LT Spice diode C-V graph

INCLUDED AT THAT URL: In the 1997 British Channel Four series "Brass Eye", host Chris Morriss campaigns against spherical cows, in a reality-meets-comedy segment, he distributes flyers and gets public opinion on the use of spherical cows as food, implying that such cows exist and are secretly being sold in British shops. Response was universally negative, but not one respondent questioned the veracity of the spherical cow's existence.

Thanks for posting that URL. I had never heard of this reference before.

The humour still eluded me as I read the description and thinking, "Of course cows in a vacuum would be spherical!" but then I realized that wasn't the point. sigh, it's been a long day.

I've been talking about LTspice's spherical cow universe on and off for the last couple of weeks.

Cheers

Phil Hobbs

No simulation is perfect. Are you dissatisfied with the results of the LTspice simulations, or does the spherical cow reference apply to all simulations?

I believe the value of LTspice is you get answers that are close enough for further study - breadboarding, or going directly to pcb. But you have to be aware of the limitations of the models used, and how to get around them. This is the same as being aware of the limitations of your test equipment - probes, scopes, dvms, etc. Nothing is perfect - you simply have to work between or around the limits.

Used properly, simulation can save a great deal of time and effort, especially if there is a flaw in the original design and it is picked up in LTspice. For example, I just spent considerable time trying to make a forward converter PWM power supply that would generate a variable positive voltage when needed, then a negative voltage at low output. The negative voltage would be produced when the PWM duty cycle was low, since the load was a sink to a negative voltage.

After fighting the various problems getting a TL494 regulator to function due to major problems with the model, I finally got the system working.

It instantly showed there was a major flaw in my idea. As the output voltage decreased, it would eventually hit the point where the output rectifier diodes started conducting due to the sink current, and the output voltage could not go below that point.

This saved a great deal of time and effort wasted breadboarding a flawed design.

However, this led to a very simple solution. I now had a functioning TL494 model, and it only took a few minutes to convert the circuit to a series PWM switch running from a positive voltage. Since it had no rectifier diodes, it did not suffer from the diode turnon, and could easily swing between positive and negative output voltages.

This effort also showed me how to balance the TL494 feedback compensation against the output filter inductor to give a well-damped response. This information is completely lacking in all the TL494 references I could find, and the procedure is much simpler than going through pages of math that could not possibly give the detailed insight that working with the model gave. You cannot see the waveforms by looking at the equations.

So you may call it a spherical cow universe. I find it to be a very useful and functional tool.

There are two general kinds of simulations, I think. The first kind is where the physics is well known but the boundary conditions are complicated. Most kinds of electromagnetic simulations are like that, and so are lens designs, FEM simulations of mechanical systems made of rigid parts, static stellar structure calculations, and stuff like that.

Most of my simulators have been like that. I've written a large-scale optimizing clusterized FDTD simulator, as well as simpler ones for thermal transfer and ray tracing.

The second kind is much squishier, and includes circuit simulations, oceanography, weather, climate, dynamic stellar structure, predator/prey studies, yada yada. Those require an enormous amount of ground truth to calibrate them, and even then you can't take anything for granted.

IC fabs work pretty hard to make their models match reality. Semiconductor companies don't take anything like that level of care in making models for board-level designs. And that's a very, very simple case compared with dynamic stellar structure and *a fortiori* climate, which is still very poorly understood and has a zillion adjustable parameters.

Yup.

Spherical cows aren't useless, they're just approximate. Board-level SPICE models are like spherical cows because the models are poor and because one rarely has a really accurate model of circuit strays. Many folks seem to have an inordinate amount of faith in whatever comes out of a computer simulation.

If you go back and look at what I actually said about spherical cows, it was in the context of somebody liking one of my simulated designs more than I thought it deserved.

Cheers

Phil Hobbs

It's hard to raise spherical cows; they keep blowing away (except in a vacuum.)

Yes, they also believe or misinterpret what their scope is telling them.

Circuit strays are a problem in simulation. They are an even worse problem in pcb layout. But with LTspice, you can quickly find which nodes are sensitive, and how much stray capacitance or inductance can be tolerated. You did this in your 1nA calibrator, and it helped you to complete the project successfully. It would be difficult to do that with conventional methods, since you are facing ground bounce, probe problems, and low signal levels.

Which one was that?

Maybe you do not give yourself enough credit. I find your designs to be uniformly superb. Maybe there might be some nits about minor details, but please be assured I am in awe of your design concepts.

Nobody here designs the way you do. I can look at your designs from all kinds of angles, and learn more with each approach. Nobody else has posted such rich designs.

That's actually a feature--remember the theme from Rawhide? (I actually know it only from the Blues Brothers.)

