IGBT full bridge issues

[TmaxElectronics]

I have an IGBT full bridge as the primary driver in a tesla coil, but am having some, let's just say issues with it.
The gate drive ICs i am using (FAN7390) keep dying and i don't know why.
The circuit i am using is a one to one copy of one in an induction cooker, where it worked flawlessly for five years.

Also the IGBTs (IXXN110N65) blew up for seemingly no reason at all, i have got a current limiting circuit in the controller that was set to stop excitation at 150A and was operating at a pulse width of only 20us, so there should not have been thermal issues. However i think that this might have been caused by a gate driver failure causing shoot through, the resulting explosion then pushed 350V down the gate driver output, blowing it up and spiking the control lines to cause shoot through on the other side as well.

I am also aware of the very common drive circuit by Steve Ward but decided not to use it as i wanted to learn by designing one from the ground up.
Attached are pictures of the gate driver output waveform (keep in mind that the frequency is NOT the actual drive frequency when operating) and the layout of the driver PCB.

Are there any issues with the design that could cause such trouble, or is the design perhaps flawed altogether?
Sorry btw if this post is hard to understand, i am quite honestly not sure how i should explain what is going on properly :/

EDIT: i have already made two changes to the design since: i have increased spacing between low and high side traces and added clamping diodes across the outputs of each channel.

[T3sl4co1l]

Show the connection to, and layout of, the H-bridge.

Tim

[Etesla]

Nothing looks wrong at a glance. In general I think it's difficult to determine weather the IGBT's or gate drivers failed first. One failing can easily cause the other to fail. Have you done a test where you run the circuit with dummy capacitors around 5000 pF in place of the IGBT gates and see if the drivers fail on their own?

[TmaxElectronics]

Show the connection to, and layout of, the H-bridge

Unfortunately i only have the attached image, as the destruction was rather violent and it disassembled itself  , because of those jucy capacitors in the foreground (the bridge with the IGBTs is in the background).
I hope the image is shows enough.

The length of the chinch cables used for the gates is quite significant, but I couldn't think of a better solution for connecting the drivers.

I will try the drivers out with a dummy load once new ones arrived, the old ones are now a part of the atmosphere.

Also I will try to mount the capacitors directly onto the primary bridge power rails next time to reduce inductance, I didn't have usable spacers yet.

[T3sl4co1l]

BWAHAHA RCA cables, is that what they are!

But anyway yeah, you've got a good probably 300nH supply inductance, about 100 of which is between the transistors.  Not clear if you have other ground paths between DC link and control.

Check the DC and transient voltage limits on that gate driver, particularly between the bootstrap common and VSS.  Consider switching speed i.e. dI/dt with respect to that inductance.

Oh yikes, and a big fat output cable running right over all the control signals, that can't be helping either.

If you're looking for a slam dunk:
Make a 4-layer PCB that screws onto the transistors.
Put DC+/- on inner layers.
OUT1 and OUT2 on bottom.
Gates on top.
OUT1/2 can be taken out to screw terminals without having to cross anything.  Use an inline current transformer maybe.
Likewise DC+/- can be poured in the middle and taken out to screw terminals without having to overlap the drivers.
Use ground plane under the drivers.  Probably use somewhat larger capacitors, too.
Use fat wide pours for all transistor connections, including the gates.
Add a few film caps to bypass the supply, 1-4.7uF, e.g. TDK B32672 family.
Keep the Kelvin gate connections, don't short the two emitter terminals together.
Put the gate drivers along the board edges, within a few cm of their respective transistors.

This at least has the possibility of getting stray inductance down below 20nH, and eliminates the resistance and inductance of the gate driver cables.  You can use smaller gate resistors (mfg recommends 2.2 ohms; plots up to 10 ohms).  The board cost will be only a few times what your existing control board was, probably the same or less if you include the labor of cutting and shaping the bus bars.

Can also consider TO-247 or TO-264 transistors, soldered into the board.  Cheaper than SOT-227s.  Tied to planes at the leads, the stray inductance is about the same either way, no performance loss.  Obviously, if you already have a tube of the latter, this doesn't help much, that's fine.

Tim

[TmaxElectronics]

BWAHAHA RCA cables, is that what they are!

Yeah, honestly i used them because i had the PCB mount sockets laying around.
I did initially consider the PCB idea, but since the currents could still go up to ~500A, I didn't.
I think i will use a circuit board for just the gate driving and keep the bus bars (which are actually just flattened copper pipe, didn't have the money to also buy proper ones).

