Interesting IGBT failure

[TMaxElectronics]

It seems that my bad luck with IGBTs struck again, but not in the way you might expect

I was testing the small coil (https://highvoltageforum.net/index.php?topic=2394) when suddenly there was a pop and the breaker tripped. Power was only around 10kw so not fill tilt, but I was running from rectified three phase turned down on my variac to ~280V on the bus. After taking apart the coil I was surprised to see none of the IGBTs blown apart and no shrapnel anywhere in the case. Although three Transistors definitely died and shorted out.

With a bit of searching I finally found what made the pop noise:


So from what I can tell this is what happened:
When installing the IGBTs I accidentally mis-aligned the insulator sheet by perhaps a millimeter to the left. Certainly not enough to cause any immediate problems with shorting out. I then put a little too much thermal compound underneath the IGBT and some squeezed out by the side a little, but once again not enough to immediately cause trouble. The first few coil runs never ran at more than a few kw continuously (that 20kw run I posted lasted maybe 3sec before the breaker tripped?) so the IGBTs never really had time to warm up.

During the last run however I had much more power to play with because of the three phase supply, so I ran the coil at those 10 or so kw for quite a few minutes. The now hot thermal compound must have gotten a bit softer and dripped towards the heatsink.
Finally, the (slightly conductive!) thermal compound bridged between the non isolated heat spreader and the earthed heatsink. The plasma that formed expanded enough to touch the gate of the adjacent IGBT killing it. You can actually see a small shiny spot on the gate pin on the second IGBT from the right. That of course also killed the two IGBTs on the low side of the bridge when those switched on. Interestingly enough the parallel IGBT did not suffer damage (at least its not shorted), probably because the adjacent one dying made it take all of the current.

So remember to not use too much thermal paste everybody

[AstRii]

Though luck :x

I always thought thermal paste were non conductive. Are you using some special one? I am using the same thermal pastes which are used for computer CPUs and never had this problem.

[flyingperson23]

Why not make the insulating sheets significantly oversized so this isn't a possibility?

[davekni]

So remember to not use too much thermal paste everybody

Yes.  Had a similar problem twice at work with 60V FETs.  Heat sink is live (to FET drains), so only thermal compound, no separate spreader nor insulator.  Excess thermal paste bridged up to gate pin.

Interestingly enough the parallel IGBT did not suffer damage (at least its not shorted), probably because the adjacent one dying made it take all of the current.

Interesting, but makes sense.  When my QCW bridge failed due to excess transient heating (running in phase shift mode), all four IGBTs of one half-bridge failed.  In my case, whichever IGBT of a paralleled pair failed first, other die was likely so hot it was on the verge of failing too.

I always thought thermal paste were non conductive.

The common old white thermal compound was not conductive.  Perhaps powdered alumina filled if I recall correctly.  Newer gray or black thermal compounds have much higher thermal conductivity, but are filled with powdered silver or graphite, so electrically conductive.  Now I find thermal compounds available with thermal performance as good as silver-filled (>8W/mK) but electrically non-conductive.  Perhaps filled with AlN powder?  We just switched to one of those for prototype builds at work to avoid previous failures due to excess compound.

[TMaxElectronics]

Why not make the insulating sheets significantly oversized so this isn't a possibility?

Well... laziness xD Didn't want to use two pieces of the insulator sheet I cut the smaller sections out of so I just cut it in half, that ended up being exactly that width.

I always thought thermal paste were non conductive.

I'm using arctic mx4. I've used various types of CPU/PC type compounds and so far only come across conductive ones actually. Arctic MX2 is definitely confirmed to be conductive, that also already killed a circuit for me.
Interestingly enough the compounds aren't a dead short or even just a high resistance. The failure of one of my LED drivers (the one with MX2 in it) happened after quite a while and there the thermal paste directly contacted multiple exposed traces. Only after running for a while did the HV side short to a logic level track and kill the controller. Perhaps the conductive particles inside don't always make enough contact to conduct?

Oh I just remembered seeing ads for non conductive thermal compound? I think its designed for use in GPUs where you really have to use an excess of paste.

In my case, whichever IGBT of a paralleled pair failed first, other die was likely so hot it was on the verge of failing too.

I think I will end up replacing that one anyway though, as it would have potentially seen some quite high voltages on the gate. After all the gates both used the same winding on the GDT so during the transient there was effectively just a 20 Ohm resistor in series with the HV on the other gate. Guess the TVS diode caught that but I wouldn't bet anything on the IGBT not having died a little ^^ Better not risk it dying later and taking the new replacement IGBTs with it.