Spice Diode Modeling of Forward Overshoot & Reverse Recovery

What does the real device do under such circumstances?

Win's post, while at a much lower frequency, would indicate nothing visible.

Besides, forward overshoot and reverse recovery are of most importance to the power supply application, not something with a "2-GHz ring".

I have 55 "learned" papers on Spice modeling forward overshoot and reverse recovery.

Their models are grotesque complex because they aren't Spice aficionados, they're equation bangers ;-)

However their models do match the measured behavior quite closely.

But so does mine at 11 lines total... a much simpler approach using controlled sources (that allow equations in their templates ;-)

I _will_ ultimately publish. What I seek yet is a way to simply plug in numbers from a datasheet and automatically parameterize the model's equations. Right now I have to manually twiddle coefficients until I get a fit.

Of note, once fit, the model behavior _does_ track slew rate _and_ peak current values. ...Jim Thompson

Sno-o-o-o-ort >:-} ...Jim Thompson

My post? Do you mean John's scope traces post?

I guess you didn't post it, but someone posted a link to your experiment forward biasing a 1N914 at various rates.

I'll see if I can re-find that link. ...Jim Thompson

-- | James E.Thompson | mens | | Analog Innovations | et | | Analog/Mixed-Signal ASIC's and Discrete Systems | manus | | STV, Queen Creek, AZ 85142 Skype: skypeanalog | | | Voice:(480)460-2350 Fax: Available upon request | Brass Rat | | E-mail Icon at

Thinking outside the box... producing elegant solutions.

As are you ;-) ...Jim Thompson

-- | James E.Thompson | mens | | Analog Innovations | et | | Analog/Mixed-Signal ASIC's and Discrete Systems | manus | | STV, Queen Creek, AZ 85142 Skype: skypeanalog | | | Voice:(480)460-2350 Fax: Available upon request | Brass Rat | | E-mail Icon at

Thinking outside the box... producing elegant solutions.

A model for the Grehkov effect would only be useful if one were designing the doping profile on purpose. But the DSRS application is obscure and it's more practical to just test available power diodes, of unknown guts, for behavior. The best Grehkov diode that I've found was the c-b junction of a power transistor, based on no more theory than a vague hunch.

...Jim Thompson

And what happened to "end of discussion" ?

My error, it was an experiment by Pease, not by you...

...Jim Thompson

-- | James E.Thompson | mens | | Analog Innovations | et | | Analog/Mixed-Signal ASIC's and Discrete Systems | manus | | STV, Queen Creek, AZ 85142 Skype: skypeanalog | | | Voice:(480)460-2350 Fax: Available upon request | Brass Rat | | E-mail Icon at

Thinking outside the box... producing elegant solutions.

AoE will be used and remembered, by the millions, for some years to come.

Are you passing your IC design skills on to anyone? I try to teach my kids everything that I know, and give away all but our few crown jewels, publically and privately. (And none is encrypted!)

When people call with a problem for which we don't have a product or time to address their problem, we help them if we can. Maybe a sketch, a link, or we suggest another source. It's nice to do and coincidentally it's good business.

So don't. I'm not going to either. I was asked for situations where SPICE would fail, and that's one. SRDs are another, as I said. Lots of people do expose those to microwaves, after all. (Improving a SRD tripler was one of my very first engineering jobs, back in 1981-2. I did it with a pencil, quite literally.)

Simulating a 10-wavelength piece of coax is still another, except that SPICE has been hacked up to deal with that specific transport problem. I invite you to try doing that with LCs and see how many it takes.

In other words, if SPICE is all you need for your work, terrific, but there's a big world out there.

It sounds like we're in violent agreement now. Weird.

But not in SPICE.

Tim, with all due respect, that's a very stupid question. And you seem to have an odd definition of "nobody", i.e. "Not Tim Williams".

If not thinking that circuit design is all there is to engineering is "being a useless dick", I plead guilty.

However, I'm pretty sure I've worked in more areas of technology than you have, so I have a different view. One example.

In order to do my antenna coupled tunnel junction silicon photonics work, I had to write my own clusterized optimizing EM simulator, which I still use and which is superior in many respects to anything you can buy at any price. (It's called POEMS, for "Programmable Optimizing Electromagnetic Simulator, and you can read the manual on my web site if you're interested.)

