From: Scott McMorrow ([email protected])
Date: Tue Nov 16 1999 - 18:22:18 PST
Fred,
I always enjoy your comments.
2 + 2 never makes 5. I have never advocated that.
Here are my claims for the record:
1) Measurement based models are the most accurate if
accurately measured. This includes measurement based
Spice and IBIS models.
2) Measurement based models based upon continuous
production measurement (i.e. device manufacturing test data)
are extremely accurate, as long as the IBIS model is updated
throughout the course of production. There are manufacturers
whom I have personally worked directly with who do use this
approach in generating IBIS models which are statistically accurate.
3) Measurement based models based upon worst case
process silicon are probably the most accurate of all. There
is at least one manufacturer whom I have personally worked
directly with who has used this approach in generating their final
release IBIS models.
4) What is shipped by a vendor is often rarely represented accurately
by the Spice output buffers models that are delivered, for a variety of
reasons. Some of these which I have personally encountered are:
a) They lie by removing circuits from the final model that
represent valuable IP.
b) They lie by removing clamp and ESD structures which
do not "simulate well." These also tend to not work well
in real systems.
c) The spice model was never updated after design finished
it, even though the process or design was adjusted.
d) The spice model reflects what was designed but does
not reflect the bin splits that occur during final manufacturing
tests.
e) The spice model is not an exact representation of the
actual circuit being used. It has been modified in some way.
f) The same claims as above can be made for any IBIS model
created from these "flawed" Spice decks.
5) Often some clueless wonder created the IBIS model because he was
lowest on the totem-pole and some free software was there which
helped him do it. I have had lots of personal experience with these.
6) Given two correctly created models, one in Spice, one in
IBIS, there are many cases where the results in simulation are
identical to any degree of accuracy which you might desire. It takes
some engineering experience and common sense to determine
if this is true for each particular case.
7) Given two correctly created models, one in Spice, one in
IBIS, there are many cases where the results in simulation are
significantly different. It takes some engineering experience and
common sense to determine if this is true for each particular case.
8) IBIS behavorial models always simulate in circuit significantly
faster than their Spice transistor level models. Often the difference
is 10-1000X. With this sort of performance advantage I can iterate
many more points in the system design space before having to
move to Spice for the final worst case corner simulations. IMIC models
may also have a performance advantage and better approximate
the actual circuit performance than IBIS does. Unfortunately, no
one is producing IMIC models.
9) IBIS modeling methodology does not currently do a good job
of modeling return path effects, package effects, and SSO/SSI
effects. For these types of detailed modeling Spice and/or
full wave simualtion is in order.
10) Most models whether Spice or IBIS are not even
close to correct. So to argue that either modeling and/or
simulation method is accurate is ludicrous. GIGO. It is silly to
believe that any model, modeling method, or simulation environment
is accurate until proven implicitly. To do otherwise is
truly far fetched.
Engineers are problem solvers and approximators. There are
many approximations, many tools and many tradeoffs. Personally
I like as many tools in my tool box as possible, as many choices to
tradeoff as possible and then approximate the hell
out of those things that I can.
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.com> > > Although there are IBIS models made from measurment, your claims seem > far fetched. If there are Silicon vendors out there making IBIS models > from measurement and making process corners from real silicon lets > here from them. Some of the best IBIS models I've encountered come from > Intel. Maybe they can comment. > > But most IBIS models are made from SPICE, so how do 2+2 make 5. > > Best Regards, > -- > Fred Balistreri > [email protected] > > http://www.apsimtech.com > > **** To unsubscribe from si-list: send e-mail to [email protected]. In the BODY of message put: UNSUBSCRIBE si-list, for more help, put HELP. si-list archives are accessible at http://www.qsl.net/wb6tpu/si-list ****
**** To unsubscribe from si-list: send e-mail to [email protected]. In the BODY of message put: UNSUBSCRIBE si-list, for more help, put HELP. si-list archives are accessible at http://www.qsl.net/wb6tpu/si-list ****
This archive was generated by hypermail 2b29 : Tue Feb 29 2000 - 11:38:57 PST