Research Institute under Grant ARI-90-201. R. W. Knight and J. S. Goodling are with the Mechanical Engineering Department and the Alabama Microelectronics Science and Technology Center, Auburn University, Aubum, AL 36849. D. J. Hall was with the Mechanical Engineering Department, Aubum University, Auburn, AL. He is now with Compaq Computers, Houston, TX 77070. R. C. Jaeger is with the Electrical Engineering Department and the Alabama Microelectronics Science and Technology Center, Aubum University, Auburn, AL 36849. IEEE Log Number 9202704.

Subscripts base Fin base. Cross sectional available for flow. C C, fin Cross-sectional of fin. ch Channel. f,i Fluid inlet. f ,o Fluid outlet. fin Fin. fluid Fluid. h Hydraulic. Surface available for heat transfer. S Surface at the fluid inlet face. S, i Surface at the fluid outlet face. s,o

I. INTRODUCTION

HE advent of high density components has required investigation of innovative techniques for removing heat

01484411/92$03.00 0 1992 IEEE

KNIGHT et al.: HEAT SINK OPTIMIZATION

833

from these devices [l]. One method is through the use of forced convection heat spreaders called a heat sinks. This paper is inspired by the pioneering heat sink work offered first by Tuckerman and Pease [l] , wherein they reported a method for cooling a chip by forcing coolant through closed channels etched onto the backside of a silicon wafer. They observed that for laminar flow in a channel, the heat transfer coefficient is inversely proportional to the channel width. A minimum in the thermal resistance expression was used to size the coolant channels subject to several restrictions. A sample heat sink was fabricated and tested with a resulting modest chip temperature rise above ambient for very high flux levels of thermal input. This paper serves to show that, under certain circumstances, conditions imposed in previous studies are not desirable in optimizing the fin configurations subject to a specified pressure drop through the heat sink channels. Since the problem presented here lends itself to generalization through nondimensionalization, the technique described herein is applicable to any closed, forced convection heat exchanger.

Sasaki and Kishimoto [7] optimized the channel dimensions of a finned heat sink constructed on a silicon chip for a given pressure drop. The Wch/Wfin ratio again was restricted to unity, and the optimal channel widths were found to be 400 and 250 pm for a pressure drop of 200 and 2000 kg/m2, respectively, again subject to the laminar flow constraint. The experimental results were claimed to match the analysis well; however, the latter was not presented. Kishimoto and Ohsaki [8] discussed a packaging technique wherein VLSI chips are mounted on a multilayered alumna substrate. Coolant channels (800 pm wide by 400 pm high) were made in the substrate at a staggered pitch of 2.54 mm. A thermal resistance of O.3l0C/W was obtained for a pressure drop of 19.6 kPa and flowrate of 1 L/min of water. Hwang et al. [9] designed a multichip, water cooled module suitable for VLSI packaging. An analytically designed 25 chip module was modeled by a 9 chip experimental module with the coolant flow rate scaled to match the 25 chip model. The laminar and turbulent flow regimes were considered with the latter yielding the least thermal resistance. The turbulent case yielded a maximum temperature rise per chip of 18C at a 11. REVIEW OF LITERATURE chip power level of 42 W. A pressure drop of 55.2 kPa with a In the original work by Tuckerman and Pease [l], the flow rate of 126 cm3/s was experienced through the heat sink optimization of the heat sink design was done subject to with channel dimensions of 5870 pm wide by 1000 pm high, several constraints: the flow through the channels was fully separated by a wall 1270 pm thick. This design is significantly developed and laminar in nature; the laminar Nusselt number different from the fin concept utilized in this paper because the and the channel to fin width ratio (W&/Wfin) were fixed, as channels are parallel to the heat source instead of perpendicular were the pressure drop through the fin array, pumping work, to it. Phillips [4] reported an innovative design method for water planar dimensions, and fin efficiency. Subject to the designed dimensions, a heat sink was constructed in a silicon wafer and cooled, microchannel heat sinks in which both laminar and a heat flux of 790 W/cm2 was achieved with a temperature turbulent flow regimes were considered for hydrodynamically developing and thermally fully developed flow where rise of 71C using water as the coolant. Since that landmark paper, many studies have been pub- restrictions were placed only on the Wch/Wfin ratio and the lished using a similar optimization technique, but with few aspect ratio. All assumptions made by Phillips are thoroughly variations on the constraints. Most notably are the works of discussed in his paper. Most results are presented pictorially Goldberg [3], Mahalingam [6], Sasaki and Kishimoto [7], with thermal resistances displayed as a function of channel Kishimoto and Ohsaki [8] Hwang et al. [9] and Phillips [4]. width. For the test case presented, the design which allowed With the exception of the latter two referenced works, all were turbulent flow yielded the smallest thermal resistance. Bar-Cohen and Jelinek [lo] optimized arrays of air cooled iterations on the theme set forth by the Tuckerman and Pease [ l ] and showed that such a design has practical applications rectangular fins to maximize heat transfer or minimize fin mass. In their scheme, fin material, air flow rate, available and can in fact be implemented. Goldberg [3] constructed three air cooled, forced convection pressure head, and either the fin depth or thickness is specified, heat sinks and tested each one. Each heat sink had a different and the remaining fin dimension (depth or thickness) and fin thickness, with the Wch/Wfin ratio restricted to unity, and number of fins are determined by a computerized search the flow limited to the laminar regime. The air flow for each procedure. In this manner, optimal designs of two air cooled heat sink was adjusted to provide a rate of 30 L/min. As fin systems are reported. Landram [ l l ] identified optimal configurations of heat sinks expected, the design with the largest pressure drop and smallest channel width yielded the smallest thermal resistance. by the use of a computational scheme. The temperature profiles Mahalingam [6] constructed a microchannel heat sink in for both the cooling fluid and the conductive heat sink were a 5 cm by 5 cm by 0.20 cm silicon substrate. The channels simultaneously determined. Optimal designs were found to were about 200 pm wide with a depth of 1700 pm, separated be strong functions of coolant to wall thermal conductivity by a 100 bm fin. A 3.8 cm by 3.8 cm thin film heat source ratio and channel aspect ratio. The flow considered was fully was mounted on the topside and experiments were carried out developed and laminar in nature. using water and air as coolants. The water experiments, at a In a recent publication, Knight, Goodling, and Hall [12] flux of 1100 W , yielded thermal resistances of O.O3C/Wand present an optimization method for sizing coolant channels O.O2OC/W for flow rates of 12cm3/s and 63 cm3/s, respec- for a given pressure drop or pumping power, given platively. Results for the air experiments were also presented, but nar dimensions, specified fin efficiency or fin length, and yielded higher thermal resistances, as expected. coolant and solid properties. The governing equations were

