TO-247 dissipation...

150 watts is safe with good heatsinking. Buy a part rated at 300 watts or more.

Water is excellent.

Water is a great insulator if it's fairly pure.

AlN is good, almost as good as common aluminum alloy and about 5x better than alumina.

One trick is to clamp the transistors to a flat copper heat spreader block and then insulate that from the heatsink. That greatly increases the footprint of the insulator.

Why two transistors? Using, say, four might save money overall.

And why not hard anodize? Below maybe 200 volts, it's hard to beat.

John

"Tim Williams"

** The exact number depends on the particular device, but expect a 1 degree C per watt rise between the heatsink surface under the device and the chip.

So 150 watts dissipation equates to a 150C rise while the chip temp limit is

175C - means the mounting surface UNDER the device must remain at or below 25C.

Lotsa luck trying to achive that !!!

** To achieve the near impossible, try a copper block immersed in dry ice. ** No matter what most of the heatsing temp is, the surface under the chip is the one to worry about. ** For what YOU are asking, there cannot be any insulator under the device.

If the heatsink cannot be at ground potential then you need to insulate it from its surroundings to allow direct contact with the drain terminal.

.... Phil

Figure about half rated, then? What about those transistors they optimistically rate at some ungodly number like 455W? Didn't you test some of those, with interesting results?

Oh for the days when tubes were rated with real numbers... and "absolute maximum voltage" meant just DC, not even DC + peak! ;-)

I'd rather not. I'd like to run tap water through this thing -- well, not forever, that shit gets nasty, but suffice it to say things would corrode rapidly given the conductivity of our water after it's soaked through five hundred million year old midwestern dolomite.

Even pure water conducts a little, which means inevitable corrosion. You can get platinum or MMO anodes to take the brunt of it, but that's more hardware. Brass fittings are fine with me, but they won't take DC.

I may consider copper heat spreaders. Need to get a slab of the stuff first, I only have 0.04" sheet on hand. Also need a few large squares of sil-pad or kapton, not hard to get, but more I need to order.

Sometimes that works. Always seems to work with caps... my induction heater tank cap is 200 pieces of MKPs, which cost me 60 bucks. A comparable commercial unit is >$100. Since my time is free, it was a good deal.

As for transistors, a whack of TO-247s is cheaper than some stupid module over $50 (seriously, who would buy that). Beefy TO-220s are a little cheaper than pussy TO-247s, but not by much (IRG4BC30FD is $3 at Digikey, wereas the PC model is $4). Beefy TO-247s don't cost much more than pussy TO-247s (HGTG20N90A4D is $5.18 at Mouser). Say, that's a good deal, I should order some right now.

Any processing is more processing, and I'm not set up to do hard anodize personally (if I even can -- AFAIK it's just cold sulfuric acid, but who knows what else goes with it). Anyway, I've got 320V rails, plus some unspecified AC component -- since the rails are just rectified 240VAC (the chassis is grounded, of course). Add in some galvanic isolation and 200V is too close for comfort.

Tim

--
Deep Friar: a very philosophical monk.
Website: http://webpages.charter.net/dawill/tmoranwms

Depends om the thermal resistance from junction to case, interface to heatsink thermal resistance and thermal resistance of the heatsink ( beware of figures derived using blown air or specific orientation.

Suggest you read ......

Esp look at Fig 35.

Whatever tickles your fancy from simple extrusions to heat pipes.

Graham

-- due to the hugely increased level of spam please make the obvious adjustment to my email address

You don't 'figure' ! You CALCULATE and then MEASURE to confirm. Even some of the cheapest Asian DMMs have acceptably accurate K-type thermocouple inputs and bead thermocouples are dirt cheap.

Graham

-- due to the hugely increased level of spam please make the obvious adjustment to my email address

We tested a lot of fets rated at 300 watts and above, to see how long they would last at 300 watts, clamped directly to a copper block. Most of the ones we tried exploded in under 100 millisecond. We were using "switchmode" fets in a linear application, but even then most of them were detonating at points below their SOAR curves.

ftp://jjlarkin.lmi.net/ExFets.jpg

Spreading thermal resistance creates hot spots on a heat sink, especially an aluminum sink that doesn't have an insanely thick baseplate.

ftp://jjlarkin.lmi.net/Infinite_Sheet.jpg

Silicon is cheap and heat sinks are expensive. So the optimum design - if you can stand the capacitance - is often to use more fets and spread them around some.

