Spice modeling of active devices (BJT, Mosfets, Diodes, etc) from datasheets?

[Fusion916]

Are you guys aware of if there are any detailed informational guides/books on how to model/test actives devices from datasheet information? Both parameters and supplied curves in datasheets.

[montemcguire]

In general, I rely upon manufacturer supplied models for components, and if they are available, those models seem to work pretty well. However, as you have found, there are sometimes no detailed models for some components, and it would be nice to make models that conform to either measured behavior or datasheet specs. I also know that this is basically what you're asking for, and I have no solution as to how that can be done.

If you're asking for how to make an accurate models from measurements, I can't help you (and I'd like to know as well!)

However, if you're just trying to use commercial components to make a well engineered device, I have found that, of all of the components and manufacturers out there, I can generally find well specified components somewhere, and I will then choose to use those components over others that do not offer detailed models. A good example is for MLCCs. Most of the time, a basic 'C only' model is OK for low frequency uses, but many manufacturers such as Murata and Kemet offer some very detailed 8-12 component models of a 'simple' capacitor, and these models are accurate enough to define the effects of the shape and size of the electrodes, and, if you work with their tool and choose a bias voltage, you can generate a model that will be accurate at a specific bias voltage, useful for dielectrics such as XR7 that 'squish up' when a bias voltage is applied. However, vendors such as AVX, who make really nice MLCCs, offer mostly useless, overly basic Spice models, so I simply choose not to design with those components.

For semiconductors, where there are often no second sources, it's a little more complicated. Many manufacturers will release decent models, but you have to examine the models carefully to see what is modeled and what is not modeled. Open the model in a text editor, see what it says that it does, and see what it's made out of to determine if it has any hope of modeling what you want to figure out. For example, op amp spice models based on the Boyle model do not couple the amplifier output currents to the power supply terminals, making the simulation very inaccurate in some conditions, but not all. At the other end of the scale, some amplifier models provide real passive and transistor devices around the IO terminals, with some idealized components within, and end up providing a very accurate model, even when things are 'going wrong' in the circuit. Sometimes you can find two parts from the same vendor that are the same but have different part numbers, probably from a consolidation of another company's product line, and they provide different models for each part number, so you can try both models and see if they agree or disagree, giving you extra confidence. An example is the LM4562 and the LME49710 - same amplifier, different models.

I don't know which simulator(s) you're using, but the LTspice Yahoo group has a few people who have ground out their own models for some components, and the group has a file archive with a large number of models that may or may not be useful for what you're doing. Join it, and search the archives to see if anything there relates to what you're working with.

Best of luck!

[T3sl4co1l]

Certainly, though I haven't read any personally (I'm not a good reference for books: I just figure things out myself).

To that end, I suggest familiarizing yourself with the functions, the details of how SPICE and models work, and how the real devices work (semiconductor behavior, and for passives, RLC equivalents and frequency response).  That's a lot to cover, but hey, you're not going to hand-write a model worth its salt by using crude simplifications.

That said, there are simple tools for generating BJT parameters, which will give a quick starting point from which you can adjust parameters in a standard test setup.

The most recent big model I edited was: a fix of the STP19NM50N model and datasheet (a model for that particular part is not provided, so I started with the STP21-, which is close enough for starters), for which the DC and AC parameters were all kinds of wrong (compared to the STP21- datasheet alone).  I took measurements of the real thing, fit curves to them, and verified them using the same measurement methods on the SPICE model.  (The DC parameters are still pretty weird, on account of SPICE not having a primitive for deep trench MOS behavior.  But they're more than good enough for switching models.)

Tim