PCB Thermal

Simulation - The State

of the Art
Alexandra Francois-Saint-Cyr
Applications Engineering Manager
Mechanical Analysis Division

March 16, 2010

 PCB Thermal Simulation –
The State of the Art

 Agenda
— Introduction to the Mentor Graphics Mechanical
Analysis Division
— PCB Design Challenges
— Electronics Failures Related To Temperature
— Heat Transfer 101
— Thermal Design Tools Review
— PCB Thermal Simulation
— Application Example
— Assessing Thermal Simulation Accuracy

© 2010 Mentor Graphics Corp. Company Confidential

2 www.mentor.com
 About the Mentor Graphics
Mechanical Analysis Division

 A new division formed in August 2008 after the

acquisition of Flomerics by Mentor Graphics
 Focused on delivering analysis and simulation
software and services for mechanical design
 Division headquarters: Hampton Court, London, UK
 Over 180 employees
 Development in UK, Moscow and Budapest
 Direct Sales and Support operations throughout the
world

© 2010 Mentor Graphics Corp. Company Confidential

3 www.mentor.com
 Principal Lines of Business

 Thermal Design of Electronics

— FloTHERM = clear market leader
— 75% of division’s revenue
 Concurrent Computational Fluid
Dynamics (CFD)
— FloEFD - a different breed of CFD
software that is fully embedded
in the mechanical design
environment
 Building Heating & Ventilation
— Optimize airflow, temperature
distribution and contamination
control in and around buildings
and in HVAC equipment
— Data Center Cooling is a
particular focus

© 2010 Mentor Graphics Corp. Company Confidential

4 www.mentor.com
 Typical Design Constraints

 Operate at Maximum Ambient

Temperature
 Meet Federal EMI
Specifications
 Reliability, Cost, Size and
Noise Considerations limit the
Number of Fans
 Meet Manufacturer
Recommended Maximum
Junction Temperature for
Components

© 2010 Mentor Graphics Corp. Company Confidential

7 www.mentor.com
 Causes of Electronic Failure

 Examples of failures related to temperature

— Coefficient of Thermal Expansion mismatches inducing
mechanical stresses
— Electrical performance lessened by changing device
parameters
— Corrosion (Encapsulant failure)
— Current leakage
— Oxide breakdown
— Electro-migration

8
© 2010 Mentor Graphics Corp. Company Confidential
8 www.mentor.com
 Causes of Electronic Failure
 Number of Failures after 1000 HRS / Million Units

Component T = 25 ºC T= 75 ºC

Thick film resistor 5 15 3X

Chip capacitor 10 25 2.5X
Power transistor 50 300 6X
Diode 1 9 9X
Logic ICs - SSI 125 1125 9X
Logic Ics - MSI 250 2250 9X
Logic Ics - LSI 500 4500 9X
Source: C.A. Harper, Handbook of Thick Film Hybrid Microelectronics

9
© 2010 Mentor Graphics Corp. Company Confidential
9 www.mentor.com
 Effect of PCB Component Layout on Junction
Temperature and Failure Rate

Layout 1 Layout 2

84.9 92.2 97.2

105.5 126.3 122.4 104.9 87.5 96.1 95

122.1 124 111.9 90.9 124 89

60.8 128.2 128.9 111.9 93 47.9 93.7

114.8 125.8 124 108.8 92.2 92.1 94.4

105.6 113.1 108.8 97.6 83 89.9 94.4 97.6

Mean Tj = 112.9°C Mean Tj = 89.8°C

Std. Dev. = 15.2°C Std. Dev. = 10.6°C

Failure rate 8 times higher in layout 1 than in layout 2

Source: Hanneman, 1977

© 2010 Mentor Graphics Corp. Company Confidential

10 www.mentor.com
 Heat Transfer 101
 Heat transfer is the transfer of thermal energy due to a
temperature difference
 Heat gets moved from heat source to heat sink by
conduction
 Heat sink transfers heat to ambient air by convection
— Heat can also be radiated to surrounding environment
Heat Sink

