The Effect of CPU Location in Total Immersion of Microelectronics

World Academy of Science, Engineering and Technology
International Journal of Electrical, Computer, Energetic, Electronic and Communication Engineering Vol:9, No:4, 2015
The Effect of CPU Location in Total Immersion of

Microelectronics
A. Almaneea, N. Kapur, J. L. Summers, H. M. Thompson
International Science Index, Mechanical and Mechatronics Engineering Vol:9, No:4, 2015 waset.org/Publication/10000942
AbstractMeeting the growth in demand for digital services

such as social media, telecommunications, and business and cloud
services requires large scale data centres, which has led to an increase
in their end use energy demand. Generally, over 30% of data centre
power is consumed by the necessary cooling overhead. Thus energy
can be reduced by improving the cooling efficiency. Air and liquid
can both be used as cooling media for the data centre. Traditional
data centre cooling systems use air, however liquid is recognised as a
promising method that can handle the more densely packed data
centres. Liquid cooling can be classified into three methods; rack heat
exchanger, on-chip heat exchanger and full immersion of the
microelectronics. This study quantifies the improvements of heat
transfer specifically for the case of immersed microelectronics by
varying the CPU and heat sink location. Immersion of the server is
achieved by filling the gap between the microelectronics and a water
jacket with a dielectric liquid which convects the heat from the CPU
to the water jacket on the opposite side. Heat transfer is governed by
two physical mechanisms, which is natural convection for the fixed
enclosure filled with dielectric liquid and forced convection for the
water that is pumped through the water jacket. The model in this
study is validated with published numerical and experimental work
and shows good agreement with previous work. The results show that
the heat transfer performance and Nusselt number (Nu) is improved
by 89% by placing the CPU and heat sink on the bottom of the
microelectronics enclosure.
KeywordsCPU location, data centre cooling, heat sink in

enclosures, Immersed microelectronics, turbulent natural convection
in enclosures.
I. INTRODUCTION
HE wide spread growth of data centers in recent years is

resulting in an increase in end use energy demand of
which more than 30% is required for their cooling [1].
Therefore more efficient cooling methods are essential, but
most data centres use air cooling techniques. However,
approaches that use liquids offer greater efficiency of heat
removal due to their density and are readily adopted in
situations of high density Datacom equipment. The revival
from the 1960s [2] of liquid-cooled datacom equipment brings
water from the edge of the data centre to the racks or brings
water into the rack to enable liquids to get closer to the main
sources of heat dissipation. There are different types of liquid
cooled microelectronics which bring water into the rack or via
water jackets in contact with a dielectrically immersed
microelectronics enclosure.
Heat transfer in total immersion of microelectronics is
A. Almaneea (corresponding author), N. Kapur, J. L. Summers and H. M.
Thompson are with the Institute of Engineering Thermofluids (iTF), School of
Mechanical Engineering, University of Leeds, United Kingdom (e-mail:
mnaalm@leeds.ac.uk).
International Scholarly and Scientific Research & Innovation 9(4) 2015
achieved by natural convection of the dielectric liquid, where

