X3T9.2/91-218 Date: December 10, 1991 To: X3T9.2 Membership From: Lawrence J. Lamers, X3T9.2 Secretary Bill Spence, Chair SPI Working Group Subject: December 10, 1991 SPI Working Group Meeting Minutes Bill Spence called the meeting to order at 9:00 a.m. December 10, 1991. He thanked Rich Bailey of Vitro Corporation for hosting the meeting. The SPI working group is the umbrella for all contact, connector, cable, terminator, and transceiver issues related to SCSI-3 Parallel Interface (SPI). Bill Spence chairs this working group which is chartered with developing a set of recommendations for the SPI standard that will enhance the reliability of the physical plant. The final agenda was as follows: FIRST SESSION--CONCEPTS 1. The Trung Le/Aeronics FPT Terminator -- Steve Ego of Aeronics, others ad lib. - Is there or is there not an MTBF issue? - Is any change or addition to the SCSI-3 Standard called for? 2. Quantification of Bus Goodness -- Robert Allgood of IOMEGA (X3T9.2/91-213) 3. Data which System Integrators Need from Chip Producers -- Bill Ham of DEC 4. The Active Negation Driver -- Florin Oprescu of Apple, Erik Jessen of Western Digital, David Steele of NCR, Sassan Teymouri of AMD, Rick Muething of Emulex, others 5. Reducing SCSI Impedance Levels -- Bill Spence of ENDL Consulting, Robert Allgood, Rick Muething, others (X3T9.2/91-209) SECOND SESSION--SPECIFIC PROPOSALS 6. X3T9.2/91-64R2 12/07/91 David Steele Fast S/E and Other Changes 7. X3T9.2/91/177 11/04/91 Bill Spence Terminator Resistor Values 8. X3T9.2/91/178 11/04/91 Bill Spence High-End Terminator Specs 9. X3T9.2/91/179 11/04/91 Bill Spence Cable Stock Specs 10. X3T9.2/91/187 11/04/91 Florin Oprescu Portable Machine Terminator 11. X3T9.2/91/198 11/13/91 Rick Muething SCSI-3 SPI Recommendations 12. Meeting Schedule The following people attended the meeting: Name Status Organization ------------------------------ ------ ------------------------------ Mr. Robert C. Herron A 3M Company Mr. Thomas Newman S Adaptec, Inc. Mr. Sassan Teymouri A Advanced Micro Devices Mr. Steven P. Ego O Aeronics Inc. Mr. Charles Brill P AMP, Inc. Mr. Bob Whiteman A AMP, Inc. Mr. Jeff Rosa P Amphenol Interconnect Mr. Jan V. Dedek P Ancot Corp. Mr. Florin Oprescu O Apple Computer Mr. Chris Nieves O AST Research Mr. Douglas Brenneke O Belden Wire & Cable Mr. David Hess P Berk-Tek, Inc. Mr. Geoffrey Hyman O Cables To Go Mr. Kurt Witte P Ciprico Inc. Mr. John Geldman S Cirrus Logic Inc. Mr. Edward Haske O CMD Technology Mr. Norman H. Harris P Digital Equipment Corp. Dr. William Ham A Digital Equipment Corp. Mr. Rick Muething S Emulex (Consultant) Mr. Paul R. Nitza A Emulex Corp. Mr. D. W. Spence A ENDL Associates Mr. Edward Lappin P Exabyte Corp. Mr. Denni Rash O FileNet Corp. Mr. Robert Liu P Fujitsu America, Inc. Mr. Steve Caron P Furukawa Electric America Mr. Tom Gafford V Gafford Technology Mr. Russ Golish O Hi-Tech Computer Solutions Mr. Tom Kulesza O Honda Connector Mr. George Penokie P IBM Corp. Mr. Gary R. Stephens A IBM Corp. Mr. Giles Frazier S IBM Corp. Mr. Larry Grasso V IBM Corp. Mr. Kevin R. Pokorney A Intellistor, Inc. Mr. Geoff Barton P Iomega Corp. Mr. Robert D. Allgood S Iomega Corp. Mr. Kenneth E. Smith V Kres Engineering Mr. Robert Bellino P Madison Cable Corp. Mr. Lawrence J. Lamers P Maxtor Corp. Mr. Doug Pickford A Maxtor Corp. Mr. Jim McGrath A Molex Inc. Mr. John Lohmeyer P NCR Corp. Mr. David Steele S NCR Corp. Mr. Bruce Anderson P NEC Technologies Inc. Mr. Bill Medlinski P Panasonic, MECA-BEC Mr. Steve Kennedy O Quantum Corp. Mr. Brian N. Davis O Quantum Corp. Dr. Tetsuro Motoyoma P Ricoh Corporation Mr. Gene Milligan A Seagate Technology Mr. Ricardo Dominguez P Texas Instruments Mr. Richard Mourn O Texas Instruments Mr. Bob Mammano V Unitrode IC Corp. Mr. Erik Jessen O Western Digital Mr. Doug Piper P Woven Electronics 53 People Present Status Key: P Principal A Alternate O Observer S Special Interest (frequent visitor) V Visitor Results of Meeting 1. The Trung Le/Aeronics FPT Terminator Bill Spence introduced this topic by stating that there are three approaches to improving signal quality on the SCSI bus: raise the cable impedance, lower voltage range swing, raise current in cable by using a Boulay or Aeronics type terminator. But questions had been raised as to whether the 60 ma at 0.2 volts typically sourced by 2 FPT terminators is acceptable under the standard and is acceptable from a MTBF standpoint. Bill stated his conclusions that the MTBF concern is insufficient to make it inadvisable to use the FPT terminator where it is needed, and that it is unlikely that a consensus can be reached to modify the standard specifically to accommodate it. Steve Ego