X3T9.2/88-144 LASER MAGNETIC STORAGE MEMO INTERNATIONAL _________________________________________________________________________ Optical Storage Division 1 November 1988 To: SCSI Standards Committee X3T9.2 From: Paul Boulay (719) 593-4323 Subject: Single-Ended Termination Alternatives I've looked at the proposal for an alternative terminator from C. Mollard of BULL systems. (Thanks for the fax, John.) I see some serious problems with the design as stated. The BULL systems scheme is: (Pardon my character graphics.) +-(Jumper)-+ | | 1N5818 | 1N4005 | + 5V ---------->|---+-------+---+----|<----+------ TermPwr | | Cap R 180 Ohm | R Gnd | +---------- Typical Signal Line Note: Add the jumper to supply | TermPwr to the bus, remove R 390 Ohm otherwise. R | Gnd First the advantages: It lowers the characteristic impediance to 123 ohms. Use of the 180/390 terminator has the potential of raising the setpoint (or, the negated signal level) of an open-collector driven signal. Now the problems: The worst problem with the BULL systems proposal is that requires the bus drivers to sink too much current. Either a slightly over voltage +5V supply or slightly lower than nominal resistance in the 180 ohm resistors requires that the driver sink in excess of 48 milli Amps. (Jim Schuessler and I did independent back-of-the-envelope calculations and both came up with 53+ mills.) The practical implications of this are that the SCSI bus driver chips will run hotter than they should and that the low voltage asserted onto the bus will be higher than the standard calls for. An additional gotcha related to excess driver current -- when the driver succeeds in sinking the current available and all lines are driven the total termination current requirement is 1.0 A. (vs. .82 w/ 220/330.) I'll certainly grant that all lines driven is impossible in practice, however, I can build a possible case for 15 of 18. With the selection of diodes as indicated, the setpoint in the BULL systems scheme varies widely (2.35V to 3.35V -- remote power w/ +5V@-5% to local power w/ +5v@+5%. The too high driver sink requirement occurs with this second case). The low end of this range doesn't allow devices to meet the minimum signal negation voltage spec on an open- collector signal. This low voltage results from starting with +4.75V, less two diode forward drops one of which is a plain 1N4005. An aside about diode specifications ( @ 25 degrees C.): Type I(fwd) V(rev) V(fwd) V(fwd) V(fwd) V(fwd) max. work. typ. max. typ. max. Silicon rept. rept. I=.1A I=.1A I=1A I=1A 1N4001 1 A. 50 V -- -- .93 V 1.1 V 1N4005 1 A. 600 V -- -- .93 V 1.1 V Schottky 1N5818 1 A. 20 V .28 V .33 V .42 V .55 V 1N5821 3 A. 30 V .23 V -- .34 V .38 V "Budgetary" prices @ 1000 pieces -- 1N4001 = $0.06; 1N5818 = $0.25; 1N5821 = $0.35. My alternative that fixes these problems: 1N5818 or 1N5821* (a) + 5V ----------(Fuse)------->|--------+------ TermPwr | 1N5818 (b) | +-------+---|<----+ | | Cap R 180 Ohm | R Gnd | +---------- Typical Signal Line Note: Add the fuse to supply | TermPwr to the bus, remove R 390 Ohm otherwise. Devices that opt R to never supply TermPwr may | omit diode (a) and provision Gnd for the fuse. (* = recommended.) By rearranging things and adding the second schottky diode (b) this corrects the problems cited above. I hasten to add that it retains only one of the advantages, but it doesn't make things worse. The characteristic terminatiom impediance of 123 ohms is retained. The setpoint remains essentially the same independent of the location of the source of TermPwr. Given a sckottky diode supplying TermPwr and a 220/330 ohm divider, this brings up the setpoint by 0.15 V. Compared to a silicon diode, the gain is .5 V. With this improved scheme, the driver current sink requirements stay (barely) within 48 mA. Also the max TermPwr current requirement is close to the current value. All this brings me to the conclusion that with clever design we can improve the situation a little. However, the constraints don't give us much lattitude. I wonder if the small gains we can achieve are worth it. Perhaps we can change the words in the standard to define the requirements for a prospective termination circuit rather than calling out 220/330 specifically. The 220/330 values could go in an implementors note. Maybe this scheme could go in an appendix.