IGBT transistor failure mode - slower closing?

[daqq]

Hi guys,

I'm dealing with a device where an IGBT transistor connects a capacitor bank to a pulse transformer, which then feeds a resistive load. The transistor is opened for about 2us, then closed, the resulting waveform is square-ish. The normal input voltage is about 300VDC, the figures displayed are from a 40V test voltage, but the issue is the same for both voltages. The current @300V is around 500Amps. The transistor is FZ900R12KE4 ( https://www.infineon.com/dgdl/Infineon-FZ900R12KE4-DS-v02_03-EN.pdf?fileId=db3a30431f848401011fb7c00c74594e ), a 1200V IGBT, being driven by the 2ED300C17 driver module ( https://www.infineon.com/dgdl/Infineon-2ED300C17_S_ST-DS-v04_03-en.pdf?fileId=db3a304412b407950112b429b99741da ). There's 2 Ohms in series with the Gate. When the device is working properly, the waveform looks like this:

There are several units, all of them have waveforms like this - a fast rise, a fast fall. However, one stands out:

There seems to be a much longer falling edge and the waveform is of a very ugly shape - not square-ish at all. The load is the same. I have measured the gate voltage waveform and it is absolutely identical to a known good device.

Now, I'm trying to isolate the problem - I'm fairly confident that this is something that showed up recently and that the device was not doing this before - we probably would have caught it during initial testing as waveform shape is checked. That said, it could have happened. The transistor looks the same (no physical damage that I can see).

My question is, is there a mode of failure of IGBT transistors that would result in a slow fall time, but keep the other parameters (gate charging, rise time etc.) the same?

There's also a few snubbers and other things, I'm checking them as well, but I can't find anything wrong with the device.

Thanks,

David

[krish2487]

Try placing a UFR diode anti parallel to the 2 ohm  series gate resistor. From what I can see, I can only agree with you.. The IGBTS are turning off too slow..
Placing a diode anti parallel with allo you to control the on time while still having a quick off time..

However, EMI might become a  potential issue with such a fast turn off..

On closer reading... I think you might have an issue with the gate driver... Can you just swap the drivers between the good / ugly boards to see if the problem remains on the same setup or not?? It might give some insight to locating the problem..

[daqq]

OK, mystery solved, it was a faulty diode in the RCD snubber. It failed and shorted out... this'll be a real headache... Now to find out why the damn thing failed, it was a DSEI2X101-12A ( http://ixapps.ixys.com/DataSheet/DSEI2x101-12A.pdf ).


It probably wasn't overvoltage or current since the resistors are perfectly fine and would have died probably... this'll be fun.

[capt bullshot]

One rough guess:
The repetitive 500A peak current into the diode creates too much power dissipation when the pulse repetition rate is too high.
Not taking the turn on transient of the diode into account, just forward voltage over peak current, the 500A roughly creates 700W peak power into the diode. So I'd recommend to calculate the Joules per pulse and check the power dissipation capabilities.

[daqq]

At a pulse repetition rate of 250pps and the average power is negligible. There are devices running, the temperature rise on the diode is around 2-3 deg C or so, nothing particularly amazing.

[MagicSmoker]

At a pulse repetition rate of 250pps and the average power is negligible. There are devices running, the temperature rise on the diode is around 2-3 deg C or so, nothing particularly amazing.

Interesting failure mode and thanks for keeping the thread updated. The SOT-227B dual diode looks like a reasonable choice for the application so I have to wonder if this is a case of too-high instantaneous dissipation due to a longer than expected forward recovery time. Basically, the voltage across the diode grossly overshoots the typical forward voltage spec during the beginning of conduction leading to thermal hotspotting in the die and eventual second-breakdown-like failure (much like dI/dt induced failure in thyristors, actually).

Unfortunately, few manufacturers of fast diodes give a forward recovery spec which makes it difficult to find an alternate just by datasheet comparison. You might want to try substituting a Schottky for the low voltage testing, and/or intentionally inserting inductance in series with the RCD snubber (add an RC damper across the inductance or you'll just be trading one overvoltage problem for another).