made a presentation on forced perfect termination (FPT) in its Light incarnation (X3T9.2/91-215). What the Light FPT does is add current limiting to keep the assertion driver current under 48 ma. In a system with the host at or near an end of the bus, FPT Light still provides an additional 400 millivolts of noise immunity compared to 220/330 or Boulay, according to the data presented. In this incarnation, there is no question about violating the standard. With respect to the original FPT design, there continued to be no consensus for endorsing the level of current sourced to an assertion driver, because of MTBF concerns. Steve stated that test and field experience data so far is very favorable and will be reported to the committee when assembled. Steve's presentation made a point of the fact that 74F38 chips have a military rating of 48 ma, a commercial rating of 64 ma, and an absolute rating of 128 ma. Two standard FPT terminators source about 60 ma at very low assertion voltages. David Steele stated that if a worst case driver can only sink 48ma, it will not be helped in its first negation step with this type termination. But the internally-regulated Boulay termination will help in all low Vterm situations. 2. Quantification of Bus Margin--Impedance Optimization Robert Allgood presented a spreadsheet analyzing noise margins provided for a wide variety of single-ended parameters. See 91-213 for details. The Excel spreadsheets used for the analysis are available on the SCSI BBS. Results were displayed in plots which generally showed the Low State noise margin increasing with increasing cable impedance, the High State noise margin decreasing with increasing cable impedance, and the optimized point where both margins are equal to be where the cable impedance is pretty near to the terminator impedance. The paper recommended that SCSI-3 set nominal impedances for the terminators and cables; reclaim the 1.6 ma used for input current for the assertion current to help the high state margin; and limit the upper (negated) voltage to define and control inrush current. Bill Spence noted that it appeared Robert's model for calculating assertion noise margin does not include the effect of assertion propagation losses. He stated that the negation signal typically improves as it propagates down the bus, but that the assertion signal degenerates. Having intervening devices on the bus helps negation, hurts assertion. He knows of no analytical model that explains this phenomena. But he thinks that this effect causes optimum impedance to be lower than Robert's paper indicates. He stated that when it is desired to maximize the bus length, his data shows cable impedances as low as 0.75 times the terminator impedance to be optimum. Florin Oprescu stated that there is a model that explains why the presence of intervening devices on a long bus works against signal assertion much more than against signal negation. He stated that we should start with slower edges (controlled rise times) and reduce the capacitance per device to 15 pf. And lower impedance cables reduce the effect of the lumped capacitance, adding further pressure toward lower impedance cables. 3. Data that System Integrators need from Chip Producers Bill Hamm asked if there is data from the chip vendors that could be shared to help analyze the problem. The fast single-ended discussion is driving this. Concerns: - What is required to go above 3 meters in fast SCSI? - Which devices can be mixed and still play OK together? - Hot plugging - How to model the chips in a system - Ground offset voltages - Design homogeneity within chip families - Where and how much is the unused margin? What we don't know today about drivers: - Open-collector or active negation? On which lines? - Guaranteed assertion levels? - Guaranteed slew rates? - Worst case skew? - Active negation drivers--Terminal characteristics --Time it is on --Transients when going to tristate - Terminal capacitance and inductance - Latchup and ESD protection What we don't know today about receivers: - DC switching threshold - Hysteresis thresholds--test conditions - Any filtering--what type--which lines - Worst case skew - Glitch rejection characteristics - Latchup and ESD protection - Tri-state leakage - Power-down leakage - High-signal-voltage damage level--power on and off Representatives of chip producers present noted these points for guidance of their marketing departments. 