================================================================================ 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 7, 2018 Meeting (* means attended at least using audio) ANSYS Curtis Clark* Cadence Design Systems Bradley Brim Intel Corp. Michael Mirmak Keysight Technologies Radek Biernacki Mentor, A Siemens Business Arpad Muranyi* Micron Technology Justin Butterfield*, Randy Wolff* SiSoft Walter Katz*, Mike LaBonte* Teraspeed Labs Bob Ross* Mike LaBonte convened the meeting. No patents were declared. Justin Butterfield took minutes. Review of Minutes: - Mike called for review of the minutes from the March 2 meeting. Randy Wolff moved to approve the minutes. Mike LaBonte seconded. The minutes were approved without objection. Review of ARs: - Mike and Arpad Muranyi to work on refining the A_gnd warning text offline. - Mike note that Arpad took the lead on this. Arpad reported this is incorporated in BIRD189.5_draft18_v3. - Bob Ross to take the BIRD189.5_draft18_v2 and make his proposed changes to the rail rules. - Bob reported that this is done in BIRD189.5_draft18_v3. - Radek Biernacki to draft text for referencing in BIRD158. - Mike noted that this was discussed in the ATM meeting. Opens: - None. Rail Rule Relaxation: Bob shared BIRD189.5_draft18_v3 that he modified to relax the rail rules. Mike asked if we could take a look at the language in the BIRD regarding these rules. Bob stated, on page 9, he added a rule that PDN models can connect to only one interface. Under the rail terminal rules, he eliminated the requirement that a rail pin_name cannot be duplicated across more than one PDN Interconnect Model. Mike asked if one of the Interconnect Models could be the through path and the other could be for the decoupling model. Bob stated that this is possible, and we would allow models without the through path. Mike asked if there could be two through paths in parallel. Bob stated that this is not prevented with his proposal. A terminal for PDN structures can be repeated in other PDN structures. The pin might be the same signal_name. He noted PDN models might have separate rail routing, as we don't require shorting rails at the buffer side. Bob also added "pin only", "die pad only" and "buffer only" to the list in the third bullet under Rail terminal rules. He also added a statement that within a group we have an assumption of terminal connectivity. Mike asked the group about the sentence "A rail terminal may be in the Interconnect Models with the following interface combinations" and if this sentence needs reworded to consider the path. Bob thought that we can wordsmith this later. Walter Katz stated that he does not agree with the changes Bob is proposing. Walter moved to table the rail rules discussion. Randy noted that Bob's proposal would be useful for on die decoupling capacitor. He can modify these models easily to work with the existing syntax, but he would like to know if there are other examples where this capability would be useful. Arpad stated that he would like to better understand the parser implications of these rules. He commented that the rules could be relaxed later. Arpad stated he is leaning toward tabling this topic, as it might slow the discussion. Bob thought we should keep these changes in draft, and he disagreed that we could relax the rules later. There was no second to the motion, and Mike declare the motion dead. Bob gave a subcircuit analogy where the Interconnect Model Group connects various PDN and decoupling models. He would like to use common node names in the Interconnect Model, but this is not possible with the current rules. Bob commented that there is no technical reason for not including this capability. He stated that relaxing the rules would allow the Interconnect Models to better follow how spice works. Randy asked if Case A works in the existing syntax, while Case B makes connections by common terminal names, but this is not permitted with the existing rules. Bob noted that the parser would reject Case B with the current rules, but Case B makes more sense for connecting spice models. Arpad showed a decoupling cap model example comparing a series and shunt approach. The shunt approach is illegal because only one interface is used. Arpad asked how the decoupling would work in an IO model with both signal and PDN but the PDN's terminals only touch one interface. Randy asked if there is anything else broken and if putting more instances connecting the same terminals is useful. He also wondered if you have overlapping terminals how this would work. Bob noted he uses bus_label to handle the overlapping terminals. Arpad asked why you would want to do this. Randy thought that he would probably include a subcircuit that includes everything wrapped together. Bob commented that there is no reason for restricting how rails are connected. Unlike I/Os, we do not necessarily expect full paths for rails. Model makers would not be forced to wrap their existing models to work with the new syntax. Mike thought that we have a lot of rules in BIRD189, and we could always lift restrictions later. He noted part of the reason for the rules was to restrict branching and forking. Bob commented that this does not apply to rails. Walter stated, with respect to these rules and the algorithms to check them, these rules are easy to check and parse. Walter will supply pseudocode for the parser development [AR]. Walter also noted that, as soon as, you have two models in the same Group that connect the same pins, it becomes incredibly difficult to deal with. Creating the PDN netlist becomes more difficult. Mike asked if the pseudocode would be helpful for the parser developer. Bob stated that psudeocode is informal. Walter asked if there is any limitation that prevents model makers from the capability that they need. Bob replied the issue is that we can create subcircuits that cannot be used. Curtis Clark commented, in the current syntax, we are not trying to consider all spice cases. The current rules would be a good compromise between what tool vendors can parse and what model makers can provide. Mike asked if will discuss A_gnd in ATM. Arpad replied that he would like to discuss his BIRD158 change. Next Meeting: The next meeting will be March 14. Arpad moved to adjourn. Walter seconded. The meeting adjourned without objection. Task List BIRD189.5 editorial additions/changes to be completed: