X3T9.2/91-30 To: John Lohmeyer and X3T9.2 11 Mar 1991 From: Bill Spence Subject: Cabling Aritcle Submission Following is the Cover Memo and Final Draft which were Faxed to Electronic Design of Friday, 3/8. ------------------------------------------------------------------------------- To: Dave Bursky, ELECTRONIC DESIGN 8 Mar 1991 Copies: John Lohmeyer and All X3T9.2 From: Bill Spence, TI Austin Subject: Cabling Article, Forwarding Cover Memo Dave, as we discussed by phone recently, several members of X3T9.2 feel that the breakthroughs in comprehension of the requirements for single-ended SCSI cabling which the past year has brought should be publicized far more quickly than we will get a new standard published. The leadership concurred and suggested that Kurt Chan and I might collaborate on something like a follow-up article to the Boulay, Chan, Schuessler article of last year. You pointed out, however, that the time-to-publication of that type article is about 6 months, as opposed to about 6 weeks for a news-type article input by the editorial staff. We all agreed that the latter was the way to go, and I undertook to put together an input to you on which you could base such an article. It's gone through 3 versions, being modified to represent a viewpoint on which all interested parties can agree. What I am forwarding to you has been endorsed by the fol- lowing list of people, which includes most everyone who has been active in the X3T9.2 cable deliberations over the past several months: Dal Allan of ENDL Marge Bacis of Montrose Bob Bellino of Madison Peter Blackford of Astro Kurt Chan of Hewlett-Packard Jim Fiala of 3M Chuck Grant of Madison Dave Hess of Berk-Tek Larry Lamers of Maxtor John Lohmeyer of NCR Tom Marks of NEK Vit Novak of Sun Bob Snively of Sun Deni Springer of 3M David Steele of NCR Wills Xu of C & M In addition, all deliberations have been publicly displayed on the SCSI Bulletin Board maintained by NCR. There has been no dissent that this final form represents a concensus of the task group's understanding. The following individual were kept apprised of this activity but has made no response: Fred Hengelhaupt of JPM. The leadership has asked me to submit three requests to you: 1. To have reprint permission to the final article. 2. To have permission to post the final article to the SCSI BB, as Computer Technology Review permitted with their SCSI-2 article by Dal Allan. 3. To be sympathetic to the difficulty we seem always to have in expressing our technical concepts so unambiguously that all readers pick up the intended meaning. After you translate what I am forwarding you into English, we would dearly appreciate a chance to review what you made of my input, just in case I managed to get the wrong meaning of something in the forefront. It also is implicit that we request that if you elect to use this material, you do so quickly. If this cannot be, we would appreciate the opportunity to seek another outlet. ------------------------------------------------------------------------------- FINAL DRAFT -- 3/08/91 -- Forwarded by Bill Spence at 512-250-6627, Fax: 512-250-7479, or E Mail: spence@csdeng.dsg.ti.com SINGLE-ENDED SCSI CABLING UPDATE ILLUSTRATION: The Boulay terminator schematic, 110-ohm version only. (Figure 4-9 of Rev 10c with Note 2 deleted) CAPTION: Tight on-board regulation permits reduced impedance and improved performance in this new terminator circuit. More than just an interconnection scheme, the SCSI bus is a high-speed signal transmission system. Facing up to the need to better specify requirements for successful transmission, several members of the X3T9.2 Task Group on the SCSI Standard have in recent months input the results of several theoretical and experimental studies. The resulting new understanding has changed several existing conceptions and clarified others. It provides important guides to more reliable, longer, higher speed single-ended SCSI implementations. Major points: - While differential SCSI remains the premier option, proper implementation of a single-ended (unbalanced) system permits enjoying the resulting savings in power, space, and cost while easing considerably the present performance uncertainties and restrictions as to speed, length, and loading. - The key enhancements are a proper relation of cable to terminator impedance, low cable attenuation, proper isolation of clock lines from data and parity lines within the cable, and improved drivers with active signal negation. It continues to be essential to limit stub lengths and receiver loading on the bus. - Cable impedance