This document describes a case study using ESI Presto and Mineset software to perform design space exploration for electronics cooling. The case study examined 7 parameters that could affect the maximum skin temperature, touch temperature, and CPU temperature of a tablet design. Over 16,000 simulations were run to understand how variations in parameters like material conductivity and system power impacted the temperatures. Machine learning techniques in Mineset, like column importance, decision trees, and parallel coordinates, were used to analyze the results and identify parameter ranges that minimized high temperatures. This allowed designers to better understand design tradeoffs and constraints for further refinement.

1. 1www.esi-group.com
Copyright © ESI Group, 2015. All rights reserved.
Case Study in Design Space Exploration
for Electronics cooling

2. Designs often depend on combinations of many of these…
Physical
parameters
• System dimensions
• Component
dimensions
• Component locations
• System/component
orientations
Material
parameters
• Skin/casing material
• Heatsink, spreaders
• Thermal interface
material
• Cooling fluid material
Operational
Parameters
• Ambient temperatures
• Ambient pressures
• Fan types
• Fan speeds
• Number of fans
• Powers
Challenge: Multiple parameters

3. Each one adds more complexities and constraints to the design cycle
Increasing
performance Increasing power
densities
Tight design cycles
Cost
constraints
Manufacturability
Reliability
Further Challenges…

4. 4www.esi-group.com
Copyright © ESI Group, 2015. All rights reserved.
Tablet design in ESI Presto
● Number of Parameters : 7
o Material – conductivity (W/m-K)
o Midplane chassis material (15, 90)
o Gappad material (0.5, 3.5)
o Graphite sheet material (15, 1500)
o PCB conductivity (15, 55)
o Geometric
o Graphite sheet thickness (100 to 500 microns)
o System power, Ambient temperature
● Total runs performed: 15
● QoI: Skin temperature, CPU max temp

5. 5www.esi-group.com
Copyright © ESI Group, 2015. All rights reserved.(*) included in the demo
Sample Case Study – Cooling Simulation
Business
Problem
Input parameters:
1. Mid Chassis Conductivity
2. PCB Conductivity
3. Ambient Temperature
4. System Power
5. Graphite Thickness
6. Graphite Conductivity
7. Gap Pad Conductivity
Output parameters:
1. Max Skin Temperature
2. Max Touch Temperature
3. Max CPU Temperature
Objectives
Understand the interplay of input parameters vs output results
Characterize which parameter ranges cause extremes in output values
Analyze the effects of individual parameters

6. 6www.esi-group.com
Copyright © ESI Group, 2015. All rights reserved.
General Approach
• Design tests for the hypothetical requirements
• Select parameter ranges: 4th order collocations means 4 values per
each parameter
• Run simulations for all evaluation points
• Transform data and export to ESI Mineset
‣ Data entered as a simple CSV file containing parameter values of each
simulation
• Iterate through the process using lessons learned from the previous
round

7. 7www.esi-group.com
Copyright © ESI Group, 2015. All rights reserved.
Case Study Design Space Exploration
Mineset + PRESTO for Chip Manufacturing
PRESTO
Sparse Grid DOE
7 Parameters
4 collocations
each
MINESET
Machine
Learning
Using 16,384
samples

8. 8www.esi-group.com
Copyright © ESI Group, 2015. All rights reserved.
to understanding the data
Statistics is the starting point
For continuous numbers look
at the distribution, extremes
and statistical markers
For categories, get a
histogram
Output temperatures
show a Bell-curve of
distribution
Each parameter value
has been seen the
same number of times

9. 9www.esi-group.com
Copyright © ESI Group, 2015. All rights reserved.
Shows the main factors affecting this target outcome
Column Importance ranking for Touch temperature
• System Power is the biggest predictor of Max
Touch Temperature, followed by Ambient
Temperature (Tamb) etc.
• The first two parameters provide an outsized
influence to the results

10. 10www.esi-group.com
Copyright © ESI Group, 2015. All rights reserved.
Not the same factors
Column Importance ranking for CPU temperature
• For the CPU temperature graphite
conductivity is shown to be the most
important factor.
• The first three parameters provide a nearly
complete predictive ability for the CPU
temperature

11. 11www.esi-group.com
Copyright © ESI Group, 2015. All rights reserved.
What-if Analysis
Evidence Visualizer for target:
CPU Temperature
Distribution of low ( )
medium ( )
and high ( )
temperatures in the
overall set
Breakdown
of the CPU
temperature
within
ranges of the
input
variables

12. 12www.esi-group.com
Copyright © ESI Group, 2015. All rights reserved.
Posterior probability is
calculated based on the
data range
Keeping the Graphite
Conductivity and System
Power no higher than
mid-ranges, cuts the
chances of overheating
in half
What-if Analysis
Interpreting the interactive results
This selection:
• Graphite Conductivity
between 254.16 and 752.5
• System Power between
1.083 and 2.05

13. 13www.esi-group.com
Copyright © ESI Group, 2015. All rights reserved.
Decision Tree Analysis
• At each branch the data is further subdivided based on what condition best separates the
outcome results
• The nodes show the distribution of temperatures based on the rules leading to it

14. 14www.esi-group.com
Copyright © ESI Group, 2015. All rights reserved.
Decision Tree Analysis
When Graphite Conductivity is
< 254.167, resulting CPU
temperature is high in 2/3 of
simulation cases
Graphite Conductivity > 254.167
and System Power < 2.05
guarantees no overheating

15. 15www.esi-group.com
Copyright © ESI Group, 2015. All rights reserved.
Parallel Coordinates – Which parameter settings result
in low CPU temperatures ?
Temperatures
below 36 are
possible only in
the highlighted
cases where:
• Sys_Power < 2
• Tamb < 31
• Graphite_cond
>= 500
Selecting records with low
CPU temperature (< 36)

16. 16www.esi-group.com
Copyright © ESI Group, 2015. All rights reserved.
Parallel Coordinates – Conditions for high temperatures
The simulations resulting in high
temperatures come only from settings of
maximum Tamb and minimum graphite
conductivity
Selecting records with
high temperature (> 50)

17. 17www.esi-group.com
Copyright © ESI Group, 2015. All rights reserved. 17
Summary
Mineset
Features
Used
Histograms and Statistics
Understand parameter distribution and
characteristics
Column Importance
Calculate attributes with highest influence on
resulting temperatures
What-if analyzer
Enable exploration of the design space in order to
determine best parameter values to minimize high
temperatures
Decision Tree
Find the rules that lead to either positive (low
temperatures) and negative (high temperatures)
outcomes
Graphic Plots For visual inspection and exploration
Actions Understand the results for better refinement on next round of simulations.

18. 18www.esi-group.com
Copyright © ESI Group, 2015. All rights reserved.
ESI Mineset :
https ://clou d. es i -gro up . com /analy tic s
ESI Presto:
https ://w w w. es i -grou p .com/software -s olu tions /virtu al -
environ ment/cfd -multiphys ics /es i -p resto
Free 14-day trial:
https://cloud.esi -group. com/mineset/trial -signup

19. 19www.esi-group.com
Copyright © ESI Group, 2015. All rights reserved.
Links:
• Mineset :
https://cloud.esi-group.com/analytics
• Free trial:
https://cloud.esi-group.com/mineset/trial-signup