Re: [SI-LIST] : Re: (Differential SCSI)

D. C. Sessions (dc.sessions@tempe.vlsi.com)
Thu, 05 Feb 1998 17:37:48 -0700

Fred Balistreri wrote:
> D. C. Sessions wrote:
> > Fred Balistreri wrote:
> > > D. C. Sessions wrote:
> > > > Mark Nass wrote:

> > > > > We are laying out a board that has differential a SCSI bus on it.
> > > > > The specs say the that the differential impedance should be 110-135 ohms.
> > > > > What is differential impedance? Is this the impedance between
> > > > > the 2 traces, or the impedance with respect to the plane?
> > > >
> > > > The two tracks form a delta network. The effective differential
> > > > impedance is the parallel of (a) the trace-to-trace impedance
> > > > and (b) the series of the two traces' impedances to ground.
> > > > If the traces are equal and well-separated this is just twice
> > > > the individual line impedance, which would mean that your PWB
> > > > traces should be in the 55-67 ohm range (call it 61 ohms plus
> > > > or minus 10%). If your native impedances are higher, you can
> > > > get into the right neighborhood by running the pairs closer
> > > > together so that the coupling capacitance lowers the total.
> >
> > > Hello DC, your statement about 2X the individual impedance for diff
> > > impedance is not correct. If you like I can email you the formula
> > > in terms of c,l and the mutuals. In order for your statement to be
> > > true the two traces would need to be seperated from each other such
> > > the the mutuals would be negligible. If that were the case you would
> > > be throwing away the benefits of the differential system and you would
> > > have a hard time achieving the desired impedance I venture to guess.
> >
> > I think I said that: "If the traces are equal and well-separated this is
> > just twice the individual line impedance."
> >
> > > The two traces must be
> > > "electrical equal" to each other in order to have a true differential.
> > > Usually the best and easiest way to insure that is to make the two
> > > the same physically with the proximity to any ground or power plane
> > > the same.
> >
> > All agreed -- I thought that this was implicit in my statement
> > about the port impedances of a delta network.

> You missed the point about what differential is all about. If you
> seperated the traces such that the mutuals are negligible you also
> throw away the benefits of running differential traces in the first
> place. Your premise then holds true but you have a bad differential
> pair. A good differential pair should be tightly coupled. The idea
> is to cancel the fields by sending a complimentary signal down the
> traces. When seperated that doesn't happen as easily. So you have
> indeed met your impedance criteria but defeated other important
> aspects that impact SI. Therefore this kind of system is really
> matched single ended not differential.

Actually the differential impedance match would be OK, but any
common-mode current would run into a mismatch. So long as the
PWB traces between the connector and the IC are short, this may
be acceptable. Unfortunately, it's VERY difficult to get a full
delta match betwen a PWB and a cable. Typically a cable has
much higher capacitance between the signal conductors than
between the signal conductors and ground; a PWB is the
opposite. The closest approach to the cable characteristics is
with two conductors in the inner pair of layers on a 6-layer
board, with the two conductors overlaying each other. Even
then it's not a very good approximation to the cable.

Since most systems don't even try to match the TERMINATORS to
the full delta net, it's silly to drive the entire PWB stack
based on the unbalanced impedance of the cable, and especially
when few signalling standards (notably SCSI) even constrain it.

Thus my point: do the SI for the REST of the board, get your
stack right, and if the maximum trace impedance is higher than
nominal for the cable then by all means lower it by coupling
signal pairs. Trace separation is much more readily controlled
than height, so fine tuning is easy. On the wild chance that
you can put bounds on the cable's unbalanced impedance then
by all means tune the terminator to match the nominal value;
the small unbalanced mismatch of the PWB traces should be
well within the noise budget.

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