Sunteți pe pagina 1din 81

IMPINGEMENT HEAT SINK OPTIMIZATION AND THERMAL

CHARACTERIZATION OF A POWER MODULE USING COMMERCIALLY


AVAILABLE CFD CODE

BY
SHILADITYA CHAKRAVORTY
BE, University of Mumbai, 2005

THESIS
Submitted in partial fulfillment of the requirements for
the degree of Master of Science in Industrial and Systems Engineering
in the Graduate School of
Binghamton University
State University of New York
2011

UMI Number: 1499658

All rights reserved


INFORMATION TO ALL USERS
The quality of this reproduction is dependent on the quality of the copy submitted.
In the unlikely event that the author did not send a complete manuscript
and there are missing pages, these will be noted. Also, if material had to be removed,
a note will indicate the deletion.

UMI 1499658
Copyright 2011 by ProQuest LLC.
All rights reserved. This edition of the work is protected against
unauthorized copying under Title 17, United States Code.

ProQuest LLC.
789 East Eisenhower Parkway
P.O. Box 1346
Ann Arbor, MI 48106 - 1346

Copyright by Shiladitya Chakravorty 2011


All Rights Reserved

Accepted in partial fulfillment of the requirements for


the degree of Master of Science in Industrial and Systems Engineering
in the Graduate School of
Binghamton University
State University of New York
2011
July 22, 2011

Dr. Bahgat G. Sammakia, Chair


Department of Mechanical Engineering, Binghamton University
Dr. Daryl Santos, Member
Department of Systems Science & Industrial Engineering, Binghamton University
Dr. Nagen Nagarur, Member
Department of Systems Science & Industrial Engineering, Binghamton University

iii

ABSTRACT
Considerable improvements in performance of semiconductor devices
have been made in recent years. This has been possible through improvements
in semiconductor design and manufacturing techniques which have facilitated
increased density of circuits. However, increased circuit densities have resulted
in consequent increase in power dissipation from these devices and in turn have
created the need for new and improved cooling solutions. Additionally, similar
improvements in electronics packaging design and manufacturing especially with
the advent of SMT have led to increased density of components on PCBs. This
has resulted in increased temperatures for entire systems, requiring system level
thermal management and optimization. Developing new thermal solutions and
optimizing existing technologies have traditionally involved only experimental
analysis, but in recent years numerical analysis using Computational Fluid
Dynamics (CFD) techniques are increasingly being used for the same. For
complex systems, however, validating a CFD model with experimental data is
essential before using it for further design and development. This study examines
and illustrates the use of commercially available CFD code in developing and
optimizing thermal management solutions.
This study is divided into two sections. The first section is about numerical
optimization of parallel plate impingement heat sink designs for semiconductor
applications with significant hot spots, using a commercially available CFD code.
In the second section, a project on thermal characterization of a power
module is presented. The project involved gathering experimental data from the

iv

power module and subsequently, creating a CFD model using commercial code.
The CFD model was then validated with the experimental data.

ACKNOWLEDGEMENTS
Firstly, I would like to submit my deepest gratitude to my committee chair, Dr.
Bahgat Sammakia, for his constant and patient supervision at all stages of this
research work. I would also like to thank him for his guidance, advice and his
encouragement from time to time that have been invaluable to me.
I would like to thank Dr. Varaprasad Calmidi for his time and his technical
guidance. His moral support throughout the course of these projects is much
appreciated.
I am grateful to Dr. Daryl Santos for his guidance on the statistical analysis
sections in this thesis. Dr. Krishnaswami Srihari and Dr. Gamal Refai Ahmed
deserve special thanks from my end for having advised me at various stages of
my research.
Finally, I wish to express my heartfelt thanks to Dr. Nagen Nagarur for agreeing
to be on my committee. I appreciate the time that he has taken out of his busy
schedule to review my thesis and offer his invaluable suggestions.

vi

TABLE OF CONTENTS
LIST OF TABLES ........................................................................................................... viii
LIST OF FIGURES........................................................................................................... ix
NOMENCLATURE ............................................................................................................ x
CHAPTER 1 .......................................................................................................................1
Introduction.........................................................................................................................1

1.1 Introduction to impingement heat sink optimization ................................................ 1


1.2 Introduction to thermal characterization study ........................................................ 7
1.3 Research Objective and Problem Statement.......................................................... 8

CHAPTER 2 .....................................................................................................................10
Numerical Optimization of Impingement Heat Sink ...................................................10

2.1 Model................................................................................................................... 11

2.1.1 Fan ............................................................................................................................... 12


2.1.2 Heat Sink ...................................................................................................................... 12
2.1.3 Chip .............................................................................................................................. 13
2.2 Method of Solution ............................................................................................... 14
2.3 Grid Sensitivity Analysis....................................................................................... 17
2.4 Results and Discussion........................................................................................ 19
2.4.1 Fin Height Optimization under Non-Uniform Power Distribution .................................. 19
2.4.2 Fin Shape Optimization under Non-Uniform Power Distribution .................................. 23
2.4.3 Observations ................................................................................................................ 25
2.5 Design Guidelines................................................................................................ 32
CHAPTER 3 .....................................................................................................................35
Thermal Characterization of a Power Module ............................................................35
3.1 Experimental Set Up ............................................................................................ 36
3.1.1 Test Environment ......................................................................................................... 38
3.1.2 Test Card ...................................................................................................................... 40
3.1.3 Air Velocity Transducer ................................................................................................ 40
3.1.4 Ambient Temperature Thermocouple .......................................................................... 40
3.2 Model................................................................................................................... 42
3.2.1 Test Card ...................................................................................................................... 44
3.2.2 Module Card ................................................................................................................. 44
3.2.3 Inductor......................................................................................................................... 45
3.2.4 FET 1&2 ....................................................................................................................... 46
3.2.5 Copper Column ............................................................................................................ 47
3.2.6 Solder Bumps ............................................................................................................... 47
3.2.7 Position and Placement................................................................................................ 48
3.3 Material Properties............................................................................................... 49
3.4 Method of Solution ............................................................................................... 53
3.5 Results ................................................................................................................ 55
3.5.1 Results and Discussion ................................................................................................ 55
3.5.2 Statistical Comparison of Experimental and Simulated Results .................................. 60
CHAPTER 4 .....................................................................................................................64
Conclusions ......................................................................................................................64
4.1 Summary ............................................................................................................. 64
4.2 Future Work ......................................................................................................... 65
REFERENCES ................................................................................................................67

vii

LIST OF TABLES
TABLE 1: MATERIAL PROPERTIES FOR FLUID ................................................................................12
TABLE 2: HEAT SINK DIMENSIONS .....................................................................................................12
TABLE 3: MATERIAL PROPERTIES (ISOTROPIC) ............................................................................13
0

TABLE 4: T (IN K) AT DIFFERENT INLET AIR VELOCITIES FOR DIFFERENT FIN HEIGHTS
.............................................................................................................................................................21
TABLE 5: P (IN PASCAL) AT DIFFERENT INLET AIR VELOCITIES FOR DIFFERENT FIN
HEIGHTS............................................................................................................................................22
TABLE 6: DIFFERENT DESIGN CONFIGURATIONS AT 2M/S INLET VELOCITY .......................30
TABLE 7: SELECTED DESIGNS FROM GROUP II AT 2.5 M/S INLET VELOCITY .................................................32
TABLE 8: EXPERIMENTAL RESULTS (STEADY STATE TEMPERATURE READINGS FROM
VARIOUS THERMOCOUPLES) .....................................................................................................41
TABLE 9: DIMENSIONS FOR VARIOUS COMPONENTS .................................................................47
TABLE 10: FLUID (AIR) PROPERTIES ..................................................................................................49
TABLE 11: MATERIAL PROPERTIES ....................................................................................................50
TABLE 12: DENSITY AND SPECIFIC HEAT FOR THE TEST AND MODULE CARDS ................51
TABLE 13: ANISOTROPIC THERMAL CONDUCTIVITY FOR TEST AND MODULE CARDS .....52
TABLE 14: HEAT TRANSFER COEFFICIENTS AT VARIOUS AIR INLET VELOCITIES .............52
TABLE 15: EXPERIMENTAL AND SIMULATED T VALUES FOR FET 1&2 .................................59
TABLE 16: EXPERIMENTAL AND SIMULATED T VALUES FOR THE INDUCTOR AND
MODULE CARD ................................................................................................................................59
TABLE 17: DIFFERENCE BETWEEN SIMULATED AND EXPERIMENTAL TEMPERATURE
VALUES FOR FET-1, FET-2, INDUCTOR AND MODULE CARD SURFACE........................60
TABLE 18: RESULTS FOR PAIRED T- TEST.......................................................................................62
TABLE 19: CONFIDENCE INTERVALS FOR DIFFERENCE BETWEEN EXPERIMENTAL AND
SIMULATED VALUES ......................................................................................................................63