834

IEEE TRANSACTIONS ON COMPONENTS, HYBRIDS, AND MANUFACTURING TECHNOLOGY, VOL. 15, NO. 5, OCTOBER 1992

AT
Fig. 1. Schematic of the fin system.

made dimensionless, and characteristic groups (not unlike the Reynolds number) were formed. A n analytical solution was presented for the idealized case of infinitesimally thin fins which showed that the optimal turbulent case offers significantly better thermal resistance than the optimal laminar flow for some pressure drops. The optimization technique is then applied to a heat sink with fins of finite thickness with some simplifying assumptions, most notably of which is that of fully developed flow. The design technique is applied to the designs of Tuckerman and Pease, and Goldberg. In that same paper [12], it was shown that the optimal solution occurs in the laminar regime when the pressure drop through the fin array is low and in the turbulent region when that pressure drop is high. In this paper, the optimization method delineated in [12] is further generalized and includes developing flow and better heat transfer correlations. Results using the present method are compared to those of three previous investigations.
111. THE MODEL

Length
Fig. 2. Temperature variation in a constant flux heat exchanger.

A schematic of the modeled heat sink being considered is shown in Fig. 1. It consists of a flat rectangular energy source with a series of channels and fins extending from a base plate. The structure has n channels and ( n- 1) fins. The tips of the fins are insulated with a flat plate used solely for containing the coolant flow. The equations presented in the following show that once certain variables are specified, the number of fins and the fin channel width ratio can be found such that the thermal resistance between the hottest point on the heat source and the coldest point in the coolant is a minimum. The factors to be specified are: 1) the thermal conductivity of the material used for the heat sink; 2) the maximum allowable planar size ( L by W ) of the heat sink; 3) the properties of the coolant used ( p , c p ,v,Pr, Icfluid); 4) the maximum allowable pressure drop and/or pumping work used; 5 ) the desired fraction of infinite fin performance or specified fin length. The first three require no elaboration in that the emphasis here is on the fin and channel geometry. Regarding condition 4 ) , it was shown by [12] that if the design is constrained by the maximum allowable pressure drop through the fins,

then the device with the minimum thermal resistance could require a pumping power which is comparable in magnitude to the amount of heat dissipated in the heat source. This high pumping power requirement is due to the high volumetric flow rates found for a minimum thermal resistance. Since such a solution is undesirable, a upper limit on pumping power is advisable in the optimization scheme. Condition 5) is a convenient option in the design procedure. The heat sink can be analyzed as a two-dimensional flow through narrow rectangular channels with a constant heat flux boundary condition at the base of the fins. Temperature profiles in the L-direction for the heat sink and coolant are shown in Fig. 2, if the heat transfer coefficient is assumed constant. The temperature difference between the solid heat source and the bulk fluid is the same at any plane in the L direction (Fig. 2). Thus the total heat transfer rate equivalent to the electrical circuit power is written as
q = h&(Ts,, - Tf,%)= hAS(Tsp- Tf,o).

(1)

This relation holds true even if the heat transfer coefficient is spatially variable, as long as the average value of h is used in (1).Furthermore, this same amount of power is transferred to the fluid and is expressed by

4 = +&,o

- TfJ.

(2)

These two equations, written in the standard notation of heat exchanger terminology [13], are combined to yield thermal resistance defined as the maximum temperature difference between the source and coolant AT = (Ts,o - Tf,*) divided by the electrical power of the source.

(3)
The two terms on the right side of (3) are referred to as the convective resistance and capacity resistance terms, respectively, in heat exchanger terminology. The latter was designated the caloric resistance by Tuckerman and Pease.

