An optimisation algorithm for minimising energy dissipation in NoC-based hard real-time embedded systems

This paper introduces an evolutionary multi-objective optimisation algorithm to facilitate fast and efficient task allocation of hard real-time embedded systems with Networks-On-Chip (NoC) as the interconnection at the early design stage, where evaluating as many as possible solutions is crucial. Our approach uses analytical fitness functions to provide fast evaluation of large number of solutions; a contrast to simulation-based optimisation technique, whereby it tends to be not only impractical when the design space is very large but also unfeasible as far as hard real-time systems are concerned. The proposed algorithm guarantees the predictability in timing behaviour of the systems whilst minimising energy dissipation whenever tasks are reallocated and their packets are rerouted, which differs from the state-of-the-art approaches. In addition, not only it can explore the allocation of tasks but also the encoding of the data packets. The evidence gathered from case studies shows that the proposed algorithm is able to find schedulable allocation of tasks, preserving it whilst further minimising energy dissipation.