Simple SCSI-1 signal trace length question, anybody know the answer?

Hi, does anybody know what the maximum allowable difference in length
between signal lines is on old, narrow, 8 bit, 50-pin SCSI-1 or fast
SCSI-2. (good old 50-pin ribbon cable style)

Please note that I'm *not* looking for the max allowable cable
length.

Rather I want to know what the largest acceptable difference between
the lengths of 2 conductors/traces in a cable/pcb is that will.

I need to know this because I'm having a printed circuit board routed
for use in an old legacy system that runs SCSI along with some other
signals on a proprietary connector, and I want to make sure I don't
violate the spec.

Somebody told me 1 inch was the limit, but I wanted to double check
before fabricating any boards.

Many thanks if anybody can help.

-Chris

expedienttel@aol.com wrote:
>
> Hi, does anybody know what the maximum allowable difference in length
> between signal lines is on old, narrow, 8 bit, 50-pin SCSI-1 or fast
> SCSI-2. (good old 50-pin ribbon cable style)
>
> Please note that I'm *not* looking for the max allowable cable
> length.
>
> Rather I want to know what the largest acceptable difference between
> the lengths of 2 conductors/traces in a cable/pcb is that will.

The connected devices don't care for length, they only "see" the
electrical propagation delay. So the answer depends on the signal speed
on the cable.

Physical:
SCSI2 specify for Fast10 transfer mode:
200ps/m maximum signal-to-signal propagation delay delta.

SPI specify in general:
5.4ns/m maximum propagation delay.
150ps/m maximum signal-to-signal propagation delay delta.


Protocol:
There is a "Cable skew" timing of 4ns for all transfer modes up to
Fast10. It is defined as:
| The maximum difference in propagation time allowed between any two
| SCSI bus signals measured between any two SCSI devices excluding any
| signal distortion skew delays.

For SCSI2 this seems to be inconsistent because a SCSI2 HVD-Bus is
allowed to be 25m long. But 25m * 200ps/m results in 5ns. IMHO this is
the reason why they have reduced it to 150ps/m in SCSI3/SPI, 25m *
150ps/m results in 3.75ns what is less then one "Cable skew" time.

> I need to know this because I'm having a printed circuit board routed
> for use in an old legacy system that runs SCSI along with some other
> signals on a proprietary connector, and I want to make sure I don't
> violate the spec.
>
> Somebody told me 1 inch was the limit, but I wanted to double check
> before fabricating any boards.

For the _worst_case_ calculation you can use the slowest allowed
propagation speed:
5.4ns/m (approx. 2/3 * speed of light)
This results in a maximum of 135ns runtime from Terminator to Terminator
(25m bus length). The cycle time for the Fast10 transfer mode is 100ns.

For every meter of cable or trace, the maximum length difference is:
150ps / (5.4ns/m)
150p / 5.4n [m] = 28mm
The 1inch is a good approximation for 1m bus length.

Note:
If you want to design a "good" board, your traces must have a defined
impedance that matches the one of the cable to avoid reflections on the
connection. Look at Annex L how to calculate the impedance of different
stripline layouts:
ftp://ftp.t10.org:21/t10/drafts/spi4/spi4r10.pdf

If your board traces are not longer than the maximum stub length (100mm)
or if your whole bus is very short, you maybe don't want to use
striplines and let your design eat up the signal margin. For
asynchronous transfer mode, the "quick and dirty" solution is to simply
ensure that the REQ and ACK signals are the longest on the backplane.


Micha