The Future of High-Level Modeling and System-Level Design . Some - PDF document

The Future of High-Level Modeling and System-Level Design ……. Some Scenarios Grant Martin Fellow, Cadence Labs Electronic Design Processes Workshop, Monterey, April 21-23, 2002 1 CADENCE DESIGN SYSTEMS, INC. Outline • Six Scenarios for System Level Design • What is the likely future? 2 1

The Cult Scenario • High Priests, Gurus, and their Acolytes • Obscure Mathematics and Jargon • Impenetrable to ordinary designers • No flow – Models have no reuse at lower levels of abstraction • There is no mainstream 3 The Niche Scenario • Best practices today • Dataflow: from algorithm to implementation • Finite State Machine capture, simulation, and generation of synthesisable code • Most of the well-understood niches have been explored 4 2

The Platform-Based Design Scenario • Co-ordinated family of HW-SW architectures promoting high levels of reuse of HW and SW virtual components • If Platform-based design succeeds : – Most designs are straight-forward derivatives – Design = configuration, selection, “soft” programming – Platforms can be designed with blood, sweat and tears – System design helps, but only a small community 5 The Hardware Scenario • Bob Brodersen/BWRC SSHAFT group are key proponents for this approach • Since HW = 100X better than SW in power and area, and 10X+ in performance, why do anything in SW? • Direct mapping from algorithms in high level models to HW implementation via automated flows • BUT: is this likely? For most products in DSM technologies? – Requires Masks, $1M NRE, a Month, risk Management – Also requires generalised behavioural synthesis, beyond the known niches 6 3

The Software Scenario 1RUPDO (P HU� &DOO &DOO �IXQFWLRQDOLW\� �IXQFWLRQDOLW\� $VN�6UF ^UHTXLUHG_GHOD\� � V` * HW�6UF 7 HDU�6UF * HW�6UF ^UHTXLUHG_SUHHP SWLRQ` �XVH� �QHHG� $63 $3, ^DYDLODEOH_GHOD\� �V` Z DLW5HT�� SULRULW\7DVN�� SURF5HT�� SUHHPSW7DVN�� $FN5HT�� RU ^SRWHQWLDO_SUHHPSWLRQ ^DYDLODEOH_GHOD\� ���V` SROO5HT�� IHDWXUH�DYDLODEOH` SURF5HT�� SROO$FN�� ��� • Au contraire , most products will be done in software only • Assumes triumph of the Platform-based approach – Derivatives will be done soft, even reconfigurable HW done “soft” • SW people will not take kindly to HW-based concepts of ‘system-level design’ or ‘high-level modeling’ • Much more likely to: – Hack C code – Use UML/SDL and the like and base flows on these – In this scenario, system level design = “software-software codesign” 7 The Optimistic System-Level Design Scenario 1 2 System System System System Behavior Behavior Architecture Architecture Optimal Mapping Mapping Mapping 3 Performance Performance Simulation Simulation Function Communication Refinement 4 A r c h i t e c t u r e Flow To Implementation Flow To Implementation • Function-architecture codesign is needed • Platforms imply implementation choice • Design space exploration is key to optimal product design • Models of computation are important ! Dataflow � control! 8 4

What are the pre-requisites for the optimistic scenario? Lingua Franca Consensus on Flow methodologie(s) and Model abstraction(s) 9 What are the pre-requisites for the optimistic scenario? Lingua Franca Flow Consensus on methodologie(s) and Model abstraction(s) 10 5

What is the likely future? • I am an optimist • System Level Design will have a Radiant Future • We’ve got to work hard to make it happen • We’ve got to keep educating people on the important issues. 11 6