Sign In
Forgot Password?
Sign In | | Create Account

So, you want to predict component temperatures do you? Part I

Prediction of component temperatures is central to electronics cooling simulation, the management of their temperatures  is central to electronics cooling design. Either way, heat is dissipated inside a component, component gets hot, if component gets too hot, component stops working. When creating an electronics cooling simulation model the question of how to thermally represent the components is key.

The vast majority of electronics cooling simulations performed today use a technology called CFD, computational fluid dynamics. A 3D CAD type definition of the components, PCBs, heatsinks, chassis, fans etc. are input. The equations that predict temperatures and air flow distribution are solved by clever number crunching executables. Finally the predicted temperatures and air flow can be examined in both graphical (CFD: color for directors) or tabular formats.

temp2

Why is there any question about how to model components, surely you would just define them as they exist in reality? It all comes down to data availability. Components (packaged silicon normally) are quite complex, incorporating many different parts, with different materials, from bond wires to lead frame, from a spreader to the die itself. Having to manually create all that for EACH component would be madness.

So, why not get such a description from the component manufacturer? Well, they tend to be quite secretive about the internals of their packages. There’s a lot of IP bound up in there, last thing they want to do is to provide a full 3D physical description of it to all and sundry.

Where does this leave the poor thermal engineer? With little data and pressing needs, a number of different component modelling methodologies have emerged over the last 20 or so years.

For this first part of this blog series we’ll cover the most common, a lumped block representation. A 3d homogeneous block with a single material.

pqfp_block

In terms of data, the one thing that is available is the package footprint size, this is required for routing and manufacturing design (as is often the case with the more unique disciplines, thermal eats the crumbs that fall from the giants table). If you’ve got a footprint shape then the only other thing you’ll need to make a 3D model is the height.  The component height can itself be quite rare,  maybe less nowadays as component libraries become more refined and the need for mechanical interference checks becomes much more common place.

In reality a component will have a range of temperatures throughout it. The die being the hottest, the peripheral corners being the coldest. Components will be specified to work up to a maximum junction (die) or case (usually middle+top) temperature. How on earth can we get a range of temperatures out of a component model that is just one lump? Well, you can’t. Why not at least dissipate the heat in that block where the die actually is? Well, don’t bother, even if you did know how big the die was there is still no advantage in accuracy as you’re not modelling all the other heat flow paths in any detail. You can’t be half accurate when trying to be accurate. You’re just going to have to lump it, literally, and spread all the heat throughout your block representation.

library2Last thing required is a material definition. A 3D thermal simulation requires that all solids have a thermal conductivity defined (to enable a steady state thermal prediction) and density and specific heat (to obtain the thermal capacitance) in addition if a transient thermal simulation is required. Well fear not young thermal engineer. We, your preferred vendor, have created a library of ‘typical lumped packages’ materials with values that will result in typical case temperature predictions when modelling your components as lumped blocks. If you buy the best you get the best, even when it helps you create a more accurate block.

Next time you see some nice piccies of an electronics cooling model, check out how the components are modelled. More than likely they’ll be blocks. Not the best approach by far, in fact it is the least accurate but easiest to define. How inaccurate? Hard to say as it varies a lot. On average I’d say between 10-30% error on case temperature rise with no indication of junction temperature at all. Not bad considering what you’re not representing.

9th October 2009, Ross-on-Wye

Component, CFD

More Blog Posts

About Robin Bornoff Follow on Twitter

Robin BornoffRobin Bornoff achieved a Mechanical Engineering Degree from Brunel University in 1992 followed by a PhD in 1995 for CFD research. He then joined Mentor Graphics Corporation, Mechanical Analysis Division (formerly Flomerics Ltd) as an application and support engineer, specializing in the application of CFD to electronics cooling and the design of the built environment. Having been the Product Marketing Manager responsible for the FloTHERM and FloVENT softwares he is now Market Development Manager for the Physical Design of Electronics in the Mechanical Analysis Division. Visit Robin Bornoff's blog

More Posts by Robin Bornoff

Comments 11

Post a Comment
"So, why not get such a description from the component manufacturer? Well, they tend to be quite secretive about the internals of their packages..." Ahem. If it's NXP MOSFETs you're after, then for several years we've made detailed models available for most of our devices. Used to be downloadable from smartparts3d, and now available direct from the NXP website. Regards...

