IBIS Macromodel Task Group

Meeting date: 22 Jan 2013

Members (asterisk for those attending):
Agilent:                      Fangyi Rao
                            * Radek Biernacki
Altera:                     * David Banas
                              Julia Liu
                              Hazlina Ramly
Andrew Joy Consulting:        Andy Joy
ANSYS:                        Samuel Mertens
                            * Dan Dvorscak
                            * Curtis Clark
                              Steve Pytel
                              Luis Armenta
Arrow Electronics:            Ian Dodd
Cadence Design Systems:       Terry Jernberg
                            * Ambrish Varma
                              Feras Al-Hawari
                              Brad Brim
                              Kumar Keshavan
                              Ken Willis
Cavium Networks:              Johann Nittmann
Celsionix:                    Kellee Crisafulli
Cisco Systems:                Ashwin Vasudevan
                              Syed Huq
Ericsson:                     Anders Ekholm
IBM:                          Greg Edlund
Intel:                      * Michael Mirmak
Maxim Integrated Products:    Mahbubul Bari
                              Hassan Rafat
                              Ron Olisar
Mentor Graphics:            * John Angulo
                              Zhen Mu
                            * Arpad Muranyi
                              Vladimir Dmitriev-Zdorov
Micron Technology:            Randy Wolff
                              Justin Butterfield
NetLogic Microsystems:        Ryan Couts
Nokia-Siemens Networks:       Eckhard Lenski
QLogic Corp.                * James Zhou
SiSoft:                     * Walter Katz
                              Todd Westerhoff
                              Doug Burns
                            * Mike LaBonte
Snowbush IP:                  Marcus Van Ierssel
ST Micro:                     Syed Sadeghi
Teraspeed Consulting Group:   Scott McMorrow
                            * Bob Ross
TI:                           Casey Morrison
                              Alfred Chong
Vitesse Semiconductor:        Eric Sweetman
Xilinx:                       Mustansir Fanaswalla
                              Ray Anderson

The meeting was led by Arpad Muranyi

------------------------------------------------------------------------
Opens:

- Arpad: Our next meeting will be Feb 5

--------------------------
Call for patent disclosure:

- None

-------------
Review of ARs:

- None

-------------
New Discussion:

Interconnect Task Group report:
- Michael M:
  - We went over David Banas' survey
  - Will meet tomorrow to for continued survey discussion & summit status report

Arpad: Fangyi is not here for the scheduled discussion

Discuss of survey results:
- David: The results were:
  - Question 1: Which point in a serial communication link is the output of Tx
    GetWave() assumed to be representative of?
    - 7 chose A) The input to the analog channel.
    - 1 chose F) Other. (Please, explain below.)
  - Question 2: Is section 10 of IBIS v5.1 a normative part of the standard,
    which must be followed, or an informational appendix only, intended to
    provide one possible example of the AMI flow?
    - 6 chose Normative
    - 2 chose Informative, one was an EDA person
  - Section 10 is a normative standard that must be followed
- Michael M: How many were model users?
- David:
  - EDA 5
  - IC 2
- Arpad: I responded Informative
- Michael M: The survey is informative, but is it normative?
  - Does this say anything about inputs to GetWave and Init?
- David: I don't think so
  - It would be good if someone else could look into this
  - In Init I convolve responses out of order
  - That should not matter, they are commutative
  - What is the point of being asked to change the point of my signal?
  - Once done with TX output, channel convolution is still the appropriate
    next step
  - Why is it different in the GetWave and non-GetWave cases?
- James: Where does this come from?
- David: From the number of A responses
  - Why does the difference between Init and GetWave exist?
- Arpad: GetWave is not required
  - It needs to have a bit stream formed
- James: Potentially the vendor can apply a portion of the TX to the channel
  response
  - In GetWave this has already been done
- Walter: Historically there was an effort to make the TX and RX functions
  symmetric
  - We could have said the DLL Init returns just the equalization
  - The RX output would have been just the equalization
  - There was resistance to that
  - Someone said it was easier to generate the combined impulse response
  - For GetWave to be symmetric they need to add to an input impulse response
  - Symmetry was the key factor, all function add equalization to an input
- James: The TX is ahead of the channel, but the RX is not
  - They are not symmetrical
- David: No one answered "C"
- Walter: Use_Init_Output required convolving GetWave with Init output
  - In one case you add just the channel response, in another Init output two
  - That caused double counting
- Walter: One model maker had trouble putting a peaking filter in an RX
  - Non-LTI TX with some EQ in TX and RX gets very complicated
  - Requiring deconvolution was a problem
- James; That is outside the scope of the 5.1 spec
- Walter: That was in 5.0, now removed
- James: You can't use both Init and GetWave in a simulation
  - Putting part of the response in each might not work
- Ambrish: Some say we should always use GetWave
  - Init might replicate what GetWave does though
- James: We have to tell the model maker what to do
- Ambrish: To simplify the flow we had to eliminate that path
- James: The old flow is not possible
- Walter: That is right
  - This may be the only thing IBIS ever deprecated because it was unworkable
- James: I don't see a way in 5.1 for the TX to choose settings
- Walter: The idea was to open the eye at the RX
  - The TX could handle eye opening that 4 to 8 years ago, by normative rules
  - Specs gave eye opening requirements
  - At even 6Gb it no longer works
    - Opening the eye at the pin or pad harms the decision point signal
  - Now the RX has to handle eye opening optimization
- James: In the field people need to know the right TX setting
- Walter: That would be an informative rule, not normative
  - Backchannel systems send test patterns to find the optimum setting
- James: The simulation must train, but it can take hours

- Michael M: Can we have more surveying?
  - David's exercise has been very helpful
  - More questions need to be asked
- Ambrish: The spec needs to say more about how things are done
- James: I disagree
  - We do not have to say how things are done
  - We mostly have trouble knowing what to do
- Walter: This can lead to Init-only RX models, leaving users stuck
- James: Agree, users have a lot of frustration
  - The minimum requirement is the results have to come out the same
  - Being faster, etc. is optional

James: We have divisions regarding analog models
- These divisions are arbitrary
- No matter where the division, TX plus channel has to give the same result
- Arpad: If the dividing line moves we need a way to tell the tool

-------------
Next meeting: 5 Feb 2013 12:00pm PT

Next agenda:
1) Task list item discussions

-------------
IBIS Interconnect SPICE Wish List:

1) Simulator directives
