====================================================================== IBIS FUTURES/COOKBOOK SUBCOMMITTEE MEETING MINUTES Date: September 1, 2005 Attendees: ---------- Cadence Design Systems - Shangli Wu, Lance Wang Cisco - Zhiping Yang, Syed Huq, Vinu Arumugham Intel - Michael Mirmak, Arpad Muranyi Mentor Graphics - John Angulo, Ian Dodd NCSU - Ambrish Varma Sigrity - Sam Chitwood Teraspeed - Bob Ross ====================================================================== Next Meeting: Thursday, September 8, 2005 9 AM - 11 AM US Pacific Time Telephone Bridge Passcode 916-356-2663 4 114-9206 Agenda: 9 - 9:05 AM Opens 9:05 - 9:10 AM Cookbook Checkpoint 9:10 - 9:45 AM BIRD100 Presentation 9:45 - 11:00 AM BIRD95 Analysis - C_comp Splitting Concerns from Lance, Shangli - Simple test cases for ground parasitics from Sam, Michael and Zhiping ======================================================================== COOKBOOK --------- Michael Mirmak called for comments on the latest text. Arpad Muranyi questioned why we need to mention extrapolation when the clamp currents are assumed zero for non-ODT cases. Bob Ross noted that the non-ODT treatment implies that spec-compliant models are less than optimal. Michael, Arpad and Bob will meet offline to resolve the final issues with the cookbook and post Draft 7 for a Sept. 16 approval vote. FUTURES --------- Zhiping Yang presented his analyses of the Micron model data set and also his suggestions on ground parasitics. Bob Ross asked how the C_comp split was determined. Zhiping commented that the Micron IBIS model iss different from Cisco's splitting results; Cisco did not use Arpad's C_comp splitting method. Lance inquired about the stimulus pattern. Zhiping confirmed 100 ps was used, same as in the Micron case. Much of the meeting was devoted to discussion of proper implementation of extraction and presentation of results. Lance commented that using the average I-t currents in the tables is a problem. For example, say the 50 ohm case is used for extraction -- the output current for 50 ohm test loads should be an exact match. If you use an average, an exact match is impossible. Bob stated that he is most interested in the lower-level process and would like to see specifics on how the data was extracted. Arpad surmised that Cisco didn't use split C_comp. In one of the first few presentations, C_comp wasn't split. Now it's split 50/50. Zhiping responded that Cisco had not verified whether 50/50 was accurate in the earlier data. Now, the split proportion has been reanalyzed. Syed commented that, overall, Cisco is satisfied with the great correlation seen in the results, even with multiple buffers switching. Quiet line cases were also analyzed. Lance commented that the treatment of C_comp and this specific case is questionable; if a different frequency of signal is running through the network, will the C_comp change? Zhiping presented on removing ground parasitics in ground pin models. He noted two areas specifically where ground pin parasitics are problematic: 1) They are not needed for high-speed, where they cause more issues than they solve. 2) IBIS is now a three-terminal device, and cannot correlate with transistor-level when the transistor level model has more than 3 terminals. For high-frequency, Zhiping stated that we only care about the local reference, NOT about cross-PCB or cross-die analysis. We can simplify our networks to remove separate references. Sam commented that the analyses are showing only half the system; the power network itself is not shown. Time ran out for discussion of other topics.