KNIGHT et al.: HEAT SINK OPTIMIZATION

835

IV. DIMENSIONLESS PARAMETERS

As with many important problems, the use of dimensionless groups generalizes the problem and allows for scaling to any physical dimension. In a previous paper [12], each of the appropriate groups for this paper were presented in algebraic detail. Each is briefly reviewed here. Equation (3) with the usual dimensions of thermal resistance (degree per watt) when multiplied by kfiuidW becomes an expression for dimensionless thermal resistance:

friction and heat transfer coefficient are written in terms of two parameters; the hydraulic diameter, Dh, which for one channel is expressed as 2w Dh = n - 1) ( W I D )

+ r(n

and the fluid flow channel aspect ratio

-D_ - n + r ( n - l )
c
WID

(9)

AT e=------

- kfluidw

kfluidw
+-T--.

hAS

mcP

(4)

A dimensionless group, N A ~is , related to the maximum allowable pressure drop experienced over the length of the heat sink. This pressure drop is dictated by the coolant pump or fan capacity and is defined as

The frontal cross-sectional area available for coolant flow in the heat sink A, becomes nWD A ,- n r ( n - 1)

Nap = .

p 1 ~ 3
PV2

while the surface area available for heat transfer,A,, is written as: nWL As = + 2qDL(n - 1). (11) n r ( n - 1)

(5)

For realistic pressure drops through microchips, this number is typically of order lo8 to Another group, Nwork, represents the amount of pumping work used by a specified coolant pump or fan. It is defined as

The first term is the area available for heat transfer at the base and between the channels. The second term is the effective fin area with the fin efficiency included. V. FIN EQUATIONS Heat transfer enhancement due to the presence of fins can be quantified with the assumptions of one-dimensional conduction along the fin length only, constant material properties, constant heat transfer coefficient, no radiation, and uniform fin base temperature. First, a convenient grouping of terms, m,is defined.

(6)
If only Nap is specified for a given design, the pumping work required to attain minimum thermal resistance might exceed the cooling required of the heat source, which is under many circumstances unacceptable. Thus the procedure required that this parameter have an upper bound. By definition, Nap and Nworkare related as follows.

Some other applicable dimensionless groups commonly used in heat transfer andfluid dynamics studies are the: Reynolds number: Nusselt number: Prandtl number: friction factor: ReDh =
NUDh
~

kfluid

UmDh v hDh

The approximate equality results from the assumption that L >> ri. There are at least two methods of quantifying the degree to which fins enhance heat transfer. The first is through the use of fin efficiency, defined as the ratio of actual fin heat transfer to that of a similarly shaped fin with infinite thermal conductivity. For straight rectangular fins of uniform cross sections and insulated tips, the actual fin heat transfer can be found analytically [13] as

Pr=

while the heat transfer of the ideal fin with infinite conductivity is

Additional groups relevant to this paper are the thermal conductivity ratio kfluid/kfin; ratio of fin width to channel width r = Wfin/Wch, the aspect ratio cy = D / l , the heat sink length to width ratio L I W and the heat sink depth to width ratio D I W . As a matter of convenience, other geometrical parameters are expressed through those described previously. For internally confined fluid flow and heat transfer, the expressions for

The ratio of these two expressions forms the fin efficiency.


rl=

tanh (mD)
mD

(15)

A second method follows the concept set forth by Tuckerman and Pease wherein performance was defined by the ratio of actual heat transfer of a fin of finite length compared to that

836

IEEE TRANSACTIONS ON COMPONENTS, HYBRIDS, AND MANUFACTURING TECHNOLOGY, VOL. 15, NO. 5, OCTOBER 1992

of a fin made of the same material and having the same cross section, but of infinite length. The infinitely long fin transfers heat according to the following relation:

A. Laminar Flow

For fully developed laminar flow, f can be expressed analytically as a simple function of Reo,.
f=-. 7 1

Reo,
The ratio of these is
, h ' = tanh(mD).

(17)

The value of 7 1 is determined from the aspect ratio of a given channel. Bejan [14] defined a parameter G as: G = ([lo)' (l/D
1 ++ 1)'

Equation (11) incorporates the symbol q, fin efficiency as defined in (15), as a convenient way of expressing effective fin heat transfer area. Equation (17) is also used in the current optimization scheme. Once a value of 77 or p is specified, m D can be found from (15) or (17). Using geometric relations and the definition of Nusselt number in (12) yields the following relation:
(mD)2 = Nuoh(kfluidlkfin)

(23)

'

A least squares fit of a straight line in G to available values for 71 yields


7 1 = 4.70

+ 19.64G.

(24)

["' ~

I &~[ ~ ' 1
(I8)

which can be rewritten as a quadratic equation in D / W

Equation (24) agrees with exact values [15] within 23%. Algebraic combination of (20)-(22) along with the definitions of NAP and Dh allow for the Reynolds number to be expressed as a function of the NAP and other parameters that are a function of the geometry only:

where C1 is defined as n r(n - 1).Consequently (19) can be readily solved for dimensionless fin length. VI. HEAT TRANSFER The fins shown in Fig. 1 are cooled by forcing a fluid through the channels formed by adjacent fins. Depending on the value of the Reynolds number, this forced convective flow is either laminar or turbulent in nature, or it will be in transition from the former to the latter. The magnitudes of pressure drop and the heat transfer coefficient depend on which of these regimes the flow is in. The Reynolds number is defined as:

The heat transfer coefficient, h, is obtained from the Nusselt number. It is a function only of the channel aspect ratio. The parameter, G, used in computing the friction factor is .used again. As before, a least squares fit to the available exact values gives NuD,, = -1.047

+ 9.326G.