In any event, observe the voltage across the RCD diode to see if this might be the culprit. Setting up the triggering is going to be tricky, but if you have a high bandwidth Rogowski or Pearson current probe the rise in current as it diverts into the RCD snubber might be the best thing to trigger on.

[daqq]

so I have to wonder if this is a case of too-high instantaneous dissipation due to a longer than expected forward recovery time.

Whatever the case, it seems to be a fluke (hopefully). These devices have been running for over a year 24/7 @ 250pps (not this particular piece, this one's new) and during testing they were working 24/7 @ 500 pps, at severely heightened temperatures (60+deg C ambient). They can survive repeated arcing on the secondary.

In any event, observe the voltage across the RCD diode to see if this might be the culprit.

I have. The diode is rated to 1200V and parallel to a 100 Ohm (pulse) resistor. I don't see how it would even be possible to get 1200V out of the system, given the 300V input. There are RC, RCD snubbers and protection diodes.  It's all very depressing.

[MagicSmoker]

In any event, observe the voltage across the RCD diode to see if this might be the culprit.

I have. The diode is rated to 1200V and parallel to a 100 Ohm (pulse) resistor. I don't see how it would even be possible to get 1200V out of the system, given the 300V input. There are RC, RCD snubbers and protection diodes.  It's all very depressing.

Hmm, it wasn't made clear that this is an outlier among a number of units that have been working for a year; that does rather complicate things.

That said, I'm still leaning towards the instantaneous dissipation during forward recovery being too high - see Fig. 6 in the diode datasheet which plots t_fr and V_fr vs. dI/dt. For example, if current is changing at ~300A/us then tfr will be ~500ns and Vfr will be ~20V... Needless to say, that's quite a bit more dissipation than might naively be assumed from the conventional forward voltage spec!

Ironically (or perversely, really), "fast" diodes often have a slower forward recovery time compared to their reverse recovery, and in some cases a standard recovery diode will outperform the fast type in forward recovery! As you have configured the snubber to limit dV/dt (R || D), rather than as a voltage clamp (R || C), you might want to try both standard recovery and SiC Schottky diodes here.

Otherwise, you may end up needing to use a higher current diode, and/or multiple diodes in parallel (with a ballast resistor for each of them).

[duak]

I don't know for sure but I would expect that current filaments could form in the diode dice under fast turn on conditions.  There is a brief description starting on page 10 in this paper:
http://www.mos-ak.org/dresden_2016/presentations/T9_Sohrmann_MOS-AK_Dresden_2016.pdf

It would be interesting to see the voltage across the diode as it is turning on.

An odd thought, I wonder if the material of the screws used with the terminals should be non-magnetic?  That is, would a magnetic screw introduce a series inductance that affects current rise and fall times?  Quick test, are all screws of the same type? - perhaps the failed diode had a different type of screw - maybe even too long.

[T3sl4co1l]

AFAIK, diodes can fail due to electromigration at high peak currents.  At least that's my guess at the mechanism.  Respect the surge rating, if given.  You'd expect the rating to be higher for very short pulses, but from what I've seen, it's not much more.  Say 45A for 1μs vs. 30A for 8.3ms.

Example, I've blown diodes rated 12A that carried a few average, and ran cool. Snubber application. They failed shorted. That board needed 30A diodes.

It would be interesting to see the voltage across the diode as it is turning on.

I don't know about this particular case, but I've seen 60V across such a diode—most of which is not attributable to lead inductance!

An odd thought, I wonder if the material of the screws used with the terminals should be non-magnetic?  That is, would a magnetic screw introduce a series inductance that affects current rise and fall times?  Quick test, are all screws of the same type? - perhaps the failed diode had a different type of screw - maybe even too long.

Nah—the impedance is very small, comparable to the resistance of the screw. It is definitely there, as can be measured in real inductors for example (a diffusion term works well for both wire AC resistance and core loss), but you'll have to use a long screw before it's very significant here I think.

Tim