setup, hold, and signal timing skew inconsistencies

Richard Moore r_moore at qlc.com
Thu May 6 10:58:42 PDT 1999


* From the T10 Reflector (t10 at symbios.com), posted by:
* Richard Moore <r_moore at qlc.com>
*
Recently I've noticed that there are still what I consider to be
inconsistencies in the SPI-3 timing budgets
(Figures 46 and 47, and Tables 31 and 32 in rev 5).
 
First, I think that the huge signal timing skew allocation for slow DT
transfers is not only unnecessary,
but inconsistent with the ST signal timing skew. For example, the latest
proposal has Fast-20 DT signal
timing skew at 13.4 ns, and Fast-5 or Fast-10 ST signal timing skew at 8 ns.
Are we saying that a cable
plant that works fine at Fast-20 DT may not work at Fast-5 or Fast-10 ST?
 
It was said at the working group meeting this week that this is because DT
is supposed to give relief
to the cable. But DT gives relief to the cable by driving it with REQ and
ACK signals that are less subject
to distortion, because they are half the frequency of the corresponding REQ
and ACK signals for the
same transfer rate in ST. By relaxing the skew we are in effect providing
redundant relief. In the end I don't
think it matters to those who are designing silicon with Fast-80 DT in mind
(except that the change may
force them to repeat all their slow DT simulations that passed with the old
timings), but it does send a
peculiar message.
 
Another inconsistency is the difference in setup and hold budgets for Fast-5
(Figure 46). Fast-5 setup
and hold times are 23 ns and 53 ns at the transmitter, and 15 ns and 25 ns
at the receiver. To account
for the differences, signal timing skew was set to 8 ns for setup and 28 ns
for hold. At the working group
meeting it was stated that this was due to active negation; this was then
restated to say that it was due
to reflected-wave signaling. This explanation does not make sense; if the
data lines used reflected-wave
and the REQ/ACK lines used incident-wave, then it is the setup, not hold,
that would be degraded.
 
Fast-5 setup and hold times changed between SCSI-2 and the first generation
of SPI. SCSI-2 called
for the sender to provide "one deskew delay plus one cable skew delay" of
setup time, and "one deskew
delay plus one cable skew delay plus one hold time" of hold time. These
phrases equate to 55 ns setup
and 100 ns hold time. A careful reading of SCSI-2 shows that this extra hold
time was budgeted to the
receiving device, not to the cable: "The initiator [target] shall read the
value on the DB(7-0,P) signals within
one hold time of the transition of the REQ [ACK] signal to true" (SCSI-2,
section 6.1.5.2). For consistency,
Figure 46 should be altered so that the signal timing skew is 8 ns for both
setup and hold, and the 20 ns
difference should be allocated to the receiver hold time.
 
I don't know why the Fast-5 numbers were so dramatically reduced going from
SCSI-2 to SPI.
 
As a footnote, I would like to repeat that SCSI does not work with
reflected-wave signaling on the data lines:
With 6-meter cables you need 60 ns for a cable round trip, and this far
exceeds the 8 ns (10 ns in SCSI-2)
allocated for Fast-5. Although electrically you may expect to see
reflected-wave signaling with 48 mA drivers
on a low impedance cable, in practice single-ended usually works anyway
because most devices drive
somewhat higher than 48 mA, and hence really do use incident-wave signaling.
Note that I said, "usually";
anyone who's been in the initiator business for 10 years has at some point
configured a system that doesn't
work, and discovered that it can be made to work by moving the offending
device from the middle of the cable
to the end of the cable.
 
 -- Richard Moore
    QLogic Corp.
 
*
* For T10 Reflector information, send a message with
* 'info t10' (no quotes) in the message body to majordomo at symbios.com





More information about the T10 mailing list