IBIS Macromodel Task Group Meeting date: 09 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: - Chulsoon said he had prepared some slides and was ready to give an update on BIRD220, time permitting. ------------- Review of ARs: Kinger: Send his "SPIM transient support in IBIS" presentation 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 25th meeting. Kinger moved to approve the minutes. Dian seconded the motion. There were no objections. -------------- New Discussion: MIPI C-phy support in IBIS: Walter summarized the recent discussions. He said there are fundamentally two proposed approaches: 1. Treat it as 3 independent single-ended IBIS Txs and channels. Those outputs go into 3 differential Rx models to analyze the system. It's more work for the EDA tool to set it up that way, but the AMI standard does not need to change. 2. Alaeddin's proposal is to combine the 3 Rxs into one single C-phy receiver with 3 inputs. We could define the impulse responses and the impulse matrix in such a way that the AMI function signatures would not have to change. However, this approach would require different ways of designing the models and interpreting the data. Fangyi and Arpad agreed with Walter's summary. Fangyi said he'd been discussing options with Sai at Intel. Arpad said that if we had a combined 3-input Rx, then the model could take care of finding the trigger points internally. If you have 3 regular differential Rxs, then they could not output clock times directly. The tool would have to do it because C-phy bases the trigger times on the first zero crossing to occur amongst the 3 waveforms. With 3 independent models, none of them would be aware of the other two waveforms' crossing points. Fangyi agreed and said C-phy specifies how the trigger points are generated, so the tools could do it in the individual model scenario. Fangyi shared a slide with a block diagram representing the independent model scenario. It included a 6-port channel model and 3 differential Rx models. Fangyi noted that for the analog channel connections of the 3 Rx models, the positive pins would be connected to Za, Zb, Zc respectively, and the 3 inverse pins would be open (high impedance). He said the tool would find the earliest zero crossing time of any of the 3 signals, and that would be the clock tick. Ambrish asked about channel characterization in this scenario. He asked whether IBIS Open Source or Open drain type models would be sufficient for the Txs. Arpad said IBIS would at least need to define some way to tell the tool which 3 buffers form a C-phy group. Fangyi said channel characterization gets a bit tricky. He said that if you look at the AB pair into the first differential Rx, the tool would have to treat the AB pair on the Tx side as if it were a differential input for the purposes of channel characterization. Ambrish asked what we do with line C. Fangyi said C would be treated like crosstalk for the purposes of the AB pair. Walter questioned whether the Rx side was really implemented in Silicon as 3 differential Rxs as is shown in the block diagrams. He asked whether it was actually implemented as a single latch with 3 inputs. There is a single clock, so is this implemented as a single device in the Silicon? Fangyi said he had asked Intel similar questions. How would these 3 Rx AMI models need to communicate to emulate the actual device? He said he was waiting for more information from Intel. Arpad and Fangyi said the question is whether 3 separate models can account for everything that might be going on in the real device. As an example, Fangyi asked what if they wanted to implement some sort of echo cancellation? He said modeling that would be a challenge with 3 separate models. Fangyi said he was discussing these types of questions with Intel. Arpad agreed that there might be scenarios in which a single combined model would be necessary. Walter asked whether anything prevented users and tools from implementing the three model approach shown in the slide right now? Ambrish expressed concern about how we would future proof the spec if we were to add new keywords, etc., to accommodate C-phy. What would we do when the next C-phy variant comes along, change the IBIS specification again? Arpad noted that C-phy could be considered a subset of chord signaling. As far as possible, we should try to find the most general way to support C-phy. Fangyi shared a second slide in which the 3 Rx models were shown as single-ended and connected to A, B, and C, respectively. He said this illustrated the analog channel connections, but in practice the tool would pass a difference waveform to the "single-ended" Rx models (for example, A-B would be passed to the Rx connected to line A). Arpad said one concern he had about the configurations shown in the two slides was the fact that in C-phy the 3 terminating impedances are not straight to ground. They are terminated to a common node, which connects to ground via a capacitor. Fangyi said the single node is similar to a ground plane. Arpad said that since the 3 lines are never all driving in the same direction, they would tend to balance each other out at the common node. We would get improperly shifted DC levels if we connected the terminating resistors straight to ground, and this would affect the zero crossing locations. Arpad said one approach might be to use an IBIS interconnect model and include the 3 lines in the on-die interconnect. He said this might seem like a hack though, so we might want a different solution. Fangyi said we would have to think more about this termination issue. Randy noted that in the 3 single-ended Rx scenario, the tool would have to know what Rx buffers are A, B, and C so it could compute the difference waveforms to pass to the Rx models. Fangyi and Ambrish said that's a tool issue, but Arpad and Randy said that at a minimum we'd need something to tell the tool which buffers make up a C-phy group. Walter agreed that we could add a C-phy group indicator similar to [Diff Pin], but he also agreed with Ambrish that this set up could be left entirely to the tool. Arpad and Ambrish raised concerns about the Tx and the channel characterization. Ambrish noted that the C-phy slides shared by Arpad had shown different slew rates for different Tx transitions. He said if we approach this with the same single lowest-to-highest transition impulse response approach, we miss all of the differing edge rate effects. Walter said that these are high frequency effects, and by the time you get through the package all the edge rate differences have been filtered out anyway. Ambrish said that if Walter's assertion is correct, and we don't need to worry about different edge rates for different transitions, then we don't have to change anything in IBIS to support C-phy. Arpad asked whether we had any consensus on what to do. Walter said that Alaeddin thought there was a compelling reason to introduce a new combined 3-input Rx model. He suggested we wait until Alaeddin can attend another meeting and convince the group that it is necessary. Otherwise, we may not need to do anything. Walter moved to table the C-phy topic until Alaeddin and Sai can get back to us. Fangyi seconded. There were no objections. T-coil in [C Comp Model]: Walter said he could do an introductory description at some point, and then he expected we might table the T-coil discussion. BIRD220: Arpad noted that we could take up the BIRD220 discussion at the beginning of the next meeting. Chulsoon agreed that he would be ready to present then. - Curtis: Motion to adjourn. - Ambrish: Second. - Arpad: Thank you all for joining. New ARs: None. ------------- Next meeting: 16 July 2024 12:00pm PT ------------- IBIS Interconnect SPICE Wish List: 1) Simulator directives