Ultra SCSI SSWG Minutes

Ravinder S. Shergill rss at berlioz.nsc.com
Thu Mar 10 14:24:57 PST 1994

To:     SFF Reflector, SCSI Reflector
From:   Robbie Shergill, National Semiconductor
Subj:   Minutes of the Ultra SCSI SSWG
Date:   March 10, 1994

Minutes of the Ultra SCSI Special Subject Working Group (SSWG)
of the Small Form Factor (SFF) Committee.

Held on March 2nd, 1994, at Milpitas, California.

Chaired by:          Jim McGrath, Quantum Corp.
Hosted by:           Jim McGrath, Quantum Corp.
Minutes recorded by: Robbie Shergill, National Semiconductor Corp.

Jim McGrath brought the meeting to order at 9:30 AM and asked each person in 
attendance to introduce themselves. An attendance sheet was also circulated.


Name               Company                  e-mail
----               -------                  ------

Wally Bridgewater  Adaptec                 wally at adaptec.com
Mike Dreitlein     Adaptec                 ljlamers at adaptec.com
Norm Harris        Adaptec                 nharris at adaptec.com
Shahe Krakirian    Adaptec                 shahe at adaptec.com
Larry Lamers       Adaptec                 ljlamers at adaptec.com
Craig Stuber       Adaptec                 stuber at adaptec.com
Mark Knecht        AMD                     mark.knecht at amd.com
Joseph Chen        Cirrus Logic            chen at cirrus.com
Bill Gintz         Conner                  n/a
Louis Grantham     Dallas Semiconductor    grantham at dalsemi.com
Jens Paetau        Dallas Semiconductor    n/a
Michael Smith      Dallas Semiconductor    msmith at dalsemi.com
Bill Ham           DEC                     ham at shwsys.enet.dec.com
Ed Kavetsky        DPT                     kavetsky at dpt.com
Ron Roberts        Maxtor                  ron_roberts at maxtor.com
Mike Farhat        National Semiconductor  farhat at rockie.nsc.com
John Goldie        National Semiconductor  cjfgsc at tevm2.nsc.com
Steve Ichikawa     National Semiconductor  tsmisc at eccc.nsc.com
Todd Nelson        National Semiconductor  ctwnsc at tevm2.nsc.com
Robbie Shergill    National Semiconductor  rss at berlioz.nsc.com
Mark Jander        NCR                     mark.jander at colospgs.ncr.com
John Lohmeyer      NCR                     john.lohmeyer at ftcollinsco.ncr.com
Harry Mason        NCR                     n/a
Ting Chan          Q-Logic                 t_chan at qlc.com
Jim McGrath        Quantum                 jmcgrath at qntm.com
Brian Davis        Seagate                 (to Gene Milligan)
Steve Finch        SSI                     5723283 at mcimail.com
Vit Novak          Sun                     vit.novak at sun.com
Pete Tobias        Tandem                  tobias_pete at tandem.com
Kasem Elkhalid     TI                      4307725 at mcimail.com
Kevin Gingerich    TI                      4307725 at mcimail.com
Paul Aloisi        Unitrode                aloisi at uicc.com
Tak Asami          Western Digital         asami at dt.wdc.com

Jim McGrath briefly described the SFF Committee's objective and its 
workings. He then presented the proposed agenda for this meeting. The agenda 
was discussed briefly and adopted as follows:

         Introductions                     Mcgrath
         Review of SFF Rules               Mcgrath
         Objective of this SSWG            Mcgrath
         Date and Place of Next Meeting    Mcgrath
         Ultra SCSI (1) Concept            Ham
         Media Results                     Ham
         Timing Budget                     Ham


Jim McGrath reiterated his needs from the Ultra-SCSI specification as 
previously stated in his Ultra-SCSI SSWG document. Jim has customer that 
need next higher level of transfer rates, and they are primarily interested 
in short distance, inside the box connections. So, the cable length can be 
traded for speed. Jim knows that there are also other people that want 
faster speeds over longer distances, outside the box. Such out-of-box 
applications have to be differential; but the short distance, inside-the-box 
applications should be kept single-ended for power and cost reasons. 

John Lohmeyer raised the issue that he doesn't want this work continuing in 
SFF. He pointed out that this work started in X3T10; so why not keep it 
Jim McGrath said that he has no problem with that - perhaps the next meeting 
in Newport Beach can be a joint SFF/X3T10 meeting. John said that he doesn't 
see a need for even that as we should have basic agreemnet today.