Component dissipating heat

PCB

Heat Transfer by Conduction

Heat Transfer by Convection and Radiation

© 2010 Mentor Graphics Corp. Company Confidential

11 www.mentor.com
 Thermal Resistance Definition
Q
Face A
 Conductive Heat Transfer Face A
at
at fixed L
• Rk = ∆T/Q = L/kA (K/W) T1 > T2 fixed
T2 < T1
• Similar to RΩ = ∆V/I (Ohms)
x
k: Material thermal conductivity T1 T2
L/kA
 Convective Heat Transfer Qh
V∞ ,T∞
— Rh = ∆T/Qh = 1/hA (K/W)
∆T = Ts - T∞ TS

Surface A

h: Heat transfer coefficient TS T∞

1/hA
© 2010 Mentor Graphics Corp. Company Confidential
12 www.mentor.com
 The Thermal Budget
 A useful design tool defined as:
— Defined as: ∆Tbudget = Q * RJA [K]
 Breaks the problem into clearly defined heat paths for a
clear design understanding
TA
RSA Sink to Ambient Resistance
Ts
RCS Case to Sink Resistance
TC
RJC Junction to Case Resistance
TJ
RJB Junction to Board Resistance
TB

© 2010 Mentor Graphics Corp. Company Confidential

13 www.mentor.com
 Thermal Design Tools
 Hand calculations/Spreadsheet
— Excellent tool for early design exploration
 Finite Element Analysis (FEA)
— 3D numerical analysis
— Typically doesn’t calculate convective heat transfer and
radiation explicitly
 Computational Fluid Dynamics (CFD)
— 3D Conjugate fluid flow and heat transfer numerical
analysis
 Lab tests
— Most value when used as a model validation - rather
than for parametric investigation

© 2010 Mentor Graphics Corp. Company Confidential

14 www.mentor.com
 In the beginning…

 20 years ago thermal

management, and therefore
simulation, was focussed on the
mechanical system level
 PCBs represented as 2D plates Q(W)
Q(W) Q(W) Q(W)
Q(W) Q(W)

 Prescribed split of heat dumped

into the air on either side
 Good enough for local Ta
prediction
— and so optimisation of air flow
partitioning in slots etc.
 No board or component
temperature prediction
— Also an assumption of uniform
heat distribution on PCB
© 2010 Mentor Graphics Corp. Company Confidential
15 www.mentor.com
 Simplest of 3D representations

 Heatsinking soon became a

common cooling method
 From a simulation
perspective this forced a 3D
representation of the board
and components
 Simple block representations
of board and components
— With power uniformly
dissipated in the component
blocks The Question
What is a good value of thermal conductivity to use in
FLOTHERM simulations for printed circuit boards (PCBs)
 What value to use for and components if you have no real data?
The Short Answer
thermal conductivity? 10 watt/m/C.
— Indicative of case [Tony Kordyban, 3rd International Flotherm user
conference , 1993]
temperature… in most cases © 2010 Mentor Graphics Corp. Company Confidential
16 www.mentor.com
 The grail of Tj and Tc modeling
 The most dependable indicator of
thermally affected reliability is the
component junction temperature (Tj)
(as well as the temperature
gradients created as a consequence)
 Increases in functional layout density
and speeds have resulted in many
more components operating at/near
their maximum operating
temperatures
 The need for thermal validation (“will
this design fail thermally or not?”) is
now an integral part of all electronic
design processes
 So, how best to predict such Tj and
Tc values…?
© 2010 Mentor Graphics Corp. Company Confidential
17 www.mentor.com
 Component Representation
 Various methods of component representation exist
 Choice most often based on data availability rather than
simulation intent!
Top Inner Top Outer
Case (top)
Side Leads
Junction Junction

θja
Bottom Outer
k=? Board Bottom Inner

Thermal Simple block

resistance model Detailed model
from junction
to ambient Good for Thermal Resistor Network model Explicit representation of
metric ‘typical’ case internal construction
temperature “2-Resistor” and “DELPHI” types geometry, materials, die
Good for prediction at Capable of Tj and Tc prediction size etc.
comparative best Generally DELPHI models are better Capable of Tj and Tc
purposes than 2-Resistor prediction
only, NOT Results as good as the derivation of
predictive the characteristic thermal Most accurate model type
resistances available