viii

LIST OF FIGURES
FIGURE 1: SCHEMATIC DIAGRAM OF THE BASELINE CASE HEAT SINK MOUNTED ON THE
CHIP....................................................................................................................................................11
FIGURE 2: POWER DISTRIBUTION IN THE CHIP IN MW. ...............................................................13
FIGURE 3: SNAPSHOT OF THE FLUENT INTERFACE.....................................................................15
FIGURE 4: HEAT SINK MODEL WITH BOUNDARY CONDITIONS .................................................16
FIGURE 5: GRID SENSITIVITY STUDY ................................................................................................17
FIGURE 6: GAMBIT SNAPSHOT SHOWING THE MESH ..................................................................18
FIGURE 7: TEMPERATURE CONTOUR OF THE CHIP SURFACE SHOWING SIGNIFICANT
HOT-SPOTS (BASE CASE OF 12 MM FIN HEIGHT). ...............................................................20
FIGURE 8: PLOT OF T AS A FUNCTION OF FIN HEIGHT. ............................................................21
FIGURE 9: PLOT OF P AS A FUNCTION OF FIN HEIGHT. ............................................................23
FIGURE 10: VELOCITY CONTOUR COMPARISON BETWEEN BASE MODEL (LEFT) AND
DESIGN "H-44" (RIGHT) FOR 16 MM FIN HEIGHT AT INLET VELOCITY OF 2 M/S. .........24
FIGURE 11: DIFFERENT HEAT SINK MODELS THAT WERE STUDIED (CHARACTERISTIC
DIMENSIONS SHOWN FOR DESIGN D ALSO CORRESPONDS TO ALL DESIGNS
FROM E TO J). ............................................................................................................................27
FIGURE 12: ISOMETRIC VIEW OF DESIGN H-43 WITH ONE END FIN REMOVED, SHOWING
THE MODIFIED FIN SHAPES. .......................................................................................................28
FIGURE 13: T VS. P FOR DIFFERENT DESIGNS AT 2M/S INLET VELOCITY. ......................31
FIGURE 14: DESIGN II HEAT SINK MODELS AT AN INLET VELOCITY OF 2.5 M/S. .................34
FIGURE 15: SCHEMATICS OF EXPERIMENTAL SETUP .................................................................36
FIGURE 16: THERMOCOUPLE POSITIONS ON THE MODULE AND TWO OPPOSITE
DIRECTIONS OF FLOW (A & B) ....................................................................................................37
FIGURE 17: SCHEMATICS OF AN OPEN CIRCUIT WIND TUNNEL [26] .......................................39
FIGURE 18: CROSS SECTION OF THE MODULE AND THE TEST CARD ...................................42
FIGURE 19: SCHEMATIC OF THE TEST CARD AND MODULE ......................................................43
FIGURE 20: CROSS SECTIONAL VIEW OF THE INDUCTOR .........................................................45
FIGURE 21: CROSS SECTIONAL VIEW OF FET-1 ............................................................................46
FIGURE 22: POSITION OF MODULE ON THE TEST CARD .............................................................48
FIGURE 23: POSITION OF COMPONENTS ON THE MODULE CARD (ALL DIMENSIONS IN
MM) .....................................................................................................................................................49
FIGURE 24: SIGNAL AND POWER PLANES IN A PCB .....................................................................51
FIGURE 25: CFD MODEL WITH BOUNDARY CONDITIONS ............................................................55
FIGURE 26: VELOCITY CONTOUR PLOT 1.05 W AND AIR INLET VELOCITY OF 0.5 M/S ......56
FIGURE 27: TEMPERATURE CONTOURS AT 1.83 W AND 0.5 M/S AIR INLET VELOCITY .....58
FIGURE 28: TEMPERATURE CONTOURS AT 1.05 W AND 0.5 M/S AIR INLET VELOCITY .....58
FIGURE 29: NORMALITY PLOTS FOR FET-1 , FET-2 , INDUCTOR AND MODULE ..............................61

ix

NOMENCLATURE
Cp

Specific heat, J/Kg-K

Heat transfer coefficient, W/m2-K

Pressure, Pa

Thermal conductivity, W/m-K

Power dissipation, (W)

Thermal resistance,0C/W

Acceleration due to gravity 9.81 m/s2

Nu

Nusselt number

Temperature, oC or oK

Fluid velocity, m/s

x, y, z Coordinates, m
L

Characteristic dimension, m

Greek Symbols
T

Surface to ambient temperature difference, oK

Density, kg/m3

Dynamic viscosity, P

Subscripts
o

Outer planes

Inner planes

Fluid (air)

CHAPTER 1
Introduction
Improved performance of semiconductor devices in recent years has
resulted in consequent increase in power dissipation and has created the need
for optimized cooling solutions. Thermal analysis and management of electronic
components and printed circuit boards traditionally involved only experimental
studies but now are increasingly being conducted using Computational Fluid
Dynamics (CFD) techniques and software, reducing the requirement of
experimentation which, in turn, saves time as well as money. Commercially
available CFD codes have become major tools in these endeavors and have
resulted in drastic reduction in development time and costs of new and innovative
thermal management solutions.
1.1 Introduction to impingement heat sink optimization
Thermal performance of systems depends on component temperatures
which, in turn, depend on total peak power and power distribution in the
microprocessor. The primary goal of thermal management is to maintain
operating temperatures below certain limits to ensure functionality and reliability
of the system.

Systems can either be air-cooled or cooled by water (or any other


coolant). Water-cooled systems include microchannel heat sink designs meant
for high-power dissipating systems. Farnam et al. have developed and
numerically analyzed transient microchannel heat sink models which use water
as a coolant [1]. Air-cooled systems, on the other hand, are commonly used for
most applications. For air-cooled systems (heat sinks) which use ambient air for
cooling, the two primary mechanisms of increasing heat transfer are the
following: (1) increasing the surface area, and (2) increasing the mass flow rate
of air. Both mechanisms have limiting constraints attached to them. Increasing
surface area is bound by physical constraints of space and extrusion limits, etc.,
while increasing mass flow rate is limited by the corresponding increase in
pressure drop across the heat sink. To counter the increase in pressure drop one
can employ large fans, but that again results in increased acoustical noise and
overall system costs [2,7]. From a manufacturing point of view, aluminum is the
material of choice for heat sinks as fins can easily be formed by simple extrusion.
This method of manufacturing is good enough for heat sinks with a fin thickness
of 1.3 mm or more and a fin height-to-gap aspect ratio of up to 6, while being
simple and cost effective [3]. With special design techniques, fin thicknesses of
0.8 mm and aspect ratios of up to 10 are achievable, but the extrusion tolerances
are compromised as the aspect ratio increases. Such heat sinks are generally
limited to 18 inches in width and 3 inches in height and are suitable for relatively
low power applications because they present relatively lower surface area for
heat transfer per unit volume [3].

For higher aspect ratios of up to 25 or more,

epoxy-bonded fins can be used [4]. These heat sinks are made of an extruded or
machined base which is flat on one side and has grooves on the other side, and
on these grooves the fins can be attached using an epoxy resin [4,5,6]. The
epoxy interface does increase the thermal resistance due to lower thermal
conductivity of epoxy but this can be overcome by soldering or brazing the fins to
the base, although, that would lead to higher manufacturing costs. Aluminum or
copper can be used to manufacture such heat sinks [4,5,6].
Increased heat transfer area does not necessarily mean a proportional
improvement in thermal performance. In fact, increased heat transfer area in
some cases might lead to reduced heat transfer and degraded thermal
performance of the system if the design is not optimal [2,7]. This is because
adding a heat sink in the path of incoming air signifies an increase in the
resistance to air flow and if there is an alternative path available in the system, air
is going to by-pass the heat sink altogether and take the path of least resistance,
in turn reducing heat transfer [4]. Another factor which can substantially limit the
enhancement in thermal performance of a system, even after installing a heat
sink, is the upstream heating of air entering the heat sink due to power
dissipation from other sources [2,4,7]. A heat sink designed for a certain air flow
rate at a particular temperature in a real system may grossly underperform if the
incoming air gets preheated. Hence, careful analysis and testing under real
system conditions is required to ensure effectiveness of designs [4]. A feasible
solution to these difficulties is to use air impingement, where air is ducted directly
to the heat sink. This helps avoid preheating of air and also prevents it from

escaping via any alternative flow path [4]. However, this approach increases
system costs as it requires separate manifolds and additional mechanical
supports. It also requires overall mechanical design of systems and additional
space within systems in a direction perpendicular to the board [2,4,7].
One of the early high-power dissipation multichip modules, which used an
air impingement type heat sink, was the IBM 4381 system [8,9]. The IBM 4381
system consisted of 22 modules with a maximum of 36 chips per module and
each chip dissipated up to 3.6 W. All the chips in the module were attached to
the substrate by a flip chip (C4) interconnect and were hermetically sealed by a
ceramic cap on which the heat sink was attached. The die was thermally
connected to the chip cap using thermal grease (TIM). Effective external heat
transfer coefficients were reported to be 1000 W/m2K [10]. In a separate study,
Sathe et al. numerically analyzed heat transfer in the IBM 4381 system and
found the results conforming well to experimental data [11]. Another air
impingement type heat sink design with pin fins was studied separately by
McPhee et al., Fitch, and Heng and Pei, where the chips were dissipating up to
30 W each and were back bonded to a copper heat spreader on which the heat
sink was attached [12,13,14]. In yet another study on impingement type heat
sinks, Bartilson analyzed miniature pin fin heat sinks, both numerically and
experimentally [15]. This particular design allowed some space saving in the
system.
Sathe and Sammakia performed numerical studies on a high performance
parallel plate heat sink with air impingement typically used for workstation

applications [16]. The results of the study were verified with experimental data.
The system involved a rectangular jet impinging on a set of parallel fins and then
turning in to cross-flow [16]. The effects of fin thickness, gap nozzle width and fin
shapes on the thermal performance (in terms of heat transfer and pressure drop)
were investigated. It was observed that material could be removed from the dead
air flow zone at the center of the heat sink in order to reduce pressure drop
across the heat sink, without sacrificing overall heat transfer rates. With fin
thickness of around 0.5 mm, channel width of 0.8 mm and appropriate fin shapes
(material removed from the center of the heat sink) high heat transfer coefficients
to the tune of 1500 W/m2K were obtained at pressure drops below 100 Pa [16]. It
was also observed that at lower pressure drops, high heat transfer coefficients
were not achievable for heat sinks operating under a similar set of constraints
[16].
Hansen et al. performed a parametric numerical study to optimize the
design of a pin fin heat sink with a staggered array of fins [17]. The fins had an
elliptical cross section. A numerical solver based on Volume Averaging Theory
was employed to solve the energy, momentum and mass conservation
equations. The numerical solver code was written in FORTRAN [17]. A design of
experiments (DOE) approach was used to do a parametric optimization of the
heat sink and the Reynolds number (Re) turned out to be the factor having the
most impact on thermal performance [17]. However, other factors related to the
fin geometry also had at least a 10% effect on thermal performance which

becomes interesting in a context where only a 10-15% improvement in


performance is required and increasing Re is not an option.
Shah et al. performed numerical analysis on impingement-type parallel
plate heat sinks with fan on top for microprocessors with uniform power
distribution [7]. The study examined effects of different fin shapes on maximum
die temperature and pressure drop across the heat sink. On comparing different
designs based on die temperatures and pressure drops, a stepped fin design
produced optimum results with significant improvements over the original
baseline case.
Gondipalli et al. studied impingement-type parallel plate heat sinks using
numerical analysis [2]. Analysis was performed on designs with different fin
shapes for microprocessors with uniform and non-uniform power distribution. For
the case of non-uniform power distribution the die was divided into a 6x6 matrix
with each section representing unique power dissipation. The heat sink was 32
mm in length and 32 mm wide with a 6 mm base and there were 16 fins, each
0.5 mm thick and 34 mm high. Fin shapes were modified for only the two central
fins. Again a stepped fin design was determined as the optimum among
different designs that were studied.
In Chapter 2 the focus is on impingement type heat sinks which are
smaller in size and have a microprocessor with non-uniform power dissipation.
Effects of different fin heights, fin shapes, air-flow rates and number of modified
fins, are considered on the thermal performance. Some general design
guidelines are developed based on which optimal designs are selected.

