Abstract
The use of micro-channels in order to cool modern high speed electronic circuits is one of the techniques
frequently adopted in current practice. A burst of publications in this area has been observed in the last decade.
However, modeling of fluid flow and heat transfer in micro-channels is still an open problem. In fact, many
deviations have been experimentally observed between well established correlations used for conventional
normally sized channels and the behavior of microchannels. These deviations increase as the channel size
decreases. In this work, observed experimental deviations are first listed, followed by a critical review of different
hypotheses advanced in the literature to explain them. One of these hypotheses is thoroughly developed in order
to build a model that explains both the orders of magnitudes and the trends of observed phenomena.
Index Terms
Inspec
Controlled Indexing
channel flow circuit simulation cooling flow simulation high-speed integrated circuits
integrated circuit measurement integrated circuit modelling integrated circuit packaging
thermal analysis thermal management (packaging)
Non-controlled Indexing
channel size fluid flow heat transfer modeling high speed electronic circuits cooling liquid
flow modeling micro-channels
Author Keywords
Not Available
PACS Codes
Not Available
References
Citing Documents
-512-
5th. Int. Conf. on Thermal and Mechanical Simulation and Experiments in Micro-electronics and Micro-Systems, EuroSimE2004
i- PIP inlet __ TaulTau inlet --- -QN/& inlet --- I
6
4
2
0
-2 !
I
L W
Figure 1 - Different roughness shapes It will also be proved in the next section that this
scenario may successfully model all other deviations
enumerated above. In fact, micro-eddies that will be
I-- PlPinlet ~ TadTau inlet ....-..-&/& inlet I provoked by roughness tips will cause a Re number
dependence of both Po and Nu, which explains another
4
observed deviation. Finally, it is evident that micro-eddies
3 are responsible for the observed early transition to
turbulence in micro-channels. Moreover, the order of
2 magnitude analysis will also show that these effects
cannot be detected for conventional channels.
1
5. Modeling friction
0 Fully developed flow will be assumed since L/9, (L
being the channel length in flow direction) is usually very
-1 high. Modeling will be done in two steps. First, very low
Re will be assumed to obtain a simple model without
-2 micro-eddies. Second, the effect of micro-eddies will be
added. Subscript cl will be used for classical results, i.e.
-3 conventional channels, lo for low Re results and mc for
the final model of micro-channels. As an idealization of
Figure 2 - Flow over circular rough elements the peaks and valleys caused by wall roughness, it will be
assumed that wall will be separated into two distinct
I-P/P inlet __ Tau/Tau inlet--- -.-.---..Qw/Qwinlet---I portions (Fig. 5). The first portion is shielded by the gas
blanket and occupies a portion 5 of the wall area, where 5
I n I I is the shielding coefficient that should typically be around
0.5. The unshielded portion occupies 1-5. The gas
blanket is of thickness 6, which is equal to the surface
roughness. This scheme is applicable as long as Re is
small enough to prevent (or at least reduce) the effect of
micro-eddies on the mean flow. Hence we need first a
simple friction model in the shielded portion, then
combine it with friction in the unshielded portion and
finally add effects of micro-eddies.
5.a Friction in the shielded portion
I The cross section constitutes a 2D domain that will be
-3
subdivided into 2 sub-domains, one occupied by the liquid
-4 0 1 , which is surrounded by the domain Ogoccupied by the
gas (Fig. 5). Fully developed flow implies that pressure is
Figure 3 - Flow over rectangular rough elements constant over both RI and Og and has a constant axial
-513-
5th. Int. Con5 on Thermal and Mechanical Simulation and Experiments in Micro-electronics and Micro-Systems, EuroSimEZOO4
gradient -B. Hence, Navier-Stokes equation and boundary VI io = VI Ci+ AV (5.9)
conditions in both domains are: But, from the definition of Po, and Eq. (5.3):
Po = -8 T,., Re/@ vavg2)= 2BD?/(pl vlavg) (5.10)
where avg denotes the average lover the cross section.
