IRF540 failed G-S

[mc172]

I've got a pair of identical amplifiers in for repair with some funny symptoms.

The amplifiers have got an IRF540/IRF5210 output pair. The IRF540 has gone short from gate to source in both of them and both of the IRF5210s are good. I've never seen this before - I've only ever seen drain to source shorts.
I've got some IRF540s knocking around so I got one out and blew it up by forcing it to dissipate a bit more power than it should be able to handle while attached to a heatsink, and thus overheat, over a period of about 2 hours and it eventually failed drain to source short.
The amplifier has got 12 V Zener diodes from gate to source, and they're all working correctly when I take them out of circuit to test them, so I doubt that excess V_GS has caused it.

Has anyone seen this before or got any ideas why this device would fail G-S short please?

[Whales]

IRF540 is a common part number used on greymarket sites.  I think the ones I bought years back didn't meet datasheet specs.

Too high of a G-S voltage could damage the part.  Any chance this could be happening?

[floobydust]

The only time I've seen MOSFET gate punch-through (assuming the circuit's zeners are working) is when the parts got damaged by ESD.
It seems to be about 2 weeks and then the parts fail. I know that sounds weird but I noticed all the techs who were careless about ESD had recalls after a couple weeks as the parts failed. I read ion migration happens across the damaged oxide layer and then it gets worse.

[magic]

Does the D-S diode still work?

[M0HZH]

I've had plenty of laterally-diffused MOSFETS (aka LDMOS) fail with low (a few ohms) G-S resistance due to Drain overcurrent / overvoltage  or even overheating; as a matter of fact, that's the only failure mode on that type of FET I've ever heard of. The Source almost surrounds the Gate on LDMOS, it's quite sensitive.

[duak]

I'll second M0HZH attributing overheating as one cause.  If you put enough power into the device, you can melt enough of the die to connect the gate and source.

I learned this while developing a 1 kW BLDC driver.  If memory serves, I used a fuse in each phase's DC link + connection for overcurrent protection.  If the fuse ever opened, the recirculating current between the phases would cause the now isolated phase to pump up its voltage until something broke down, usually a MOSFET.  The other two phases would continue to switch, driving current until a load imbalance was detected.  The MOSFETs were in metal cases and one struck an internal arc between the source pin and the case, burning a 1mm dia hole in it.

[mc172]

Thanks for the comments chaps.

I think I may have misguided myself by not testing everything the same way when I had the opportunity to. It didn't occur to me at the time but hindsight is brilliant as they say.

It turns out that both the FETs I received in the amplifiers and the one I deliberately blew up are short between all pins but to varying degrees, which gave some strange looking results on the meter in diode mode, when compared with the next device.
For example, one of the FETs from the amplifiers gives the following in diode mode:
30 mV G-S
30 mV S-G
501 mV S-D
1192 mV D-S
555 mV G-D
1202 mV D-G
However, when I put this FET on a power supply, it will happily draw as much current as you can give it S-D or D-S, and will draw about 600 mA S-G, G-S at 10 V.

For comparison, the one I blew up deliberately, shows:
102 mV G-S
103 mV S-G
22 mV S-D
23 mV D-S
124 mV G-D
125 mV D-G
This FET will again draw as much current as you can throw at it between D-S or S-D, but will draw only 250 mA S-G, G-S at 10 V.


Magic, your suggestion to test the body diode is what tipped me towards sticking the devices on a power supply and just brute force them G-S to see what is shorted and what isn't, at an arbitrary 10-ish V. I had done this D-S before but for some reason didn't think to do it between all pins.

Thanks everyone for your input, I really appreciate it. I think the conclusion then is overheating.

[magic]

short between all pins

That's the favorite failure mode of MOSFETs, so that the driving circuitry also gets a chance to receive some of that juicy capacitor bus discharge

[JohnPi]

It's not clear what your voltage measurements are, but generally power MOSFETs fail from overvoltage or overheating.

With 12 V zener, you are unlikely to have exceeded the 20 VGS rating. If the VDS rating is exceeded (IRF540 is rated at 100 V, and it is avalanche rated), you could get D-S breakdown to begin and this will lead to overheating and subsequent permanent failure (not because the silicon melts, but the internal aluminium interconnects melt and get absorbed by the silicon creating a short). But if the power supply is < 100 V, this is unlikely. Actual FETs are quite rugged and don't fail (melt) until the internal temperature rises above 350 deg. C.

Defective devices often fail with D-G shorts internally which prevent the external gate from being able to actually switch the device off. This leads to overheating and massive failure of the device.

Therefore likely your failure is simple overheating -- too much DC power dissipation. Is the load shorted ?

[CaptDon]

There is one particular model of Ampeg bass guitar amplifier
that is known to eat FETs on a regular basis. First off, they run
so hot they'll burn off your finger prints, second, when you
turn the amp off there is such a hellish thump that it drains
all of the energy from the capacitor bank. I find the fets have
all sorts of weird failure modes and indeed they have the
gate to source protect zeners. BTW, zeners are somewhat
slow to hit their regulation voltage and you would be surprised
how high the voltage of a fast transient can be and the zener
never reacts to it. The most outrageous spikes pass through
on startup when the zener is not already 'on'.

[JohnPi]

No, zener are not slow to regulate -- they regulate within nanoseconds of applying a voltage. Conversely gate-source oxide on a FET can withstand 1.5x the rated voltage for many milliseconds.

[CaptDon]

We tried using 'run-of-the-mill' 1/2 and 1 watt Zeners as transient suppressors.
We never saw 'nanosecond' response time, at best maybe tens of nanoseconds.
This measurement was taken going from zero volts / zero bias to an overvoltage
but with resistance to limit current to a safe peak. We gave up on a zener-only
type answer and went with zener plus two stages of R/C filtering. There may be
faster zeners today but the 1n7XX stuff and the 1n3xxx/1n5xxx stuff just wasn't
doing the trick. For that matter the transorbs were even slower 1KPxx stuff and
their clamp voltage drifted like crazy downward with heat.