“All models are wrong, but some are useful” Part III
“All models are wrong, but some are useful” Part III
The next two parts in this series focus on the thermal model representations of electronic objects. The first being packages, or should that be components, or chips? From a mechanical analysis perspective they're the things that are plonked onto a PCB that get hot. These things do clever electrical stuff, or should that be electronic stuff? If you haven't already guessed I'm a mechanical engineer with a penchant for fluid and thermal effects, apologies for my somewhat uneducated use of electronical terminology ;)
Either way, from the heat's perspective the package is just the first stage in a series of resistances it experiences as it makes it way to the ambient (conversely it's the last barrier the cold has to go through before it can quench the heat source). Under such conditions the heat flux is very high (this is the area into which the heat has to squeeze out (W/m^2)) and so any resistance the heat experiences will have a disproportionate effect on the source (junction) temperature. So, getting these thermal resistance right is key. When modelling a package there are two ways in which you can represent these resistances.
Such models contain explicit 3D representations of the internal construction of a package. A 3D object representing the die, with the correct material properties, with the power being set to dissipate on its surface. All the other important internal objects such as die attach, bond wires or leadframe, encapsulant etc. with their material properties. Very few assumptions about the thermal resistances are therefore made and thus the resulting accuracy is as good as you can achieve.
OK, so why are such package models rarely used? Not through want of trying that's for sure. It comes down to the fact that component suppliers are unwilling to divulge what is seen as proprietary information regarding their package. What, like small minor design issues? No, whopping big ones like die size.
We provide a tool called FloTHERM.PACK (formerly Flopack) that has the ability to create thermal models of packages which are parametrically specified. The following is part of the design sheet for a TBGA:
Once parametrically specified a detailed model can be downloaded and used in FloTHERM or FloTHERM.PCB to conduct the actual thermal simulation. If things like die size are not known then sensible defaults are assumed based on the experience we have of the more common package constructions.
Some assumptions, or rather simplifications, are made for the detailed package model. Grid count can be minimized by representing all solder balls as a single lump of homogeneous material with a thermal conductivity representative of the fact that there is both air and solder in that volume of space (a common trick used often for lots of electronics thermal models).
Thermal Resistor Network
A thermal resistor network (sometimes referred to as a CTM (compact thermal model)) is the alternative method of package representation.
"As if CFD based electronics thermal simulation wasn't involved enough. Why do you guys go and complicate things with lots of options and methods? Jeesh."
The main advantage of a CTM is that such a description does not give away any proprietary information. It is 'compact' in that it's a very simple representation but still 'behaves' (thermally) like the real thing. It is a bit abstract though in terms of getting your head round the concept.....
The inside of the package is represented as a collection of thermal resistances (degC/W) that link abstract points (nodes) up together. The topology of the network of resistances (how many nodes there are and how they are connected) is intended to represent the dominant heat flow paths the heat follows as it leaves the package. There will be internal nodes, e.g. one that represents the junction and peripheral nodes, e.g. one that represents the top of the package, one that represents the bottom etc. With the power dissipation value assigned to the junction node a clever program like FloTHERM or FloTHERM.PCB can solve this network to find out how much of the heat flows through the resistances and what the temperature values are at the nodes. All without requiring die size or any other explicit physical property of the package. Suppliers are happy, thermal engineers are happy. Everyone hug.
"Hey, hold on, everyone knows you don't get something for nothing, what's the catch with these CTM thingamajigs?"
Not surprisingly it's accuracy. The CTM should behave the same as the detailed model regardless of what environment they are placed in. In reality if a great big heatsink is placed onto a package the heat removal paths will be different compared to when that same package is put on a board that is wedgelocked into a sealed enclosure where the heat is designed to be sucked down out through the board. It can well be that the resistor topology does not account for the (surprisingly) wide variations in heat removal paths. It may not be Boundary Condition Independent (BCI). A good CTM will be BCI. I won't go into detail here about various ways in which a CTM can be derived. JEDEC have just issued some relevant guidelines for both DELPHI and 2-resistor CTM types if you want to read more.
Block models and no model at all models
Both detailed and CTM models are capable of predicting junction and case temperatures thus providing the ability to judge thermal compliance. But what, if like the most of the real world, you can't find any such models? The best you'll be able to do is to model your package as a single block with a single thermal conductivity and assume the power is dissipated throughout that volume.
"Wow, sounds much easier. All I need is the footprint size, height and thermal conductivity.... Hey hold on, you nearly had me there again, so what's the catch!?"
