IBIS Macromodel Task Group Meeting date: 30 July 2019 Members (asterisk for those attending): ANSYS: Dan Dvorscak Curtis Clark Cadence Design Systems: Ambrish Varma Ken Willis Intel: * Michael Mirmak Keysight Technologies: * Fangyi Rao Radek Biernacki Ming Yan Stephen Slater Maziar Farahmand Mentor, A Siemens Business: * Arpad Muranyi Micron Technology: Randy Wolff Justin Butterfield SiSoft (Mathworks): * Walter Katz * Mike LaBonte SPISim: * Wei-hsing Huang Teraspeed Labs: * Bob Ross The meeting was led by Arpad Muranyi. Mike LaBonte took the minutes. ------------------------------------------------------------------------------- Opens: - None ------------- Review of ARs: - Fangyi to create BIRD197.4 incorporating changes from his new BIRD and comments. - Done -------------------------- Call for patent disclosure: - None. ------------------------- Review of Meeting Minutes: Arpad asked for any comments or corrections to the minutes of the July 23 meeting. Mike LaBonte noted that the minutes incorrectly gave the meeting date as July 16. Michael Mirmak moved to approve the minutes. Bob Ross seconded the motion. There were no objections. ------------- New Discussion: Jitter HF/LF components BIRD draft: Michael Mirmak showed a draft BIRD "Separating High and Low Frequency Tx Random Jitter Parameters". Michael said channel amplification of jitter applied more to higher frequencies than lower frequencies. The BIRD would split jitter into high and low freq bins. The user would not have to re-budget jitter, moving some jitter to the Rx side. Tx_Rx would not be deprecated, and it would be synonymous with Tx_Rj_HF. Tx_Rj_LF would be new. The new parameters would not be legal before version 7.1. The cutoff frequency would not be specified, that left to the model maker. There had been suggestions of a jitter distribution function, but Michael felt that was not yet needed. Arpad Muranyi asked if Tx_Rj_HF would be required if Tx_Rj_LF was present. Michael said it should be, and it would be OK to have a rule that if Tx_Rj were present the others were not allowed. Arpad said we had been trying to avoid parameters with double meaning, preferring mutually exclusive use. Walter Katz likened the proposal to Rx_Noise and Rx_Gaussian_Noise, an "alias". He suggested having it that way for Tx_Rj and Tx_Rj_HF, keeping them synonymous. Michael suggested that the presence of Tx_Rj_LF would then require Tx_Rj or Tx_Rj_HF. Walter Katz asked if Tx_Rj_HF would adjust the stimulus going into the Rx. Michael Mirmak said they had statistical analysis in mind, and he needed to check with experts on their preference for bit by bit mode. In statistical mode a distinction between LF and HF required users to edit the Rx jitter to include Tx LF jitter. Walter asked if there was no Rx_Rj, and the Tx had Tx_Rj_LF, would the EDA tool act as if the Rx had that LF jitter? Michael said it would. Walter gave an example of how an EDA tool might handle jitter data, with picosecond values. Michael noted that the BIRD avoided specifying the exact math. Walter felt it would be helpful for EDA tools to have more detail on the requirement. Arpad agreed. Fangyi suggested giving an example of one way to do it, allowing tools to innovate. Another method would be to say it is statistically uncorrelated. Arpad noted that a CDR tracking out low frequency jitter might negate that component. Michael said that sounded like we would effectively be providing Tx_Rj_HF only. The allocation of the LF/HF would matter. He will ask others about that. Arpad said there could be limits to the ability of a CDR to follow the LF jitter. Michael suggested we might specify a frequency dependent sensitivity threshold. Fangyi asked if the Tx vendor would need to know how the Rx functions. Michael Mirmak said the Tx should not need to know details of the Rx, but certain assumptions would be needed. Walter Katz noted that Sj is similar, and generally it is an LF component. The shapes are different; Sj has no tail on the distribution curve. Had Sj been tried as an alternative means of representing LF jitter? Michael said he was working on getting that answer. His understanding was that Sj constituted a single tone. Walter said it amounted to a dual Dirac function at the Rx. The question was whether or not it was bounded. Sj was usually due to power supply noise, which usually is bounded. Standards for jitter specified tests that can be implemented by scopes, and standard distributions are usually applied. Arpad Muranyi said they measure frequency modulation and DCD, analyzing whether they were moving and with respect to what. Mike LaBonte asked if Tx modelers had to consider only the channel, or the Rx too. Michael Mirmak felt we might need to specify a distribution, not an unspecified cutoff point. The Tx should not have to know exactly what the Rx is. Arpad asked why the cutoff need not be known. Michael said the cutoff point should not make much difference. The idea was hat it would be unlikely for someone to connect Ethernet and PCIe devices together for example, so it was simpler to just say one bucket of jitter is affected by the channel and another bucket isn't. If the Rx must be considered, a distribution curve might be better. Fangyi asked if statistical analysis would not model the CDR, and be more affected by the jitter. Michael Mirmak said that could not be assumed, impacts on clock recovery could be modeled statistically. Digital signatures: Michael Mirmak said digitally signing DLLS was a good practice, but not something to be specified by IBIS. For security, users should be able to verify the source of a DLL. Arpad Muranyi said one question was who would be the gatekeeper. Would IBIS issue the signatures? Mike LaBonte said that as an unincorporated organization, the IBIS Open Forum might not have standing to procure a signing certificate. Also, IBIS would then become the issuer guaranteeing the origin of the models, It would be best for model suppliers to sign the models. Wei Hsing Huang described the certification process, which involves business verification. Walter suggested we could describe the steps for model makers to handle the digital signatures, to help them get started. Wei Hsing said web browsers had the ability do the checking when models were downloaded. Arpad asked if this should remain on the agenda. Michael said it could be removed from the agenda, but the topic would make a great summit presentation. PAM4 threshold consistency: Arpad Muranyi showed IBIS 7.0 page 255. He asked if we required associated PAM4 thresholds to all have the same Usage. He noted that Out and Info were both used in two examples, but the Usage was consistent within each example. Yet, this was not specified. Walter Katz suggested the PAM4_Upper_Threshold and PAM4_Lower_Threshold could be Out, for example, and the center could be Info because it would always be assumed to be zero. Arpad asked if a BIRD should be written. Walter suggested no BIRD was needed. Walter Katz moved to adjourn. Michael Mirmak seconded. Without objection the meeting adjourned. ------------- Next meeting: 06 August 2019 12:00pm PT ------------- IBIS Interconnect SPICE Wish List: 1) Simulator directives