Hence, by taking the average of (5 9) and using (5.10):
PO&= PoCl/ (l+AV/v,i) = PO,! / ( 1+p) (5.1 1)
p = PoClE (p1/pg-l)/8 (5.12)
where Po5 denotes Po for the shielded portion.
n 5.b- Friction for low Re
an Combining friction on shielded and unshielded
portions gives:
POI,=P O ~(1-6) I + 6 PO< =PoCl(l-\P/(p+l)) (5.13)
5 .c-Effect of micro-eddies
According to the mixing length theory [39], ‘turbulent
.)x like’ eddies can produce an additional shear Ted:
Ted = -p h2 Idv/dyl dv/dy (5.14)
where h is the mixing length. It can be expressed as:
h=c6 (5.15)
Figure 5 - Simplified problem geometry
where c is the mixing coefficienl that is expected to be
around 0.5 as seen from Fig. 5. The average dv/dy at the
V2v,= -B/p, in Ri (i=g or 1) wall could be obtained from the viscous wall shear stress
v, = 0 on dR ,vi = v, on do, (5.1) corresponding to the low Re case obtained above (5.10):
where vi is the axial velocity of gas (i=g) or liquid (i=l), p pldv/dylw= -w,= PolopvavpZ/(8Re)
is the dynamic viscosity and dCli is the liquidgas Hence:
interface. Since the gas layer has a small thickness Ted / Twi0 = Polo (c&)*Re/8 (5.16)
compared to section dimensions, it may be approximated
The total wall shear stress will thus be assumed as the
as a 2D channel of infinite width, having a solid boundary
sum of viscous and eddy stresses:
at its bottom, and a liquid boundary at its top. Across Q,
5mc = ‘cwlo + Ted = Twlo (1+Ted/Kwlo)
the shear stress should vary linearly with y the coordinate
normal to the channel wall, due to the fully developed i.e. Po, = Po,,( 1+Pol,(cs)’Rel’8) (5.17)
assumption. Momentum balance over any layer extending This equation (where POI,has, been defined by 5.13)
from 0 to y gives: constitutes the full friction model of micro-channels. It
(L - ~ ( y ) dx ) dP y explains why friction has been found to be less than
1 (5.2)
where x is the coordinate along the flow. Momentum expected in investigations performed at low Re, and why
it was higher at high Re. In the next paragraph it will be
balance over the whole cross section R yields:
quantitatively compared with published results.
~~p dx = dP A 5.d- Experimental verification
i.e. T~ =dP/dx Alp = -B Dh/4 (5.3) Test results will be confronted with the postulated
where p is the section perimeter and A its cross-sectional model given by Eq. (5.17). It contains 2 parameters that
area. From Eqs (5.2,3) we get an expression for z(y): depend on the nature or shape of roughness elements,
~ ( y=) B (y-Dd4) (5.4) namely 6 and c. Note that these are NOT adjustable
Integrating Eq. (5.4), and noting that: coefficients that may take any value to fit test results.
~ ( y=) - p g dv,/dy; ~ ~ (=00) (5.5) They do have a physical meaning that implies a narrow
yields: range of expected values around 0.5 for each of them. The
first comparison will be made with the results of Pfahler et
VdY) = B(YI)IJ4-Y2/2)/P, (5.6) al. [ 121 that were obtained at very low Re.
Hence, the velocity at the interface is constant and
Figure 6 shows the effect of size on friction. Typical
(assuming a small ratio &=6/Dh)is equal to:
values of G=c=0.5 were taken. It is clear that for
v = E B D2/4 pg (5.7) Isopropanol there is a perfect match. But for Silicon oil,
If the domain figwas filled with liquid, the velocity at the trend was correct with a small shift. This means that a
y=F would have been the same as that given by Eq. (5.7), small modification of 6 may be needed for this particular
but with pl instead of CL,. Hence, the presence of a gas type of fluid/wall combination. Effect of Re on friction is
layer increases the liquid velocity at the interface, for the compared in Fig. 7 with experimental results [14,15,20]. It
same wall shear stress, by the constant amount: is astonishing to notice that ai 1 available test results
AV = E B D2/4 (l/p,-l/pl) (5.8) compared well with (5.17) if we take k=c=0.5, except one
The variable liquid velocity in the whole domain RIat case (Harms et a1 [20]) with 5=0.4 and ~ 0 . 5 .This
low Re, V Ilo can thus be approximated by: constitutes a strong proof of the validity of the proposed
-514-
5th. Int. Con$ on Thermal and Mechanical Simulation and Experiments in Micro-electronics and Micro-Systems, EuroSimE2004
scenario, as well as the simple model issued from it. to 1. The coefficient was correlated in terms of test
Intensive test series may give a more precise estimate of 6 section geometry and Re against their experiments using 4
and c depending on wall andor liquid type as well as adjustable coefficients and an expression that was not
machining procedure used. But, in all cases, Eq. (5.17) based on a physical model or scenario. The advantage of
seems a suitable form for a practical precise correlation, such expressions is that they can easily fit a given set of
as well as a rough estimate for designers with the typical test conditions, but the disadvantage is that one cannot
values of t=c=0.5. guarantee the satisfaction of other sets. In particular, this
correlation cannot predict cases for which POP% < 1.
