SAGA: Synthesis technique for guaranteed throughput NoC architectures

We present SAGA, a novel genetic algorithm (GA) based technique for synthesis of custom NoC architectures that support guaranteed throughput traffic. The technique accepts as input a communication trace graph, amount of data, period, and deadline for each trace, interconnection network architecture elements, and generates a custom NoC topology, and routing and schedule of the communication traces on the architecture. SAGA minimizes both the energy consumption and area of the design by solving a multi-objective optimization problem. We present a detailed analysis of the quality of the results and the solution times of the proposed technique by extensive experimentation with realistic benchmarks and comparisons with optimal MILP solutions. SAGA is able to generate solutions that are as good as the optimal solutions produced by the MILP formulation. Whereas the MILP formulation run time rises exponentially for even moderately sized graphs, SAGA generates solutions for large graphs in reasonable time.