Bill Ham pointed out that the SPI-2 project proposal is being prepared as we 
speak and Ultra-SCSI transfer rate can be part of it. John Lohmeyer added 
that the SPI spec has only one public comment against it and it is from 
Lohmeyer on behalf of the committee. It is to include SCAM into SPI. SPI-2 
project proposal will be posted to reflector within a few days.

Joe Chen asked about expectations over time. Jim McGrath said that he 
doesn't expect complete agreement today because people would want to go home 
and analyze. But he does expect this spec to be completed in 3 months' 


The next meeting of this SSWG will be held on March 31, 1994. It will be 
held at this same location (Crown Sterling Suites Hotel in Milpitas, CA) and 
will be hosted by Quantum Corp. 

(Note that the next discussion on Ultra-SCSI will take place in the SCSI 
working group in Newport Beach, CA, on March 16, 1994).


Bill Ham (DEC) took floor next and quickly went through his original 
presentation to X3T10. Key points of his proposal are:
 - intended as informative annex to SPI.
 - fully backward compatible with current SPI.
 - uses exactly the same physical connection.

Bill then presented the results of DEC's lab work. Bills' tester can go up 
to 30 mega transfers/sec (MT/s) and uses separate transceiver chips for both 
differential (National, DS36954) and single-ended operation (TI: 9-channel, 
BICMOS, SPI-compatible with 24-mA active-negation and termination on-board). 
The system sweeps across various data-to-clock time values and measures 
error rates.

Bill's differential system operated error-free (millions of transfers) with 
data-to-clock times of about 6ns to 40ns at 20 MT/s; and about 6ns to 27ns 
at 30 MT/s. 

More discussion took place on the single-ended results. With a shielded 
cable, 5 ft long, signals look good at 20 MT/s with little jitter and data-
to-clock times of about 7 to 44 ns allowed error-free operation over 
millions of transfers. With 1.5m flat-cable, thirteen wires-only stubs on 4" 
centers, and driving and receiving at mid-bus, times degrade only to about 8 
to 43ns at 20 MT/s. However, when Bill mixed 1.5m round cable with 1.5m flat 
cable with its associated stubs, the signal got very bad. The Vol level was 
only about 1v and jitter was high. data-to-clock times went to about 14 to 

Bill concluded that short distance and controlled rise and fall times appear 
to provide a very friendly environment for 20 Mt/s operation. Key question 
is "can we control the silicon well enough ?" 
(Bill will provide hard copies of his results at the Newport Beach meeting).


Bill Ham outlined a set of "ground zero" assumptions for further discussion:

1. The Ultra-SCSI device will maintain chip timing and other specs that are 
as good or better than the present (SPI) spec when operating in a non-ultra 

Larry Lamers asked if we are going to allow old devices (pre-SPI) on the 
bus?  Jim McGrath and Bill Ham are willing to go along with this limitation 
if needed.

2. We have almost a clean sheet of paper once the two devices are in Ultra-
SCSI mode (data transfer).

3. specific areas to look at:
   - window around slew rates
   - receiver input spec
      Larry Lamers mentioned that glich filter requirements have to be
   - max assertion level (Vol) 
      Wally Bridgewater (Adaptec) stated that perhaps the Vol should be
      spec'd in an AC way. paul Aloisi (Unitrode) doesn't want to put more
      energy into the bus.
      He and Bill Gintz (Conner) asserted that the focus should be on the
   - setup and hold times
   - active negation parameters
   - clarifying how timings are defined.

John Lohmeyer asked about any possible emi issues.Bill Ham said that he 
hasn't looked at it yet, but older data says that data traffic is not where 
emi peaks are.
Ting Chan said that it may not be a concern at 20MT/s but should be 
considered for rates above that.

4. Specify narrower chip (system) operational parameters that are needed in 
normal office/computer room environment.

5. Chip specs that can be met by at least two siliocn suppliers using 
established cost structures - ie, no BC process. 

6. Stay with present packaging technology, but pinout changes are okay.

With the above list agreed to, Kevin Gingerich (TI) asked that we narrow 
down the system operational parameters before we get into details of 
silicon. This was agreed to and Bil Ham prioritized the above list in order 
of discussion.

However, the discussion soon moved into skew budget issues.
Bill Ham said that at least two cable manufacturers are telling him that 
cable skew (currently 4ns) can be cut to 2ns. Taking 2ns each from cable and 
distortion skew, we can gain 4ns in the budget.

The discussion moved into actual operational parameters such as loading and 
power supply range until Larry Lamers asserted that we're getting too deep 
in; let's back off and look at the skew budget.

