X3T9.2/92-100 Date: April 28, 1992 To: X3T9.2 Membership From: Lawrence J. Lamers, X3T9.2 Secretary Bill Spence, Chair SPI Working Group Subject: April 28, 1992 SPI Working Group Meeting Minutes 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. Opening Remarks 2. Old Business 2.1 Latest Update on Skew Budget [Chan] 2.2 Glitch Filter Proposal [Geldman] 2.3 Wording re Glitch Immunity During Power Cycle [Ham] 3. New Business 3.1 Possible Considerations in 3.3-volt Operation [Spence] 4. Agenda for Next Meeting 5. Meeting Schedule The following people attended the meeting: Name Status Organization Phone Number ------------------------- ------ ------------------------------ -------------- Mr. Norm Harris P Adaptec, Inc. (408) 945-8600 x2230 Mr. Bob Whiteman A AMP, Inc. (717) 780-7481 Mr. Russell D. Moser V AMP, Inc. (717) 986-5022 Mr. Jeff Rosa P Amphenol Interconnect (607) 786-4222 Mr. Michael Wingard A Amphenol Interconnect (607) 786-4241 Mr. Richard M. Ross O Amphenol/Spectra-strip (203) 281-3200 Mr. Eric Lawrence O Brand-Rex Company (203) 456-8000 Mr. Christian Mollard O BULL Systems (33)130807128 Mr. Clifford E. Strang O BusTek (408) 492-9090 Jr. Mr. Kurt Witte P Ciprico Inc. (612) 559-2034 Mr. Michael Smith V Dallas Semiconductor (214) 450-0457 Mr. Skip Jones A Emulex Corp. (714) 668-5058 Mr. Steve Caron P Furukawa Electric America (404) 487-1234 Mr. Kurt Chan A Hewlett Packard Co. (916) 785-5621 Mr. Russell Smith A Hewlett Packard Co. (508) 256-660 Mr. Denis Chaves O Hewlett Packard Co. (508) 436-5355 Mr. David McFadden P Honda Connector (708) 913-9566 Mr. Tom Kulesza O Honda Connector (708) 913-9566 Mr. George Penokie P IBM Corp. (507) 253-5208 Mr. Otis Greene O IBM Corp. (914) 892-5597 Mr. Larry Grasso S IBM Corp. (512) 838-3672 Ms. Gricell Co V IBM Corp. (512) 838-3664 Mr. Geoff Barton P Iomega Corp. (801) 778-1000 Mr. Robert Bellino P Madison Cable Corp. (508) 752-7320 Ms. Donna Pope P Maxoptix Corp. (408) 954-9700 x315 Mr. Lawrence J. Lamers P Maxtor Corp. (408) 432-3889 Mr. Raymond C. Yule P Micropolis Corp. (818) 718-7803 Mr. Ken Erickson A Micropolis Corp. (818) 718-5113 Mr. John Lohmeyer P NCR Corp. (719) 596-5795 x362 Mr. David Steele S NCR Corp. (719) 596-5795 Mr. Stephen F. Heil P Panasonic Technologies, Inc (201) 348-7064 Mr. Gene Milligan A Seagate Technology (405) 324-3070 Mr. Mike Holt O Silicon Systems (714) 573-6871 Mr. Robert L. Simpson P Sony Corp. of America (408) 944-4348 Mr. Gregory Green O Stratus Computer (508) 490-6473 Mr. Robert N. Snively P Sun Microsystems, Inc. (415) 336-5332 Mr. Richard Mourn O Texas Instruments (214) 997-3426 Mr. Richard J. Krass P The JPM Company (404) 631-8888 Mr. Leonard Dore A The JPM Company (717) 524-8200 Mr. Harvey Waltersdorf P Thomas & Betts (803) 676-2905 Mr. Peter Dougherty P UNISYS (714) 380-6270 Mr. Mark Jordan S Unitrode Integrated Circuits (603) 429-8628 Mr. Tak Asami O Western Digital (714) 932-7621 Mr. Doug Piper P Woven Electronics (803) 967-1751 Mr. Steve Drake P Yamaichi Electronics, Inc. (408) 452-0797 45 People Present Status Key: P Principal A Alternate O Observer S Special Interest (frequent visitor) V Visitor Results of Meeting 1. Opening Remarks Bill Spence, the Chair, called the meeting to order at 9:00 a.m., Tuesday April 28, 1992. He thanked Chuck Brill of AMP for hosting the meeting. As is customary, the people attending introduced themselves. A copy of the attendance list was circulated for attendance and corrections. 2. Old Business 2.1 Latest Update on Skew Budget [Chan] Primarily, what was at issue was 9 numbers in Table 12, p. 33 of the SPI Working Draft and p. 89 of the April mailing. Behind these numbers is an Informative Presentation on Setup and Hold Timing in Annex C, p. 55 of the SPI Working Draft and p. 111 of the April mailing. The following table presents these numbers at the beginning and at the end of the discussion at the meeting. All these numbers are considered to be still subject to review by affected parties, and the ones marked with ? have been specially identified as being tentative. Timing Description Before After Assertion Period 90 ns 80 ns Negation Period 90 ns 80 ns Receive Hold Time 31 ns 22 ns Receive Setup Time 21 ns 12 ns Receiver Delay Skew (recommended) 6 ns 6 ns ? Transmit Hold Time 35 ns 35 ns Transmit Setup Time 25 ns 25 ns Transmit Delay Skew (recommended) 6 ns 6 ns ? Asymmetry, req/ack * ----- 6 ns ? * [ Par. 7.21 ] |tPHL - tPLH|req/ack <= 6ns time of prop delay high to low minus time of prop delay low to high of the req/ack signal should be less than 6ns. Underlying the above decisions was a great deal of input, some of which is detailed below, as