the heat is transferred from all of the microelectronic
components, including the CPUs, to the water jacket on the
opposing surface. This study develops a theoretical method for
quantifying the improvements of heat transfer when the heat
sink and CPU vertical location is varied as shown in Fig. 1.
The analysis presented in this paper is based on natural
convection of liquid in an enclosed cavity, where there is heat
flux into and out of the cavity on each vertical wall; heat is
transferred by natural convection, without the use of pumps.
Previously, [3] established the first theoretical benchmark for
heat transfer via laminar flow in a square enclosure where the
left wall is cold and the right wall is hot. Tian [4] performed
benchmark experiments for the turbulent flow in a square
enclosure and he plot thermal counter and vector for first time
in turbulent enclosure. In the case of rectangular enclosures,
[5] has investigated the fluid behaviour in both cases of
laminar and turbulent flows, where it was found that for aspect
ratios between 1 and 40 and Prandtl numbers between 1 and
20 the fluid flow becomes turbulent when the Rayleigh
number, Ra>107. For the localised heat source in an enclosure
[6] numerically investigated a vertical rectangular cavity with
three heat sources on one wall and the opposing wall is
maintained at a cold temperature. This study determined
theoretically the optimum spacing between the heat sources
and reported spacing arrangements that can drop the
temperature by 10%. Heat sources in enclosures have been
investigated by many further studies [7]-[9]. However, the use
of fins to increase surface area has been the focus of much
fewer studies. Nada [10] carried out an experimental study on
finned vertical rectangular enclosures. In this case horizontal
fins were adopted on the hot wall side and the opposing wall
was cold. This study also investigated the effect of different
Rayleigh numbers (Ra) and different fin lengths and spacing.
It was found that increasing the fin length increased the
Nusselt number and improves the heat transfer.
In the present work a model of turbulent natural convection
flow and conjugate heat transfer is used to investigate
vertically placed microelectronics in a sealed enclosure filled
with dielectric liquid. The model geometry and water jacket
inlet temperature remain constant. The CPU heat flux and
location are varied with the aim of quantifying heat transfer
improvements based on heat sink and CPU location that
reduce the CPU case temperature.
II. THE MODEL DESCRIPTION
The geometry and boundary conditions of the liquid
immersed microelectronics model is shown in Fig. 1. Heat is
431
scholar.waset.org/1999.5/10000942

geenerated from the CPU undderneath the hheat sink and cooled
c
byy the water thhat passes throough the solidd block repressenting
the water jackket at the floow rate 3x10-7 m3/s andd inlet
mperature 3033 K. The conjjugate heat moodel with sym
mmetry
tem
planes for the immersed liquid approacch is simulateed via
OMSOL 4.3. The simulatioon model is vvalidated againnst the
CO
exxperimental woork for naturaal heat convecttion in [10], w
where a
3220mm x 200m
mm x 40mm ennclosure with fins attached to the
hoot wall. The buuoyancy driveen flow inside the enclosure is due
to the temperatuure differencee between the hot and cold walls.
Thhe average errror between the simulatioon model presented
heere and the expperimental woork in [10] is 44.8%.
F
For natural coonvection insside the sealeed enclosure,, the
diellectric fluid floow behaviour can be indicatted by consideering
the Rayleigh num
mber [12], Ra
wheere H is the eenclosure heigght, g is the aacceleration duue to

gravvity, cp is speccific heat capaacity, D is dieelectric densityy, D
is tthe dielectric thermal connductivity, D is the dieleectric
visccosity, is the dielectriics coefficiennt of volum
metric
expansion. All thhese fluid theermal properties are taken from
Chi et al [13] andd listed in Tabble I. The Ra iss equal to 1.8xx108,
i
thereefore that the flow
whiich is greater tthan 107 and indicates
is tuurbulent [5].
TAB
BLE I
WORKING FLUID THERMAL
H
PROPERTTIES
Abbbrev.
Properties
Dielectric liquid
S
Specific heat capaacity
Cp
1140 (J/kg.K)
Thermal expansiion
1.151496x10-3 (K-11 )
Dynamic viscosiity
1.124782x10-3 (Pa..s)
D
T
Thermal conductivity
D
6.9x10-2 (W/m.K))
Innside the encloosure the fluidd flow is turbuulent and goveerned

