IBIS Macromodel Task Group

Meeting date: 23 September 2014

Members (asterisk for those attending):
Altera:                       David Banas
ANSYS:                      * Dan Dvorscak
                            * Curtis Clark
Avago (LSI)                   Xingdong Dai
Cadence Design Systems:     * Ambrish Varma
                              Brad Brim
                              Kumar Keshavan
                              Ken Willis
Ericsson:                     Anders Ekholm
Intel:                        Michael Mirmak
Keysight Technologies:        Fangyi Rao
                            * Radek Biernacki
Maxim Integrated Products:    Hassan Rafat
Mentor Graphics:            * John Angulo
                            * Arpad Muranyi
Micron Technology:          * Randy Wolff
                              Justin Butterfield
QLogic Corp.                  James Zhou
                              Andy Joy
eASIC                         Marc Kowalski
SiSoft:                     * Walter Katz
                            * Todd Westerhoff
                            * Mike LaBonte
Synopsys                    * Rita Horner
Teraspeed Consulting Group:   Scott McMorrow
                            * Bob Ross

(Note: Agilent has changed to Keysight)

The meeting was led by Arpad Muranyi.

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

- None

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

- None

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

- Ambrish send updated BIRD 147 draft to Mike for posting.
  - Done

- Todd produce slides for co-optimization requirements discussion next week.
  - Not done.

- Todd send BIRD 147 notes to Ambrish.
  - Done

- Arpad to review IBIS spec for min max issues.
  - In progress.


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


BIRD 147.1:

- Todd: SiSoft sees two optimization needs:
  - 1: Emulating what hardware back-channel does.
  - 2: Finding optimal settings for hardware without back-channel capability.
    - This is not hardware emulation.
- Arpad: In the second case the tool is doing the user's job?
- Todd explained the two needs again.
- Walter: The RX can give the TX settings directly, without iterating.
- Todd: Most adaptation complexity is in the RX.
  - Some people turn off TX EQ and let the RX do it all.
  - That doesn't work well.
  - The point is that if all TXs are nearly the same, we can have smart RXs
    sending generic "prescriptions" to the TXs.

- Arpad: We have to decide whether to recommend this BIRD for a vote.

- Ambrish: Do we want to emulate the hardware protocol?
- Walter: The TX and RX have to communicate their capabilities.
- Arpad: We should not care if the hardware has the capabilities the model has.
- Walter: A model might communicate settings in a way that is not approved.
  - This BIRD can't communicate tap indexes.
  - You can't have "ibis_" in the name unless it matches an approved protocol.
- Todd: Cadence changed the BIRD to allow both index and coefficients.
- Walter: I don't want top be prevented from finding another way.
  - I sent an example which was deemed not approved.
- Todd: The goal is to determine how to configure the TX and RX, and the margin.
  - We emulate the path only because it might go astray.
- Ambrish: It might not converge.
- Todd: Taps have not only ranges, but also granularity.
- John: Co-optimization might require just getting the TX and RX talking, without a protocol.

- Arpad: Can BIRD 147 be modified to satisfy SiSoft?
  - If not we might recommending rejecting it.
- Walter: BIRD 147 would be acceptable if:
  - 1: both Init and GetWave allow optimization.
  - 2: There is a way to have models approved as standard.
- Todd: We are fairly close.
- Ambrish: The BIRD can cover any protocol.
- Arpad: Do Todd's comments to Ambrish resolve anything?
- Todd: They were about style and clarification.
  - We are requesting another training modes.
- Ambrish: You can do that.
- Todd: Training says what you are doing, but also what the model can do.
  - The model can initialize in Init and continue in GetWave?
- Ambrish: Yes.
- Todd: When we call GetWave we will have to change values and the model must recognize that.
  - There is no way to know the model can't do that.
- Walter: Is it a List or a Value?
- Todd: It's a List
- Walter: We need to handle this in one simulation.
- Ambrish: Yes that can be supported.
- Walter: There are words saying what can be approved by IBIS.
- Todd: Cadence will stand by this interpretation?
- Ambrish: If SiSoft wants to run Init then GetWave there is nothing illegal about that.
- Todd: The next question is what's allowed in approved BCI messages.
  - What would be an example?
- Walter: A TX transmitting min and max taps, index values, and tap coefficients for the Init response.
- Todd: You want both the mapping and current state?
  - Can we define things beyond a tap list in a BCI file?
  - We might have a "framis", a List of 3 floating point numbers.
  - Can I write a BCI file that will be approved?
- Ambrish: Yes
  - Protocol_Specific can have any parameter.
- Walter: I will write an example BCI file for the task group to check.

- Radek: I would prefer to have increment/decrement in the BCI file, not Training_Parameter.
- Ambrish: That needs to be in the specification.
- Walter: An RX model can have 2 protocols?
- Ambrish: Correct.
- Walter: One protocol will have increments, the other coefficients.
  - How will it know?
- Ambrish: That will require two files.
- Walter: I need models that do training first in Init, then in GetWave.
  - How to I tell the model I want to train in GetWave?
- Todd: AMI_parameters_out becomes an In.
- Walter: Do the parameters sent to Init get sent to GetWave AMI_parameters_out?

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

1) Simulator directives