There's very little in the signal path. And the 20ns delay to *the MOSFET output* is about as fast as you'll find. My only complaint is that the FETs are too big, too much capacitance. When switching a 50-ohm cable-matching output resistor, a low 70-milli-ohm Ron is serious overkill.
C2 across R2 is anti-helpful. The problem, in lay terms, is that your op-amp is reacting very quickly, but to old news.
That is, it's over-reacting to feedback that is delayed in time. So you're creating a situation where you are already driving Q2 appropriately, but Q1's output hasn't moved yet, and your op-amp then tries to drive Q2 even harder even though its existing drive level was already perfect (if it had only waited long enough to see). C2 makes that worse.
The easy solution is to eliminate C2 and add in Rcomp. Rcomp, sufficiently large, slows the op-amp response until the feedback delay is inconsequential in comparison to the op-amp's now-gradual corrections. In that way, the op-amp is reacting to a realistic representation of the results of that op-amp's last output, and can make appropriate new adjustments.
Another way is to accelerate the feedback to the op-amp so that the op-amp's information isn't so stale, such as with feed-forward compensation. Here, that would mean putting C2 across R1. But if you don't need the maximum speed possible, just slow down the op-amp. It's less ticklish.
Slowed down, the op-amp will make corrections gradually, then have plenty of time to see the effects of those corrections, then make even more corrections. Tada! that's also known as 'closed-loop feedback'. :-)
When you take tpd as of the driver+FET, 20nS isn't bad. But I need sub-nS edge-timing accuracy, so 20nS and a resistor-set slew rate are rather scary to me at a time when I'm worried about logic threshold drift affecting my timings.
I think I can manage about half that delay using hand-picked parts, without getting too fancy, and then having the whole signal path available and under my control.
Also, mine is a totem-pole driver. I'd have to float the high-side LMG3410, pass its logic commands through an isolator that can handle insane slew rates, and power it all with a floating supply.
With the signal isolator, my delay would be approaching 30nS. I'd have to use a similar isolator on the low side to match the high-side delays.
All of that is do-able, naturally. But I'm trying to avoid it, here.
And I don't trust most of the digital isolators when it comes to jitter--who knows how much those modulated transmission schemes jitter? Excess jitter is something this application can't abide.
I do love the LMG3410 though--it is a very tempting concept, and a tempting part.
Wow, Navitas NV6113, GaN, 200V/ns, only $3.38 at Digi-Key. And Octopart doesn't even know about them yet! It looks hard to get heat out of the package, they say limited to 2MHz switching rate.
Hah, you have to register, complete with password, just to see a product list. Well, at least Digi-Key shows three single-FET switches available now. But their two-FET half-bridge would be far more interesting to those of us who like to make HV pulsers.
BTW, WRT their stated 2MHz operation spec, the Co(er) spec, which is 16pF at 400V, that's only P = C V^2 f = 5 watts at 2MHz. But, oops, with Rth-ja = 50C/W on 1-in-sq copper, that's dT = 250C.
But, that's assuming the driver dissipates power in both directions, which isn't true. When the FET is off and something else is charging Co, the driver dissipation is essentially zero. So, the actual 2MHz /
400V dynamic dissipation, unloaded, should be about 1.3 watts, right?
Au contraire! Two amps is way more than I need, and I much appreciate the reduced capacitances. (Not all of us are trying to drive big metal all the time, or launch EMI out into space. :)
There are lots of big wide-gap devices for big-power stuff. But we don't have many choices yet when it comes to making a dinky lil' signal generator.
ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here.
All logos and trade names are the property of their respective owners.