(26)

Equation (26) agrees well with analytical results [15]. Laminar dimensionless capacity and convective thermal resistances are formulated respectively as

where U,,, is the mean velocity, Dh is the hydraulic equivalent diameter of the enclosed conduit, and v is the kinematic viscosity of the coolant. The generally accepted value of the Reynolds number for internal duct flow at which the transition from laminar to turbulent flow begins is Reo, E 2300 [13]. 2 4000. The For the flow to be assured turbulent, Reo, transitional regime exists between these two values. Pressure drop can be obtained through the definition of a dimensionless Fanning friction factor:

and kfluidW-

hA,NuD,(L/W)[C1
+

2C1 (w/D)l[n

2q(D/w)(n - l ) c l ]

(28)

where C1 = n - 1). Dimensionless work, Nwork, is expressed as an explicit function of dimensionless pressure drop, Nap, as follows:

+ r(n

KNIGHT et al.: HEAT SINK OF'TIMIZATION

837

B. Turbulent Flow When flow originates from a reservoir or manifold to

be channelled downstream as is the case here, the velocity profiles will grow to their final steady-state form some distance downstream from the entrance. For this hydrodynamically developing turbulent flow, Phillips [4] recommends a friction factor of the form f = B ReC where

As with velocity profiles, the thermal profiles must also develop from the entrance downstream. For thermally developing Zukauskus [191 Offers the

5 = Nu,

o.48(L/Dh)-0.25

. exp (-0.17LIDh)

3600

d m

(37)

(30)
(31)

B = 0.09290 +
and

1.01612
~

(LIDh) 0.31930
(L/Dh)

C = -0.26800 - .

(32)

which is valid for 4000 5 ReDh 5 5 X lo5, 0.7 5 Pr 5 1.0; and L/Dh 2 0.06. The fully developed Nusselt number is Nu,. For thermally developing turbulent flow using coolants with Prandtl numbers not covered by the Zukauskus correlation, the Gneilinski equation for thermally developing flow is used. Gneilinski [18] offers the correlation

When L/Dh becomes large, (30) yields values similar to those found by other investigators for fully developed turbulent flow [16]. Equation (30) is valid for 2300 5 Re 5 28000 and is used for hydrodynamically developing turbulent flow in this paper. Equation (30) is for round enclosures. Jones [17] recommended that an equivalent diameter (Des) should be used in lieu of the hydraulic diameter (Dh) for rectangular ducts. The expression for their ratio developed by [17] is

which is valid for 2300 5 ReDh 5 lo6; and 0.6 5 Pr 5 lo5. This correlation does not include the effects of ReDh or Pr. For thermally developing turbulent flow (37) is used for gases (low Prandtl Number fluids) and (38) for liquids (high Prandtl Number fluids). The dimensionless capacity and convective thermal resistances for turbulent flow are written, respectively, as

This is valid when the channel aspect ratio, l / D , is less than one. The equivalent diameter concept used in circular duct correlations reduces the scatter of rectangular duct, turbulent flow friction factors from 520% to 25% [16]. Once a friction factor is determined from the pressure drop and geometry, the mean velocity and the Reynolds number can be found. When f is written in the form of (30), U , becomes

and

kfluid~

hA,NuDh(L/W)[Cl

[%]Dr-c)vc 2pB

a+c
'

+ (w/D)][n + 2'dD/W)(n

2C1

- l)C11'

(40)

(34)

As was done for laminar flow, an expression for the relationship between N a p and Nworkfor turbulent flow is derived:
Nwork

From the definition of the Reynolds number, hydraulic diameter, and (34), an expression for the turbulent Reynolds number is written as

= N ~ ~ ( L / W ) ( D / W ) ( ~ / C+ ~ (W/D)]}-1/2. ){~[~

(41) Table I is a recapitulation of the applicable equations for designing a heat sink. For any given n r, mD, and maximum allowable N a p , Nworkand dimensionless fin length ( D I W ) , there are five unknowns to be found: Reynolds number, friction factor, Nusselt number, actual fin length, and actual N a p or Nwork.The unknowns are found from the simultaneous , iterative solution o f the five nonlinear equations for friction factor, Reynolds number, Nusselt number, fin length relation, and pumping work given in the table. The solution method is delineated in the following. The solution procedure for laminar flow for a given n, I?, mD, and maximum allowable N a p , Nwork,and D/W is as follows.

