Spicemod??

Come to think of it, can anyone recommend a good book on building SPICE models?

I've never found the executables advertising themselves as model "builders" or "creators" were very adequate. In fact I've found them hilarious... take data, make a model, take data from that model, you get a different model :-(

If you're trying to model semiconductors, you really need to understand the math behind the device model, because parameters tend to be interactive in unexpected ways. My son Aaron and I wrote an executable _many_ years ago (under DOS) that takes a bipolar data stream and gets all of the DC parameters. I probably should resurrect that with a GUI ;-)

Though, now-a-days, MOS is king, and extraordinarily difficult to model.

For chip functions behavioral modeling can be pretty good. Again it's just understanding the math that separates the men from the boys.

I have several books on the subject, none of which has been terribly helpful.

My most common comment is to make sure that the functions you create have no discontinuities... all derivatives exist. Thus I'm fond of fitting TANH to just about everything ;-)

What are you trying to model? ...Jim Thompson

IGBT, in linear mode to boot.

I hear you about understanding the math -- I would hope that any book that I did get would be maybe 50% about what each parameter in each SPICE model means, with the rest devoted to helping you decide when .MODEL isn't enough and a subcircuit must be done, and examples of how to do so with various semiconductor types.

Hymowitz shows a horrendously complex behavioral model on the IntuSoft site.

I'd tend toward using a 2N7000 model (Supertex appears to be best at sub-threshold) plus a PNP model, then scale each until you get a match to the data sheet. ...Jim Thompson

The models that I have do that. Or at least they do the NMOS + PNP, with a few diodes and maybe an NPN for verisimilitude.

I just don't know how much they can be trusted in linear operation, when that's not usually how IGBT's are used.

Suddenly struck me, WHY do want an IGBT in linear mode? What characteristics are you trying to emulate? You might be better off with an MOS/Bipolar combination. ...Jim Thompson

-- | James E.Thompson, CTO | mens | | Analog Innovations, Inc. | et | | Analog/Mixed-Signal ASIC's and Discrete Systems | manus | | Phoenix, Arizona 85048 Skype: Contacts Only | | | Voice:(480)460-2350 Fax: Available upon request | Brass Rat | | E-mail Icon at

Because that's what the customer is insisting on.

A single honking-big device (we weren't able to find Really Big MOSFETs, and would have to use _lots_ in parallel).

Strange, that's what I told my customer!

They _do_ have a record of success with an IGBT in the circuit -- I'm just helping them drive it better -- so it's hard to argue against it.

If I could find a bipolar transistor in a single honking-big module that may fly with them.

Voltage rating? Current rating? Required saturation voltage? Is load inductive? IGBT's are inherently snap action devices. How do you propose keeping them from "snapping" ?? ...Jim Thompson

'bout 100V

'bout 400A

Not very important -- several volts would do OK. Keep in mind that it's used as a linear pass element for the most part.

Resistive

Well, it's that last that is exactly what worries me, it is why I tried to talk the customer into MOSFETs, and it is why I'd be interested in some other alternative. My original proposal to keep them from "snapping" was to keep them out of the circuit entirely -- that didn't work.

I do know, however, that they've had these devices in-circuit in a previous product and are happy with it, so it's hard to budge them.

Got any 100V/400A NPN's lined up, or are you planning paralleling? ...Jim Thompson

I wish I did! I don't think I'm going to budge them off of their IGBT decision unless things start going "BANG!" in the lab. I'm hoping to convince them to do tests on prototypes, though.

It's probably possible to make a linear IGBT with judicious tub and isolation placement.

Can you curve trace one of these dudes? ...Jim Thompson

You probably know this, but just in case you don't, i ought to warn you. IGBTs are thermally unstable in linear mode, I had a guy from IR explain this to me years ago. When driven hard into saturation, they are just barely thermally stable, ie. the current roughly shares over the device area, at least enough to allow rated currents. When allowed into linear mode, even for microseconds, they don't share well at all, and the current hogs to the hottest spot of the die. This allows thermal runaway, even if it only persists for us every on-cycle. So, even if significantly derated and run on cold heat sinks, they will have a HUGE tendency to self-destruct when run in the linear region. When part of the die gets hot, the SOA is reduced, and the (partial) transistor has difficulty turning off.

Jon

I'd use an N-channel MOS plus an NPN:

Drain connected to Collector Source connected to Base Gate as the Input Emitter to a Ballast Resistor (*) Collector/Drain to Load

(*) Probably going to need emitter ballast resistors to split the load up amongst multiple NPN's. ...Jim Thompson

At 100 V, IGBTs have no advantage. Their ONLY advantage, in fact, is low IR drop in saturation. MOSFETs are inherently self-balancing, and most designs share current well on their dies too, eliminating many of the thermal runaway difficulties with IGBTs. Paralleling

My guess is the long-term survivability of IGBTs in this mode will not be good, the thermal instability will greatly shorten the life of the devices. This is a real hidden problem, no measurement from the outside will reveal how much of the die is conducting. The transistors will appear to be handling the load fine until they fail shorted suddenly.

Jon

Tim, What approach did you decide on? ...Jim Thompson

I think we're going to hand fit, and I'm going to be conservative about my design so that it'll still work even if the IGBT's are slower than expected (and I'm _not_ going to be responsible of they latch up!!).

I'm fond of telling clients, "I disagree with your decision, please sign this release acknowledging that I so advised you."

I love it!

John