Current status and issues of in-memory accelerators for deep neural networks

Abstract

Deep neural networks (DNNs) have been widely employed for a large number of applications such as image recognition and natural language processing. However, DNNs require a large amount of computation. In-memory accelerators are one of the promising solutions in order to ease this heavy burden. In this article, the current status and issues of in-memory accelerators compared with typical digital-based accelerators are described with some examples. Additionally, we discuss the main considerations and necessary steps in order to make in-memory accelerators practical, and also how to perform effective benchmarking.