A Deadlock-Free and Adaptive Prime Perspective Turn Model for 3D-Mesh Based Network-on-Chips

With the tremendous demands for core-based digital systems, routing algorithms play a key role in improving metrics like performance parameters (delay, throughput) and cost effectiveness of network-on-chips (NoCs). A 3D-mesh-based NoC essentially stacks many layers of 2D-meshes integration to achieve greater performance, a high integration density, shorter interconnects, and lower power consumption. We propose a 3D-mesh routing model that is adaptive, deadlock-free, and that forbids certain turns depending on the prime number of stacks up layer as well as vertical and horizontal directions without virtual channels. The proposed routing model extends the two dimension repeating turn model (RTM-2D) into 3D-mesh by applying certain additional rules based on prime perspective, and it balances the degree of adaptiveness accordingly as prime perspective plane. According to the simulation results, our routing method outperforms in terms of performance metrics to the six typical routing systems in the uniform, transpose, and hotspot cases.