Transcript Metrics
Foundations for Understanding Achievable Design: Ground Truths, the Bookshelf, and Metrics Theme Summary 1 Mission Enable the understanding of achievable design 2 Precepts System design is based on : – design optimization – prediction GSRC must prove : – real collaboration (“combine strengths”) ; whole > sum of parts – critical mass and influence to create community-wide culture change – its effect on the real-world practice of design and test This Theme’s answer: – ground truths – the bookshelf – metrics 3 1. Ground Truths Fundamental facts and data points that anchor the process of bounding the achievable envelope of design – ground truths must be identified and propagated Can be with respect to: – manufacturing process, materials, physical phenomena – specific CAD optimizations of circuit topology/embedding – system architecture and packaging Are properly extrapolated via: – "inference chains" – response surface modeling and parameter optimization Drive the EDA vision of future design issues, methodology – fundamental limits, fundamental truths, stakes in the ground – current distribution, inductance extraction, UDSM testing, … 4 Example Ground Truths What is the maximum possible clock frequency for a given process and die size? When does inductance matter? What design tradeoffs must be made to maintain reasonable supply currents? What is the necessary number of package pins/balls for power/ground distribution? At what geometries, supply voltages will domino lose most advantages over static CMOS? What is an optimum design strategy from a manufacturing cost point of view? 5 2. The Bookshelf The Bookshelf – framework to enable shared, community-wide maintenance of state for key pieces of the (CAD algorithms) R&D leading edge – remove barriers to entry for algorithm research -- and adoption of algorithm research -- at the leading edge Bookshelf Slot: “key problem for the VLSI CAD field” Bookshelf = repository for leading-edge innovations -- and their implementations -- for solving this problem Literally, publication medium + review process for implementations – electronic infrastructure permits archival, “arbitrarily large” publications – credit for leading-edge work via respectable, referable nature – facilitates review, returns for fixes to preserve quality – backed by other incentives, openness, culture change 6 The Bookshelf (cont.) For given Bookshelf slot, several basic types of entries – – – – canonical problem definition (e.g., C++ class + supporting packages) reference solver implementations benchmark data heuristic evaluation and comparison methodology Bookshelf framework allows : – improved effectiveness and impact of heuristic algorithm research in VLSI CAD – more rapid communication between research groups – more rapid adoption of research advances by industry – path to evolvable GSRC implementation flows/methodologies – implicit definition, testing of underlying data model 7 3. Metrics Design optimization must be founded on – an understanding of what should be optimized by which heuristic – an understanding of design as a process “Metrics” supports ideal of "measure, then improve” – design becomes less of an art and more of a formal discipline – design process optimization enabled through framework of recording, mining, measuring, diagnosing, and then improving Infrastructure – hoped-for: “proactive” initiative from EDA vendors – reality: for now, will use Perl scripts around existing tool logfiles – guidance from designer and design tools R&D communities: (i) what should be measured, (ii) data mining / visualization / diagnosis infrastructure, (iii) project-specific design process data collection 8 Staffing Andrew Kahng 100% Wayne Dai 100% Wojciech Maly 50% Jason Cong 33% Kurt Keutzer 20% 9 Ground Truths Status Basic engine and infrastructure – 2 Ph.D. students, one staff – dialogue with RPI, Georgia Tech, Berkeley, Rockwell, … – bring in experts to help define / vet models and rules chains Key differentiators – models of optimizations (BIS/WS-optimized, multi-terminal global interconnects; block packing; optimal layer assignment, …) – cost modeling, cost implications of design decisions – flexible use model, platform independence, many “canned” previous studies, design of system with adoptability in mind Working through key software issues: – software architecture (C++/Java, multi-platform) – engine, GUI codesign – flexible definition of rules chains, studies, etc. 10 The Bookshelf Status Initial bookshelf slots: – – – – – hypergraph partitioning block packing / floorplanning standard-cell placement single-tree interconnect synthesis/optimization grid-graph based global routing Extremely active convergence on data formats, tool linkages – UIC, UCLA, SUNY Binghamton, UCSC in initial loop – 2 Ph.D. students + outsourcing to interested/engaged researchers – will likely end up being a practical driver for efforts toward GSRCendorsed data model and API Standards – standards (build system, platform, software, etc.) near-converged Commercial backplane – tools (back-end implementation flow) from Synopsys, Cadence – compare against, integrate, mix-and-match with mini-flow above – industry data also sought 11 Metrics Status Metrics infrastructure – – – – – substantial IP recently obtained from OxSigen LLC (used at Siemens) Metrics warehouse: data model, schema, API, servlets Metrics transmitter: applets to write metrics, embeddable in tools Off-the-shelf standard components: Oracle8i, XML, Java, Chai OxSigen scripts + extensions wrapped around today’s tool logfiles : metricizes the baseline (Synopsys/Cadence) GSRC back-end flow – 1 Ph.D. student will do thesis on Metrics in EDA Need more buy-in from EDA companies Need more pull from EDA customers = GSRC sponsors 12 Future Plans Ground truths system, GTX1.0 – GTX1.0: arbitrary tradeoff studies, parameter optimizations – platform-independent, GUI-enabled – accurate models of optimization effects in layers between individual wires/devices and system architecture – determine how correct/incorrect our views of ground truths may be (e.g., white papers on clock jitter/skew, wire estimation models, …) Bookshelf – culture/behavior goals, standards: publicize, internalize, externalize – 3-5 slots instantiated with entries from multiple investigators – mini-flow built around partitioning / soft-block packing / cell placement / interconnect opt / global routing Metrics – integration of OxSigen IP – GSRC endorsement/proposal of standard Metrics schema and API – GSRC sites running metricized std implementation methodology 13