====================================================================== IBIS INTERCONNECT TASK GROUP http://www.ibis.org/interconnect_wip/ Mailing list: ibis-interconnect@freelists.org Archives at http://www.freelists.org/archive/ibis-interconn/ ====================================================================== Attendees from March 16 Meeting (* means attended at least using audio) ANSYS Curtis Clark* Cadence Design Systems Bradley Brim* Cisco David Siadat Intel Corp. Michael Mirmak* Keysight Technologies Radek Biernacki*, Ming Yan* Mentor Graphics Arpad Muranyi* Micron Technology Justin Butterfield*, Randy Wolff* SAE ITC Maureen Lemankiewicz, Logen Johnson Signal Integrity Software Walter Katz*, Mike LaBonte* Teraspeed Labs Bob Ross* University of Aveiro in Portugal Wael Dghais Michael Mirmak convened the meeting. No patents were declared. Michael called for comments on the minutes of the March 9 meeting. Radek Biernacki noted that the minutes contained a few imperfections and inaccuracies but the gist was correct. For example, the word “ports” was used instead of “pins” in a few places. Radek moved to approve the minutes of the March 9 meeting. Mike LaBonte seconded the motion. The minutes were approved without objection. No opens were raised. Brad Brim asked about the intent of the Touchstone shortcut. He suggested tabling it, though Walter Katz was not present to provide his perspective. Brad asked Mike about Walter’s intent. Brad reported that he had not asked Walter offline regarding the shortcut, and asked the team whether spending more time on the shortcut was worthwhile. Radek stated that, if we don’t want to put restrictions in place, including Touchstone functionality, instead of using an IBIS-ISS subcircuit, he had no objections to tabling the proposal. He suggested not using as the default global N+1 terminal referencing, and not using 2N terminals. Bob Ross stated that the issue is whether to include the Touchstone shortcut, and that the fundamental issues regarding N terminals, N+1 terminals and reference, etc. still need to be settled. Mike noted that, on latter point, Radek’s concern makes sense; the treatment should be consistent with a single-S-element implementation of a shortcut. We don’t want something hard to implement. Bob stated that every situation discussed can be implemented in IBIS-ISS. Brad agreed that the discussion is whether to table the proposal; he would call it a definitional issue. If you don’t hook up the N+1 or the 2N terminals, they will float; 99% of those who float them are doing something they do not intend. Restrictions are not for convenience, but instead are implemented because the majority of files are of specific forms. Inside an IBIS-ISS subcircuit you don’t even have to wire up the reference node. For the model-maker, the shortcut eliminates having the wrapper around the S-element. Brad added that, if we don’t have a rule about connecting the reference, do we just assume global ground? Bob replied that we can state this as a consequence. The N+1 approach allows you to declare a reference node. We might add a rule prohibiting reference node to be one of the existing ports. Radek recalled Arpad’s presentation regarding using a positive pin as a reference node on a port. The use of an isolated node as a reference might be acceptable, but this depends on the model maker. If the node is *not* isolated, this would be a mistake. If we want to cover all the cases, we must face the consequences, in several specific cases: N nodes -> N+1st node is global ground, and is part of the topology 2N nodes -> we don’t gain anything; instead we get underdetermined system of equations and an incomplete circuit description; some information will be generated differently between simulators Radek added that he would prefer the N+1 node approach be explicitly stated, and the node be available for connection (including local ground). Brad asked whether hooking this up should be required. Radek suggested yes. Brad replied that Walter has a concept of unconnected terminations being automatically connected to a reference. Radek replied that he was opposed to this, because if a circuit is open, then is should remain open. A reference impedance termination is very convenient for the EDA tool, but the tool should not look for the easy way for itself. Reference impedance is a mathematical concept; it does not signify a termination. This approach is not that reliable unless the data is never changed. Brad added that he does not really see a huge value to avoiding a wrapper around a single S-element. This is not much work. As for termination by reference impedance, on the other hand, it’s easy to say that, for a 20 node netlist, one can hook up only 10 of them and leave the rest to be automatically connected. But 99% of what we get are signal lines. You don’t know the reference impedance without looking inside the IBIS-ISS. That’s probably