The model presented here was derived from problem
physics according to a given scenario. Hence it contained
-
-0- Pfahler et al(l991) Isopropanol only 2 parameters having a physical meaning as well as a
Model Isopropanol very narrow range of variation.
Pfahler et al (1991) Silicon oil 6. Modeling transition to turbulence
---- Model Silicon oil At a certain intensity level, micro-eddies may be
strong enough to induce and sustain turbulence in the
1.1 I I I 1
mean flow. The corresponding model could be inspired
from experimental results of turbulence initiation due to a
cylindrical obstacle (a wire) having an axis normal to flow
direction and attached to a wall [39]. Turbulence is
initiated when the cylinder diameter (seen here as the
0.6 roughness height) reaches a value that was experimentally
0.1 1 10 100 determined:
Dh bm)
6 = K p /(pu*) (6.1)
Figure 6 - Effect of channel size on friction for low Re where K is an empirical constant that equals 7 for the case
-
from which the criterion for transition to turbulence could
-Mala & Li 50micr. Model
.-.*-e-
be derived in terms of the critical Re:
Mala & Li 80micr. .- .- k . . - Model Recr= 8 (Us)’ POI, (6.3)
i-- Mala & Li 100micr . - - + - - Model
- The value of K for the case of distributed roughness
still has to be determined experimentally. It is evident that
it should be much smaller than the case of a single wire
(i.e. 7) but still probably of the order of 1. Unlike the case
2.0 ... A
of friction, literature is both unanimous about the early
transition and lacking data on roughness values. To the
authors’ knowledge, only one article [20] has provided
both the value of roughness (~=2%)and the transition Re
(Re,,=1510). Substituting in (6.3) gives K=1.92. It is
certainly not possible to recommend a value of K based
on only one experiment, especially for the transition to
turbulence. But the reasonable value of K pledges in favor
0.0 ~ I I I
of the proposed scenario. It is hoped that once the form of
0 500 1000 1500 2000 a suitable criterion has been derived (Eq. 6.3), new
Re
experiments could use it to obtain reliable correlations.
Figure 7 - Dependence of friction on Re It is interesting to note that an equivalent criterion
could be obtained using another approach. In fact, if we
Surface roughness was not specified by Papautsky et define transition to turbulence as the point at which eddy
al. [I41 hence, a value of 1% was assumed which is and viscous shear stresses are of equal importance, then
typical for this fabrication technology. Surface roughness we get from (5.16):
reported by Mala & Li [ 151 was not measured, but based Recr= 8 / ((cE)~ POI,) (6.4)
on manufacturer data. Hence, the smallest channel (50 The amazing fact is that if we apply it with the
pm) had a greater error. They have stated that Po may recommended value of c=0.5, that has been validated
decrease with Re in a certain range, which is not normal. from friction data, we get ReCr=1630,which is very close
The decrease lies however within error bars. They have to the observed value. The value of c=OS in fact
also presented a model based on surface roughness. It corresponds to the value of K=2 in equation (6.3).
used an added viscosity, called ‘roughness viscosity’. It Although this result was obtained for the only quantified
contained an adjustable coefficient that varied from 0.0 1 test case available, which does not constitute a formal
-515-
Sth. Int. ConJ on Thermal and Mechanical Simulation and Experiments in Micro-electronics and Micro-Systems, EuroSimE2004
proof, it is still a good indication of the validity of the
scenario proposed.
7. Modeling heat transfer
1 - Harms et al -- Model I
I I
Available heat transfer data are much less than those
of friction, and in addition some of them are unusable.