There exists a multitude of heat transfer correlations for turbulent flow in enclosures. The Gneilinski equation for hydrodynamically developed turbulent flow [181 is chosen here due to its excellent corroboration with experimental data [161. Nu,

( f / 2 ) ( R e ~ ,- 1000) Pr
1.0

+1

2 . 7 0 (Pr2l3 - 1)

(36)

838

IEEE TRANSACTIONS ON COMPONENTS, HYBRIDS, AND MANUFACTURING TECHNOLOGY, VOL. 15, NO. 5, OCTOBER 1992

EQUATIONS FOR

THE

TABLE I OF'TIMlZATlON OF HEATSINK DESIGN

Heat Sink Relations Laminar: f Friction Factor

e: =
y1
eDh

4.70

+ 19.646:

6=

(f/D+l)
(LIDh)

Turbulent: f = B Rebh:

B = 0.09290

w:

C = -0.26800 -

Laminar: R ~ D ,= (4/*,1 ).vA,[n Reynolds Number Turbulent: Reo, = { ( 4 / B ) > V ~ , l n

+ r ( n - 1)+ ( u - / D ) ] - ~
1

+ r(n - 1) + (W'/D)]-'}=

Laminar: Kunh = -1.047

+ 9.3266
-inon

(f/z)[.e.,
Turbulent: Xu, Nusselt Number

1 0+12 7 @ [ P r 2 / 3 - 1 ]

~r

Fin Length Relation

Capacity Thermal Resistance

Convective Thermal Resistance

Assume a D / W and N a p . Maximum allowable values are convenient starting points. Find t/Dh from (9). Find G from (23). Use (24) to find 7 1 . Calculate the Reynolds number from (25). The resulting Reynolds number is checked to ensure that it falls in the range of applicability for the laminar flow correlations used here. Use (29) to find N w o r k . If Nwork exceeds the maximum allowable value, solve (29) for the N a p value which results in the maximum allowable Nwork. If N a p is changed here, return to step L5). Calculate the Nusselt number from (26). Solve (19) for D / W . If D / W exceeds the maximum allowable value, use the maximum allowable D / W .

L9) Repeat steps L2)-L8) until D / W and Nap are converged. L10) Calculate convective resistance from (40), capacity resistance from (39), and sum them to obtain total thermal resistance. Steps L1)-LlO) are repeated for a wide range of n and r values, and the geometry resulting in laminar flow which yields the minimum laminar thermal resistance is found. The solution procedure for turbulent flow for a given n, r, m D , and maximum allowable N a p , Nwork, and D / W is as follows. T1) Assume a D/Wand N a p . Maximum allowable values are convenient starting points. T2) Find L/Dh from (8). T3) Find B and C from (31) and (32).

KNIGHT et al.: HEAT SINK OPTIMIZATION

839

Researcher

r
1

kPa (in H20)

til,

Watt

ir,

liter/min

A@

c ~ ~ \ ! ~ ~ f ~ Present ~ Study s same Ap


Present Study same U: Goldberg Present Study same U 12.5

c ~ = l ~ , ~ Present ~ ~ Study s same A p


Goldberg Present Study same Ap c a ~ = l ~ , ~ ~ ~ s Present Study same U.

I I

Goldberg

25

0.435 0.390
1

1.17 (4.68) 1.17 (4.68) 0.73 (2.92) 0.29 (1.17) 0.29 (1.17) 0.26 (1.06) 0.047 (0.19) 0.047 (0.19) 0.075 (0.30)

0.583 1.747 0.583


0.146 0.239 0.146

30 89.9 48.2 30 49.2 32.8


30 22.4 19.1

32.4% 15.4% 18.4% 11.4% 38.6% 46.2%

0.315 0.390
1

0.250 0.320

0.024 0.018 0.024

Calculate the Reynolds number from (35). The resulting Reynolds number is checked to ensure that it falls in the range of applicability for the turbulent flow correlations used here. Find friction factor from (30). Use (41) to wind Nwork. If Nwork exceeds the maximum allowable value, solve (41) for the N a p value If which results in the maximum allowable Nwork. Nap is changed here, return to step T4). Calculate the Nusselt number from (33), (36) and (37) or (38). Solve (19) for D / W . If D/W exceeds the maximum allowable value, use the maximum allowable D / W. Repeat steps T2)-T8) until D / W and N a p are converged. Calculate convective resistance from (40), capacity resistance from (39), and sum them to obtain total thermal resistance. Steps Tl)-T10) are repeated for a wide range of n and r values, and the geometry resulting in turbulent flow which gives the minimum turbulent thermal resistance is found. The geometry yielding the minimum thermal resistance, considering both laminar and turbulent flow, can be determined from a comparison of the laminar and turbulent thermal resistances.

VII. RESULTS
The optimization scheme described previously was applied to three published studies; one by Goldberg [3] where air cooled, copper heat sinks were designed and tested; one by Tuckerman and Pease [ l ] using a laminar flow, water cooled, silicon heat sink; and one by Phillips [4] which was also a water cooled, silicon heat sink designed without the laminar flow restriction. In all three of these cases, the original authors restricted the channel to fin width ratio (r)to unity. Further a (the aspect ratio of the coolant channels) was fixed in both Goldbergs and Phillips studies. The results presented here show notable improvement in the performance of the heat sink by relaxation of the previously imposed restrictions. Goldberg designed, built, and tested three narrow channel forced convection copper heat sinks cooled by room temperature air. The size of the square heat source was 0.635 cm by

