why do i always manage to kill IGBTs *_*

[TMaxElectronics]

As I've posted in another thread I'm currently building a small (30cm secondary) tesla coil. I've built the inverter exactly as planned here: https://highvoltageforum.net/index.php?topic=2394.0
Here's the problem though... the IGBTs do end up dying once I put the coil on mains voltage. Despite everything being pretty rosy all the way up to 300-325V Vbus.

Testing the coil with low bus voltage everything looks beautiful, gate voltages look good (tvs diode is clamping during low resistance switch off), phase lead is tuned nicely to just slightly lead the current waveform and IGBT temperature when running at a high test load is ok too.




There is some phase mismatch during the first few cycles though, which leads to fairly significant voltage overshoot during switch off.
This would make be beleive there is an overvoltage issue, but when I look at the waveforms that other people post here they are much further off ideal phase and their inverters don't seem to care
Also I must say I'm surprised that there seems to be that much overshoot. I have 40uF of foil cap on the six layer pcb. One layer each for bridge outputs and bus +/-...

The overvoltage is not all that significant in this scope shot, but when turning up to the full current (150A for the current setup with only one IGBT per position), the overshoot increases to around 80% (which feels like a lot to me? I'd have expected more like 30-40%). That however still only results in 630V (from 350V VBus), which the IGBTs should survive. Also I must admit, that I'm not 100% sure a tiny little overvoltage transient would result in enough power dissipation to kill the IGBT.

One thing I can't explain is the weird voltage curve (marked in red) both during switched off state and switched on state of the IGBTs. I'd kind of suspect measurement error from poor probing, but it does not seem to change at all even when I place ferrite plates right next to the probes. The inverer output and the primary coil is also around 25cm away from the probes and no significant current carrying traces are near them, except the bridge output itself. Also, since it's not sinusoidal, I don't beleive it would be caused by the primary current. In addition it isn't at all present in the gate voltage, which is probed in a slightly worse way. Maybe this is where the problem lies?

Tail current of the IGBT that previously switched off might still be present at this time, but I also doubt that would be the cause here, as it takes much to long for it to drop off. I'd expect maybe 1us tail current time as the absolute worst case for these fast IGBTs, and much much less when soft switching them like this. Diode reverse recovery current shouldn't be an issue either as (at least during most pulses) the lead time prevents commutation of the current to the wrong diode.

Bus Voltage seems to not be oscillating either (at least not like in the last inverter I built that kept killing its transistors). Though I do begin to suspect that a very low resistance, but relatively high inductance connection (like for example the two brass standoffs seperated by ~10mm I use) between inverter and bulk caps might be detrimental to inverter operation due to the high q of the created resonant circuit. I might try either adding some tiny resistance or making the connection between bulk caps and inverter itself laminar too. Nontheless, I still doubt that would be the issue here

Here's the measurement setup, all nice and close to the IGBT itself:


Oh and before I forget: The IGBTs are STGW30M65DF2 from ST https://www.st.com/content/ccc/resource/technical/document/datasheet/6b/6d/06/78/c0/d5/44/d3/DM00177695.pdf/files/DM00177695.pdf/jcr:content/translations/en.DM00177695.pdf

Maybe somebody has some good ideas what might be causing this

[davekni]

Here's the problem though... the IGBTs do end up dying once I put the coil on mains voltage. Despite everything being pretty rosy all the way up to 300-325V Vbus.

What are the differences between 325Vbus that looks "rosy" and the presumably-325Vbus from mains voltage?  Are you running higher duty cycle or higher peak current on mains?  What do you have for mains inrush current limiting?

Testing the coil with low bus voltage everything looks beautiful, gate voltages look good (tvs diode is clamping during low resistance switch off), phase lead is tuned nicely to just slightly lead the current waveform and IGBT temperature when running at a high test load is ok too.

What is the current scaling for these scope captures?

The overvoltage is not all that significant in this scope shot, but when turning up to the full current (150A for the current setup with only one IGBT per position), the overshoot increases to around 80% (which feels like a lot to me? I'd have expected more like 30-40%). That however still only results in 630V (from 350V VBus), which the IGBTs should survive. Also I must admit, that I'm not 100% sure a tiny little overvoltage transient would result in enough power dissipation to kill the IGBT.