1.2 Introduction to thermal characterization study


Thermal characterization studies of electronic packages are important for
determination of cooling requirements and for overall system design. Thermal
models once validated with experimental data can be used to predict thermal
performance for a different set of ambient conditions as well as for different
component power dissipations.
Sathe and Sammakia had performed a thermal characterization study of
an IBM TBGA package and determined cooling requirements at the die level,
card level, and system level [18]. Thermal performance predicted by CFD
analysis was compared with data collected during experimental studies [18]. The
impact of radiation was noted to be significant for natural convection and at low
flow velocities [18].
Burogos et al. performed a component level thermal characterization
study on three leaded surface mount plastic packages with 68, 84, and 164
leads, respectively, under forced convection conditions using a commercially
available CFD code [19]. Initially, a thermal model of the package with 84 leads
was created and simulation results were obtained at a flow velocity of 1.52 m/s
[19]. These results were validated with experimental data and certain modeling
guidelines were established. Consequently, these guidelines were applied to
model the other two components and the models were processed over a range of
flow velocities (0.7-3.05 m/s) [19]. The results were compared with experimental
data to validate the modeling guidelines.

In a separate study, Han performed thermal characterization of a 35 mm x


35 mm TBGA package and studied thermal performance at different die sizes
and heat spreader remaining thickness [20].
Pauley, in his report for Texas Instruments, illustrated methods for
calculating thermal resistance for components and indicated the differences
between military standards (MIL-STD-883) and JEDEC-51 standards for thermal
characterization of packages [21]. An explanation of the inconsistencies in
measurements between manufacturers and package types is also provided.
Pauley also points out that new JEDEC methods have potential for greatly
improving the accuracy and reliability of semiconductor-package thermal
characterization.
The study in Chapter 3 focuses on thermal characterization of a complex
power module (a DC to DC converter) with multiple components.
1.3 Research Objective and Problem Statement
The objective of this study is to examine the use of commercially available
CFD code in design and development of thermal management solutions. In order
to analyze this, numerical optimization of an impingement heat sink is attempted
in Chapter 2, while, in Chapter 3, thermal characterization of a power module is
undertaken.
The main objective in Chapter 2, is to improve thermal performance of an
impingement heat sink for a given semiconductor application by making design
modifications to an existing baseline case. And in Chapter 3, the aim is to

validate the CFD model of the power module with data collected during
experimental analysis.

CHAPTER 2
Numerical Optimization of Impingement Heat Sink
As processing capabilities of microelectronic devices increase, so does
the power dissipated by them. To add to that, power distributions are non uniform
which may create significant hot spots. These devices require improvements in
thermal performance of the entire system due to their higher power dissipation
and non uniform power distribution. This study looks at impingement type heat
sinks which are commonly used for desktop microprocessors and other
applications with the objective of improving thermal performance. One key
feature of the heat sinks studied here, is their smaller size compared to the ones
that are commonly used for desktop applications. This would save real estate on
the board and the system. Detailed parametric studies are conducted covering a
broad range of different air flow rates, shapes and fin heights. The studies are
based on a numerical analysis using a commercially available CFD code.
Conclusions are drawn from the results outlining some general design guidelines.
The objective of the different designs is to reduce chip temperature and the
pressure gradient across the heat sink. The pressure gradient at the center of the
heat sink, near the base tends to be especially high which reduces the airflow
and hence the transport in that region. The study examines different fin shapes
for parallel plate fins with material removed from the central region of the heat
10

sink resulting in improved thermal performance, lower operating temperature and


pressure gradient.
2.1 Model
The model represents a parallel plate heat sink with air impingement,
mounted on the chip by 100-m thick layer of indium, which is used as a TIM
(thermal interface material) between the die and the heat sink. The conjugate
conduction convection problem is solved at steady state using a commercially
available CFD code using a finite volume approach.

Figure 1: Schematic diagram of the baseline case heat sink mounted on the
chip.

11

2.1.1 Fan
The impingement fan is modeled in 2D (two dimensions) with a constant
flow velocity and hence a constant flow rate. The model does not consider a fan
hub as it is modeled only in 2D. The inlet fluid (air) temperature is considered to
be constant at 300K. The ambient air enters through an inlet face located 0.5 mm
above the fins and impinges on to them and it exits through the gap between the
fins on both sides of the sink (Figure 4). The fluid properties are given in Table 1.
The model operates in the laminar flow regime.

(Kg/ m3)

Material

Air

1.225

Cp

(J/ Kg K)

1006.430

(W/ m K)

0.0242

Table 1: Material Properties for Fluid


2.1.2 Heat Sink
The base model is a parallel plate heat sink with 15 fins built completely in
aluminum. The dimensions for the base and fins are given in Table 2, while the
material properties for aluminum used for the purpose of modeling are provided
in Table 3.
Base

Fins

Width (mm)

14.5

0.5

Height (mm)

2.0

8.0-24.0

Length (mm)

14.5

14.5

Number

15

Table 2: Heat Sink Dimensions

12

2.1.3 Chip
Total peak power dissipation from the chip considered for this study is 38
W. The chip is 10 mm in length, 10 mm in width and 0.75 mm thick. It is divided
in to 16 equal sections arranged in a 4 x 4 matrix (Figure 2). The power map is
derived from simulations of an Intel P4 Northwood microprocessor operating at a
clock rate of 2 GHz [22]. Each section has separate power dissipation (Q) which
represents non-uniform power distribution. The heat sink is mounted on the chip
and is thermally connected to it by the TIM. Material properties for the chip
(silicon) and TIM (indium) are given in Table 3.

Material

(Kg/m3)

Cp (J/ Kg K)

k (W/m K)

Silicon

2330

712

149

Indium (TIM)

7310

233

82

Aluminum

2719

871

202

Table 3: Material Properties (Isotropic)

Figure 2: Power distribution in the chip in mW.

13

2.2 Method of Solution


The present heat transfer problem was solved for an incompressible fluid
at steady state flow. The problem domain and boundary conditions of the studied
heat sink model are detailed in Figure 4. The conjugate conduction convection
problem is governed by the Navier Stokes equations of continuity, momentum
and energy for fluids and only the energy equation in solids [16]. The equations,
as stated are given below.

The fluid is assumed to be incompressible with laminar flow and constant


material properties (Table 2). The heat sink, chip and TIM were also assumed to
have constant and isotropic material properties (Table 3). We did not include
buoyancy effects in any of our models as they were relatively small and the flow
was dominated by forced convection.
The Navier Stokes equations were solved at steady state using a
commercially available CFD code, FLUENT, which uses a finite volume approach
[23].
14

The numerical models had to be imported and boundary conditions had to


be set to start the solver. FLUENT uses a pressure based segregated solution
algorithm called SIMPLE (Semi-Implicit Method for the Pressure-Linked
Equation) to solve the Navier Stokes equations in three dimensions [23]. The
solver iterates until convergence criteria are met or a pre specified number of
iterations are complete and then the results can be examined in the post
processing steps. The convergence criterion for continuity residuals was set to
10-4. It was also set to the same value for x, y and z- velocity residuals. And for
the energy residuals it was set to 10-6. The model was assumed to be operating
under standard atmospheric conditions. Hence the operating pressure was
defined as 101325 Pa. Figure 3 below shows a snapshot of the FLUENT
interface.

Figure 3: Snapshot of the FLUENT interface

15

The model was built on a commercially available preprocessor for CFD


applications called GAMBIT. GAMBIT provides a graphical user interface (GUI)
platform which was used for creating the heat sink model, meshing the model
and assigning zones. All volumes were meshed with Hex elements and no
smoothing. The mesh was then exported to and solved in FLUENT.

Figure 4: Heat sink model with boundary conditions

16

2.3 Grid Sensitivity Analysis


A grid sensitivity study was performed for the baseline case to configure
the meshing scheme and validate computational accuracy of different variables.
Figure 4 represents velocity magnitude at a certain height from the base for four
different cases representing the number of nodes between two fins (in the interfin gap).

Figure 5: Grid sensitivity study


The velocity profiles for 9 nodes and 15 nodes were observed to virtually
coincide. This leads to the conclusion that the computational accuracy of field
17

variables is independent of the number of nodes (and correspondingly grid cells)


as long as we have 9 or more nodes. Hence 9 nodes between two fins was
chosen as a standard for all the models as this would lead to fairly accurate
results in reasonable amount of computational time. The meshing scheme of 9
nodes between fins translated to an interval spacing of 0.05 mm along the Z axis
(in the direction of air flowing out of the system). The interval spacing along X
axis was 0.25 mm and along Y axis it was 0.125 mm for all the volumes except
for the TIM. The interval spacing along Y axis for the TIM volume was 0.05 mm.
A snapshot of the GAMBIT user interface shows a heat sink model meshed with
the meshing scheme mentioned above (Figure 6).

Figure 6: GAMBIT snapshot showing the mesh


18

2.4 Results and Discussion


2.4.1 Fin Height Optimization under Non-Uniform Power Distribution
For the purpose of this study the chip was divided in to 16 sections with
individual power dissipation, as mentioned earlier. The maximum power
dissipation from a section was 7892 mW with the mean being, = 2380.688 and
the standard deviation being, = 2657.433. The coefficient of variation in this
case is cv= /= 1.116. This showed that relative to the mean the standard
deviation was very high and represented a highly non-uniform power distribution.
Due to this, one can see significant hot spots on the die. This can be observed
from Figure 7, which shows a temperature contour plot of the chip surface for a
heat sink with 12 mm fin height.
Different heat sink models were tested at different air- flow velocities (flow
rates) and the impact of different flow velocities on maximum die temperature
and pressure drop across the sink were studied. The dimensions of the heat sink
were as mentioned in Table 2. Fin height was gradually increased from 8 mm to
24 mm (in steps of 8, 10, 12, 13,14,16,18 & 24 mm) for the different air-flow rates
and an optimum fin height was observed in each case. The results of this study
are presented in Table 4 where T for different heat sink models is recorded at
different air inlet velocities. Figure 8 shows a plot of T as a function of fin height
for different air flow velocities. Temperature rise (T), is defined as the difference
between maximum die temperature and the ambient temperature. The ambient
temperature for the purpose of this study has been considered to be 300 0K. No
preheating of air was considered and the inlet air temperature was also assumed

19

to be constant. This assumption was made in order to draw comparisons


between different heat sink designs even though in a real system the inlet air
temperature might vary from time to time and there is also a possibility of
preheating of the inlet air.
Figure 8, shows that there is an optimal fin height for each air flow rate at
which T is minimum. It can also be observed that the marginal reduction in die
temperature reduces as air-flow rate increases.
To study the impact of fin height on pressure drop across heat sinks the P
values for different fin heights at different air inlet velocities were recorded, as
shown in Table 5. Pressure drop (P) is defined as the difference in pressure at
the heat sink inlet and the atmospheric pressure at the heat sink outlet. Figure 9
shows a plot of P as a function of fin height for different air inlet velocities and
from the plot one can observe that the marginal increase in P is higher as we
increase the inlet velocity. One can also observe that P reduces up to a certain
fin height and then it essentially becomes constant.

