GaN With the Wind

Slewing a kV in a couple ns may not impress the Lords of Picoseconds as much of a big deal, but the amps/ns sure makes big impressions on PCB traces everywhere!

Grins, James Arthur

And SiC at the upper voltage end. Even the reasonably cheap ones have

800mOhm at 1700V -- what kind of magic is that?!

There is no such problems with the SiC parts, but asymetric HV gate drive is.

I have just completed a 117-900VDC input 47W PSR flyback based on a SiC part and 3C95 planar ferrites. At low line and full load the converter is not even warm. This is insane.

Best regards, Piotr

1 amp/ns is 1 volt into 1 nH. You'd never miss that.

I could tell my drift-step-recovery diode story again.

My Pockels Cell driver ain't bad:

The Pockels cells don't like long-term DC, so the AC coupling option is a couple of caps in parallel. The real problem was toasting the Phoenix terminal block from dielectric losses. The green gunk they use has worse losses than FR4!

--

John Larkin         Highland Technology, Inc 
picosecond timing   precision measurement  

jlarkin att highlandtechnology dott com 
http://www.highlandtechnology.com

I had to make my own GaN gate driver, to drive a Cree SiC fet. -5 to

+18 in about a nanosecond. That took a few tries to get right.

Tiny logic >> GaN >> SiC.

--

John Larkin         Highland Technology, Inc 
picosecond timing   precision measurement  

jlarkin att highlandtechnology dott com 
http://www.highlandtechnology.com

I'm pulling about 10A/ns from the cap. Since I need both lots of bulk storage *and* a precision voltage step, that means I'm stuck with using combinations of caps to get the overall desired effect. A 2012 1uF x 2kV cap with 10pH ESL sure would be handy.

Now *that's* a cool story. But then you're always burning things up and tormenting innocent electronic parts, aren't you? (You've inspired some of that in me, too!) Brute!

I've still got to do the overvoltage-the-MOSFET-gate-for-a-ns trick some time. But I think GaN has made that project moot.

Cheers, James Arthur

Fireworks are illegal in California. Gotta have fun somehow.

I found that adding a small inductor in series with a SiC gate helped a lot. SiC fets have a lot of internal gate series resistance (some, a very lot) so a little L makes the internal/actual gate voltage, hidden behind that resistance, a lot snappier.

C2M0280120D, 27 nH.

--

John Larkin         Highland Technology, Inc 

Science teaches us to doubt. 

  Claude Bernard

The Wolfspeed/Cree C2M1000170J is my all-time favorite FET, a real workhorse. Magnificent package, too.

Cheers, James Arthur

I'm using a C2M0280120D (TO247, 1200V, 280 mOhm on) part on a water-cooled plate with an aluminum nitride insulator. The actual chip inside is tiny, so that huge package is just a heat spreader.

--

John Larkin         Highland Technology, Inc 

Science teaches us to doubt. 

  Claude Bernard

Yes, I have ended up with exactly this part. Started with SCT2750, but the Cree part is smaller and has the Kelvin gate drive pins, which improves reliability.

Best regards, Piotr

just a side comment...

1 ps is pretty fast for ESD switching. I recall seeing a photo I have just completed a 117-900VDC input 47W PSR flyback based on a SiC part and 3C95 planar ferrites. At low line and full load the converter is not even warm. This is insane.

Best regards, Piotr

Good work

Tony

I spent many hours doing similar things, but with Rg=25 ohms there's only so much a boy can do before frying the die.

Lower Rg would sure help -- both native speed, and for extreme over-the-top measures.

Cheers, James

Cree specs the internal gate resistance of my C2M0280120D as 11 ohms, an outrageous figure for fast switching. I've seen other SiC fets that were much higher.

Cree's Spice models look reasonable (except for the substrate diode) and the inductive peaking advantage is similar in simulation and in real life.

I truly believe that if an RC time constant slows down the edge of the actual gate structure on the chip, and I add an inductor to peak it up, no harm is done. I have destroyed a very modest number of fets to confirm that.

3 pF of Crss sounds pretty small, until you swing 1200 volts in a few ns.

--

John Larkin         Highland Technology, Inc 
picosecond timing   precision measurement  

jlarkin att highlandtechnology dott com 
http://www.highlandtechnology.com

Diallyl phthalate?

Joe Gwinn

Looks like it. Lossy!

--

John Larkin         Highland Technology, Inc 

Science teaches us to doubt. 

  Claude Bernard

I don't think so:

Our max frequencies are normally 2-300 MHz from fast edges in inverter drives, so from that point, that is really not so bad, I guess, if we were to use such a stackup

We also do some BLE, that gets trickier

Cheers

Klaus

I've got the fabricated board on Friday. :-( Fabricated here in .de. 10 minutes of drive between my customer and the board house.

All I wanted was 0.635mm of Rogers TMM6 between Layer1 and 2 because on layer 1 there live the coplanar waveguides.

Rest of the stack was don't care: 2, 3 are gnd, 4 is just the lowpass of the synthesizer PLL. The rest could be FR4 of because TMM6 is mechanically stable like candle wax.

They succeeded to put the Rogers layer between layer 3 & 4 which carries just some AF.

AAARGHHHH!

I have cracked the first sample of the 10 GHz puck oscillator board which is only 0.65mm TMM6 when I fastened the first SMA connector. OK, my fault. TMM6 is like wax.

I have TDRed the longish coplanar WG output line of the 2nd sample. No difference to the internal 50 Ohm line of the Agilent

54754A. ADS line calc seems to work as promised. Just the SMA laucher needs a smaller footprint and less tin. Dips down to 40 Ohms over some mm.

Gerhard

Am 22.06.20 um 20:26 schrieb Gerhard Hoffmann:

I made some boards with microwave lam in just the top layer. They came curled up like potato chips.