Metrics for multiprocessor system on chip

Generally, system designers use their own experience to define SoC architecture. However, the system design space is growing increasingly, due to the great choice of potential hardware and software resources. This characteristic leads to raise the complexity of defining an efficient and flexible architecture for actual and future systems. Hence, fast, automatic (or interactive) and efficient exploration of the system design space is mandatory. To define an appropriate architecture for an application, a thorough analysis of the application is necessary. In this paper, we define several metrics that allow analyzing the system characteristics at a high level of abstraction. They enable also to evaluate the impact on the system performances of tasks clustering. As the proposed metrics have an important impact on the final system performances they are considered during the first steps of the design flow before any architecture definition and hardware-software partitioning. Several experimental results performed on an UMTS application demonstrate the efficiency of these metrics to exhibit the main characteristics of an application. The results also show the benefit of considering during the first steps of the design flow different metrics in order to build an architecture that leads to an efficient implementation of the application.