This is due to different definitions of effective
temperature differences and areas, as well as omitting
roughness data. Hence, the heat transfer model that will be
developed here is not intended to validate the proposed
scenario. It is believed that above arguments were
sufficient. But it is rather intended to understand the 0 1000 2000 3000 4000
implication of the proposed scenario on thermal Re
resistance. At very low Re, the gas blanket acts as an Figure 8 - Comparison of model with Harms et al.
insulating layer. Shielded and unshielded areas can be
considered as two parallel thermal resistances. The
unshielded portion has a Nu number that is the same as + Wangsec.3 ~ Nbdel 0.5mm
that of the classical problem Nbl. The resistance of the --twangsec5 ~ Pdbdel 0.3 mm
shielded portion can be estimated as 2 series resistances.
The thin gas blanket could be considered as a layer 10, k
conveying heat mainly by conduction. The liquid core has
a Nu number that is slightly greater than N u . But the
difference is not big and may be neglected, especially
when added in series with the rather insulating gas
blanket. Adding these three resistances yields:
Nulo= N&i( 1-ty/( 1+y)) (7.1) " 0 -
-516-
5th. Int. Con$ on Thermal and Mechanical Simulation and Experiments in Micro-electronics and Micro-Systems, EuroSimE2004
transfer effects. Another test series is due to Wang et a1 [4] Ueno, K., Kim, H-B. & Kitamura, N., “Channel shape
[17]. Unfortunately they did not report roughness values. effects on the solution-flow characteristics and the
A value of 2% was assumed here. They have tested 6 liquidliquid extraction efficiency in polymer
different test sections. Sections 1 and 2 were disregarded microchannel chips,” Analytical Sciences, 19, (2003),
since the flow was not fully developed. pp. 391-394.
The temperature difference used to calculate h was [5] Keblinski, P. Phillpot, S.R., Choi,S.U.S. & Eastman,
unusual: outlet wall temperature-inlet fluid temperature. J.A., “Mechanism of heat flow in suspensions of nano-
Fortunately, enough data were given to enable us to re- sized particles (nanofluids),” Int. J. Heat Mass
estimate it by the standard way for constant heat flux case: Transfer, 45, (2002), pp. 855-863.
wall-fluid temperature, both taken at outlet. Comparison [6] Souders, D., Khan, I., Yao, G.F., Incognito, A. &
results are shown in Fig. 9 for test sections 3 and 5 . Corrado, M. “A numerical model for simulation of
Taking into consideration the large scatter in experimental combined electroosmotic and pressure driven flow in
data, the model was at least able to predict the correct microdevices”. 7Ih Int Symp on Fluid Control,
order of magnitude as well as the correct trend. Sections 4 Measurement and Visualization, Flucom’03, Perronte.
and 6, were identical to 3 and 5 respectively, but with [7] Pamula, V.K. & Chakrabarty, K. “Cooling of
thinner walls. Differences were significant, thus proving integrated circuits using droplet-based microfluidics,”
that conjugate heat transfer effect is important, as GLSVLSI’O3,April 2003, Washington, DC, USA.
indicated elsewhere [311. [8] Wexler, E., Lout@, R., Ortega, A., Wallinger, D. &
8. Conclusions Prindiville, T. “Novel Boiling-Enhanced Multi-
Liquid flow and heat transfer in micro-channels were Channel Heat Sinks,” The 6Ih ASME-JSME Thermal
studied. Deviations of flow and heat transfer from those of Engineering Joint Conference, March 2003
normal size channels were reviewed. A critical analysis [9] Papautsky, I, Ameel, T & Frazier, A.B. “A review of
was made of previously advanced hypotheses to explain laminar single-phase flow in microchannels”, ASME
deviations of liquid flow. A hypothesis relying on surface international Mechanical Engineering Congress and
roughness was made, based on which simple models were Exposition, November 200 1
derived to predict most of the phenomena encountered in [ 101 Sabry, M.N., “Dynamic compact thermal models
micro-channels. In particular, the reduction of both used for electronic design: A review of recent
friction and heat transfer compared to predictions of progress,” INTERPACK2003-35 185, July 2003,
classical theory for low Re, as well as the increase of both Hawaii
Po and Nu numbers with Re in the laminar range were all [ l l ] Pfahler, J., Harley, J., Bau, H.H. & Zemel, J.
explained. At relatively high Re, both Po and Nu may be “Liquid and gas transport in small channels,” ASME-
higher than the classical theory as has also been observed. DSC, vol. 19, ed. CHO et al., (1990) pp. 149-157.