0.635 cm (1/4 by 1/4 in) with a fixed fin length of 1.27 cm (1/2 in). The designs by Goldberg mandated a flow rate of 30 L/min, thereby establishing the capacity component of thermal resistance. The heat sink designs were not optimized, but rather the channel and fin thicknesses were systematically varied at values of 5 , 10 and 25 mils (corresponding to 0.127, 0.254, and 0.635 mm and aspect ratios of a = 100, 50, and 20). The most severe restrictions were mandatory laminar flow and r = 1. In Goldbergs design procedure the Nusselt number was fixed at a value of eight. When the restrictions of a volumetric flow rate of 30 L/min, Nusselt number of eight, and the geometries specified by Goldberg are examined in (3) or (4), the obtained predicted thermal resistances agree with those found by Goldberg. In the present analysis, the overall heat sink dimensions of Goldbergs design were kept, but those regarding fixed r, a, N u D h , flow regime and volumetric flow rate were relaxed. The results for each case are presented in Table 11. For each of the three cases compared, the best design was determined either at the same pressure drop through the device or the same power consumed by the fan as those set or measured by Goldberg. In each case, the optimal solution occurred in the laminar region due to the allowable pressure drops. Likewise, the choice of value was significantly less than unity. The last column in the table indicates that significant improvement in the optimal design (11-39% in thermal resistance) is achieved when the r = 1 restriction is dropped. These results show that this condition is neither necessary nor desirable. Tuckerman and Pease designed, built, and tested a 1 cm by 1 cm water cooled silicon, microchannel heat sink. Their optimization procedure included assumptions of laminar flow, fixed pressure drop through the heat sink, r = 1, a fully ) 6, p = 0.76, developed and fixed Nusselt number ( N u D ~ = and a friction factor for infinite parallel plates (71 = 96). Their study indicated that the best design occurred with channel widths of 57 pm and channel depths of 365 pm for the pressure drop considered. At a power level of 790 W/cm2 through the heat spreader, a resulting AT of 71OC was measured. Pumping power was 0.3% of dissipation, or 2.27 W (corresponding to 11 cm3/s flow rate at 206.8 kPa, or 30 lbf/in2 pressure drop). In a previous study by Knight et al. [12], the optimal laminar flow design under the above

840

IEEE TRANSACTIONS ON COMPONENTS, HYBRIDS, AND MANUFACTURING TECHNOLOGY, VOL. 15, NO. 5, OCTOBER 1992

TABLE 111 COMPARISON OF PRESENT RESULTS TO THOSE OF TUCKERMAN AND PEASE [l] Tuckerman and Pease
~~ ~

Present Study

CONSTRAINTS Size, Length (L) by Width (W) pressure drop, Ap % infinite fin performance, 3 Fin length, pm Coolant Prandtl number, Pr Fin Material
kfluldlkfin

Fin to channel thickness ratio, r' Nusselt Number Type of flow -fl, laminar friction factor DIMENSIONLESS GROUPS

1 cm x 1 cm 206.8 kPa 76% unrestricted water 3.71 Silicon 0.00464 1 6 laminar 96

same same unrestricted 365 same same same same unrestricted unrestricted laminar or turbulent not applicable

1 2.82 x 10" 3.62 x 1013 CALCULATED RESULTS


n , number of channels Depth, D , pm Fin to channel width ratio, r' Fin thickness, pm Channel thickness, pm Reynolds Number Volumetric Flow Rate, cm3/s Aspect Ratio, cy Nusselt Number Laminar friction factor - r i
1 Y ' Ap
AVwork

same same unrestricted Turbulent 22 365 0.215 81 317 8459 59.2 0.97 85.6 not applicable 2.82 x 10" 1.95 x 1014 0.006 0.050 0.056 35%

Laminar 88 365 1 57 51 730 11 6.4 6 96 2.82 x 10'' 3.62 x 1013 0.022 0.064 0.086

Capacity Thermal Res, -C/W Convective Thermal Res, OC/W Total Thermal Resistance, "C/W Change in Thermal Resistance, A@

restrictions was obtained using the design method reported in this paper. The obtained dimensions and results were all within 5% of those obtained by Tuckerman and Pease. In the present paper, the pressure drop and overall package dimensions (planar dimensions of heater and fin length) were maintained the same while the restrictions on the state of flow development (fully) and the values of I?, yl, and NuDh were relaxed in order to find optimal channel and fin dimensions. An optimal solution for turbulent flow was sought. Table 111 is a presentation of results. When turbulent flow is allowed, the thermal resistance is reduced by 35% from that of Tuckerman and Pease. The relatively wide channels found for the best turbulent solution allow, for fixed pressure drop and same fin height, a greatly increased mass flow, thereby significantly reducing the capacity resistance term. Likewise, the heat transfer coefficient increases due to the presence of turbulence, reducing the convective term by 22%. A maximum pressure drop of 206.8 kPa (30/bf/in2) is common to both cases. Pumping power for the turbulent case is increased to 12.2 W or 1.6% of heater power, still a negligible amount.

