GaN FET in linear mode?

We are now designing a multisection snap-apart 4-layer proto board with a bunch of different GaN and SiC circuits. Different engineers will contribute circuits, manufacturing will build some boards, and we'll break them up and distribute them for testing. We hope to have them built in 3 to 4 weeks.

Lots of the parts couldn't be soldered by hand, but my production people are really good.

We can add a tile or two for you if you get us input.

Consider a BUF602 as the gate driver.

How did you get on? I'm doing something similar.

Wound up using a PNP high-side driver instead. It made life easier, and it turned out that there was an easier method to achieve wide, stable bandwidth in my gizmo--the cube rooter in my recent "Cute power law circuit" post. (See also the "Thermal Faraday Shield" thread for more detail.)

When implemented in Class H, the output of the thermal Faraday shield gizmo wants to go like the cube of the Class H voltage: heating goes as V**2, and the RTD bridge output goes like V, making V**3 overall. Assuming a 1-pole rolloff, the loop bandwidth wants to go like V**2, so for a 10:1 range of heater voltage, the bandwidth would vary 100:1.

In a MCU-supervised loop, one can use an MDAC to change the loop gain over a wide range like that, but it's harder in analog.

The cube rooter makes the bridge output voltage linearly proportional to the control voltage instead, and so keeps the bandwidth pretty well constant. Since it's inside the loop, its offset voltage doesn't matter, and the slope doesn't have to be that accurate either--I wouldn't care about a +-20% gain variation over the full heating range.

Using three dual transistor packages, with a total cost of 30 cents or so in modest quantity, makes a cube rooter that's easily good enough for the job. The guaranteed datasheet specs, plus a very little thermal design, are enough to keep the gain variation at the 10% level. The error from the cube root curve is better than that.

Cheers

Phil Hobbs