Minutes, IBIS Quality Committee

11 Apr 2006

11-12 AM EST (8-9 AM PST)
Phone: 1.877.384.0543 or 1.800.743.7560
Meeting access: Meeting ID: 90437837

ROLL CALL
Adam Tambone
Barry Katz
Benny Lazer
Benjamin P Silva
*Bob Ross
Brian Arsenault
*Eckhard Lenski
Eric Brock
Gregory R Edlund
Hazem Hegazy
John Figueroa
John Angulo
Katja Koller
Kevin Fisher
Kim Helliwell
Lance Wang
Lynne Green
*Mike LaBonte
*Moshiul Haque
Peter LaFlamme
Robert Haller
*Roy Leventhal
Sherif Hammad
Todd Westerhoff
Tom Dagostino
Kazuyoshi Shoji
Sadahiro Nonoyama

Everyone in attendance marked by  *

NOTE: "AR" = Action Required.

-----------------------MINUTES ---------------------------
Mike LaBonte conducted the meeting.

Review of minutes for 28 Mar 2006:

- Minutes approved.

Announcements - EIA Call for Patent Disclosure: 

- No company declared a patent.

AR Review:

- AR: Mike send parser bug report to ibis-bug@eda.org
  Done, PBUG94. IBIS 

- AR: Mike clone macromodel web page for IQ
  TBD

- AR: Bob contact Moshiul Haque about the next meeting
  Done

Opens for new items:

- Mike LaBonte announced that a different phone bridge
  service may be used for future teleconferences, one that
  is integrated with a computer desktop sharing feature.

Second review of parser bug 94 for IQ 2.2 check:

- The IBIS committee has assigned number 94 to our ibischk
  bug report for IQ 2.2, but has not yet classified it.

- Arpad suggested that the combined I/V curve warning could
  be enhanced to print the voltage(s) at which the non-monotonic
  regions appear. So the bug is returned to this group.
  We agreed with Arpad's idea.

- Testing shows that when there is non-monotonicity in the
  combined I/V curves of an [IBIS ver] 4+ file, ibischk4 
  already prints both the individual curve warning as
  well as the combined curve warning.

- We don't any want warnings when the combined I/V is OK, so
  the single curve warning should be suppressed in this case.

- Mike has source code for the parser, hasn't looked into it yet.

AR: Mike rewrite bug and submit.

Continued discussion of IQ level designations:

- Recap of last week:
  - Existing system has only 2 "reasonable & correct" levels, 0 and 1
  - Level 0 sounds like "no quality", and the checks are somewhat trivial.
  - Level 1 consists of all other checks, and is difficult to achieve.
  - Other levels are inconsistent and confusing
    - 2a and 2b based on Greg Edlund's accuracy work.
    - 3 is the combination of 2a and 2b.
  - To fix this we could switch to a new system using letters

- This is a major change.
  - How much of a legacy does the existing system have?
  - Micron has been using IQ
  - Roy Leventhal's book explains IQ.

- Possible top level designations, discussed at the last meeting:
  (I)bischk passes
  (Q)quality checklist
  (M)easurement - test against measurements
  (S)simulation - test against SPICE run
  e(x)ception - There are exceptions/waivers to some checks

- We could break (Q)uality into 4 areas:
  (S)ignal integrity
  (T)iming
  (P)ower delivery
  (E)MC
  (F)ull checks including pulse immunity?

- Bob Ross proposed that (M)easurement and (S)imulation could become
  (P)hysical and (S)oftware.
  - This leaves us with 2 (S)s.

- Is this new proposal too detailed?
  - The designations might be handy as an indicator of what the
    file contains.
  - Another idea is to go back to having 2 levels:
    - Checks that every IBIS file should pass
    - More obscure items
  - There are DDR-specific parameters in IBIS. Do these need a letter?

- Moshiul asked if individual checks have number/letter designations.
  - This question was not addressed directly.

- IQ goals, to help us evaluate the new ideas:
  - 1) Users can quickly understand how useful an IBIS file is.
  - 2) Vendors can make a model that passes non-trivial checks.

- To evaluate the new proposals we need to go through the existing checks
  and categorize them.

AR: Mike will send the IQ spec document and checklist to Roy.
AR: Roy will attempt to categorize the checks.

Final topics

- We may want to consider fixing the hierarchical numbering system for the
  checks to make future updates easier.
  - Simple is best.

- Sometimes having few users can be a plus, since extensive deployment
  would make discussion of changing the IQ "legacy" difficult.

- Bob Ross has a model with absolute garbage data, that passes ibischk!
  - Every value in it is either 0 or 1f.

Next meeting:

Apr 25 2006
11-12 AM EST (8-9 AM PST)
Meeting access info to be supplied later.

Meeting ended at 12:12 PM Eastern Time.