by the conservaation equationns for mass,, momentum and
enerrgy for turbulent natural coonvection of thhe dielectric liiquid
dom
main.
C
Continuity equuation
0
M
Momentum equuation
s
simuulation model ussed in
Fig. 1 The geeometry of the symmetry
COMSOL ((a) Isometric vieew (b) Side view
w (c) Front View
w
III. MATHEM
MATICAL MOD
DEL
In this study,, the model coonsists of two types of fluidd flow,
naamely dielectrric liquid thatt circulates innside the encclosure
duue to natural convection and the watter that is puumped
through channeels to carry thhe heat away from the encllosure.
model, the twoo fluid
Beefore solving the full conjugate heat m
floows are checkked to determiine whether thhe flow is turbbulent.
Foor the water fflowing in thhe channel thiis is determinned by
caalculating the Reynolds
R
num
mber, Re
whhere D is the water channeel diameter (3..5x10-3 m), V is the

avverage flow veelocity (1.2x100-2 m/s), w (9996.6 kg/m3) aand w
(77.96x10-4 Pa.ss) are the ddensity and vviscosity of water,
respectively. Thhe Re is 53 forr a flow rate off 1.15x10-7 m3/s and
caan therefore bee considered aas laminar [11]].
2
k
3
T
The body forcee, F, depends mainly on thee density variaation.
For an increase iin temperaturee there is a ddecrease in dennsity
whiich creates thee driving buoyaancy force, exxpressed as:
T
wheere g = (0, 0, ).
T
The dielectric fluid
f
has density-temperaturre variations
(T) = 17716.2 -2.2T
Inn turbulent natural
n
conveection in encclosures, the k-
turbbulent model hhas been founnd to be a robbust model which
w
can offer solutionns that are cloose to experim
mental observattions
model
as investigated in [14]-[16]]. The k- turbulent m
introoduces two additional variiables; turbuleent kinetic eneergy,
k, aand specific disssipation rate, . The two additional transsport
equations are based on [17], whhich are:
432
.
.
.
.

The turbulentt viscosity cann be defined ass

DTT
The productioon term is fouund from the flluid velocity aas

p
2
3
2
k .
3
The energy equation wouldd be written ass:
whhere D is therrmal conductivvity and PrT iss a turbulent Prandtl

P
nuumber based oon the Kays-Crrawford equattion [17].
The empiricaal turbulent moodel constants parameters arre
13
,
25
1
,
2
1
,
2
9
,
125
9
,
0
100
0.41,
Fig. 2 CPU locaations, where S is varying as shhown in Table II

I
TAB
BLE II
CPU LO
OCATION
5.2
S (mm)
For the pum

mped water chhannel which is a laminarr fluid
doomain, the govverning equations are:
Continuity eqquation
.
Momentum equation
e
Energy equattion
c
In the solid ddomain, only thhe energy equuation is requirred

