[SI-LIST] Re: [SI-LIST]: Which tool is the best
- From: "Muranyi, Arpad" <arpad.muranyi@xxxxxxxxx>
- To: <si-list@xxxxxxxxxxxxx>
- Date: Wed, 5 May 2004 10:53:49 -0700
Chris,
To give you and the rest of the readers a little background and
history, I didn't "invent" IBIS so that I could go to exotic
places. I developed my first behavioral model using HSPICE's
controlled sources when my group was running the signal integrity
simulations for the very first (33 MHz) PCI bus. This was done
before anyone had a buffer design done, so the best we could do
was put together an IV curve and edge rate based behavioral model
to show that the reflected wave switching (and in those days we
also looked at the incident wave switching version) PCI bus will
work. And the reason this behavioral HSPICE model turned into the
IBIS specification was because this model was successful and there
was a great demand for it, even by our customers.
Regarding tweaking the design and the time it takes to convert to
IBIS, in my experience it was the other way around. We (the SI
team) would come up with the characteristics of the buffer using
behavioral models because it would be a lot easier to tweak the
IV curves, the edge rates, and the die capacitance values of the
behavioral model than keep tweaking the transistor parameters and
the circuit design, and we would then give the behavioral model to
the designers as a specification to be implemented in silicon.
Also, what I hear from various experienced engineers is that the
latest and greatest high speed buffers are becoming easier to
model (behaviorally) because they tend to be more linear than
ever before, and due to the high speeds they also have to be
kept simple.
By the way, there is a great deal of myth going around out there
about the accuracy of transistor level models. People tend to
think that it is the most accurate way of modeling circuits because
it has "all" the detail. The only thing they forget is that if they
don't have the correct process parameters and/or SPICE options,
they can get a lot of garbage. I have had this experience numerous
times. Even our design engineering team has been hurt by this=20
problem on more than just one occasion... So the great demand
you are referring to is not a proof for the goodness of SPICE
models. I am not saying this to put down SPICE, it has a lot
of merits, but people should not trust it as blindly as they do.
I have a question for you: Have you ever wondered whether we
should start using those models and simulators which are commonly
used by process design engineers? They do their work with even
more detail, involving quantum effects, crystal structures, doping,
electron mobility in the crystal, etc... If "more detail =3D higher
accuracy" is true, we should all start using these models for our
bleeding edge high speed buffers, instead of SPICE since SPICE is
just a behavioral abstraction of those..., correct? Hmmm... how
come we don't do it?
Arpad
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D
-----Original Message-----
From: Chris Cheng [mailto:Chris.Cheng@xxxxxxxxxxxx]=20
Sent: Tuesday, May 04, 2004 6:30 PM
To: Muranyi, Arpad; si-list@xxxxxxxxxxxxx
Subject: RE: [SI-LIST] Re: [SI-LIST]: Which tool is the best
Arpad,
At the end of the day, our differences lie in your believe that =
abstracting
an accurate I/O model is trivia and running the analysis dominates the
design process.=20
I believe the tasking of abstracting an accurate I/O takes longer time =
than
the actual analysis itself especially when the I/O is constantly tuned =
to
optimized for the best performance.
Whose right or wrong ? I don't know. All I know is when presented with =
the
same set of design accuracy criteria and given a choice of SPICE or =
other
tools, ALL my customers turned to SPICE instead of a behavioral model =
and
those who tried to abstract the I/O into behavioral models takes longer =
time
to generate the abstraction model than the actual topology simulation =
time.
And by the way, behavioral vs. SPICE models is even less than days in a =
week
and I can still go to church on Sunday, not the 300 times you claim.
Believe it or not, I am not as single minded as you think. I actually =
spent
the time to benchmark both cases because I want ALL my customers to be
successful. Had they wanted SPICE, I provided them, had they wanted =
IBIS, I
sent them IBIS too. That's why I have the comparison between them and =
have
the data to back up my claims.
Whether I need a hole punch or missile is a subjective discussion we =
won't
go in. But like I say above, when given a choice between SPICE or IBIS, =
ALL
my customers want SPICE and no one end up using IBIS model. That says
something about the nature of the project isn't it ?
I always suspect many SI engineers are closet SPICE fans but their =
vendors
just limited their choice to IBIS only models. It may work a few years =
ago,
but at the speed I am dealing with nowadays, many vendors realize the
importance and start handing out HSPICE encrypted models. That's a right
step in direction in my book. Give them a choice and let's see what the =
use.
I know, I know, NIH and you certain don't get to go to standards =
conferences
in exotic places.....
------------------------------------------------------------------
To unsubscribe from si-list:
si-list-request@xxxxxxxxxxxxx with 'unsubscribe' in the Subject field
or to administer your membership from a web page, go to:
//www.freelists.org/webpage/si-list
For help:
si-list-request@xxxxxxxxxxxxx with 'help' in the Subject field
List FAQ wiki page is located at:
http://si-list.org/wiki/wiki.pl?Si-List_FAQ
List technical documents are available at:
http://www.si-list.org
List archives are viewable at:
//www.freelists.org/archives/si-list
or at our remote archives:
http://groups.yahoo.com/group/si-list/messages
Old (prior to June 6, 2001) list archives are viewable at:
http://www.qsl.net/wb6tpu
Other related posts:
