X3T9.2/91-31

 MEMORANDUM    --    12 Feb 1991

 TO:	       John Lohmeyer, Chairman, X3T9.2

 FROM:	       Bill Spence, TI

 SUBJECT:      Cable Issues; Fast SCSI Ops

In working with other members of the Task Group, I found continuing  concern
with two issues:
    1.	What is the ideal impedance for a SCSI cable working with 110-ohm
	terminators?
    2.	What cable characteristics should be specified to implement the ideal?

Both Kurt's and my results suggest that in the most demanding case--longest bus,
most loading--80 ohms is close to the ideal using present chips, i.e., release
on negation drivers.  The basis developed for the tradeoff is mainly this:
better release levels (negation levels) are secured with higher impedance
cables, better transmitted low levels (assertion levels) are secured with lower
impedance cables.  As with most all tradeoff situations, the optimum is broad.
It is not easy to quantify any great advantage for 70, 80, or 90 ohms as opposed
to the other two.  I offer the following comments, based on my observations.
     1.  The more devices on the bus, the better the release levels and the
	 worse the assertion levels.  Thus, the more critical condition to
	 optimize to is the fully-loaded bus, and observed waveforms in this
	 case favor 80 ohms and below.	NOTE: this effect cannot be fully
	 simulated in any way I know of, since it is significantly contributed
	 to by silicon effects that may not even be documented yet.
     2.  The emergence of positive negation drivers in future single-ended
	 protocol chips will give a big boost to negation levels, so that cable
	 design can be tilted further toward favoring improved assertion levels.
	 This will even more strongly point us toward staying below 80 ohms.

What complicates the implementation of this decision--what is behind the second
issue presented in the first paragraph above--is the loss of impedance and
increase of signal propagation delay in the outer layer pairs of our round
shielded twisted-pair cables.  If one desires to keep the minimum cable
impedance above the high 70's, say, one may have to specify a foamed dielectric
with a nominal high 80's impedance, to keep the lowest impedance of any outer-
layer pair from dropping too low.  Similarly, if one wishes to limit the delay
skew in nanoseconds between the core and the outer layers, one may have to
specify foamed dielectric so that all propagation delays are reduced, thus
reducing the nanosecond skew also.




To make the outer pairs more similar in performance to the core pairs, it would
help to have a thicker buffer layer between the outer pairs and the shield.
But this may conflict with the OD requirements of the high-density connectors.
We may individually or collectively want to ask our cable suppliers to revisit
this matter and advise us of any unexplored options.

FAST SCSI P.S.:  In my plant we have just had our first opportunity to operate
at 10 MB/s, using a WD33C93B in our Host Adapter with a new disk at the other
end of the bus.  There has not been time to mount a full suite of tests.  In our
first results, acquired just in time to make the X3T9.2 mailing, using 110-ohm
terminators and a single 68-foot "80-ohm-SCSI-Standard" cable (Madison 4179), we
have achieved waveforms as shown in the attached pictures.  It will be seen that
the high quality -ACK waveform was carried to the disk is good shape, whereas
the low quality -REQ waveform was received at the host adapter in a form very
vulnerable to double clocking*.  The system ran poorly, but it appears this was
because of the poor -REQ signal, not the cabling.  It also appears that an even
lower ratio of cable to terminator impedance would might have been beneficial.
			     *I am told that the 33C93B sample being tested did
			      not have its hysteresis fully sorted out yet.