IBIS Macromodel Task Group Meeting date: 05 June 2018 Members (asterisk for those attending): ANSYS: Dan Dvorscak * Curtis Clark Cadence Design Systems: Ambrish Varma Brad Brim Kumar Keshavan Ken Willis eASIC: David Banas GlobalFoundries: Steve Parker IBM Luis Armenta Trevor Timpane Intel: Michael Mirmak Keysight Technologies: Fangyi Rao * Radek Biernacki Ming Yan * Stephen Slater Mentor, A Siemens Business: John Angulo * Arpad Muranyi Micron Technology: * Randy Wolff * Justin Butterfield SiSoft: Walter Katz Todd Westerhoff * Mike LaBonte SPISim: * Wei-hsing Huang Synopsys: Rita Horner Kevin Li Teraspeed Consulting Group: Scott McMorrow Teraspeed Labs: * Bob Ross The meeting was led by Arpad Muranyi. Curtis Clark took the minutes. -------------------------------------------------------------------------------- Opens: - None. ------------- Review of ARs: - Michael M. to prepare a draft BIRD allowing Rgnd and Rpower for the Input Model_type. - In progress. -------------------------- Call for patent disclosure: - None. ------------------------- Review of Meeting Minutes: Arpad asked for any comments or corrections to the minutes of the May 29 meeting. Bob moved to approve the minutes. Randy seconded the motion. There were no objections. ------------- New Discussion: Enhanced C_comp Modeling: - Randy: [sharing his "Overview of C_comp Modeling using IBIS-ISS or Touchstone" presentation] - Reintroducing the topic with this presentation and discussion instead of merely reviewing the most recent draft BIRD text. (most recent draft can be found at the ATM archives dated August 15, 2017). - slide 2 [Bird Overview] - Alternative to simple [C_comp] model. - Allows more complicated frequency dependent models. - [C_comp Model] - new keyword scoped by [Model] - [C comp Corner] - will be required if using [C_comp Model]. - [C_comp Model] used during actual simulation. [C comp Corner] used by the simulator for K-t curve generation. - Modification of BIRD191.2. If [C_comp Model] is used, then the "Die" value for Si_location or Timing_location refers to the Buffer_I* terminals of the [C_comp Model]. - slide 3 & 4 [[C_comp Model] Keyword Sub-parameters] - Mode - Similar to Submodel language (Driving, Non-Driving, All). - Up to two [C_comp Model]s may be defined (one for driving and one for non-driving). - This could cause issues with dynamic switching of a [Model] from Input to Output mode during simulation (e.g. bus switching). This needs more consideration. - Param - Corner or single Value type params can be passed to an underlying IBIS- ISS subcircuit. - Number_of_terminals - definition similar to BIRD189. - Terminal lines - at least two required (one signal and one reference). - slide 5 [C_comp Model] - Shows direct C_comp replacement. - Single connection to the Buffer_I/O terminal. - Connections to all the *_ref rails. - slide 6 [C_comp Model] - Shows C_comp replacement containing series elements. - Breaks the buffer up into sections. - Needs additional internal terminals Buffer_O and Buffer_I. - May be used for Input buffer isolation/filtering. - May be used for buffer-level interconnect not modeled as on-die interconnect with BIR189. - slide 7 [C_comp Model] - Shows C_comp model for differential buffers ([Pin]s associated with [Diff Pin]). - Could allow for true differential buffer models. - [C_comp Model] could contain anything you want. - Do we want this added complexity in IBIS? - Last time this BIRD was discussed there was some interest in this. - inv_pin would have to use the same [Model], otherwise you have issues with different Pins using different [Model]s and getting different [C_comp Model]s. - slide 8 [Issues to Discuss] - Reference requirements: - Should A_gnd be an allowed reference? - Differential C_comp Models - Pseudo-differential buffers typically have limited interaction except at the input buffer. - Differential capacitance not usually needed. - Could [C_comp Model] create true-differential buffer models that won't work in practice? - C_comp compensation algorithms - Is [C comp Corner] enough for EDA tools to do compensation? - Separate Input and Output mode [C_comp Model]s - Will cause issues with Enable/Disable simulation of I/O buffers. - Discussion: Bob noted that the presentation contained some concepts not in the most recent BIRD draft. Bob noted that the problem with a generalized C_comp is that anything could be in the black box, and the EDA tool wouldn't know how to compensate. Bob noted that if a model had a straight path for the Output mode then the existing compensation algorithm would work fine. Radek noted that the differential concept figure showed more nodes defined for terminals than there are terminals in a given [Model], and he asked how this could work. Randy noted that both the positive and inverting pins in the [Diff Pin] pair would have to use the same [Model], and their terminals would be connected as shown in slide 7. Arpad noted that a similar situation exists with the true differential [External Model]s. Bob noted that true differential is only supported in IBIS by [External Model] and requires one of the *_diff Model_type(s). Arpad noted that [C_comp Model] might provide a hybrid with pseudo-differential buffers and a true diff [C_comp Model]. One might be able to fake a true differential buffer with this proposal. Bob agreed. Arpad noted that he understood why the [C comp Corner] was required and would be used for K-t compensation. But he feared that the bigger the difference between the full [C_comp Model] and the [C comp Corner] the worse the normal compensation algorithms would perform. At that point, might we be better off without a traditional compensation, but then how would one come up with the proper drive strength to get the measured v(t) waveform edge rate? Mike L. noted that even with BIRD189 we still only define a single terminal for the buffer I/O. What Randy's proposal allows is recognition that the output transistors and the sensing inputs on an actual I/O buffer may in fact be separated by bits of interconnect that matter more and more as speeds go up. Bob noted that the syntax supports a C_comp Model for the Output and Input. His understanding was that it was for the Input section that more detail was required and a more detailed "C_comp" model including series elements might be most useful. For the Output side this was typically not an issue. Randy agreed. Bob noted that he would prefer to have the "box" on slide 6 shown as two separate boxes, one for Output and one for Input. This could leave one with a simpler model on the Output side and reduce the problems with the K-t compensation. Bob noted that A_gnd doesn't appear in this proposal, but we don't know if it appears within the IBIS-ISS subcircuit. Arpad wondered if we might only discuss A_gnd for use with Touchstone files and define it for use only as the N+1st reference terminal. Randy noted that it had not been limited to that use in BIRD189. Arpad said he understood the desire to utilize the same rules as BIRD189. However, unlike BIRD189, in this [C_comp Model] context we should never have unused ports we need to terminate, and we should not have a reference we need to tie to A_gnd because we have the Model's *_ref rail terminals available. Randy said it wouldn't be a problem to disallow A_gnd for any other purposes in the BIRD. Wei-hsing asked if a [Diff Pin] pair and a Series Model could behave like a differential capacitance, and noted that the IBIS Cookbook says to model differential capacitance with a Series Model. Randy asked if the differential [C_comp Model] could be useful as a more complicated version of the Series Model approach. Arpad noted that a Series Model was just several possible RC circuit combinations, and that they were not acting as drivers and did not require C_comp compensation. Arpad asked if Randy needed more input from the group. Randy took the AR to make another minor revision of the BIRD draft's text. - Mike L.: Motion to adjourn. - Curtis: Second. - Arpad: Thank you all for joining. AR: Randy to update the enhanced C_comp Model BIRD draft. ------------- Next meeting: 12 June 2018 12:00pm PT ------------- IBIS Interconnect SPICE Wish List: 1) Simulator directives