The results here differ from those presented previously [12]; not in the magnitude of improvement but rather in the heat sink design configuration. This is due primarily to the inclusion of developing flow for both the fluid flow and heat transfer equations. The current modeling program now includes developing turbulent flow, a laminar equivalent hydraulic diameter, and a better correlation for turbulent heat transfer in rectangular channels. Fig. 3 graphically quantifies the influence these limitations have on the optimal thermal design of heat sinks. Three sets of dimensionless thermal resistance values are plotted as a function of number of channels, n. Common to all are the materials, fluids, pressure drop, and overall dimensions of the heat sink (1 cm by 1 cm by 365 pm fin length) used by Tuckerman and Pease [l].The equations employed for analyses are those presented in Table I. A dashed line is drawn for flow with r = 1. The optimum in the laminar regime occurs at n = 93, a value close to that found by [l],the difference being due to the laminar Nusselt number in this paper being variable with aspect ratio. The solid line in the laminar regime is drawn with the J? restraint lifted and results in a 17% improvement in thermal performance

KNIGHT et al.: HEAT SINK OPTIMIZATION

841

TABLE I V COMPARISON OF PRESENTRESULTSTO THE WORK OF PHILLIPS [4]

Phillips
CONSTRAINTS Size, Length ( L ) by Width ( W ) pressure drop, Ap Aspect Ratio Fin Length, microns Coolant Fin Material Fin to channel thickness ratio, r Nusselt Number Correlation Type of flow DIMENSIONLESS GROUPS 1 cm x 1 cm 68.9 kPa 4.0 unspecified water Silicon 1 (37). laminar or turbulent

Present Study

same same unspecified 1889 same same unrestricted same same

L/W
Maximum Nhp Maximum Nwork CALCULATED RESULTS
n, number of channels Depth, D , pm Fin to channel width ratio, r Fin thickness, pm Channel thickness, pm Volumetric Flow Rate, cm3/s
NAP
Nwork

1 8.86 x lo9 1.45 x 1014

same same same

11 1889
1

472.1 472.1 145 8.86 x 1.45 x 0.0016 0.064 0.066

lo9
1014

19 1889 0.66 215 323 145 8.86 x IO9 1.45 x 1014 0.0016 0.049 0.052 21%

Capacity Thermal Res, "C/W Convective Thermal Res, "C/W Total Thermal Resistance, "C/W Change in Thermal Resistance, A 0

0.0010

0.0009
0.0008 0.0007

0. r=i 0 , r=Optimum Optimum


/
\

11.2

- 1.0 - 0.8 - 0.6 - 0.4 - 0.2


0

0 0.0006
0.0005

0.0004 0.0003

0.0002

'

25

50

75

100

125

150

0.0 175

'

Number of Channels
Fig, 3. Thernial resistance as a function of the number of channels for the heat sink described in Table 111.

with = 0.32 being the best value. The solid line in the turbulent regime depicts thermal resistance values for turbulent flow, again with no constraint on r. Here the performance is improved by 32% over that for laminar flow and fixed r, with the optimal value of r= 0.215.

The dotted lines in Fig. 3 shows the value of J? which yielded the lowest thermal resistance for a given n. The slope of these lines is seen to change sharply near an n value of 45 and again near an n value of 70. Between these two n values, the r value which minimizes the equation for thermal resistance yields a Reynolds number between 2300 and 4000. Since the correlations used are not valid in this regime, such results are viewed as unacceptable. So, between n of about 45 and the end of the laminar flow region, the reported gives the highest allowable Reynolds number for laminar flow, 2300. This yields the lowest thermal resistance with the Reynolds number in the range where the correlations used are accurate. In the turbulent flow regime, the given I? yields the lowest acceptable Reynolds number for turbulent flow when n is less than 70 for the same reason. Since these breaks in slope are not near the laminar or turbulent minima, there is no effect on the resulting predicted overall optimal design. Phillips devised a scheme for microchannel heat sink design to include the possibility of turbulent flow and accounted for hydrodynamic flow development in both regimes, but = 1 and a specified aspect maintained the restrictions of ratio. Phillips used water as the coolant and silicon as the heat sink material with all properties evaluated at 27C. When the optimal geometry determined by Phillips is examined using the equations presented in this paper, excluding the effect of thermally developing flow, a thermal resistance is predicted

whose value is within 1% of that given by Phillips. When the geometry recommended by Phillips is analyzed, accounting for developing flow, the predicted thermal resistance is lowered by 18% from the thermally fully developed case. Only a comparison with his turbulent optimal solutions is made here to show the effects of mandating r and a. For our case, the fin length is identical to that of Phillips, but the fin thickness to channel width ratio (r), and therefore, the aspect ratio is determined for the best turbulent case. The pressure drop, volumetric flow rate, and pumping work are identical to those used by Phillips. The present design has the same overall exterior dimensions as that of Phillips but as Table IV shows, improved performance by 21%. The main reason for this improvement comes from the inclusion of the effect of thermally developing flow. Additional improvement comes from the relaxation of the r constraint. The current design has a much smaller I (0.66 compared to 1 for Phillips), more channels (19 versus 1I), and consequently substantially more area available for heat transfer for the same fin length. It is recognized that the optimal design scheme described here could lead to a design which could be impractical due to the channels or fins being too thin to manufacture. This optimization scheme necessarily incorporates a constraint on minimum thickness of either fin or channel.

