How do manufacturers create SPICE models of their IC's?

[shredder929]

There's always requests for various SPICE models, but how are the manufacturers making these models? Is it all part of the process of designing the IC, creating the schematic and then adding in stray capacitance/inductance/resistance in SPICE and then exporting it as a binary subcircuit?

If I were to design an analog IC like a simple OTA in SPICE, could I create a model that users could implement?

 

[sleepless_nights]

Hi shredder,

I don't know if I properly understood your question, so my apologies in advance if I got it wrong.

If you are referring to the modelling of each device, the process engineers extract the physical parameters of the devices. Then, using model parameters well known and some adjust parameters, they basically model the devices doing a "fit"of the extracted parameters curves with the model parameters.

I'm not exactly an expert on it, so maybe someone else can explain it better. But, as a reference, I can suggest you to take a look at Daniel Foty's book called "Modelling MOSFETs with SPICE".

Best regards!

Vitor

 

[dick_freebird]

There's a difference between the "model" supplied to
customers for their use, and the "model" I make over
the course of chip circuit design. My "model" is transistor
level and complete down to the individual devices and
their detailed SPICE / Spectre models, from data pulled
from physical samples and fitted by someone who can
run the automated model software (most models are
"empirical" and have too many params for humans
to fit; not like the old Level=3 days).

What's provided to customers (IME) has been a
macromodel made of controlled sources and passive
elements. Foundries do not let their proprietary device
models out without a NDA, so publicly available
apps collateral is scrubbed of all that. In the process
of macromodeling many subtleties are lost, that the
design circuit model would show (usually) if asked.
Of course even "real" transistor models often fail to
emulate things like breakdown conduction unless
somebody supplements them (e.g. a subcircuit model
that adds a couple of diode primitives, which do show
breakdown, on top of the BJT or FET that does not).
In places I've worked, the applications engineering
group had dudes who would gen up an op amp
model that acted like the datasheet. They may or may
not have done much work to verify that it matches
silicon, any further than that. After all, it's only a
marketing tool provided for free.

 

[shredder929]

Thanks for the response! I see, so the process or verification engineers figure out the specs and response curves that will go in the datasheet, and then use that to create a SPICE model of controlled voltage and current sources and passive elements that closely simulate how the actual IC would respond, but don't use the actual transistor level design in any way. Do I have that correct?

Would it be accurate to say that SPICE is useful for IC design and quickly simulating passive elements but not as useful/robust for board level design with IC modules?

 

[d123]

Hi,

In the free version of Tina-TI you can model transistors (no idea how complex or simplistic as that level is way beyond me) and even better, make block models to reduce whole circuits to IC-sized parts, it's pretty comprehensive, I am under the impression it can slow down simulations sometimes. You can create netlists, afaik.
All SPICE-based sim tools must have this option somewhere in a menu.

Imagine the details and time needed to make just a pcb track + solder joints (Seebeck Effect) emulate pcb hot and cold spots, and so on, give more realism to the simulation and results.

I think simulation tools are best-foot-forward approximations to mostly perfect realities, presumably idealized and unable to emulate an IC/a passive component/a board/circuit 100%: part-to-part variations, layout, operating environment, etc... and should be taken as accurate guides to component and circuit operation only.

Things are always much better, and sometimes fantastically misleading in simulation world, as you know - e.g. at the extreme end of unbelievable, I can make 1ppm voltage references from 0°C to 100°C with four resistors, two BJTs and one generic op amp ...

 

[d123]

Hi again,

Just wanted to add, I think that if you understand all the options you can modify in a decent simulation tool, and by tailoring passives at least with the multiple options offered for each component in e.g. Tina, and other meticulous attention to detail, setting simulation temperature, etc. you might be able to get a quite accurate approximation to a real board - but that level of detail and knowledge is for the experts, not myself.