Well, knowing what thermal conductivity value to use that best represents the package as a whole aint easy. Somewhere between 0.5 and 10 W/mK. Buy FloTHERM or FloTHERM.PCB, we've supplied a whole bunch of values based on package style. Even then due to non-BCI behaviour accuracy will not be perfect (but then again, that's the whole point of this series...).
The final component modelling 'level' is 'no model at all'. For small passive surface mounts capacitors, resistors and the like they are so thermally passive that they will not effect the heat around them, they will simply assume the temperature that is coming from other more thermally dominant packages near by. Often you won't care about such passives but if you do then just ignore them for the simulation but note the board temperature where they sit in reality, that will be = case = junction temperature.
"For a series that's meant to be about accuracy you've just made me read 1100 words and I'm still none the wiser."
Sorry, I do have a tendency to ramble but the following summary would not have much sense without the package modelling level background. All % errors themselves should have error bars associated with them. The following is a VERY ROUGH guideline. The errors are specific to just the component model and have been learnt through experience as opposed to formal study. Based on dTj rise over ambient, compared to 'reality':
- Detailed ~5%
- DELPHI ~10%
- 2-R ~ 20%
- Block - ~20% on case temp rise, unable to provide junction temp
These average errors should set expectations only. There are times when a 2-R model can be very accurate (for certain package styles when placed in certain environments).
Having the accuracy of a detailed model with the ease of definition, distribution and hiding of proprietary data of a CTM would be a great combination. Yep, sure would.
5th June, Ross-on-Wye
More Blog Posts
- Why Not Just Shove a Heatsink on Top of it? Part 2: Heat Flow Budgets
- Why Not Just Shove a Heatsink on Top of it? Part 1
- Hot Off the Press
- Experiment vs. Simulation, Part 5: Detailed IC Package Model Calibration Methodology
- CFD - Colourful Friday Distractions
- Experiment vs. Simulation, Part 4: Compact Thermal Models
- Some Like it Hotter
- Experiment vs. Simulation, Part 3: JESD51-14
- 13 mm in 3 Months
- Experiment vs. Simulation, Part 2: TIM Thermal Conductivity
- May, 2013
- April, 2013
- March, 2013
- February, 2013
- January, 2013
- December, 2012
- November, 2012
- October, 2012
- September, 2012
- August, 2012
- July, 2012
- June, 2012
- May, 2012
- April, 2012
- March, 2012
- February, 2012
- January, 2012
- Bottlenecks and Interface Materials; Part 2 - When TIMs Go Bad
- Bridging the Simulation Supply Chain; NXP Semiconductors, a Case in Point
- Bottlenecks and Interface Materials; Part 1 - Great Thermal Bedfellows
- Is Pipe Insulation Effective?
- Emails, more Emails and Jeff Bridges
- LEDs; The future's bright and hot.
- December, 2011
- November, 2011
- October, 2011
- September, 2011
- August, 2011
- July, 2011
- June, 2011
- Come, meet FloTHERM/VENT/EFD users, learn and enjoy!
- The Future of Air Travel
- PC Overclocking and Aftermarket Modding. Part III - Power vs. Frequency?
- Applebee's switch to LED lighting and saves massively
- Mechanical U2Us are Here!
- PC Overclocking and Aftermarket Modding. Part II - Liquid Nitrogen Overclocking, How Cool is That?
- May, 2011
- April, 2011
- March, 2011
- February, 2011
- January, 2011
- FloEFD HVAC Module - Taking Built Environment CFD Simulation to the Next Level
- Beer Fridge - A Case Study in Thermal Design. Part 5 - Time for a FloBEER
- Driving Fun with CFD
- Beer Fridge - A Case Study in Thermal Design. Part 4 - FloBEER
- Beer Fridge - A Case Study in Thermal Design. Part 3 - Side Up or Upside Down?
- December, 2010
- November, 2010
- Beer Fridge - A Case Study in Thermal Design. Part 1 - A Gift
- A transient in methods: the RthJC transient method
- On a Wing and a Prayer - Part Deux
- What Can You Learn When You Turn It On?
- Oh these pilots...
- How thermal testing can help increase reliability of electronic systems?
- Measuring the real world - at least the thermal properties of semiconductor device packages
- On a wing and a prayer…
- We Love FloTHERM - 8 Reasons to Upgrade to V9.1
- October, 2010
- September, 2010
- August, 2010
- Lies, Damned Lies, and “CFD Comparison Charts” – Part IV
- Lies, Damned Lies, and “CFD Comparison Charts” – Part III
- Good Things in Little Packages
- Mind Your P’s
- Lies, Damned Lies, and “CFD Comparison Charts” – Part II
- How many frogs does a horse have?
