Disapating 750W of MOSFET heat for under $100

[nAyPDJ]

Well, I've been stuck on this issue for quite a while. I'm trying to dissipate 750W (I'm running the FETs in linear mode).

Here are my current thoughts:

- 8x parallel MOSFETs (13N50H, TO-263 (1.14°C/W ea)
- 2x Chinese CPU fan (supposedly 0.18°C/W)

My idea was to place 4 MOSFETs per heatsink by bolting one of these to each heatsink:



which would have worked out (marginally) fine with +110°C @ 750W, but apparently I did my math wrong and it's actually +174°C @ 750W.

as an aside, here's the math:

Code: [Select]

((1 ∕ ((0.18°C/W)^−1 × 2)) + (1 ∕ ((1.14°C/W)^−1 × 8))) × 750W
Now, I can place two more MOSFETs on each heatsink, but even then I'm still at +138.75°C @ 750W.

So I clearly need a new package. The heatsinks have a reasonable thermal resistivity, but 1.14°C/W on the MOSFET is just crazy.

The TO-3P package is much more reasonable, with the KIA Semicon 8150A having a thermal resistivity of 0.38°C/W, which works out to +103°C  @ 750W with 8 of them, and a nice, cool, +91°C @ 750W with 12 of them.

However I don't see a good way to mount them. I can't screw them into the base of the heatsink because of the heat pipes. Thoughts on how to solve this problem? I was thinking of baking the heatsink in the oven @ 450°F & soldering the MOSFETs on it then… but that sounds like a bad idea.

[wraper]

100W per mosfet in linear mode, good luck with that. Hot spots on the die guaranteed. Even if used as switchers, it would be highly questionable. Not to say that max power rating is given at 25^oC case temperature. Look at figure 10 to cool down your enthusiasm. Also:

Total power dissipation
derate above 25ºC 1.56 0.39 1.56 W/ºC

At your intended case temperature basically near zero power dissipation is allowed.

[coromonadalix]

why not use the to-3p package instead

[duak]

I wouldn't try to solder to the cooler or try to heat it in an oven.

The previous posts are giving good advice.  There are recent postings on this board (as well as many manufacturers websites) discussing the use of MOSFETs in linear mode.  Also, look up "Electro-Thermal Instability".  Also, you really have to discount the specs that manufacturers give until you can test your parts under your conditions.  There's usually some gotcha somewhere.

Anyway, you should be able to press the TO-3P transistors to the cooler with some sort of plate - something that uses the original CPU and cooler mounting hardware.  Each device may require an individual spring, washer and locating pin to balance the pressures.  Many power semiconductors are mounted on heat sinks with flat springs and if done properly, works well.

If the pad on the cooler is too small you might be able to use a heat spreader that accepts the TO-3P transistors that are held to it with screws.  The spreader is then held to the cooler as above.  Copper is expensive and hard to work.  Aluminum is cheaper but not as good.

[nAyPDJ]

why not use the to-3p package instead

I want to, but I don't see a good way to mount it onto the heatsink. usually, a screw would be used, but putting a screw into a heat pipe will let the magic vapor out.

100W per mosfet in linear mode, good luck with that. Hot spots on the die guaranteed. Even if used as switchers, it would be highly questionable.

Sounds like the kind of knowledge only gained through experience .

Not to say that max power rating is given at 25^oC case temperature.

I was under the impression that power derating was only necessary when ambient is above 25°C, but I guess ambient as far as the chip is concerned is the package temperature.

Look at figure 10 to cool down your enthusiasm.

Fig 10 in the 8150A DS is Rdson vs Junction Temp, which indicates an overall resistance of 250-300mΩ with 8 MOSFETS @ 125°C. Still sounds fairly reasonable.

Fig 10 in the 13N50H DS is max current vs case temp, which is also fine since I will only be putting 3.125A through each MOSFET in an 8x. I should have mentioned that I've decided to design for 25A--I don't want to be dealing with wires as thick as my fingers.

Also:

Total power dissipation
derate above 25ºC 1.56 0.39 1.56 W/ºC

At your intended case temperature basically zero power dissipation is allowed.

Oh, I see, that makes sense. However the TO-3P package would be fine in a 12-mosfet configuration:

Code: [Select]

750W/12 − (333W−91°C×2.63(W/°C))=−31.17 W, or 31W of headroom each.


