Re: [SI-LIST] : How to identify SSO

D.C. Sessions (dc.sessions@tempe.vlsi.com)
Mon, 12 Jan 1998 09:00:53 -0700

Lfresearch wrote:
>
> Chaps,
> I've been following this ( see bottom of message ) discussion with a lot of
> interest. I have faced an ASIC problem recently ( hope to publish what I've
> found very soon ) which is directly related to the issue of noise on NON
> switching signals.
>
> The subject is low cost controllers, an ASIC was a great way to combine
> discrete devices to recoup board space etc etc from earlier versions. However,
> the ASIC is used not only to control many lines of I/O leaving the controller,
> but signals between the uP, its memory, and an onboard A to D convertor.
> Because all signals use the ASIC, the I/O lines have remnents of the really
> fast switching on them.
>
> I can't afford to filter each line, the whole product DMC is less than $30! I
> have looked at several vendors parts and discovered there is a big difference
> between them from this aspect. Being an EMC guy rather than a Silicon guy I
> don't fully appreciate why. I'm nervous though that if my current good
> supplier goes, my whole controller range will fail EMC tests.
>
> I guess this posting was to make sure that folks designing chips know that the
> EMC chaps are very interested in things like "crosstalk" between lines being
> switched. Even if it's not considdered of primary importance to the SI
> engineer.

Gate arrays don't give you much flexibility in this regard. For our
standard-cell designs we isolate the supply connections of sections
that need to be quiet, such as printer and keyboard ports. The
main constraint here is that the isolated sections have to be large
enough to allow sharing of ESD transients; fewer than ten I/Os or
so tend to have increased vulnerability.

As far as isolating high-speed sections which never "leave the box"
from each other, it's a waste of effort. Physical separation leads
to onchip resistance and inductance, while all of the victim lines
share the induced noise. The net effect is pretty small compared
to the self-generated noise on (for instance) a PCI bus.

-- 
D. C. Sessions, telecommuting :-)
dc.sessions@tempe.vlsi.com