Re: FW: FW: Spice Model of IBIS

Jon Powell wrote:
>
> Well, I just couldn't resist.
>
> I certainly wouldn't argue with anyone that SPICE (actually some SPICE
> commercial derivative) is the premier circuit simulation technology that
> is currently publically available. I also wouldn't argue with anyone who
> said that IBIS has some problems that need to be addressed.
>
> However, the market has shown that there is a need for a PCB SI
> simulation technology that offers features that transcend the current
> SPICE node based syntax. Think of the bigger problem. You need to
> simulate the SI and Crosstalk effects of every net on a PCB. You need
> MIN MAX timing from every driver to every receiver. You need SI and
> Noise margin numbers for every net. And you need them quickly enough so
> that you can use the data to turn the design a couple of times. I
> believe these issues are difficult to solve using SPICE as it (they)
> exists on the market today.
>
> Of course, I use SPICE every day. I think it is great. I just believe in
> using the right tool for the right job and I have more than one job to
> do.
>
> Here is a conceptual example:
> you have a 32 wire bus. Each wire couples to it's neighbors. You need to
> simulate every wire and the effects on that wire by it's immediate
> neighbors. You do not want to waste simulation time by simulating
> effects of the wires that are more than a couple of conductors away.
>
> How would you solve this problem using SPICE?
> Can it be solved automatically?
> How would spice represent the names of the different nets? (ie. what is
> the crosstalk on D13?).
>
> As for accuracy:
> Leave the argument of which approach is more accurate for the moment.
> what is the accuracy of your inputs? (PCB stack-up, SPICE model data,
> Dielectric constant, trace width).
> If you are simulating to a higher accuracy than your data inputs then
> you are wasting time and deceiving yourself. I think that a teacher told
> me this in 5th grade but I never really believed it until I started
> doing SI simulation.
Surprise! Every item you mention above has been sucessfully acomplished
by many of our customers and other Spice vendors as well. Using IBIS
or behavioral models a whole board can be done in reasonable times using
Spice. The problem of simulating wires by distance or number of
conductors
is related to the user interface control and the field solver not a
simulator issue at all. This has been done by the way in our system
since
1992 and some other SPICE vendors a bit later. Node names are not an
issue
in modern Spice programs. Your dealing with 2.6 from Berekley circa
1968.
PCB stackups have nothing to do with Spice. It's an issue on integrating
a
good field solver with your simulation engine. Again its been done by
several Spice vendors. You are either trying to confuse customers by
these
issues you raise or you truly don't understand modern day Spice tools.
Of course Spice is not perfect and there are plenty of problems. But
neither
is IBIS. It has its place but at the moment the capabilites are a bit
on the lite side...and there are plenty of problems as you yourself have
raised. We'll leave accuracy aside for the moment as you mentioned. How
about flexibility, capability? Can IBIS handle Analog? How about more
sophisticated output stages that dynamically change current
sourcing/sinking
capability? Where's the information about the power supply currents?
Heck
some IBIS data doesn't even contain switching information. How about
ground
bounce? Since there is no mention of the gate or base of the
transistors,
once noise is introduced via the power/gnd what's the voltage on the
base/
or gate of the I/O output transistors? Without that you don't even know
which transistor is on or off. I'm for bringing out these issues and
solving
them using IBIS not covering them up or changing the argument. Spice is
just a simulator. It will survive or die on its own merits. How about
improving IBIS?
 

Best Regards,

-- 
Fred Balistreri
fred@apsimtech.com
http://www.apsimtech.com