I think you missed my point: the Vce[sat] for most IGBTs that are rated at 1700V or less is in the range of 1.2V to 3.0V, and it increases with Ic as one would expect for a bipolar device (that is, 60-120mV/decade of current, plus a bit more due to the various resistances), so triggering desat at 7V is needlessly abusive. For example, even if Vce[sat] is 3.0V at rated Ic, it will likely only increase to about 3.5V - 4V at 2x Ic, and maybe 4V - 4.5V at 3x Ic, etc., but once desaturation occurs - which could be anywhere from 2x to 6x rated Ic, depending on the precise construction of the die - Vce will shoot up rapidly, to the point that the device will act more like a constant current source. Triggering desaturation protection a couple volts lower than the standard 7V threshold isn't so much to grant the protection circuit more time, it is to reduce the peak power experienced by the switch. Besides, to make a reliable desaturation protection circuit you have to ignore Vce when the gate is first brought high for 200ns - 1us, anyway, to avoid false triggering.
If it's rated for 10us of short circuit, what's the difference? It still survives.
Lower and lower thresholds are more sensitive to errors, and prone to false triggering; and you want to avoid blanking and filtering schemes where possible.
That said, if the fault is cross-conduction of a half-bridge directly fed from a dc link capacitor then there is almost no hope of protecting the IGBTs, even if the Vce threshold is lowered to 5V, as fault current might very well zoom up to the desaturation-limited value even during the blanking time for the desat detection circuit, and then it's more a case of, "Abandon Hope, All Ye Who Enter," etc.
Whose IGBTs are hopeless?
That assumes this is a power supply employing current-mode control, when it could be a power supply using voltage-mode control with cycle-by-cycle overcurrent protection, or a PFC stage using average current mode control, or a motor drive also only using cycle-by-cycle current limiting, or a motor driving using average current mode control and relying on desaturation detection as well as a peak-detector on a Hall-effect transducer for overcurrent protection, a UPS running a fixed PWM waveform without a control loop at all - you may shudder at this kind of sloppy design but I've actually seen it in a commercial product, etc.
The predecessors to the designs I was working on, were basically wide open loop, old school, verbose logic designs. They didn't even use oscillators, they slapped together a bunch of monostables to do it. Lots of CD4000 or 74AC logic, on the premise that the higher voltage, or lower output impedance, provides noise immunity. Lots of opamps, for reasons I never quite understood (only a couple are really needed, for a full analog+logic design of that capability). The old guard laughed at our choice of FPGA with its 1.1V core; but go figure, they didn't know anything about EMC, they just cookbooked everything.
Point being, the gate drivers similarly were just that, a transformer to supply DC power, an opto to couple signal, and an IRF(9)540 H-bridge to drive the gate to +/-15V. No desat protection (would be hard to implement with that H bridge output, anyway), no current sensing, aside from the one current transformer (common to all inverters in parallel) for feedback (phase and amplitude, this was a resonant system). Controller terribly slow, dominant pole compensated, you could watch the dials as it started up (settling time in the fractional seconds).
So, needless to say, they burned a lot of transistors on those designs. But service is so expensive that they really don't care about transistors as such, and operation is usually so consistent (industrial process equipment) that they can get years of continuous operation despite the dumb design.
There was also the even older generation of SCR inverter units, which I had nothing to do with, but saw them come in for service from time to time. Think the controls were even more limited on those, but again they were only used on very stable loads so if it starts up once, and stops once, successfully, who cares...
Yep, this is a good plan of attack if your budget allows such a braces and suspenders approach (substitute a Hall-effect transducer for current transformer, where appropriate).
Yes, which since we're talking $200 IGBTs here, the added cost is quite small, and the savings (to the customer) in service calls is huge.
Tim