Spice Diode Modeling of Forward Overshoot & Reverse Recovery

I've used it fairly often, most recently in that $2 diode sampler, where I had a 50-ps stub with a mismatched load to help shape the sampling pulse. Being able to twiddle that in SPICE was a great help--breadboarding with 0402s in mid air is hard. (Chester W and prototyped it like that back in December--worked great.)

I'm really not knocking LTspice, which is a really nice and very capable program that I use all the time. It just isn't the One True Numerical Solution To Everything, even for circuits.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs 
Principal Consultant 
ElectroOptical Innovations LLC 
Optics, Electro-optics, Photonics, Analog Electronics 

160 North State Road #203 
Briarcliff Manor NY 10510 

hobbs at electrooptical dot net 
http://electrooptical.net

I like to run a realtime junction temperature simulation, analog or digital, to protect mosfets. That lets one safely push them a lot harder than a simple current or foldback limiter.

My usual model is scaled power dissipation driving a 1st order lag, added to heatsink temperature. Tau's are typically in the 100 millisecond range. We calibrate the model by destructive tests.

Here's an analog mosfet power dissipation computer:

--

John Larkin         Highland Technology, Inc 

lunatic fringe electronics

Yes, that's a powerful technique, and it you feed it to the right kind of computing circuit, or a micro ADC + program, you can protect the MOSFET.

--
 Thanks, 
    - Win

You can also implement the ideal equivalent to that in Spice, add a heatsink model, and do more things than the crap thermal stuff presently included in the present models.

I've even implemented SOA display for MOS devices with PSpice (*) macros, including hot electron problem coverage for very short channel devices.

(*) Probably doable also in LTspice, I just haven't tried. ...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 http://www.analog-innovations.com |    1962     | 

     Thinking outside the box... producing elegant solutions.

Great!

I don't see anyone arguing that SPICE has doesn't have limitations*.

*Jim might not appreciate the distributed-state property of your examples, like TLs or transport. But that's his problem.

Feel free to call that battle a success. ;-)

Okay, cool.

So, how far can you take it?

Suppose a model can be created, and that it has a finite number of free variables. (Whether approximating it in SPICE, or in an environment that can actually solve transport phenomena.)

How much data do you need to fix all those parameters?

And how many of those can be read off the datasheet, or measured easily?

Without this critical information, your model is again just a curiosity. You have the means to solve it, but no data to enter into the solver.

I suspect this information will be hard to come by, for SRDs. So, though it might be a valuable application, can a usefully accurate model even be created (in any environment)?

(And that's without discussing the PCB, where even trace widths and corners matter to these picosecond signals.)

Tons of things have been simulated successfully with SPICE. That hasn't stopped anyone, despite their models being trivial, utterly unrealistic toys. :)

That wasn't phrased very well on my part -- I wasn't referring to transport (or the simulator you made!), but "some amazing toy (...) that no one needs".

As long as you've brought it up -- "nobody" is a rather peculiar word. Its meaning seems plain, but it's used most often to mean: "no one that I know of". Language is shitty like that...

I've read about it before -- pretty amazing stuff, Phil!

Duly explained above (I suspect the physics-bits aren't well enough documented to make it reasonable to fit, even when a solution can be constructed).

So, you have no concerns about the ability to model a 1N914 at limited dI/dt..? Yes it would almost certainly be easier with transport, but if that's not in your toolkit, how would you do it?

There seem to be two behaviors of interest here: SRD behavior, and the lossy-inductor regime.

Are these behaviors so deeply inseparable, from the underlying phenomenon of charge diffusion, that one must solve for that, to have any reasonable success at modeling just one or the other?

That seems to be what you're saying, in regards to the 1N914. Or if not, then... please, share the model?! :D

That's why you put the assumptions in the liner notes!

If there's one thing I hate, it's models that don't even say what their limits are!

Trivial example: PCB trace impedance calculators that dutifully spit out negative impedances. Ferrite bead models are awful, too, and never specify their useful range. Let alone nonlinearity.

A diode, modeled with forward recovery, with the attached note saying "invalid above 1000 A/us", would be lovely!

Anyone who is sufficiently experienced at SPICE to appreciate the bounds of their models, should be grateful to see something like that. Maybe the limit precludes accurate analysis of a particular application (say if a particular monolithic switcher accidentally spits out SRD pulses..), but it will still be applicable to others (like a more relaxed inverter that has enough stray inductance to limit dI/dt).

Tim

--
Seven Transistor Labs, LLC 
Electrical Engineering Consultation and Contract Design 
Website: http://seventransistorlabs.com

Perhaps if there was a model and solver that could use that data, then there would be more incentive to provide it. Chicken, egg.

Clifford Heath

I have retrieved a few SRD papers.

Looks like I can even model "transport" in Spice.

Nothing is impossible to model with Spice.

Watch this space >:-} ...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 http://www.analog-innovations.com |    1962     | 

     Thinking outside the box... producing elegant solutions.