Figure 7: Temperature contour of the chip surface showing significant hotspots (base case of 12 mm fin height).
20

Air inlet velocity


Fin height
(mm)

1.0 m/s

1.5 m/s

2.0 m/s

2.5 m/s

3.0 m/s

3.5 m/s

8
10
12
13
14
16
18
24

223.10
218.77
216.47
218.00
218.46
220.07
221.54
231.98

169.22
164.22
161.05
161.99
162.04
162.87
163.84
171.87

142.03
135.44
131.74
132.11
132.62
132.95
134.18
140.88

125.65
119.06
114.15
114.34
114.13
114.20
115.49
122.10

114.42
107.82
102.94
102.76
101.90
101.74
101.98
108.85

106.23
99.79
95.07
94.63
93.71
92.70
92.44
97.87

Table 4: T (in 0K) at different inlet air velocities for different fin heights

Figure 8: Plot of T as a function of fin height.


21

At this particular fin height the air flowing out of the channels probably
stops feeling the resistance offered by the base of the heat sink. This fin height is
optimal as increasing the height beyond this point has virtually no impact on the
pressure drop. Another interesting observation is that the optimum fin height for a
particular inlet velocity from a P point of view seems to coincide with the
optimum fin height from a T point of view. In other words, there exists a single
optimum fin height for a particular air inlet velocity. Based on these
considerations a base model for further optimization was chosen, which exhibits
a moderate pressure drop and the lowest T for that pressure drop. We
considered an air inlet velocity of 2 m/s, and for that velocity a fin height of 12mm
turned out to be the optimum with a T of 131.74 0K and a P of 26.82 Pascal.

Air inlet velocity


Fin Ht.
(mm)

1.0 m/s

1.5 m/s

2.0 m/s

2.5 m/s

3.0 m/s

3.5 m/s

11.13

20.54

32.50

47.01

64.08

83.69

10

10.19

18.39

28.67

41.05

55.55

72.18

12

9.78

17.40

26.82

38.08

51.21

66.23

13

9.68

17.13

26.29

37.20

49.89

64.40

14

9.60

16.94

25.91

36.56

48.93

63.06

16

9.52

16.73

25.47

35.79

47.73

61.33

18

9.49

16.63

25.26

35.40

47.09

60.39

24

9.46

16.55

25.08

35.08

46.53

59.49

Table 5: P (in Pascal) at different inlet air velocities for different fin
heights

22

Figure 9: Plot of P as a function of fin height.


2.4.2 Fin Shape Optimization under Non-Uniform Power Distribution
In a conventional parallel plate heat sink (like the base models) at the
central region between two fins, near the base of the sink one can observe a
zone where there is virtually no transport and a high pressure drop (Figure 10left). This zone has been referred to as the "dead" zone in literature [2,6]. The
lack of transport in this zone renders the enhanced surface area for heat transfer
in this region completely useless and increases the overall P across the heat
sink. In order to improve transport in the dead zone we modified fin shapes by
removing material from the central zone and the effect can be observed in Figure
10 (right) where we see improved air flow and considerable reduction in P
which was expected. We also observed an improvement in thermal performance
in terms of reduction in T in some of these designs.
23

Figure 10: Velocity contour comparison between base model (left) and
design "h-44" (right) for 16 mm fin height at inlet velocity of 2 m/s.
Ten different designs (as shown in Figure 11) were numerically studied
with shape modifications being made to varying number of fins. Figure 12 shows
an isometric view of one such design without one end fin. Fifty one different
design configurations were tested for thermal performance based on the two
factors (P & T) at an air inlet velocity of 2 m/s, and the results are recorded in
Table 6. The different designs were categorized (from a to j) such that no
more than one characteristic dimension in a particular category is different from
the other designs in that category. The designs were also segregated in to three
groups. Group-I consisted of different heat sink models with a maximum fin
height of 12 mm, Group-II consisted of models with a maximum fin height of 16
24

mm and Group-III of 18 mm. It should be noted that the models "d-33" & "d-34" in
Group II are slightly different from the rest, because out of the 11 fins cut, cuts on
the 5 central fins were moved +0.5 mm and -0.5 mm respectively in "x" direction.
Also model "d-35" was unique as it had one channel on each end blocked to
prevent any airflow through those channels.
2.4.3 Observations
A plot of T values against P values for the different models is shown
(Figure 13). From Figure 12 we made certain observations, which are listed
below.
Observation 1: Making through and through cuts on the fins is not
appropriate as it increases the temperature dramatically, even though the
pressure reduction is considerable. This happens because cutting all
fifteen fins creates a path of least resistance and allows the air to escape
out through this path without encountering most of the enhanced surface
area.
Observation 2: Design "c", where we completely remove the central fin is
not optimal as it results in comparatively higher T.
Observation 3: Making cuts too close to the fin edge leads to
comparatively higher T.
Observation 4: Blocking channels (in our case design "d-35" had one
channel on each edge blocked) considerably increases P.
Observation 5: Heat sink models with modified fin shapes which have a
maximum fin height equal to the current optimum fin height (Group I)

25

exhibit a relatively higher T and P compared to models which have a


slightly higher maximum fin height (Group II or III). In other words one
observes greater improvement in thermal performance by making cuts on
a heat sink with fin height slightly greater than the optimum fin height.
Observation 6: In the current study all (except for design a-20) designs in
Group II & III with 11 or 13 fins cut were observed to exhibit comparatively
lower T as well as low P. Hence these heat sink models can be
considered optimal in comparison to the base case.
Based on these observations all heat sink models from Group II
which showed a relatively low P and T compared to the base case and other
models were selected and consequently analyzed numerically at an air inlet
velocity of 2.5 m/s. The reason for selecting Group II models while leaving out
Group III models, in spite of displaying a similar drop in P and T, is the lower
fin height of Group II models. Lower fin height would be an obvious advantage in
presence of space constraints. The results from this analysis are presented in
Table 7. These results were plotted (T vs. P) along with the results previously
obtained for these models at an inlet velocity of 2m/s and also the base case at
2m/s inlet velocity with 12 mm fin height (Figure 14). From Figure 14 it was
observed that P at 2.5 m/s inlet velocity for all the heat sink models was higher
compared to P at 2 m/s, while T at 2.5 m/s was comparatively low, which was
expected. It was also observed that some models exhibited a lower P
compared to the base case and at the same time, comparatively, had a very low
T. All these models were an improvement to the base case and hence are

26

indicated as optimal designs in Figure 14. In reality however these designs are
probably only suboptimal, although they represent a significant improvement on
the base case. Among these optimal designs, h-43 had the lowest temperature
rise while h-48 had the lowest pressure drop. Model h-43 with its modified fin
shapes showed a 17.40% decrease in T and model h-48 showed 16.55%
reduction in P as compared to the base case.

Figure 11: Different heat sink models that were studied (characteristic
dimensions shown for design d also corresponds to all designs from e
to j).
27

Figure 12: Isometric view of design h-43 with one end fin removed, showing
the modified fin shapes.

28

Design
No. of
Dimensions
&
P (in Pa) T (in K) Fins
(All dimensions in mm)
Sl No.
cut
Group I (Fin height 12 mm )
a
1
26.57
132.03
1 x=1
2
24.45
131.35
5 x= 2.5
3
20.80
130.33
13 x= 2.5
b
4
23.39
137.48
5 x1= 1.25 ; x2= 5
5
24.56
133.76
5 x1= 1.25 ; x2= 3
6
24.59
132.67
5 x1= 1.25 ; x2= 2
7
24.61
132.15
5 x1= 1.25 ; x2= 1.5
8
25.44
131.69
3 x1= 1.25 ; x2= 1
c
9
23.54
142.58
1 Entire central fin removed
x3=4.5 ,y3=1.25; x2=2,y2=5;
i
10
22.16
128.96
11 x1=1,y1=5.75
x3=4.5 ,y3=1.25; x2=2,y2=5;
11
21.23
128.84
13 x1=1,y1=5.75
Group II (Fin height 16 mm)
a
12
25.19
132.23
1 x= 1
13
22.69
129.46
5 x= 2.5
14
22.26
129.62
5 x= 3
15
23.16
129.49
5 x= 2
16
17.39
127.22
13 x= 3
17
15.74
129.08
13 x= 4
18
18.38
126.72
13 x= 2.5
19
22.20
127.68
13 x= 1
b
20
22.85
130.61
5 x1= 1.25 ; x2= 2
21
23.81
131.40
3 x1= 1.25 ; x2= 2
22
22.85
130.52
5 x1= 1.25 ; x2= 1.5
23
22.86
130.12
5 x1= 1.25 ; x2= 1
c
24
22.24
143.06
1 Entire central fin removed
x3=4.5,y3=1.75; x2=2,y2=6.75;
d
25
24.80
132.04
1 x1=1,y1=7.5
x3=4.5,y3=1.75; x2=2,y2=6.75;
26
23.80
130.25
3 x1=1,y1=7.5
x3=4.5,y3=1.75; x2=2,y2=6.75;
27
22.84
129.18
5 x1=1,y1=7.5
x3=4.5,y3=1.75; x2=2,y2=6.75;
28
21.88
128.34
7 x1=1,y1=7.5
x3=4.5,y3=1.75; x2=2,y2=6.75;
29
20.89
127.93
9 x1=1,y1=7.5
x3=4.5,y3=1.75; x2=2,y2=6.75;
30
19.85
125.95
11 x1=1,y1=7.5
29

