IBIS Macromodel Task Group

Meeting date: 21 October 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:

- Todd produce slides for co-optimization requirements discussion next week.
  - It will take a couple more weeks.

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


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

BIRD 128:

- Arpad: BIRD 128 is still tabled.


BIRD 147.1:

- Arpad: SiSoft still is not comfortable with the BIRD.
  - We are waiting for slides from Todd about industry needs.
- Ambrish: We addressed the two question Todd had posed.
- Arpad: We need to address industry needs.
  - SiSoft said the BIRD could not be improved to do this.
- Todd: We have tried to resolve discrepancies in the BIRD.
  - Cadence answered the two questions that were left.
  - For a while SiSoft had not been certain about whether we would write a new BIRD.
  - Now we think there should be an alternative BIRD.
  - It will not come out until the beginning of December.
- Arpad: Some people have said BIRD 147.1 is not ready for a vote.
  - There are issues other than those posed by SiSoft.
- Ambrish: We would like to know what the issues are.
- Ken: We are 3 years into BIRD 147, it should get a vote.


Filter/analog driver (C_comp related):

- Arpad: Michael Mirmak made presentations about this.
  - BIRD 79 is about this.
  - There are presentations about this.
- Arpad show a presentation by Luca Giocotto.
- Arpad: This shows non-linear capacitance at high frequencies.
  - A circuit is proposed.
  - Tests show good correlation.
  - It shows what it would look like in IBIS.
  - Voltage dependence is also modeled.
  - It adds up to 24 sub-parameters.
  - The BIRD was rejected in 2003.
- David: This is a hard-wired alternative?
  - Why would EDA vendors support this instead of [External Model]
- John: Those replace legacy IBIS, they don't just add to it.
- Arpad: It is more challenging to write AMS models.
  - I have put example models in the public domain.
- Walter: IBIS-ISS was created for interconnect.
  - It would need PWL sources to be enhanced to cover this.
  - That would be better than AMS.
- Arpad: It still would not have event-trigerrable sources.
- Walter: IBIS-ISS could have that, and it could be translated.
  - Receivers are easy.
  - A driver with C_comp corrected K factor can be done with a SPICE circuit.
  - It requires a node with time since last translation.
- Arpad: It would be easier with a triggered source.
  - Should we work on adding to IBIS-ISS?
  - Or should we do something with keywords for the short term?
- Randy: We should have done this long ago.
  - It would be useful for DDR4 and other technologies.
- Walter: For DDR4 it is simple.
  - IBIS-BSS could be used, with an event trigger element.


Back-channel proposals:

- Walter: We will have a new BIRD, will have vendors review it.
- Ken: Will it preclude the back-channel work we have done?
- Walter: No.
- Todd: The flows are similar but different.
  - We want the committee to look at the BIRDs side by side.
- Ken: Maybe there will then be a third proposal, taking more time.
- Todd: We are not confident we have a handle on this yet.
- Ambrish: Can we see something next week?
- Todd: No, we have to complete it.
- Arpad: Voting without an alternative is different than with an alternative.
- Ambrish: We have had other proposals.
- Arpad: BIRD 147 is the only one for back-channel.
- Ambrish: We are concerned about the time.
  - We have been misled.
- Todd: We have tried in good faith to close the gap.


New Redriver Flow BIRD:

- Walter showed page 8 of Redriver_Flow_BIRD.docx.
- Walter: This shows an Init-only flow.
  - RX2 Init gets the TX2 output into just channel 2.
  - If TX2 is a redriver then TX1 to RX1 equalization must be added in.
- Arpad: Agree with this.
- John: Yes, for redrivers but not retimers.
- David: It's a fully analog path, not digital.
- Walter: Fangyi agrees with this flow.
  - We are not sure how much difference it makes however.
  - We had a problem with time domain simulations with an Init-only RX2.
  - It requires Init_Returns_Impulse along the whole path, plus convolution.
  - Analyzing all possibilities you end up with 81 combinations.
  - Handling all of these unique flows would be painful.
  - The BIRD only handles the case where every model has Init_Returns_Impulse.
- Fangyi: Where does it say that?
- Walter: It doesn't yet.
  - IC vendors believe this is the right flow.

- Arpad: Can we submit this now, addressing mixed cases later?
  - This is a flaw in the specification, it should be fixed immediately.
- Walter: We could do that.

- Fangyi: Retimers are used because they are not linear.
  - We may not often see Init-only models.
- Todd: More than that, they can be adaptive.
- Arpad: That might limit the number of possible permutations.
- Fangyi: RX1 and TX1 do not both require GetWave.
  - Only one would be needed, they are part of the same redriver.
  - There can be optical channels in the middle.

- Bob: Is this a flow change?
- Walter: It can be more complicated if TX2 equalization depends on channel 2.
  - Of course the silicon never does that.
  - A parameter could say the TX is not adaptive.
  - RX2 can't deal with non-linearities that happen before it if it is Init-only.
- Arpad: We should write a BIRD soon to correct this flow problem.

AR: Walter write formal Redriver Flow BIRD

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

1) Simulator directives
