IBIS Macromodel Task Group Meeting date: 11 May 2021 Members (asterisk for those attending): Achronix Semiconductor * Hansel Dsilva ANSYS: * Curtis Clark * Wei-hsing Huang Cadence Design Systems: * Ambrish Varma Ken Willis Jared James Google: Zhiping Yang Intel: * Michael Mirmak Kinger Cai Alaeddin Aydiner Keysight Technologies: * Fangyi Rao * Radek Biernacki * Ming Yan Todd Bermensolo * Rui Yang Luminous Computing David Banas Marvell Steve Parker Micron Technology: * Randy Wolff * Justin Butterfield Missouri S&T Chulsoon Hwang Siemens EDA (Mentor): * Arpad Muranyi SiSoft (Mathworks): * Walter Katz Mike LaBonte Teraspeed Labs: * Bob Ross Zuken USA: Lance Wang The meeting was led by Arpad Muranyi. Justin Butterfield took the minutes. -------------------------------------------------------------------------------- Opens: - None. ------------- Review of ARs: - Walter to send his PAMn proposal to Randy to submit to the Open Forum as an official BIRD. - Arpad reported that this is done and the draft is submitted to the IBIS Open Forum as BIRD213. - Walter to create a BIRD211.2_draft4 with today's changes and send it to ATM. - Done. -------------------------- Call for patent disclosure: - None. ------------------------- Review of Meeting Minutes: Arpad asked for any comments or corrections to the minutes of the May 4th meeting. Walter moved to approve the minutes. Bob seconded the motion. There were no objections. ------------- New Discussion: BIRD211.2: Walter made editorial changes based on the last meeting. He added Fangyi as a co-author. Bob noted he has additional comments. Hansel asked, when deconvolution is used, why not have the model use the pulse response as an alternative. Walter replied, if GetWave exists, there is no need to run deconvolution. The only reason for deconvolution is if GetWave does not exist and you want to run time-domain simulation. Any GetWave function could continue to adapt over time. If the GetWave function is continuing to adapt, there could be issues getting the correct pulse response from the data. Bob commented, when you define a parameter, the equal sign should not be used in the text. We should use the word "is" in these cases. Bob also noted it is ambiguous to say "IBIS 7.0". An IBIS-AMI 7.0 model could be called by an IBIS 6.1 file. Walter changed the equal sign to "is". Walter asked about the issue with the "IBIS 7.1" phrase. Bob noted on page 3, the mention of "IBIS 7.1" is not necessary as the parameter already states illegal before IBIS 7.1. Walter asked if we should remove this line from the text. Bob and Radek agreed. Walter asked about the "IBIS 7.1" on page 6. Bob asked whether this is talking about the Rx models without the Tx_Impulse_Input reserved parameter, as the IBIS-AMI model can be called from other versions of IBIS. Walter suggested to remove the "IBIS 7.1" from this statement. Fangyi disagreed with this, as we need a qualifier for the statement that follows. Walter stated this section is not talking about the new parameter. If you do deconvolution, you can use the unit impulse response. For pre IBIS 7.1, deconvolution can be used. He thought the statement is correct as is. Arpad suggested to say IBIS-AMI version, as the IBIS file could have a different version. Bob stated the parser will support different versions of the IBIS and IBIS-AMI. Ambrish asked if there is a situation where an EDA tool would need to do deconvolution for IBIS 7.1. Walter replied for IBIS 7.1 there should not be a need for deconvolution, as the EDA tool can use the unit impulse response. Walter commented this is detailed in the section about the reference flow for the time-domain when GetWave_Exists is false. Bob commented that we have the existing flow and there are three additional flows documented in the BIRD, but there are only three flow diagrams. He suggested we should add a figures for the existing flow. Walter agreed we need to add a diagram for the Upstream flow. Bob suggested we need a figure for the existing flow as well. Walter stated the default existing flow is wrong and needs to be deprecated. Ambrish objected to this statement and noted we need to say this applies to the existing redriver flow only. Walter commented that we need to fix the redriver flow whether the new parameter exists or not. Bob noted all tools need to change even for IBIS version 6.1 models. The other option is to keep the old flow, and the only way to use the new flow is to have the new Tx_Impulse_Input parameter. Arpad asked about the case where he has an IBIS-AMI model with no AMI_Version in it making it an IBIS-AMI version 5.0 model, but the model maker really intended to use the IBIS 5.1 flow. He noted we have a similar case for the redriver flow. Walter commented that anyone who follows the old flow will get incorrect results. Arpad said we should say this in the standard. Ambrish asked if this is only true for statistical flow with Init models. Ambrish asked is the expectation that the EDA tool should use the new flow. Bob agreed that the original flow may be incorrect, but it should be documented. Arpad suggested we should sate that the corrected or new flow should be used even for old models which have AMI_version older than 7.1. Ambrish suggested we need to be specific about what is broken and what we are fixing. Bob commented that the specification needs to be clear for all cases. Arpad asked whether we should make a similar statement about IBIS-AMI 5.0 models expecting to be used with the 5.1 flow. Walter will send out an updated draft of BIRD211.2. PAMn BIRD: Walter would like to discuss the important issues with the PAMn BIRD. Currently, the IBIS specification has PAM4 thresholds. The proposal is to put the thresholds in a table and have the thresholds be a function of the modulation levels. Walter noted there was a comment on the symbol definition. For PAM4, it was simple, but it is more complicated for PAM3 or PAM5. He added PAM_Mapping_Name and PAM_Mapping_Table parameters to map binary numbers to the symbol patterns. Walter gave an example of 7-11 mapping with 11 binary bits mapped into 7 symbols. Fangyi thought the table approach might not work for all duo binary cases, because there are variants which do not use static mapping, i.e. the new state depends on the previous state. - Ambrish: Motion to adjourn. - Walter: Second. - Arpad: Thank you all for joining. AR: Walter to add a diagram for the Upstream flow to BIRD211.2. AR: Walter to send out an updated draft of BIRD211.2. ------------- Next meeting: 18 May 2021 12:00pm PT ------------- IBIS Interconnect SPICE Wish List: 1) Simulator directives