Thank you (and everyone else) for taking the time to answer my questions,
I owe ya....
I wasn't thinking that by having the synchronous bus all switching at the same time (or at generally the same location along the line) that the other 6 inches or so of trace (until the next "clock") would be roughly at DC levels, which don't contribute to crosstalk. Beyond that, the "DC" effect could be trashed by reflections, but proper terminations and no stubs can help with that. (I'm just repeating what everyone said)
So we are running the busses together and increasing the clearances to control lines and other noisy stuff, but I just wanting to thank you guys for the GREAT answers!
Jack
-----Original Message-----
From: David Price [mailto:[log in to unmask]]
If a bus is synchronously timed, and matched in routing path and length
well enough, and if the timing is designed right, then within the bus (one
net coupling to another), they will all get 'quiet' within enough settling
time that when the inputs on the bus look at the signal level, everybody in
the bus will be stable. In other words, if one net in the bus couples to
another net and produces crosstalk, the crosstalk will have enough time to
dissipate so that when the bus signal matters, crosstalk within the bus
will have gone away.
This is one underlying justification for 'bus routing'.
If the bus flows in the same fashion, stubs alike, within similar lengths
and impedances, etc. - well, if it looks like a duck and walks like a duck
- all the nets perform alike. So treating the nets in a bus similarly is
more important that routing the bus as a bus, at least to us.
As a practical matter in SPECCTRA, for synchronous busses, we accumulate
crosstalk from other nets but turn off crosstalk within the bus, allowing
the router the opportunity to nest the nets closely within the synchronous
bus if it's advantageous to routing but not forcing it.
SPECCTRA has new bus 'bundling' feature, but in all instances and
strategies the bus can be managed for crosstalk.
We do a lot of this and have 'modelless' tools for managing real crosstalk
in SPECCTRA, so please contact us offline if you're interested.
Regards,
David Price
|