IBIS Macromodel Task Group Meeting date: 20 June 2023 Members (asterisk for those attending): Achronix Semiconductor: Hansel Dsilva Amazon: John Yan ANSYS: * Curtis Clark * Wei-hsing Huang Aurora Systems: Dian Yang Cadence Design Systems: Ambrish Varma Jared James Google: Hanfeng Wang GaWon Kim Intel: * Michael Mirmak * Kinger Cai Chi-te Chen * Liwei Zhao Keysight Technologies: * Fangyi Rao Majid Ahadi Dolatsara Stephen Slater * Ming Yan Rui Yang Marvell: Steve Parker Mathworks (SiSoft): Walter Katz Graham Kus Micron Technology: * Justin Butterfield Missouri S&T: Chulsoon Hwang Yifan Ding Zhiping Yang Rivos: Yansheng Wang SAE ITC: Michael McNair Siemens EDA (Mentor): * Arpad Muranyi * Randy Wolff 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: Bob: Send information about the recent IEEE SPI IBIS Summit presentation on PSIJ to the ATM list. - Done. -------------------------- Call for patent disclosure: - None. ------------------------- Review of Meeting Minutes: Arpad asked for any comments or corrections to the minutes of the June 13th meeting. Michael moved to approve the minutes. Lance seconded the motion. There were no objections. -------------- New Discussion: PSIJ Sensitivity Discussion: Kinger shared draft11 of the proposal, in particular figure 4 and equation 8, in order to discuss issues raised by Fangyi in an email reply to draft11. Arpad asked whether the 3 VCC_m names shown in figure 4 represented 3 independent supplies. Kinger said they represent the same supply rail applied to 3 distinct blocks. Fangyi asked where PSIJ_VCC_m_CKTi was measured in figure 4. Fangyi asked whether the intent was to measure the jitter at the output of each of the cascaded blocks, and then the total jitter at the final output is some combination of them? In that case, Fangyi said he understood the intent. He said the measurement location(s) should be described in the BIRD. Kinger agreed. Fangyi said he thought "multiplication" was probably not the right term to use with regard to combining PSIJ_VCC_m_CKTi terms. He noted that equation 8 would have issues with the final units for PSIJ_VCC_m, given that it specified a possible mix of multiplication and addition of the individual terms. Referring to figure 5, he said the curves combined via "multiplication" looked more like typical filter behavior in which each of the cascaded stages had a unitless jitter transfer function applied to its input. Fangyi asked how we would use this information in a transient simulation. Kinger said the goal was to enable an evolution of high-level SI/PI cosimulation. Traditionally we totally separate SI and PI simulations and then use a simple peak to peak noise design target. More recently people had been attempting to do SI/PI cosimulation by simulating everything together, but this can be nearly impossible to do if we consider full SPICE models of each of the circuit blocks. He said that with this proposal we are thinking of a pure SI simulation with ideal voltage sources and then an independent simulation to determine the noise on each power rail considering all the circuit blocks. We convert each power rail's noise to a power spectral density, multiply with the PSIJ Sensitivity curve for that rail and a particular signal to get the jitter spectrum, and can then use IFFT to compute the peak-to-peak jitter distribution. This can then be applied to the results from the ideal SI simulation. This provides a holistic approach that can cover all supply rails and all circuit blocks. Kinger said he would set up an offline meeting with Fangyi to review his suggestions. Kinger briefly discussed the recent IBIS Summit presentation on PSIJ from the Indian Institute of Technology Jodhpur. He said he thought it was focused on a simulation algorithm for computing the jitter contribution of an individual circuit block, where Kinger's proposal is more of a high-level system approach. He said he did not think his proposal conflicted with the summit presentation or BIRD220. The group discussed Fangyi's questions about the presentation, which he had sent to the ATM list. Bob agreed to forward Fangyi's questions to the authors. AMI Support for [Test Data]/[Test Load]: Michael reported that, based on feedback from the previous week's discussion, he was planning on altering the draft to move any test load interconnect details over to the traditional [Test Load] keyword, and then he would create a new [AMI Test Data] keyword for AMI. Per the suggestions, [AMI Test Data] would only provide the channel impulse response to be passed to the AMI executable model by the tool. The impulse response waveform itself will be provided in a separate file, which is referred to by the [AMI Test Data] keyword. Arpad recalled that during the previous week's discussion Ambrish has suggested that if we provide the impulse response, then we can extend the use of the original [Test Load] to cover AMI. Michael said that if we want to make the existing [Test Load] more useful, we need to support IBIS Interconnect Model syntax to properly model modern high-speed lossy channels. He said that would decouple the issues of channel modeling and channel characterization from AMI. However, he noted that this would enable people to compare channel characterization results from different tools. So, it would not get us around the contentious issue from the previous meeting about comparing different tools' channel characterizations. It would only decouple the issue from AMI. Michael said one of the reasons that his proposal originally provided a more detailed channel model, and not the impulse response itself, was a concern for the model maker's ability to provide the impulse response itself. He said he was concerned that tools might not make their computed channel impulse response readily available to the user (the model maker in this case). So, a model maker might not have an easy way to capture an impulse response waveform if they had to provide one in the [AMI Test Data]. Arpad recalled that he had previously suggested that instead of inventing a new new syntax to provide the AMI parameter settings, we could instead specify the exact parameter string itself. Michael said his concern here was similar to the concern about the impulse response. He was worried about expecting the model maker to create the fully formed parameter string from scratch. He said the BIRD would have to encourage tools to make their generated parameters strings readily accessible to the user. This would allow model makers to capture the generated string and include it in their model. Wei-hsing asked whether model makers should take that up with their chosen tool vendors if they needed easier access to impulse responses, parameter strings, etc., that had been generated by the tool. Michael said this was a good point, but he said he generally considered IBIS to have three main audiences: the end user doing system simulation, the model creators, and the EDA tool vendors. He said discussions in ATM were often biased toward the EDA tool vendor audience. To facilitate the adoption of this feature by model makers, we would need to help ensure that the parameter strings and impulse response waveforms generated by the tools were more accessible to users. - Curtis: Motion to adjourn. - Michael: Second. - Arpad: Thank you all for joining. New ARs: Kinger: Work offline with Fangyi to address Fangyi's suggestions on the PSIJ Sensitivity draft11. Bob: Forward Fangyi's questions to the authors of the IEEE SPI IBIS Summit presentation on PSIJ. ------------- Next meeting: 27 Jun 2023 12:00pm PT ------------- IBIS Interconnect SPICE Wish List: 1) Simulator directives