the biggest convenience of using the shortcut. Brad agreed with Radek that the reference impedance itself isn’t meaningful. Bob responded that IBIS-ISS has a default termination of “open” if the termination is not explicitly connected. He detailed two issues: 1) Technical – the issue is whether we use N+1 terminals as a shortcut for IBIS interconnect; whether we explicitly terminate the reference terminal as the N+1 terminal. If you are given an S-parameter with the assumption of a reference impedance of global ground, definite vs. indefinite matrices are an issue. Brad replied that the S-parameter data is the same if you have N nodes and global ground, N+1 nodes or 2N nodes. Netlisting is the only thing that changes. Radek recommended avoiding the 2N case. 2) For the second issue, Bob referred to Appendix B of the Hall & Heck book, where 4-port transformations are described. He supports a global ground assumption and statements by Brad. Brad noted that the authors assumed a global ground. Industrial practice is to connect a 4-port as 3-port-plus-ground. Traditionally we have an even number of ports for inputs and outputs. So long as we have N+1, with N terminal connections, it would work. Radek related an example: imagine you have a 2-terminal RL network, which is a two-port S-parameter data with respect to a floating node as reference. Vladimir Dmitriev-Zdorov described this elsewhere; the restriction is that any current flowing in will be the current flowing out. For a two-port network, this is not required. Walter replied that there’s always a capacitance physically, even in such a network. The example is invalid. Radek responded that the floating node is truly unconnected, and there is no return current. You can view currents in and out of floating node as zero. You can netlist this data with three nodes, per Arpad. Capacitance is irrelevant here. The circuit can be netlisted as 1, 2, 2; 1, 2, 1; or 1, 2 0 and all be equivalent; a 1,2,0 arrangement will only work if the netlist uses a truly isolated node. This is correct from circuit theoretical perspective. Walter replied that this is not true in reality. There is a capacitance, a t-line, and therefore there is a reference node. Brad suggested you can manipulate the data and convert the data to an S2P file; he also agreed with Walter that it’s meaningless. One can take an S1P data set and netlist it with nodes 1 and 2, and use a third node as part of the subcircuit definition. Radek replied that the data does not support what you expect to see. Brad responded you don’t support KCL if you leave the terminal floating. Bob stated that S-parameters are models of something; forgetting relativistic effects, assume we are modeling a simple series inductor, this can be inserted using IBIS-ISS as a two-node device. This can also be modeled as a two-port, with assumed global ground. Walter replied that S-parameters do not assume global ground. This circuit can be described with a return terminal for every port. Radek disagreed – you cannot hook it up any way you want. Consider a 1-port between 1 and 2. Bob replied that one can transform this to a 2-port. Radek stated that one needs N+1 terminals for N-port data. Arpad replied that Radek is correct on N+1 terminals vs. N-port data. However, he asked whether N pins means N terminals plus reference, or should this be described by terminal-1 ports? Should a 6-pin model use a 5-port approach or a 6-port-plus-one? Michael asked whether the shortcut leaves its interpretation up to the EDA tool, but IBIS-ISS leaves it up to model-maker. Radek replied that even what’s inside the IBIS-ISS circuit is up to the model maker. Bob stated that terminal N+1 shall be connected to a POWER or GND pin, per the proposed rules; the text could also add “or left unconnected”. He added that he is very uncomfortable with absolute rules about reducing 6-port to 5 connections with reference. He additionally noted that he doesn’t see odd-port S-parameters today. Walter suggested that the EDA tool can select to hook up to ideal node 0. Arpad replied that the terminal rules in Draft 30 are very specific about connections. Early in the conversation, Walter explained how the measurements are made, and six-port models are shown with six terminals. Walter stated that the number of ports and connections tend to be equal. Arpad noted that draft text does not say this. Michael had to leave the meeting. Mike continued running the meeting. Arpad showed draft 30 of the Interconnect BIRD, and read a sentence on page 11 saying the number of terminals shall be N+1. Arpad stated that he felt N was the number of pins. Walter stated N was the number of ports. Bob asked if a port could be hooked up to the reference node. Radek said this could be done only if the data was generated that way. The discussion will continue in the next meeting. Brad moved to adjourn. Arpad seconded the motion. The meeting adjourned.