IBIS Macromodel Task Group Meeting date: 12 January 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 Mentor, A Siemens Business: * Arpad Muranyi Micron Technology: Randy Wolff * Justin Butterfield SAE ITC Jose Godoy 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: - None. ------------- Review of ARs: - Fangyi to add comparisons of proposed variants of Init function input/output data to his presentation. - Done. Fangyi to present the new slides today. -------------------------- Call for patent disclosure: - None. ------------------------- Review of Meeting Minutes: Arpad asked for any comments or corrections to the minutes of the January 5th meeting. Lance moved to approve the minutes. Ambrish seconded the motion. There were no objections. ------------- New Discussion: Redriver Flow Issues: Fangyi shared his updated presentation "AMI Redriver flow". Fangyi briefly reviewed slides 3 and 4, unchanged from last week's presentation, which introduce the terminology he uses (slide 3) and provide a diagram of one scenario and some of the various crosstalk paths (slide 4). slide 6 - Comparison of Inputs to AMI Init (new slide) Note: variant 1 - Fangyi's proposal variant 2 - Walter's proposal from last week's meeting. variant 1: - Tx and Redriver Tx Init Inputs - IR Matrix columns consist of: - Direct downstream IR (same as existing flow) - Cross talk IR terms (same as existing flow) - Upstream cumulative IR (new column) - Rx and Redriver Rx Init Inputs - IR Matrix columns consist of: - Direct upstream IR (same as existing flow) - Crosstalk IR terms (same as existing flow) - Upstream cumulative IR (new column) variant 2: - The upstream cumulative IR is the only column of the IR Matrix input to Init for both Tx and Rx. slide 7 - Comparison of Outputs of AMI Init (new slide) variant 1: - Tx and Redriver Tx Init Outputs - IR Matrix columns consist of: - Direct downstream IR equalized (same as existing flow) - Crosstalk IR terms equalized (same as existing flow) - Upstream cumulative IR equalized (new column) - Redriver Rx Init Outputs: - IR Matrix columns consist of: - Direct upstream IR equalized (same as existing flow) - Crosstalk IR terms equalized (same as existing flow) - Upstream cumulative IR equalized (new column) - Terminal Rx Init Outputs: - IR Matrix columns consist of: - Direct upstream IR equalized (excluding DFE (differs from existing flow)) - Crosstalk IR terms equalized (excluding DFE (same as existing flow)) - Upstream cumulative IR equalized (AFE and DFE included) (new column) variant 2: - Tx and Redriver Tx Init Outputs - IR Matrix columns consist of: - Upstream cumulative IR equalized - IR of the equalization itself - Redriver Rx Init Outputs: - IR Matrix columns consist of: - Upstream cumulative IR equalized - IR of the equalization itself - Terminal Rx Init Outputs: - IR Matrix columns consist of: - Upstream cumulative IR equalized (equalization includes AFE and DFE) - IR of the equalization itself (AFE only, no DFE) Ambrish asked why the first column of the Terminal Rx Init Output in variant 1 does not include DFE (this is how it differs from the existing flow). Fangyi said we only want DFE applied to the through channel IR, and that is what is returned in the last column of the Impulse Matrix in variant 1. We want the direct upstream IR without DFE (column 1) so the EDA tool can calculate the contributions from all of the various crosstalk paths. slide 8 - Post processing steps (new slide) This slide enumerates various post processing steps the EDA tool will have to do in the case of variant 2 to compute all the quantities that the model returns in the case of variant 1. Walter noted that, at first glance, he thought the steps listed for variant 2 might be double counting the upstream channel, but he needed to review them more carefully. slide 9 - comparison of the capabilities of the two variants (new slide) Fangyi noted that two things variant 1 supported that variant 2 didn't support were a Tx Init optimizing based on the downstream IR or any Init optimizing based on crosstalk terms. Walter mentioned that one other feature of variant 1 is that it provides some IP protection by not directly returning the equalization IR, but he wondered if the IP vendors thoughts this was a real benefit. He noted that one advantage of variant 2 is that it lets the EDA tool create a proxy GetWave for an Init-only model by using the IR of the equalization itself, which is returned by Init in variant 2. Arpad suggested this be added to the comparison table. Fangyi added a new row to the table "Allows EDA tool to construct GetWave (without DFE) for the model based on the Init output?" (variant 1 - no, variant 2 - yes). The comparison table indicates both variants support the GetWave flow and even support the case of models with a GetWave that depends on the input to Init, as long as "all models are Dual". Arpad noted Walter's previous statements that Init-only models were okay upstream, but every model downstream of the first model with a GetWave had to have a GetWave. Fangyi agreed with this statement, but said he wanted to avoid the complicated language and instead had gone with the simpler "all models are Dual". While reviewing the pros and cons, Walter and Fangyi came up with a hybrid of the two variants. This hybrid approach could be controlled by two new independent Reserved Parameters. One parameter (true/false) as Usage In would be used to select, or as Usage Info would indicate, whether the model takes an additional column in the IR matrix containing the cumulative upstream IR (variant 1). A second parameter (true/false) as Usage In would be used to select, or as Usage Info would indicate, whether the model returns an additional column at the end of the IR matrix containing the IR of the equalization itself (without DFE). If both parameters are true, the model provides the benefits of both variants, but the parameters could be used independently. If both parameters are false or don't exist, then the model supports the legacy flow. Fangyi took an AR to update the new comparison slides with information for this hybrid approach. Walter asked that we make sure to note that the Terminal Rx model need not be Dual, as it is reasonable to have a terminal Rx that does not want to return an IR from Init because it wouldn't include the DFE effects. Ambrish asked that we add a line noting that if all of the models are GetWave- only, then none of this is a problem. Arpad asked if we should start on a BIRD or wait until we see Fangyi's new slides next week. Fangyi suggested waiting until next week. Note: At last week's meeting Arpad suggested we hold off on posting the presentation slides and post them after Fangyi had added the comparison info. We will again hold off on posting and wait for next week when Fangyi has added the new hybrid variant information. - Ambrish: Motion to adjourn. - Walter: Second. - Arpad: Thank you all for joining. AR: Fangyi to add information on the new hybrid variant to his presentation. ------------- Next meeting: 19 January 2021 12:00pm PT ------------- IBIS Interconnect SPICE Wish List: 1) Simulator directives