0
. T
whhere, the s iss the thermal cconductivity ffor the solids which
usse properties oof copper for the heat sinkk and aluminiuum for
the water jackett and its fins.
The workingg fluids in thiss paper are waater and a dieelectric
liqquid (Table I)). The water is used as a cooling liquiid that
paasses throughh the channells of the waater jacket annd the
m
microelectroniccs enclosure iss filled with tthe dielectric liquid
foor which the thhermal properrties are kept constant exceept for
the density whicch is a functioon of temperatuure.
IV
V. EFFECT OFF CPU LOCATIION ON HEAT TRANSFER RESULTS
The followinng investigatioon determines the effect of the
CP
PU and heat sink locationn on the heat transfer withhin the
im
mmersed microoelectronics ennclosure. Thee CPU and heaat sink
locations are vaaried as shownn in Fig. 2. Thhe variation iss from
the top to the boottom of the eenclosure and the CPU locaation is
where the values are listed inn Table II.
deefined as (S) w
Case 1
Case 2
Case 3
300
200
100
T
The effect of the CPU andd heat sink llocation on N
Nu is
pressented in Fig.. 3. The heat flux load (qL) is also varrying
from
m 70 to 100%
% (where heat flux q is 88888.88 W/m2 at heat
fluxx load qL 100 %) while the flow rate andd inlet temperaature
werre kept constannt at 3x10-7 m3/s and 303 K respectively.. The
highhest Nu impliies the best heat transfer performance annd it
occuurs when the CPU is locateed toward the bottom at S=
= 100
mm
m which is for case 3. Fig. 3 shows that thhe Nu is increeased
by 89% betweenn case 1 and 3, where S= 300 and 100 mm
resppectively. Thiss indicates thaat the CPU andd heat sink loccated
tow
ward the bottom
m of the encllosure has greeater effect onn the
heatt transfer perfo
formance.
A
As the dielecttric liquid is heated its ddensity decreaases;
whiich drives the buoyancy indduced circulattion. At the toop of
the enclosure it is directed ttowards the ccold plate byy the
pressence of the ttop wall wherre it transfers heat to the coooler
watter jacket annd then its density
d
increases causing the
diellectric liquid to travel bacck down to tthe bottom off the
encllosure. This bbuoyancy forcce creates the circulation of the
fluidd and hence transfers the heat from thee heat sink too the
coldd water jackett. The locationn of the CPU ddictates the sizze of
the flow domainn over which meaningful heat
h
transfer ttakes
most clearly w
with the buoyyancy
placce. Fig. 4 illuustrates this m
drivven eddy existting at the levvel of the CPU
U and above. The
fluidd below the C
CPU is not heeated and therrefore does pllay a
rolee in the global flow.
W
When the CPU
U is located toowards the topp of the channnel, a
largge section of thhe channel doees not experiennce any flow the
reciirculation sits between the C
CPU and the ttop of the walll and
the fluid beneath it hardly movves. Consequenntly there is onnly a
smaall surface areea on the coool wall for reemoving the heat.
Witth the CPU located at thhe bottom of the channel,, the
reciirculation fillss the entire cavvity and the eeffective area over
433

whhich meaningfful heat transffer takes placee on the cool wall

w is
larrgest hence the temperatuure of the dieelectric fluid w
will be
coooler when it ffirst meets the CPU.
REFER
RENCES
[1]
[2]
[3]
[4]
[5]
[6]
Fig. 3 Variattion of heat fluxx load for differrent CPU locatioon
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
Fig. 4 Effect of the CPU locaation on the vellocity for the thrree
ddifferent cases. T
These velocity profiles
p
are seleected for the besst heat
t
transfer
perform
mance when the heat flux load iis 100%, flow rrate is
3x10-7 m3/s and waterr inlet temperatuure is 303 k
[15]
[16]
V. CON
NCLUSION
This paper has
h investigateed heat transffer of encloseed and
im
mmersed micrroelectronics in a dielecttric liquid. N
Natural
coonvection and conjugate heaat transfer for a vertical encclosure
is simulated ussing COMSOL
L v4.3. Sincee the Ra>107 in the
microelectroniccs enclosure, the dielectriic fluid regime is
m
turbulent and thhe k- turbulennce model is eemployed. The fluid
inside the wateer channel is llaminar becauuse Re is veryy low.
mpared with previous pubblished
Thhe model sett up was com
exxperimental results and show
wed that the model
m
in this sttudy is
in good agreement with previious work [10]].
Different cases were modeeled to establissh the CPU annd heat
sinnk location thhat can yield the best heatt transfer. Thhe best
CP
PU and heat sink
s
arrangem
ment was able to show a booost in
the Nu number by 89%.
[17]
434
Shah, A., et all. Impact of rackk-level compactioon on the data ccenter

