The Effect of CPU Location in Total Immersion of Microelectronics

7/26/2019 The Effect of CPU Location in Total Immersion of Microelectronics

http://slidepdf.com/reader/full/the-effect-of-cpu-location-in-total-immersion-of-microelectronics 1/4

Abstract — Meeting the growth in demand for digital servicessuch as social media, telecommunications, and business and cloudservices requires large scale data centres, which has led to an increasein their end use energy demand. Generally, over 30% of data centre

power is consumed by the necessary cooling overhead. Thus energycan be reduced by improving the cooling efficiency. Air and liquidcan both be used as cooling media for the data centre. Traditionaldata 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 heatexchanger, on-chip heat exchanger and full immersion of themicroelectronics. This study quantifies the improvements of heattransfer specifically for the case of immersed microelectronics byvarying the CPU and heat sink location. Immersion of the server isachieved by filling the gap between the microelectronics and a water

jacket with a dielectric liquid which convects the heat from the CPUto the water jacket on the opposite side. Heat transfer is governed bytwo physical mechanisms, which is natural convection for the fixedenclosure filled with dielectric liquid and forced convection for thewater that is pumped through the water jacket. The model in thisstudy is validated with published numerical and experimental workand shows good agreement with previous work. The results show thatthe 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.

Keywords — CPU location, data centre cooling, heat sink inenclosures, Immersed microelectronics, turbulent natural convectionin enclosures.

I. I NTRODUCTION

HE wide spread growth of data centers in recent years isresulting 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, butmost data centres use air cooling techniques. However,approaches that use liquids offer greater efficiency of heatremoval due to their density and are readily adopted insituations of high density Datacom equipment. The revivalfrom the 1960s [2] of liquid-cooled datacom equipment bringswater from the edge of the data centre to the racks or bringswater into the rack to enable liquids to get closer to the mainsources of heat dissipation. There are different types of liquidcooled microelectronics which bring water into the rack or viawater jackets in contact with a dielectrically immersedmicroelectronics 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 ofMechanical Engineering, University of Leeds, United Kingdom (e-mail:[email protected]).

achieved by natural convection of the dielectric liquid, wherethe heat is transferred from all of the microelectroniccomponents, including the CPUs, to the water jacket on theopposing surface. This study develops a theoretical method forquantifying the improvements of heat transfer when the heatsink and CPU vertical location is varied as shown in Fig. 1.

The analysis presented in this paper is based on naturalconvection of liquid in an enclosed cavity, where there is heatflux 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 forheat transfer via laminar flow in a square enclosure where theleft wall is cold and the right wall is hot. Tian [4] performed

benchmark experiments for the turbulent flow in a squareenclosure and he plot thermal counter and vector for first timein turbulent enclosure. In the case of rectangular enclosures,[5] has investigated the fluid behaviour in both cases oflaminar and turbulent flows, where it was found that for aspectratios between 1 and 40 and Prandtl numbers between 1 and20 the fluid flow becomes turbulent when the Rayleighnumber, Ra>10 7. For the localised heat source in an enclosure[6] numerically investigated a vertical rectangular cavity withthree heat sources on one wall and the opposing wall ismaintained at a cold temperature. This study determinedtheoretically the optimum spacing between the heat sourcesand reported spacing arrangements that can drop thetemperature by 10%. Heat sources in enclosures have beeninvestigated by many further studies [7]-[9]. However, the useof fins to increase surface area has been the focus of muchfewer studies. Nada [10] carried out an experimental study onfinned vertical rectangular enclosures. In this case horizontalfins were adopted on the hot wall side and the opposing wallwas cold. This study also investigated the effect of differentRayleigh 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 investigatevertically placed microelectronics in a sealed enclosure filledwith dielectric liquid. The model geometry and water jacketinlet temperature remain constant. The CPU heat flux andlocation are varied with the aim of quantifying heat transferimprovements based on heat sink and CPU location thatreduce the CPU case temperature.

II. THE MODEL DESCRIPTION

The geometry and boundary conditions of the liquidimmersed microelectronics model is shown in Fig. 1. Heat is

The Effect of CPU Location in Total Immersion of

MicroelectronicsA. Almaneea, N. Kapur, J. L. Summers, H. M. Thompson

T

World Academy of Science, Engineering and TechnologyInternational Journal of Electrical, Computer, Energetic, Electronic and Communication Engineering Vol:9, No:4, 2015

431International Scholarly and Scientific Research & Innovation 9(4) 2015 scholar.waset.org/1999.5/10000942