VIII. CONCLUSIONS
The governing equations for fluid dynamics and heat transfer have been presented in a generalized, dimensionless form along with applicable geometrical relationships. These can be used to determine the dimensions of any microchanneled heat sink of rectangular coordinates such that the resulting thermal resistance is minimized. Comparisons of present results with those obtained by previous investigations show that unnecessary and undesirable restraints were imposed on their design procedures (laminar flow, fixed fin thickness to channel width ratio and/or fixed aspect ratio) and that relaxation of these restraints leads to significant improvements in designed thermal resistances. Depending on whether the pressure drop or pumping power was maintained the same, the Goldberg heat sink design improvement ranged from 11.4 to 46.2% in thermal resistance. The most notable changes occurred in the designed values of I?(- 0.3) and the number of fins. The redesign of the Tuckerman and Pease laminar heat sink rendered a much smaller value for r (0.215 as compared to I), the number of fins (reduced from 88 to 22), and the nature of the flow (turbulent rather than laminar) with a resulting decrease in thermal resistance of 35%. Refinement in Phillips design effected a decrease in the r value from unity to 0.66 and more than a 50% increase in the number of fins resulting in a predicted decrease of 21% in thermal resistance.

[2] D. B. Tuckerman, Heat transfer micro-structures for integrated circuits, SRC Tech. Rep. No. 032, SRC Cooperative Research, Research Triangle Park, NC, 1984. [3] N. Goldberg, Narrow channel forced air heat sink IEEE Trans. Comp. Hybrids Manuf. Technol., vol. CHMT-7, pp. 154-159, Mar. 1984. (41 R. J. Phillips, Micro-channel heat sinks, Advances in Thermal Modeling of Electronic Components and Systems, Volume 2. A. Bar-Cohen and A. D. Kraus, eds. New York: ASME, Ch. 3, 1990. [5] A. Bar-Cohen and A. D. Kraus, Advances in Thermal Modeling of Electronic Components and Systems, Volume 2, New York: ASME, 1990. [6] M. Mahalingam, Thermal management in semiconductor device packaging, Proc. IEEE, vol. 73, pp. 1396-1404, Sept. 1985. [7] S. Sasaki and T. Kishimoto, Optimal structure for micro-grooved cooling fin for high-power LSI devices, Electron. Lett., vol. 22, no. 25, pp. 1332-1334, 1986. [8] T. Kishimoto and T. Ohsaki, VLSI packaging technique using liquidcooled channels, 36th Electronics Components Conf. P roc., May 1986, pp. 595-601. [9] L. T. Hwang, I. Turlik, and A. Reisman, A thermal module design for advanced packaging, J . Electron. Mat., vol. 16, no. 5, pp. 347-355, May 1987. [lo] A. Bar-Cohen and M. Jelinek, Optimum arrays of longitudinal, rectangular fins in convective heat transfer, Heat Transfer Eng., vol. 6, no. 3, pp. 68-78, 1986. [ l l ] C. S. Landram, Computational model for optimizing longitudinal fin heat transfer in laminar internal flows, Heat Transfer in Electron. Equipment, vol. 171, pp. 127-134, 1991. [I21 R. W. Knight, J. S. Goodling, and D. J. Hall, Optimal thermal design of forced convection heat sinks-Analytical, J . Electron. Pack., vol. 113, no. 3, pp. 313-321, Sept. 1991. [13] F. P. Incropera and D. P. DeWitt, Fundamentals of Heat and Mass Transfer. New York: Wiley, 1990. [14] A. Bejan, Convection Heat Transfer. New York: Wiley, pp. 75-82. 1151 W. M . Kays, and M. E. Crawford, Convective Heat and Mass Transfer, New York McGraw-Hill, 1990. 116) S. Kakac, R. K. Shah, and W. Aung, Handbook of Single-phase Convective Heat Transfer, New York: Wiley, 1987. [17] 0. C. Jones, Jr., A n improvement in the calculation of turbulent friction in rectangular ducts,J. Ffuids Eng., vol. 98, pp. 173-81, June 1976. [18] V. Gneilinski, New equations for heat and mass transfer in turbulent pipe and channel flow, Int. Chem. Eng., vol. 16, pp. 359-368, 1976. [ 191 A. Zukauskas, High-Performance Single-phase Heat Exchangers, J. Karni, ed. New York: Hemisphere, 1989

Donald J. Hall was born in Columbus, GA, in 1965. He received the B.S. degree in applied mathematics and the M.S. degree in mechanical engineering from Auburn University in 1987 and 1991, respectively. Since 1991, he has been employed by Compaq Computer Corporation of Houston, TX, as a mechanical engineer. Mr. Hall is a member of the American Society of Mechanical Engineers.

Roy W. Knight, for a photograph and biography, please see page 760 of
this issue.

John S. Goodling, for a photograph and biography, please see page 760 of this issue.

REFERENCES
[ I ] D. B. Tuckerman and R. F. W. Pease, High-performance heat sinking for VLSI, IEEE Electronic Device Lett.. vol. EDL-2. DD. 126-129. May 1981.
,
I I

Richard C. Jaeger, (S68-M169-SM78-F86), raphy, please see page 821 of this issue.

for a photograph and biog-

Das könnte Ihnen auch gefallen