A structural analysis of evolved complex networks-on-chip

Abstract

Designing large-scale heterogeneous Networks-on-Chip (NoCs) for irregular applications often involves sophisticated optimization techniques that lead to unstructured networks. Such networks are hard to understand because they were not built with common engineering knowledge. In this paper we use tools from complex network analysis, such as community detection, and small-worldness, to analyze the structure of optimized heterogeneous NoCs. Our results show that communities evolve robustly and that heterogeneous link types are efficiently establishing inter- and intra-subnet connections. We confirm that networks optimized under cost pressure are more modular. Finally, we determine the small-worldness and observe that small-world networks evolve as a result of a trade-off between performance and cost. Our results are relevant for the understanding of heterogeneous NoCs and for designing optimal communication fabrics for emerging technologies.