I n t e r n a t i o n a l S c i e n c e I n d e x , M e c h a n i c a l a n d

M e c h a t r o n i c s E n g i n e e r i n g

V o l : 9 ,

N o : 4 ,

2 0 1 5 w a s e t . o r g / P u b l i c a t i o n / 1 0 0 0 0 9 4 2

http://scholar.waset.org/1999.5/10000942


http://waset.org/publication/The-Effect-of-CPU-Location-in-Total-Immersion-of-Microelectronics/10000942



g bthte

pl

Ce3htoTh

ndthBflFca

wa(7reca

nerated fromthe water th

e water jac perature 30

anes for the

MSOL 4.3. perimental w0mm x 200t wall. The b the temperate average ere and the ex

Fig. 1 The gCOMSOL (

In this study,mely dielecte to naturalrough channefore solvingws are checr the waterlculating the

ere D is theerage flow v.96x10 -4 Pa.spectively. Tn therefore b

the CPU unat passes thr et at the fl K. The conjimmersed li

The simulatiork for naturam x 40mm eoyancy drivere differenceror betweenerimental wo

eometry of thea) Isometric vi

III. MATHE

the model coic liquid thaconvection

ls to carry ththe full conjed to determiflowing in thReynolds nu

water channelocity (1.2x1) are the de Re is 53 for considered a

erneath the hugh the soli

ow rate 3x1ugate heat mquid approac

n model is vl heat convecclosure with

n flow inside between thethe simulatirk in [10] is 4

ymmetry simuw (b) Side vie

MATICAL MO

nsists of twot circulates iand the wate heat awayugate heatne whether te channel thi ber, Re

μ

l diameter (3.0-2 m/s), ρ w (ensity anda flow rate o

s laminar [11]

eat sink and block repres0-7 m3/s andodel with syh is simulat

alidated agaiion in [10],fins attachedthe enclosurehot and coldn model pre.8%.

lation model us (c) Front Vie

EL

types of fluiside the enc

er that is pfrom the enclodel, the twe flow is tur s is determi

.5x10 -3 m), V96.6 kg/m 3)iscosity of

f 1.15x10 -7 m.

ooledenting

inletmetryd via

st thehere ato theis duewalls.sented

ed in

flow,losurempedosure.

fluidulent.ed by

is thend μ w water,/s and

dielthe

whgrais tvisexpChiwhiis t

S

I

byenedo

ρ

Forwhi

wh

Iturbcanasintr k, aequ

or natural cectric fluid flRayleigh nu

re H is the eity, c p is spe

he dielectricosity, β isansion. All tet al [13] anch is greaterrbulent [5].

W

Properties

pecific heat capaThermal expansiDynamic viscoshermal conducti

side the enclthe conserv

rgy for turbulain.ontinuity equ

omentum eq

. .p

he body forcan increase i

ch creates the

re g = (0, 0,he dielectric

turbulentulent modeloffer solutioinvestigated

oduces two and specific diations are bas

ρ .

ρ . ω

onvection inw behaviour

ber [12], Ra

nclosure heiific heat capthermal conthe dielectriese fluid the listed in Tabhan 10 7 and

TAORKING FLUID T

Ab

cityontyvity

osure the fluition equatioent natural co

ation

.ρuation

μ μ

, F , dependsn temperatur driving buoy

).luid has dens

ρ(T) = 17

atural convhas been fous that are cloin [14]-[16]

dditional varissipation rate,ed on [17], w

.μ μ

.μ μ

ide the sealcan be indica

μ

ht, g is the acity, ρ D is diductivity, μ D

c’s coefficiermal propertile I. The Ra indicates ther

LE IERMAL PROPER

brev.

C p Β 1µD 1λ D

flow is turbs for mass,

nvection of t

0

23μ

mainly on thee there is a dancy force, e

ρT

ity-temperatu

16.2 -2.2T

ction in end to be a ro

se to experim. The k-ωables; turbule ω. The two aich are:

∗ k p

ω α

ed enclosure,ed by consid

cceleration dlectric densit is the diel

t of volues are taken

equal to 1.8fore that the

IES

Dielectric liquid

1140 (J/kg.K).151496x10 -3 (K -1

.124782x10 -3 (Pa.6.9x10 -2 (W/m.K

lent and gov momentume dielectric li

. 23ρk

density variaecrease in de pressed as:

e variations

losures, theust model

ental observaturbulent

nt kinetic endditional tran

∗

p

thering

e to, λ D

ctricetricfrom108,

flow

1 ).s)

rnedand

quid

tion.nsity

k-ωhichtionsodelrgy,

sport





V o l : 9 ,

N o : 4 ,







wn

d

wusth

li pmfoth

I

Cilothd

The turbulen

The producti

p μ The energy e

ere λ D is thember based oThe empirica

1325, ∗ 12 For the pumain, the goContinuity e

Momentum

ρ . Energy equa

In the solid d

ere, the λ s ise propertiese water jackeThe workinguid (Table Isses throughicroelectronicr which the te density whi