Increasing Tj and Tc predictive accuracy

© 2010 Mentor Graphics Corp. Company Confidential
18 www.mentor.com
 From source to ambient:
Modeling heat flow paths

 Heat is dissipated in the component and travels to a (colder) ambient

 From a design perspective it’s important to get the heat out

easily/quickly (choose your analogy), conversely get the cold in
easily/quickly
 Similarly, from a simulation perspective, it’s important to model the
critical heat flow paths accurately
— In terms of the thermal resistances the heat experiences as it leaves by:
– Convection (heat removal by air)
– Conduction (heat removal through solid)
– Radiation (heat removal by em radiative transfer from solid to solid surface)
 Only a full 3D CFD simulation captures all modes of heat transfer
accurately

© 2010 Mentor Graphics Corp. Company Confidential

19 www.mentor.com
 Modeling the thermal resistance of the PCB
stack-up

 Arguably the most critical thermal resistance the heat experiences

(outside of the package) is the conductive resistance of the PCB itself
 Various methods exist for the representation of the PCB stack-up
 All of which are predicated on the theory of serial and parallel thermal
resistances theorem (Thévenin’s equivalent electrical circuit) such that:
R1
R2 Reffective

R3 1/Reffective = 1/R1 + 1/R2 + 1/R3

R1 R2 R3
Reffective Reffective = R1 + R2 + R3

 For conductive thermal resistances, R = d/kA

— d = length of heat flow, k = thermal conductivity, A = cross sectional area of heat flow
— So, for a given collection of resistances (e.g. FR4 and Cu) a single ‘effective’ thermal conductivity
may be derived

© 2010 Mentor Graphics Corp. Company Confidential

20 www.mentor.com
 The fully lumped approximation

 A refinement of the lumped “10 W/mK” single block approach

 The PCB is represented as a single block and effective thermal
conductivities (orthotropic, different in different directions)
applied in the x, y (in-plane) and z (through plane) directions

Y
X

21
© 2010 Mentor Graphics Corp. Company Confidential
21 www.mentor.com
 The fully lum ped approximation

Top layer
 A refinement of the lumped “10 W/mK” single block approach
 The PCB is represented as a single block and effective thermal
FR4
conductivities (orthotropic, different in different directions)
applied in the x, y (in- plane) and z (through plane) directions
GND

Through Plane (z)

FR4

kz (W/mK)
Effective
VCC

Z
FR4

Y
X
Bottom layer

© 2010 Mentor Graphics Corp. Company Confidential

22 www.mentor.com
 The fully lumped approximation

 A refinement of the lumped “10 W/mK” single block approach

 The PCB is represented as a single block and effective thermal
conductivities (orthotropic, different in different directions)
applied in the x, y (in-plane) and z (through plane) directions

In Plane (x = y)
Top layer

FR4

GND

FR4
Effective
VCC
kx = ky
FR4 Z
(W/mK)
Bottom layer
Y
X

© 2010 Mentor Graphics Corp. Company Confidential

23 www.mentor.com
 The fully lumped approximation

 Based on the assumption that all heat is either conducting through or in

the plane of the board
 Accuracy reduced when spreading occurs, e.g. from small high powered
surface mount actives
 But numerically efficient and simple to specify (%Cu, board dimensions)

Y
X

© 2010 Mentor Graphics Corp. Company Confidential

24 www.mentor.com
 Discrete layer representation
 Next level of refinement is a more
explicit representation of the change in
thermal resistances through the
thickness of the board
 Each layer is modeled as a separate
object with its own thermal conductivity
 Metallic layers are still composites of
FR4 and Cu traces, pads etc.
— Metallic layer effective thermal conductivity
can be derived assuming thermal resistances
in parallel for all 3 directions, just %Cu
required
 Numerically more intensive but provides a
better accommodation of spreading effects
— Especially if the distribution of Cu on metallic
layers is relatively uniform
© 2010 Mentor Graphics Corp. Company Confidential
25 www.mentor.com
 Full metallic distribution representation
 Current simulation
state of the art is a
representation of the
change in thermal
resistances in X, Y
and Z directions
 To achieve this a
detailed description
of the distribution of
Cu in both metallic
and dielectric layers
is required
— Necessitating import
of layer artwork
descriptions from
EDA tools into the
simulation tool

