Thanks Mike Labonte for your help with the minutes this time. Doug
IBIS Macromodel Task Group Meeting date: 24 oct 2006 Members (asterisk for those attending): * Arpad Muranyi, Intel Corp. Barry Katz, SiSoft Bob Ross, Teraspeed Consulting Group * Doug White, Cisco Systems * Hemant Shah, Cadence Design Systems Ian Dodd, Mentor Graphics * Joe Abler, IBM * John Angulo John Shields, Mentor Graphics Ken Willis, Cadence Design Systems * Kumar, Cadence Design Systems Lance Wang, Cadence Design Systems Michael Mirmak, Intel Corp. * Mike LaBonte, Cisco Systems Paul Fernando, NCSU * Randy Wolff, Micron Technology * Richard Ward, Texas Instruments Sanjeev Gupta, Agilent Shangli Wu, Cadence Todd Westerhoff, Cisco Systems * Walter Katz, SiSoft Vuk Borich, Agilent * Vikas Gupta, Xilinx ------------- Review of ARs: - Michael will try to get access to the IEEE document they are voting on. He will check into access restrictions first. If we get access to it, we will need a couple of volunteers to read the document and see if it is worth our while. - TBD - Macro Library Documentation - Mike will get to this in Nov ------------- New Discussion: Time zones: we will keep meeting at the same time (12:00p PT) after the US time zone change. VHPI We need to discuss this more before we define a new API. Arpad showed a presentation: "Channel Analysis Flow with IBIS" There are 2 ways to get a pulse response, characterize the channel and convolve, or run a plain analog simulation with drivers and receivers. Arpad explained his view of Kumar's optimization process. All simulated impulses are really ramps, although very short. Convolution can be done in AMI_Init. Some simulators will process the actual waveform in AMI_GetWave to initialize filter coefficients. Walter: there are actually 4 steps. Both TX and RX have init functions. 1 call TX init with channel and TX params 2 call TX AMI_GetWave with stimulus pattern, generates V(RX) 3 call RX AMI_Init to setup RX for tap coeffs 4 call RX AMI_GetWave Optimizer algorithm needs access to model parameters. Special optimization features for ICM models would be good. Should be able to use *-AMS API for waveform exchanges. Walter: won't work for time domain simulators because it requires waveforms in time chunks, not one timestep at a time. There is no live communication between the SPICE engine and optimization engine. Walter: receiver might have a Framostat (black box) Kumar: you could write a wrapper to make the black box mdel work in AMS. AR: Mike investigate call dropouts. Arpad has been bounced out with busy signal twice, Michael Mirmak once. Kumar added 3 or 4 slides to his presentation. Nice diagrams showing step-by-step operations. Walter: how will we document in that model how it works? Kumar: parameters are passed by character string. Waveforms a vector of doubles. Aprad noted that the optimization of an eye could be accommodated in IBIS with a small BIRD proposal with: -range and step definitions (for parameters) added to IBIS spec for external model and external circuit calls Question arose about the possibility of the optimization of the passive channel itself, utilizing interconnect models (ICMs). In the case of behavioral representations like s-parameters, this becomes difficult if not impossible. Arpad reiterated his view that the AMS languages appear to have their own structures which could accommodate the data exchange between simulator entities and the prospective models, in the same way that Cadence's API proposal does, and with such changes to IBIS as the "step" and "range" additions mentioned above. Question the group had was whether we wanted the IC vendors to be forced to generate AMS models? Kumar answered by saying that his API proposal is meant to be agnostic to IC vendor models. They can be anything of any complexity but must have a standard interface to the simulator. He called this "simulator-neutral". It was asked whether the AMS API could work like this (have these characteristics). Kumar presented some more detailed slides of Cadence API proposal. INIT: send the impulse response to the black box model, as a bunch of vectors. GETWAVE: Vector waveforms are sent to RX model, which modifies the waveforms. Sidenote: Impulse response can be obtained by doing: Output pulse/Ideal input pulse area It was asked whether we could obtain voltages at intermediate nodes with this approach. AR: Kumar to present an elaborated description of the data constructs in the next meeting. AR: Mike to post Cadence powerpoint (presented today) on website. ------------- Next meeting: Tuesday 31 Oct 2006 12:00pm PT