Minutes for IBIS Quality Committee meeting. May 14, 2002 11:00am EST Adam Tambone Adam.Tambone@fairchildsemi.com *Barry Katz bkatz@sisoft.com Benjamin P Silva benjamin.p.silva@intel.com *Bob Ross bob_ross@mentor.org *Eckhard Lenski Eckhard.Lenski@icn.siemens.de *Eric Brock ebrock@sisoft.com *Gregory R Edlund gedlund@us.ibm.com Hazem Hegazy hazem_hegazy@mentorg.com John Figueroa jfigueroa@apple.com *Kevin Fisher kfisher@sisoft.com Kim Helliwell kimgh@apple.com *Lynne Green lgreen@cadence.com Mike Labonte mlabonte@hhnetwrk.com Peter LaFlamme plaflamm@amcc.com *Robert Haller rhaller@cereva.com Roy Leventhal Roy_Leventhal@3com.com *Sherif Hammad sherif_hammad@mentorg.com Todd Westerhoff twester@hhnetwk.com Tom Dagostino tom_dagostino@mentorg.com Everyone in attendance marked by * Minutes: -Roll call -Review of Action Items Siemens is still working on spec. Should be ready by next meeting. Kim Helliwell is working on proposal for open source of golden parser. A draft will be sent out this week for review. Roy to send out pointer to Accuracy documents. -Discussion of Greg Edlund's email about a figure of merit. Greg provided clarification. Figure of merit would apply to accuracy section of checklist -- not to the IBIS model as a whole. Warrants further discussion when we get to the HSPICE correlation section of the checklist. -Review of VT/Ramp section of checklist. Current list of items in the checklist reviewed. Ramp checks include: Are dV 20%-80%? (IBIS committee specified that dT is the only value of interest. dV included as informational/verification only.) Check for min/typ/max swizzle. No negative values for Ramp. No Zero values for dV or dT. VT curves do not have to be generated under same load conditions as Ramp data however, when they are they should be consistent and crosschecked. Eckhard noted when VT curves have been created in-correctly, and the Ramp data was derived from the VT curves, then both the VT curves and ramp data are useless. Discussion on whether we should recommend a termination for people to use when generating VT curves led to discussion on portability vs. accuracy. Lots of issues here. A general consensus is that high-quality models will be portable across multiple simulators -- even at the expense of some accuracy. For portability across simulators 50ohm to VCC/GND RISING/Falling waveforms in general seems to be the best. However, exceptions were pointed out (ie differential LVDS ). Miscompares between IV and VT curves should be documented as to why. Currently the IBIS Golden parser generates a warning between 2% and 10% miscompare, and error at >10%. V4.0 of IBIS will allow up to 1000 V-T points potentially reducing the amount of discrepancies. (Currently constrained to 50 points.) Visual inspection of VT curves:. Are DC states achieved? Is curve Monotonic? This is expected, so if it's not, then it should be noted. Note: Improper Fixture values can produce non-monotonic waveforms. This may be OK for documentation purposes (ie lab measurements), but not for driving a simulator. For portability, leading/trailing delays from curves should be removed. This may be at the expense of accuracy in some simulators, but with the benefit of providing consistency across all simulators related to potential frequency limitations of a model. Is removal of the delays consistent across rising and falling waveforms. This a complicated issues that will require further discussion and may alter or conflict with preceding paragraph. (Can also cause simulator dependencies.) Want to avoid multiple sets of VT curves. -All errors/warnings need to be explained in checklist. Bob pointed out that some warnings are put in place to flag certain values that may or may not be of concern to any particular user. These explanations would go in the notes section associated with warnings and errors in the checklist. -Next meeting will be in two weeks, May 28, 2002. 11:00am EST. Bridge Info Call: 408-974-8478, access CONFERENCE SERVER 2, enter Conference ID: 4203 or Call toll free: 800-249-5618