heat sinking EPC GaN fets

Datasheet says 80C temp rise for 1 in^2, 2 oz copper.

That sounds very typical, and you've tested it yourself with other small parts. What's wrong with that?

It'll be even better with vias, or a bit of goopy thermal pad. The squishy stuff, say 1mm thick, would smoosh down pretty thin over the device, and fill in around it almost like potting, touching the surrounding copper. With thermal vias and similar heatsinking on the other side, it'll probably get below 20C/W that way.

To achieve optimal smoosh, you'd need a fairly stiff spring on a heatsink. Possible that one of those square-with-two-legs kinds, with the spring loaded push pins, would do well enough. Or the pin-array kind with a strap shaped spring.

Otherwise, a screw-down one would be possible, given that screws are rather firm, so you should include spacers (or use press-in or soldered PEM nuts, the kind with a collar that extends through the board just the right amount), keeping the heatsink at just the right height (say 1.2mm for a 1mm pad plus various passives underneath it).

Seven Transistor Laboratories, LLC is proud to introduce a brand new product line:

Heatsinks designed specifically for CSP devices, including the world's most efficient and compact switching regulators, EPC eGaN(R) transistors and more!

Products start at $150 in single quantities. Design support available. Call today!

Tim

--
Seven Transistor Labs, LLC 
Electrical Engineering Consultation and Contract Design 
Website: https://www.seventransistorlabs.com/

Me, I'd solder some copper foil at right angle to the PCB.

| | |

Thousands of them, 1.6x1mm. Not that bad.

Excuse me my curiosity, but what's the point in using these tiny parts if you have to supplement them with a relatively huge heatsink? Wouldn't SOT223/DPACK solve the issue without all that complexity?

Best regards, Piotr

A little pcb with copper and vias that clamps on top?

George H.

We have made a PCB with a hole in it, and dead bug flip the device and drop it in and attach the heat sink above onto the exposed mating surface.

Easier to use a bigger package OR even a little tiny heat sink radiator type which you thermally epoxy onto the top of them individually.

The problem with heat sinks is that they get tweaked during handling and detach from their ideal coupling schema at assembly.

So, tiny stand off at all four corners to attach it to the PCB and a surface which sits on top of the device to be cooled that matches up with the difference beteen the stand offs and the package surface so as to not place ANY undue pressure on it. In Other Words, the mating dimension between the stand off mounting and the coupling to the device to be cooled is the critical dimension.

The only low-cost option I know of aside from running the whole enchilada in a liquid is heat-conductive foam. I have used it in the past but with larger dimensions. 3M, Bergquist and Saint-Gobain and main manufacturers.

However, the thermal properties are such that there needs to be enough heat sinking into the board and sufficient flat area around those FETs for the pad to pick up some of that heat. There also shouldn't be very tall or bulky components in that "pick-up area". The FETs are simply too small to pick up a lot of heat from the FET body itself.

Metal-plating on the top of the board where possible also greatly helps. I think you guys use Ni-Au a lot which might work.

--
Regards, Joerg 

http://www.analogconsultants.com/

It's a bare die, so how about silver-epoxying a heatsink to the top?

The GaN is grown on a layer of AlN(*), so they already put the thermal pad in there for you!

(*) Fig 1:

Cheers, James Arthur

The industry has used Silver filled thermal and electrically conductive epoxy for die attach for many years, and the majority uses this:

I'd much rather buy packaged parts. The phemts were SOT89, the ideal size. But the EPC parts are the way they are.

--

John Larkin         Highland Technology, Inc 

lunatic fringe electronics

That's the problem: the available top-of-part area is so small that the silicone stuff won't carry much heat away. The best thing to do would be to push some metal or AlN against the top of the chip, then transport the heat away from that somehow.

A really thermally-good PCB layout isn't electrically best. And I don't have square inches of PCB surface to spare as heat sinks. One interesting circuit has several of these little buggers clustered together, burning a watt each.

That 3M stuff is 2 w/m-k, which is only fair. There is gap-pad stuff that claims 6 or even 8. AlN is like 160.

The pyrolytic carbon idea could work here.

--

John Larkin         Highland Technology, Inc 

lunatic fringe electronics

That could work. I could put three fets in a triangle, or use dummy parts, to define a plane parallel to the board. Then add epoxy and push down a heat sink. It would be messy, and rework would be impossible. Might stress the BGA balls, too.

A heat sink purely glued on the top of one fet wouldn't survive handling.

Well, the AlN is a thin layer; the thermal path is mostly silicon. But theta from the top to an infinite heat sink is low, under 4 k/w. Through the PCB side, with a reasonable layout, we could wind up over

100 k/w.

--

John Larkin         Highland Technology, Inc 

lunatic fringe electronics

On Feb 4, 2018, John Larkin wrote (in article):

What I?ve seen done to handle this and to reduce mechanical shear load on solder balls is to underfill the component with a suitable resin - glue the part to the circuit board (including the copper coming to the leads). Even unfiled resin is a far better heat conductor than air. And silicon rubber is a better conductor than non-silicon resins.

One problem with underfill is to increase capacitance between the terminals and the substrate, but this device already has pretty high capacitances for its size. Grounding the substrate may help.

Joe Gwinn

031690

ad

One of these statements has to be wrong:

undamentals.pdf "EPC?s GaN transistors are insulated from the sub- strate. This allows monolithic fabrication of mul- tiple transistors in any configuration and efficient, common heatsinking WITHOUT THE NEED for an insulating layer between the device and the heat-sink."

"Because the FET substrates must be connected to source potential, the heatsink, must be electrically isolated from at least one of the die or thermal interface material MUST also be an electrical insulator." (emphasis added)

Good top-side info in that second reference, though.

Cheers, James Arthur

What's a boy to do?

The assumptions on page 6 of the last doc are silly and contradict the data sheet. Fig 10 has obvious challenges.

This thermal stuff is always awful. You can't believe data sheets, there is no good way to simulate, and testing is a pain.

--

John Larkin         Highland Technology, Inc 

lunatic fringe electronics

How do you underfill? After soldering? How about rework?

One problem in the GaN fet situation is that the underfill can only dump heat where the pcb is hot already.

--

John Larkin         Highland Technology, Inc 

lunatic fringe electronics

They make those, pick-and-place-able. Look up D(2)PAK SMT heatsinks. :)

Tim

--
Seven Transistor Labs, LLC 
Electrical Engineering Consultation and Contract Design 
Website: https://www.seventransistorlabs.com/

On Feb 5, 2018, John Larkin wrote (in article):

Feed catalyzed resin in from one side using a needle, so the resin wicks under the component.

. .

This is done last, after soldering but before any conformal coating. Conformal coatings can also underfill.

Rework depends on the underfill resin chosen. For big ball-grid components, rework is not often successful, but for a single transistor, it shouild be possible to get a bad transistor off the PCB without damaging the PWB.

Use enough thick copper for the leads, to spread the heat. One can also use the underfill resin to attach a copper foil top strap to component and nearby PWB.

Joe Gwinn

I doubt that the body of the FET has good thermal conductivity as a main path for heat.

What I will do on a current design is add a couple of layers which are not to be used for routing and won't be interrupted for power plane islanding. Then some vias without thermal reliefs from device pad to that plane. Assemblers moan and groan if I do that with 2oz copper but with 1oz or less they are usually ok.

Or have a Westinghouse ice dispenser drop a cube every 10mins :-)

--
Regards, Joerg 

http://www.analogconsultants.com/

spring-loaded heatsinks with grease over bare dice are popular in PC hardware

--
This email has not been checked by half-arsed antivirus software