IBIS Macromodel Task Group Meeting date: 16 July 2024 Members (asterisk for those attending): Achronix Semiconductor: Hansel Dsilva Amazon: John Yan ANSYS: * Curtis Clark * Wei-hsing Huang Aurora System: * Dian Yang Raj Raghuram Cadence Design Systems: * Ambrish Varma Jared James Dassault Systemes: Longfei Bai Google: Hanfeng Wang GaWon Kim Intel: * Michael Mirmak * Kinger Cai Chi-te Chen Liwei Zhao Alaeddin Aydiner Sai Zhou 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: - None. -------------------------- Call for patent disclosure: - None. ------------------------- Review of Meeting Minutes: Arpad asked for any comments or corrections to the minutes of the July 9th meeting. Dian moved to approve the minutes. Michael seconded the motion. There were no objections. -------------- New Discussion: BIRD220 update: Yifan reviewed a presentation on the latest work related to BIRD220. She first discussed some of the questions and feedback from IBIS ATM on BIRD220. 1. Will it affect [Composite Current]? No, this proposal does not alter any power-aware IBIS keywords. It is only focused on modifying the Ku(t) and Kd(t) scalers, which control the output current. 2. Will time-averaged supply voltage noise affect the driver delay? The results will be checked and verified against transistor-level models. Yifan also noted that the new proposed extraction process is simpler, and the new proposal can cover a wider range of supply voltage deviations. These changes also address feedback from ATM reviewers. Yifan reviewed the current limitations of IBIS. She said that even power-aware IBIS models can't reproduce the shifts in the rising/falling edges caused by variations in the power supply. IBIS's existing Ku(t) and Kd(t) scalers can reproduce the edge rate changes and DC level shifts of the output stage in response to supply variation, but they do not capture the edge shifts that occur in the pre-driver chain. The initial BIRD220 proposal introduced two additive terms to the existing Ku(t) and Kd(t) equations. These terms are linear and second order functions of the difference between the time averaged power rail voltage Vcc(t) and Vcc nominal. This approach had limitations and stability issues as the time averaged Vcc(t) deviated further from nominal. The linear and second order terms started to dominate the K(t) equations and introduce non-physical oscillations in the edges. Yifan introduced a proposed new approach that no longer uses the additive terms. Instead, the new approach directly computes a time shift to be applied to the existing Ku(t) and Kd(t). That is: Ku(t) = Kuo(t + delta V * DC_Jitter_Sensitivity) where Kuo(t) is the time varying coefficient for the nominal Vcc0 case, delta V = (time avg value of Vcc(t) since the last transition - Vcc0), DC_Jitter_Sensitivity = jitter sensitivity in s/V. In addition, to model non-linearities in DC Jitter Sensitivity as the range of voltage variations increases, Yifan proposed that the new sensitivity keyword(s) could contain a table of DC Jitter Sensitivity values at different voltage deviations from Vcc nominal. Arpad noted that the current BIRD220 provides a single value for DC Jitter Sensitivity (one for the rising edge and one for the falling edge), but the new proposal adds a table of values to address non-linearities. Yifan agreed. Randy noted that several years ago he had presented some data in ATM for a DDR5 device. He recalled that Jitter Sensitivity was reasonably linear over about 50 to 100mV. Outside that range it became more non-linear, but he said this is likely very dependent on the technology and also on the Vcc nominal at which you start. Randy suggested that the Jitter Sensitivity should probably be provided for the normal pvt (process, voltage, temperature) corners. Arpad suggested separate rising and falling keywords, each of which would contain 4 columns (voltage, typ, min, max) for the Jitter Sensitivity. He said this would be consistent with the I-V keywords, for example. Fangyi asked for confirmation that the A(t) and B(t) terms were no longer in the new proposal (A(t) and B(t) are the coefficients of the linear and second order additive terms). Chulsoon noted that equation was presented in the original BIRD220 to provide the derivation, but it was not really part of BIRD itself. He confirmed that those equations will not appear in the new proposal. Fangyi noted a non-monotonic dip in Ku(t) in the middle of the edge in some of the validation and comparison plots (slide 15). He asked whether that dip was physical. Yifan and Chulsoon noted that the dip appeared in the original Ku(t) itself, prior to the Jitter Sensitivity modifications. Ambrish asked whether both rising edge and falling edge tables were required. Yifan said they expected to require both. Chulsoon noted that if a buffer is truly symmetric then you could assume the falling edge is an inverted version of the rising edge. However, the buffers they had studied were not symmetric in this way. Chulsoon said we could consider an option to only provide one table, but he thought the model maker could simply duplicate the table and provide both. Kinger asked why the keyword specifically has "Pre-driver" in its name. Randy said this proposal is intended to capture delay effects introduced by the pre- driver. Yifan and Randy reviewed the "current IBIS limitations" slide and noted that slew rate changes and DC levels from the final stage were already captured by IBIS. Arpad noted that IBIS has I-V data for the output stage, and this can be scaled with respect to supply variations by the ISSO PU and ISSO PD keywords. That only changes the impedance (DC levels). The edge rates come from the V-t waveforms, but the V-t waveforms are used to compute multiplicative scalers to the I-V data and don't provide the edge's shift in time. Kinger asked why we need to worry about such large variations in the supply voltage. He said typically +-10% is all you see. Anything outside of that, the chip probably won't work anyway. Randy agreed but said the table allows the model maker to provide data over any range they want. Randy noted that ISSO tables also provide data over a very wide range, much of which will likely not ever be used. Chulsoon said it doesn't hurt anything to give the model maker the option to provide data over a wide range. He noted that the CMOS driver they had been studying is fairly linear, but he had seen papers on much more non-linear drivers. He said he thought it was better to cover a range to help model non-linear behavior. Kinger said that a table to help cover non-linear behavior was fine with him. He just wondered how wide a range we really needed to cover. Randy said the range over which the data is provided is entirely up to each model maker. IBIS doesn't have to specify any range. Randy and Arpad said this new proposal looked like a big improvement. Arpad asked what the next step would be. Chulsoon and Yifan said they could update the BIRD and prepare a BIRD220.1 draft for the next meeting. - Ambrish: Motion to adjourn. - Curtis: Second. - Arpad: Thank you all for joining. New ARs: Yifan: Prepare a draft of BIRD220.1 and send it to the ATM list. ------------- Next meeting: 23 July 2024 12:00pm PT ------------- IBIS Interconnect SPICE Wish List: 1) Simulator directives