My first guess is that overvoltage may be your problem.  I recently ran some detailed turn-off spike tests on a fast IGBT:
https://highvoltageforum.net/index.php?topic=2498.msg18359#msg18359
Your STGW30M65DF2 parts are specified comparatively slow for current fall time, at 10 ohms gate and Vge going from 15V to 0V.  Drive from 18V to -18V is causing faster current fall, probably fast enough to cause those spikes across just IGBT lead and package inductance.  Also, turn-off spike voltage is roughly proportional to current at IGBT turn-off.  Spikes are low at your well-tuned small phase lead.  Any change in phase lead during operation could cause turn-off at much higher current.
Concerning power dissipation for spikes:  Your STGW30M65DF2 parts do not have any avalanche energy specification.  Same is true for the parts in my test.  I fried the first part testing at 200A at only 1kHz.  Avalanche-rated IGBTs and FETs have uniform breakdown voltage across die.  Power is spread out uniformly.  Non-rated parts often fry at much lower avalanche energy.  I suspect breakdown voltage is lower some places, perhaps at edges of die.  Much less energy and power capability when avalanche breakdown is in a limited spot on die.

One thing I can't explain is the weird voltage curve (marked in red) both during switched off state and switched on state of the IGBTs. I'd kind of suspect measurement error from poor probing, but it does not seem to change at all even when I place ferrite plates right next to the probes.

Are you referring to the light red horizontal line at the top of Vge and Vce traces?  Don't know what current is, but curvature may be just IGBT Vce voltage drop and a bit of inductance in IGBT leads.  ECB plane resistance can add some too.  Planes on inner layers are often 17um (1/2oz) copper.  If using thermal reliefs, adds a bit more resistance.  High part of Vce plot may also include some resonant voltage on Vbus as you discuss.  I don't think Vbus resonance is significant to your IGBT frying.  (Can't absolutely rule out resonance matching some high interrupter frequency, but very unlikely.)

The IGBTs are STGW30M65DF2 from ST

My second guess to frying issues (after turn-off Vce spikes) is that these IGBTs are not robust at 150A, especially if running high duty cycle or long on-times.  Power dissipation rises quickly at 150A per spec graphs (though hard to extrapolate to 18Vge).  Transient thermal impedance may not be sufficient.  And/or thermal coupling to heat sink may be insufficient.  Using the screw hole does not apply pressure efficiently over active IGBT die location.  Can even squish insulating pad under screw, causing lead end of IGBT to tip up very slightly, enough to make terrible thermal contact.  I'd consider something more like this example:
https://highvoltageforum.net/index.php?topic=2191.msg16147#msg16147

Good luck with debug!

[TMaxElectronics]

What are the differences between 325Vbus that looks "rosy" and the presumably-325Vbus from mains voltage?

No 325V or so would be from my variac. Mains voltage gets it to 350V on the Bus. That extra little bit seems to be too mich for the igbts


I'll read the rest of the reply later today

[Hydron]

Might be useful to see the PCB layout of the bridge and gate drive components. I've been battling some turn-off overshoot issues myself and PCB layout made a huge difference to their magnitude (and yes I did see some overshoot with 80% magnitude in some of my earlier layouts).

Couple of other options maybe worth trying if you think it's definitely the turn-off spikes killing the IGBTs:
- Using a snubber like the one in the attachment - would need a tight layout for the D and C to minimise stray L (resistor doesn't sit in the fast current path so isn't critical), and a diode that starts forward conduction very quickly. This type of snubber only clamps voltages above the steady state bus voltage so power dissipation is not a big concern as it can be with other styles. This is on my list to try if I do another PCB spin (especially if I can find some appropriate parts in JLC's assembly list).
- If you're turning the IGBTs off via a gate resistor (as opposed to just totally bypassing with a diode for turn-off) then you might be able to tame the spikes a little by using TVS diodes to clamp the voltage between gate and collector (NOT emitter) to ~400V or so (will end up clamping a fair bit higher because of TVS behaviour). This will help reduce spikes by slowing turn-off when it causes high Vce voltages, though in extreme cases it may risk some cross-conduction though if it delays turn-off too long.
- Have a go with Dave's GDT buffer - it did improve the behaviour on my coil: https://highvoltageforum.net/index.php?topic=2389.0

Improved PCB layout made a much larger difference than trying the TVS clamping trick in my case though so I'd start there. 4 layer boards are crazy cheap these days and give you a lot of flexibility in reducing loop area etc (though take care not to end up with too much stray C between gate and collector). I'll try and tidy up my own latest half-bridge board design and release the PCBs this weekend, though I haven't gotten them going at full bus voltage yet either (or with the full planned 8 half-bridges - just testing with 2 for now and a 4*higher-Z tank).
I'm also trying to hard switch >100A too (without avalanche rated parts) so I'm playing on hard mode, as the 4 IGBTs I've killed so far attest to.

