Approximate communication for energy-efficient network-on-chip

Abstract

Approximate Computing is being touted as a viable solution for high-performance computation by relaxing the accuracy constraints of applications. This trend has been accentuated by emerging data intensive applications in domains like image/video processing, machine learning, and big data analytics that allow inaccurate outputs within an acceptable variance. With the increasing communication demand as well as the optimization bottleneck of NoC performance and energy consumption, approximate communication, which leverage relaxed accuracy for energy-efficiency Networks-on-Chip (NoC), have become the accepted method for connecting a large number of on-chip components. We, respectively, proposed approximate designs for traffic regulation, bufferless NoC, and multiplane NoC. These designs improve network performance and reduce power consumption by reducing network load, optimizing data transmission, and optimizing network architecture design. The approximate communication designs show a huge improvement in energy-efficient NoCs while maintaining low application error.