X3T9.2/91-168R2 Attn: SCSI Driver and Terminator Specialists John Lohmeyer, Chairman, X3T9.2 22 Oct 1991 From: Bill Spence Subject: Minutes of the 10/10 Irvine S/E Driver/Terminator Conference Note: Rev 1 adds the roster of attendees, the input about the Methode SLIM terminator on p 3, the P.S. re the Albuquerque working group meeting, and the follow-up FAX from Jean Kodama in digital form. The date of the November WG meeting was corrected in this revision. The date when this topic will be covered should have been 11/13/91. BACKGROUND Minutes don't usually have backgrounders, but this one does. Many system integrators have solved their s/e SCSI bus problems by using the three remedies developed in X3T9.2 over the past year: polyolefin cables, the Boulay termina- tor, and proper signal placement in the cable. By a combination of good design and good fortune, they find that they can fully load the bus and extend it well past 6 meters with good reliability and acceptable signal waveshapes. Other system integrators have not fared as well. They, along with PC, work- station, and peripheral vendors who wish to assure s/e SCSI bus performance without having control of the bus implementation (incredible task), are most eager for more powerful remedies. Two such are active negation drivers and the Trung Le/Aeronics signal-limiting, high-current-sourcing terminator. Because of perceived conflicts between these devices and between active-negation drivers and the Boulay terminator, the subject conference was convened. RESULTS SUMMARY Minutes don't usually have Summary sections, either, but the minutes below are so detailed that many may be grateful for a summary of what happened. Briefly: - There is near universal opposition among the silicon people to permitting any increase in the 48 ma driver current limit, even at the 0.3 v and below assertion voltages achieved. In other words, they reject the use of the Trung Le/Aeronics terminator with existing SCSI drivers. - They allow that a new generation of drivers could have higher limits. One participant, however, has presented a set of cogent arguments for not taking that path. - There are quite divergent opinions as to what characteristics active negation drivers should exhibit. No one foresees incompatibility with the Trung Le/ Aeronics terminator in the negated state, however. But some designs do conflict with the Boulay terminator, and some are not self protecting against shorts to ground. - A new, potentially serious problem with active negation drivers was presented: damage from overvoltage transients if operated with a terminator missing. - Action items include presenting new proposals at the Albuquerque working group meeting. DETAILED MINUTES The meeting was called to order by myself on schedule at 10:00 am. Skip Jones of Emulex, who served as host, was not able to be present, but there was a large Emulex presence. The meeting arrangements were excellent and Emulex should have been thanked therefor, whether it was or not. The attendees introduced themselves and identified their function as driver designers, terminator designers, system integration engineers, or superior beings. Sign-in logs were circulated a number of times. Larry Lamers kept notes but cannot be blamed for these minutes. The final agenda was as follows: A. The driver in assertion. 1. Should the SCSI-3 standard still limit the current sourced by a termi- nator to 22.4 ma, or can the limit be raised to 24 ma? 2. Below 0.5 volts, can the current limit be replaced by a power limit, either 11.2 or 12 mw? If so, up to what absolute max current? Note: Obviously, the purpose of this item is either to validate the extra current sourced by the Trung Le terminator (and possibly others to come in the future) or to limit it. 3. Should the standard attempt to address driver fall times? B. The high current sourcing terminator. 1. Do we need any limit on the current which a terminator may draw from the TERMPWR line? 2. Is there a graceful way now that we could reclaim the line opposite the TERMPWR line and make it a second TERMPWR line (in the A cable)? Note: If extra current sourcing is applied to more than 2 or 3 lines in the bus, and specially if the S/E length limit is relaxed, a single AWG 28 conductor becomes increasingly unsatisfactory as the TERMPWR carrier. 4. Should the standard speak further to TERMPWR voltage levels than the the 4.25 v min at the TERMPWR source? Note: There is now available, from an unnamed company (initials TI), a 2.85 v regulator that regulates with TERMPWR voltage down to 3.45 v. But permitting quite low TERMPWR voltages may seriously impact the performance of the Trung Le terminator. 5. Should the standard speak to the benefits from stronger current in the region above 0.5 v? 6. Should the standard substitute the 187/267 ohn terminator for the 220/330 ohm terminator. C. The driver in active negation. 1. Should the standard speak to the minimum levels of active negation? 2. Is this minimum level compatible with the Boulay terminator? Ref: X3T9.2/91-130 and X3T9.2/91-151. 3. Should the standard speak to the maximum levels of active negation? 4. Are active negation drivers self-protecting when shorted to ground? Should they be? 5. Is this maximum level compatible with the Trung Le terminator? 