neat paper on SiC history

As 'history' goes, it's woefully incomplete. SiC abrasive, and grinding wheels, wereused for years before someone clamped electrodes onto a wheel and made the first MOV-like device. Lightning arresters were the big SiC semiconductor application for many years.

And they've been using silicon on beaches for a billion years at least.

--
John Larkin         Highland Technology, Inc 
picosecond timing   precision measurement  

.pdf.pdf

g

el

That's silicon dioxide, which has it's own applications in electronics, but only as an insulator.

The silicon carbide used in grinding wheels hasn't got that much to do with the sort of semi-conducting silicon carbide used in electroncis, but at le ast it is the same compound (and while it does occur naturally, there isn't a lot of the natural stuff round).

--
Bill Sloman, Sydney

A couple of things I've noticed or measured about the SiC power fets:

1. They need a *lot* of gate drive voltage to turn on hard. And they need serious negative gate voltage to turn off fast. May as well go abs max in both directions. Which is related to...
2. Many parts have high internal gate resistances, some above 100 ohms. That makes it impossible to drive the gate+miller capacitance fast.
3. The body diodes are fairly slow.

So, in a lot of cases, one may as well stick with mosfets.

What I want is some kilovolt GaN parts, enhancement things that only need a volt of gate drive.

--
John Larkin         Highland Technology, Inc 

lunatic fringe electronics

The standard spec is +20, -4 volts. But for really fast switching it's useful to limit the gate swing a bit, I've had good results with +16 and -4.

There are also pretty good parts. My RIS-764 6A 600-volt pulse generator, using CREE C2M0280120D, had a 5ns rise / falltime until I slowed it a bit. It makes 10ns pulses across a 160pF load, for femtosecond-laser pulse picking, and can run at a 10MHz repetition rate. SiC rocks!!

--
 Thanks, 
    - Win

I ruined my weekend making a dremel prototype with that same Cree part. My gate driver was primitive, but it did slam from +22 to -10 reasonably fast, so I could at least see the turn-off behavior.

The Cree spice model is insane. It has 5 pins: s g d tc tj, where tc and tj are temperatures but are actually tangled up with the substrate somehow. Spice reports kiloamp currents on tc, in a circuit where kiloamps just aren't available. The prototype is teaching me what parts of the spice model I can trust, and some that I can't.

Is your circuit public?

--
John Larkin         Highland Technology, Inc 

lunatic fringe electronics

It's still a work in progress, but here's a schematic of one of the later versions.

--
 Thanks, 
    - Win

Cool; thanks for sharing that.

--
John Larkin         Highland Technology, Inc 

lunatic fringe electronics

FWIW, I'm driving C2M1000170J's 'on' from -5 to +19V in 8nS through

20 ohms and 'off' from 19V to -5V in 5nS through 10 ohms and a diode.

I chose the highest reasonable voltages in part so my driver would slew through the switching region with a higher dV/dT, and also to ensure maximum enhancement.

That level of enhancement switching 1.3kV into a few tens of pF makes nearly ideal linear ramps, 13nS (rise) / 16nS (fall).

My gate resistors are slowing it down a little, intentionally, but not much.

So, IME this FET is a bit 'soft', but fortunately good enough for the task at hand.

Cheers, James Arthur

Pardon me for being crass, but what companies that make SiC devices do you think will be around in ten years and making money? I Have been retired for 20 years now and don't do any design work. But do buy stocks and pick companies that I think have good management and are in fields that are growing. I look at everything I reasonably can, but always make up my own mind ( it other words I like to know what people think, but may not follow their advice ).

I apologise for posting off topic.

Dan

SiC is so good that it should be around for a while. The figure of merit, Cout*Rds_on or someting like that, just blows mosfets away.

As the cited paper explains, it took a long time to get it right.

High voltage GaN, or doped diamond, would be better.

20 years is a long time in this business.

Cree looks pretty good to me. They also make a lot of nice LEDs.

--
John Larkin         Highland Technology, Inc 
picosecond timing   precision measurement  

Right. Cout = Coss, is an important parameter for making efficient off-line converters, etc., where P-loss = Coss V^2 f. V is the PFC bulk voltage, 300 to 400V. We like f to be as high as practical.

--
 Thanks, 
    - Win

You're switching ~1kV/10 ns into 10 pF loads? Can you say anything about the app?

(Kerr cell/ optical something or other? or more mundane?)

George H.

You can speed it up a little (not that you want to) by adding a little inductance in series with the gate. It makes the step response, measured at the actual internal gate, second-order. RC becomes RLC.

--
John Larkin         Highland Technology, Inc 
picosecond timing   precision measurement  

"into a few tens of pF"

I actually don't know. The customer wants to keep that on a need-to-know basis for the time being, and I didn't need to know.

Cheers, James Arthur

I'm using 2.2 and 1.5 ohms for the high side, with a pair of sot-89 BJTs capable of 4A gate currents.

--
 Thanks, 
    - Win

I don't know if driving this FET harder will make any difference. It looks to be Miller-limited, just as a first guess--it looks like it's integrating the square drive.

It ramps beautifully--add some fast 1.3kV comparators and Highland could use it for a DDG timing ramp. (Well, not quite.)

If these were MOSFETs we could do the crazy stuff like driving the gate to

80V for a few nS, then backing off. (That uses the bond-wire as your series inductor.) But I don't need it, and I don't want to hurt the FETs. They're too nice.

The DPAK-7 is excellent. Great creepage clearances, for one. And, there's a Kelvin connection to the source so that the gate driver doesn't have to fight the source's load-current-induced voltage drop inside the part.

I also thought about driving it cascode with a Si MOSFET. That eliminates the need for dual supplies (I think), and the Si FET could be fairly small and fast.

Cheers, James Arthur

I'm using the same driver as you, no buffer. It doesn't seem to need one. The driver edges are silly fast (it's barely trying), and the waveform at the FET gate is just a couple nS behind, slewing 6V/nS even with my largish gate resistors. SiC rocks.

I slowed switching down deliberately, to stay under the gate driver's

100V/nS guaranteed CMTI spec.

My FETs are smaller than yours, though.

Cheers, James Arthur

At these speeds, everything has to be Z = 50-ohms to the connecting coax. I source the output with 50-ohms back terminated. Thus the FET switches see 100 ohms, or a 6A load at 600V, until a hi-Z reflection returns from the far end of the coax. So we need a fairly serious driving transistor.

--
 Thanks, 
    - Win