cooling ensembble. in Thermal aand Thermomechhanical Phenomeena in
Electronic Systeems, 2008. ITHER
RM 2008. 11th Inntersociety Confeerence
on. 2008: IEEE.
Anderson, D., F
F. Sparacio, and R.
R M. Tomasulo, The IBM System
m/360
model 91: Machhine philosophy annd instruction-hanndling. IBM Jourrnal of
Research and Deevelopment, 19677. 11(1): p. 8-24.
de Vahl Davis, G., Natural conveection of air in a square cavity: a bbench
mark numerical solution. Internattional Journal for Numerical Methoods in
Fluids, 1983. 3(33): p. 249-264.
Tian, Y. and T. Karayiannis, Low
w turbulence naturral convection in an air
c
part I: the thermal annd fluid flow ffields.
filled square cavity:
International Journal of Heat annd Mass Transferr, 2000. 43(6): p.. 849866.
MacGregor, R., Free convection through vertical plane layers-modderate
T
ASME, Jouurnal of Heat Traansfer,
and high Prandtl number fluids. Trans.
1969. 91: p. 3911-403.
Phan-Thien, Y.L
L., Nhan, An opttimum spacing prroblem for three chips
mounted on a vertical substraate in an enclossure. Numerical Heat
A: Applications, 20000. 37(6): p. 6133-630.
Transfer: Part A
Heindel, T., S. Ramadhyani, and F. Incroperra, Conjugate natural
n
m an array of prootruding heat souurces. Numericall Heat
convection from
Transfer, Part A Applications, 19996. 29(1): p. 1-188.
Keyhani, M., L. Chen, and D. P
Pitts, The aspect ratio effect on natural
n
convection in ann enclosure with protruding
p
heat soources. Journal off Heat
Transfer (Transaactions of the AS
SME (American Society of Mechanical
Engineers), Seriies C);(United Staates), 1991. 113(44).
Wroblewski, D. and Y. Joshi, Liiquid immersion cooling of a subsstratemounted protruusion in a three--dimensional encclosure: the effeccts of
geometry and boundary condittions. Journal off heat transfer, 1994.
116(1): p. 112-1119.
Nada, S., Naturral convection heat
h
transfer in hhorizontal and veertical
closed narrow enclosures with heated rectanguular finned base plate.
International Journal of Heat annd Mass Transferr, 2007. 50(3): p.. 667679.
R.H. Turner, and JJ.M. Cimbala, Funndamentals of theermalengel, Y.A., R
fluid sciences 20008. p. 534.
Holman, J., Heaat transfer, 9th. 20002, McGraw-Hilll. p. 335-337.
Chi, Y.Q., Jonaathan Summers, P
Peter Hopton, Keiith Deakin, Alan Real,
Nik Kapur and Harvey Thompsoon, Case Study of
o a Data Centre U
Using
Enclosed, Immeersed, Direct Liqquid-Cooled Servver, in Semicondductor
Thermal Measuurement and Management Sympossium (SEMI-THE
ERM).
2014, IEEE.
Zitzmann, T., ett al. Simulation oof steady-state naatural convection using
CFD. in Proc. of the 9th Interrnational IBPSA
A Conference Buuilding
Simulation 20055. 2005: Montrall: IBPSA.
Rundle, C. and M.
M Lightstone. Vaalidation of turbuulent natural conveection
in a square cavity for applicationn of CFD modelliing to heat transfeer and
fluid flow in atria geometriess. in 2nd Canaadian Solar Builldings
Conference, Callgary. 2007.
Aounallah, M., et al., Numeriical investigationn of turbulent natural
n
convection in an inclined squuare cavity withh a hot wavy wall.
International Jouurnal of Heat andd Mass Transfer, 2007. 50(9): p. 16831693.
Wilcox, D.C., Turbulence moddeling for CFD. Vol. 2. 1998: DCW
industries La Caanada, CA.

The Effect of CPU Location in Total Immersion of Microelectronics

Caricato da

Informazioni sul documento

Copyright

Formati disponibili

Condividi questo documento

Condividi o incorpora il documento

Opzioni di condivisione

Hai trovato utile questo documento?

Questo contenuto è inappropriato?

Copyright:

Formati disponibili

The Effect of CPU Location in Total Immersion of Microelectronics

Caricato da

Copyright:

Formati disponibili

World Academy of Science, Engineering and Technology