STAR: A Mixed Analog Stochastic In-DRAM Convolutional Neural Network Accelerator

Editor’s notes: This article addresses the optimization of data movement in accelerating machine learning workloads, one of the most critical issues of state-ofthe- art computing platforms. It presents a novel in-DRAM accelerator for convolutional neural networks using mixed analog-stochastic optimizations and shows significant energy-efficiency improvements. —Umit Ogras, University of Wisconsin, USA