FARSI: An Early-stage Design Space Exploration Framework to Tame the Domain-specific System-on-chip Complexity

Domain-specific SoCs (DSSoCs) are an attractive solution for domains with extremely stringent power, performance, and area constraints. However, DSSoCs suffer from two fundamental complexities. On the one hand, their many specialized hardware blocks result in complex systems thus high development effort. other system knobs expand complexity of design space, making search optimal difficult. Thus to reach prevalence, taming such complexities is necessary. To address these challenges, this work, we identify necessary features early-stage space exploration framework that targets provide instance refer as FARSI. FARSI provides agile system-level simulator speed up accuracy 8,400× 98.5% compared Synopsys Platform Architect. also efficient heuristic achieves 62× 35× improvement convergence time classic simulated annealing (SA) modern Multi-Objective Optimistic Search. This done by augmenting SA architectural reasoning locality exploitation bottleneck relaxation. Furthermore, embed various co-design capabilities show that, on average, they have a 32% impact rate. Finally, demonstrate using development-cost-aware policies can lower complexity, both terms component count variation much 60% 82% (e.g., Network-on-a-Chip subsystem), respectively.