IGBTs dying silently, but why?

[xxThorx]

Hi everybody,
I'm building my second DRSSTC with this configuration:

Full Bridge of IXGN60N60C2D1
Filter cap: 5100uF 400V
Snubber cap: 2.5uF 1kV
MMC: 149nF made with 3S 3P CDE942C20P15K-F
Primary: 7.5 turns of 6mm copper tube
Frequency: 110kHz
OCD set to 300A (planned 400-450A)
10kohm 50W across outputs
Resistor between GDT and Gate: 10R
Schottky diode between GDT and Gate: 1N5819
Zeners between Emitter and Gate: 28V 3W
GDT made with 10 turns of CAT5 cable over EPCOS N30 toroid

I'm using UD 2.5 driver, with feedback transformers made with turns ratio 1:625 (OCD 5R1, FB 51R)(2x 2 EPCOS N30 toroids)
I've burnt in total 7 IGBTs, they keep failing silently, shorting E, G and C all together.
The first 6 failed with a bus voltage of around 120VDC.
I've tried using 5R1, 8R2 and 10R resistors between Emitter and Gate, with these results (0VDC bus)



After these failures, I've rebuilt the GDT and feedback transformers with the EPCOS N30 toroids (I was using the yellow toroids from an ATX, really wrong choice, I admit).
I thought I had found the source of the failures, but today I've tried pushing the coil to 200VDC, yet another silent failure..
I plan to use it at full power (325VDC, 450A OCD), but I can't keep throwing away IGBTs, as it take 1 month to arrive from China..
For Phase Lead I'm currently using Coilcraft 7m3 153 inductors (maybe I have to tune it better?)
I'll attach some pictures of my overall setup
Does somebody have any suggestion?
Best regards
Daniel Rinni

IMAGES:

[oneKone]

welcome to the forum!

When i built my first drsstc i also used chinese ixgn60n60 devices, Like you i had zero success with them. I think i killed like 12 of them before i gave up. In the end i ended up ordering genuine 'nos' (bit of a stretch calling them that ) devices, they were expensive but they worked flawlessly.

imo if you can (and i definitely don't want to deter you from using the packages you want) have a look at skm modules from aliexpress, I've successfully used skm100gb in a fullbridge at 650a @ 180khz, although now they're sitting at 500a limit.

these are the ones i ordered:
https://www.aliexpress.com/item/32226740068.html?spm=a2g0o.productlist.0.0.48205ec3N0Us4f&algo_pvid=adc8b0d0-59f2-4b2e-9e15-c57bf2dfb687&algo_expid=adc8b0d0-59f2-4b2e-9e15-c57bf2dfb687-1&btsid=0ab6f81615852262872864010e4291&ws_ab_test=searchweb0_0,searchweb201602_,searchweb201603_

sorry i couldn't give the answer you were looking for,     

[xxThorx]

Thank you very much for your answer!
Well, it is quite strange that all these modules have broken, and they all come from China..
I read that also Mads have had some problems with these modules, were these also from China, brand new? Or they were an old stock..?
It should be a discontinued/obsolete product, maybe they're producing them again, but not with the same quality?

Anyway, I will consider switching to these half bridge modules, but first I would like to search some similar modules with the same package, I don't really want to rebuild the aluminium busbars, they're laser cut 😅
Any suggestions on similar, not discontinued IGBTs?

[davekni]

I'll agree that counterfeit Chinese parts are the most likely cause.  As you mentioned, drive phasing is another possible cause.  That you can measure with your scope.

One small point that isn't likely to be a real problem:  Those IGBT packages are intended to have one emitter terminal used for the power connection and the other for gate return.  It's better not to have connections between the two emitter terminals outside of the IGBT package itself.  Outside connections add some of the voltage drop (due to parasitic inductance) to the gate-drive signal.  Your connections are with aluminum bars directly on the packages, so not likely an issue.  I'd still suggest leaving off those little bars.

I couldn't tell from the images just how much wiring inductance there may be between the 2.5uF snubber cap and the bulk filter cap.  With 600V IGBTs, inductance there can be problematic at the end of each enable pulse, when the H-Bridge changes from drawing power to sourcing power (from the Tesla primary back to the bulk supply).  I had to redo my bulk-cap wiring for that reason, as my DRSSTC also uses 600V parts (10x TO247 for each leg).  Bulk cap inductance makes a large voltage spike on the snubber at that end-of-enable transition.  Try scoping across the snubber and triggering off the trailing enable edges.