31

18.75

126.11

13

32

11.98

167.71

15

33

19.85

127.06

11

34

19.85

127.26

11

35

31.01

130.05

11

36

22.80

129.18

37

18.65

126.20

13

38

12.04

166.36

15

39

19.47

126.74

11

40

18.30

125.93

13

41

17.59

125.53

13

42

17.33

125.34

13

43

16.18

124.97

13

44

16.73

126.42

13

45

16.44

125.52

13

46

15.95

125.13

13

47

15.75

125.80

13

48
15.59
126.83
Group III (Fin height 18 mm)
a
49
21.88
129.24

13

50

15.40

125.06

51

15.18

125.05

x3=4.5,y3=1.75; x2=2,y2=6.75;
x1=1,y1=7.5
x3=4.5,y3=1.75; x2=2,y2=6.75;
x1=1,y1=7.5
x3=4.5,y3=1.75; x2=2,y2=6.75;
x1=1,y1=7.5
x3=4.5,y3=1.75; x2=2,y2=6.75;
x1=1,y1=7.5
x3=4.5,y3=1.75; x2=2,y2=6.75;
x1=1,y1=7.5
x3=4.5,y3=1.75; x2=2,y2=7.5;
x1=1,y1=6.75
x3=4.5,y3=1.75; x2=2,y2=7.5;
x1=1,y1=6.75
x3=4.5,y3=1.75; x2=2,y2=7.5;
x1=1,y1=6.75
x3=5.5 ,y3=1.75; x2=2,y2=6.75;
x1=1,y1=7.5
x3=5.5 ,y3=1.75; x2=2,y2=6.75;
x1=1,y1=7.5
x3=5.5 ,y3=1.75; x2=2.5,y2=6.75;
x1=1,y1=7.5
x3=5.5 ,y3=1.75; x2=2.5,y2=6.75;
x1=1.5,y1=7.5
x3=5.5 ,y3=1.75; x2=3.5,y2=6.75;
x1=1.5,y1=7.5
x3=5.5 ,y3=1.75; x2=3.5,y2=6.75;
x1=0.5,y1=7.5
x3=5.5 ,y3=1.75; x2=3.5,y2=6.75;
x1=1.0,y1=7.5
x3=5.5 ,y3=1.75; x2=3.5,y2=6.75;
x1=2.0,y1=7.5
x3=5.5 ,y3=1.75; x2=3.5,y2=6.75;
x1=2.5,y1=7.5
x3=5.5 ,y3=1.75; x2=3.5,y2=6.75;
x1=3.0,y1=7.5

13 x= 1
x3=5.5 ,y3=2; x2=3.5,y2=7.5;
13 x1=1.5,y1=8.5
x3=5.5 ,y3=2; x2=3.5,y2=7.5;
13 x1=2,y1=8.5

Table 6: Different design configurations at 2m/s inlet velocity


30

Figure 13: T vs. P for different designs at 2m/s inlet velocity.


31

Design
P (in
T (in
No. of
Dimensions
&
Pa)
K)
Fins cut
(All dimensions in mm)
Sl No.
Group II (Fin height 16 mm)
a
16
24.89
111.23
13 x= 3
17
22.71
114.06
13 x= 4
18
26.20
110.40
13 x= 2.5
19
31.30
110.07
13 x= 1
x3=4.5,z3=1.75; x2=2, z2=6.75;
d
30
28.18
109.02
11 x1=1,z1=7.5
x3=4.5,z3=1.75; x2=2, z2=6.75;
31
26.65
109.49
13 x1=1,z1=7.5
x3=5.5 ,z3=1.75; x2=2, z2=6.75;
f
39
27.65
109.69
11 x1=1,z1=7.5
x3=5.5 ,z3=1.75; x2=2, z2=6.75;
40
26.02
109.11
13 x1=1,z1=7.5
x3=5.5 ,z3=1.75;x2=2.5, z2=6.75;
g
41
25.08
108.87
13 x1=1,z1=7.5
x3=5.5 ,z3=1.75;x2=2.5, z2=6.75;
42
24.71
108.96
13 x1=1.5,z1=7.5
x3=5.5 ,z3=1.75;x2=3.5, z2=6.75;
h
43
23.17
108.82
13 x1=1.5,z1=7.5
x3=5.5 ,z3=1.75;x2=3.5, z2=6.75;
44
23.92
109.96
13 x1=0.5,z1=7.5
x3=5.5 ,z3=1.75;x2=3.5, z2=6.75;
45
23.53
109.13
13 x1=1.0,z1=7.5
x3=5.5 ,z3=1.75;x2=3.5, z2=6.75;
46
22.86
109.20
13 x1=2.0,z1=7.5
x3=5.5 ,z3=1.75;x2=3.5, z2=6.75;
47
22.60
110.02
13 x1=2.5,z1=7.5
x3=5.5 ,z3=1.75;x2=3.5, z2=6.75;
48
22.38
111.18
13 x1=3.0,z1=7.5
Table 7: Selected designs from Group II at 2.5 m/s inlet velocity
2.5 Design Guidelines
The observations made in the course of this study along with the obtained results
facilitated a general understanding of the principles governing heat transfer within
systems using impingement type parallel plate heat sinks. Based on this
32

understanding certain general design guidelines were formulated which are given
below:

Based on P specifications one may choose an appropriate air inlet


velocity.

Based on the inlet air flow velocity one can find out the optimum fin height
for that velocity.

If the corresponding T meets or exceeds specifications then there is no


need for further optimization. If not, then one can try to alter fin shapes to
further reduce T.

If increasing fin height is an option then one should choose a fin height
slightly greater than the current optimum and modify fin shape as shown in
Figure 11.

To determine the number of fins to be modified one may incrementally


increase the number of fins cut and observe the impact on T.

Locating the cuts centrally also seems to be a good idea as it helps to


prevent air escaping out through the sides without interacting with the
enhanced surface area in the middle, rendering it redundant.

Modifying all the fins (making a through and through cut) is not advisable
as this provides a path of least resistance to the air and it escapes, thus
rendering most of the fin surface redundant.

Design c, which involves removing the entire central fin, is also not
optimum as this considerably reduces available surface area for heat
transfer while providing an escape route for the air.
33

Figure 14: Design II heat sink models at an inlet velocity of 2.5 m/s.

If fin shape modifications considerably reduce P, one may increase the inlet
air velocity which would significantly reduce T. One can then select the
designs which exhibit a P lower than the base case (or specifications).

34

CHAPTER 3
Thermal Characterization of a Power Module
Improved performance of semiconductor devices in recent years has
resulted

in

consequent

increase

in

power dissipation.

Hence

thermal

characterization of components becomes important from an overall thermal


design perspective of the system. This study looks at a high performance nonisolated point of load power module (a DC to DC converter) meant for advanced
computing and server applications. Thermal characteristics of the module were
experimentally analyzed by placing the power module on a bare test board inside
a wind tunnel with thermocouples attached to it. There were three devices on
this module that dissipate power. There were two FETs (Field Effect Transistors)
and an inductor which can be considered as sources. The consolidated power
dissipation from the module was calculated by measuring the input voltage and
input current while keeping the output voltage and current constant.
Temperatures at various points on the module and the test card were recorded
for different air flow velocities and overall power dissipation. Consequently, this
set up was numerically analyzed using a commercially available finite element
analysis code (using a finite volume approach) with the objective of comparing
results with experimental data previously obtained.

35

3.1 Experimental Set Up


The experimental set up shown in Figure 15 consisted of the power
module on a bare test board (with no insulation) clamped to the support stand
placed inside a wind tunnel. Four thermocouples were attached to different key
points on the module and one was attached to the test board as well to monitor
temperatures for different air flow rates and power dissipations. The wiring was
done such that it did not significantly interfere with normal air flow either over or
under the test board [24].

Figure 15: Schematics of experimental setup

36

The airflow velocity was constantly monitored inside the wind tunnel using
a thermal anemometer placed upstream from the module and the test card and
regulated by controlling the rotational velocity of the wind tunnel fan using a VFD
(Variable Frequency Drive). There was an ambient air temperature probe with a
thermocouple which monitored the ambient temperature throughout the
experiment. Data was recorded by a data acquisition system (on a PC)
connected to a data logger.
There were three devices on this module that dissipated power. There
were two FETs and an inductor which could be considered as sources. The
consolidated power dissipation (heating power) for the module was measured as
a product of heating current and heating voltage [25]. The consolidated power
dissipation from the module was calculated by measuring the input voltage and
input current while the output voltage stayed constant at 3 V and the output
current stayed constant at 5.5 A. The output currents were measured across two
resistances connected in series (0.4 Ohm and 0.1 Ohm). The effective resistance
including the resistance from the wires came out to be 0.55 Ohms.

Figure 16: Thermocouple positions on the module and two opposite


directions of flow (a & b)
37

Five thermocouples were attached on the module and the board. Their
positions are as shown in Figure 16 above. Thermocouples 1&2 were placed on
FET 1&2 respectively, while the third thermocouple was placed on the inductor.
The fourth thermocouple was placed on the surface of the module substrate
between the two FETs. The fifth thermocouple (not shown in the figure) was
placed at a distance from the module on the test board to capture the
temperature on the evaluation board.
Since the module was not placed at the center of the test board, the
orientation of the board played a part in the observed temperature readings. The
temperature readings for air flow direction as shown in Figure 16 (a) were
considerably lower than the air flow direction shown in Figure 16 (b). This can be
observed from the temperature readings for the 1.43 W power dissipation for the
two orientations in Table 8.

The rest of the readings (at different power

dissipation values and air flow velocities) were all recoded for an airflow direction
as shown in Figure 2 (b) as we would want to test for the worst possible field
conditions.
3.1.1 Test Environment
These tests were conducted in an open circuit in-draft type wind tunnel
which complied with JEDEC standards for low velocity applications (<10 m/s)
[26]. The following figure (Figure 17) shows a schematic of the wind tunnel. It
also had a honeycomb structure at the inlet to reduce lateral velocity differences
and screens which promote uniform axial velocity due to their high pressure drop
along the direction of flow [25]. These elements produced an overall flow regime

38

with desired flow characteristics namely a flat velocity profile and reduced
turbulence intensity.
The contraction section of the wind tunnel allowed for acceleration of flow
without affecting the flow quality. The test was conducted in the test section while
the transition section conveyed the flow from the test section and in to the fan
inlet.
The test section was 12"x12" in cross section and was 20" in the direction
of flow. Within the test section minimum clearance was maintained with respect
to the test card as required by JEDEC standards.

