Minutes, IBIS Quality Committee

16 October 2012

11:00-12:00 EST (08:00-09:00 PST)

ROLL CALL

Altera				  * David Banas
Cisco Systems:                      Tony Penaloza
Ericsson:                           Anders Ekholm
Green Streak Programs:              Lynne Green
Huawei Technologies:                Guan Tao
IBM:                                Bruce Archambeault
                                    Greg Edlund
Intel                               Michael Mirmak
IOMethodology:                    * Lance Wang
Mentor Graphics:                    John Angulo
Micron Technology:                  Moshiul Haque,
                                    Randy Wolff
Nokia Siemens Networks:             Eckhard Lenski
QLogic Corp.:                       James Zhou
Signal Consulting Group:            Tim Coyle
Signal Integrity Software         * Mike LaBonte
Teraspeed Consulting Group:       * Bob Ross
Texas Instruments:                  Pavani Jella

Everyone in attendance marked by *

NOTE: "AR" = Action Required.

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

Call for opens and IBIS related patent disclosures:

- Mike asked about email

AR Review:

- Mike post spreadsheet plus CSV files to web archive
  - Done

- Mike add FOM to spreadsheet
  - Done

- Mike contact David Banas for an update on his FSV method
  - Done, David joins us today

- Lance check on China patent status for correlation methods
  - It is pending in both China and the U.S.
  - The attorney said it would be easiest to cancel the patent
  - Lance will have a meeting with the attorney

New items:

Mike showed 2007
- Slide 9:
  - David: The Accuracy Handbook hints at the possibility of FSV
- Slide 12:
  - Mike: 3 metrics show no errors
    - Maybe lower thresholds are needed?
  - David: In 2006-2007 s2ibis was hardcoded to produce only 100 points
    - That was insufficient, but at Xilinx we didn't know this at the time
    - We were stuck looking at overlay and envelope metrics
    - I was tasked with exploring FSV
    - IC designers suggested these 5 metrics
    - We looked at how sampling extends the observed rise/fall times
    - Improving resolution fixed the problem
  - Mike: Was this diagram before or after the improvement?
  - David: Not sure
  - Lance: Is the same x-axis time used for both SPICE and IBIS?
  - David: We use (dt_IBIS - dt_SPICE) / dt_SPICE
    - The absolute time was never critical
    - The data rate was slow enough so picking an edge was not critical
  - Lance: Is the stimulus multiple pulses?
  - David: We used a square wave, several periods
    - We insured high and low levels were the same
- slide 13:
  - David: We explored the possibility our models were too strong or weak
    - That would produce both rise/fall and high/low differences
    - We were seeing only rise/fall differences
  - Bob: The test load is the same as the standard load?
  - David: We used a different load
- slide 17:
  - David: We automated it using HSPICE .measure statements

Mike: Have you expanded on that work since 2007?
- David: No, I am dealing with SerDes
- Mike: Many simulations with different settings have to be run
  - The IBIS FOM is convenient because you can plot it on the Y axis
  - I showed a presentation about how the numbers and the appearance match up
  - The appearance tends to be worse than the numbers imply
- David: The comparison is better at the RX end
  - Above 10Gb/s high frequency content may show large differences at the TX
  - But they don't matter at the RX
  - Correlation looks worse when simplistic channel models are used
- Mike: Do we need more complex test channels?
- David: It can be simple, but it needs to be apropos to your customer's needs
- Mike: Would it be reasonable to turn the 5 metrics into one?
- David: That would be counter-productive
  - You want to retain the details for debugging
- Mike: We need a single goodness metric when delivering good models
- David: What does the customer need?
  - We use eye area where a single number is needed
- Mike: That probably would apply to non-SerDes models too

Mike: We could apply the 5 metrics at how our test waveforms
- David: I would be interested in seeing if test waveform set is sufficient

Mike: ack leaves the denominator up to the user
  - The common approach seems to be the enclosing rectangle of the reference
  - Another approach is the area under the reference
- David: It is not clear what to use there
  - I like to measure in terms of power, so area squared may help
- Mike: A charge accumulation formula might be good

Review of group meeting schedule:

Next meetings:
- The next two meetings will be Oct 30 and Nov 13

- Meeting ended at 12:04 ET
