20250122_A-102_Kabir.PDF

1、Using AI to Improve EDA Tools for Multi-Die ArchitecturesChiplet SummitKamal Desai,Product ManagerSutirtha Kabir,Sr R&D Director1/22/25 2025 Synopsys,Inc.2Agenda AI Assisted,Early Architecture Exploration AI Enabled System Analysis&Optimization Flow 2025 Synopsys,Inc.3Power models and characterizati

2、onTypical Performance&Power Analysis FlowRoot-cause analysisSoftware andtracesmapApplication specificationApplication workloadWorkload trace and statisticsModel librariesHardware platformSensitivity analysisDesign space explorationand collaborationParameter SweepingManual StepsRequires expert design

3、 knowledge 2025 Synopsys,Inc.4Design Space Exploration,Time to Results ChallengesToo many scenarios for coverage of the entire design spaceNeed expert knowledge of parameter impact on KPIsManual,tedious tasks of identifying root causesRequires significant compute resources for entire design space 20

4、25 Synopsys,Inc.5Machine Learning for Design Space Exploration and Sensitivity Analysis Automatically setup all relevant scenarios and execute relevant ones only Nominal expert knowledge required on IP and design parameters Analysis results in fastest time while minimizing compute resources Expedite

5、d DesignRequirements with constraintsDesign with workloadExploration Space 2025 Synopsys,Inc.6AI Enabled Performance&Power Analysis FlowSimulate&AnalyzeRequirements Met?Executable Architecture Specification&Architecture Performance ReportYesNoRoot Cause Analysis +Structural ChangesStarts from requir

6、ements on KPIs,which guides the AI assisted explorationExploration space for the architecture needs to be Defined:which parameters should be modified?Constrained:which parameter values and which parameter value combinations are allowed?If there is a parameter configuration meeting all requirements T