A criterion was also derived for the transition to [ 121 Pfahler, J., Harley, J., Bau, H.H. & Zemel, J., “Gas
turbulence that predicted the observed early transition. and liquid flow in small channels,” ASME-DSC, vol.
The model was successfully confronted with available test 32, ed. CHO et al., (1991) pp. 49-60.
results from different authors giving both the order of [13] Jiang, X.N., Zhou, Z.Y., Huang, X.Y. & Liu, C.Y.,
magnitudes and the trends. It is important to note that the “Laminar flow through microchannels used for
model contained no adjustable coefficients. It contained microscale cooling,” Proc. 1st Electronic Packaging
though two parameters having a physical meaning and Technology Conference, 1997, pp. 119-122
thus an expected limited range of values. Very good [14] Papautsky, I., Brazzle, J., Ameel, T.A. & Frazier,
agreement between model and experiments was obtained A.B., “Microchannel fluid behavior using micropolar
for parameter values lying in the middle of this range. fluid theory,” Proceedings of the Eleventh Annual
Future work in this area may give a more precise value of International Workshop on Micro Electro Mechanical
these parameters depending on the nature of walls, liquid Systems, 1998, pp. 544 -549.
and machining procedure used. [ 151 Mala, G.M. & Li, D., “Flow characteristics of water
References in microtubes,” Int. J Heat and Fluid Flow, Vol. 20,
[l] Tuckerman, D.B. & Pease, R.F.W., “High (1999) pp 142-148.
performance heat sinking for VLSI,” IEEE Electron [16] Choi, S.B., Barron, R.F. & R.O. Warrington, “Fluid
Device Letters, Vol. EDL-2, No 5 (1981) pp, 126-129. flow and heat transfer in microtubes,” ASME DSC-vol.
[2] Kandlikar, S.G. & Grande, W.J., “Evolution of 32, ed. Cho et al., (1991) pp. 123-134.
microchannel flow passages thermohydraulic [17] Wang, B.X. & Peng, X.F. “Experimental
performance and fabrication technology”, ASME Int. Investigation on Liquid Forced Convection Heat
Mechanical Engineering Congress & Exposition, Transfer through Micro-channels,” Int. J. Heat Mass
November 2002, New Orleans, Louisiana Transfer, Vol37, Suppl, 1, (1994) pp 73-82.
[3] Erickson, D. & Li, D., “Numerical simulations of a [18] Peng X.F. & Peterson, G.P. “Convective Heat
low power microchannel thermal cycling reactor”, Transfer and Flow Friction for Water Flow in
International Journal of Heat and Mass Transfer, 45 Microchannel Structures,” Int. J. Heat Mass Transfer,
(2002) pp. 3759-3770 V0139, NO. 12, (1996) pp 2599-2608.
-51 7-
5th. Int. Con$ on Thermal and Mechanical Simulation and Experiments in Micro-electronics and Micro-Systems, EuroSimE2004
[19] Zeighami, R., Laser, D., Zhou, P., Asheghi, M., 331 Ng, E.Y.K., Poh, S.T. and Chai, J.C.K., “Modelling
Devasenathipathy, S., Kenny, T., Santiago, J. & of pressure-driven liquid flows in conjugate
Goodson, K., “Experimental investigation of flow rectangular microchannel with electric double layer
transition in microchannels using micron-resolution effects,” Proceedings of Yd Electronics Packaging
particle image velocimetry,” The 7‘h Intersociety Technology Conference, 2000, pp. 419 4 2 4
Conference on Thermal and Thermomechanical 341 Tso, C.P. & Mahulikar, S.P., “The role of the
Phenomena in Electronic Systems, ITHERM 2000, Brinkman number in analysing flow transitions in
vol. 2, pp. 148 -153. microchannels,” Int. J. Heat A4ass Transfer, Vol. 42,
[20] Harms, T.M., Kazmierczak, M.J. & Gemer, F.M. (1999) pp. 1813-1833.
“Developing Convective Heat Transfer in Deep [35] SABRY, M.N., “Scale effects on fluid flow and heat
Rectangular Microhannels,” Int. J. Heat Fluid Flow, transfer in microchannels,” Components and
V0120, (1999) pp 149-157. Packaging Technologies, IEEE Transactions on
[21] Bower, C., Ortega, A., Skandakumaran, P., Packaging and Manufacturin,g Technology, Part A,
Vaidyanathan, R., Green, C. & Phillips, T. “Heat vol. 23, no. 3, (2000) pp. 562 --567.
