Exploration of Power Domain Partitioning with Concurrent Task Mapping and Scheduling for Application-Specific Multi-core SoCs

This paper proposes a novel approach to explore the design space of Power Domain (PD) partitioning in the architecture definition phase of heterogeneous SoCs. By formulating an Integer Linear Program (ILP), task mapping and scheduling is determined concurrently while considering power-off dependencies among cores in the same PD and the power-gating break-even time. Compared to state-of-the-art approaches aiming at design phases where task mapping and scheduling has been frozen, our proposed approach shifts joint exploration into earlier design phases, creates more power-gating opportunities for PD partitions, and thus identifies better trade-offs in terms of energy consumption and design costs.