Arc Model and DDG: Deadlock Avoidance and Detection in Torus NoC
Surajit Das, Chandan Karfa · IEEE Embedded Systems Letters · 2021
Wraparound channels in Torus network-on-chip (NoC) help in reducing overall hop counts traversed by traffic. However, the cyclic paths created by the wraparound channel make Torus NoC deadlock prone. The Turn model and channel dependency graph (CDG) are two classical approaches used for detecting and avoiding deadlock in NoC. In this work, we propose an Arc model for avoiding deadlock in Torus NoC. The Arc model is an extension to the Turn model and is useful for deadlock avoidance in Torus. Directional dependency graph (DDG) is also presented in this work by combining both the Turn model and CDG for detecting deadlock in Torus NoC. DDG is a simpler approach for identifying deadlock scenarios, formulating deadlock avoidance, and showing deadlock freedom while using the Arc model along with a set of Turns.