Subject: Re: removing non-monotonic points?!
From: Scott McMorrow (scott@vasthorizons.com)
Date: Thu Feb 07 2002 - 08:50:38 PST
Arpad,
We are all in agreement that the ibis checker should be fixed to report
non-monotonicity in the summed curves, also.
It is our experience that when these points come well below or above the
voltage rails, say 800 mV above or below the rail, they just do not
matter. For an output to achieve an extreme voltage this far below the
rail while operating would be quite extraordinary. We have yet to see a
real device in a real system where the output is pulled this far below
ground. For normal device operation, this non-monotonic behavior (which
also often does show up in the summation) will never be seen. It is
perfectly acceptable to smooth out the curves as we do.
In our spice to ibis process, we perform this data smoothing in the
non-monotonic output curves below and above the rails, and then perform
a correlation to the original spice models in a real circuit with
reflected waves, to be sure that under normal operating conditions the
model is still faithful and accurate.
Theoretically, it does matter what the curves do 700 mV below the ground
rail and above the power (or clamp) rail. But as a practical matter, it
does not alter real simulations one iota. For my money, anything can be
done to a simulation model, as long as the approximation results in
accuracy that is accurate enough for the work we are doing.
The easiest way to check this is to compare simulations with the
original spice, or to compare the results to measurements.
best regards,
scott
-- Scott McMorrow Principal Engineer SiQual, Signal Quality Engineering 18735 SW Boones Ferry Road Tualatin, OR 97062-3090 (503) 885-1231 http://www.siqual.comMuranyi, Arpad wrote:
>Prabhu, > >You are correct, the recommendation in that presentation will result >in double counting the clamps. What the author of that paper really >should have recommended is to check the monotonocity after summing >the pullup with the power clamp and the pulldown with the GND clamp >IV curves. Unfortunately the IBIS checker doesn't do this, and in >my opinion it should be corrected. It has mislead way too many >people already. > >When I brought this up in the meeting Scott's response was that your >signals will never go into that range at the driver anyway, so it >really doesn't matter. Even though this maybe true in most cases, >the correct way for monotonocity check is after the summation. > >Arpad Muranyi >Intel Corporation >==================================================================== > > >-----Original Message----- >From: Mohan, Prabhu [mailto:Prabhu.Mohan@actel.com] >Sent: Wednesday, February 06, 2002 6:59 PM >To: ibis-users@eda.org >Subject: removing non-monotonic points?! > > >Hi all, > >I have a few questions regarding a paper that was presented at the last IBIS >Summit, that suggested various methods to remove non-monotonic points from >the IBIS model file. > >http://www.eda.org/pub/ibis/summits/jan02/bell2.pdf > >For example, on page11, the pulldown curve is just extrapolated in the -ve >region to remove non-monotonic behavior. > >Question#1: Won't we double count the clamp current by doing this (since >simulation tools add the Pulldown/up curves with the clamp curves)? > >Question#2: What do we do if the silicon really does behave like this? (true >in my case. i do get a similar -ve bump when i subtract the 'tristate' curve >from the 'ouput low' curve) > >Question#3: Is it advisable to remove the non-monotonic points completely, >since they almost always seem to occur outside the normal operating range? >ie. flatten the -ve/+ve bumps? > >There might not be a simple solution. But I'd really like to know what the >possibilities are. > >Thanks in advance, >Prabhu/~ > >
This archive was generated by hypermail 2b28 : Thu Feb 07 2002 - 09:08:34 PST