Re: [SI-LIST] : skin effect and transmission line simulators

Ray Anderson ([email protected])
Thu, 4 Sep 1997 10:41:02 -0700 (PDT)

Just a couple comments on the questions posed by Chris Simon
regarding time domain modeling of frequency dependent lossy
transmission lines.

From the sounds of the requirements stated, I'd guess that he is
trying to do eye pattern analysis or something similar on various
transmission lines at gigabit speeds.

The newer versions of Hspice (97.1 and 97.2) have the new W
element which supports frequency dependent losses. However
I don't don't believe that frequency dependent inductance
is accounted for. Perhaps someone from Avant! can comment on
this aspect of the W element. The W element allows you to
specify the RLGC parameters of a line. (with R and G in a
normalized form) and then it extrapolates the lossy element
values with respect to frequency. At present you need to
provide the RLGC matrix values as inputs. (you need to use
a field solver or other means to get the values). Purportedly
an integrated field solver will be provided within Hspice
in some later release. The W element is extremely fast as
compared to some of the other HSpice transmission line
models as well as being accurate. I've heard some users
have had good correlation with lab measurements up through
about 10 GHz.

HP MDS Release 7.0 with the Picosecond Modeling Option supports
IBIS models of receivers and drivers. The frequency dependent
transmission line models provided in the Multilayer Interconnect
Model Library provide for frequency dependent analysis taking
into account the dispersion caused by skin effect and dielectric
losses ( I believe most of the MDS t line models support frequency
dependent behavior). I'm not sure any of the MDS models account for
frequency dependent inductive variations. MDS uses a convolutional
approach to their freq. dependent tline modeling. It may be possible
to define a frequency dependent tline utilizing SDD's or FDD's from
measured data, however attempts to make that scheme work so far
have failed for MDS users here at Sun.

The option of using a "black box" representation of a lossy
line in spice (for example creating a Pade' approximation
in the form of the ratio of two polynomial defined voltage
controlled voltage sources ala ContecSpar) from measured
s-parameters doesn't work as the black-box model doesn't
account for delays and is only useful for electrically
short devices.

Ray Anderson
VLSI Engineering Signal Integrity Group
Sun Microsystems Inc.

> X-Sender: [email protected]
> Mime-Version: 1.0
> Date: Thu, 4 Sep 1997 11:36:06 -0600
> To: [email protected]
> From: [email protected] (Chris Simon)
> Subject: [SI-LIST] : skin effect and transmission line simulators
> I am looking for a time domain lossy transmission line simulator for
> digital signals with bit rates of 1Gbit/sec and higher. I would like to
> get information on simulators that you have experince with.
> The simulator I'm looking for should include all of the following:
> 1. The results can be viewed in the time domain.
> 2. Transmission line models that include series resistance and
> inductance which are functions of frequency to ACCURATELY model skin
> effect. (i.e., a frequency dependent line model that is used to get time
> domain simulation results)
> 3. SPICE CMOS models and/or IBIS type non-linear models can be used
> as drivers and receivers
> 4. Arbitrary driver data patterns can be used (i.e., not limited to
> repetative signals)
> There is one other criteria, but I'm not certain that this one is required
> for signals with 150 to 200 pSec rise and fall times. Any opinions?
> 5. Includes the effect of dielectric losses.
> I am aware of several products that claim to meet these critria. Upon
> close inspection several of the lossy line models didn't really meet number
> 2. This seems to be the tough one.
> I am going to take a detailed look at Microwave Design System (MDS) from
> Hewlett-Packard, although someone told me that the current version doesn't
> meet number 3 above.
> I would be interested in hearing which of the above criteria are met by
> other tools, or whether a future revision will incorporate these features.
> If you know how the following tools stack up, especially with regard to
> number 2, I would appreciate hearing from you.
> Quad Design, Quantic, Interconnectix/Mentor Graphics, Pacific Numerix,
> Cadence SigNoise, ApSim/Contec, EMC-workbench, others?
> Chris Simon