Power BJT failure mode

[pg]

I have two KORAD KD3005P Lab PSU. One of them started failing in a very strange way.
Up to about 1A, everything works fine. For higher current, the output starts 'flickering' in an irregular pattern, the output current (tested with output shorted) was coming and going unpredictably for 0.1 .. 10s seconds. After a while, the output current stayed at 16mA.

I first thought of a bad connector contact or solder joint. I tracked the error down to the two parallel (with 50 mOhm load sharing resistor) 2SD1047 main power transistors. In the failed state, I measured (directly an the transistors pins) VBE of about 9V.
At the working 1A condition, VBE was about 0.78V but the voltage across one of the 50 mOhm load sharing resistor was practically zero.

I concluded, that one of the transistors had completely released it's magic smoke and the other has this strange failure mode.

Is this a common or known failure behavior? How common?

My wild guess is, that it could be caused by the heavy thermal cycling in this application causing stress on the bond wired or something like this.

I now got some replacement D1047 and replaced the bad ones in the failing unit, that now works again.

I then opened the other unit to check whether both transistors are working and found that about 90% of the current went through one of them and 10% to the other. No fair thermal sharing. So I replaced them too. (not that easy to do, mechanically)

I now have 4 removed, old transistors and expected one to be completely dead and one having some kind of changed parameters.
I did a short test with IB 20mA, VCE 1.3V and got about IC 1.6..1.8A for ALL of them!
How is this possible? Did the thermal and mechanical stress on the pins of removing them fix the problem?
Very strange.

[magic]

My first thought was bad solder joints, but this doesn't explain 9V directly across transistor pins.

Given that the transistors show similar characteristics, I suppose the silicon is good in all of them. This leaves us with package problems, maybe it's indeed what you suspect that thermal expansion tore a bond wire somewhere and it goes open circuit when the package reaches a particular temperature (for example).

[Wolfram]

What's the expected degree of load current sharing with 50 mohm sharing resistors? Where are they placed in the circuit? Just to rule out this being a design failure rather than just a failed component.

[pg]

Total current is max 5A, so ideally 2.5A per transistor/resistor. The resistors are ok, as can be observed with the new transistors. There is only a short track on an aluminum PCB to the 50 mOhm SMD resistors.
A failure of the resistors would not explain the VBE of 9V.

[magic]

This is obviously some sort of internal failure. Maybe it could be reproduced by "warming up" the bad specimen with a heat gun.

Or maybe you aren't testing enough. You say you found them to have about equal β, but what about Vbe, which is the parameter that actually matters when paralleling transistors?

Another question is if it's possible that bad (asymmetric) heatsinking could cause one transistor in the pair to thermally runaway from the other, and maybe fail later. Say, 10°C difference translates to 22mV, which requires ~0.5A collector current difference to compensate. Now, is such current difference enough to increase power dissipation sufficiently to maintain the 10°C difference forever or increase it further?

[tom66]

I don't know if it is related but I had a power FET fail in a linear regulator circuit in a weird manner that I still cannot fully explain, but it sounds distantly related to your situation.

The FET was used to regulate 200V DC down to about 100V at around 50mA rms.  The circuit was used in an LG plasma TV sustain board and the transistor was well heatsinked.  As time went on the transistor became leakier which resulted in the output voltage climbing up slowly causing issues with the picture.  As it got hotter it got leakier and the voltage continued climbing upwards of 150V, before the TV shutdown due to sustain overcurrent (the panel became incredibly bright with distorted image and TV was pulling 900W from the wall, spicy!)  All throughout this the regulation circuit would be pulling gate-source low, so I know it was the FET that was bad.  However if tested at room temperature the FET showed millions of ohms between drain and source.  It was only as it got hotter that the defect appeared.

[richard.cs]

I've had bond wire failures with thermal flashing/flickering in high power LEDs, so yes it feels plausible. It gets hot, it opens, it cools down, cycle repeats.