6. Should the standard attempt to address driver rise times? D. Items re Fast S/E SCSI not already covered above from David Steele's proposal X3T9.2/91-64R1. ATTENDEES: Skip Strang Adaptec ..................... 408-957-4863 262-2533 Steve Ego Aeronics ..................... 512-258-2303 258-4392 Sassan Teymouri AMD ..................... 408-235-8616 235-7076 Jano Banks Apple ..................... 408-974-6417 Florin Oprescu Apple fao @netmanl.apple.com 408-974-2354 985-9893 August Hoecker Autologic ..................... 805-498-9611 499-1167 John Geldman Cirrus Logic johng@cirrus.com 415-623-8300 226-2170 Jim Komarde CIS/Emulex ..................... 415-556-7431 Bill Ham DEC ham@subsys.enet.dec.com 508-841-2629 841-2604 Tom Jones Emulex ..................... 714-668-5361 668-6654 Jean Kodama Emulex ..................... 714-668-6621 Rick Muething Emulex ..................... 510-248-0787 Paul Nitza Emulex p_nitza@emulex.com 216-236-8504 236-3660 Bill Spence ENDL/TI bill.spence@hub.dsg.ti.com 512-255-0339 255-0339 Denni Rash FileNet ..................... 714-966-3209 966-3288 Larry Grasso IBM ..................... 512-838-3672 838-3703 Trung Le IBM ..................... 512-823-5792 823-7544 Robert Allgood IOMEGA ..................... 801-778-3709 778-4667 Larry Lamers Maxtor 71540,2756 408-432-3889 Frank Samela Methode ..................... 708-867-9600 867-3149 Willard Tu Motorola ..................... 602-962-2657 962-2090 David Steele NCR ..................... 719-596-5795 597-8225 Brian Davis Quantum ..................... 408-894-5017 943-1735 David Browning Silicon Systems ..................... 714-731-7110 731-0974 Vit Novak Sun vit.novak@sun.com 415-336-2455 Rob Teigen Sun robert.teigen@sun.com 415-336-1830 Richard Mourn TI ..................... 214-997-3426 Dean Wallace TI ..................... 214-997-5973 997-5962 John Gaumont Unitrode ..................... 603-424-2410 424-3460 Tak Asami Western Digital ..................... 714-932-7621 932-7796 Nam Nguyen Western Digital ..................... 714-932-7115 932-7798 Erik Jessen Western Digital ..................... 714-932-6612 932-7798 RESULTS OF THE MEETING: Topic A--The Driver in Assertion: Re the somewhat trivial question of whether the standard should continue to limit the current sourced by a terminator to 22.4 ma (as opposed to 24 ma), there was a fair concensus that by the time SCSI-3 is widely available, concern about having a significant number of TTL input devices on a bus will no longer be warranted, and 22.4 ma (or 44.8 ma from two terminators) should disappear. Robert Allgood of IOMEGA repeated a good deal of his presentation from the St. Paul meeting, stressing two points: - With the vast majority of SCSI shielded cables in use, any terminator which respects the 22.4 (or 24) ma current limit provides inadequate current to provide safe margin in the first step of a negated signal. - The Trung Le terminator, as provided to the industry by Aeronics, corrects this inadequacy, but in so doing it stresses SCSI drivers above their spec limits. I presented my experimental results and reviewed my many previous inputs to X3T9.2 supporting the same viewpoints. Driver designers were asked to respond to the pressure for more current in the asserted driver. David Steele spoke for NCR, Erik Jessen spoke for Western Digital, Jean Kodama and Jim Komarde and Paul Nitza spoke for Emulex, and Sassan Teymouri spoke for AMD. All but Erik were united that they could never condone exceeding 48 ma per driver for any single-ended SCSI driver produced to date. Erik differed only in suggesting that it was possible that study of each individual product might show that some had sufficient margin that some current limit increase could be considered. Others present chimed in supporting the no increase position, including some user types concerned about preserving MTBF. On the other hand, most or all agreed that if the industry wants higher current limits in s/e drivers, they can be provided. The question of in which year was not addressed. Subsequently, Jean Kodama of Emulex has presented a set of arguments for not raising the driver current limit (attached). My proposal that the 48 ma at 0.5 v limit be replaced by a 24 mw limit below 0.5 v, with a max max current limit of perhaps 60 ma, was rejected, with respect to all existing devices, on several bases: - Higher currents open up the possibility of metal migration. - All calculations and testing of existing devices were for 48 ma, would have to be repeated, and obviously would fail in every case where the existing margins are at the lower limit of acceptability. - Existing test facilities are not readily adaptable to such limits. It was pointed out that the Methode SLIM terminator provides diode clamping on the -REQ and -ACK lines without exceeding the driver assertion current limit. Robert Allgood presented the viewpoint, with supporting waveforms, that a s/e driver asserting a typical SCSI shielded cable line must sink considerably more than 48 ma for the time interval until the first reflections are received from the ends of the bus--because the cable impedances are lower than the 100+ ohms necessary to limit to 48 ma from a line charged to 2.85 v. For a 6 m bus, the longest time until the first reflection is received from the farthest end is about 60 ns. There appeared to be concensus among the driver designers