Application driven routing for mesh based Network-on-Chip architectures

Expertise and intuition of chip design engineers supplemented by a unique design objective governs the choice of routing algorithm in Network-on-Chip(NoC) based systems. However, the same choices when considered for more than one performance parameter of the network, do not yield expected results since different applications exhibit different network performances for the same routing technique. Thus, we advocate use of an application pattern driven routing approach instead of the usual static routing strategies. In this paper, we propose an application pattern driven dynamic routing approach for NoC based systems while committed to optimize desired performance parameters. Our proposed methodology serves a dual purpose of improving the network performance of the NoC based system as well as relieving the design engineers from making the complex and tedious routing decisions. A machine learning based predictive approach is employed in our approach to classifying traffic patterns and it achieves 94.78 % accuracy while our overall proposed methodology achieves significant improvement in power consumption and throughput as compared to state of art works available in literature.