Figure 17: Schematics of an open circuit wind tunnel [26]


39

3.1.2 Test Card


The test board consisted of two outer (signal) copper planes and two inner
(power) copper planes. Both inner planes were 35 mills in thickness while the
outer planes were 18 mills thick. Each copper layer had approximately 90%
coverage. The board was 101.6 mm x 101.6 mm and the board (dielectric)
material was FR4. The overall thickness of the card was 1.6 mm.
3.1.3 Air Velocity Transducer
The inlet air flow velocity was measured using a temperature
compensated and calibrated thermal anemometer (air velocity transducer). The
anemometer reported a mass flow velocity equivalent to air at standard air
conditions as per JEDEC standards. In other words, the velocity of air reported
by the anemometer was as if air was at standard conditions with the air density
being 1.2 Kg/m3 [26]. The standard air density corresponds to dry air at 101.325
kPa (760 mmHg) at 21 C or moist air at 50% humidity at the same pressure and
20 C [26]. The anemometer was placed upstream from the test card at a
location where it did not directly interfere with the air flow in to the test card and
hence did not cause any turbulence in the air reaching the device.

Minimum

accuracy of the anemometer was as per JEDEC specifications of +/- 4% of


reading and +/- 0.05 m/s [26].
3.1.4 Ambient Temperature Thermocouple
The room in which the wind tunnel was housed had temperature and
humidity control which provided a uniform ambient temperature. Sufficient
clearance was provided at the inlet and exit of the tunnel to prevent any

40

turbulence in air flow within the wind tunnel. The ambient temperature was
measured using a thermocouple located upstream and below the plane of the
test board and was placed at a sufficient distance from the walls of the wind
tunnel, so that it does not interfere with the boundary layers on the walls. It was
supported by a structure from the nearest wall.

The ambient temperature

remained fairly steady throughout the course of the experiments (between 19.0
C and 19.6 C) and allowed for an adequate steady state thermal performance
measurement. This can be observed from the results recorded in Table 8 below.
Q
(W)

V
(m/s)
3.0
2.0
1.43 (Fig
1.5
2-a)
1.0
0.5
3.0
2.0
1.43
1.5
1.0
0.5
3.0
2.0
1.80
1.5
1.0
0.5
3.0
2.0
1.05
1.5
1.0
0.5

T1 (0C)
30.7
33.0
35.0
37.0
39.8
33.4
36.7
38.0
39.3
41.6
38.9
42.9
44.6
46.1
48.9
29.8
31.9
32.8
33.8
35.3

T2 (0C)
31.0
33.3
35.3
37.1
39.8
32.1
35.3
36.9
38.4
40.3
36.6
40.4
42.5
44.3
46.7
28.9
30.9
31.9
33.1
34.3

T3 (0C)
37.4
39.8
42.1
44.0
46.6
38.1
41.5
43.0
44.5
46.8
45.6
49.9
51.9
53.8
56.6
31.4
33.5
34.5
35.6
37.0

T4 (0C)
29.6
32.3
34.7
37.1
40.2
32.1
35.6
37.4
39.1
41.2
36.9
41.1
43.4
45.5
48.1
28.7
30.9
32.1
33.4
34.8

T5 (0C)
20.8
21.4
22.1
22.8
23.9
19.9
20.7
21.2
21.8
23.0
20.6
21.3
22.0
22.7
24.1
20.0
20.3
20.6
21.1
31.9

Amb (0C)
19.3
19.4
19.6
19.6
19.5
19.0
19.2
19.2
19.2
19.4
19.4
19.3
19.5
19.4
19.5
19.2
19.3
19.3
19.3
19.3

Table 8: Experimental results (steady state temperature readings from


various thermocouples)

41

3.2 Model
A cross section of the actual test card (Figure 18) shows an insertion
mount module placed on the test card. There were three components on the
module that dissipated power; an inductor and two FETs. All three components
were surface mount on the module substrate. From the cross section it can also
be observed that both the module substrate and the test card substrate had two
outer planes and two internal planes each. There were other surface mount
components on the underside of the module substrate as well. However these
components werent considered for our numerical CFD analysis.

Figure 18: Cross section of the module and the test card
Our model below (Figure 19) represents a module card with an inductor
and two FETs mounted on a test board. Each of the three components on the
module had a separate thermal flux (Q) and heat transfer from these components
was primarily governed by conduction and convection. However the radiation
42

effects were important as well and hence couldnt be ignored. The overall heat
transfer problem was solved numerically with a commercially available CFD code
and steady state temperatures at various points were analyzed.
Instead of modeling the entire wind tunnel a much smaller air enclosure
(volume) which impacted heat transfer around the test card was considered in
the CFD model. This small enclosure was further divided in to six separate but
interconnected volumes to reduce computational complexity. Additionally the air
gap between the inductor and the module card as well as the air gap between
the module card and test card were modeled as separate volumes connected to
the other volumes of the enclosure and formed a part of the overall enclosure.
This was done to accurately model the thermal resistance that occurs due to lack
of transport in these regions. Similarly, there were other special considerations
for modeling the test card, the module card, the inductor, the two FETs, the
copper columns and the solder bumps that were taken into account.

Figure 19: Schematic of the test card and module


43

3.2.1 Test Card


The test board was modeled as three separate interconnected volumes for
the purpose of making it computationally less cumbersome. The interconnected
volumes took the shape of the card, i.e. they formed a rectangle. The overall
dimensions of the card as considered in the model are described in Table 9. The
copper and epoxy layers werent modeled separately and overall material
properties for the entire card were used instead. Additionally there were some
vias that werent considered in the model. Heat transfer from the underside of the
test card to ambient air was not modeled in order to make the model less
cumbersome. Instead the surface was assumed to be a flat plate and an overall
heat transfer co-efficient was calculated and incorporated in the model. Free
stream temperature of ambient air was also specified in order to evaluate steady
state temperatures.
3.2.2 Module Card
Similar to the test board, the module card was modeled as four separate
interconnected volumes for computational simplification. The overall dimensions
for the interconnected volumes are described in Table 9. As in the case of the
test board, overall material properties were used instead of modeling copper and
epoxy layers separately. Module substrate and test board interconnect was
achieved by through-hole copper columns and this was incorporated in the model
as well. The two FETs and the inductor were also placed on the module card at
their actual locations as on the original test specimen. The components on the
module card substrate are shown in Figure 19.

44

3.2.3 Inductor
The actual inductor on this module had a copper coil wound on a ferrite
core. Figure 20 shows a cross sectional view of the inductor. The dimensions for
the inductor are recorded in Table 9. The inductor was a leaded component with
three leads and had surface mount soldered interconnect with the module card. It
was modeled as a single solid block with uniform material properties of the ferrite
core. Since heat was generated at the center of the core and the ferrite offered
the path of least resistance for heat conduction to the top and bottom surfaces of
the inductor this was suitable approximation.

Figure 20: Cross sectional view of the inductor

45

3.2.4 FET 1&2


There were two FETs on the module as can be seen from Figure 19.
Figure 21 below shows a cross sectional view of one of the FETs. It has a silicon
die on top of a copper base with thermal interface material between the two. The
copper base is soldered to the pads on the module and the die is covered with
over-mould compound. However in our model the FETs are represented as solid
blocks with equivalent material properties for the sake of computational
simplicity. Dimensions for the FETs are given in Table 9.

Figure 21: Cross sectional view of FET-1

46

3.2.5 Copper Column


The module card was attached to the test board with 11 copper columns
using through hole interconnect. The module card was wave soldered to the test
board. The copper columns due to their high thermal conductivity enhanced heat
transfer between the module substrate and the test board. Each copper column
was modeled as two separate interconnected volumes one on top of the other
forming a through hole interconnect between the module card and the test board.
While the volume on top was buried into the module substrate the bottom
structure was buried into the test board. Overall dimensions for the copper
columns are given in Table 9.
3.2.6 Solder Bumps
The inductor was surface mount on the module substrate with three solder
bumps and these were represented in the model (Figure 19). The dimensions for
each solder bump are given in Table 9.
Components

Dimensions (in mm)


Length

Width

Height

Test Card

100.00

100.00

1.60

Module Card

22.00

19.00

0.80

Inductor

10.25

7.00

4.80

FET 1

5.50

5.00

0.80

FET 2

5.00

5.50

0.80

Copper Columns

1.50

1.50

4.80

Solder Bumps

1.75

1.50

0.80

Table 9: Dimensions for various components


47

3.2.7 Position and Placement


The placement of the module on the test card in the model is based on
measurements made on the actual test card. Figure 22 below shows the position
of the farthest point on the module card with reference to the coordinate system
for the test card and direction of air flow.

Figure 22: Position of module on the test card

Similarly the position of different components on the module card was also
measured with respect to the closest point on the module card to the origin of our
chosen coordinate system and the direction of air flow. The positions of different
components with respect to this new coordinate system are shown in Figure 23.

48

Figure 23: Position of components on the module card (all dimensions in


mm)
3.3 Material Properties
The fluid (air) entering the wind tunnel was at different inlet velocities was
at an ambient temperature of around 292.5 oK. The maximum temperature
reached by the air was approximately an additional 20-25 oK. The fluid properties
(except for density) within that range of temperature were assumed to be
constant for the purpose of this study and are given in Table 10 below.
Material

(Kg/ m3)

Cp(J/ Kg K)

k(W/ m K)

Air

1.2074
(Boussinesq)

1006.4300

0.0242

Table 10: Fluid (air) properties


49

The air density was determined by the Boussinesq approximation and


hence was temperature dependent. The value given in Table 10 is the operating
density value.
For the sake of computational simplification effective material properties
were used for the inductor and FETs in the model. Their material properties
along with material properties for the solder bumps and copper columns and FR4
are given in Table 11 below.

Material

(Kg/ m3)

Cp(J/ Kg K)

k(W/ m K)

Inductor

4649.1

548.9

10.0

FET (1&2)

3383.6

826.4

40.8

Copper

8960.0

390.0

401.0

Solder

7400.0

230.0

63.2

FR4

1850.0

60.0

0.3

Table 11: Material properties

These material properties were considered to be constant and isotropic


over the range of operating temperatures for the purpose of this study.
For the test board as well as the module card we have used effective
material properties in our model. While we considered isotropic and constant
density and specific heat, we used anisotropic thermal conductivity for both.
Table 12 shows the density and specific heat values and Table 13 shows the in
plane and out of plane thermal conductivity values for both.