Wow, between that and the Donald, I may OD on popcorn. ;)

Cheers

Phil Hobbs

I did that last night... and now I am suffering some intestinal discomfort.... I really shouldn't eat popcorn... seems the wine turns it to concrete :-(

But you do indeed need to watch this space ;-) ...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 http://www.analog-innovations.com |    1962     | 

     Thinking outside the box... producing elegant solutions.

Sounds like a job for a nonlinear multi-parameter least squares curve fitti ng procedure.

My Ph.D. project used Fletcher-Powell - the Fortran package solution availa ble from the computation department was based on Marquardt but couldn't be shoe-horned into my software, so I had to write my own, and the Fletcher-Po well approach seemed more suitable for my particular problem.

la

Working within a restricted parameter space, a model doesn't have to be phy sically realistic to work. Physically realistic models can be expected to hold up better when used outside the operating range within which the model was fitted.

--
Bill Sloman, Sydney

Too much popcorn will lacerate your plumbing. Blood.

--

John Larkin         Highland Technology, Inc 

lunatic fringe electronics

[snip] [replying to Phil Hobbs]...

Over dinner last night I was musing about Spice models, my diode model covering forward overshoot and the resulting hysteria amongst the "elite".

It struck me that most of the lurkers on this group are engineers.

All that an engineer cares about in regard to a Spice model is that it's terminal behavior be the same as the real device behaves in his system/PCB.

The engineer doesn't give a damn whether the model duplicates the actual physics of the device or not, only that it performs properly.

I would argue that attempting to model a device following the complete physics of a device produces a cumbersome, difficult to converge and slow-to-simulate model.

I have 55 "learned"/peer-reviewed papers in my forward-overshoot/reverse-recovery research file that demonstrate my point ;-)

I would point out, for instance, that the Spice model for a bipolar device is not a model that conveys the actual physics of the device, but a simulatablely-tractable approximation.

The elitists amongst us would insist that a proper model can't be created unless it "solves" the underlying physics equations.

That is nonsense.

In my musing, not even committed to paper yet, just rolling the function around in my head, I realized that a controlled charge accumulation, metered release could easily model the sacred SRD... no solution of a transport equation involved.

When I posted the waveforms of my forward recovery model, the resident pompous asshole reared his ugly head and pronounced, "...only waveforms".

Sonnuva gun, I do believe that's what engineers want to see on their simulator and their 'scope... not a scrolling useless paper extolling that the physics equation has been solved.

Lastly, I think many people here are using LTspice, and the elitists and the resident pompous asshole hang their hats on the results...

I would suggest you check out the accuracy of the resident Spice models... all based on nearly ideal approximations to emphasize simulation speed... not accuracy.

HTH >:-}

...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 http://www.analog-innovations.com |    1962     | 

     Thinking outside the box... producing elegant solutions.

Not necessarily.

Gummel-Poon doesn't capture inverted transistor behaviour that makes Baxandall class-D oscillators "squeg" when built with bipolar transistors and a too-big feed inductor.

It is. A proper model must respect the underlying physics as much as it can - and the underlying physics is often a royal route to a good model - but - as the Gummel-Poon example makes clear - an imperfect model can often be very useful.

It's nice if it's imperfections are spelled out, as they rarely are.

Phil Hobbs did spell out why this might not work as well as you like to think, and might have given you a clue to an approach that might work better.

Waveforms are the integrated output of complex processes. A waveform has one value at one moment. The process that generates the waveform has several variables all of which are varying with time.

That's what they want to see. This doesn't mean that it's what they ought to be looking for.

We all know that LTSpice outputs can be misleading. The trick is to know when, where and why.

Gummell-Poon is pretty accurate, where it works. VBIC - which LTSpice will run - is supposed to be better, if you can get the parameters for a specific part. Semiconductor manufacturers treat them as trade secrets.

--
Bill Sloman, Sydney

Most of the foundries I deal with (that have processes with bipolar devices) _do_ provide VBIC models (PSpice runs them as well).

Unfortunately most _discrete_ device manufacturers don't provide adequate models, if they provide any at all. ...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 http://www.analog-innovations.com |    1962     | 

     Thinking outside the box... producing elegant solutions.

You missed that my point was about LTspice... LTspice's default models are gross simplifications. ...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 http://www.analog-innovations.com |    1962     | 

     Thinking outside the box... producing elegant solutions.

"Update required"

Update of what?

Google is evil. Post pix to Dropbox or something.

I recently made a pulse generator using a Cree SiC fet. Their Spice model assumes zero forward recovery time of the substrate diode. Wrong.

--

John Larkin         Highland Technology, Inc 

lunatic fringe electronics

Spice Modeling of forward AND reverse recovery is difficult due to lack of real data. Thus my original post...

Message-ID: ...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 http://www.analog-innovations.com |    1962     | 

     Thinking outside the box... producing elegant solutions, 
              by understanding what nature is hiding. 

"It is not in doing what you like, but in liking what you do that 
is the secret of happiness."  -James Barrie