Re: [SI-LIST] : RE: [SI-LIST]: Long bus or star?

D. C. Sessions (dc.sessions@vlsi.com)
Mon, 29 Mar 1999 16:12:08 -0700

Shayle Hirschman wrote:
> D.C. Sessions wrote:
> >
> >How do you propose to keep several legs ALL less than 12mm long?
> >(at least 60ps/cm, risetime of less than 300ps, line length
> >Tr/4 => 1.25cm) Note that this is a foolishly optimistic
> >calculation.

> D.C.,
>
> Right, of course, with psec rise times. I'm more used to a couple of ns of
> rise time, such as which Chris was asking about. In that case, 4 to 6 inch
> legs of a star may be possible, but only worth the effort if the equivalent
> length sum of a daisy-chain adds undesirable delay or skew.

As I (perhaps too frequently) note, you have to take extraordinary steps
to get anything slower than 500ps or so any more. Time to recalibrate
your thinking.

As a fer-instance, have a look at the V/T tables in any reasonable
IBIS model. (Make sure that the loading conditions are somewhere
near reality, though: 50 ohms R_fixture, 4-5 pf C_fixture, etc.
There are some really useless IBIS models out there.)

> I guess the question I'm posing is: Would, in that case, reflections then
> add, acceptably, while the signal is still
> > transitioning? Or would this lead to too large an amplitude due to multiple
> > simultaneous reflections from each leg?
>
> Shayle
>
> >> At 10:58 AM 3/29/99 -0800, you wrote:
> >> >>Chris Bobek wrote:
> >> >
> >> >
> >> >>I have an address bus that is 75ohm, and 19" long. There are 8 loads
> >that
> >> >>are daisy-chained along the trace. The edge from the driver (which
> >> >>is at one end) is 3ns.
> >> >
> >> >>I was wondering if I would be better off with a star pattern, where >each
> >> leg is short (less than 4"). Would this work and not require
> >> >>any termination?
> >> >
> >> >
> >> >>Thank you very much,
> >> >
> >> >
> >> >Star patterns are a nightmare. There are so many reflections in
> >> >the system that it easily becomes unstable. If the star is not
> >> >perfectly balanced your signal integrity will degrade rapidly.
> >> >Just the process variations for each agent can be enough to degrade
> >> >your signal integrity to an unacceptable level. Furthermore, if
> >> >you have a GTL-like bus a star wont work because it would require
> >> >too many pull-up resistors. Star topologies can be made to work,
> >> >but not without a lot of pain. They should be avoided for a high
> >> >speed bus unless all other options have failed.

-- 
D. C. Sessions
dc.sessions@vlsi.com

**** To unsubscribe from si-list: send e-mail to majordomo@silab.eng.sun.com. 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 ****