The 12V level shifter saga

Old-timey planar processing is far from the whole story these days--folks are using most of the periodic table for various tasks in silicon.

Many of the advantages of planar processing (especially the reduction of surface recombination and other stuff like that) can be reproduced using epitaxy to do junction isolation, with deposited films on top of that.

I have no idea how far folks have got with epitaxial processing of diamond, but "no good oxide" is no longer a cogent argument against a semiconductor system, and hasn't been for a good many years.

Cheers

Phil Hobbs (Who used to do his own silicon processing, back in the tunnel junction days.)

So am I. But not in a straight-forward way, I have already checked that and the results are far worse than expected.

I know nothing about short HV pulse generation, but I am sure that you don't make it out of a handful of BC847-s.

This rejection is a result of complexity evaluation against an alternative working solution, which requires 1 part. The booster is composed of 3 parts, which makes it 3x more complex than an LM5101.

Best regards, Piotr

I certainly understand that, and believe me, I feel your pain.

The LM5101 would be great, but it doesn't do d.c. (as you say you need), and it only drives 1.6A up and 1.8A down, well shy of your 12A goal.

It sure seems like someone /should/ have a 12A high-side driver. But apparently (going by your reports), they don't so we've got to do the best we can with what we've got.

Cheers, James Arthur

It's not the only argument. Diamond is metastable at ambient conditions, not stable, and any dopant you can add might transform 'metastable' to 'unstable'. The chemistry problem is quite complex, and most of our silicon trickery won't transfer over. Chemical etching for V-grooves for sidewall MOSFETs? Solders that can be used for mounting to metal? Ohmic contacts? There's a bigger list of problems than solutions.

I don't claim that diamond processing is simple. Even in silicon, relatively small departures from known conditions commonly produce completely different results. (When I was doing my own silicon processing to make SOI waveguides and MIM tunnel junctions, I had about the best possible advisors but there were a lot of things I had to discover the hard way even so.)

If you can get high-level doping, you can make ohmic contacts that way.

Doped diamond seems to be quite a lot more difficult to make than doped silicon. P-type is okay , but N-type presents serious problems .

Fun.

Cheers

Phil Hobbs

That's OK. We need fast PNPs.

LM5101 certainly can do DC, you just need to power the HB pin from an external supply. One thing that bothers me is the integrated bootstrap diode -- it will make the power supply rail selection unpredictable, with possible overcurrent on the bootstrap path. But there are HB drivers without such a diode, so it is not a big problem in practice.

Best regards, Piotr

I recently did a cheap solution with a cascode BJT driving a BJT current source level shifter and a tiny logic buffer bootstrapped on the high side

Cost was just 4 cents, which I considered hard to improve

Rail was 40V and had a 150ns propagation delay to high side

Low side had a bit simpler construction due to lack of bootstrapping

Cheers

Klaus

You can speed that up, to a few ns, and improve common-mode rejection using an LVDS line receiver on the high side, single-ended like you are probably doing but with a small receive resistor up there. Really spiffy would be to do differential current drive up to the LVDS. But all that would cost more.

The LVDS driver idea is good, but a little expensive

It may also have trouble when the switch node is moving with 50V/ns

Could be fun to try it out

Cheers

Klaus

Things like this

work surprisingly well.

I have a similar thing, actually a totem pole, swinging 80 volts in about a ns. Simple circuits that are absurdly naiive with mosfets or bipolars often just work with GaN.

Fun. What do you use for the 4V DC-DC?

Cheers

Phil Hobbs

I usually just buy a cheap brick. The Murata has super low capacitance.

The windings are PCB traces and vias, and the ferrite toroid is buried inside the PCB. Clever.

I like the Recoms and CUI bricks too. There is a common footprint for all three.

A couple ferrite beads on the output reduce capacitive loading even more.

Here's a 50 volt pulse, 25v into 50 ohms.

This one is deliberately LC lowpass filtered to get prettier symmetric edges.

Oh, one gotcha on those cheap unregulated dc/dc things it that they will make outrageous voltages at no load. Add a 1K dumper or something if you drive just CMOS gates.

Nice, thanks.

Cheers

Phil Hobbs

What low-Vgs, low Ciss parts do you like to use?

The EPC bga parts. They are hard to handle, hard to rework, physically fragile, and wonderful.

What's really low is the Crss. EPC2037 is good for 100 volts, 1.7 amps, and drain-gate capacitance is 0.1 pF. As Phil has noted, terminal inductances are picohenries.

I'll have to get a few and measure their noise. I could bootstrap a solar panel with one of those. ;)

Have to ask Simon to do some breakouts in his abundant spare time.

Cheers

Phil Hobbs

We did a bunch of EPC breakouts. I could send you a few, or the Gerbers.

It's hard to solder the BGAs to the boards. These little boards were OK, but we are now doing better "solder mask defined pads."

I now know some things now that aren't on the data sheets.