ALF meeting minutes, Feb.9 1999, 9AM-5PM at Cadence, San Jose Attendees: Tim Baldwin Cadence Ted Elkind Cadence Mike Andrews Mentor Graphics Tim Jennings Motorola Jose DeCastro LSI Logic (by phone) Arun Balakrishnan NEC Wolfgang Roethig NEC Sergei Sokolov Sente (by phone) Shir-Chen Chang Synopsys Tim Ehrler VLSI Next Meeting: March 29-31 Tentative location: LSI Logic, Boulder CO Fallback location: Motorola, Phoenix AZ * review of minutes and old A.I. -------------------------------- All action items pertinent to document corrections were completed prior to the meeting, unless mentioned otherwise. old A.I. better wording for RETAIN -- Ted Elkind new A.I. change wording for RETAIN definition as follows: "time during which an output will retain its value. Always in conjunction with DELAY." -- Wolfgang Discussion for IEEE standardization continued. Making ALF an IEEE standard requires the following: - approval by OVI (Vassilios Gerousis) - application for PAR (Victor Berman) - a sponsor A.I. Contact Vassilios Gerousis @ OVI for approval and directions -- Wolfgang Status: in progress A.I. Contact Victor Berman @ DASC for information purpose -- Mike Status: done A.I. Contact Steve Wadsworth @ VHDL for information purpose -- Ted E. * review of ALF 1.0.9 --------------------- Chapter 2: Verilog and VHDL examples were updated to point out the difference between model for synthesis and model for accurate simulation. A.I. Explain concurrent behavior refering to execution clause in VHDL LRM -- Ted E. Wireload model example was changed cosmetically. A.I. start discussion on more accurate wireload -- OLA work group Chapter 3: A.I. correct non-commutative &> in vector operator table -- Wolfgang A.I. annotation term explanation to avoid misunderstanding, since ALF "annotation" is not the same as SDF "annotation" -- Ted Question about the term "static" in LRM. ALF definition: static measurement === measurement independent of time Some people in the industry refer to "static power" meaning "short-circuit power" which is transient. Other people distinguish between "static power" and "leakage power" which are both static in ALF sense. A.I. cross-reference 3.6.5 and 3.6.11 for definition of "static" -- Wolfgang A.I. explain calculation of RMS in figure 3.18 write "for illustration purpose, definition is up to library developper" -- Wolfgang A.I. spell check: parameterizable -- Wolfgang A.I. some parenthesis correction p.130, p.131 -- Wolfgang Dynamic templates as means for describing parameterizable cells are included in the spec. for ALF 1.1. However, they will not be supported by OLA 1.0, since there are no OLA-compliant design tools for parameterizable cells today. Synthesis, timing, power analysis work on fixed cells. Design planning tools work on parameterizable cells. Since such tools are very new and do not have well-established native libraries in many cases, they may adopt ALF as their native library. * vector modeling review ------------------------ Clarification was achieved on the position of ALF relative to Verilog/VHDL simulation. The ALF goal is not to replace native simulator languages but to complement it by a canonical functional description language for cells, from which Verilog and VHDL models can be generated automatically. However, we don't want to standardize a particular way of translation, since there is a 1-to-many correpsondence between ALF and Verilog/VHDL. ALF describes only the common denominator between all of those models, without defining simulator-specific features. Those features should be in a product specification and not restricted by a standard. Examples: pessimistic or accurate conflict resolution, X-handling, glitch handling ... Arun and Wolfgang presented new documents for vector modeling. Arun proposed a change in the BNF which preserves syntax rules but exposes the semantics in a better way. edge_expression ::= edge_operator boolean_expression vector_expression ::= edge_expression { vector_binary_operator edge_expression } etc. A.I. define associativity for sequence of edge_expression -- Wolfgang A.I. change wording "toggle" into "transition" -- Wolfgang An important misconception about vector expression was cleared: Vector expressions are not meant to be self-sufficient for complete specification of simulation models and finite state machines. However, vector expressions are used in conjunction with other ALF language constructs to provide such specifications, and they make the specification language richer. As for all ALF constructs, vector expressions must be interpreted within their context. For instance, the non-event semantics apply only for power analysis today. There is no semantic meaning for non-events in the context of timing analysis. A.I. Define the purpose of vector expressions upfront -- Wolfgang Scope of vector expressions was also discussed. The suggestion was made to allow additional annotation values for SCOPE in addition to "measure" and "behavior" The proposed interpretation for event-sequence AND was accepted for event sequences of equal length. The proposed left extension for misaligned event sequences has to be approved yet. The detection mechanism for non-events by exclusion of the next event was approved to be correct. Mike voiced a practical concern for instantaneous power analysis. If the next event is not guaranteed to happen within a bound timeframe, instantaneous power waveforms cannot be provided during simulation. Wolfgang suggested that this is an implementation issue for the analysis tool which can be resolved by providing timeout periods for event detection. A.I. put all approved items from vector modeling discussion into spec. -- Wolfgang A.I. maintain the vector modeling document as working document for discussion and put all pertinent items there -- Arun * misc. ------- INITIAL_VALUE is approved as new 1.1 keyword. It shall apply for all downstream modeling tools. If vendors wish to specify tool-specific initial values, they should put those into custom property fields. A.I. put Tim Ehrler's proposal into spec. -- Wolfgang ALF spec. will be frozen by March 1st. A conference call will be scheduled for Monday March 1st for last-minute changes and corrections. The corrected document will be submitted to OVI for official review and approval. A.I. put minutes on the web -- Mike A.I. edit ALF 1.0.10 by Feb.19 -- Wolfgang A.I. put ALF 1.0.10 on the web as candidate for ALF 1.1 approval -- Mike A.I. review ALF 1.0.10 by March 1 -- all A.I. schedule conference call on March 1 -- Wolfgang