The accidental thermal engineer – Simulation and reality

The accidental thermal engineer – Simulation and reality

A large part of thermal analysis work is concerned with carrying out ‘what if?’ analyses. For example, if I have a PCB with dissipating components on it I would probably like to know what temperatures those components will run at. But what if I add more components? What if I mount them closer together? What if I add a cooling fan? What difference does any of this make?

Similarly, as a MOSFET manufacturer, we might sometimes want to know the thermal consequences of varying a package design or changing the types of materials used in the package and so on.

Of course, it is possible to answer these questions by building and testing actual prototypes, and at some stage in the design process we will have to do this anyway. But in the early stages of a design or feasibility study it makes much more sense to carry out the investigations using simulation software, where changes to a configuration can be made easily and analysis of results viewed quickly. The thermal simulation package that we use for this purpose is Mentor Graphics’ FloTHERM® (other simulation packages are available), which allows thermal models to be built and simulated in a 3D CAD-style environment.

An obvious question at this point is: do we trust the simulations to give us reasonably accurate answers? I would say ‘yes’ for two reasons:

  • the physics underpinning the simulations is well understood, and
  • the people writing the software presumably know what they are doing.

So, assuming we use the software properly, we should be fairly confident of the results. Even so, from time to time it’s still good to carry out some calibration or ‘sanity check’ exercises, and it’s an example of one such exercise that I’d like to share in this post.

A real-life example

For this exercise, identical MOSFET devices were mounted on PCBs of the same overall size but with varying amounts of copper coverage. See Fig. 1 and Table 1.

Mosfetnew

 

board

 

board2

 

For each PCB configuration the MOSFET device was powered with a constant 1 W and allowed to reach a steady-state temperature. Each MOSFET had a calibrated on-die temperature sensor which allowed us to take extremely accurate measurements of junction temperature (TJ) in the steady state. The ambient temperature during each test was also recorded. The four PCBs were then recreated in simulation, with die dissipation set to 1 W, and the steady-state temperatures recorded. The real and simulated results are recorded in Table 2, expressed as thermal resistances as described in the JESD51 series of JEDEC standards.

table

 

So how did we do?

Well, the A, B and C boards came very close to the measured result, but G still needs a little work. We must remember that not all the simulation input parameters can be known with absolute certainty – this being particularly true of material thermo-physical properties.

It is also worth remembering that our measurement of reality may not be perfect: all real-life test equipment has some degree of error associated with its measurements. There isn’t enough space in this post to explore these potential errors in detail, but it could certainly be the subject of a future blog post!

Next time: how simulation can help us look at what goes on in a device die at very short timescales.

 

Christopher Hill
Christopher Hill
With over 20 years experience as an applications engineer specialised in power MOSFETs, Chris has dealt with a multitude of power semiconductor design-in challenges. One of the recurring themes in these challenges has been ‘thermal’, and therefore he has spent much of his working life immersed in questions of a thermal nature. He has authored numerous conference papers, magazine articles and application notes on a variety of power semiconductor topics. In his free time he likes nothing better than plodding through muddy fields in the North of England.

3 Comments

  1. Avatar Ewold van Geffen says:

    So how does the result now support the story that simulation can do a good prediction? I find the explanation for result G slightly unsatisfactory. I would have expected to go over the simulation again adapt the model for board G and then find a result that is within 5 degrees difference for this simple example.
    I am left with some questions
    Was radiation switched on in Flotherm. ?
    What about the surfaces finish, was polished or smooth copper selected in Flotherm (poor radiation properties) ?
    Did you add in simulation that the top and bottom copper have solderresist, because that improves the radiation properties of the copper surfaces is wat we found.

  2. Avatar Chris Hill says:

    Hello Ewold,

    Firstly, thanks for taking the time to read my blog and post a response.

    Due to space limitations I was not able to further explore certain aspects of the story that I posted recently, although I may be able to return to some of those aspects in later blog posts. The points that you raise are certainly related to some of those aspects.

    As you may already know, one of the hardest parts of building a thermal simulation is finding reliable thermo-physical data for the materials which you wish to include in the simulation. Such data would include thermal conductivity, density, specific heat, surface emissivity/finish (which you mention), etc., etc. For some common materials, such as copper, this data is already well known but for others, such as plastic or PCB FR4 material, the data may be unreliable or perhaps not available at all. In those cases, the only option you have is to use estimated numbers from similar materials. You can then try altering your estimates by perhaps +/-50% to see if there is any effect on the temperature results which you are interested in. In some cases, if you are lucky, even though you may not know precise data for a material, it might be that the particular parameter does not appear too critical anyway. Unfortunately, we are sometimes forced to take this approach. Incidentally, an example of such an exercise would make a good blog post in itself – so thanks for the idea!

    To respond to your other points:

    I did explore the “G” result further and found that the FR4 thermal conductivity had a significant effect on simulated temperatures. I believe this is because the PCB copper area in this case was very small and so the “heat spreading” from the device tended to rely more on the thermal conductivity of the FR4. The available data for FR4 was a little vague, unfortunately, so some estimation was necessary as described above.

    Radiation: yes, this was switched on. In natural convection cases such as this it is important to include radiation heat transfer in the simulation. My experience is that, if radiation is switched off, then simulated temperatures may by >15% higher than when radiation is switched on.

    As regards the surface finish of the PCB and copper, I agree that this is important. We attempted to recreate the solder resist on our PCBs although again, not having specific data, it was necessary to use more general, published information and perform some sensitivity analysis as already described. In an ideal world we would like to know all the thermos-physical properties of our materials with a high degree of confidence, but in reality it is rare that this is completely possible.
    I hope this answers the points you raise!

    Best regards,

    Chris Hill.

  3. Avatar Pitt Chao says:

    Thermal Resistance Tester(TF130), to determine the thermal resistance, thermal conductivity, and CLO value of various fabrics, quilting products, and other heat insulation materials.

Buy now