[SI-LIST] : Re: delay lines with PCB traces

Michael Chin (mchin@cisco.com)
Wed, 24 Sep 1997 17:49:21 -0700

Brett,

> From bgrossma@td2cad.intel.com Wed Sep 24 17:01:57 1997
>
> One suggestion in Howard Johnsons 'Black Magic Handbook': there is a
> suggestion to implement delay lines with a chamfered corner or a rounded
> corner. The argument is based on increased capacitance on a right-angle
> due to the change in effective trace width. I am sure the praises have
> been sung enough for this book, but I make sure all my designers get it.
> A very good working understanding without wallowing in the e-mag world.

Thanks for the reference to Howard's book. This was exactly
what I did. In fact, by going to a wider gap, the rounded radius
become more smoother. I followed this rule for all PCB trace corners,
although a lot of my colleagues did not think that these would make
a big difference.

> Some additional questions:
>
> 1. Does anyone on the LIST have PCB design rules for this geometry
> (serpentine) that they have validated on the bench? (e.g. not just a
> rule of thumb)

Well, I valididated on the bench the two delay line routings.
The first case was a 4 mil trace with 6 mil gap in the serpentine,
and about 10% decrease in the expected trace delay was measured.

When the gap was increased to 24 mil in the serpentine, the trace
delay was in line with our expectation.

We did not have time to develop a PCB design rules but a rule of
thumb I used is that as the trace gap is at least 3x the trace
width, the coupling between two adjacent tracks dimenished to
some insignificant level.

> 2. Or, have a decent model to comprehend the effects of this geometry
> (cross-coupling due to parallelism, any additional degradation of rising
> edge, etc?)

We once had suspected a coupling problem between two traces in
a similar situation. And, we started to work on a model to
simulation this behavior. But, after the trace gap was widen
to 3x the width and the design worked, our energy was diverted
to other areas. Thus, we never completed the modelling effort.

What we learned was that this kind of coupling does occur even on
slow signal (50Mhz range) if the edge rate was fast enough and
the trace separation was not at least 3x the width.

Michael Chin
Cisco Systems, Inc