Can also consider TO-247 or TO-264 transistors, soldered into the board.

From the way i understood it, the TO227 packages have higher terminal and bonding current limit that the TO247s, right?

[T3sl4co1l]

With the switching speed and current ratings, you aren't going to be running it continuous either way.  Bus bars don't matter very much.  And PCB traces can be wider, and they make 4oz+ if you still don't feel comfortable with it.

Incidentally, make sure to custom order a more evenly spaced stackup (e.g. 3 x 20 mils dielectric).  Standard proto will probably handle high voltage on the 5-10 mil dielectric, but it's rather marginal.  Eh, I guess that'll more like quadruple rather than double the cost, but still, it's an easy way to get something with high confidence.

Tim

[profdc9]

If you're interested, I have PCBs for my Tesla coil at

https://github.com/profdc9/DRSSTC-PCB-Pack

You may be specifically interested in

https://github.com/profdc9/DRSSTC-PCB-Pack/tree/master/full-bridge-multi

The layout minimizes inductance and places the DC link capacitors in between the two sides of the bridge.  I also used a "woven" connector for the DC bus, with three sets of positive and negative connections intertwined to minimize the loop size.

There are also DRSSTC drivers based on Ward's design, interrupters, etc. in the github.

[TmaxElectronics]

Have you done a test where you run the circuit with dummy capacitors around 5000 pF...

The replacements just arrived and i gave them a hard time, driving 4.7nF with 24V input at 300kHz (over 3 times the expected frequency) and they survived without issue.
Temps got a little high (~ 70°) but the chips were totally fine.

One thing i forgot to mention is that the drivers only started to fail once mains voltage was used, for initial testing (and safety while doing so) is used an old amplifier transformer giving ~110V across the bridge with which the drivers were fine (I only ran it for maybe 10 mins. but with mains they died in less than one).
But i don't know how the higher voltage would affect the drivers enough to destroy them, they are rated for 600V bootstrap voltage after all.

[Phoenix]

I have a few points in no particular order:

1. Your bootstrap diodes need to be high speed diodes (1N4007 are slow diodes). The reverse recovery of the diode will prevent it turning off and blocking the discharge of the bootstrap, and possibly causing other problems. UF4007 is an easy replacement.

2. Is your controller providing appropriate deadtime between the high and low side switch operation?

3. The gate drive waveform you've provided appears to be without any DC voltage applied. It only shows that the drive chip is doing something sensible, but nothing about the quality of the gate drive.

4. The general quality of construction leaves a lot to be desired, and as others have said could cause all sort of problems too.

5. Depending on the cable lengths and IGBT you might need a negative drive rail to prevent turn-on shoot through problems too. You can't bootstrap this.

6. Are you expecting to switch those IGBTs at 100kHz? That's not trivial...

[moffy]

I agree with T3sl4co1l. The only point I would make is that the driver chips need to be right next to the IGBT's. Also the drive waveform shows quite a slow RC component. Maybe dropping the value of the 33 Ohm gate resistor, and checking that the drivers have enough current output to drive the IGBTs quickly. What does the power supply waveform to the drivers look like?

[moffy]

Just one last point, with the gate resistors so far from the IGBTs, would they be prone to parasitic VHF oscillations like MOSFETs are? I know this can be destructive of the MOSFETs.

[T3sl4co1l]

Probably not; the resistance of the cable, and the 33 ohms terminating it (not exactly matched, but a far sight better than a completely open stub), should suppress that mode alright.

I've not seen IGBTs oscillate quite that fast, though I also haven't gone out of my way to try.  I would guess it's safe to assume the MOS contribution to current flow can change as quickly as in any other MOSFET of similar generation, so the potential is the same.

Tim

[moffy]

I looked a little online and they appear to have both a low frequency and higher frequency mode of oscillation, perhaps a contribution from the Bipolar and MOSFET components. It didn't appear as bad though as with just MOSFETs.

[TmaxElectronics]

Thanks for the help guys, I will redo the design once I am done with my exams

I also realised I kind of forgot about dead time     
In the small coil that the primary driver design came from I kept reducing the dead time further and further to increase the maximum drive frequency. At some point I switched to an isolator which had no dead time at all, which was kind of ok for the small-ish currents I guess, but obviously the larger IGBTs didn't like it and just shot through.
I do feel kind of stupid for forgetting about that

[moffy]

Yeah, it's kind of embarrassing when you forget something important  We've all done it at one time or another.