Power layout for GaN FET

The PCB material doesn't matter, because surface leakage is the critical thing. But 100 volts will be fine.

How much power will that fet dissipate?

There is a book that might help...

by the EPC folks. Lots of stuff about layout and assembly.

Be careful: these parts are not avalanche rated and an over-voltage spike may zap them. L di/dt could get interesting. A series gate resistor might be prudent, to give you some control over edge rates and oscillation.

GaN has been coming along *very* slowly. Products that were announced years ago still aren't available.

The 100V FET is 15m Ohms, so around 5W for 20A.

The 100V version is reasonable for $2, but the 150V and 200V are $10 ea.

They are available at digikey.

Not for UL. If you can get solder mask in there, it might be okay, but you may have a hard time finding a manufacturer that'll do such fine-pitch work. (Or maybe not. How common and cheap is chip-scale stuff getting, anyway?)

It would be prudent to keep the voltage low, perhaps a lot lower. Like

Also... you aren't getting 5oz. copper in there -- it can't be etched / plated to fine enough tolerances. Check with your PCB house for details.

Tim

UL is not too happy about 600V MOSFET in TO-220 either. The pins/holes are at most 40 mils apart.

I guess we can add solder mask or something after adding solder.

Then we would not need GaN. Plenty of MOSFET for low voltage.

If I'm not mistaken UL does not care about insulation spacing EXCEPT where it is a saftey issue, i.e. from line to chassis or line to output side that someone can contact.

If you are talking about spacing between nodes within the circuit that are not connected to chassis or "touchable" UL does not care about the spacing.

Of course you should care becasue you don't want it to fail, but that not failing spacing criteria in not the same as ULs saftey spacing criteria.

Mark

Probably even less enthusuastic about a 1200 volt DPAK.

We can pot it with plastic or ceramic. Or perhaps we can just die bound it in a standard TO-220. Wondering why EPC does not offer TO-220 like everyone else.

You can also conformal coat where spacing is tight to greatly increase the voltage spacing if necessary.

boB

Well, that's not as bad if it's the usual two pin model.

Tim

TO-220 is a relatively old and high-inductance package. Short leads tracing a well-defined and predictable path become more attractive as components get faster and people mount them more carefully.

Speed is not so much of a problem. But to get so much current in/out a tiny chip, we need special considerations. Die make sense for high volume, while TO-220 would make sense for low volume or prototypes. They should have both available.

Den tirsdag den 19. august 2014 19.55.17 UTC+2 skrev snipped-for-privacy@gmail.com:

but why GaN?

-Lasse

So why at all do you need a GaN-based part? The GaN HEMT/MOSFET cascodes exist exactly because of their far superior switching abilities.

As Bill told you, the high intrinsic inductance of TO-220

• extremely low T_on/T_off of the cascode => makes no sense.

Best regards, Piotr

GaN is phenomenally fast and easy to drive. The switched-power/gate-charge ratio is absurd compared to mosfets. And the EPC package can get the source inductance way down, where source inductance is usually the killer for really fast switching.

But GaN is expensive, fragile, and sole-sourced, so it's a risk to design in. Unless you need to switch tens of amps in a ns or less, or do some military wideband RF thing, it's probably prudent to stick with mosfets.

Less than 500pF input capacitance and less than 10nC gate charges.

Perhaps 2W average with 1/3 duty cycle.

The dice size is not particularly small, compared to other MOSFET. PCB heat dissipation problem is the same as other TO-220. It's just that EPC does not deal with the packaging issues, so we have to.

For higher volume, we would prefer to do the packaging. But the upstart cost is higher, it would only make sense to package at least 1000 in TO-220 or DIP carrier. EPC can probably do higher volume and much cheaper as well.

I am thinking about a 4 pins metal case DIP around 0.3" square.

If you repackage the EPC part, be careful about adding source inductance.

cost is higher, it would only make sense to package at least 1000 in TO-22

Yes, of course. We can live with a little lower switching frequency, but s till much higher than MOSFET. It would be like additional layers with high er current, perhaps with embedded gold wire. It's easier to do it with a s mall carrier board than with the main PCB.

My production people say that conventional assembly (solder paste, p-n-p, reflow, wash) wouldn't be a problem with the EPC parts. I may use them soon as gate drivers for some kilovolt mosfets.

Mosfets have a sort of charge gain-bandwidth parameter. At some speed, the peak gate current gets to be equal to the drain current, so if you want to switch 100 amps in a couple of ns, you need 100 amps of gate drive. These EPC parts look like good mosfet gate drivers.

I like the low gate drive voltage of these GaN parts. You can do fun stuff driving them from TinyLogic buffers.