MosFETS on the same heatsink

When mosfets are sharing a heatsink how does this effect the total power handling capability of the fets.

For example if I calculated an allowable 140W dissipation for a single fet mounted on a large 150mm x 100mm for a max ambient of 50C and max junction of 110C With a fan 1 x 42CFM.

Would it be possible to get 300 to 400W total dissipation if I parallel 2 to 4 FETS on the same heatsink? Or would I have to buy 2 or three more of the large heatsink?

The mosfets are operating in linear mode it's for a variable electronic load.

To avoid thermal runaway in a fet due to Vgs( th) differences between devices is it best to use dedicated opamps per fet or large source resistors? I've read several papers but thought I'd ask here for someone who has maybe done something similar.

Reply to
Hammy
Loading thread data ...

Hammy schrieb:

Hello,

if your calculation is right, two of those fets mounted on two of those heat sinks will dissipate 140 W each and 280 W together. Why do you think that one heatsink of the same size will be enough for

300 to 400 W?

Bye

Reply to
Uwe Hercksen

I put 8 x TO220 FETs direct (non-insulated) on flat aluminium plate,

3mm thick by 40mm to spread heat onto one side of 80mm square fancooled heatsink rated 0.3'C/W and could easily handle 400W, with an 8 x 0R33, 50W metal pack resistors on other side of heatsink. Two parts of the heatsink mated to form an 80mm by 200mm tunnel that one bolted a fan to, I used a 90mm fan via adapter. 0R22 source resistor too small to balance FETs better than 100%, so I selected best match eight from batch of 20, expensive and still not a good performer for intended use, okay for manual operation.

If I built one like that again I'd use individual opamps, once saw a site with that method, but didn't save the reference. An opamp version with pair of LM324s didn't work very well, so I think you may need better, faster than LM324.

Also tried 8 NPN transistors instead of N-channel MOSFETs, problems with drive, leakage, temperature drift :(

What I ended up with was fine for a manually adjusted active load, but proved unsuitable for the power DAC I was hoping to convert it too. Got a PWM temp controlled fan, so it's quiet until driven to high power, and, it could suck 400W all day, 500w for short periods, with a temperature sensor to shutdown on overheat, of course it's not properly documented :(

Many rebuilds, and the cap banks I had to hang off it for stability with some power sources like a mains commutated SCR controlled power supply, ugh!

Current version power DAC I'm building is saturated FETs driving resistor banks: P channel for a hi/lo range switched resistor banks.

N channel FETs for thermometer code drive for MSB 3 bits, then binary weighted resistor banks filling out to 63/64, then a couple power opamps (3A max output) catching the fine 1/64 end, another for trimming the 1% resistor bank slop. Opamps driven by dual

8bit DAC chip, entire hybrid power DAC by a PIC chip with 16bit resolution ADC for feedback. Forgot to order crystals for PIC,

Grant.

Reply to
Grant

The total thermal resistance from junction to ambient is (110-50)/140=0.43 C/W, which sound very optimistic :-).

The thermal resistance from junction to ambient Rth (j-a) consists of the thermal resistance from junction to case Rth (j-c) in series with the thermal resistance from case to ambient Rth (c-a). The thermal resistance from junction to case can be found from the transistor data sheet.

Assuming that the Rth (j-c) is 0.20 C/W (what kind of package is this good?), thus Rth (c-a) would be 0.23 C/W, since 0.20+0.23=0.43 C/W

Putting two transistors on the same heatsink will effectively divide the Rth(j-c) by two, but it does not affect Rth (c-a) thus Rth (j-a)=

0.20/2+0.23=0.33 C/W. Thus P=(110-50)/0.33=182 W.

With 4 transistors on the same small heatsink (assuming the extra transistors do not disturb the air flow) Rth (j-a)=0.20/4+0.23=0.28 C/W and hence P=214 W.

Reply to
Paul Keinanen

Actually, it might. One transistor dissipating, say, 100 watts on a heatsink will suffer from hot-spot effect, namely the heatsink lateral spreading thermal resistance. That same 100 watts shared among two transistors, at 50 watts each, would have lower case temperatures. The issue is that heatsinks are usually specified assuming a uniform heat load, but transistors are very local spots of heat. The thinner the baseplate, the worse hot-spots get.