that this amount of overcurrent for this short a time was not a danger. Robert's point that there is a disharmony between s/e SCSI cables and s/e SCSI drivers was well buttressed. The final position of the meeting was that no concensus for a recommendation for a driver current limit increase for existing devices is possible. A clari- fication of the standard was suggested that the s/e driver current limit is 48 ma at any voltage and that the asserted voltage limit is 0.5 v max. It was pointed out, however, that even Boulay terminators would source more than 48 ma into any driver which pulled down to essentially 0 v. Probably some clarification of the meaning of the present current limit should and will eventually be offered. No remedy was offered for the fact that already thousands of installations depend on the Trung Le/Aeronics terminator for successful functioning. As of now, it violates the proper reading, although not the only reading, of the standard. It was foreseen that those now depending on that terminator will no doubt continue to do so. One line of reasoning is that there is no way that the chip people will ever agree to increasing the stress on their parts, and that it will be up to users to assess their risks and make their choice. Rise and fall times are treated under Topic C below. Topic B--The High Current Sourcing Terminator: No increase in the current sunk by drivers having been agreed upon, there was little incentive to tackle the question of TERMPWR current limits for terminators. There was general agreement, however, that the A cable TERMPWR situation is unsatisfactory, and that an attempt should be made to find a way to migrate the standard toward both center contacts in the A cable connectors being permitted to carry TERMPWR. It was pointed out that insulation-displacement connectors add as much as 2 ohms to total TERMPWR line resistance. It was pointed out that in some cases, existing power supply limits do not allow for increasing the TERMPWR current as required by high-current-sourcing terminators. No concensus emerged for a change in the standard's requirements on TERMPWR voltage. At one extreme, August Hoecker of Autologic reported using a separate p/s regulator to provide a TERMPWR supply such that after the drop through the series diode, the TERMPWR voltage remains close to the max limit of 5.25 volts. This is an approach to getting the line current up toward the max limit using conventional terminators. At the other extreme, people using the Boulay terminator with the new 0.6 v drop regulator are almost indifferent to TERMPWR voltage levels. Any level from 3.45 v up is okay. I presented my arguments for the benefits of increased current from the terminator in the region above 0.5 v, a-la the Trung Le terminator and a-la the ideal terminator characteristic presented in my X3T9.2/91-126. No disagreement was noted, but a lot of glazed eyes were. I also raised a question about the effect of such terminator characteristics in increasing driver stress. Robert Allgood questioned whether that driver characteristic has any effect on the driver stress. The glazed eyes started to close. No concensus emerged. Somewhere along here Robert Allgood and I got into a noisy argument as to the need to match to cable impedance to the terminator impedance. It became clear that the protagonists should step out into the alley and resolve their differ- ences out of the range of tender ears. I undertake an action item to study the 10-page paper which Robert made available and then contact him to try for a reconciliation of viewpoints before Albuquerque. I raised the question of substituting the 187/267 ohm terminator for the 220/330 ohm terminator for low-end applications. Robert Allgood preempted debate on the topic by insisting forcefully that both terminators, and by inference the Boulay and all other legal terminators, are inadequate. This topic still needs addressing, but the normal X3T9.2 working group is probably the proper forum. Driver design and terminator design specialists are not required. It may be indicated, however, to resurrect the cable working group to deal with this and other questions arising from the 10/10 conference. Topic C--The Driver in Negation: The very sharp difference between WD's and NCR's approaches to active negation drivers was clearly brought out. NCR believes an adequate driver can be designed which will respect the Boulay limits presented in my X3T9.2/91-151 and which will be self-protecting from a short to ground. WD believes in a design which does neither. The majority preferred, of course, that Boulay protection and self protection both be achieved, but they had no idea of whether this is compatible with adequate strength in negation. In this debate, Florin Oprescu of Apple exploded a large bomb. He predicted that if the far terminator is left off of a bus with active negation drivers that drive above about 2.85 v, the resulting double-voltage reflectiom will do permanent damage to the first device the reflection encounters. This degree of risk is not acceptable to Apple. The driver people were pretty unanimous that a 2.85 v upper limit is not feasible. Florin and his follow anarchist from Apple, Jano Banks, then concluded that a current limit exists for an active negation driver which will reduce the energy available