At this point, John Lohmeyer presented NCR's proposal on skew budget.

  total cable skew = 4ns
  tx chip s/h = 12/17
  board skew still 1ns at each end
  rx chip s/h = 6/11ns

  total cable skew = 4ns
  tx driver skew = 4ns
  rx receiver skew = 5ns
  board skew still = 1ns each
  tx chip s/h = 16/21
  rx chip s/h = 1/6

John asked if differential transceiver skew geting cut in half is okay with 
TI and National as this was a hotly debated SPI spec. Kevin Gingerich (TI) 
said that this is okay if the environment parameters are okay. John Goldie 
(National) said that John's numbers look feasible.

Larry Lamers showed Adaptec's proposed skew budget numbers. Adaptec shows 
assertion time degrading by only 2ns from tx to rx side. John Lohmeyer 
observed that Adpatec numbers will have trouble when applied to the 
differential case.

General sense of the group was that NCR and Adaptec proposals are close 
enough that any differences can be worked out after further analysis. NCR 
and Adaptec will bring revised proposals to the Newport Beach meeting.

Larry wanted to discuss the receiver switching points before dissecting the 
skew numbers any further. In particular, Adaptec wants to increase 
hysteresis to 0.4v and redefine spec measurement points with hysteresis 
considerd. Adaptec also wants to eliminate the glitch filtering requirement 
if the group agrees to increase hysteresis to 0.4 volts. No one else offered 
any objections on these points.

Steve Finch (SSI) drew some timing diagrams that served as basis for further 
discussion on timing measurement points. After significant discussion, the 
following measurement points were agreed to:

At the receiver and at the driver:

      --------                                 --------
              \                               /
               \               |       2.0v -/-
                \              v            /
                 \         --------- 1.6v -/-
                  \        nom. hyst      /|
                  -\- 1.2v ---------     / |
                    \          ^        /  |
                    -\- 0.8v   |       /   |
                     |\               /    |
                     | ---------------     |
                     |                     |
                     |<- min assert time ->|

                      /                       \
               2.0v -/-                        \
                    /|                          \
                   / |                           \
                  /  |                            \ 
                 /   |                            -\- 1.2v
                /    |                             |\
               /     |                             | \
              /      |<---- min negation time ---->|  \
     --------/                                         \---------

     ------        -------------------------------------        ------
           \     -/- 2.0v                               \      /
            \    /|                                1.6v -\-   /
             \  / |                                      |\  /
 Data         \/  |                                      | \/
              /\  |                                      | /\
             /  \ |                                      |/  \
            /    \|                                1.2v -/-   \
           /     -\- 0.8v                               /|     \
     -----/       |\-----------------------------------/ |      \------
                  |                                      |
                  |<--- setup time --->|                 |
                         ---------     |                 |
                                  \    |                 |
                                   \   |                 |
                                    \  |                 |
                                     \ |                 |
 Clock                                \|                 |
                                 1.2v -\ |<- hold time ->|
                                   0.8v -\-

Rise and fall times were discussed next. Should the spec remain in terms of 
10% and 90% points, or should it be changed to specific voltage levels such 
as 0.5 volts and 2.5 volts? There was agreement that the current SPI 
specification needs to be enhanced in this this area as the 0% and 100% 
("full amplitude") voltage levels are not clearly defined. 

Paul Aloisi (Unitrode) pointed out that maximum high voltage has to be 
specified, if 10/90 % points are used, so that your tr/tf absolute numbers 
don't kill the budget.

Robbie Shergill (National) pointed out that the good thing about 10% and 90% 
points is that the full transition range, where the greatest slewing takes 
place, is covered. Many people agreed that the best thing to do is to 
specify the maximum slew rate. Finally, the group agreed on 400mV/ns maximum 
on the test circuit. This is roughly the same as the current SPI spec.
Tak Asami (WD) wanted the present load circuit to be changed. All agreed 
that this needs to be looked at.

Bill Gintz (Conner) wanted receiver test conditions also to be specified. 
All agreed that this needs to be looked at.

Robbie Shergill wanted the System Deskew Delay time to be defined. John 
Lohmeyer clarified that this time is defined as half of total skew in system 
contributed by the two nodes (other than cable plant).

It was also mentioned that ground voltage offset spec. for the single ended 
case should be defined as well and the current RS-422 based spec for the 
differential case should be refined.

In the end, the group came up with the following wish-list for further lab 
work on Bill Ham's tester:

  a. stubs actually loaded with devices (all SPI-compliant, 16-bit devices).
  b. receiver a minimum distance (4") away from driver, which is at the end.
  c. cable distance of 1.5m and 3m.
  d. loaded with up to 16 devices.

The meeting was adjourned at 3:45 PM.

More information about the T10 mailing list