Heterogeneous network-on-chip design through evolutionary computing

Abstract

This article explores the use of biologically inspired evolutionary computational techniques for designing and optimising heterogeneous network-on-chip (NoC) architectures, where the nodes of the NoC-based chip multiprocessor exhibit different properties such as performance, energy, temperature, area and communication bandwidth. Focusing primarily on array-dominated applications and heterogeneous execution environments, the proposed approach tries to optimise the distribution of the nodes for a given NoC area under the constraints present in the environment. This article is the first one, to our knowledge, that explores the possibility of employing evolutionary computational techniques for optimally placing the heterogeneous nodes in an NoC. We also compare our approach with an optimal integer linear programming (ILP) approach using a commercial ILP tool. The results collected so far are very encouraging and indicate that the proposed approach generates close results to the ILP-based approach with minimal execution latencies.

Keywords:

• NoC
• evolutionary computing
• genetic algorithm
• heterogeneous

Acknowledgements

This research is supported in part by TUBITAK grant 108E233, by a grant from IBM, and by a Marie Curie International Reintegration Grant within the 7th European Community Framework Programme.