Jerry's questions re: [Rising Waveform], [Falling Waveform]

Hi Jerry,

>Hello, I'm seeking some advice about the subject keywords. I have included
>them in a model I've developed for a customer (since the drivers I'm modeling
>do have slew rate control), but I'm not entirely sure how simulators use
>the V/T tables versus using the [Ramp] data. In particular, I have the
>following questions:
>
>1) What xxx_fixture values should I specify for my SPICE simulations (and
>hence place into the IBIS model)?
I would suggest a value that reflects typical usage of your device.

>2) Should the xxx_fixture values vary for each customer based on their
>application?
For simplicity I would recommend creating a single V/T for use by
all customers unless your outputs are relatively high impedance. Say > 15
ohms.
Most CMOS outputs today are less than 10 ohms, many are less than 3 ohms.
(This is a rough guess so don't shoot me if I am off a few ohms)

>3) What would happen (in other words, what would a simulator do?) if the
>V/T tables use a different loading assumption than the assumptions used
>to produce the dV/dt_r and dV/dt_f values? (a response that this is not
>appropriate for an IBIS model would be acceptable to me).
It depends on the simulator. V/T curves are provided to augment the data
available in the dV/dt_r and dV/dt_f fields. Some simulators ignore V/T
tables
since dV/dt_r and dV/dt_f are required.
Some use only the first table found
Some will attempt to regressively construct a driver model from all the tables
provided under all the load conditions provided.
I don't know of any simulator that does a perfect job of the later. The
biggest
problem is that getting all the rise/fall data and the tables to match is
very complex
and highly prone to error. Many routines will fall back to the simplest
methods if
they detect discrepancies with the complex methods.

>4) Would the simulations be any less accurate if I simply didn't include
>the V/T tables?
Probably but it may not be a large difference, it is also possible to have
more
accurate simulations without V/T tables. There are conditions that could
cause the
smart algorithms that interpret the V/T data to create slew curves that are
worse than
the result provided with a simple dV/dt_r spec.

The goal of V/T tables is two fold:
 1) Provide a "specification waveform" for the output into a load.
 2) Provide data to "improve" the output characteristics.
Item 1 could be used by a model developer as a manual verification when
testing models
to insure that the dV/dt_r and dt_f data is working properly.
Item 2 could be used to "tune" the simulation to achieve better matching
for the specified load.

-------------------------------------------------------------------------
Have a great day...
Kellee Crisafulli at HyperLynx Inc.
kellee@hyperlynx.com http://www.hyperlynx.com
-------------------------------------------------------------------------
 
Received on Sun May 4 15:03:54 1997