. EFFECT O

The followiU and heatmersed micr

cations are ve top to the bfined as (S)

viscosity ca

μD

on term is fou

: quation woul

ρc T .mal conductin the Kays-Cl turbulent m

, 12, ped water cerning equatiuation

.quation

.pion

ρcomain, only t

0 the thermalf copper forand its fins.fluids in this

). The waterthe channel

s enclosure ihermal propech is a functio

CPU LOCATI

g investigatisink locationelectronics eried as show

ottom of thehere the valu

be defined a

nd from the fl

23 .

be written a

λ c ity and Pr T iawford equatdel constants

9125, ∗ 910annel which

ons are:

0

μ

.λ T e energy equ

.λ T conductivitythe heat sink

paper are wis used as as of the wa filled withties are keptn of temperat

ON ON HEAT

on determine on the heatnclosure. The in Fig. 2. Tnclosure andes are listed i

uid velocity a

23ρk .

:

Ts a turbulention [17]. parameters a

, 0.41,is a lamina

ation is requi

or the solidsand alumini

ater and a diecooling liquiter jacket ahe dielectricconstant excre.

RANSFER R E

s the effecttransfer with

CPU and hehe variation ithe CPU loca Table II.

s

randtl

e

5.2

fluid

ed

whichm for

lectricd thatd theliquid pt for

SULTS

of thein thet sinkfrom

tion is

prefrofluwer hig

occmm

byrestowhea

whithe

prewatdielencfluicolthe

pladrifluirole

lar recithesm

Witreci

Fig. 2 CPU loc

CaseCaseCase

he effect ofented in Fig. 70 to 100

load q L 100e kept constaest Nu impli

urs when thewhich is for

89% betweenectively. Thiard the bottot transfer perf s the dielec

ch drives theenclosure itence of the ter jacket anectric liquidlosure. Thisd and hence

water jacketflow domaine. Fig. 4 illen eddy exisd below thein the globalhen the CP

e section of trculation sitsfluid beneathll surface ar

h the CPUrculation fills

ations, where S

TACPU L

1

the CPU an. 3. The heat

(where heat%) while thet at 3x10 -7 m

es the best h

CPU is locatcase 3. Fig. 3 case 1 andindicates tha of the encl

ormance.tric liquid is buoyancy inis directed

op wall wher d then itsto travel bauoyancy for transfers the. The locatio over whichstrates thising at the lePU is not h

flow.is located to

e channel do between theit hardly mova on the co

located at th the entire ca

is varying as s

LE IICATION

S (mm)

300200100

d heat sink lflux load (q

flux q is 888flow rate an3/s and 303eat transfer p

d toward theshows that t

3, where S=t the CPU anosure has gr

heated itsuced circulatowards thee it transfersdensity increk down to te creates theheat from th of the CPUmeaningfulost clearlyel of the CPated and the

wards the toes not experie

PU and the tes. Consequel wall for r

e bottom ofvity and the

own in Table I

ocation onL) is also va8.88 W/m 2 at

inlet temper respectively.erformance a

bottom at S=e Nu is incr

300 and 100heat sink lo

ater effect o

ensity decreion. At the told plate bheat to the cases causinghe bottom ocirculation o heat sink tictates the sieat transferith the buoy

U and above.efore does pl

of the channce any flowop of the waltly there is omoving the

the channel,ffective area

I

u isyingheat

atureThe

nd it

100asedmmatedthe

ases; p of

theolerthethe

f thethe

e ofakesancyTheay a

el, atheandly a

heat.

theover





V o l : 9 ,

N o : 4 ,







wlaco

t

icoismtuinTein

siC

th

ich meaninggest – henceoler when it

Fig. 3 Varia

Fig. 4 Effectifferent cases.ransfer perfor

3x10 -7

This papermersed micnvection andsimulated us

icroelectronicrbulent and tside the watee model set

perimental re good agreemDifferent cask location tU and heat

e Nu number

ful heat transthe temperat

irst meets the

ion of heat flux

of the CPU lochese velocityance when them3/s and water

V. CO

as investigatoelectronicsconjugate heing COMSOs enclosure,e k- turbuler channel is

up was cosults and shoent with previes were modeat can yield

sink arrange

by 89%.

er takes placre of the die

CPU.

