SMT-based Multi-objective Optimization for Scheduling of MPSoC Applications

Network-on-Chip (NoC) is a promising interconnecting paradigm in the state-of-the-art multi-core architectures. Its communication network can increase the capacity of parallel data transfer such that system performance is improved. In the design of MPSoC-based applications, multiple objectives exist, such as minimizing time and energy consumption, which may conflict and certain trade-off needs to be evaluated. Heuristic-based methods such as evolutionary algorithms are always adopted to find near-optimal solutions for such applications. However, it is hard to evaluate the accuracy of those solutions. As most of the constraints on the mapping and scheduling process of NoCs can be described as logic formulas, we apply SMT-based methods for the multi-objective optimization of NoC-based MPSoCs. Moreover, to improve the scalability of the optimization problem, we propose to reduce the search space with respect to the symmetry feature of NoC architecture, and to decompose the search process according to the feature of non-dominated solutions. Extensive experimental results from random and real-case benchmarks demonstrate the accuracy of SMT-based methods in finding all the Pareto-fronts, and the efficiency of the proposed strategies.