Abdullah Hanif Muhammad

1. Introduction Deep neural networks (DNNs) are widely used in most of the application domains including computer vision, data mining, healthcare, robotics and language processing [1]. While these networks offer state-of-the-art accuracy for many applications, they are highly compute-intensive and energy hungry. To address these challenges, a large body of work has been carried out in building DNN accelerators for improving the performance and energy-efficiency of the DNN-based applications [2].