IR is notorious for giving parts insane ratings, like DPAKs rated for

110 amps and 200 watts.

MCM has machinable copper.

Sil-pads and kapton are both terrible thermally.

TO-220s are a rotten package for high power.

Maybe bolt the transistors to aluminum blocks and insulate the blocks from the heat sink. But the best insulator is no insulator.

This is an electrically-hot heat sink with copper heat spreaders.

ftp://jjlarkin.lmi.net/Amp.jpg

John

Yeah, I can put in a hunk of 1/4" plate, or use 1/8" copper. Actually, those are about the same, but the aluminum's cheaper (and I have it)...mmm...

What I'm concerned with is getting ~150W per. Is that good enough for a TO-247 rated at, say, 300+W (even IR watts)?

Yes. But that power is spread out, so it's low power on each. That's what you do when paralleling things, use smaller parts. In this case, the TO-220s still look too crappy though.

Yeah, but how much power do you run through them?

Tim

--
Deep Friar: a very philosophical monk.
Website: http://webpages.charter.net/dawill/tmoranwms

That's a complicated question. It can peak at well over 300 watts per fet for roughly 10's of milliseconds. It depends on the customer's load and waveforms.

There's a microprocessor that samples everything about 2000 times a second and runs a realtime simulation of fet junction temps, and shuts down if it computes them to get too hot.

Peak output is 120 amps at up to about 150 volts. It drives MRI gradient coils, resistor+inductor type loads.

John

"John Larkin"

** Shame John had never heard of Hitachi's lateral mosfets ( released in 1979 ) when he concocted that over complicated POS. Laterals are expressly intended for linear applications like high powered audio and RF amplifier stages.

Using them would have eliminated over 90% of the components on that PCB of his - including all the circuitry being used for device current sharing, SOA protection and thermal protection.

So no need for the off board micro either.

What a waste.

..... Phil

Do I have to give all the money back?

John

I remember playing with some Hitachi TO-3 case FETs around that time frame. I just replaced the BJTs in our audio power amp and made some slight biasing adjustments and they worked fine. That was just to try the "new" technology at the time. That was a 100 Watt amplifier module.

boB

when he concocted that over complicated POS. Laterals are expressly

You must've tested them in some rudimentary (as opposed to complex) way. I can't imagine building something like that and never giving it a big whopping sine wave into a resistor. What'd that do?

Or running DC. Gradients are DC (not that gradient coils generally stand still very long). 150V is plenty of drop to make a large continuous dissipation (if the coil is fairly low resistance).

Tim

--
Deep Friar: a very philosophical monk.
Website: http://webpages.charter.net/dawill/tmoranwms

** Makes you one up on Larkin, then.

Far as the septic fool is concerned, lateral fets are same wacko audio parts that originated outside of the USA.

Therefore they simply do not exist.

..... Phil

You can calculate it from the device thermal resistance figures on the datasheet. That's what they are there for.

Name the device and we could tell you !

Also read ...

And then you'll understand. It's quite simple. Also take into account pulsed operation. That why the data sheet (ought to ) give device thermal time constant curves. Pulsed operation will allow greater dissipation during the pulse period since dissipation is essentially an integrated figure over a second or so.

Graham

-- due to the hugely increased level of spam please make the obvious adjustment to my email address

Never tried that. That's one thing it will never be called on to do.

Imaging waveforms are weird and complex. High duty cycle, too.

John

Cool: I can get rid of all the temperature sensors, ADCs, software, and individual closed-loop gate drivers. So all I need are some nice lateral P and N-fets, each good for 500 volts, 8 amps, and that can each dissipate 300 watts on a 200 degree C heat sink. Oh, I'll need gate capacitance below 200 pF, including Miller c.

Part numbers, please.

John

I already gave some examples in an earlier post. Add IRG4PC50UD to the list.

Tim

--
Deep Friar: a very philosophical monk.
Website: http://webpages.charter.net/dawill/tmoranwms

I wonder how much of that is due to moisture in the packaging. If you have to bake them before soldering them, how can they expect to withstand big heat transients in use?

--
Ben Jackson AD7GD

http://www.ben.com/

If they are going out into the big bad world, with humidity and stuff, that's the condition they'll have to work in. I can't expect my customers to bake their fets before they run the amps.

And has anyone seen a recommendation to bake a TO-247 before soldering it? FPGAs, yes.

John