Professional Documents
Culture Documents
Tony A. Asghari
Motorola
21440 W. Lake Cook Rd.
Deer Park, IL 60010 USA
480-200-3448
ABSTRACT INTRODUCTION
In order to achieve a low thermal resistance path directly beneath A statistical factorial design of experiments is an efficient means to
high power dissipating transistors, a thermal via and printed circuit simultaneously study several input factors' effects on an output
board (PCB) optimization study was performed. The thermal response and determine optimal settings for them. In addition, any
resistance normal to the PCB surface served as the primary potential interactions between the factors can be evaluated. By
reference output, while the calculation parameterization was varied varying the levels of the factors, simultaneously, the calculation is
to minimize this thermal resistance. extremely cost-effective, as fewer computer thermal simulation
models need to be run in order to determine the optimal response
output.
A fractional factorial designed experiment (DOE) was developed
using MINITABTM statistical software for the following thermal via
and PCB factors, evaluated at two levels: via diameter, pitch, and
barrel thickness as well as PCB top Cu pad area and number of Factorial designs answer questions such as:
metalized layers.
• Which variables exercise the most influence on
The results for each DOE treatment combination were simulated as the output response?
a parametric run using a finite volume computational fluid dynamics • What factor settings will optimize the results?
(CFD) software tool – Icepak®. Steady-state thermal resistance
values, from a power source on top of PCB to the bottom of board, Previous work includes coupling DOE and CFD to linearly optimize
were determined. various aspects of thermal performance at the system-level of a
network server [1]. Advantages of using DOE to reduce process
This demonstrates a methodology of coupling statistical DOE with variation and improve quality have been demonstrated [2].
thermal CFD to efficiently optimize the thermal performance during Response Surface Methodology takes the study one step further by
the early design process. using central composite designs to determine an optimal point on a
non-linear curve between the end point levels [3].
KEY WORDS
Design of Experiments, Factorial Design, The purpose of this optimization study is to select the factors (via
MINITABTM, Icepak® CFD, PCB, Thermal diameter, pitch, top copper (Cu) pad thickness, via barrel thickness,
Resistance, Thermal Via, Optimization and number of Cu layers) and select values for them that will result
in reducing the overall thermal resistance of a PCB.
NOMENCLATURE
DIAM = thermal via diameter
This paper shows a technique of combining statistical DOE with
PITCH = adjacent thermal via center to center distance thermal CFD to efficiently optimize the thermal resistance of a PCB
during the early design process.
TOP_CU = PCB top copper thickness
BARREL = thermal via copper barrel thickness
LAYERS = PCB metalized copper layers
PROCEDURE the PCB, to the bottom of the board was calculated using
Icepack 4.1 CFD software. • A lower PCB thermal
MINITABTM resistance was desired.
THERMAL
VIA
VIA
225
Machine Translated by Google
ANOVA
Figure 2: Icepak® ¼ Symmetry Isometric view of model Factor Type Levels Values
showing temperature gradient from source thru PCB as PITCH fixed 2 25.35
well as thermal via configuration. BARREL fixed 2 1, 1.4
LAYERS fixed 2 forty six
Source DF Seq SS
PITCH first 0.4489
BARREL first 0.44223
LAYERS 0.2304
TOP_CU 11 0.0576
Error 11 0.13467
Total 15 1.3138
R-Sq = 89.75%
Table 3: ANOVA Results
226
Machine Translated by Google
Normal Probability Plot of the Effects Interaction Plot (data means) for Thermal Resistance (C/W)
(response is Thermal Resistance (C/W), Alpha = .10) 25 35 2 2.5 first 1.4 4 6
99
2.4
DIAM
Effect Type
14
Not Significant 2.2
B DIAM 16
95 Significant
2.0
90 Factor Name
A DIAM PITCH
2.4
80 B PITCH 25
C 2.2
70 TOP_CU PITCH 35
60 D BARREL 2.0
E LAYERS
50
40 2.4 TOP_CU
Percent 2
30 2.2
TOP_CU 2.5
20
C 2.0
ten E
2.4
BARREL
5 D first
2.2
BARREL 1.4
2.0
first
2.0
The d in the y-axis of the optimizer chart is the level of
14
BARREL
16 25
LAYERS
35 2 2.5
desirability. This desirability value is between zero (low)
2.3
and one (high). For this study, the desirability value was
2.2
0.838, resulting in the lowest thermal resistance output
2.1
Mean
value of 1.71 oC/W.
2.0 The optimization performed for this analysis was linear
first 1.4 4 6 (analyzed at 2 end points for each factor). This study
could be taken one step further to include center-points
Figure 5: Main Effects Plot Figure
for each of the factors, allowing one to determine the
5 shows a main effects plot. A more extreme slope indicates a more
optimal point on a non-linear curve (Response Surface
significant effect on PCB thermal resistance. Note the thermal via
Diameter factor has a horizontal slope; Therefore, it is not as Methodology).
Figure 8 is a cube plot showing PCB thermal resistance values as a
significant.
function of the various factors.
The lowest thermal resistance value of 1.71 oC/W
Figure 5, also, shows the effect on PCB thermal resistance
occurs with a 14mil via Diameter on a 25mil Pitch, with
for each factor at the corresponding high
& low levels. 2.5mil top Cu Pad thickness, with a via Barrel thickness
of 1.4mil, and a 6 layer Cu thick PCB.
Figure 8, is a good representation of a fractional factorial,
where all corners of the Cube Plot do not
227
Machine Translated by Google
have output trial results. A full factorial would have output REFERENCES
results on all corners of the Cube Plot.
[1] T. Stewart and DW Stiver, “Thermal Optimization of
CONCLUSIONS Electronic Systems using Design of Experiments Based on
Numerical Inputs”, presented at the 20th IEEE SEMI-
This study demonstrates how to use MINITABTM THERM Symposium, 2004.
statistical software to perform a DOE study to optimize a [2] A. Zargari and W.R. Grise, “A Comparison of Traditional Methods
PCB normal direction thermal resistance. of Setting-Up the Experiments with the Utilization of Computer-
The study indicates that in order to achieve a desired lower Based DOE Software: Validation of a Scientific Experiment”,
PCB thermal resistance, the design engineer needs to: Electrical Insulation Conference & Electrical Manufacturing & Coil
Winding Conference, 1999.
1) Decrease thermal via pitch size. [3] BA Zahn, “Steady State Thermal Characterization and
2) Increase via Cu barrel thickness. Junction Temperature Estimation of Multi-Chip Module
3) Increase number of PCB metalized Cu layers. Packages Using the Response Surface Method”, IEEE
Transactions on Components & Packaging Technologies,
Vol. 23, No. 1, March 2000.
Using ANOVA methodology, the relative sensitivity of PCB
thermal resistance to each of these significant parameters [4] DC Montgomery, Design & Analysis of Experiments.
was determined. New York: John Wiley & Sons, Inc.,
The design engineer, early in the program, can work with 2001.
PCB suppliers to achieve a balance between thermal [5] FP Incropera and DP DeWitt, Fundamentals of Heat
performance, PCB cost, and supplier capability. and Mass Transfer. New York: John Wiley & Sons, Inc.,
1996.
This methodology can be used to optimize any similar [6] B. Ryan and BL Joiner, MINITABTM Handbook,
system. 4th Edition. Pacific Grove, CA: Duxbury, Thomson Learning,
2001.
[7] Icepak® 4.1 Users Guide. Fluent© Inc, 2005.
2.35 2.12
1.99 1.71
1.95 1.81
6
LA YERS
2.58 2.03
4
2.79 2.36
35
14 16
DIAM
first 1.4
BARREL
228