Re: IBIS Connector group status concerning ac losses

Kellee,

It is in the area of extending the model for use with packages (which can be
quite lossy), cables (which do have significant losses), and pcboard traces
(as an extension to the current ebd non-lossless model) that some
sort of frequency dependent loss model is needed. The Hspice W-element
model with a frequency dependent dielectric and skin effect loss is generally
good to the 1 to 2 GHz range, within a small margin of error. Yes, there are
frequency dependent changes in L and C, however, they are quite small
in the frequency ranges where IBIS simulations are generally run.

AC losses are a significant factor in real systems with sub-400ps edge rates,
and high frequencies. I see these effects in source synchronous bus simulations
in the 200 to 400MHz range, where the edge rate harmonics approach 1.5 to
2 GHz. They significantly effect noise margins and timing of high speed signals,
and
definitely effect the eye patterns of fiber channel and gigabit ethernet signals.

I would suggest that the committee take the lead and define an extensible method
of describing ac losses. One that could be enhanced by different methods of
modeling
in the future. For example, using the current matrix models that you have for
lossless transmission line sections, you could easily extend this to an RLGC
matrix
with an Rs, which is the frequency dependent skin effect resistance, and Gd, which

is the frequency dependent dielectric loss. Other, possibly more advanced,
modeling
types could be included in the standard in the future.

I agree that, in general, ac losses are insignificant for most connectors.
However, if
we are creating a coupled matrix template to be used in other areas, as I have
suggested above, then it makes sense to consider inclusion of frequency dependent
losses in this standard. Otherwise, it will be years before the chance will occur
again.

If we think of this as a way to enhance ebd and package modeling, then I believe
adding ac losses to the coupled matrix descriptions makes undeniable sense.

If you don't beleive, then try to correlate a 400MHz memory bus simulation to
measurements using an IBIS simulator with currentebd models of the RIMM modules.
It can't be done.

Regards,

Scott

> 2) There is no way to describe lossy R,L,C,G effects.
> The sub-group responded by saying we had considered it but felt
> is was too complex to include in the first version of the specification.
> We agreed to reopen this for discussion in the sub-group.
> See below for details.
> HyperLynx-Teradyne meeting summary:
> 1) How do we describe the 3 major types of resistive loss
> across frequency (R matrix). General methods in use today only focus
> on one or two and require geometric data as input.
> 2) How do we describe L,C frequency effects
> 3) How do we describe dielectric loss v.s. frequency (G matrix).
>
> We discussed several methods and agreed that a full loss model
> was very complex. We determined 3 possible courses of action:
> 1) Continue without loss in the first release (follow up later)
> 2) Create a very simplified loss method for the first release
> 3) Continue discussions until a full featured loss model is found.
>
> It was mentioned several times that because connectors are very short
> typically less than an inch the loss with todays uses is very small
> and we could probably get by without including loss until the next release.
> It was also mentioned that acquiring the data from real connector models
> to describe several aspects of loss is very difficult.
> We also discussed the possibility of using the IBIS connector model for
> doing cable models.
>
Received on Wed Feb 24 20:50:41 1999