© 2010 Mentor Graphics Corp. Company Confidential

26 www.mentor.com
 Full Metallic Distribution Representation
 Computational resources
are currently not at a
level to support explicit
representation of each
individual Cu feature
 Even at this level an
effective thermal
resistance approach is
taken
 Each layer is subdivided
into a tessellated array of
‘patches’
 Effective orthotropic
thermal conductivities are
calculated for each ‘patch’
by examining the
distribution of Cu/FR4
within that area © 2010 Mentor Graphics Corp. Company Confidential
27 www.mentor.com
 Full Metallic Distribution Representation

 Trace data imported directly from major EDA tools into

FloTHERM/FloTHERMPCB

Discretized Cu distribution for top layer

© 2010 Mentor Graphics Corp. Company Confidential

28 www.mentor.com
 Full Metallic Distribution Representation
 Black and white image of Cu distribution is attached to
each layer and is transformed into patches of various
thermal conductivities

© 2010 Mentor Graphics Corp. Company Confidential

29 www.mentor.com
 Application Example: JCI PCB

 Board provided by JCI

Automotive Group, MI
 Dashboard PCB tested in 26ºC
still air environment
 Total power dissipation = 9 W
 Analysis done using FloTHERM

FloTHERM PCB Top Side FloTHERM PCB Bottom Side

© 2010 Mentor Graphics Corp. Company Confidential

30 www.mentor.com
 Application Example: JCI PCB
Bottom trace before discretization
Bottom trace after discretization
Trace 4 before discretization
31
 PCB contains 6 copper layers
— Each layer becomes a pixelized
representation of the thermal
conductivity based on the
Copper distribution
— Each pixel is assigned a value
for kx, ky, and kz as determined
by image processing and
becomes a layer patch
— A similar approach is used for
the via representation
Trace 4 after discretization
© 2010 Mentor Graphics Corp. Company Confidential
www.mentor.com
 Application Example: JCI PCB

 Results
— Temperature
distribution very
similar between
experimental and
numerical model
PCB bottom surface temperature (IR)
— Top layer results
also compared
very well

PCB bottom surface temperature (FloTHERM)

© 2010 Mentor Graphics Corp. Company Confidential

32 www.mentor.com
 Application Example: JCI PCB

 Results
— 3.4% Average
discrepancy
between
numerical and
experimental
results
— Maximum error
of 10.7%

© 2010 Mentor Graphics Corp. Company Confidential

33 www.mentor.com
 Detail leads to accuracy?
 “A detailed PCB representation will increase my results
accuracy”
 A true statement but caution is required
 Increased detail will lead to longer simulation solution times
 So, when is a detailed PCB representation necessary?
— When the critical heat flow path is through the board
– Natural convection or conduction cooled environments
– Small high powered surface mount actives where the local PCB
copper acts as a heatspreader
 So, when should a simpler PCB representation be used?
— When the board is not on the critical heat flow path
– Forced convection cooled environments
— When your component representation is not detailed enough
– Why model the PCB Cu content in detail when all components are
modelled as simple blocks?

© 2010 Mentor Graphics Corp. Company Confidential

34 www.mentor.com
 A final word on Accuracy
 “How accurate is the thermal simulation?”, a common
question for obvious reasons
 Sources of error fall into three categories:
1. Data availability
– Dictates both component and PCB modeling approaches
2. Data accuracy
– If there are just 3 inputs to define accurately then they are power
dissipation, power dissipation and power dissipation
3. Numerical modelling
– CFD grid not fine enough on the air side around components and/or
in the board just under components
– Incorrect assumption made about the environment the PCB is
placed in
 A good model(er) that is aware of the above 3 issues should
produce Tj – Ta results within 10% of experimental

© 2010 Mentor Graphics Corp. Company Confidential

35 www.mentor.com
 PCB Thermal Simulation –
The State of the Art
 Questions?

© 2010 Mentor Graphics Corp. Company Confidential

36 www.mentor.com
 www.mentor.com

© 2010 Mentor Graphics Corp. Company Confidential

37 www.mentor.com