load for differ

ation on the velrofiles are sele

heat flux loadinlet temperat

NCLUSION

d heat transin a dielectt transfer for

v4.3. Sincethe dielectr

nce model isaminar beca pared withed that the

ous work [10led to establithe best heaent was able

on the coollectric fluid

ent CPU locati

ocity for the thcted for the bes 100%, flow r re is 303 k

er of enclosric liquid.a vertical enc the Ra>10 7

c fluid regiemployed. Thse Re is ver

previous pubodel in this s.h the CPU at transfer. Thto show a b

all isill be

n

eet heatate is

d andaturallosurein theme ise fluid

low.lishedudy is

d heate bestost in

[1]

[2]

[3]

[4]

[5]

[6]

[7]

[8]

[9]

[10]

[11]

[12]

[13]

[14]

[15]

[16]

[17]

Shah, A., et alcooling ensembElectronic Systeon. 2008: IEEE.Anderson, D.,model 91: MacResearch and Dde Vahl Davis,mark numericalFluids, 1983. 3(Tian, Y. and T.filled squareInternational Jo866.MacGregor, R.,and high Prandt1969. 91: p. 391Phan-Thien, Y.mounted on aTransfer: Part AHeindel, T., Sconvection froTransfer, Part AKeyhani, M., Lconvection in aTransfer (TransEngineers), SeriWroblewski, D.mounted protrugeometry and116(1): p. 112-1

Nada, S., Natuclosed narrowInternational Jo679.Çengel, Y.A., R fluid sciences 2Holman, J., Hea

Chi, Y.Q., Jona Nik Kapur andEnclosed, ImmThermal Measu2014, IEEE.Zitzmann, T., eCFD. in Proc.Simulation 200Rundle, C. andin a square cavifluid flow inConference, CalAounallah, M.,convection inInternational Jo1693.Wilcox, D.C.,industries La Ca

R EFE

. Impact of racle. in Thermalms, 2008. ITHE . Sparacio, andine philosophy avelopment, 196

G., Natural convsolution. Interna): p. 249-264.

Karayiannis, Loavity: part I:

urnal of Heat a

Free convectionl number fluids.-403.., Nhan, An opvertical substr

: Applications, 2. Ramadhyani,

an array of pr Applications, 19. Chen, and D.

enclosure withctions of the A

es C);(United St and Y. Joshi, Lsion in a three-

boundary condit19.al convection h

enclosures withurnal of Heat a

.H. Turner, and J08. p. 534.

t transfer, 9th. 20

than Summers,Harvey Thompsrsed, Direct Li

rement and Man

al. Simulationof the 9th Inte. 2005: Montréal

. Lightstone. Vty for applicatioatria geometriegary. 2007.

et al., Numerian inclined sqrnal of Heat an

Turbulence monada, CA.

ENCES

-level compactiand Thermomec

M 2008. 11th I

. M. Tomasulo,nd instruction-ha. 11(1): p. 8-24.ction of air in aional Journal for

turbulence natuthe thermal ad Mass Transfer

through verticalrans. ASME, Jo

imum spacing pte in an enclo00. 37(6): p. 61

and F. Incropetruding heat so

96. 29(1): p. 1-1itts, The aspectrotruding heat sME (Americantes), 1991. 113(4quid immersiondimensional encions. Journal o

eat transfer inheated rectangud Mass Transfer

.M. Cimbala, Fu

02, McGraw-Hil

eter Hopton, Ken, Case Study ouid-Cooled Ser

agement Sympos

f steady-state narnational IBPSA: IBPSA.alidation of turbu of CFD modelli. in 2nd Cana

cal investigatioare cavity wit Mass Transfer,

eling for CFD.

n on the dataanical Phenometersociety Conf

The IBM Systedling. IBM Jour

square cavity: a Numerical Meth

ral convection ind fluid flow, 2000. 43(6): p.

plane layers-mournal of Heat Tr

roblem for threeure. Numerical-630.a, Conjugaterces. Numerical.ratio effect onurces. Journal o

Society of Mech).cooling of a sublosure: the effe heat transfer,

orizontal and vlar finned base, 2007. 50(3): p.

damentals of th

. p. 335-337.

th Deakin, Alanf a Data Centreer, in Semiconium (SEMI-TH

tural convectionConference Bu

lent natural convng to heat transf dian Solar Bui

of turbulenta hot wavy

2007. 50(9): p.

Vol. 2. 1998:

centerna inrence

/360nal of

benchods in

an airields.

. 849-

deratensfer,

chipsHeat

aturalHeat

aturalHeat

anical

trate-ts of1994.

rtical plate.. 667-

rmal-

Real,UsinguctorRM).

usingilding

ectionr and

ldings

aturalwall.

1683-

DCW





V o l : 9 ,

N o : 4 ,





The Effect of CPU Location in Total Immersion of Microelectronics

Documents

Transcript of The Effect of CPU Location in Total Immersion of Microelectronics