IGBT vs MOSFET for dummy load

[Kleinstein]

There can be some scattering in when the parts will actually fail.  However there is little chance to find a part that is much better - more like a small chance to find some early failing parts from even for types that normally have a good SOA.

The process of thermal instability formation can be described the following way: 
We assume a given local heating to start the process, that will cause a temperature peak. The higher temperature will concentrate (in a device prone to instability) the current to that hot spot which than causes additional local heating. If this extra local heating is stronger than the initial heating assumes in the process, the system will likely run away to destruction unless possibly other nonlinear effects to stabilize it at a higher temperature state. So the thermal instability will not need some initial faults or inhomogeneity to start - it is more like feedback in an oscillator, one high enough it will start to oscillate (and is usually limited in amplitude by nonlinear effects).

So no chance find some magical much better samples, more like a chance to have some faulty ones that at some point's are even worse.
There is not magical limit to always allow some < 3% or similar in the linear region. It depends on voltage (higher voltage is usually worse) and the device type.

When going towards saturation there are 2 effects that help. One is less heat from a given increase in current and the second is that in many devices there would be no more current localization as the TC changes sign.

[duak]

I ran across this paper when searching for "electro-thermal instability IGBT" https://www.nrel.gov/pv/assets/pdfs/2015_pvmrw_131_das.pdf  Pictures on p.29 shows localized heating.

There are quite a few papers on the subject although not too many address linear operation specifically.

Cheers,