RE: [SI-LIST] : Serpentine traces

Mellitz, Richard ([email protected])
Thu, 2 Sep 1999 14:41:28 -0700

Any words of wisdom about current crowding in the corners and radiation
losses at say at 10 to 100 ps edges.

... Richard Mellitz
Intel

-----Original Message-----
From: Howard Johnson [mailto:[email protected]]
Sent: Thursday, September 02, 1999 1:44 PM
To: Elya B. Joffe; [email protected]; Vinu
Arumugham; [email protected]; John Lin - TAO; Howard Johnson; Gert Winkler;
Ron Matthews; Roland F. Portman
Subject: Re: [SI-LIST] : Serpentine traces

I've scratched around the edges of this problem for several
years, and
not yet come up with a final answer, but here are a few
thoughts.

I. The effect of inter-section crosstalk...
(1) The coupling between serpentine sections is a strong
function of
trace separation.
(2) The coupling generated as the main signal passes through
section N
is generated mostly from sections N-1 and N+1.
(3) Because the direction of flow on adjacent sections is
always opposite,
the crosstalk from N to N-1, and from N to N+1, will be
near-end
crosstalk (NEXT). It will have a positive polarity. The
total crosstalk
generated as a step edge passes through section N will be a
short burst
of NEXT preceeding the main wave (from section N+1),
followed be a short
burst of NEXT following the main wave (from section N-1).
The impulse
response of the whole mess resulting from the main signal
traversing
section N will look like H = [A*exp(2Ts) + 1 +
(-A)*exp(2Ts)]
(4) Each section does basically the same thing, so you get a
total
system transfer function that looks like H raised to the
power of N,
where N is the number of trombone section (actually N-1 is
more accurate
because the sections at the end are missing some side
partners).
(5) Overall, what this does is to advance the appearance of
the
rising edge (reduce the total circuit delay).

II. The effect of right-angle bends...
(6) There is a tiny excess capacitance present at each
right-angle bend.
Since a serpentine has so many of these, it may be worth
taking them into
account.
(7) This effect will increase the total circuit delay, and
slightly reduce
the effective trace impedance.

III. The skin effect...
(8) The skin effect will disperse the rising edge at the
output of the
serpentine.
(9) The effect is to increase the total circuit delay.

Overall, my impression is that effect I is the largest, with
II and III
trailing not too far behind. If you want to predict
serpentine accuracy
to within 1% or better, you will need all three corrections
(plus
probably some more I don't know about yet). Alternately,
you could
do what many people do and just build one, and then scale to
fit.

A last note concerns the relation between the trombone
section length
and signal quality. As long as the delay of each individual
trombone
section is less than 1/10th the signal risetime, the
crosstalk is fairly
well-behaved and has only the effect of lowering the overall
delay.
As the delay of each individual trombone section approaches
1/2 of
a signal risetime, the frequency response of the crosstalk
effect
develops nasty resonances in the passband of your signal,
resonances
that will make hash out of the resultant waveform. My advice
is
to use more sections, with a shorter delay each, as needed
to conform
to the 1/10th risetime rule of thumb.

I would be very interested to see any experimental results
measuring
delay versus trace separation. It's clear to me that if the
lines
are separated too widely, we are just wasting space on the
board.
On the other hand, if the traces are too close, the
crosstalk
effect shrinks the effective trace delay. Somewhere in
between
there is a value that will optimize the actual amount of
delay per square inch of board area.

Best regards,
Dr. Howard Johnson

At 01:00 PM 2/11/98 -0500, you wrote:
>Has anyone ever tried to simulate the effect of serpentine
coupled lines in
>SPICE?
>The scenario is 6 mil lines running through many slots of
2mm shielded
>connectors. With two tracks per channel, these data traces
are serpentine,
>and couple closely every time they go around ground
contacts, with 6 mils
>spacing and 8 mils spacing to ground.
>I was thinking of creating 3 distinct LRC tline models and
cascading them,
>any thoughts on the accuracy of this?
>These data traces are routed to vias at most slots, so
there will be
>capacitive coupling from the vias also. Has anyone found
this situation to
>create significant crosstalk (>3%)?
>
>Thank you very much,
>
>Fabrizio Zanella
>
Sr.
>Design Engineer
>
>Hardware Engineering
> _/ _/
email:
>[email protected]
> _/_/_/_/ _/ _/ _/_/_/ _/ voice: (508)
435-1000 Ext.
>4645
> _/ _/_/ _/_/ _/ _/ Fax: (508)
435-8949
> _/_/_/_/ _/ _/ _/ _/ _/_/_/
> _/ _/ _/ _/ EMC
Corporation
> _/_/_/_/ _/ _/ _/_/_/ 171 South
Street
> Hopkinton,
MA 01748-9103
>
>
>
_________________________________________________
Dr. Howard Johnson, Signal Consulting, Inc.
tel 425.556.0800 // fax 425.881.6149 // email
[email protected]

http://WWW.sigcon.com -- High-Speed Digital Design books,
tools, and workshops

**** To unsubscribe from si-list: send e-mail to
[email protected]. In the BODY of message put: UNSUBSCRIBE
si-list, for more help, put HELP. si-list archives are accessible at
http://www.qsl.net/wb6tpu/si-list ****

**** To unsubscribe from si-list: send e-mail to [email protected]. In the BODY of message put: UNSUBSCRIBE si-list, for more help, put HELP. si-list archives are accessible at http://www.qsl.net/wb6tpu/si-list ****