It was fun, but it took most of a year, and I wouldn't have done it unless I had to--at the time there were zero full 3D EM simulators that could optimize device geometry, and I needed one. (I started with a big wrapper script around Berkeley TEMPEST, but that ran out of stooch pretty fast because it was single-threaded, which is why I had to do my own clusterized FDTD engine.) It needed to support free-electron metals, because I was using gold and silver antennas. Vanilla FDTD can't do those because they have an epsilon that is very nearly a negative real number, and the Yee (FDTD) algorithm blows up. You have to add state, usually an auxiliary differential equation, to each volume element containing such materials. Transport could be put in the same sort of way if I needed it.

What I got out of it was the opportunity to develop a completely new optical interconnection technology that nobody thought would work--until it did. You tell me if that's value or not.

I mentioned the 1N914 because that's what Jim's behavioural model is trying to represent. Please tell my why you ignored the SRD example.

And where have I said anything about analytical accuracy?

I'm talking about the general tendency of any sort of unphysical fitted model (numerical or analytical) to become inaccurate rapidly once you exceed some neighbourhood of where the fitting occurred. This isn't specifically a SPICE problem--it happens all over, and usually doesn't announce itself. If there's one thing I hate, it's reasonable-looking wrong answers.

My post was a response to Jim's question about exactly why SPICE couldn't solve transport problems such as carrier dynamics in silicon.

And merely having a settled opinion that isn't the same as yours is not being obstinate. Neither of us is 18 years old.

Cheers

Phil Hobbs

Sure, one doesn't do that sort of thing for no reason. The photodetection example I gave upthread is maybe more apropos--you really do want to use that to guide the design of the photodiode.

I have a device from Excelitas whose propagation delay varies no more than 30 ps over the full width of the active area. Some C-V measurements showed that it has an APD-style structure inside, i.e. a buried layer of high doping that generates a huge electric field when you deplete it fully. Bolting a transport capability onto SPICE would allow optimizing that and trading it off against capacitance and series resistance, and seeing what the whole circuit would do. I'd give my eye teeth to be able to do that.

Cheers

Phil Hobbs

At 1 mA, the rise is slow and capacitances dominate, so he didn't get much overshoot. Slam it harder, and you can forward-bias a 1N914 to lots of volts. Pease's observed change of overshoot with rep rate suggests some very slow diffusions going on.

I've run power diodes at 48 volts forward bias, for a while.

We used (or, actually, couldn't use) some laser diodes that acted like they had big inductors or something inside. A sneak-up-on-it drive pulse would create a big voltage overshoot and a long wait for light to come out. If it was pulsed at maybe 1 MHz, the effect disappeared. One of the layers was acting like a slow PIN diode in series with the rest.

Don't use mine. It was all faked.

They exist.

For example, Infineon BSC014NE2LSU Spice model includes an elaborate thermal section.

Unfortunately it was written by "equation bangers" with a lot of IF statements... a total disaster, so I'm trying to decipher it and remove the thermal crap so I can manage to do a transient analysis that doesn't hang :-( ...Jim Thompson

Actually, the hunch was guided by the observation that diffused non-epitaxial junctions have a better chance of blundering into a good doping profile. Boff discovered the step-recovery effect like that, lucky diffusion profile. Theorists then followed up and optimized it.

Epitaxial junctions don't seem to snap.

The PD case would involve inputting 3D doping profiles and anode metalization patterns and such, hardly Spice stuff. I guess that, knowing all that, you could do a better Spice model. Do you think that Spice model would benefit from some sort of diffusion element, hacked like the transmission line? I just kluge up something that behaves like the real thing over a narrow operating range.

Hacked or not, LT Spice does transmission lines well, and it's a lot of fun and education to play with them.

One gotcha: it ignores the impedance of the low side, like a coax shield, to the universe, so it behaves as if there is an ideal transformer inside the line. That can be useful in itself.

Phil, (I feel like I'm barley touching bottom math-wise, so be gentle.) But can't you sorta solve transport equations by putting in initial conditions, turn the differentials into differences and then step the solution along, with a small enough time step. ? George H.

Manufacturer's models aren't of much use beyond the millisecond territory, because they don't / can't include one's heat-sink and PCB thermal information.

Ugh! Removed the thermal crap and discover that the core model is LEVEL=1. How the frick do designers get reliable results with such crap? ...Jim Thompson