50

Material
Test Card

(Kg/ m3)
2785.0

Cp(J/ Kg K)
1098.0

Module Card

2658.0

1011.0

Table 12: Density and specific heat for the test and module cards

The in plane and out of plane thermal conductivities were calculated using
the following formulae:

kinplane =

L1k1 fo + L2 k 2 + L3 k3 fi + L4 k 4 + L5 k5 fi + L6 k 6 + L7 k 7 fo
L1 + L2 + L3 + L4 + L5 + L6 + L7

k out of plane =

L1 + L2 + L3 + L4 + L5 + L6 + L7
L
L
L
L
L
L
L
( 1 + 2 + 3 + 4 + 5 + 6 + 7)
k1 k 2
k3
k4
k5
k6
k7

Layers L1 to L7 correspond to planes in a PCB as shown in Figure 24


below. In the formulae L is the thickness of a particular plane while k is its
thermal conductivity. The coefficients fo and fi are fraction coverage of outer and
inner planes respectively.

Figure 24: Signal and power planes in a PCB


51

Material

k(W/ m K)
In Plane

Out of plane

Test Card

60.95

0.36

Module Card

47.20

0.39

Table 13: Anisotropic thermal conductivity for test and module cards

Since the underside of the test board was not modeled, an overall heat
transfer co-efficient was required to be specified. The heat transfer coefficient
was calculated using the following correlation:

Here, k f is the thermal conductivity of the fluid (air), L is the characteristic


dimension which is the length of the board in the direction of air flow in our case
and Nu is the Nusselt number. Since Nu depends on the air stream velocity, the
heat transfer co-efficient varied as velocity was changed. Table 14 below
provides the h values at different inlet velocities.
Velocity (m/s)

h (W/m2 K)

0.5

4.43

1.0

6.27

1.5

7.68

2.0

8.87

3.0

10.86

Table 14: Heat transfer coefficients at various air inlet velocities

52

3.4 Method of Solution


Overall heat transfer from the inductor and the two FETs was controlled
primarily by conduction and convection; however radiation effects were
substantial as well. The heat transfer problem was solved for an incompressible
fluid with a laminar flow regime. Heat transfer from the inductor and the two FETs
was analyzed and steady state temperatures at several key points were
recorded. Problem domain and boundary conditions for the model are described
in Figure 25.The overall heat transfer problem was governed by Navier Stokes
equations of continuity, momentum and energy [18]. The equations are given
below.

The fluid was assumed to have constant thermal conductivity. Other


material properties were also assumed to be constant within the range of
operating temperatures. Properties for the fluid and other materials are given in

53

Tables 10-13. Although buoyancy effects were negligible at higher velocities and
heat transfer was dominated by forced convection they were substantial at lower
velocities and hence were included in the simulation. The Boussinesq model was
employed to capture buoyancy effects.
The Navier Stokes equations were numerically solved for incompressible
flow and steady state thermal properties were analyzed using the commercially
available CFD code called FLUENT [23].
The model itself was created using the commercially available preprocessor for FLUENT called GAMBIT. GAMBIT was used to create the
geometry as well as the discretization scheme. The geometry was meshed into
finite number of control volumes, following which different volumes were
assigned into solid and fluid zones.
The numerical model had to be imported into FLUENT and boundary and
other cell zone conditions had to be set up before initializing the solver and
running a simulation. A pressure based segregated solution algorithm called
SIMPLEC was used to enforce mass conservation and iteratively solve the
Navier Stokes equations in three dimensions. The iterations continue until
convergence criteria are met or a certain pre defined number of iterations are
completed. Results of the simulation run can then be examined in the post
processing steps [23]. The SIMPLEC (SIMPLE-Consistent) algorithm is a variant
of the SIMPLE algorithm used for the heat sink optimization project (in Chapter 2)
and is useful in achieving convergence in less number of iterations due to
increased under-relaxation that can be applied [23].

54

Figure 25: CFD model with boundary conditions


3.5 Results
3.5.1 Results and Discussion
Since the actual power distribution within the module was unknown, the
total power was equally distributed within the three components (two FETs and
the inductor). The power distribution ratio was determined by trial and error and
55

the simulated results seemed to best fit the experimental data with this ratio.
After completing a simulation the models were analyzed using FLUENTs post
processing features. Figure 26 below shows a velocity contour plot on a plane
cutting across the inductor and FET 2 at an inlet air velocity of 0.5 m/s and a total
power of 1.05 W. Formation of boundary layers on the test card surface, the
module card surface, the inductor and FET 2 can be observed from the figure. It
can also be observed that there is no significant transport under the module card
and the inductor, especially at low air inlet velocities. This indicates that heat
transfer from the inductor to the module card and from the module card to the
test board is controlled primarily by conduction through the solder bumps and
copper columns respectively (especially at low velocities). Hence modeling the
inductor-module card interconnect and the module card-test board interconnect
accurately was important in order to get good agreement of simulation results
with experimental data.

Figure 26: Velocity contour plot 1.05 W and air inlet velocity of 0.5 m/s
56

Fifteen separate cases were considered in the numerical study, each


corresponding to an experimental case for which data was collected previously.
Temperature values obtained from the simulation study at specific points on the
module, the two FETs and the inductor were recorded. Table 15 compares
experimental and simulated T values for FET 1&2 for three separate power
dissipation values and at different air inlet velocities while Table 16 shows the
comparison of experimental and simulated T values for the inductor and the
module card for same power dissipations and air inlet velocities. Temperature
rise (T) is defined as the difference between surface temperature of a particular
component at the specific point of measurement and the ambient temperature.
Figures 27 and 28 show temperature contours on the surface of the
module card along with temperature contours on top of the two FETs and the
inductor at air inlet velocity of 0.5 m/s for power dissipation of 1.8 W and 1.05 W,
respectively. A comparison of the two cases show that the simulation results
closely matched experimental temperature values for the two FETs and the
inductor at higher power dissipation and agreement for the point on the surface
of the module card was not as good; while at lower power dissipation there was
good agreement of simulated and experimental values for the FETs and the
module card surface and not so good agreement in case of the inductor.

57

Figure 27: Temperature contours at 1.83 W and 0.5 m/s air inlet velocity

Figure 28: Temperature contours at 1.05 W and 0.5 m/s air inlet velocity
58

Power

1.8W

1.43W

1.05W

Inlet Velocity
(m/s)

FET 1
T (in K)

3.0
2.0
1.5
1.0
0.5
3.0
2.0
1.5
1.0
0.5
3.0
2.0
1.5
1.0
0.5

Exp
19.58
23.62
25.10
26.73
29.46
14.42
17.49
18.82
20.13
22.25
10.60
12.60
13.50
14.50
16.00

FET 2
T (in K)
Sim
21.15
22.93
24.05
25.83
28.89
16.84
18.28
19.25
20.77
23.06
12.38
13.50
14.31
15.42
17.26

Exp
17.27
21.14
22.95
24.93
27.17
13.06
16.02
17.66
19.22
20.99
9.70
11.60
12.60
13.80
15.00

Sim
19.51
21.25
22.46
24.29
27.51
15.54
16.93
18.00
19.53
21.96
11.45
12.53
13.35
14.49
16.46

Table 15: Experimental and simulated T values for FET 1&2


Power
1.8W

1.43W

1.05W

Inlet Velocity
(m/s)
3.0
2.0
1.5
1.0
0.5
3.0
2.0
1.5
1.0
0.5
3.0
2.0
1.5
1.0
0.5

INDUCTOR
T (in K)
Exp
Sim
26.25
26.03
30.62
29.13
32.43
31.60
34.37
34.30
37.13
38.50
19.10
20.26
22.28
23.32
23.75
25.39
25.31
27.60
27.44
30.80
12.20
15.20
14.20
17.28
15.20
18.69
16.30
20.35
17.70
22.81

MODULE
T (in K)
Exp
17.56
21.86
23.92
26.12
28.61
13.07
16.40
18.19
19.92
21.84
9.50
11.60
12.80
14.10
15.50

Sim
13.48
16.00
17.00
19.00
22.00
10.77
12.35
13.55
15.20
17.39
8.00
9.26
10.01
11.29
13.09

Table 16: Experimental and simulated T values for the inductor and
module card
59

3.5.2 Statistical Comparison of Experimental and Simulated Results


The experimental and simulated results for each specific combination of
input power and air flow velocity can be considered as paired data as they
represent thermal performance of the power module under the same set of
conditions. Even though it is understood that all physical conditions and material
properties are not accurately represented within the CFD model, analyzing the
results as paired data allows for quantifying the difference between the two
results, which can then be attributed to the inaccurate representation of physical
and material properties.

The difference between simulated and the

corresponding experimental results in each specific case for all the four
components are noted in Table 17, below.
Power
1.8W

1.43W

1.05W

Inlet Velocity (m/s) FET-1


FET-2
INDUCTOR MODULE
3.0
1.57
2.24
-0.22
-4.08
2.0
-0.69
0.11
-1.49
-5.86
1.5
-1.05
-0.49
-0.83
-6.92
1.0
-0.90
-0.64
-0.07
-7.12
0.5
-0.57
0.34
1.37
-6.61
3.0
2.42
2.48
1.16
-2.30
2.0
0.79
0.91
1.04
-4.05
1.5
0.43
0.34
1.64
-4.64
1.0
0.64
0.31
2.29
-4.72
0.5
0.81
0.97
3.36
-4.45
3.0
1.78
1.75
3.00
-1.50
2.0
0.90
0.93
3.08
-2.34
1.5
0.81
0.75
3.49
-2.79
1.0
0.92
0.69
4.05
-2.81
0.5
1.26
1.46
5.11
-2.41

Table 17: Difference between simulated and experimental temperature


values for FET-1, FET-2, inductor and module card surface
FET-1, FET-2, INDUCTOR and MODULE represent the difference between
simulated and experimental temperature values for FET-1, FET-2, inductor and
60

the module card respectively. A normality test of these values yields a high p
value in each case indicating that all the four quantities are normally distributed.
Results of the normality tests along with the normality plots are shown in Figure
29, below.
Since the four quantities are normally distributed it is appropriate to do a
paired t-test to compare the simulation results with the experimental data.
Comparison of simulated and experimental T values displayed fairly good
agreement between the two. It seems that the simulation results overestimated
temperature values of the two FETs by 1 0K and the inductor by 2 0K while
underestimating temperature of the module card surface by 4 0K.