Thank you both for all the great advice

[wraper]

Fig 10 in the 13N50H DS is max current vs case temp, which is also fine since I will only be putting 3.125A through each MOSFET in an 8x. I should have mentioned that I've decided to design for 25A--I don't want to be dealing with wires as thick as my fingers.

As power dissipation is limited due to temperature rise, only small voltage is allowed at that current, not even close to 30V. Safe operation area is also very important but basically not readable in 13N50H datasheet.

[BravoV]

Alternative of TO-3P is much better SOT-227B.

Also securing the fet to heatsink base, you will need the one that has base plate instead of exposed pipes, to spread the concentrated heat. Securing can use two pieces of metal clamping both heatsink base to the FET body, and use some creativity of course. 

Example of a linear fet IXYS IXTN46N50L 500V VDSS 46A on OCing CPU heatsink that has metal plate at the base.



As others had pointed out, you need to watch closely on SOA, the devil is in the details, example of above "linear" mosfet SOA, watch for the yellow highlighted area.

[nAyPDJ]

As others had pointed out, you need to watch closely on SOA, the devil is in the details, example of above "linear" mosfet SOA, watch for the yellow highlighted area.

That's a good point, I didn't even notice that the 8150A (for example) doesn't even specify a temperature in the SOA

Example of a linear fet IXYS IXTN46N50L 500V VDSS 46A on OCing CPU heatsink that has metal plate at the base.

That's a nice setup! The air gaps between the mosfet & cooler at the edges are making me nervous though--you know that air is 0.03W/(m·K), right?

There's usually some gotcha somewhere.

As a @BravoV kindly pointed out, SOAs depend on temperature, and the manufacturer of the 8150A unfortunately chose not specify temperature in their SOA.

Anyway, you should be able to press the TO-3P transistors to the cooler with some sort of plate - something that uses the original CPU and cooler mounting hardware.  Each device may require an individual spring, washer and locating pin to balance the pressures.  Many power semiconductors are mounted on heat sinks with flat springs and if done properly, works well.

There's nothing that quite fits my requirements available off the shelf, but I think I might be able to fabricate my own out of sheet metal. I've attached some initial ideas, I'll try and build both of them and see how it goes. I've never worked with steel before, so this should be fun

If the pad on the cooler is too small you might be able to use a heat spreader that accepts the TO-3P transistors that are held to it with screws.  The spreader is then held to the cooler as above.  Copper is expensive and hard to work.  Aluminum is cheaper but not as good.

I'd rather not take this approach because additional material means more thermal resistance, and more importantly, I don't own mill. Thanks for the suggestion though, maybe someone who owns a mill will come across this post and make use of it!

[BravoV]

Example of a linear fet IXYS IXTN46N50L 500V VDSS 46A on OCing CPU heatsink that has metal plate at the base.

That's a nice setup! The air gaps between the mosfet & cooler at the edges are making me nervous though--you know that air is 0.03W/(m·K), right?

That photos were shots when it was not secured.

Of course there will be TIM in between, I used thermal pad instead of grease.

[Circlotron]

Just a quick note about current sharing with paralleled mosfets; if they are all switched fully on with very low voltage drop then the one that takes the highest current will heat up more and so it's RDSon will increase, leading to a reduction in current for that particular mosfet. If they are in linear mode though, with a relatively large voltage drop, a slight increase in RDSon will do practically nothing. What will happen though is the higher temperature will lead to a lower gate threshold voltage, making the hotter mosfet turn on even more... This can be an issue not just with side by side mosfets but also across the area of an individual mosfet die. A slightly hotter spot asks for more current and this makes it hotter still... This might be an over simplification but I think it is mostly correct.

[T3sl4co1l]

100W per mosfet in linear mode, good luck with that. Hot spots on the die guaranteed.

Could you at least look at the datasheet before jerking knees?

I'm rather more concerned with heat sinking a D2PAK.  What's that board made of?  Is there solid copper inside it?

The better solution is to use resistors, which are smaller and cheaper than heatsinks for the same wattage.  There are methods to maintain linear control while switching resistors.

Tim

[magic]

Certainly use TO-220, TO-3P, TO-247, TO-264 or similar cases. Do the math for thermal resistance from junction to case, case to sink, sink to air and you will know why.

