Half bridge current shoot through... Why?

[EugenioN]

Hi all.

I'm repairing a faulty ICEPOWER ICE1000A audio amplifier (datasheet here https://icepoweraudio.com/download/2408/)
It's a H bridge with IR2010 high and low side gate drivers, four FDP2532 mosfet (datasheet here https://www.onsemi.com/pdf/datasheet/fdi2532-d.pdf) and a PNP bjt gate turn off:

" alt="" class="bbc_img" />

I've found one leg of the H bridge with high side mosfet completely destroyed (vaporized source and drain pins) and the low side mosfet shorted (gate, drain and source all tied toghether).
Since the FDP2532 is discontinued and not available from reputable distributors, I've replaced the broken two with an IPP075N15 (datasheet here https://www.infineon.com/dgdl/Infineon-IPP075N15N3-DS-v02_06-en.pdf?fileId=db3a304319c6f18c0119cd76cc527ab6)

They should be approximately the same: lower Rds,ON for the replacement, and almost the same gate and miller charges. The same for the dynamic behaviour.

Unfortunately, the new mosfets heat up *a lot*, and a closer inspection revealed a crisp and cleart shoot through, in the range of 6 or 7 amps during turn on. This current is virtually no influenced by the bus voltage (why?). Increasing the turn on gate resistor from 12Ohms to 68Ohms completely eliminates the shoot through, but this results in poor distorsion.

In this image, top traces show gate voltage and drain current with original mosfets; bottom ones gate voltage and drain current of the replaced mosfet.
Same 12Ohm gate resistors.
[gate ringing is due to poor probe grounding; horizontal time is not consistent with the waveform: I'm showing off saved waveforms, this is what I get from a 20+ year scope]
" alt="" class="bbc_img" />

No deadtime is inserted at gate driver input signals: the small deadtime comes from different turn on and turn off resistors.

Why I need almost a tenfold increase in gate resistor to avoid shoot through with almost identical mosfets?
What could I check?

[Andy Watson]

Are the two PNP transistors functioning correctly? They should have gain, however, if the collector is not sinkng the required current the turn-off time will be impaired. There is a current path to discharge the gate via the emitter-base diode - so the turn-off could still appear to be rapid but not as fast as it should be with the extra collector current.

[T3sl4co1l]

Did you replace the IR2010? Transistors?

Think about what you said: one failed as a three-way short.  Meaning, at least momentarily, some fraction of the +120V was applied to the gate driver and associated components.  This often fries the driver, and could fry stuff further upstream of that (say by failing as a multi-way short in turn).

Shoot-through isn't necessarily problematic; it needs to be designed in, however.  If used, it's likely a very small amount, 10s of ns, and switching loop inductance likely will be modified to support this (snubbers).

Tim

[David Hess]

That bottom waveform looks like it has a snivet displayed at the rising edge of the gate waveform on a DSO where aliasing occurred.  It does *not* look like poor probe grounding because the same thing is not shown at other fast edges.  A snivet is a type of negative resistance oscillation, and would easily explain high shoot through and heating.  This is consistent with your report of solving the problem by increasing the gate resistance.

Ideally the gate resistance is close to the body of the MOSFET, so how is the layout?

Take a close look at the waveform in analog mode or peak detection mode to see if the rising edge is oscillating at a high frequency.

I might add a ferrite EMI beat at the MOSFET gate lead to try and suppress the oscillation if that is what is going on.

[langwadt]

Trr of the body diodes?

[EugenioN]

Trr of the body diodes?

105ns for the original mosfets, 146ns for the replacements.
+40ns of recovery time can be the culprit?

[EugenioN]

Are the two PNP transistors functioning correctly? They should have gain, however, if the collector is not sinkng the required current the turn-off time will be impaired. There is a current path to discharge the gate via the emitter-base diode - so the turn-off could still appear to be rapid but not as fast as it should be with the extra collector current.

I'm wondering why the mosfets failed in the first place. Turn-off bjts losing some gain maybe caused the fireworks: Now I changed the gate resistors and a slower turn-on is "hiding under the carpet" this issue. I'll get some replacement bjts.

Thank you!

[langwadt]

Trr of the body diodes?

105ns for the original mosfets, 146ns for the replacements.
+40ns of recovery time can be the culprit?

one is max the other typical, but who knows

[EugenioN]

Did you replace the IR2010? Transistors?

Think about what you said: one failed as a three-way short.  Meaning, at least momentarily, some fraction of the +120V was applied to the gate driver and associated components.  This often fries the driver, and could fry stuff further upstream of that (say by failing as a multi-way short in turn).

Wishful thinking: gate drivers survived that explosive event.
I'll try to replace both of them: I've got plenty of IR2110S: they seems good enough for the job

Shoot-through isn't necessarily problematic; it needs to be designed in, however.  If used, it's likely a very small amount, 10s of ns, and switching loop inductance likely will be modified to support this (snubbers).

Interesting: do you have any reference/application? I'm curious now!
Looking at the "good" bridge leg I can measure ~100ns of deadtime, and a small/negligible drain current spike during the gate voltage plateau: everything as expected, so I assume no black voodoo magic.

Side curiosity: this class D amplifier has a bridge tied speaker with a custom coupled inductor (common mode) and a capacitor (differential) as LC output filter. Quite uncommon, I've always seen two independent LC filter, one for each leg.

[EugenioN]

That bottom waveform looks like it has a snivet displayed at the rising edge of the gate waveform on a DSO where aliasing occurred.  It does *not* look like poor probe grounding because the same thing is not shown at other fast edges.  A snivet is a type of negative resistance oscillation, and would easily explain high shoot through and heating.  This is consistent with your report of solving the problem by increasing the gate resistance.

I must confess my ignorance: I don't quite get what "snivet" means!
When properly probed with the ground clip connected to the source, the gate voltage is perfect in both legs: nice turn-on with barely visible voltage plateau during turn on, steady 10V voltage during on state and a fast turn-off ramp. No ringing.
I'm sorry for this poor scope screenshot, but it was the only leftover in the waveform memory. My fault

Ideally the gate resistance is close to the body of the MOSFET, so how is the layout?

Take a close look at the waveform in analog mode or peak detection mode to see if the rising edge is oscillating at a high frequency.

I might add a ferrite EMI beat at the MOSFET gate lead to try and suppress the oscillation if that is what is going on.

Well, actually one leg is working fine as is. No special care can be observed in pcb design: maybe it works because the gate driving circuitry dimensions are less than a couple of squared centimeters, so even sparse components cannot be that harmful.
I don't know: ony the leg with new mosfets is shooting through, and the pcb design is a cut&paste of the working one. Before the explosions everything was fine.