X3T9.2/92-43 Date: February 18, 1992 To: X3T9.2 Membership From: Lawrence J. Lamers, X3T9.2 Secretary Bill Spence, Chair SPI Working Group Subject: February 18, 1992 SPI Working Group Meeting Minutes Bill Spence called the meeting to order at 9:00 p.m. February 18, 1992. He thanked Carl Zietler of IBM for hosting the meeting. The SPI working group is the umbrella for all contact, connector, cable, transceiver issues related to SCSI-2 Parallel Interface (SPI). Bill Spence chairs this working group which is chartered with developing a set of recommendations for the SPI standard that will get the physical plant to a reliable state. The final agenda was as follows: 1. Old Business 1.1 Review of the new draft of the SPI working document --Kurt Chan 1.2 Glitch filter analysis--Florin Oprescu 2. New Business 2.1 Fast SCSI Timing Problems and Solutions--Kurt Chan 2.2 Note on effect of bus length--Bill Ham 2.3 Driver and receiver data revisited--Bill Ham 2.4 Proposal re termination of reserved lines--Bill Spence 2.5 Studies in hot plugging--Bill Ham The following people attended the meeting: Name Status Organization ------------------------------ ------ ------------------------------ Mr. Alan R. Olsen P 3M Company Mr. Robert C. Herron A 3M Company Mr. Thomas Newman P Adaptec, Inc. Mr. Tony Castillo O Aeronics, Inc. Mr. Charles Brill P AMP, Inc. Mr. Bob Whiteman A AMP, Inc. Mr. Ken Schreder S AMP, Inc. Mr. K. Talentimo V Amphenol Mr. Jeff Rosa P Amphenol Interconnect Mr. Richard M. Ross O Amphenol/Spectra-strip Mr. Scott Smyers P Apple Computer Mr. Florin Oprescu O Apple Computer Mr. Douglas Brenneke O Belden Wire & Cable Mr. Eric Lawrence O Brand-Rex Company Mr. Edward Hrvatin P Burndy Corp. Mr. Clifford E. Strang, Jr. V BusTek Mr. Kurt Witte P Ciprico Inc. Mr. Joe Chen P Cirrus Logic Inc. Mr. John Geldman A Cirrus Logic Inc. Mr. Edward Haske O CMD Technology Mr. Stephen R. Cornaby P Conner Peripherals Mr. Michael Smith V Dalla Semiconductor Mr. Charles Monia P Digital Equipment Corp. Dr. William Ham A Digital Equipment Corp. Mr. Paul Hanmann P Emulex Corp. Mr. Skip Jones A Emulex Corp. Mr. D. W. Spence A ENDL Associates Mr. Edward Lappin P Exabyte Corp. Mr. Stan Sharp P Foxconn International, Inc. Mr. Joel Urban O Fujitsu Components of Amer. Mr. Robert Liu P Fujitsu Computer Products,Am Mr. Steve Caron P Furukawa Electric America Mr. Kurt Chan P Hewlett Packard Co. Mr. Tom Kulesza O Honda Connector Mr. George Penokie P IBM Corp. Mr. Larry Grasso S IBM Corp. Ms. Gricell Co V IBM Corp. Mr. Behrouz Rasoulian V IBM Corp. Mr. Geoff Barton P Iomega Corp. Mr. Robert Bellino P Madison Cable Corp. Mr. Lawrence J. Lamers P Maxtor Corp. Mr. Raymond C. Yule O Micropolis Corp. Mr. Robbie Shergill P National Semiconductor Mr. John Lohmeyer P NCR Corp. Mr. Stephen F. Heil P Panasonic Technologies, Inc Mr. Gene Milligan A Seagate Technology Mr. Robert L. Simpson P Sony Corp. of America Mr. Gregory Green O Stratus Computer Mr. Vit Novak O Sun Microsystems, Inc. Mr. Pete Tobias O Tandem Mr. Ricardo Dominguez P Texas Instruments Mr. Richard Mourn O Texas Instruments Mr. Sam Washburn O Wangtek Mr. Doug Piper P Woven Electronics 54 People Present Status Key: P - Principal Member A - Alternate Member O - Observer S,V - Visitor Results of Meeting 1. Review of SPI Draft Document Gene Milligan raised the question of how this working group is going to get its recommendations into effect at the plenary. Bill Spence responded that the draft document will contain the less controversial consensus items and those that are controversial may require a separate plenary vote. John Lohmeyer suggested that significant individual items have a plenary vote. Specifically the removal of the 220/330 terminator will be an agenda item for a confirming vote. Section 5.2 -- Max allowable variation of impedance within a cable: the previous meeting's decision for 8 ohms did not stand up to further investigation. Bob Bellino indicated that Madison cables are within 10 ohms. Bill Spence had a "max max" possible variation of 13 ohms for foamed polyolefin cables from David Hess of Berk-Tek. Kurt had chosen 12 ohms for the draft standard, and that was not changed. -- It was agreed to allow using AWG 30 conductors for TERMPWR lines in P and Q cables because there are multiple TERMPWR lines. The four AWG 30 lines have less resistance than one AWG 28 line, 2.06 ohms to 5.2 ohms, for the same length. Section 5.2.3 -- It was pointed out that the attenuation spec fails to identify the