Transcript Document
C.A.D.: Bookshelf June 18, 8:00am-11:00am Outline • Review: [some of] bookshelf objectives • Where we want to go vs what we have now • Invited presentations – Herman Schmit, CMU – Paul Villarrubia, IBM Technology Group • On-going work • Current challenges • Future foci ? Review: [some of] our objectives • to provide benchmarking infrastructure relevant for – research – publishing – industrial use • offer high-qualify design tools – preferably open-source • integrate design tools into tool flows Where we are now • So far, mainly focused on physical design • Available now – – – – – – partitioning, floorplanning and placement benchmarks floorplanner in Java (no source?) partitioners (MLPart, hMetis) placers (Capo, Dragon, Feng Shui) a global router (Labyrinth) a DB with LEF/DEF parsers and PERL/Tcl/Python interfaces – scan chain slot with codes – RSMT/RMST and BST slot with codes, etc Where we are now (cont’d) • Know-how regarding integration with Cadence and IBM P&R tools • Links to lots of related general-purpose goodies – network-flow solvers – LP and non-linear solvers – etc, etc... We need • Feedback from the industry and use by the industry • More integration, esp. with commercial tools • Wider participation and adoption • Need to refine future focus Industrial participation/requirements and vertical benchmarking • Paul Villarrubia, IBM – "An overview of important features for industrial placement problems" – Issues: relevant benchmarks and industrial adoption • Herman Schmit, CMU – "The Vertical Benchmarking Project at CMU" – Issues: relevant benchmarks and design tool integration On-going work • • • • Ivan Kourtev: optimal clock skew scheduling C.-K. Cheng: interconnect delay/timing analysis John Lillis: SITS Integration and comparisons with commercial tools – Cadence Pearl, WarpRoute (UCLA, UMich) • e.g., CapoT > Pearl > WarpRoute > Pearl – IBM ChipBench (UMich, IBM) • e.g., CapoT > EinsTimer > XRouter > EinsTimer On-going work (cont’d) • “Simple" (but not easy) benchmarks w/o all bells and whistles – solvable with both commercial and academic tools – can give apples-to-apples comparisons – WL-driven and timing-driven placement (UCLA, UMich) – routing benchmarks (UMich) On-going work (cont’d) • New tools – an open-source floorplanner in C++ (UMich) – more versatile open-source routing tools (UMich, SUNY, etc) – UCSD is committed to filling in special engines (clock, power, test, area fill, etc.) that are needed to get reasonably complete layouts Discussion: current challenges • “Separating" global and detail routing (data format, evaluations, at least one engin for each) • “Merging" floorplanning and large-scale placement • Correlating and anti-correlating – placement wirelength with routability and routed WL – placement wirelength with timing objectvies – formulating and validating "simple"/clean design metrics and optimization objectives for consistent research in multiple groups (e.g., the overflow metric used by Majid Sarrafzadeh's group) Future foci ? • • • • More attention to data-modeling ? Fully open-source CAD design flows? Comprehensiveness Formulations of open problem submitted by the industry ? – remember the "Top-ten list for Physical Design" @ ISPD ? • More formal and automated interface – peer reviews for some of bookshelf content – hit statistics • Lobbying for an official status with DAC ? Future foci(?) cont’d • “Same old" – more benchmarks ? – better tools ? – more empirical comparisons ? Conclusions…