[davekni]

No 325V or so would be from my variac. Mains voltage gets it to 350V on the Bus.

Is your local line voltage 248Vac rather than 230Vac?  Or is 350V a power-on transient?  (That's why I asked about inrush current limiting.)

- Using a snubber like the one in the attachment - would need a tight layout for the D and C to minimise stray L (resistor doesn't sit in the fast current path so isn't critical), and a diode that starts forward conduction very quickly. This type of snubber only clamps voltages above the steady state bus voltage so power dissipation is not a big concern as it can be with other styles.

This form of snubber can be quite useful, though doesn't seem common.  I used a similar circuit (same except resistors are parallel with capacitors) on an IGBT-brick H-bridge at work.  Helps with the inductive spike across the IGBT who's internal diode is turning on.  Does not help with portion of spike developed across lead inductance of IGBT that is turning off.  More common R+C snubbers help with both, at the expense of much higher power dissipation in the resistors.

- Have a go with Dave's GDT buffer - it did improve the behaviour on my coil: https://highvoltageforum.net/index.php?topic=2389.0

If the problem is turn-off spikes, my buffer will not help with that.  Buffer is designed for very fast turn-off (to minimize IGBT Eoff) and lower gate drive power.
Edit:  After a bit more thought, possibly it could reduce spikes a little on three-lead IGBTs, since Vge swings down to 0V and not negative.  I'm using it on four-lead IGBTs with Kelvin emitter connections.  With three leads, emitter lead inductance might slow turn-off a bit, as inductance adds to internal Vge during current fall.

I'm also trying to hard switch >100A too (without avalanche rated parts) so I'm playing on hard mode, as the 4 IGBTs I've killed so far attest to.

I'm also playing in the same realm.  That's why I designed the buffer, to minimize switching energy.  Does make for very fast turn-off, and is causing me the same turn-off spike issues.  I've fried only one part so far in my turn-off test fixture:
    https://highvoltageforum.net/index.php?topic=2498.msg18359#msg18359
In the process of adding R+C snubbers to my phase-shift QCW H-bridge now.

[Hydron]

This form of snubber can be quite useful, though doesn't seem common.  I used a similar circuit (same except resistors are parallel with capacitors) on an IGBT-brick H-bridge at work.  Helps with the inductive spike across the IGBT who's internal diode is turning on.  Does not help with portion of spike developed across lead inductance of IGBT that is turning off.  More common R+C snubbers help with both, at the expense of much higher power dissipation in the resistors.

The style I posted could in theory still help with turn-off in the situation that the C+D part of the snubber is lower inductance and has a diode which starts conducting faster than the full path through the other IGBT's diode and the Vbus capacitors. Would really depend on the reason for the spike though, and whether it can absorb enough of the spike given that the capacitor starts with Vbus across it (the advantage of that is minimising dissipation though).
In my situation at one point I suspected that either the antiparallel diodes in the opposing IGBTs were too slow ("forward recovery voltage" is a thing), or that the physical layout I'd chosen was inherently too inductive, but a newer PCB suggested that this might not be the issue (and these snubbers may be futile) - I should really look at other styles as well though (e.g. try a RC snubber and live with the additional dissipation). I think in theory with correct timing (i.e. always switch a bit before zero crossing for ZVS turn-on) just a capacitor across the IGBT might work (still in theory a RC in this case - R being the ESR/parasitics).

If the problem is turn-off spikes, my buffer will not help with that.  Buffer is designed for very fast turn-off (to minimize IGBT Eoff) and lower gate drive power.
Edit:  After a bit more thought, possibly it could reduce spikes a little on three-lead IGBTs, since Vge swings down to 0V and not negative.  I'm using it on four-lead IGBTs with Kelvin emitter connections.  With three leads, emitter lead inductance might slow turn-off a bit, as inductance adds to internal Vge during current fall.

Yeah this surprised me a bit too, wasn't expecting a big difference in turn-off spikes, more just a cleaner faster waveform, but it looks like I got a bit of both. I am using 3-pin devices (and am annoyed with myself that I forgot to order the 4-pin ones you've got while Farnell had them at a cheap clear-out price).

[Intra]

Please, post photos of parts of your inverter with 2 sides of each pcb. Nothing to I could understand from photo.
What is parameters of primary with mmc and secondary? JavaTC log please.
And the whooooole schematics.
supply and grounding environment desc

[davekni]

The style I posted could in theory still help with turn-off in the situation that the C+D part of the snubber is lower inductance and has a diode which starts conducting faster than the full path through the other IGBT's diode and the Vbus capacitors.

