We can't ignore the effect of vias on a differential interface since
the reflections they cause are "seen" and not "rejected" by the
differential receiver. Vias aren't very long electrically, but if you
get enough of them in a trace, the effective impedance of the trace
will drop noticeably.
You really can't avoid vias in a standard multilayer PCB. You need two
per trace just to pin-escape a part. I suspect that you'll end up with
4 more when you hook up the terminations. Either hand-route or
severely constrain the autorouter to do what you would do if
Just a note, if the customer is trying to drive a cable with this
differential interface, worry about loss (turning the signal into heat
- more signal making it to the cable means longer cable). Depends on
trace length and application of course, but long 4mil traces on 1/2oz
copper can lose an appreciable amount of the signal. Trade-off plane
spacing and perhaps overall board thickness for wide (e.g 10mil)
traces at the correct impedance.
There is a paper "The effects of vias on PCB traces", by the folks at
UltraCad Design, Inc. that you might find interesting. Try
'http://www.eskimo.com/~ultra/tech.htm' and download the pdf.
Thank you |
Don Abernathey |
On Mar 31, 6:09pm, Jeff Seeger wrote:
> Subject: vias in ECL @ >1Ghz
> Dear SI-land,
> I'm pondering the wiring of some >1Ghz differential ECL, with
> wire lengths reaching perhaps 10 inches. It has become clear
> that this wiring could save alot of length (and other torture)
> if vias were used, at least one and possibly two (per side).
> These would be signal routing vias, not pin-escapes neatly
> located near a discontinuity to a package.
> Should I be concerned about the performance of these vias?
> Anecdotal history tells me to expect a loading of perhaps 1pF
> each, which I would not expect to be an issue. If each side
> of the pair passes through a via at approximately the same
> point in time, will the issue take care of itself?
> Alternatively, is there a simply way to manage the "hit" these
> vias will produce, perhaps by specific geometries or proximity
> to reference (gnd)? I can envision quite a 3-d solver excer-
> size attempting to use pad sizes versus planes versus clear-
> ances etc.
> Thanks in advance, from an electrically challenged layout
> Jeff Seeger Applied CAD Knowledge Inc
> Chief Technical Officer Tyngsboro, MA 01879
> firstname.lastname@example.org 508 649 9800
>-- End of excerpt from Jeff Seeger