IBIS Macromodel Task Group Meeting date:22 June 2021 Members (asterisk for those attending): Achronix Semiconductor: Hansel Dsilva Amazon: * John Yan 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. Curtis Clark took the minutes. -------------------------------------------------------------------------------- Opens: - Bob noted that the parser developer for ibischk7.1 had asked some questions about checking for the existence of AMI_Impulse in the executable model depending on the values of BCI_Training_Mode (BIRD201.1). Arpad suggested we could talk about this at the beginning of the meeting, as it potentially affects the delivery of an IBIS7.1 parser. ------------- Review of ARs: - None. -------------------------- Call for patent disclosure: - None. ------------------------- Review of Meeting Minutes: Arpad asked for any comments or corrections to the minutes of the June 15th meeting. Ambrish moved to approve the minutes. Randy seconded the motion. There were no objections. ------------- New Discussion: AMI_Impulse existence check question: Bob relayed a question from the parser developer working on ibischk7.1. If BCI_Training_Mode has allowed values of "Impulse" or "Both", then the new AMI_Impulse function must also exist? The group agreed it should and that ibischk should check for the existence of AMI_Impulse under those conditions. Another question was whether Init_Returns_Impulse should be True under the same conditions. Randy noted that it's clear from BIRD201.1 that AMI_Impulse will modify the IR as part of its training. Is there a case where AMI_Init would not modify the response (Init_Returns_Impulse set to False) and AMI_Impulse exists? Ambrish said you would assume AMI_Init would also modify the IR if AMI_Impulse exists, for example, in case there's an EDA tool that doesn't support the BCI statistical training. Radek noted that the BIRD currently doesn't say anything about Init_Returns_Impulse. He suggested a simple solution would be to state that if you have BCI training parameters for statistical mode training, but Init_Returns_Impulse is set to False, then training will not proceed. Walter said you could conceivably have an Rx doing training that does not return an impulse from its AMI_Init because the Rx is training the Tx. He said a Tx doing BCI statistical training would have to return a modified IR from its AMI_Init, or it wouldn't make sense. Ambrish suggested that for simplicity we should just require that Init_Returns_Impulse be True for the Tx and the Rx if BCI statistical training is advertised/allowed. Walter said he had no objection to requiring it for both the Tx and Rx. Bob also said he had no objections. Randy asked if this would be an update to the Usage Rules for BCI_Training_Mode. Arpad agreed it would, but he said it's a technical change and asked if it could be handled in Editorial. The group decided it would be best to consider a BIRD for this change. Bob to work out the details with the Quality task group [AR]. Radek asked whether IBIS 7.0 explicitly states that Get_Wave_Exists must be True for time domain BCI training. Ambrish confirmed that it does (in the Usage Rules: for BCI_Protocol). BIRD211.3 draft: Fangyi reviewed and shared the new post-BIRD211.2 draft he had emailed to ATM the day before the meeting. He had captured all the changes from the previous meeting's discussions of Ambrish's comments. From Fangyi's email's summary of the changes: 1. Clarified that for a terminal Tx the upstream impulse response is a unit impulse response 2. Clarified that if the EDA tool adds a unit impulse response to aggressor columns then it shall also increase the value of argument aggressor by one 3. Clarified that "Tx filter impulse response" is "determined" by the EDA tool 4. Separated statistical and time domain flows and removed mention of "initialization flow" 5. Consolidated repeated notes about old flow in Redriver flow section. 6. For symmetry, added this sentence: "By setting Tx_Impulse_Input to "Downstream", the model maker is declaring that the Tx initialization (AMI_Init) function does not have the ability to adapt itself based on the upstream channel." The group then discussed additional suggestions from Ambrish. 1. Ambrish asked if "unit impulse response" was now being used in the text before it was defined. 2. Ambrish noted that consistency between existing and new figures could be improved. He said the Tx and Rx devices were shown as polygons in legacy figures but as rectangles in some of the new figures. Arpad said the polygons were used in top level physical channel figures, but he agreed that we should be consistent. Ambrish said he actually preferred the polygons. Fangyi agreed the Tx and Rx in physical channel figures should be shown as polygons. 3. Ambrish asked if the "Latch" location shown in some new figures was necessary. Fangyi said he thought "Latch" should remain in the flow diagrams but not be shown in the physical channel diagrams. 4. Ambrish suggested that including the figure for the existing Redriver statistical flow, which is being replaced anyway, will just cause confusion. He said that illustrating it, only to then say it is being replaced, is not necessary. Arpad agreed and said that this section had caused confusion when he first read it. They also questioned the text and the "pre 7.2" AMI_Version discussion. Arpad said the concept was clear but the language was labored. Ambrish suggested we simply remove the legacy diagram and the "pre 7.2" related sentences. Curtis said that we could avoid the use of the version specific text, because "If Tx_Impulse_Input is not present" implicitly covers all models prior to the version in which Tx_Impulse_Input is introduced. Fangyi agreed that we could work on streamlining this section. 5. In the Definition of the Issue section of the BIRD, Ambrish noted that the third item in the list of known issues in the Redriver statistical flow is not restricted to that flow alone. The group added text to the third item stating that it also can affect the regular single channel flow. 6. Bob suggested that we have so many names, e.g., AMI functions, arguments, Reserved Parameters, that it might be helpful in some places to explicitly state "Reserved Parameter" before using a parameter name, etc. Fangyi said he thought some of these suggestions could be handled in Editorial. Bob and Arpad mentioned that this BIRD would now be relative to IBIS 7.1, not IBIS 7.0, so some references would have to change. - Curtis: Motion to adjourn. - Ambrish: Second. - Arpad: Thank you all for joining. AR: Bob to work with the Quality task group on a BIRD to address the AMI_Impulse existence requirement and the Init_Returns_Impulse set to True requirement for statistical BCI mode. AR: Fangyi to create a new draft of BIRD211.3 with the changes discussed in today's meeting and some additional comments from Arpad via email. ------------- Next meeting: 29 June 2021 12:00pm PT ------------- IBIS Interconnect SPICE Wish List: 1) Simulator directives