is much lower in single-ended mode than differential and must be specified appropriately to its intended use. - A "standard" design of shielded, twisted-pair cable has emerged which meets most needs of both single-ended and differential SCSI and which is practical in assembly, reasonably economical, and widely available from quality suppliers. There is strong assymetry in requirements on the received signals in single-ended SCSI, and it turns out that they are best served where the single- ended impedance of the cable is less than the output impedance of the terminator by about 25%. The original 220/330-ohm resistor-network terminators, with an output impedance of 132 ohms, work best with cables whose single-ended impedance is in the range of 100 ohms. The only such cable widely used is 0.050-inch pitch AWG 28 PVC ribbon cable. But busses made up mainly of such cable are typically so short as not to require optimized terminator impedance matching. For longer buses, there are shielded twisted-pair cables available with impedances of up to 90 ohms using foamed dielectrics (100 ohms if AWG 30 conductors are used). Most of these have excellent characteristics where their impedance is desired. They are not as widely acceptable in assembly operations, however, as solid dielectric cables. The preferred solution is to take advantage of the lower impedance and better control of open-circuit voltage of the 110-ohm regulated terminator (see illustration) being introduced in SCSI-2 and now in full commercial production. It is ideally matched to the so-called "80-ohm SCSI standard" shielded cable stock and to the 0.025-inch pitch AWG 30 PVC ribbon cable stock preferred for the new high-density SCSI connectors. And, unlike the original 220/330 ohm terminator, its performance does not change with TERMPWR voltage. The 110-ohm terminator works well with the 90-ohm cables also, and a great deal of success has been enjoyed with that combination. If it is desired to improve the accomodation between the 110-ohm terminator and 0.050-inch pitch ribbon cable, AWG 26 conductors may be specified. The "80-ohm SCSI standard" cable referred to incorporates 25 twisted pairs of AWG 28 conductors, 7/36 stranded, with solid polyolefin (polypropylene or poly- ethylene) insulation to 0.033-0.035-inch diameter. The cable bundle is wrapped in a polyolefin-based tape buffer layer providing low capacitance isolation of the outer pairs from the overall shield layers. (The outer pairs exhibit lower impedance and greater propagation delay then the inner pairs.) Examples of such cables are Astro 52-107-C, Berk-Tek 271212, C & M C801/25, Hitachi 8213, Madison 4099 and 4179, Montrose 7251, and NEK J0517-1. They have single-ended impedances of about 80 ohms and differential impedances of about 120 ohms (which is ideal), and their attenuations at SCSI signal frequencies are acceptably low. Note that these levels of performance are jeapordized unless the manufacturing quality is high. For instance, the twist pitch of the twisted pairs in a given layer must be very uniform if propagation delay skew is to be minimized. The greatest hazard found to single-ended SCSI is double clocking of the -REQ or -ACK signals due to noise coupled from data line switching. The recommended solution is to provide crosstalk isolation within the cable by placing the -REQ and -ACK signal pairs in the core of the cable and the data and parity pairs in the outer layer. This allows for a buffer layer of pairs inactive during data transfer in between the data and parity signals and the -REQ and -ACK signals. But it requires a level of assembly control which has not been widespread in the industry up to now. It also requires in the new "P" cable an 18-pair outer layer. (X3T9.2 has pretty will firmed up the requirements for the P cable, which provides single-cable 16-bit operation.) Good operation of fully loaded single-ended SCSI buses at lengths well above 6 meters has been demonstrated in both asynchronous and synchronous operation by implementing according to the guidelines laid out above. (X3T9.2 has not made any decision yet, however, about increasing the 6 meter limit on single-ended SCSI in the standard.) Testing is getting under way to establish requirements in "fast" SCSI, to 10 megatransfers per second. Further performance improvements are foreseen as new SCSI protocol chips become available with active-negation drivers (as opposed to the presently prevalent release-on-negation drivers). And controlled ramp rate in drivers also has improved performance by reducing overshoot and ringing.