=============================================================================== IBIS INTERCONNECT TASK GROUP Mailing list: ibis-interconnect@freelists.org ================================================================================ Attendees from October 23, 2024 Meeting (* means attended at least using audio) ANSYS Curtis Clark, Juliano Mologni Arista Networks Jim Antonellis Broadcom James Church Intel Corp. Michael Mirmak*, Xiaoning Ye Keysight Technologies Ming Yan Marvell Steve Parker MathWorks Walter Katz* Micron Technology Justin Butterfield Siemens EDA Weston Beal*, Arpad Muranyi*, Randy Wolff* Simberian Yuriy Shlepnev* ST Microelectronics Aurora Sanna Synopsys Ted Mido, Edna Moreno University of Illinois Jose Schutt-Aine Zuken USA Lance Wang Michael Mirmak called the meeting to order. No patents were declared. During the review of the October 2 minutes, Michael noted that Jim Antonellis was incorrectly listed as a participant in the October 2 meeting. Arpad Muranyi moved to approve the minutes with this change; Weston Beal seconded. The minutes of the October 2 meeting were approved without objections. Michael noted that the October 16 meeting minutes were incorrectly identified as from October 17. Randy Wolff moved to approve the October 16 minutes with a corrected date; Arpad seconded. The minutes were approved with this change with no objections. Michael noted that he has a standing, open AR to clarify the language in the Touchstone proposal that there is no support for port order significance; ports are only ordered numerically. Michael noted that his AR to add Arpad's examples to the draft proposal has been closed with the latest document posted to the reflector. Michael summarized the Touchstone referencing issue: how to reconcile local probing "ground" for measurement with need for multiple references for multi-port systems, including systems which had been modeled or measured using different references far away from each other. This is a need for EMD and IBIS Interconnect, where package, die substrates, and buffers may be characterized using very different referencing schemes. Some implication exists that power-aware analyses require "losing" a reference location. Arpad added that some parts of IBIS only support one reference node for all ports, but some descriptions feature a single reference for each port. Walter Katz noted that all EDA tools for SI and interconnect assume node 0 is the reference node for all measurements. Putting into [Pin Mapping] a reference location is a convenience to the person generating or using the S-parameters, to see what point was the reference point of the probe. But always assume that it's a ground node (physical, like VSS) which is close to the measurement point. Arpad showed his IBIS EMD package interfacing drawings, stating that his preference is for Touchstone to use a node-zero-independent way of referencing, but recognizing that measurements using such a scheme may be impossible. Yuriy answered that he thought this discussion was resolved some time ago, in the early 2000s. Multi-port systems are usually defined with respect to a local ground. Some voltage at the port is measured in or out. Currents go in at measurement nodes and out from reference terminals. We don't know the potential at the reference node. For multi-port, we lose the information about what is happening on the reference node. This means that we can make the reference node whatever we want, if we do not violate the principle of current in = current out for a port (node,reference pair). We can set the reference to "zero potential" with no violation of this assumption. On possible conversion: de-referencing the "reference" node requires some other reference. This is impossible in measurement, and impossible in simulation. Using a "virtual ground" changes the problem, and changes the solution. Close to DC, anything is possible. When you take 2-port S-parameterss and turn them into a 4-node circuit, it's technically useless and can lead to a violation. We lose information about the ground. Consider power-delivery systems in IBIS: two voltage rails plus a signal. If you define ports for power-distribution, they are defined locally. The signal path has a completely different reference definition; both planes are references for the signal. The change in reference happens at the vertical transitions, and only there. You don't need to de- reference anything, or create additional nodes. Read the specifications, as they might be useful in the case of customers creating package models where references are not clear. One must match the connector footprint, as the references have to be physically consistent. Walter suggested that, for this case, with separate companies, what is the location of the probe X,Y position vs. the board? Yuriy asked how the models are created. Where is the port located? Walter stated that the reference is not used, except when mating connectors together. Yuriy replied that only one signal and one reference exists that you cannot de-reference; two conductors, each with some resistance. You can compute Y- or S-networks of admittances, both are sum of impedances or resistances. References become resistance-less. The reference can assume any potential; zero is best to use. Arpad asked where in a 6-port would you put the references? Weston requested the clarification: is this in simulation or the laboratory? On the U-element case, Walter stated that the "ref_in" label is a name and a node, not a reference. Yuriy added that, if you can conduct some measurements, that's OK. Whatever path you create, the probe ports are so small, they have to be conservative; voltages at the tips do not depend on the paths between the tips. In reality, you cannot measure it with respect to some ideal ground. You instead use Vss or Vdd; it doesn't matter. The signal will be referenced to both. It's a transformation. Weston asked if we created S-parameterss with 6-ports, could we connect to an IBIS Rx and Tx, where the voltages are between ports? Yuriy replied, yes, whatever it useful. This is a DC approximation; for HF, the number of ports have to be sufficient to solve the problem. Michael summarized this as meaning local references are preferable for each sub-system, but that the refernece points must be at the transition locations for the system to be consistent. Walter replied that he was not sure about this Yuriy stated that one can reference to VDD. In this case, to get signals referenced to VDD, you measure the signal on one side with Vss reference, and the other side with Vdd reference. But you cannot connect them because you lose information. One cannot reclaim this information; here, you need a true reference ground with no ambiguities. Arpad summarized a 5-port vs. 6-port example; in circuit simulation you can do whatever you like. Weston added that, comparing Vss-in vs. Vss-out, with each case using a virtual pin with reference to each side, the references are entirely separate (a node 0 does not exist in this case). Yuriy replied that the setup depends on what is convenient; Vdd, Vss are all distributed planes. In high-speed, signals are pretty much detached from packages today. He added that he would have to think about it further. Michael noted that this case was highly similar to that of a transformer with no ideal DC path to node zero on either side. Yuriy added that all information about referencing is lost in multi-port simulations. Contradictions show up always when performing DC analysis. Any de-referencing creates a dead-end, as you are dealing with distributed systems. One has to separate circuit domain and distributed domain (or multi-port domain); in circuit domain, you reference everything to an ideal location. In distributed domain, how ports are created is important. This effort looks like you are trying to come up with a universal system for both. As a result, you will have problems. Yuriy noted that he could present a few slides to show how this would be done, adding that a power-deliver network is referenced differently to other devices. Interactions only take place at transitions. Circuit models are missing this part, which is why we have contradictions. There is no loop. Interactions are between signal and power distribution at specific locations; the references change at those specific locations. Michael thanked Yuriy for his participation and very active assistance. Arpad moved to adjourn; Randy seconded. The meeting adjourned without objection. The next meeting will be held October 30, 2024. ================================================================================ Bin List: 1) Complete port naming proposal (Katz et al) 2) Complete/revise Touchstone 3.0 draft outline (Mirmak) 3) Complete ISS-IRD 1 Draft - enable cascading of S-parameters through W-element (Mirmak) - TABLED Tabled ARs: - Arpad to give an example of the physical connectivity needed for EMD automation.