Transfer In Water-cooled Silicon Carbide Milk [36] Lee, W. Y., Wong, M. & Zohar, Y., “Flow separation
Channel Heat Sinks For High Power Electronic in constriction microchanne:ls,” The 14‘h IEEE
Applications”, ASME International Mechanical International Conference on Micro Electro
Engineering Congress & Exposition Washington, Mechanical Systems, 2001, pp. 495 4 9 8 .
D.C., November 2003 [37] Hwang, D.P. & Biesiadny, T.J., “Experimental
[22] Eringen, A., “Simple microfluids,” International evaluation of the penalty associated with micro-
Journal of Engineering Science, vol. 2, (1964) pp. blowing for reducing skin friction,” NASA Technical
205-2 17. Memorendum, 1 13174, (1997) AIAA-97-0677.
[23] Arkilic, E.B., Schmidt, M.A. & Breuer, K.S., [38] Kandlikar, S.G., Joshi, S. & Tian, S., “Effect of
“Gaseous slip flow in long microchannels,”Journal of Channel Roughness on Heat Transfer and Fluid Flow
Microelectromechanical Systems, Vol. 6, no. 2, Characteristics at Low Reynolds Numbers in Small
(1997) pp. 167-178. Diameter Tubes”, 3Sh National Heat Transfer
[24] Chu, W.K.H., Hsu, U.T., Wong, M. & Zohar, Y., Conference June 200 1, Anaheim, California
“Heat transfer in a microchannel flow [substrate [39] Schlichting, H., Boundarv laver theom, McGraw
cooling],” Proceedings of IEEE Electron Devices Hill, (1968) pp. 747.
Meeting, Hong Kong 1994, pp. 38-43. [40] Incropera, F. & Dewitt, D., Fundamentals of heat
[25] Fan, Q. & Xue, H., “Compressible effects in and mass transfer, John Wiley, New York, Toronto,
microchannel flows,” Proceedings of 2nd Electronics Singapore, (1996), pp 886, 1996.
Packaging Technology Conference, 1998, pp. 224 - [41] Weisburg, A., Bau, H. And Zemel, J.N., “Analysis of
228. microchannels for integrated cooling,” Int. J Heat
[26] Harley, J.C., Huang, Y., Bau, H.H. & Zemel, J.N., Mass Transfer, Vol. 35, no. 10, (1992) pp 2465-2474.
“Gas flow in micro-channels,” J. Fluid Mech., Vol. [42] SABRY, M. N., “Transversi: Temperature Gradient
284, (1995) pp.257-274. effect on fin efficiency for microchannel design,”
[27] Raju, R. & Roy, S., “Numerical Study of Heat ASME, J. Electronics Packagirzg, (200 l), pp. 344-350.
Transfer in High Speed Microflows ”, 36th AIAA
Thermophysics Conference Jun 2003 Orlando, Florida
[28] Raju, R. & Roy, S., “Modeling Single Component
Fluid Transport through Micro Channels and Free
Molecule Micro-Resistojet”, 4TdAerospace Sciences
Meeting and Exhibit January 2004 Reno, Nevada
[29] Li, W-L., Lin, J-W., Lee, S-C. & Chen, M-D.
“Effects of roughness on rarefied gas flow in long
microtubes”, Journal Of Micromechanics And
Microengineering 12 (2002) 149-1 56
[30] Richter, M., Woias, P. & Weiss, D. “Microchannels
for applications in Liquid Dosing and Flow-Rate
Measurement,” Sensors and Actuators A, Vol 62,
(1997) pp 480-483.
[31] Toh, K.C., Chen, X.Y. & Chai, J.C., “Numerical
computation of fluid flow and heat transfer in
microchannels”, International Journal of Heat and
Mass Transfer, 45 (2002) 5 133-5 14 1
[32] Mala, G.M., Li, D., & Dale, J.D., “Heat Transfer and
Fluid Flow in Micro-channels,’’ Int. J. Heat Mass
Transfer, Vol. 40, No. 13, (1997) pp 3079-3088.
-518-
5th. Int. Conf. o n Tllermal and Mechanical Simulation and Experiments in Micro-electronics and Micro-Systems, E~rroS1mE2004