"Keep 'em rollin, rollin, rollin, Though the streams are swollen, Keep those dogies rollin, rawhide! All kinds of weather, hell bent for leather, Keep them dogies rollin, rawhide!"

Much easier with spherical cows, at least if you're going downhill. ;)

Cheers

Phil Hobbs

Larkin posted a Wikipedia URL for the round cow, but now I seem to have missed how that applies to LTspice ?? as a round cow universe ?? Arrrggg I think I'll give up subtle thought.

A very young Clint Eastwood played Rowdy Yates. I remember that, back when half the shows on TV were Westerns.

I met Clint once. He was mayor of Carmel, and owned the Hog's Breath saloon. He'd tend bar now and then. I also met Johnny Weismuller (in a bar in New Iberia) and was kissed by Hopalong Cassidy (at a Mardi Gras parade, when I was just a baby.)

Yes.

All models / simulations are approximations. No one really knows if there's anything beneath quarks and such -- Standard Model stuff -- but that just means, at the energy level and precision that atom smashers currently have, we can't tell if there's anything more. It's the most exact we can determine at this time.

That said, computational models (FEA I suppose?) can evaluate stuff up to the level of atoms. Not terrifically well I guess, but I don't know what the literature is up to on that.

Presumably, one can do a complete three-of-four-fundamental-forces simulation up to the level of atoms or even molecules, at which point, hot damn is your model ever going to be slow, you really need to start making approximations because you're calculating a lot of redundant information -- on the time and space scale of electron orbitals, the nucleus doesn't change or move at all (unless you're waiting for that rare moment when an unstable nucleus poops something out, but there are better ways of playing with that sort of thing).

Theoretically possible, but computationally intractible, to do anything more -- say, simulating a transistor from the atomic level to the electrical and thermal bulk level.

So from the lowest level to the highest level, you pick the most important part and go with it for your problem at hand. Just be careful that you're taking the right parts, and that the other parts are sufficiently small not to mind.

This sort of thing has always been in physics. While putting together an equation you have to integrate, you can tell which terms will be more or less important under different situations, temperature or scale or whatever. Then instead of doing one intractible integral, you do two (or more) hard integrals, and say "for this range we use this, in that range we use the other one, and in this middle range, funny stuff happens". Which really is usually just interpolation, but sometimes not so nice (e.g. the discontinuity in Cp on 1st order phase transitions, a notoriously hard problem; which I don't know offhand if it's considered a solved problem now, and to what level, or what).

Tim

spherical cows are easy to model. as are diodes that doen't break down,

In the real world it hard to grow round cows* and petavolt diodes.

(*) cows start out spherical but by the time they're born they are considerably less ideal.

This is a bunch of highlights, interesting, brief and easy to understand.

go

Ummmm, no. Comparing octave to Python is similar to comparing a spreadsheet to Python; or apples to lamb chops. Just not the same class of thing. Octave is more of a functional language and python is an imperative language.

That said, i have seen some damn strange bits of software: like an extension to Lotus 123 to do WYSIWYG word processing, an extension to Dbase 3 to do financial calculations including compound interest and PV, an extension to WordPerfect 5.1 (DOS mode) to do presentation graphics, and there are a couple others that i forgot while typing this up.

?-)

where the physics is well known but the boundary conditions are complicated. Most kinds of electromagnetic simulations are like that, and so are lens designs, FEM simulations of mechanical systems made of rigid parts, static stellar structure calculations, and stuff like that.

optimizing clusterized FDTD simulator, as well as simpler ones for thermal transfer and ray tracing.

studies, yada yada. Those require an enormous amount of ground truth to calibrate them, and even then you can't take anything for granted.

making models for board-level designs.

I suspect that they do, but only for the VBIC the models they make for their own use. The Gummell-Poon models they make available on their web-sites aren't a good.

Specifically, they don't model inverted mode operation of bipolar transistors very well - as I found out when I tried to model "squegging" in the Baxandall Class-D oscillator.

Baxandall only refers to "squegging" in a footnote at the bottom of page 752, and the name is misleading - it is reported that it doesn't happen if you use MOSFET switches, though that creates its own problems.

I'd love to see whether a VBIC model of a bipolar transistor would squeg in an LTSpice simulation - there must a non-proprietary VBIC mdoel around somewhere, but I've yet to come across one.

they run happily at -100 KV.

Since all that John Larkin knows about climate modelling is what he has learned from denialist propaganda planted in the Murdoch media, his imagination probably is the best data source he has got, not that he knows enough to appreciate this.

Actually, I'm being unfair here. John Larkin has actually posted links to denialist propaganda posted in the UK Daily Telegraph which isn't a Murdoch newspaper, though owned by people who are no less right-wing.