I am still interested though, I want to see where this is going (not only what some industry decided to do)
Industry practice varies.
When I started at Radyne, they were using 20+ year old Inductoheat designs with wide open loop switching. Literally a TL494 going into drivers, except not using a TL494, but drawn out with discrete logic (gates, timers) on a 10" wide PCB (with tons of free space, but still dozens of chips). No desat protection: each gate drive was an opto, a linear supply (transformer, rectifier and filter cap), and an H-bridge (using generic MOSFETs) to drive the IGBT module gates. So, +/-15V and a modest number of amps.
When I left, they were using new controls with gate drives using SMPS power, planar transformers, magnetic isolators (ADUM series, it was a few years later that alternatives arrived from other manufacturers), IXDD614CI drivers, and desat protection, acting within a couple microseconds.
In the first week I worked there, a technician had made a wiring mistake inside a cabinet, I think a 300kW model. He turned on the power, hit the "start" button... KABOOM!
When I left, there was probably a dozen events (that I was aware of) where a unit had been started, and it just... clicked. Press the "start" button, "tick". Huh. Press it again, are you sure it's working? "Tick." Hrm. Maybe it's... did we wire this one right? Oh... oohhhhh.
That includes the intentional events, where we looped a fine Rogowski coil inbetween the IGBT bus bars and observed the transient. A nice smooth ramp up to 6kA or so, then off within a couple microseconds. Just as expected.
Only popped one IGBT module throughout development, and that included a large (~1MVA) research unit.
Is there joule estimations on what can fuck up a fresh IGBT? Possible to derate on heat and put slightly inadequate bypass (I assume you mean average junction temperature and not a joke)?
IGBTs are often rated for short circuit of some duration, usually 5 to 20us.
As in my above example, take 10us * 700V * 6000A = 42J. For a 1200V 600A IGBT (not a single die, but a few arrayed, I forget how many). It's basically die heating, same as desat protection on MOSFETs (a slightly less urgent but nonetheless applicable protection method -- we did this also on the smaller power supplies, at up to 400kHz), which are sometimes rated for short circuit durations, and almost always rated for avalanche energy which has a similar limit (die temperature).
In any case, the limiting factor is die temperature, so use the transient thermal curve to find out.
Tim