IBIS Macromodel Task Group Meeting date: 02 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: - Mike will not be able to attend next week - Curtis will take minutes. - Arpad: -------------------------- Call for patent disclosure: - None ------------- Review of ARs: - Radek work with Walter and Ambrish on BIRD 128.1 clarifications. - Done - Arpad send double quote BIRD draft to Mike for posting - Done - Mike post double quote BIRD draft - Done - Arpad send MinMax BIRD draft to Mike for posting - Done - Mike post MinMax BIRD draft - Done - Ambrish investigate other ways to communicate tap coefficients - Ambrish would like to discuss today. - Arpad to review IBIS spec for min max issues. - In progress ------------- New Discussion: BIRD 128.1: - Arpad: Were Radek's comments incorporated? - Radek: It was too late. - Arpad: Then it can not be voted in the next Open Forum meeting. BIRD 147 - Ambrish motioned to untable BIRD 147. - Bob seconded the motion. - There were no objections. - Ambrish showed BIRD 147_draft12.docx - Ambrish described changes. - Integers are used to specify different Training mode - There is a separate Init mode. - A coefficient passing method is described. - Arpad: How does the information get from TX to RX? - Also what happens when the increment sizes are incompatible? - Ambrish: It tries until a satisfactory result is achieved. - Arpad: Should the granularity be communicated? - Ambrish: It will go to the nearest valid value when it has to. - Arpad: Will the TX report what was actually used? - Ambrish: Yes. - Arpad: The RX may send requests and the TX reports the value has not changed. - Walter: This is peripheral to the PCIeG3 spec. - The BCI details do not belong in BIRD 147. - The communication mechanism should be in this BIRD. - Why should a new BIRD be needed to create a new "ibis_*" standard protocol? - Ambrish: The BCI file specifies the protocol. - Walter: This may not work for fiber channel, for example. - How does the TX advertise the constraints on tap coefficients? - TX Init could generate a BCI branch, and the RX could send one back. - This would be documented for each of the flows. - Ambrish: It could be done in the first GetWave. - Use_Init_Output ??? - Todd; Walter wants to optimize using Init and GetWave in the same run. - This BIRD outlaws that. - Ambrish: They can't be used in the same simulation. - That is similar to the Use_Init_Output flow, which was deprecated. - Walter: They can. - Todd: Knowledge gained in Init processing might help GetWave. - The BIRD has training for Init-only and GetWave-only. - Ambrish: Multiple Training modes can be used. - Todd: That requires models to retain settings from Init, but the BIRD doesn't allow that. - Walter: The spec allows the RX to optimize itself in Init, then in GetWave. - Ambrish: Where is that specified? - Todd: It is not explicitly allowed, but is not disallowed. - Ambrish: Init and GetWave are supposed to be used separately. - Todd: The spec allows them to be used together. - Walter: Pages 172-173 of IBIS 6.0 describes how this can work. - It says RX Init in time domain can contain an optimization algorithm. - Ambrish: Do others agree with this? - Todd: In bit-by-bit simulation GetWave will be used to get the eye. - There is no reason not to get the eye from Init too. - Walter: IBIS page 176 says both can be used. - Todd: Terms like "simulation output" are not precise. - Simulations can be "jump started" with data from Init. - Ambrish: This BIRD is written literally using what IBIS 6 specifies regarding Init and GetWave flow for the same model. - Todd: We disagree. - Ambrish: You are trying to have the BIRD support something that your tool implemented. - Walter showed IBIS 6 page 172 and 173. - Walter read the language in the paragraph between those pages. - Todd elaborated on how both Init and GetWave are used in time domain flow. - Ambrish: My BIRD should not have to specify that. - What you are doing is not part of the reference flow or application scenarios described in the specification. - Todd: We do not want it disallowed. - Ambrish: Models can implement PCIeG3 using this. - Walter: This can be accomplished by passing BCI branches. - Much of what is in the BIRD can be left up to the protocol. - Otherwise it may take time to find we don't support other protocols. - Ambrish: All but PCIeG3 require only increment/decrement. - Walter: Fiber channel uses coefficients. - Arpad: Would this BIRD as is allow ?? in the GetWave function? - Todd: The flow described had punting in Init and training in GetWave. - Arpad: Is it possible? - Todd: Yes, but with a performance penalty. - Ambrish: The question is if we move communication to the BCI file. - Then we would have another document to read to implement the back-channel communication. - It would be better to have it all in the IBIS specification. - Walter: We don't want to have to wait for IBIS 7. - Todd: This should be discussed within Cadence. - Ambrish: Others should weigh in too. - Arpad: This will be on the agenda next week. - Walter: Radek may wish to discuss it with Ambrish. ------------- Next meeting: 02 September 2014 12:00pm PT ------------- IBIS Interconnect SPICE Wish List: 1) Simulator directives