follows: Proposed Skew Budget ------------------------------ TX Chip 32 setup/43 hold trace 1ns driver 6ns (recommended) TX connector cable 4ns RX connector distortion 4ns driver 6ns trace 1ns RX Chip 10 setup/20 hold Bill Ham raised a question of data dependent cross talk. Kurt maintained that this is accounted for in the 4ns. Rise time degradation effects are also in the 4ns. Gene Milligan objected to the softness of having recommendations. Bill Spence asked how many folks would be impacted if these were requirements. David Steele stated recommendations are fine, since it gives him opportunities to fine tune the silicon. Tak Asami stated that flexibility is needed to meet the variety of implementations and environments. John Lohmeyer pointed out that there is very little margin in the 4 ns cable skew delay number. Should we care so much about the absolute propagation delay? The only numbers that matter are the maximum cable propagation time (for arbitration) and the delta prop delay between signals. David Steele wanted about 5ns of slop, taken out of the receiver side, and spec'd as a separate value. This could be used for cable mismatches or system tolerance. A waveform allowance, jitter allowance. The silicon folks believed they could live with 5 ns setup and 10 ns hold on the RX chip. It was also suggested that the 4ns distortion time should be moved to the other side of the connector. The result of these changes was as follows: Revised Skew Budget ------------------------------ TX Chip 32 setup/43 hold trace 1ns driver 6ns (recommended) TX connector cable 4ns distortion 4ns tolerance 5ns RX connector driver 6ns trace 1ns RX Chip 5 setup/15 hold David Steele noted that in slow mode the assertion period and negation period are more realistically 80 ns. David won this one without a fight. CABLE PROPAGATION DELAY. It was pointed out that the table at the top of p. 13 of the SPI Working Draft has mathematical mistakes. Further discussion led to a demand for further changes, as follows. Delete reference to a minimum propagation delay (as an information resource, this figure could be changed to the propagation delay of light in vacuum, 3.34 ns/m). The maximum should be increased, to perhaps 7 ns/m, or eliminated completely. The skew number is correct at .15 ns / meter. In any case, it was decided that the following limits need to be established: Maximum end-to-end propagation delay for the bus: 200 ns Maximum propagation delay skew for the bus: 4 ns Inserting these limits could eliminate the need for the 25 m maximum overall bus length. But it was pointed out that the need remains for provisions about maximum overall bus lengths for single-ended operation. The conversation vectored off into consideration of using A cables and whether or not the A cable should be specified in SCSI-3 Parallel Interface document. Some members expressed surprise that it was proposed that the SPI document should standardize only the P cables. It was explained that the great complexity associated with the A cable, with its numerous connectors, was adequately documented in the SCSI-2 Standard, which will not be made obsolete by the SPI Standard. John Lohmeyer pointed out that wording in the SPI document is intended to make its electrical characteristics applicable to the A cable. Perhaps this wording should be clarified. 2.1.1 Active Negation Driver Strength Norm Harris raised questions on section 6.1.1: the 7ma figure at VOH. If the driver is negated and loaded to 7ma, the terminal voltage is to be between 2.5 and 3.24 Vdc--that is the intended reading. In the discussion, it was realized that the 24 ma line should not have a minimum voltage specified, only Florin Oprescu's limit of 3.0 Vdc. David Steele argued that there had been no minimums in the original discussion leading to these specifications, and he strongly questioned the need for a minimum. David did not want to have to supply 2.5 volts at 7ma, only to not to exceed 3.24 volts at 7ma. If there were to be a minimum voltage at 7 ma, he wanted it to be 2.0 Vdc. One reason to have a minimum is that it provides a definition of what active negation is. A straw poll indicated that 8 folks favored a 2.0 Vdc min, 5 folks favored no min. John Lohmeyer advocated elevating this issue to a plenary vote for resolution. 