4. The Active Negation Driver Bill Spence went over Florin Oprescu's theoretical analysis of active negation drivers (in X3T9.2/91-188). It offered models for active negation drivers which would support operation with a wide range of cable and terminator characteristics and yet not cause damage to a unpowered SCSI device on a SCSI bus with one terminator missing. Florin wants to limit active negation to prevent damage to devices from reflection on an unterminated bus. Others disagreed that the hazard exists significantly. Finally it came out that the hazard exists when there are devices on the bus which are not powered up. Florin stated his position that avoiding this hazard is his No. 1 priority, exceeding his concern for the strength of the active negation driver. Florin and the driver people huddled over lunch. After lunch Florin stated that his concerns are met if this condition is met: Voh < 3 v @ Ioh > 24 ma. There was consensus that this limitation on negation driver strength is acceptable. David Steele pointed out, however, that this limitation is too loose to protect the 2.85 volt bus in a Boulay terminator from being driven up out of regulation. Final consensus was pretty strong that the negation driver strength should not be additionally limited sufficiently to fully protect the Boulay regulator. It was stated that no present Boulay regulator will become unstable because of a net infeed of current from the lines. And some Boulay terminators will sink some amount of net infeed before losing regulation. This leaves it up to the system integrator to evaluate the risks to his assertion drivers. Most feel the risks are small, but this resolution is not ideal. A request was made for greater guidance about active negation characteristics than the single point of 3.0 volts @ 24 ma. Florin stated that he favors a limit of 3.0 volts for all currents above 24 ma and a limit which can be expressed as 3.6 volts less 0.025 times the current, expressed in ma, below 24 ma. The matter of a minimum strength requirement in active negation drivers was brought up, but nothing pertinent developed. Here as well as at other times during the day the conflict came up between spec'ing exactly how to implement SCSI and establishing limits which allow for a range of implementations. Bill Spence pointed that the usual concept of a standard is that it presents a minimum set of requirements for interoperability. But in a multidimensional space like SCSI, any practical set of limits is almost bound to allow implementations which don't work, as well as point toward those which do. 5. Reducing SCSI impedance levels. Bill Spence pointed out that most of the current efforts in the physical layer could be viewed as impedance adjustment--increasing the line impedance or decreasing the impedance requirements by raising the driver current limit. But an alternative approach is to reduce the required receiver voltage swing-- as long as the hysteresis is not reduced. Specifically, maintain the low receiver requirement at 0.8 volts but move the high requirement down to 1.4 volts, with a hysteresis of 400 millivolts. It was agreed that this approach is not feasible with present receiver fabrication techniques. Whether a market could develop which would justify a more expensive SCSI chip with an improved receiver is a question. (Ref: X3T9.2/91-209) Erik Jessen suggested that if we were to go this far, we should make REQ and ACK differential. If this could be interoperable with older SE, then we could have a solution. REQ and ACK are the only edge-triggered signals. 6. Single-ended Recommendations [Steele] (91-064R2) Scn 5.1 Physical Description. No objection was voiced to anything in this recommendation. It was little discussed, however, in favor of more quantitative questions. Scn 5.2 Cable Requirements. This recommendation does not address the distinction between S/E and differential impedance (which is addressed in 91-179 below) but rather what the standard should say about S/E impedance. It was agreed that it should list a target figure plus a tolerance, and 95+/-15 ohms was suggested. Bob Whiteman raised the point that this rules out 0.025 pitch, 28 AWG, PVC ribbon cable (75 ohms) currently being used with high- density connectors. It was pointed out that the use of TPE (thermoplastic elastomer) or other high-grade dielectrics overcomes this problem. Bill Spence spoke in favor of a lower span--perhaps 90+/-15 ohms. He also favored the use of AWG 26 wire in 0.050 pitch, PVC ribbon cable to get its impedance down to better levels (89 vs 108 ohms). In the end it appeared that there was a consensus in favor of 95+/-15 ohms. We should remind Kurt about adding Jim Fiala's 90-134 on Cable Test Procedures as an appendix. Scn 5.2.3 Cable Requirements for Fast Synchronous Data Transfer. No objection was voiced to any part of this recommendation except that 95+/-15 ohms should be substituted for 80 to 100 