An important point oft not appreciated is that the actual silicon die in the package can change its temperature very quickly in response to loading. The thermal time constant may be in the tens of milliseconds region. If a really excessive current is applied it will reach breakdown temperature (usually >200C) before there is any noticeable warming of the case or heatsink. Sometimes, faster than a fuse can respond.
What's more, note that the thermal time constant is much longer than the transient events in question: transistors die in 10s of microseconds, while the heat takes tens of milliseconds to leave the die (and hundreds to leave the package).
(Reminder that fuses take thermal time constants to fail, even under heavy fault conditions!)
For transient events much shorter than the time constant, the temperature can be modeled as a pure integrator: apply an energy pulse, and temperature rises in a step.
The volume being heated, is the junction itself: not even the whole die, and certainly not the base plate and packaging! This is very little material, hence why most avalanche ratings are only in the mid to upper mJ range.
Typical example, a schottky diode that might be rated 40mJ avalanche. In this case, the avalanche structure is the guard rings around the schottky junction itself. Only a perimeter is dissipating that energy, so very little of total die area is heating up.
Power MOSFETs tend to have higher ratings (100s mJ, large ones pushing over 1000mJ), partly because of the longer pulse duration and partly because they utilize more of the junction. (Vertical TrenchMOS has the tiny source, substrate and gate features tightly integrated on one face of the die, constructed edgewise (vertically), while the body of the die is the lightly doped drift region, where drain voltage is dropped, and avalanche current is dissipated. The backside of the die is strongly doped to form the drain connection.)
On that subject, I'm not sure why IGBTs aren't rated for avalanche (or very much, if ever, at least). They would be subject to much more... interesting breakdown behaviors, potentially including the rapid avalanche switching behavior that BJTs exhibit, and maybe inducing subsequent 4-layer (SCR) latchup (despite valiant efforts to prevent latchup even under fault current conditions). As far as I know, they just die when ratings are exceeded -- the usual explanation here is a hole burned in the die where all the avalanche current flowed, resulting in excessive off-state leakage.
On that note, the typical pinhole failure looks like a resistor in parallel with an otherwise okay transistor. I've caused MOSFETs and BJTs to fail in such a way that C-E / D-S looks like a resistor (usually ~ohms, sometimes 10k's ohms), but if you can apply base/gate voltage (without it being loaded down by excessive short-circuit current), the 'ON' region still works just fine.
One way to reliably destroy power BJTs: set up an avalanche pulse generator with a proportionally-sized pulse capacitor. Avalanche spreads out quickly in small-signal transistors (hence why the humble PN2369 can sink a couple amperes in this operation), but not in power transistors, which have wide junctions. Trying to deliver, say, 10A or more through that small avalanche channel, burns a hole in the die. Pretty reliably, switchmode and HOT types fail after one or a few pulses, with a final R_CE(off) ~ 40kohms.
Tim