IBIS INTERCONNECT TASK GROUP Mailing list: ibis-interconnect@freelists.org ============================================================================== Attendees from June 10, 2026 Meeting (* means attended at least using audio) Chipletz Stephen Newberry Intel Corp. Michael Mirmak* MathWorks Walter Katz* Siemens EDA Weston Beal*, Arpad Muranyi*, Randy Wolff Synopsys Ted Mido, Edna Moreno ============================================================================== Michael Mirmak called the meeting to order. No patents were declared. During the AR review, Michael noted that Randy Wolff had completed his AR to provide Example 14 updated text. Arpad Muranyi asked when Touchstone 3 would be released. Weston Beal suggested asking people to review it at the most useful time, before release. Arpad suggested collecting feedback on the Port Mapping proposal from the IBIS Summit. Does the new text have all the features needed by industry? Michael replied that he could provide no prediction as to the completion time of the Port Mapping proposal. Arpad agreed, adding that the team kept "catching things" to change. Michael replied that, per Weston's suggestion, having more reviewers means more chances to catch things. Arpad presented his findings on extended net definitions and EMD. The problematic text appears on p. 388 of IBIS 8.0. Arpad cited and showed a memory chip example, which was forced to connect an EMD model and 3 Designators by using the same net name. He noted the IBIS 8.0 text: "I/O Pins in the [EMD Pin List] and the Designator Pin List that have the same signal_name are considered 'connected' by the content of the [EMD Model]." The same page mentions "extended net" ("The model maker can choose to either..."). Arpad noted that Randy stated the network would identify individual net names separately. Walter Katz asked why the specification needed the concept of an extended net in EMD. Michael noted that signal interactions occur by area (crosstalk) or by connectivity; for the former case, EMD shouldn't care, but the latter is a problem for tracing where the signal is intended to go. Walter replied that all we have in these specifications is CAD nets connected between components. Arpad stated that, in EMD, we have the EMD pins and Designator pins; we have several columns in the latter, and one of the columns is the signal_name. The same is true for [EMD Pin List]. Walter stated that these are connected by copper. Arpad disagreed, stating that these are connected by the model; inside the box we don't know what is connected or not connected. Weston added that the extended net might be included inside the S-parameters or the EMD model. Walter maintained that these are still separate CAD nets. Weston observed that they may be connected by metal or by some component sitting between them. He suggested two conditions: 1) combine both in an extended net where the device is included in the net 2) create separate EMD mdoels for each CAD net, with a termination with its own designator Arpad replied that neither condition covered his problem situation. Michael stated that connectivity is up to the model maker, not to the tool at simulation time. Arpad stated that no EMD syntax exists to describe an extended path with different constituent path names. Weston replied that different signal names (net names) are not considered connected. If you have a Touchstone file, you can make assumptions based on electrical behaviors; you cannot do that by SPICE reliably. This is why the signal names need to be the same; you connect them logically. Walter added the caveat that they be considered connected if they are physically linked by metal or a low impedance at Nyquist frequency. Weston agreed with Walter about connectivity if there is a declared series element. Michael asked whether we need an "Xnet" column in the specification. Arpad stated yes but Weston suggested this overcomplicates things. Walter noted that, if your subcircuit is only interconnect, then everything is OK. If you have a series capacitor or resistor, it wouldn't be inside the subcircuit, it would be between them. Arpad asked for clarification as to what "them" is. Walter replied that, if SPICE or S-parameters, the components are just metal interconnect (Arpad added that this corresponded to the red dotted lines in his drawing). The problem is, if you allow series components (L, R, buffer, etc.) inside the SPICE or Touchstone structure, then you need an extended net. Arpad noted that the series switch is an IBIS element. Walter added that this is like a multiplexer with multiple extended nets. R, L, C are simple. Michael stated that an initial scan would work for Series Element connections, but would be very difficult for general SPICE or Touchstone files. Weston added that signal_name is the information you need. Walter stated that a new rule would be needed: one either cannot have series components inside an interconnect element (either an IBIS-ISS subcircuit or Touchstone) OR if there is one, you have to have a separate list of "extended nets". Weston advocated for using signal_name, which is not a CAD net or an Xnet (extended net). There can be unidentified electrical behavior in EMD models. Michael asked whether net name and signal name are not the same. Weston replied that signal_name is just a label within the EMD; this corresponds to the red dotted lines in Arpad's diagrams. Weston suggested that the customer who made the file didn't follow the rules. Arpad replied that the structure he created works in another tool. Weston replied that the specific tool mentioned has an enhancement we can consider. Walter noted that Arpad's drawing features a figure in the lower left that makes the most sense. Michael and Arpad suggested that a "termination component" was not defined to include series elements. They have no argument with Walter's logic. Walter stated that the specification doesn't support components that aren't instantiated through an IBIS file. We don't have in IBIS the capability to describe a VRM, for example. This is really an analog device; yet, we want to be able to generate a load. Arpad noted that the designator here can be another EMD file, not just an IBIS file (the specification supports such nesting up to five levels down). He advocated: 1) Applying existing rules strictly 2) Add an "xnet" column to EMD 3) Supporting termination components with connectivity checking, plus series elements Walter stated that bills of materials usually list a designator plus a corporate part number; this maps into a vendor part number (this represents an AI opportunity). Arpad stated that the specification needs a terminological explanation of "extended net" vs. CAD net. One cannot distinguish between extended and "regular" nets. Michael's clarification BIRD will handle this, but we are not yet clear on the definitions of signal_name and extended net. Further, the definition of a termination component doesn't mention series components. Walter suggested forgetting nested EMDs for a moment; parallel terminations do not combine CAD nets into extended nets. We should be considering a series switch as more of an analog component. Arpad suggested there are too many optional columns in IBIS already. Michael replied that such data could be part of a separate list. Arpad noted that the specification could have add data to bus_label for signals. Michael stated that the team needs to clarify what signal_name means using a clarification BIRD. Arpad replied that Port Mapping should allow automation of EMD generation. This is in part why there is a name/value pair called "net". Michael suggested that "net" is really extended net. Arpad advocated adding another name/value pair called "extended net". Michael asked about grouping the nets. In Port Mapping, "Port" number is the unit identifier. What is it for EMD? For IBIS Interconnect, it's the pin. Arpad asked whether "signal name" is the fundamental unit identifier. Michael asked whether signal_name and net name are different, suggesting this is partially why clarification is needed; the team is actually discussing signal_name vs. "wire name". He added that "net2" and "net3" are confusing in the diagram. The meeting adjourned.