izvorni znanstveni rad

Memory-aware multiobjective design space exploration of heteregeneous MPSoC

Vlado Sruk

Sažetak

This paper discusses multiobjective exploration of heterogeneous MPSoC design space using a method based on NSGA-II evolutionary algorithm. Key feature of the proposed method is separation of computation and communication which enables exploration of mapping computation to processors and communication to memory elements. In this paper two approaches to mapping and scheduling are presented and compared: (1) single-phased, where tasks and communication channels are mapped to processor and memories simultaneously, and (2) two-phased, where mapping of tasks to processors is done in first phase, followed by mapping of communication channel to memories in the second phase.

Ključne riječi

genetic algorithmsintegrated circuit designmicroprocessor chipsrandom-access storagesystem-on-chipcommunication channelmemory-aware multiobjective design space explorationNSGA-II evolutionary algorithmheterogeneous MPSoC designTask analysisProgram processorsResource managementCommunication channelsSpace explorationProcessor schedulingEvolutionary computationdesign space explorationMPSoCevolutionary algorithm