[ibis-macro] Minutes from the 13 Jan 2015 ibis-atm meeting
- From: Mike LaBonte <mike@xxxxxxxxxxx>
- To: 'IBIS-ATM' <ibis-macro@xxxxxxxxxxxxx>
- Date: Fri, 16 Jan 2015 16:01:19 -0500
Minutes from the 13 Jan 2015 ibis-atm meeting are attached.
Mike
IBIS Macromodel Task Group
Meeting date: 13 January 2015
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
IBM * Steve Parker
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
Teraspeed Labs: * Bob Ross
(Note: Agilent has changed to Keysight)
The meeting was led by Arpad Muranyi.
------------------------------------------------------------------------
Opens:
- Arpad: We will meet Jan 20, but not Jan 27.
- We should return to our normal schedule in February.
- Bob: I'd like to ask about DesignCon presentations.
- Will Todd have a presentation on co-optimization?
- Todd: Yes
- Bob: The deadline is next week.
- Arpad will have an ATM committee report?
- Arpad: I can put one together
- Bob: Randy has one?
- Randy: I will have one on C_comp
--------------------------
Call for patent disclosure:
- None
-------------
Review of ARs:
- Todd update slides for co-optimization requirements.
- Done.
- Arpad to review IBIS spec for min max issues.
- In progress.
-------------
New Discussion:
Co-optimization:
- Todd showed a presentation IBIS-AMI and Co-Optimization.
- slide 9:
- Todd:
- We had requests from three people:
- David: Using legacy Tx models in co-optimization.
- Radek: Letting the EDA tool run co-optimization.
- Arpad: Call models in correct order to capture channel response without
TX EQ.
- We put these into a new scenario, Scenario 3.
- Radek: It is users who are asking to use Scenario 3.
- Todd: For Scenario 3, meta data about the TX will have to be supplied.
- slide 10:
- Todd: We want to avoid running blind sweeps, but without recompiling new
DLLs.
- slide 11:
- Todd: The requirements are similar to other scenarios because users want to
do the same things.
- It is important that the TX DLL do what it's told, no other tap
adjustments.
- Todd: Does this satisfy the requests of Radek and David?
- David: It seems to satisfy mine, but the details will be important.
- Radek: This looks good, but we want to operate with new models too.
- It should be a user choice, whether the model or EDA tool is the master.
- Walter: There are 3 scenarios and 4 configurations
- The 3 scenarios are:
- The models do what the hardware does.
- The RX is in control.
- The EDA tool is in control.
- The 4 configurations are:
- Legacy TX, legacy RX.
- Legacy TX, new RX.
- New TX, legacy RX.
- New TX, new RX.
- In Scenario 3 the RX may have capability but it is not used.
- We want to support all scenarios and configurations.
- Todd: No one is talking about Scenario 2 with models that don't understand
co-optimization.
- Walter: Scenario 2 could work with a legacy TX.
- Scenario 1, it's unlikely a legacy model can be used for bit-by-bit
emulation.
- Todd: There is confusion between 2 and 3.
- Walter: The EDA tool will do the best it can given model limitations.
- Todd: Legacy DLLs would not accept GetWave AMI_parameters_in.
- Walter: They might abort.
- David: The difference between 2 and 3 is who is driving?
- Todd: That is a major difference.
- David: Scenario 2 says "without hardware backchannel".
- Ambrish: Scenario 1 is for hardware that really has a backchannel.
- Walter: The meta file to describe how to run the TX model is important.
- The format will have to be carefully defined.
- Whether the RX is optimizing or the EDA too, it could work.
- For legacy models it will involve closing and reopening TX Init.
- Todd: In Scenario 3 the simulator steps in for one or both models.
- Walter: That could be done in Scenario 2 also.
- Todd: That may be possible but not as easy.
- David: This opens a lot of possibilities.
- Scenarios 1 and 3 are clear.
- Scenario 2 needs clarification.
- There may be more than 3 scenarios.
- Walter: Scenario 2 confuses flows with function.
- It does not have to find an optimal point using control loops in the
silicon.
- David: So Scenario 2 means it's not limited to what the hardware on the bench
can do.
- It could be called "model-only optimization".
- Ambrish: Should Scenario 2 be called "Init-based"?
- Walter: It can be done in either Init or GetWave.
- It should not be confused with statistical vs. silicon optimization.
- The legacy model in Scenario 3 can only do Init optimization.
- Todd changed the title of Scenario 2.
- Todd: Scenario 1 is the most literal
- This is what started the discussion.
- Scenario 2 is to solve the same problem where the silicon doesn't automate
it.
- The new requests got called out in their own scenario.
- The distinction between 1 and 2 is small, from a functional view.
- Walter: Scenario 2 might even be used where hardware optimization exists.
- This would be because it can be hard to do.
- The RX can have a proxy TX to get channel information otherwise unavailable.
- Todd: Scenario 2 makes us debate whether it is worthwhile to add this
optimization capability.
- Whether or not there really is hardware might be just a footnote, not so
important.
- David: These 3 scenarios seem to cover all cases.
- Ambrish: Scenario 1 and 2 are driven by the RX, not so different.
- Radek: And Scenario 3 is driven by the EDA tool.
- And using legacy models is one application of that.
- Todd: For Scenario 3 we assume one or both models are legacy.
- Radek: That does not have to be the case.
- Walter: It could be neither.
- Todd: It is better to say one or both might not participate.
- It is not automatic that the EDA tool is in control for Scenario 3.
- The control function must exist somewhere.
- When I say "not participating" I mean a legacy model is not doing special
things for training.
- It knows nothing about back-channels.
- Walter: The TX participates in the simulation, but not using the new features
in the BIRD.
- Todd: We assumed the same message mechanism could be used for Scenario 3.
- But the EDA tool would stand in for part of what one or both models need to
do.
- Radek: That gives us the distinction between scenarios and configurations.
- The presentation mixes those things and needs to be clarified.
- Walter: We should agree it should be possible to use legacy TX models, with
meta-files added.
- Also the exploration can be done by the RX or the EDA tool.
- The BIRD should support all of that.
- Scenario 3 can't be done by a tool that does time domain optimization.
- Ambrish: Are you saying Scenario 3 is the primary scenario?
- Walter: No, number 1 we want the case where both models fully support
optimization.
- Where neither do this, we are limited to Scenario 3.
- But Scenario 3 can use models that do support optimization, it just won't
use it.
- Radek: That is in agreement what what I thought.
- Walter: Arpad made the point about the CTLE need a channel response with no
EQ.
- This supports that by requiring a TX that can be told what to do.
- Todd: 3 scenarios and 4 configurations sounds hard to explain.
- We will have to present this carefully.
-------------
Next meeting: 20 Jan 2015 12:00pm PT
-------------
IBIS Interconnect SPICE Wish List:
1) Simulator directives
Other related posts:
- » [ibis-macro] Minutes from the 13 Jan 2015 ibis-atm meeting - Mike LaBonte
