IBIS Macromodel Task Group Meeting date: 23 March 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 updated Merged_166_210 presentation to the ATM list. - Done. - Walter to draft a BIRD210.1 proposal. - Walter reported he has a new BIRD proposal, which we may want to call as BIRD210.1 or BIRD211 depending on the outcome of this meeting. -------------------------- Call for patent disclosure: - None. ------------------------- Review of Meeting Minutes: Arpad asked for any comments or corrections to the minutes of the March 16th meeting. Michael moved to approve the minutes. Ambrish seconded the motion. There were no objections. ------------- New Discussion: BIRD 210 Redriver Flow Issues: Walter stated the existing redriver flow is fundamentally flawed. This BIRD proposes to replace the flow where the Tx2 has everything upstream but does not optimize itself. The Rx2 ends up having everything upstream. A new parameter called Tx_Requires_Downstream_Channel is added. If this parameter is true, there is an additional column for the downstream channel in the impulse matrix. The Tx2 uses the downstream channel in this flow. David asked about the call to the Tx_Init function and how it works if the driver is part of the same redriver. Walter replied they have different AMI DLLs for the Rx and Tx. Arpad asked if we have a case where both Tx and Rx are in the same DLL. Walter replied that you could do this if you have a pass through for the Tx. Fangyi asked if Tx2 in the modified flow now has the Rx1 output. He also asked, in time-domain when Tx2 is Init only, would this work for GetWave. Walter replied the GetWave flow does not change. We need to include the Tx filter either by deconvolution or by a unit impulse response. The time-domain waveform can be convolved with the unit impulse response from the Tx. Fangyi asked, regarding the Rx2 as Init only, if it would have the same problem. What happens if Rx2 is trying to optimize crosstalk. Walter replied, if you want to do time-domain simulation in this case, you would have to add the unit impulse response. Fangyi stated he does see cases where the crosstalk is considered in the Rx, where the Rx needs to cancel the crosstalk from its own Tx. Walter commented this is a much simpler solution than BIRD210. Crosstalk cancellation is a very interesting problem. It is important to ignore the unit impulse response in the matrix. Walter noted the BIRD describes the new flow shown in the diagrams. The model modifies the through channel and the crosstalk channels, but the additional column for the downstream impulse response is not modified. The BIRD contains an example. Both statistical and time-domain flows are detailed in the BIRD. Both retimer and redriver flows are considered. The EDA tool will have to change how they handle the Tx. The Tx AMI models will have to change if there is a need optimize based on the downstream channel. Walter would like to replace BIRD210, but we can vote to decide which approach to take. Fangyi commented this BIRD leaves the crosstalk issue unaddressed. Walter asked why it is unaddressed. If the Rx is not Init only, then the EDA tool can add the additional column. The model will have to ignore the additional column. Fangyi replied the model will have to guess which lines to cancel the crosstalk. Walter stated the user or the EDA tool must tell the model which crosstalk lines to cancel. Radek stated there is no way for the model to tell which lines are the crosstalk. Walter commented the model can tell from the data. Radek commented the BIRD210 approach is very precise in this regard. Walter will send this document out as BIRD211, then we can debate which to submit to the Open Forum. Arpad asked for feedback from model makers on which direction they prefer. Michael stated he would need to study the issue, but it is difficult to tell without models actually using these new flows. Fangyi gave an example of an optical channel, where the prediver and VCSEL act as redrivers. The current is used to adjust the slew rate in the VCSEL. Walter asked if there is any feedback in this system or if it is adapted in simulation. Fangyi replied there is some training but he did not know the details. Walter stated this may be a case where we need Tx optimization. David asked, regarding the use of AMI in photonics, if the proposed changes fix all the issues or if there are other issues. He was wondering how much trouble lies ahead for modeling photonic links with AMI. Fangyi replied one of the reasons for the redriver flow is to model photonic link systems. Arpad asked if the redriver flow changes will be enough. Fangyi was not sure on this, but AMI is the right direction. Walter commented it is better for the model to adapt itself rather than do an exhaustive sweep. Fangyi gave another example of a full duplexed channel. One way to increase the bandwidth of the channel is by both driving and receiving at the same time. Crosstalk is received directly from the Tx. The Rx can use the pattern from the Tx to cancel the crosstalk. Walter asked, in this case, what the impulse response is and if there is another column for the other aggressor. Fangyi replied, in practice, the Rx model can determine from the amplitude which is crosstalk. Ambrish asked if the current flow can support this. Fangyi replied the strongest crosstalk will be from its own driver. Walter suggested to put the burden on these model makers. Walter will send out the BIRD draft. Arpad asked if Walter will submit this to the Open Forum. Walter will submit the BIRD to the open forum. Arpad said it may come back to ATM to make a recommendation. Walter has a PAMN BIRD he would like to discuss next time. - Walter: Motion to adjourn. - Michael: Second. - Arpad: Thank you all for joining. AR: Walter will send out BIRD211. ------------- Next meeting: 30 March 2021 12:00pm PT ------------- IBIS Interconnect SPICE Wish List: 1) Simulator directives