BTW, I have compiled a table of "typical" case to sink resistances quoted by datasheets for "greased flat surface and normal mounting pressure". Start looking for those numbers in datasheets and you will find them. Realize that at hundreds of watts even case to sink resistance can become significant.

Code: [Select]

TO264  .15 °C/W
TO247  .21 °C/W
TO220  .40 °C/W
You may consider GPU coolers too, some are designed to dissipate 200~300W or more.

For mounting, I would try to use the original mounting system of the sink, with a stiff metal plate where PCB normally goes. No idea how heatpipes would handle reflow soldering.

[Siwastaja]

I would question dissipating 750W in semiconductors, or dissipating it at all. What are you building? There might be a better way.

[grifftech]

aluminium pot https://www.amazon.com/dp/B001PZBEKE/ref=twister_B002KMJ8GI?_encoding=UTF8&psc=1

[tszaboo]

You can either
1.) use a FETs which are qualified for linear operation
2.) test a lot of different FETs and have most of them blow up because thermal runaway, hotspotting, FBSOA, and stuff which is not in the datasheet
3.) use a lot of FETs at low temperature (separate gate drive and feedback)
4.) have someone do the homework for you

[mikeselectricstuff]

Might be worth looking at PC watercooling hardware

[nAyPDJ]

I'm rather more concerned with heat sinking a D2PAK.  What's that board made of?  Is there solid copper inside it?

Nope, the plan was to mount the heatsink on top of the PCB, so that the heat is transferred through the plastic chassis. Unfortunately I did the math wrong before designing that board, and that package has too much thermal resistance.

However mounting TO-3Ps directly to the heatsink seems like it may work.

The better solution is to use resistors, which are smaller and cheaper than heatsinks for the same wattage.

You'd think so, but the exact opposite is true. MOSFETs rated for 200W can be found for about $1, but chassis mount resistors rated to 80W cost around $5.

aluminium pot https://www.amazon.com/dp/B001PZBEKE/ref=twister_B002KMJ8GI?_encoding=UTF8&psc=1

I've definitely considered a bucket full of oil

I guess that's my fallback. If I can get continuous power dissipation, that'd be great, but otherwise a big bucket full of oil would be fine for (I did the math!) about 45 minutes.

Realize that at hundreds of watts even case to sink resistance can become significant.

Absolutely. That design that I have in the OP was made before I realized that case-to-sink resistance existed. It marginally would work with 0°C/W case-to-sink, but fails completely with the actual  ~.5°C/W.

You may consider GPU coolers too, some are designed to dissipate 200~300W or more.

This is what really bugs me about cooler manufactures: they do a crap job of actually providing specs for their products. They'll sometimes publish a power rating--meaningless, since they never publish the max temperature. However, they also *never* publish any kind of thermal resistance numbers.

1.) use a FETs which are qualified for linear operation

I'm going to pick this one . Both the mosfets which I listed are qualified for linear operation, my problem is efficiently removing the heat.

3.) use a lot of FETs at low temperature (separate gate drive and feedback)

That gives me an idea... take a look at the attachment. 0.166°C∕W for the whole thing--but unfortunately I'd be a bit over-budget with 375x KIA Semicon KNH9120A costing $232.

[PartialDischarge]

The better solution is to use resistors, which are smaller and cheaper than heatsinks for the same wattage.

You'd think so, but the exact opposite is true. MOSFETs rated for 200W can be found for about $1, but chassis mount resistors rated to 80W cost around $5.

Like the W from a mosfet are the same as W from a resistor...
Some $1 resistors can also withstand hundreds of W, for a few ms.

[OM222O]

if you don't need requirements for near zero resistance, just add a few power resistors in series to drop the voltage before mosfets, that way they have to do less work. for example 1ohm power resistors drop 1volt per amp. using 5 or 10 of them in series would allow you to easily drop about 100 watts with just a basic fan to keep them cool. their absolute value and temperature stability is also not important as they are kept outside of the current sensing circuit.

[Mechatrommer]

Might be worth looking at PC watercooling hardware

his problem is to spread the heat on bigger surface. so i suspect he will need few watercool based anyway as cpu copper contact area is that small. few water based can multiply the cost.