Chris Hill
10:19 AM Oct 12, 2009

Always the exception that proves the rule! I'm going to be getting on to 'detailed models' at the end of this blog series. But yes, NXP is a rare exception in that they do supply (detailed) models of their packages. It would be interesting to know what it is about NXP's view of their MOSFETs that results in them supplying them, i.e. why no IP, why is it that there is nothing to hide? Is it the relative simplicity of the packaging style? What I'm getting at is why does NXP supply such packages and hardly anyone else does!!!

Robin Bornoff
10:50 AM Oct 12, 2009

About 20-15 years ago the most experience people had was to use 10 W/mK for a lumped rep of a component. Over the years, with an increase of application of CFD to electronics cooling and the subsequent experience, the approach has been refined. Even today there is a 'hierarchy of accuracy', from block models all the up through compact thermal models to detailed models. In terms of recommendation I recommend detailed models (do you use FloTHERM.PACK http://www.flopack.com ?), but that was not your question! If you can not find any more info and so have to model the component as a block I would advise the 'Typical lumped packages' materials library. This is a refinement on the 5-15 W/mK range, based on some detailed vs. lumped characterisation work we conducted. The values of k have been calibrated to provide block predictions of case temperature and better reflect the types of lower conductivity packages found today compared to the types that were around 20 years ago. For a bit more info please check out: http://www.mentor.com/products/mechanical/multimedia/pcb-thermal-simulation-the-state-art_reg.cfm?int=1&id=3d15bc1d-fe24-457c-88e8-9859681f8bbd

Robin Bornoff
9:09 AM Nov 6, 2009

[...] to make creating package thermal models easy. I’d recommend you read Robin Bornoff’s blog ‘So, you want to predict component temperatures do you?’ for an on-going discussion of package thermal [...]
[...] current blog series titled “So, you want to predict component temperatures do you?” is a big chunky theme that will require quite a few more parts until completion. As a break I [...]
Can you mail the “PCB Thermal Simulation - The State of the Art” to me? formelong@gmail.com Thank you!

long mi
1:23 PM Dec 14, 2009

Some value of "Typical lumped packages" are so low such as ChipArray=0.1,TSOP=0.1...most of that are 0.x.But as we know the Flotherm commended that plastic package=5,ceramic package=15.Both the value are bigger than most of package in "Typical lumped packages".so why and what method is better?

long mi
3:24 PM Dec 14, 2009

In the sliding scale of model accuracy and representation any block method is going to be less accurate than other better methods of representation (e.g. thermal resistor network or detailed). The 'Typical lumped package' materials are preferable to the typical ceramic or plastic materials that are a throw back to earlier times. Best advise: use FloTHERM.PACK or hassle your component supplier for good quality characteristic thermal data.

Robin Bornoff
3:04 PM Dec 17, 2009

This can be accessed via: http://www.mentor.com/products/mechanical/multimedia/pcb-thermal-simulation-the-state-art

Robin Bornoff
3:16 PM Dec 18, 2009

[...] a range of methods available to thermally represent IC packages that I covered in this previous blog series. The derivation of such representations is the thermal characterisation obligation of the suppliers [...]
[...] a range of methods available to thermally represent IC packages that I covered in this previous blog series. The derivation of such representations is the thermal characterisation obligation of the suppliers [...]

Add Your Comment

Please complete the following information to comment or sign in.

(Your email will not be published)

Archives

 
Online Chat