MOSFET Dissipation (for Power Scaling / VVR)

[mhuss]

How hot can a MOSFET safely get before becoming unreliable/failing? Most data sheet have a high end spec of 150C, but that's about the same temperature as a tube envelope.

I have a '50 watt' amp and and at half voltage reduction, which is the worst case from a dissipation point of view, the (TO-3P) case gets up to 100C pretty quickly, even on a large-ish heat sink in free air. These devices are now available in impressive voltage and current ratings, but you still have to dissipate the watts.

Has anyone done a reliable 50 watt or larger amp using PS/VVR? Maybe using a CPU chiller?

[Leo_Gnardo]

My understanding of power MOSFET = admittedly incomplete = is that they become less conductive with rising temperature, a self-limiting feature that tends to save them from self destruction compared to ordinary junction transistors that tend to "runaway" at high temps. Of course it's still a good idea to provide plenty of heat sinking and moving air to keep them cool.

[olddawg]

With a lot of active components it comes down to heat conduction and surface area. If you have a large aluminum heat sink with fins and the devices attached properly with heat sink compound, it makes a big difference tather than simply being screwed onto steel. In any case, whisper fans are cheap, quiet, and plentiful. Even if you use 2. Better safe than sorry.

[loudthud]

100C is too hot for reliable operation. It would seem that you need a much bigger heatsink. Because current flows in spikes, the amplitude of which is governed by filter capacitance, calculating the MOSFET dissipation is very complex, best left to a modeling program like LT Spice.

I used a CPU chiller on a class A power amp to dissipate 150W. It works but will melt down if (when) the fan stops.

[R.G.]

Like so many things involving heat transfer, it depends. Sorry - I know that's not what you came for.

First off, kudos for recognizing that (1) a heat sink is needed and (2) a *big* heat sink is needed and (3) measuring what you get. That seems like simple stuff to you by now, but it puts you off into the stratosphere compared to where many people start.

What matters to semiconductor junctions is the chip temperature. The chip generates heat, and its temperature goes up until it reaches thermal equilibrium, meaning that as much heat is going out as is being generated.

Heat flow is more complex than this simplification, but it's good for a first look. Heat flow and temperature can be modeled like a thermal "Ohm's Law". The temperature difference across some thermal resistance is equal to the power (analogous to current) flowing through some thermal resistance, which is denominated in degrees C per watt of power transferred.

There exist grain-of-wheat incandescent bulbs that dissipate only 1/2 to 3/4W, but the filament gets up to temperatures that make it glow white-hot. That's because the bulb is designed to stop heat flow out by conduction, and the only way power can get out is by radiating out as light. On the other hand, hair dryers may generate a kilowatt or so, but the waste heat comes out as 120F air, not yellow-white light. The difference is that the thermal resistance of flowing air over wires is remarkably lower than from the inside of the bulb.

The grain-of-wheat bulb also illustrates that heat flow problems are usually heat *spreading* problems. The bulb can't spread its heat. The dryer has a moving volume of air to carry it away.

But back to semiconductors. The fractional-nanoacre chip generates heat, and most of that has to be conducted out by the solder mounting of the chip to an internal heat spreader (if it's been well designed) and then to the transistor's external metal heat conduction plate. Every transistor worth using will specify this resistance to getting heat out as a resistance of X degrees C per watt from chip to case.

There is then a thermal resistance from that metal plate through some insulator to the heatsink itself. And finally from the heat sink to the external air.

The thermal resistances add, and the internal junction temperature puts a cap on the internal temperature. If you assume a 150C chip temperature, and the external air is at 110F/43C, then you only have (150-43) = 107 C rise you can tolerate. If the chip has to dissipate 40W, then the total device + insulator + heat sink thermal resistance has to be less than 107/40 = 2.67 C/W. Easy-peasy.

Er, until that resistance needed gets under about 2.5C/W. The size of the heat sink explodes. When that happens, you can get into fancier and fancier designs with forced air, even forced *water* like some gamers like to run their CPUs.

If you have the option (as you do at the design stage) you can get a huge advantage by electrically paralleling the power devices. If you have to dissipate 100W, it may not be practical without forced-water + forced-air cooling in one chip, but if you split this to two chips, you can use a multifin well-designed heat sink to get rid of 2x 50W, and getting rid of 4 x 25W spaced out on a sink may not even require fins. It's how DENSE the power is, and how hard you have to work to spread it out.

I heavily advise looking at whether you can use 2-4 MOSFETs in parallel to spread the heat out.

[J M Fahey]

Please post amp HV current at full blast, squarewave clipping into a 20% reduced load, so as to have worst case consumption, and then do the math with that same current and half the voltage, so as to predict worst case.

I bet the results must be hair rising.

Only armed with such data can we design a cooling system (even a passive heatsink) but from the beginning, I consider the steel chassis the worst heat sink available.

