I somewhat
disagree. Typically, you won’t
find a diff pair that suffers from the weave effect. The probability is
low. But
once in a while, “in a galaxy far far away…” Oh no no no J . Anyhow,
when you do,
the effect is right at the top of the Pareto list. It can be hundreds
of ps on a
long route. I have personally seen 50 ps on 8 inch routes on computer
-like boards.
For many years, I tossed this off as “anecdotal” but have
come to understand the cause was inhomogeneous dielectric effects. Not
withstanding the weave effect, I can compensate most conductor
geometries with “good”
design practice and yes, even twists, turns, and connectors. Most of
the time
these design practices serendipitously even be choke out the weave
effect.
The good
news is that back plane have
wider trace widths which are less sensitive. The line cards are a
different
story.
… Rich
All,
The differential pair skew per inch is from many factors, the least of
which is
glass construction. The greatest, outside of the connector, is trace
length mismatch from route direction changes (45s or rounds). The
second
greatest is from length matching where a toroidal effect is not
considered.
I would offer that there comes a time when we have to stop writing the
standard
for poor design practices. A designer should always be targeting less
then 5ps point to point. My feeling is that 20ps is fine and does not
violate any US patents in any system I have been involved with. I
wouldn't want to even try and debug a system with 50ps skew ... that
would be
horrid and unpredictable.
As I said ... sooner or later we have to consider reasonable
engineering as a
solution.
-joel
Scott McMorrow wrote:
I am not currently aware of
any systematic study of
differential pair skew on production backplanes, due to the problem of
accurate
instrumentation of hundreds of channels. However, there are occasional
anecdotal measurements that have been made by some connector vendors on
customer backplanes which have shown 40 to 50 ps skew from time to time
on high
BER channels. Teradyne and Teraspeed Consulting will begin conducting
a systematic
study of backplane length differential pair skew this summer, on a
variety of
standard backplane stackup and trace configurations. That data should
be
available by fall, and allow us to set some fundamental skew limits on
weave-parallel routed tracks and off-angle tracks. I believe that
Intel
Labs is also involved in a similar study. However, I believe thate
Rich is correct. A theoretical upper skew bound of 6 ps/in does
currently exist
for many stackup, material and trace geometry combinations. In
addition,
a lower skew asymptotic bound for optimal routing strategies w.r.t. the
weave
has been shown to exist, but has not been adequately quantified by
anyone that
I am aware of. However, this may eventually prove to be elusive due to
low level defects in the fiberglass structure and non-uniform epoxy Er.
It is my belief that when laminate weave skew is placed into the mix
and common
mode conversion is properly accounted for, non-NRZ encoding schemes may
have
enough UI headroom to ultimately outperform NRZ signaling, in the worst
cases. However, this is only a hunch and not yet substantiated.
However, I think it behooves the committee to consider laminate weave
skew in
the specification. It is my considered position that a total
differential
skew specification of 20 ps in an end-to-end system is not achievable
on
epoxy/glass composites without extra ordinary efforts and license of
(or
violation of) at least one major patent in this area.
There is yet another possibility that may hold promise. Electronic
skew
pre-compensation may very well allow channels to work in the presence
of
moderate amounts of differential skew.
Regards,
Scott
Scott McMorrow
Teraspeed Consulting Group LLC
121 North River Drive
Narragansett, RI 02882
(401) 284-1827 Business
(401) 284-1840 Fax
http://www.teraspeed.com
Teraspeed® is the registered service mark of
Teraspeed Consulting Group LLC
b_panos wrote:
Yea, S McMorrow et al.
The thing is, it really depends on things like material, bundle size,
glass
dia. to trace width and separation, as well as the orientation and the
length of trace. To be quite honest, if skew is bad enough, it wouldn't
meet
ISI levels and the eye wouldn't be recoverable. The RT of a 10G signal
is
going to be on the order of 50ps. so any skew approaching this level,
your
RL is going to look terrible, not to mention your S21 results. Just my
2cents
Regards
Bill
-----Original Message-----
From: owner-stds-802-3-blade@ieee.org
[mailto:owner-stds-802-3-blade@ieee.org]On
Behalf Of Mellitz, Richard
Sent: Friday, June 10, 2005 1:43 PM
To: STDS-802-3-BLADE@listserv.ieee.org
Subject: [BP] Dielectric Weave effects
Hi All,
I started reviewing the draft doc and came across a differential skew
spec of 20 ps. Has any one actually measured skew on the published
backplanes?
It's possible to get up 6ps/inch skew in a pair due to bundle weave. Do
we need to address this issue?
These effects have been published before.
Rich Mellitz
Intel Corporation