Re: Scaling IBIS models

Paul Gregory wrote:
>
> I have a v2.1 model from a supplier. The models is of
> a CMOS 24ma driver. After simulations, I find that the
> driver is too strong, that is, there is ringing on the
> transitions, and the transition is faster than is
> necessary. So I request the supplier to reduce the drive
> strength by 10%, rebuild the IBIS models and send me the
> new stuff.

Quick sanity check: the clamp curves should be pretty much
unchanged. This is because the total amount of output
transistor is determined mostly by ESD requirements, so
when a driver is sized the drain area is left unchanged
but several gates are tied off.

> I get the new models. I find that the V/I tables show
> about a 10% reduction. That is what I expected. But,
> the transition times and rising and falling waveforms
> are not slower, but faster than the original (by about
> 10%). I think about this. Perhaps its because the
> capacitance of the the gate is reduced. No, the C_comp
> value is not changed.

They didn't change the predriver. Output transition times
are dictated by the rise and fall times of the final drive
gate voltages, which in turn are a function of the predriver
output currents and the gate capacitance of the final drive
stage. When they tied off 10% of the final stage gates, the
ones that were left switched faster, giving a higher di/dt
at the output.

> So the questions: Why does a model with reduced drive
> capacity have faster edge rates? Is this a model error
> or should it really be so?

It's real.

> Second, if you need to do some "what-if'ing", and you decide
> to edit the ibis model to scale it, what things should be
> changed and in what direction (that is increased or decreased)?
> Should everything scale by the same factor?

IMNSHO (This is my job, here) each predriver should be
individually sized to give adequate results with the intended
final stage. Failing that, you can *almost* expect a constant
value for di/dt irrespective of final drive size as long as
the predriver doesn't change. The clamp curves shouldn't
change at all, and the final drive saturation (short-circuit)
current should scale.

-- 
D. C. Sessions
dc.sessions@vlsi.com