Maybe a large (at least 4"x4") 2mm aluminum "heat spreader" , riveted to the steel chassis and with thermal grease in between may start to do someting useful.

FWIW I use such heat spreaders even on my SS amp aluminum chassis, simply because besides area, thickness rules.

[mhuss]

I was wondering about paralleling. I know the die temperature is always a lot hotter than the case, which is why I stopped when I reached 100C on the latter. Would you need to use source resistors (like emitter resistors in a big bipolar amp) to compensate for imperfectly matched devices? Or, since it's basically a source follower, would that matter?

My hesitation is stringing 450VDC and MOSFETs along like Christmas lights on heat sinks -- outside of the chassis. Unfortunately, the heat makes shrink wrap shrink away from the body and expose the leads a bit.

[octal]

Here's a thought. Rather than using a mosfet to drop the entire voltage, how about using a power transformer with tapped secondaries? You could have a "high power range/low power range" switch and then use the VVR gadget for fine-tuning your output power beyond that. This could drastically reduce what your mosfet has to dissipate. Dropping hundreds of volts across a device is going to equal some crazy heat even at low draws.

[mhuss]

I'll get some numbers, but I'm already using a ~4x4 multi-fin black aluminum heat sink (an old one I had lying around).

The position isn't optimal for airflow, but it was just a quick experiment.

[mhuss]

That's a interesting solution, but we already have a box full of shiny new not-cheap PTs...

[mhuss]

The other problem is that the solution has to fit inside the existing amp cabinet, and exist in a fairly warm environment (with some toasty EL34s) even in the best of times. This might be why R.G. was using 105F as a starting temperature. :-)

[Jazz P Bass]

I'll get some numbers, but I'm already using a ~4x4 multi-fin black aluminum heat sink (an old one I had lying around).
[ATTACH=CONFIG]34245[/ATTACH]
The position isn't optimal for airflow, but it was just a quick experiment.

That looks like 1/2 of a heat sink from an ElectroVoice 100M.

[mhuss]

That looks like 1/2 of a heat sink from an ElectroVoice 100M.

Or maybe an old Dynaco

[J M Fahey]

Here's a thought. Rather than using a mosfet to drop the entire voltage, how about using a power transformer with tapped secondaries? You could have a "high power range/low power range" switch and then use the VVR gadget for fine-tuning your output power beyond that. This could drastically reduce what your mosfet has to dissipate. Dropping hundreds of volts across a device is going to equal some crazy heat even at low draws.

Excellent idea.
Some amps (Musicman and TinybTerror as example) have switchable supplies to achieve thatn ,and in a more efficient and cooler way.

FWIW I'd make *all* tube amps available with a 1/2 voltage tap switch, instant power taming and to boot, lowering heat and prolonging life.

Making a 100W amp and then padding it down to bedroom or garage levels looks crazy to me.
Would much have , as suggested above, a tapped transformer and if necessary, Mosfet fine tuning.

[R.G.]

Hmmm. Just doing some speculation here.

You do have a fairly good heatsink there already. It's an old favorite of mine. I could look up the number if you need it. It's about 2C/W IIRC. There are thermally-better sinks available, but they get much bigger.

You're dealing with quite a bit of heat. When that happens, it's a good idea to consider a couple of approaches. One is brute force - more devices, more heat sinks, and so on. (Before I forget it, paralleled MOSFETs would probably not need much in the way of source resistors to equalize, maybe none.)

At some point you come to needing switching techniques. These come in range switching (the transformer taps idea is one of these), resistor switching to force the dissipation into things you can just let get HOT, and switching power supply techniques, like stepdown converters and ... wait for it... phase control.

I'm going to talk about the phase control idea, as it's fairly heretical. You can get SCRs that are big enough to dim city blocks. In fact, the power industry uses them to do what amounts to just that. So you could adapt a circuit intended to control routers and such to do what you want to do. They're noisy, but there are ways to filter/snub and otherwise quiet their sudden turn-on.

A better way might be to put a very capable MOSFET in series between the rectifiers' (+) output and the first filter cap. A little timing and control/delay circuitry and you can turn on the MOSFET at some point later than the peak of the AC mains wave, and the output voltage goes down proportionately, including all the way to zero if you want, and anywhere in between. The MOSFET stays relatively cool, as it's only on for a short time, and is "saturated" during that time, at high current but low voltage across it, not high current and high voltage like the linear dropping approach.

And since the MOSFET is **FAST**, you can tailor the turn-on response by the waveform you feed to the gate. This will cut down the generated harmonics by slowing down the speed of the current edge. This might not even be necessary, as you still have the normal diodes turning on as slowly as they normally do. I'd have to think about that for a while.

In any case, you get smooth control as far down as you want it to go, and much lower dissipation in return for your efforts.