John

Reply to
John Larkin

They spec a thermal restiance of 0.08°C/W using the fan for this

890SP-01500-A-100 Heatsink .

formatting link

This is unfamiliar teritory for me I've never had to dissapate a large amount of power, so is that realistic? ;-)

I'm using these FQA24N50 which has a RJC = 0.43 C/W and RCS = 0.24 C/W. TO-3PN package PD @ 25C is 290W.

I see I forgot to add RCS.

I see so I would need two of those sinks with 2 to 3 FETS per. Thanks for the detailed explanation.

Reply to
Hammy

Actually my suspicions were correct. See Paul and Johns post below.

If each transistor isnt carrying the full power I figured it must have some impact on the thermals. I just wasnt sure how much.

Reply to
Hammy

Assuming that you mean that you bolted, or otherwise clamped 8 transistors onto a thin aluminum plate, which you then attached to an aluminum heatsink, how's that going to "spread the heat"?

If I understand correctly what you did, that's going to make the situation worse than attaching the transistors directly to the heatsink. Extra thermal resistance at the interface between the plate and the sink.

--
"For a successful technology, reality must take precedence 
over public relations, for nature cannot be fooled."
 Click to see the full signature
Reply to
Fred Abse

It would be safer to use three or four fets on that heatsink to spread out the heat and reduce the effective Tc-s. Keep the dissipation below

80 watts per, maybe.

Also blast the air *into* the fins, not just around them. That sink has high fin density, so air will prefer to flow around the fins instead of between them.

John

Reply to
John Larkin

Shifting some of the heat away from silicon and into power resistors is a good deal, when conditions allow, which means serious resistance in the source and/or drain. I like these:

ftp://jjlarkin.lmi.net/Welwyn.JPG

An opamp per fet is a good way to balance dissipation accurately.

John

Reply to
John Larkin

That air flow is about 20 l/s or about 24 g/s of air, thus the air will be heated with 4 C for each 100 W dissipated. If there are multiple transistors in the direction of air flow, the last transistors will receive warmer air than the first transistor.

Looking at the picture, it appears that the fin and the air channel are both about 1.5 mm wide.

Just wondering, how such narrow channels will behave in the presence of dust :-).

If the channels are as wide as the fins, the combined equivalent channel would be about 25 x 50 mm or 0.125 dm², thus with 20 dm³/s air flow, the air speed would be 16 m/s, which would be quite noisy.

This depends of the mounting.

Anyway Rth j-s would be at least 0.67 C/W and with Rth s-a 0.05 C/W and the total resistance Rth j-a 0.72 C/W.

With 50 C ambient and 110 C junction, the maximum dissipation is 83 W.

With two transistors 0.67/2+0.05=0.39 C/W, P=156 W (78 W/transistor). With four transistors 0.67/4+0.05=0.22 C/W, P=276 W (69 W/transistor).

80 W/transistor would indeed be the maximum.

Or install the fan in the middle blowing downwards and cover the remaining area, so that the hot air escapes from the ends of the fins. Install the transistors on both sides of the fan, so that the input air temperature is the same.

Reply to
Paul Keinanen

Fine. My processor gets a bit dusty, I blow it out every so often. The = CPU temperature is identical, before and after...

It's probably the first bit of dust that's the worst, the part that you = can't un-stick from the metal surface without thorough cleaning. Like = biofouling in aqueous systems, you try cleaning it off but the worst = part is the hardest to clean.

Tim

--=20 Deep Friar: a very philosophical monk. Website:

formatting link

Reply to
Tim Williams

I've got good results from the configuration you describe, putting the fan in the middle of the heat sink, blowing directly down into the fins, with air squirting out both ends. This is called "impingement cooling" I think. The fets in the middle have a less air velocity but better lateral spreading effect, and the fets on the ends get a lot of air velocity, so it works out pretty well to distribute the fets fairly evenly, just avoiding the very ends of the heatsink.

I have this crackpot theory that the back pressure caused by the air scraping the fins should reduce the fan air volume to about half of the fan's zero-back-pressure flow spec. Sort of an impedance matching thing.

The other important thing is to keep Tc-s low by having the heatsink be really flat, and not use insulators.

If you're really compulsive, use an opamp per fet to really control the currents, and tune the gains to equalize real-world fet temperatures.

John

Reply to
John Larkin
[SNIP]

I'm not going to be using insulators. I might use two fans mounted directly overtop.