in the reflected signal below the level which can cause permanent damage. Others were not sure that the effect could be as bad as Florin predicted, but there was no outright opposition to his stance. Jano accepted an action item to furnish a recommendation re his current limit to the Albuquerque working group. David Steele and Erik Jessen also agreed to furnish recommendations to Albuquerque as to what the standard may constructively say re active negation. Paul Nitza of Emulex suggested that they may have a contribution to make in this area also. All other contributions welcome. The proponent of strong active negation, Erik Jessen, stated flatly that there is no conflict between his active negation driver and the Trung Le/Aeronics terminator. The current which the terminators sink from the driver is within acceptable bounds. He stated that it is less than the 30 ma per terminator which I has postulated in my earlier notes. No concensus was reached on a spec for rise and fall times or slew rates, or even if the standard should attempt to address the issue. But there was unanimity that such limits are very important. Florin Oprescu pointed to a strong connection between rise time and a subsequent dip in the negation signal. And it was pointed out that driver design is involved even in the absence of active negation. The rate of negation is controlled by the rate at which the open collector driver turns off. Topic D--David Steele's Fast s/e SCSI Proposal: This topic got somewhat short shrift, inasmuch as many people were more interested in continuing previous discussions. David Steele pointed out that the objective underlying his proposal was to minimize other requirements by setting a short limit on bus length--3 m. In the face of some opinion that this limit is too short, he suggested that other bus characteric specs could be tightened in compensation for increasing the length. WINDUP It was agreed that at the Albuquerque working group meeting the further need for the activity of this group should be assessed. It was suggested that if a further meeting seemed indicated, the Wednesday of the San Diego plenary meeting week, 12/11, might be a good prospect. The meeting adjourned itself about 4:15 pm. P.S.: John Lohmeyer has scheduled a driver/terminator/cable session for Wednesday, 11-13, at the Albuquerque working group meeting, starting about mid-morning. The most important agenda item, in my opinion, for that session is what the standard should include regarding active- negation drivers. Action items from Irvine include recommendations on the subject from Apple, Emulex, NCR, Western Digital, and perhaps others. There are a number of other issues for the SCSI-3 physical standard which need discussion also, on some of which I will present proposals to join other proposals already in the hopper. ------------------------------------------------------------------------------- This is the message from Jean Kodama to Bill Spence on discussions had at the Driver Meeting here in Irvine, Ca. As a clarification to the last sentence in this meesage Jean meant she doesn't have a commitment from Emulex to dedicate the resources to this yet. PWH 10/16/91 Bill, I wanted to expand on my reluctance to increase the Iol current requirements for existing and new parts. I knew there were other points bothering me but I didn't think of them until later. As was mentioned in the meeting, existing normal and fast SCSI parts have been designed to the existing spec of 48 mA. Although most drivers can provide more current, it is because they have been overdesigned so that Iol under worst case conditions can be guaranteed. Furthermore, the existing spec was also used to determine the number of ground pins needed on a chip. Insufficient grounding leads to noise problems which are only aggravated by increasing the current that is switched. We can't ignore the existing chips, especially the more recent ones that have been designed for fast SCSI2. But even if we could ignore them, increasing Iol for SCSI3 means that drivers must now be overdesigned to the new spec. This poses new problems. First is the problem of cell size, which increases proportionally with the current requirement. This may lead to increased die size (cost), either by increasing the size of a pad-limited design or pushing a design over to being pad limited when it was previously core-limited. The problem is made that much worse when going to wide SCSI. Another problem is again the number of ground pins. Although it may not be as much of a problem on the host side, pin count (package size) is a real killer on the target side. On either side, the increased power dissipation requirements would be a major concern, again exacerbated by wide SCSI. Finally, there is the issue of active negation. Designing an active negation driver is tough enough without throwing in a monster pulldown, where switching transients may make Vdd an issue. Of course, extra Vdd pins would be dear on the target side. If SCSI3 must depart from SCSI2, it would seem wise to consider alternatives other than increasing the current requirements. Unfortunately, there aren't a lot of variables that can be tweaked. I can't speak for Emulex, but I'm hoping they will be able to devote resources to investigate other solutions in time for the upcoming committee meetings.