- Lies, Damned Lies, and “CFD Comparison Charts” – Part I
- It's a wireless world! No it isn't.
- Measuring the Real World
- July, 2010
- Are you using 'Smart' in a way I am not familiar with?
- An Interview With... Clemens Lasance
- Minimizing Late Stage PCB Respins
- I was led to believe we'd have flying cars by now
- Mechanical Analysis Products Now in Mentor’s Higher Education Program
- Red Hot Electronic Thermal Analysis?
- Avoid Late Stage Design Changes
- Sony Vaio laptop in mass 'recall'
- June, 2010
- Data Center Airflow Management: Part I
- Zero to 60 ... in 8 weeks?
- The art of modelling using CFD. Part VI - Peripheral Boundary Conditions
- LEDS and Reliability
- Using Social Media for Engineering
- Olympians and Engineers - the Right Mix
- Nearly Back to Business as Usual
- The art of modelling using CFD. Part V - Grid
- May, 2010
- Quick, Easy, and Mostly Right Heatsink Performance Estimate
- The Chilling Facts
- The art of modelling using CFD. Part IV - Fans
- Using LED Lights for Data Transmission
- The art of modelling using CFD. Part III - TIGs
- CFD in High Schools
- The art of modelling using CFD. Part II - Grilles
- The art of modelling using CFD. Part I - What happens if you cross art with science?
- April, 2010
- ¡Arriba! ¡Arriba ¡Ándale! ¡Ándale!
- Not Another Volcano Story...
- The Debate about Liquid Cooled Data Centers
- Mission Impossible: Finding 100W Light Bulbs
- How much do ‘U-Value’ good thermal insulation? Part VII - "Ooo, shut that door"
- "A Faster Horse" - Mentor 'IDEAS for Mechanical' driving product development
- MicReD Technology Wins Highest Technical Honor
- Living in a Wireless World
- March, 2010
- Dad, the AV guy
- Sticking Plaster and Light beats Skin Cancer
- Leaky Valves May Delay Space Shuttle Launch
- IBM Work to take Moore's Law to 2025
- Commercial CFD Starts Here ...
- It's Spring (almost)
- Concurrent CFD Explained (Part IV)
- How much do ‘U-Value’ good thermal insulation? Part VI - revenge of the radiative heat flux
- Roundup of SEMI-THERM, FloTHERM IC launch and JEDEC
- IC package representation is central to Electronics Cooling
- The Next Industrial Revolution
- How much do ‘U-Value’ good thermal insulation? Part V
- February, 2010
- Semiconductor Package Thermal Characterization and Design as Easy as 1-2-3
- Modelling and Simulating a PC with FloEFD (Part 3)
- Gain Insight the Faster Better Way
- NEW ElectronicsCooling Magazine Website
- How much do ‘U-Value’ good thermal insulation? Part IV
- Liquid Cooling – Are We There Yet?
- EDN Hot 100 Award for FloEFD
- Foresight and X-Ray Vision or Hindsight and Regret?
- The Secret of Doing Fewer PCB Respins – Unveiled!
- Vision at 40+
- "Sledgehammer CFD" - The Best Approach?
- How much do ‘U-Value’ good thermal insulation? Part III
- FREE Exhibition at SEMI-THERM, Santa Clara February 23-24
- How much do ‘U-Value’ good thermal insulation? Part II
- Force Prediction with Concurrent CFD
- January, 2010
- How much do 'U-Value' good thermal insulation? Part I
- The recession is over, finally
- Keeping the caveman warm - HVAC blog
- FloVIZ, the free FloTHERM/FloVENT CFD results viewer, try it, it's free
- Champcar Exhaust Analysis
- Quantitavely Seeding the Flow
- Concurrent CFD Explained (Part III)
- 'Heat Trees' - taking a leaf out of natures book
- May the force be with you
- Interested in Indy Car?
- The Most Extreme CFD Model Ever Ever - Explained
- Concurrent CFD Explained (Part II)
- FloTHERM and its new XML neutral file format
- The Most Extreme CFD Model Ever Ever
- Beginning at the Beginning ... 5 Parabolic or Elliptic? - Or Somewhere In Between?