100W per mosfet in linear mode, good luck with that. Hot spots on the die guaranteed. Even if used as switchers, it would be highly questionable.

Sounds like the kind of knowledge only gained through experience

datasheet can looks charming until we learn about heat transfer rate. 400W of that size, and then there's pcb FR4 in between heatsink is not going to cut it anywhere close. you can burn mosfets (while the heatsink still cool) however many you like until the shop is empty. 100W on small package is tough. i saw youtube a while ago that turns mosfets into a nasty water heater in a jar.

[jonroger]

I expect that you need something about 10x bigger than this 70W load:

https://www.tindie.com/products/Kaktus/mightywatt-70w-electronic-load-for-arduino/?utm_source=hackaday&utm_medium=link&utm_campaign=fromstore#specs

Try to figure out an alternative way to do whatever it is you are trying to do.

[max_torque]

Commercial high power DC loads share the power disipation across a LOT of pass elements. I have a Chroma 5kW Dc load that has 296 individual mosfets in it, across  8 parallel "cards" each with a massive high power fan blown heat sink.

If you want lots of power in a small package, then to get the necessary heat flux you'll need a high deltaT, and of course, to get a higher deltaT than the max junction temp of your power silicon you'll need to use sub zero forced cooling.  That might be as extreme as liquid nitrogen chilling if you are really pushing the envelope...


The advantage that dropping power across a dumb resistor is of course the fact that the resistor, being a dumb passive element, can run at very elevated temperatures without failure. 

For example, here is a 50 watt "resistor" that operates at something like 2,500 degrees centigrade




 

[magic]

Nope, the plan was to mount the heatsink on top of the PCB, so that the heat is transferred through the plastic chassis. Unfortunately I did the math wrong before designing that board, and that package has too much thermal resistance.

Frankly, your math is still wrong. The numbers you use are junction-to-metal and metal-to-sink. Junction-to-plastic and plastic-to-sink would surely be worse. It's not completely useless - I have improved cooling of my motherboard's VRM by gluing heatsinks to the plastic tops of SMD FETs, but we are talking maybe 10W, not 750W.

You'd think so, but the exact opposite is true. MOSFETs rated for 200W can be found for about $1, but chassis mount resistors rated to 80W cost around $5.

FETs rated for 200W are rated so if you keep their metal plate at 25°C

This is what really bugs me about cooler manufactures: they do a crap job of actually providing specs for their products. They'll sometimes publish a power rating--meaningless, since they never publish the max temperature. However, they also *never* publish any kind of thermal resistance numbers.

The only real source of information on computer coolers are reviews.
Also, thermal resistance may depend on operating point so it's hard to define, those things aren't simple slabs of metal as you probably already know. The heatpipes need to reach some temperature for optimum efficiency.
Typical operating temperatures of those coolers are tens of °C, at Tj=90~100°C most chips start to throttle clocks or shut down, they can't handle more than that.

By the way, Fairchild UniFET line seems to be usable for linear operation too, here's the SOA of their beefiest parts. Disclaimer: I've never used them, only considered building something similar and gave up because I only really needed <100W of power at the time, which makes things a lot easier.

[EmmanuelFaure]

"Disapating 750W of MOSFET heat for under $100"

The Rth of a common CPU cooler, costing about $20, is about 0.5°C/W. Let's say you want a max delta T = 50K (Between junction and air), this leads to 100W dissipated for $20, 750W for $150. That doesn't take into account all the custom hardware to mount your FETs, the assembly work, etc.

For this range of dissipated power, there's no any heatsink cheaper per watt dissipated than CPU coolers, they're manufactured by millions. Any solution with industrial parts will cost more than that.

[Mr.B]

This idea will not meet your $100 budget unfortunately.
This is what I did when I built my programmable DC load.
The FETs are IXYS IXTH30N50L2 (Single N Channel 600 V 215 mO 240 nC 400 W Power Mosfet - TO-247)
These FETs are expensive - ~NZD 17 each.
They are mounted with the case against the PCB and their thermal tabs against the CPU water cooler.
There are four FETs, two each side of the PCB and two CPU water coolers, one each side clamped together.
If the SOA in the datasheet is to be believed, I should be able to max it out at 200w per FET (Tc=75C).
However, I have not pushed it beyond 100w per FET, 400w for the whole module.
Pics below.