Figure 29: Normality plots for FET-1 , FET-2 , INDUCTOR and MODULE
61

The t-statistic values for paired t-tests that were conducted comparing
experimental results to the simulation values are provided in Table 18 along with
the respective p-values. The null hypothesis in each case is that, the difference
between experimental and simulated results is as stated in the table. The high pvalues (> 0.15) indicate that we could be reasonably confident about our
estimates and would fail to reject our null hypothesis at a confidence level of 95%
in each case. Additionally the power of the test is calculated at a difference of 1
0

K in case of the two FETs and 2 0K in case of the inductor and the module

surface. A high (1-) values in each case indicates that there is a very low
probability for the test to not reject the null hypothesis when it is false. In other
words the probability of a Type II error is very low if we have an additional
tolerance of 1 0K in the case of the two FETs and 2 0K in case of the inductor and
the module surface. Hence we can be reasonably confident that our estimated
difference in each case is fairly correct especially considering an additional
tolerance. It is also important to note that maximum difference (including
additional tolerance) of 6

K in case of the module surface represents a

difference of less than 2 % of average operating temperature.

Component
FET-1
FET-2
Inductor
Module

Estimated difference
between experimental &
simulated values
-1 0K
-1 0K
-2 0K
4 0K

t-stat
1.50
0.82
0.41
0.37

p-value
0.16
0.42
0.68
0.72

Power of the
test at 95%
confidence (1-)
0.98
0.99
0.97
0.99

Table 18: Results for paired t- test


62

It should however be noted that the mean difference between


experimental and simulated values in each case will lie anywhere within the
confidence intervals given in Table 19 below 95% of the time.
Component
FET-1
FET-2
Inductor
Module

Mean difference
between experimental
and simulated values
-0.608 0K
-0.810 0K
-1.798 0K
4.172 0K

95% confidence
interval
(in oK)
(-1.170, -0.046)
(-1.305, -0.315)
(-2.856, -0.740)
(3.162, 5.182)

Table 19: Confidence intervals for difference between experimental and


simulated values
Aside from the fact that there were multiple assumptions made for this
simulation study, the mesh especially at the module card level was probably not
fine enough. As a direct consequence difference in T values were particularly
high for the module card surface. However, a finer mesh would make the model
computationally much more cumbersome and hence we choose the current set
of results instead.

63

CHAPTER 4
Conclusions
4.1 Summary
Thermal management of electronic systems involves creating cooling
solutions at the die level, the package level, and eventually at the overall system
level. Commercially available CFD code has already been stated to be useful in
creating these solutions. CFD analysis can be used to evaluate new designs and
optimize existing designs. However, the importance of experimental analysis
cannot be ignored and often, for complex systems, there is a need to validate
CFD results with experimental data.
In Chapter 2, thermal performance of impingement type parallel plate heat
sinks attached to a microchip with significant hot spots was numerically analyzed.
Optimization of thermal performance of these heat sinks was attempted by
varying fin height and air inlet velocity, and also by modifying fin shapes and
varying the number of fins modified. Based on the results of this optimization
study, some basic design guidelines were formulated for parallel plate
impingement type heat sinks. The study illustrates the process of design
optimization using numerical and analytical techniques.
In Chapter 3, thermal characteristics of a fairly complex power module
placed on a test card were experimentally and numerically analyzed. The
64

experimental studies were conducted in an open circuit in-draft type wind tunnel
which complied with JEDEC standards. Primarily, the experiments focused on
temperature measurements at specific locations for comparison with the
modeling results. The Navier Stokes equations representing conservation of
mass, momentum and energy were numerically solved for incompressible flow at
steady state conditions. Simulation results were compared to experimental
results and they were close. This study illustrates that it is possible to get good
agreement of simulation results with experimental data even when details like
exact power distribution or material properties for each and every component are
unknown.
4.2 Future Work
In Chapter 2, a parallel plate impingement type heat sink design was
optimized using CFD techniques. It would be interesting if some experimental
data is generated and then compared to the simulation results. Additionally, a
comprehensive study could be performed to evaluate the design guidelines
proposed in the study and draw a comparison with guidelines proposed in other
similar studies.
In Chapter 3, a CFD model of a power module was created using
equivalent material properties and the model was validated with experimental
data. It would be interesting to see if the CFD model of the power module can
predict thermal performance of the module within a reasonable margin of error in
an actual field application set up. A study comparing thermal performance using
the current or an equivalent model with data from the module installed in an

65

actual application would certainly be interesting. It would also be a good idea to


improve upon the accuracy of simulated values for the module card surface. To
achieve this, a detailed model of the power module with a higher mesh density
can be created with boundary conditions from the current model.

66

REFERENCES
[1] Farnam, D., Sammakia, B. and Ghose, K., Development of a complete
transient microchannel heat sink model, Proc. of 11th Intersociety Conference
on Thermal and Thermomechanical Phenomena in Electronic Systems, ITHERM,
Orlando, FL, 2008, pp. 113-120.
[2] Gondipalli, S., Sammakia, B., and Ahmed, G.R., "Improving the performance
of an impingement heat sink by modifying the fin shapes", Proc. of IPACK'09,
ASME, San Francisco, CA, 2009.
[3] Biber, C., Choosing a heat sink-some tips and recommendations, EDN
Access Mag., vol. 40, no. 21, 1995, pp.125-128.
[4] Sathe, S. and Sammakia, B., A review of recent developments in some
practical aspects of air-cooled electronic packages, Journal of Heat Transfer,
vol. 120, no. 4, 1998, pp. 830-839.
[5] Kiley, R.F. and Soule, C.A., Engineered heat sinks, Powertech Mag., July
1990.
[6] Soule, C.A., Air and liquid cooled techniques for high power density
components, Power Conv. Intel. Motion, vol. 19, no. 11, 1993.
[7] Shah, A., Sammakia, B., Srihari, H., and Ramakrishna, K., A Numerical
Study of the Thermal Performance of an Impingement Heat Sink-Fin Shape
optimization, IEEE Transactions on Components and Packaging Technologies,
vol. 27, no. 4, 2004, pp. 298-306.

67

[8] Dunn, R.M., Schulman, M. and Timko, N., Electronic circuit module cooling,
U.S Patent 4,277,816, Filing date: May 29, 1979, Issue date: Jul 7, 1981.
[9] Biskeborn, R.G., Horvath, J.L., and Hultmark E.B., Integral cap heat sink
assembly for the IBM 4381 processor, Proc. of 4th Annual Int. Electronic
Packing Soc. Conf., Baltimore, MD, 1984, pp. 468-474.
[10] Oktay, S., Dessauer, B., and Horvath, J.L., New internal and external
cooling enhancements for the IBM 4381 module, Proc. of IEEE Int. Conf.
Computer Design: VLSI in computers (ICCD' 83), Port Chester, NJ, 1983.
[11] Sathe, S.B., Kelkar, K.M., Karki, K.C., Tai, C.C., Lamb, C.R., and Patankar,
S.V., Numerical prediction of flow and heat transfer in an impingement heat
sink, in Proc. ASME Int. Electronics Packing Conf. EEP, Binghamton, NY, 1993,
vol. 4, no. 2, pp. 893-898.
[12] McPhee, J.M., O'Toole, T.S., and Yedvabny, M., Cooling the VAX 9000,
Proc. of Electro'90, Conf. Rec., Boston, MA, 1990, pp. 288-292.
[13] Fitch, J.S., A one dimensional thermal model for the VAX 9000 multichip
units, Proc. of Thermal Modeling and Design of Electronic Systems and
Devices, ASME Winter Annual Meeting, Dallas, TX, 1990, vol. 153, pp. 59-64.
[14] Heng, S. and Pei, J., Air impingement cooled pin-fin heat sink for multichip
unit, Proc. of Nat. Electronic Packaging and Production Conf., Des Plaines, IL,
1991, vol. 2.
[15] Bartilson, B.W., Air impingement on miniature pinfin heat sinks, Proc. of
ASME Winter Annual Meeting, Atlanta, GA, 1991.

68

[16] Sathe, S.B. and Sammakia, B. An analytical study of optimized performance


of an impingement heat sink, Proc. of ASME/AIChE Nat. Heat Transfer Conf.,
Portland, OR, 1995, vol. 1, pp 43-49.
[17] Hansen N., Catton I., Zhou F., Heat Sink Optimization; A Multi-Parameter
Optimization Problem, Proc. of International Heat Transfer Conference,
Washington D.C., 2010.
[18] Sathe, S. and Sammakia, B., Interaction of the System and Module Level
Thermal Phenomena: A Flip Chip BGA Example, Electronics Cooling, vol. 4, no.
2, 1998, pp. 14-22.
[19] Burgos, J., Manno, V.P. and

Azar, K., Achieving accurate thermal

characterization using a CFD code-case study of plastic packages, IEEE


Transactions on Components, Packaging and Manufacturing Technology Part A,
vol. 18, no. 4, 1995, pp.732-738.
[20] Han, J.B., Thermal Characterization of Tape BGA Package by Modeling,
Proc. of SPIE 4229: Microelectronic Yield, Reliability, and Advanced Packaging,
Singapore, 2000, pp. 202-208.
[21]

Pauley,

R.A.,

Package

Thermal

Characterization

Methodologies,

Application Report No.SZZA003, Texas Instruments, Dallas, TX, 1999.


[22] Farnam, D.,Sammakia, B. and Ghose, K., Thermal Design Criteria for
Extraordinary Performance of Devices Cooled by Microchannel Heat Sink,
Journal of Thermal Science and Engineering Applications, vol. 2, no. 4, 2010.
[23] FLUENT Users Guide, USA: ANYSIS, Inc, 2003

69

[24] JC-15 Committee on Thermal Characterization, Extension to JESD51


Thermal Test Board Standards to Accommodate Multi-Chip Packages, JEDEC
Standard No. 51-32, JEDEC Solid State Technology Association, Arlington, VA,
2010.
[25] JC-15 Committee on Thermal Characterization, Glossary of Thermal
Measurement Terms and Definitions, JEDEC Standard No. 51-13, JEDEC Solid
State Technology Association, Arlington, VA, 2009.
[26] JC-15.1 Committee on Thermal Characterization, Integrated Circuit Thermal
Test Method Environmental Conditions - Forced Convection (Moving Air),
JEDEC Standard No. 51-6, JEDEC Solid State Technology Association,
Arlington, VA, 1999.

70

S-ar putea să vă placă și