ohms. However, Robert Allgood pointed out that 0.23 ohms per meter DC resistance precludes use of 30 AWG, and that while a DC resistance is ok for TERMPWR, attenuation should be spec'd for signals. Bill Spence pointed out that the present attenuation spec is for differential mode at 5 MHz, vs S/E mode at 10 MHz which would be appropriate for a S/E Fast SCSI spec, and that no existing cables could meet 0.95 db/m in such conditions. But the present spec serves to rule out PVC and other low quality dielectrics, and so it probably meets the need. Scn 6.1.1 Single-Ended Output Characteristics. This recommendation led to considerable debate about both slew rate and method of specification. Florin Oprescu spoke in favor of specifying 10-90% points rise times and in favor of relatively long (6 ns) required times. David Steele pointed out that in Fast SCSI such times are too long. Rick Muething's 91-198 proposes a test structure. David agreed to rework this issue. The recommendations about ACK, REQ, and DATA BUS signals elicited no objection at this point, although earlier there had been a consensus that requiring active negation drivers not to be strong enough to cause a net inflow of current into Boulay terminators was overly restrictive. Scn 6.1.2 Single-Ended Input Characteristics. There was discussion over the method of specifying the parameters but no objection to David's proposed values. David stated that the following paragraph on rejection of noise spikes was not easily come by. Sassan Teymouri mentioned that AMD is working a patent on their "glitch-eater." He was requested to furnish further information to John Lohmeyer. Florin Oprescu volunteered to bring in a proposal on the noise spike rejection subject. 7. Terminator Resistor Values [Spence] (91-177) Robert Allgood calculated terminator resistor values which produced an even lower impedance value than the 187/267 ohms values Bill had calculated, making it applicable with lower TERMPWR voltages. George Penokie pointed out that the fuse or other current-limiting element, not to mention contact resistances, typically are even more damaging to TERMPWR voltage than is the Schottky diode, and that TERMPWR voltage is quite unreliable. Many agreed. A consensus emerged that rather than change the resistance values in the Alt 1 terminator, we should drop that termination from SCSI-3. 8. High-end Terminator Specs [Spence] (91-178) In the end, this proposal was accepted with the deletion of the 3rd bullet-- the IMPLEMENTORS NOTE. There was an initial objection to the reference to 2.5 rather than 2.85 volts. But later there was a consensus that the 2.85 volt level in terminators needs to be migrated downward. So the recommendation in 91-178 that current be sourced when the voltage is below 2.5 volts is okay as is--it does not preclude some current being sourced at higher voltages. This leaves unaddressed the question of the high sink currents in assertion drivers when low impedance cables are discharged from 2.85 volts. 9. Cable Stock Specs [Spence] (91-179) There was no objection raised to anything in paragraphs 1 and 2 of this proposal, except that the previous consensus for 95+/-15 ohms pre-empts the 75-110 ohms proposed and except for the single-ended length spec. There was much concern about permitting longer than 6 meter implementations. Bill Spence reiterated his observation that there is a general increase of difficulties with increasing bus length, but that there is no significance to 6 meters and that it is far below a reasonable limit. Bill Ham stated his observation that difficulties seemed to increase markedly above 6 meters. Several others expressed concern about encouraging extension past 6 meters. George Penokie suggested eliminating the prohibition above 6 meters but retaining a recommendation for 6. David Steele's dance around 3 meters in the Fast SCSI section seems applicable. 10. Portable Machine Terminator [Oprescu] (91-187) Discussion of this proposal was limited both because of the time and because of obviously limited audience familiarity with the topic. There was considerable interest evidenced. 11. 91/198 SCSI-3 SPI Recommendations [Muething] (91-198) Rick had run out of time and had left by the time this item was reached. He had input his ideas in connection with several of the previous items, however. 12. Meeting Schedule The SPI Working Group meeting for January is on Tuesday, January 14, 1992, at 1:00 pm at the Holiday Inn Astrodome in Houston, TX. It is expected that it will run well past 5 pm. The SPI Working Group meeting for February is on Tuesday, February 18, 1992, at the Doubletree Hotel in Austin, TX. The meeting adjourned about 5:30 pm.