Why would a 20 year old MOSFET fail?

Question

I have an old MOSFET power amplifier which dates from (I think) 1997 or so. It was made by a well known British manufacturer. I bought it used about a decade ago, and it has worked faultlessly.

A few weeks back it didn't power om, the cause turned out to be the fuse, in turn caused by an output power MOSFET, IRF540, that was dead short.

No big problem, replace, carry on, right? But why would a MOSFET which has been stable for years, and which was not subjected to any heavy duty use or otherwise, suddenly do this?

Maybe other components drifted out of spec? Perhaps spikes coming through the mains?

As far as I know there is no real "aging" mechanism in transistors, so long as they are used within their normal limits, am I correct?

Answer 1

Given the vintage of the amp, I'll assume that it's Class-AB (linear), not Class-D (switching).

Linear MOSFET power amplifiers often aren't designed particularly well. Many of them suffer from parasitic oscillations caused by the large gate-source capacitance of the output power MOSFETs, which are quite difficult to dampen and can get worse over time as surrounding components (such as compensation capacitors) age. This is a common theme in MOSFET amp repairs which typically causes blown-up / shorted power FETs. If you replace the FETs without fixing the underlying design issue of the amp, they'll most likely die again.

Another thing that's often done wrong is the selection of suitable MOSFETs. Your amp gets this wrong as well - the IRF540 is not rated for linear operation, as evidenced by its datasheet lacking a DC curve in its SOA plot. It's a HEXFET, which means that it's constructed out of hundreds or even thousands of tiny hexagonal MOSFET cells connected in parallel on its die. When driven into its linear operating region, these cells will not conduct current equally, causing hotspots to develop. This causes the MOSFETs to fail rapidly.

So in short, it's not actually surprising that the IRF540 failed here. It's more surprising that it even worked for a while in the first place.

Proper MOSFET-based amplifiers typically use planar MOSFETs, which don't suffer from the same hotspot problems as HEXFETs when driven into the linear region. IXYS (now Littelfuse) has a series of such FETs if I remember correctly.

International Rectifier has an application note detailing how HEXFETs fail when operated in linear mode, which also includes a thermal image of a hotspot developing on a FET die. They also note that more modern devices are more likely to fail in this way, which likely also includes later revisions of classic part numbers if there has been a die re-spin to reduce manufacturing cost.

Answer 2

Yeah, I heard it before "These MOSFETs are not rated for linear use!"

There are millions of amps using oldskool FETs like IRFP240/9240 in linear mode.

More modern FETs are a different story, they will blow up if used in linear mode. But while these old FETs are not "officially" rated for linear mode, everyone uses them, everyone knows they will work absolutely fine and are extremely rugged.

But... there's of course a limit to how bad the design can be...

Let's look at the amp first.

What the hell? Where's the heat sink? It's a class AB amp, so there should be a heat sink, right? Are they... these ridiculously tiny thumb-sized things on the back of the amp?

It's a 2x50W amp so you'd expect at least 4x bigger heat sink, in a place where there's actually airflow. Compare with this random amp:

I prefer heat sinks on the sides though. They still work when the cat sits on the enclosure.

Even if the heat sinking was adequate, one pair of TO-220 devices is not enough for a 50W amp. First, TO220 has little thermal mass, so it heats up too quickly. Second, the contact area at the back is tiny, so the thermal interface between device and heat sink is always a problem.

Class AB amps have reasonably low average dissipation but peaks can be high, and power peaks tend to occur at bass frequency, which means they last long enough to heat a TO-220 to dangerous levels.

TO-247 packages offer a good amount of thermal mass, because the chip is mounted on a big chunk of copper, which helps immensely to soak up dissipation peaks. This keeps the peak junction temperature down, so while TO247 is a bit more expensive, they will get away with a smaller cheaper heatsink.

One big weak point of TO220 is the thermal interface between the device and the heat sink. The more contact area with the heat sink, the lower thermal resistance, and TO-247 wins by a wide margin simply because it's huge. These will take a lot of abuse. TO-220, not so much.

So if you use TO220 you need several in parallel. But paralleling devices with negative threshold voltage temperature coefficient is always a headache. That's why I expected to see the usual P/N pair of TO-247 IRFP240-9240 devices per channel.

Well, in this amp's case, it doesn't matter, because there's no heat sink.

A quick look at the service manual reveals a complete absence of any protection circuit of any kind: output short, overcurrent, overtemperature, etc.

Likely cause of death is exceeding max Tj repeatedly and at length. In fact the MOSFETs did pretty well if they survived that kind of abuse for that long, I'm impressed!...

So basically, it failed because the design is garbage.

Answer 3

All electronic components are subject to aging, even MOSFETs. This is directly related to temperature, so a FET with an inadequate heatsink, package, thermal interface, or all three will be vulnerable to failure.

A 1997 vintage amp is likely Class AB. It's still a popular approach to making high power (e.g., 200W) amplifiers as a hobbyist thing owing to its simplicity, and because FETs are fairly easy to connect in parallel. Nowadays you can roll your own Class D if that floats your boat.

That said, Tripath introduced their Class D amplifier in 1996 (fun fact: in that timeframe I met Tripathi and saw his demo of a 2x20W amp - impressive.)

What happened to your amp though? High-power MOSFETs are vulnerable to Spirito Effect, which shows up with high transconductance FETS (e.g., trench FETs) as formation of localized unstable hot spots due to uneven cell matching. It's kind of like bipolar transistor mismatch in output stages, but on a more micro scale.

A bit more here: Mosfet thermal problems in linear applications

The takeaway: FETs operated in linear mode are even more vulnerable to Spirito effect, so require derating of their Safe Operating Area (SoA).

As for your amp, it was a good run while it lasted. If you can, maybe replace the FETs with physically bigger ones, and work up an improved heat sinking system.

Feeling adventurous? Could be an opportunity to explore SiC, GaN or GaN-on-SiC.

Answer 4

The operating mode of a power MOSFET amp is PWM, with the MOSFET alternating between blocking large voltage and admitting large current. That causes very little heat dissipation.

The question is what happens during power-up and power-down. When the gate voltage range gets smaller, the MOSFET may no longer open completely, meaning that its losses get much larger.

Depending on how power-up and power-down of the whole amp (including preamp) are sequenced and whether this sequence is messed up when the power switch bounces or otherwise the power sequence is interrupted, there may be moments of extra strain, like when the amp tries to route a preamp pop to the speakers during supply voltage low levels, further depleting the power supply capacitors. If the power supply capacitors hold more energy than the MOSFETs can safely dissipate, the sequence matters, and brown-outs might lead to multiple restarts of the sequence while the MOSFET junction has not yet dissipated the heat from the first shutdown.