Yes, I wasn't completely clear.  D+C snubbing definitely helps with turn-off.  Just the contribution to spike voltage caused by opposite IGBT's lead and package inductance and/or forward recovery time.  Presuming both IGBTs in a half-bridge are identical, at best that snubber reduces spike seen by turning-off IGBT in half.  If diodes and caps are perfect, there is no spike voltage on ECB.  But there is still spike voltage on the internal die of turning-off IGBT due to that IGBTs package and lead inductance.  External diode snubber doesn't reduce that half.  On the other hand, an R+C snubber can, because it can keep the ECB voltage below Vbus for long enough for IGBT to turn off.  Then turning-off IGBT's spike adds to a lower external voltage, so can be kept below Vbus.  In other words, there is still a spike internal to turning-off IGBT with R+C snubber, but the peak voltage is lower because not adding to external voltage that's already at Vbus.
That internal spike is what I measured in the link posted previously, and why I'm adding R+C snubbing.

[TMaxElectronics]

thanks for the replies. I only managed to get working on this issue further a few days ago and just got some results today. Preliminary load tests seem to suggest the issue is fixed and the IGBTs are no longer expensive popcorn

I will supplement this post with some scope captures if and when I make some, for now you'll just have to beleive what I'm writing here. I was in quite a bit of haste to put the coil back together as we will be setting up a show for this weekend starting tomorrow and I want to have the coil shown there, so I didn't take any pictures either

First of all, the issue did not seem to be overvoltage afterall. I ran tests at very low pulse frequencies (1Hz) at full current and 350V VBus and observed only <100V of worst case switchoff overvoltage during some bad switching transients, most of the time the voltage was smaller. Seems like the six layer board was worth it after all

The real reason for the very fast death is much more likely a thermal one. A combination of me being a little overambitious with the current limiter and a very low impedance primary circuit. I had the limiter (for pulseskipping) set for 250A and was pushing 800us ontimes. That combined with the higher conduction losses of the diode (which during pulseskip are actually the deciding factor due to the low tank impedance) caused the IGBTs to die a hot death. I remember testing the inverter on the bench and getting to ~90° case temperature at only 100V or so on the bus, which likely ment TDie was over 135°. As the voltage increased and the diodes started taking the current more often I think it just ended them.

Especially because the tank circuit could in theory allow for a massive current overshoot of ~150A absolute worst case, which would result in a peak of around 400A (200A / IGBT), which is even above the saturation level of the bigger IGBTs I have now. Calculations for theoretical maximum currents (without switching losses!) allow me to get to around a sine wave with 130A peak current and 100A for the diode, so I was definetely pushing it too far.

To further help reduce thermal troubles I changed out the IGBT mounting. Instead of the stackup being [TO247 | thermal paste | insulator sheet | paste | heatsink] I added an aluminium heatspreader inbetween the igbt and the heatsink like this: [TO247 | paste | 1.5mm heatspreader | insulator | paste | heatsink]. That increases the surface area contacting through the (thermally not tooo conductive) insulator sheet. It is intended for thermal conduction but is not great at it. Incidentally I'm never sure if you need to put thermal paste onto those insulator sheets or not certainly helps keep them aligned though.

That combined with a slightly lower ocd threshold (150A vs 250A) got me to some nice and cool IGBTs even when running 10% dutycycle 800us pulses (at whatever frequency that results in, 125Hz or something I think).
Case temperature got to 50°C with the heatsink at 30°C, much better than 90°C TCase at the same heatsink temperature I had before.

And now some replies I forgot to make

The snubber approach is certainly an interesting idea. I'll be keeping it in mind if I ever need some hard switching inverter

Is your local line voltage 248Vac rather than 230Vac?  Or is 350V a power-on transient?  (That's why I asked about inrush current limiting.)

No its 230V, but I assume a slight mains overvoltage to be on the safe side
I also have a 800uH choke behind the bridge rectifier, that might increase the voltage a little too.

4 layer boards are crazy cheap these days and give you a lot of flexibility in reducing loop area etc

6 Layer boards are too
The bridge I made uses the outer layers for signals and earth and the inner four for VBus +&- and the bridge outputs, just completely poured all over. JLC also offers 70um copper on the inner layers for not too much money.

What is the current scaling for these scope captures?

That was 1mV/A

[Intra]

Great to see you got troubleshoot this issue.

As I slightly mention here https://highvoltageforum.net/index.php?topic=188.msg18390#msg18390

With a heat issue which contains also the switching losses heating sub-issue I started to see to do not make inverters with over current modes. Especially that starts to be important when starts pulse-skipping.