Today, chip multiprocessors (CMPs) that accommodate\udmultiple processor cores on the same chip have become a\udreality. As the communication complexity of such multicore systems\udis rapidly increasing, designing an interconnect architecture\udwith predictable behavior is essential for proper system operation.\udIn CMPs, general-purpose processor cores are used to run\udsoftware tasks of different applications and the communication\udbetween the cores cannot be precharacterized. Designing an efficient\udnetwork-on-chip (NoC)-based interconnect with predictable\udperformance is thus a challenging task. In this paper, we address\udthe important design issue of synthesizing the most power efficient\udNoC interconnect for CMPs, providing guaranteed optimum\udthroughput and predictable performance for any application to be\udexecuted on the CMP. In our synthesis approach, we use accurate\uddelay and power models for the network components (switches\udand links) that are obtained from layouts of the components using\udindustry standard tools. The synthesis approach utilizes the floorplan\udknowledge of the NoC to detect timing violations on the NoC\udlinks early in the design cycle. This leads to a faster design cycle\udand quicker design convergence across the high-level synthesis\udapproach and the physical implementation of the design. We\udvalidate the design flow predictability of our proposed approach\udby performing a layout of the NoC synthesized for a 25-core CMP.\udOur approach maintains the regular and predictable structure of\udthe NoC and is applicable in practice to existing NoC architectures.
展开▼