2.2 Glitch Filter Proposal [Geldman] John Geldman was not able to attend. Bill Spence summarized the current status on glitch filtering, which breaks down into simple R-C filtering or more complex, probably time-domain, filtering. One problem: an R-C filter with a corner frequency high enough to provide negligible attenuation of a 10 MHz signal will provide little protection against noise glitches. Should we have a glitch filter in the SPI document? Or a time window in which a transition is ignored. David Steele spoke of rejecting a 10 ns spike. This would require a 3ns time constant for the RC filter. The skew effects of using such an RC filter can cause problems. Thus some form of time-domain filtering seems to be necessary. These fall into two types: glitch filtering per se and transition suppression. At previous sessions, Sassan Teymouri has spoken in favor of true glitch filtering. Florin Oprescu gave us an input filter spec which requires transition suppression for about 40 ns after a transition. This was for use with transfer rates up to 2.5 Mtransfers/sec. Bill Spence pointed out that reducing Florin's specified suppression period to 10 ns, while it obviously reduces the filter's effectiveness, does maintain it as effective as possible while permitting 10 Mtransfers/sec--i.e., reduce Florin's time scales by a factor of 4. David Steele said that from all the trouble reports he has received, it seems we should focus on suppressing double clocking on the negation edge. He proposed adding the following words to the input characteristics section: "From a valid negation edge (crossing 1.4 Vdc), for a period of 10ns, a receiver should not respond to any transitions on the bus." This raised questions about devices which might transition on a signal which does not reach 1.4 volts. Bill Spence said that this obscures the point and that the reference to 1.4 Vdc should be deleted. From either a test or a design standpoint, the point is that after a negation transition is forced, the receiver will not register any other transition for 10 ns. From his hot plugging perspective, Bill Ham was still interested in suppression of a 10 ns or narrower glitch of magnitude 700 mv or less, which is not necessarily associated with a valid transition. But on further analysis, he stated that with the current wording, hot plugging glitches should not be seen. The hot plugging assumption is that the SCSI bus is not quiesced during the hot plugging operation. 2.3 Wording re Glitch Immunity During Power Cycle [Bill Ham] The point is that modern SCSI protocol chips not glitch the SCSI bus. All single-ended drivers (both passive and active) shall maintain a high impedance state during power up and power down cycles until enabled. Bill proposed specific wording to that effect for 6.1.1 and 6.2.1. 3. New Business 3.1 Possible Considerations in 3.3-volt Operation [Spence] Bill Spence presented his paper exploring the issues surrounding 3.3 volt operation. He reported a demand that 3.3 v devices operate with battery supply voltages ranging from 3.6 v down to 2.7 v. The main consideration is terminator power. There are two environments: one with a 3.3 volt only and one with a 3.3 volt initiator connected to a target that has AC power. The later case is easily solved. In the first case, in order to minimize terminator power, the terminator needs to be modified. The series regulator should be removed or modified. John Lohmeyer suggested that 48 ma drivers be retained. There are other system issues, such as going external, length of bus, power management and power consumption that need to be addressed. The question of having changed the hard limits on total bus length in single- ended SCSI to merely strong recommendations was brought up again. Bill Spence recalled the process by which that compromise had been arrived at and objected to the SPI group spending more time on it. If there are objections , they should be brought up at the June plenary, when the draft document is next up for approval, or at least in the X3T9.2 full working group. 3.2 Pin Capacitance of an SCSI Controller LSI [Asami] Tak Asami presented a paper supporting the return to 25pf max node capacitance. 4. Agenda for Next Meeting Silicon issues will be dealt with if raised at the May meeting, but the plan is to focus on finalizing cable and terminator issues. Parties interested therein are urged to be present! The following list of items to be discussed owes much to Bill Ham. 1) One silicon/system issue: which lines should have active negation? 2) A-Cable in SPI ? 3) Final trimming of impedance figures. 4) Minimum bend radius. Near connector. Away from connector. 5) Length and magnitude of Zo deviations. 6) Zo distribution throughout a cable. 7) Single-ended vs differential Zo and crosstalk in a single P cable. 5. Meeting Schedule The SPI Working Group meeting for May is on Tuesday, May 19, 1992 at 1:00 pm at the Holiday Inn in Harrisburg, PA. The meeting adjourned at 4:30 p.m.