Final note:  For future posts, would you mind uploading pictures directly to this forum rather than linking?  The linked image host has intrusive advertising, and was initially blocked by my anti-virus software.

[xxThorx]

I'll agree that counterfeit Chinese parts are the most likely cause.  As you mentioned, drive phasing is another possible cause.  That you can measure with your scope.

Here are some scope shots, I've changed the burnt IGBT with a new one.




Yellow trace is the DC bus voltage (210VDC, peaks of 384V), blue trace is one output of the full bridge. I've tried every setting for the phase lead, I really can't get rid of these huge spikes, maybe I need a smaller inductor.. or maybe it is caused by the same input?
However, 384V peaks should be under the 600V IGBT rating, hence I don't think it is the real problem here.. right?

One small point that isn't likely to be a real problem:  Those IGBT packages are intended to have one emitter terminal used for the power connection and the other for gate return.  It's better not to have connections between the two emitter terminals outside of the IGBT package itself.  Outside connections add some of the voltage drop (due to parasitic inductance) to the gate-drive signal.  Your connections are with aluminum bars directly on the packages, so not likely an issue.  I'd still suggest leaving off those little bars.

I'll do it in my next coil, thank you very much for the suggestion! I can't do it in this one because I can't modify the negative side, I would remove too much material to leave uncovered the emitter for the gate return..

I couldn't tell from the images just how much wiring inductance there may be between the 2.5uF snubber cap and the bulk filter cap.  With 600V IGBTs, inductance there can be problematic at the end of each enable pulse, when the H-Bridge changes from drawing power to sourcing power (from the Tesla primary back to the bulk supply).  I had to redo my bulk-cap wiring for that reason, as my DRSSTC also uses 600V parts (10x TO247 for each leg).  Bulk cap inductance makes a large voltage spike on the snubber at that end-of-enable transition.  Try scoping across the snubber and triggering off the trailing enable edges.

I'll post another picture of the rectifying assembly:



could it be the problem?
Code for the filtering capacitor is ALS70A512MF400, code for the snubber is C4BSNBX4250ZAMJ


PS: I don't know why, but I don't manage to put the pictures in the right direction 

[davekni]

I'll take up-side-down images over the links.  It's not hard to invert them during viewing.

I agree that the bus voltage ring is not likely the issue, at least not the lower-frequency peak that I had problems with.  Thank you for the image;  I now see that the bulk cap is bolted to your bus bars, not at the other end of the cables as I was initially thinking.  That is good.

It's hard to tell phase lead based on the attached scope traces.  H-Bridge output phase appears to be roughly at the current zero-crossings.  Slightly before current zero-crossing is ideal, so slightly more phase lead.  More phase lead requires a larger inductance on the driver board.  (I have UD2.7 schematics, so am presuming that UD2.5 is similar.)  Hopefully phasing the H-Bridge output slightly before current zero-crossing will reduce the spikes.  It usually does.  It is possible that those high-frequency spikes are even higher voltage at the IGBT die themselves than where you can scope.  If the spikes can be reduced by adding a bit more phase lead, that might make the IGBTs last longer.

[xxThorx]

Little update:
I tried adding another 1.5KE440CA TVS between the snubber cap, to decrease any transient voltage.
It worked well for a while at 200BPS, 60us on-time, 300A OCD, I even managed to reach 325VDC input voltage for 5 minutes, until.. quite a nice explosion, it broke in half an IGBT.
It was the only IGBT I had in the bridge from 5 years ago, a used one.. which was quite suspectful.
The other 3 IGBTs appeared to be shorted between all the ports, so I decided to do an autopsy on the different IGBTs, which revealed a different package, with probably inferior quality and design compared to old ones.. I think that 2-3 IGBTs failed short, leaving the other one do all the job. In conclusion, looks like the diode semiconductor failed, exploding

(Bottom, short-failing, new IGBT. Top, exploded, old IGBT. Notice the different internal config and the different labeling)



In conclusion, I discarded completely these IGBTs, since I can't find any reliable component source, and accepted oneKone suggestion.
I'll buy two used SKM100GB125DN modules, add another bus cap to increase ripple current from 22.3A to 44.6A (10kHz), rebuild the full bridge with aluminium laminated busbars and hope for the best!