- Concurrent CFD Explained (Part I)
- So, you want to predict component temperatures do you? Part VII
- Winter Woes and LED Lights
- December, 2009
- Cooking with CFD
- So, you want to predict component temperatures do you? Part VI
- Cool Youtube Video
- The Secret’s Out!
- Hands Free CFD
- Why Thermal Characterization of Integrated Circuits (ICs) is Important?
- Wanna Know a Secret?
- So, you want to predict component temperatures do you? Part V
- Thermal Design: Who’s Job Is It Anyway?
- November, 2009
- Free Visualization Software for Effortless Communication
- More on Concurrent CFD in Product Design
- Optimizing Flow Fields with CFD
- Concurrent Design and Thoughts on 'Flows'
- A trip to MPH and Top Gear Live
- Mind Your Head(room) Again
- So, you want to predict component temperatures do you? Part IV
- Solutions Expos – Going MAD in the UK
- So, you want to predict component temperatures do you? Part III
- Mind Your Head(room)
- Advection-Is that what you call it?
- Where do you fit in?
- Beginning at the Beginning ... 4 The Birth of Practical Turbulence Modelling
- October, 2009
- Solutions Expos – Just Back
- No Substitutes, please!
- Web slashes and missing polar ice
- Going MAD at European Solutions Expos
- Process Compression is Queen
- So, you want to predict component temperatures do you? Part II
- Optimization Or Else
- So, you want to predict component temperatures do you? Part I
- What’s black and stuck on a PCB?
- Time Compression is King
- Underfloor Thermal Insulation; Why? Part III
- September, 2009
- Once a Petrolhead ...
- Underfloor Thermal Insulation; Why? Part II
- Exercising my DVR
- Can MCAD designers really do CFD?
- Underfloor Thermal Insulation; Why? Part I
- Try the latest thing in CFD – Free!
- Moore's Law, still holding true
- How to survive a recession
- No one taught me about heat spreading...
- Don't Mistake "Simple to Use CFD" for "Simple CFD".
- Sometimes things just have got to work
- Contemplate This
- ‘Simulating and Optimizing’ – A Series of Free Web Seminars
- CFD ... Or Not CFD?
- Is all Software Rubbish?
- August, 2009
- Thermatronic Stagnation (nothing to do with male deers)
- Fluid Dynamics = Fun (Just back from holiday!)
- Beginning at the Beginning ...3 Improving SIMPLE
- Fractals: Gods Artwork, Part III
- Thermatrons Must Leave
- It's a Struggle
- Unexpected conduction paths and transient analysis in FloTHERM
- Engineers Spend 60-80% of Work Time Changing Existing Designs
- Modelling and Simulating a PC with FloEFD (Part 2)
- Beginning at the Beginning ... 2
- Hayfever: Stopped by a Red Light
- Betz' law- no ifs, ands, or buts
- No more old fashion light bulbs in all EU countries
- 40x Performance Enhancement
- Free Thermal Management Design Guide
- July, 2009
- At the Speed of Heat
- Where did CFD come from?
- Beginning at the Beginning ... 1
- Thoughts about heat, sinks, and the stuff between
- A Load of HVAC TLAs
- Brighter Shade of Green
- Formula 1 and KERS
- How-to: Invert your thermal model to good effect
- Invert your thermal model to good effect
- Did you know you're doing quantum physics with the Flo* products?
- Mind Your Thermal Management To Improve Reliability
- Lower the Surprise Ratio
- Make FloEFD the way you want it!
- "I work with computers"
- Designing In The Box
- It’s Magic
- Effusivity Smackdown or Standoff?
- Tennis – it’s a rough sport
- Fluid Dynamics and BBQs
- “All models are wrong, but some are useful” Part V
- The Heat is On
- Wimbledon's Centre Court Roof
- June, 2009
- TIM2 Performance:Heat Spreading in Action
- Show Me the Money
- Air – Is it Running Out of Gas?
- 3D Electronics Cooling CFD, with FloTHERM, in Pictures
- The Lights... they're so beautiful
- The Deal with Electronics Cooling CFD - Meshing
- Design Sensitivity
- Wanna Dance?
- “All models are wrong, but some are useful” Part IV
- Bringing CFD to the Design Engineer
- X-Ray Vision Glasses … Here and Now
- The Deal with Electronics Cooling CFD: Geometry (Lots!)
- Motorbikes, racing & CFD- Part 2
- “All models are wrong, but some are useful” Part III
- Modelling and Simulating a PC with FloEFD (Part 1)
- Thermal Design Stages-Phase 2
- What’s the Deal with Electronics Cooling CFD?
- May, 2009