Capacity optimized NoC for multi-mode SoC

Network-on-Chip (NoC) is an evolving interconnection architecture addressing the rising complexity of system-on-chips (SoCs). We present a model for the cost of a NoC for a multiple use-case SoC, i.e., a system with distinct modes of operation, each having a unique traffic pattern. Specifically, we formulate an optimization problem capturing the fact that different use-cases can share capacity. We evaluate the proposed scheme using synthetic and real-life traffic, showing a substantial reduction of up to 27% in the required NoC resources, both when using a new algorithm we present and when using (a somewhat heavier) simulated-annealing procedure.