Thats what I was thinking. I've been looking at Newark for a decent quad op-amp but the pickings is pretty slim. They just have old 324'S or similiar.

Reply to
Hammy
[snip]

The P4 Heatsink for my PC is the same and dust does stick, it doesnt even come off with compressed air. I have to use cue tips to rub it off.

I'm not useing any insulators the heatsink will be live.

Its going to be one of those trial and errors to see where the fan performs the best. I may use two.

Thanks for the examples. :-)

Reply to
Hammy

Lowered the MOSFET tab temperature, compared to bolted directly to the heatsink. Problem is that some of the nice looking heatsinks don't conduct heat very well.

No, a heat spreader improves things, oddly enough, by spreading the heat. I started with insulated tabs, no go at all :( This is where practical or empirical knowledge will beat your calculations hands down.

While I can't tell you what the different alloys were, there is a difference in colour and chip forming on drilling the aluminium plate and the heatsink material I was working with.

And, if you look up difference in pure aluminium vs the various common alloys, there's a great difference in thermal resistance.

On top of all that, I work with materials easy to find, I'd really enjoy being able to mill out a heatsink from chosen material as I recall Jim T. doing for his fan cooled power resistor.

I didn't want to use oil-cooled or water cooled because they're too messy for indoors. Boiling water is very efficient for wasting a lot of energy.

If one is game, hooking up a fully controlled bridge and pumping the energy back to the mains is even better :) But the potential fault currents stop me going there.

Diesel trains dump their motion into heating air, seems it's the best method that scales.

Grant.

Reply to
Grant
[snip]
[snip]

My fan-cooled heat-sink was simply a conventional finned shape, except I had 2X 6-foot lengths of it. Someone traded it to me for an alternator. (When I did alternator regulator designs I had Ford, GM, American Motors and Chrysler alternators in piles in my garage ;-)

I milled the edges (for a bolt-together fit) so that I had 4-way inward-facing fins with a square outline that fit a muffin fan on the end exactly.

Total length was about a foot.

If I still have some of that extrusion I'll post a photo. ...Jim Thompson

--
| James E.Thompson, CTO                            |    mens     |
| Analog Innovations, Inc.                         |     et      |
 Click to see the full signature
Reply to
Jim Thompson

Nice, they're not cheap though.

I'm using cheaper ceramic air cooled resistor banks for a hybrid power DAC I'm putting together.

Yes, but I think something better than LM324s :) I didn't have much joy with them, but the friend I built it for gave me a handful of logic MOSFETs, they were difficult to get going in linear mode :(

At least with the construction methods I was using back then.

I had a very nice 300W ultrasonic oscillator at one point, didn't manage to stabilise it for the range of power converters I was trying it out with.

I put couple photos of my eight transistor linear active load up here:

formatting link

Showing the metal 50W resistors I used, and a view of the transistor side, complete with a ridiculous current shunt made from quite a few

1% 1/2W resistors.

Grant.

Reply to
Grant
[snip]

I have some dual LM7322MA use 1 for each FQA24N50. I still might use a buffer though 4.5nF input capacitance fets so it wouldnt hurt.;-) I want to be able to do load steps. Probably use a NJT4031 and NJT4030P push-pull NPN-PNP. Those are 3A bipolars with a minumum gain of 100. Should be able to pop those FETS around quickly.

That is quite the beast. :-)

I want to keep the size down maybe shoebox size or there is no point in me doing it. That way I can just put it on the bench vs lightbulbs and fifty wires running all over the place when I want to test something.

I was just figureing on two resistors per leg for shunts. 1 resistor

3.3 ohm for upto 1.5A 1W with another resistor 0.1ohm 2W UPTO 10A. Those current values would be divded by 4 so power dissapation is resonable.A simple toggle for switching ranges.
Reply to
Hammy

If I've learned anything about air flow, it's that I don't know much about it. Air is peverse and counter-intuitive. When we do an enclosure design, we use air flow meters and incense sticks to tell how much air is flowing and even in what direction it's flowing. I've seen card cages with a big fan tray just below the boards, blowing a hurricane of air up, and found a few card slots where the air was leisurely moving *down*. We make wood and cardboard mockups and push them around util we measure stuff that will work.

So, make provision for moving those fans around, and measuring temperatures, and do whatever works.

John

Reply to
John Larkin

ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here. All logos and trade names are the property of their respective owners.