mode of measurement. The mode intended when the spec was first placed in the standard was differential, and this needs to be recognized therein. -- The previously adopted figure of 0.10 ns max per meter for pair-to-pair propagation delay delta was raised to 0.15 ns max per meter. -- Chair's note: There probably is a consensus that, at least within the recommended max length of 3 meters, AWG 30 conductors are acceptable for fast SCSI. No useful purpose is served by giving a dc resistance spec; it should be replaced by a requirement that the conductors be no smaller than AWG 30. It may be that the use of resistance was an attempt to internationalize the spec; but for it to be meaningful, all the many other references in the standard to the American Wire Gage would also have to be changed to resistance form. -- It was again agreed that methods of measurement of cable specs shall be included in an annex. Additional specs proposed for inclusion included low and high frequency figures for attenuation, rise time degradation, and perhaps others from the proposals of our earlier studies. It was proposed that Florin Oprescu bring in a proposal on the subject. -- Bill Spence stated that since our earlier studies, the situation may have changed. Clearing up the confusion about measurement mode for specs-- particularly impedance--and establishing proper impedance and attenuation limits have had the practical effect of requiring that the round, shielded, twisted-pair cables used in longer buses use polyolefin or better dielectric. This effect, plus the assignment of specific signals to specific cable layers to control crosstalk, may be all the cable specs that the standard needs. More may simply add useless burden. Section 6.1.1 -- The subject of driver signal rise and fall times came up again. John Geldman had objected to the proposal that 10-90% rise and fall times be 5 ns min, but he was pretty much overridden. But the new draft standard came in with the min times applying to 20-80%--even more hostile to fast SCSI. Florin supported the new limits and suggested raising the figure to 7 ns min. John Geldman again objected. The matter was left open for San Diego, in hopes that more silicon people will be present. Section 6.1.2 -- Bill Ham questioned whether or not the entire 0.8 - 2.0 v receiver input voltage span is really needed. Bill Spence had also raised that question (92-209), without eliciting much response. He now supported Bill Ham's question. John Geldman had reservations on this, saying it would burn up more power. Previous objections had been that the degree of precision involved was impractical. No action taken. -- John Geldman next seemed to argue on the other side in objecting to the reduction of the nominal switching threshold to 1.2 v from 1.4 v. He rejected the argument that it was a moot issue, since receiver design precision has been held to be barely adequate to ensure that the transition will occur somewhere in the 0.8 to 2 volt span. In the end, it was agreed to put the nominal threshold back to 1.4 volts. Section 7.0 Tbl 12 -- Setup and Hold times. The amorphous magical DESKEW time has a circular definition in the Appendix. Specification was written with silicon in mind but it requires measurements at the connectors resulting in ambiguity. Kurt will work on it. 2. Glitch Filter Analysis [Oprescu] Florin Oprescu presented a proposal for glitch immunity. The figures represent real world observations of signals on a 6 meter single ended SCSI bus, pretty much worst case. He shows the areas of signal excursion which should be rejected by a receiver filter, working in time domain or frequency domain or both. He stated that glitch eaters exist that meet the envelopes. He further suggested reducing node capacitance some more (to 15pf), and increasing transition time (to 7ns, 20-80%) (see above). Several objections were raised: the glitch specs could not be applied at 5 MHz, much less 10. The 15 pf would rule out most of even the best boards in use today, and the 7 ns min transition time is probably impractical as a fast SCSI spec. Florin's replies addressed the needs of Apple more than the needs of the standard: he is not at this time interested in running faster than 2.5 MHz (he will switch out his glitch eater if he wants to go faster); he is designing boards with entry and exit SCSI ports to permit minimal stub length and lumped capacity, and 7 ns is compatible with his 2.5 MHz. Florin is requiring his glitch eater for Apple systems. Note re node capacitance: High node capacitance devices can look good individually because they have smoother transitions; the effects in a multiple peripheral system are horrendous. Rules of thumb: trace--3 pf per inch, connector junction--5 pf, ic node--5 pf. Bill Spence asked about a consensus for reducing the 20 pf to 15, it did not happen. The glitch screen question first raised by David Steele in 91-64R3 thus remains open. Florin was asked if he would recast his filter spec (1) for the tightest filter which would permit 5 MHz operation and (2) the tightest filter which would permit 10 MHz operation. 3. Note on effect of bus length [Ham] Bill Ham spoke on SCSI bus integrity, bus length, and propagation delays. He showed signal traces of a seven meter and an eleven meter bus in a fully loaded system. The signal at the driver does not get the second step effect on 5 MHz pulses (and thus may fail to achieve 2.0 volts) in the eleven meter bus. The longer cable has a longer propagation delay. The noise margin is significantly better if the cable length is below 6 meters (5 Mhz) or 3 meters (10 MHz). 4. Driver and Receiver Data revisited [Ham] Bill Ham spoke on the needed specifications for SPI transceivers. Most of his issues have been addressed in the last few months. He added one new one, that power up/power down be glitch free. Bill Spence stated that this would need a plenary vote. What a glitch is needs to be determined. Bill Ham will work on this. Do we want to embark on specifying input filtering of receivers? Florin stated that a hysteresis spec does not do him any good at a system level. Bill Spence gave a little speech on the universe of parameters, winnowing them down to offer some guidance to implementors, but not being able to eliminate good system engineering. John Geldman pointed out that if a time domain is developed then a hysteresis spec is not needed. Clamping is a chip spec issue. ESD is a chip spec issue. Latch up is a chip spec issue. The maximum high state input voltage level as a standard issue suggests 6.0 volts. 5. Fast SCSI Timing Problems and Solutions (92-032R0) [Chan] Need separate specification for silicon. Appendix B needs better description. Proposal is to establish a transmit setup time, a transmit hold time, a receiver setup time, and a receiver hold time. For differential transceivers should add note to operate off of same voltage and also indicate a temperature range. Need to establish a skew budget for all skew behind connector except protocol chip. Suggested 8ns for transmitters and 13ns for receivers. Assumes 25 meter systems. The worksheet did not come out as Kurt thought, it indicated that there is a problem with skew for at least one vendor. This area still needs a lot of work. 6. Proposal regarding termination of reserved lines [Spence] Bill Spence stated that it would be better to terminate normally the reserved lines rather than grounding them. Grounded lines can radiate. Florin objected to this, as did Bill Ham. Debate was cut short in favor of spending time on hot plugging, but there is more to come next month. 7. Studies in hot plugging [Ham] Bill Ham's fundamental question is "How do you absorb the energy during hot plugging?". Bill assumes that you are adding a certain small capacitance, the device being hot plugged is not active on the bus, the bus is running, the device has ground integrity, but the make/break of the power is asynchronous with the make/break of the bus conductors. Bill claims worst case is: high device capacitance, fast transitions, devices nearby, insertion or bounce with data state change, high impedance cables. Worst case hot plugging disturbance is 700 mv for 5ns. DSSI uses a low pass filter to eat this glitch in slow operation. This filter is built into the silicon. Florin stated that in his experience hot plugging causes system failures. Bill Spence brought out that this occurred in the absence of establishing grounding first. Where do we go from here? The type and magnitude of glitches due to hot plugging if done right are no worse than what the normal bus typically sees due to stubs. Bill Ham agreed to assemble his data into a paper for the next mailing. 8. Meeting Schedule The SPI Working Group meeting for March is on Tuesday, March 17, 1992 at 1:00 pm at Humphreys Half Moon Inn in San Diego, CA. The SPI Working Group meeting for April is on Tuesday, April 28, 1992 at 9:00 am at the Tradewinds Hotel in St. Petersburg, FL.