Vector Matrix Multiplication Using Crossbar Arrays: A Comparative Analysis

摘要

Vector Matrix Multiplication (VMM) is a demanding operation that exposes a weakness of current digital hardware when applied to Artificial Intelligence (AI) algorithms. Crossbar arrays were proposed as a means to get more effective analog implementations. These arrays are often composed of circuits based on memristors or other custom designs. The Dot Product Engine and the Resistive Processing Units are two proposed solutions. Either design can reduce the training time to O(1) per VMM layer giving a boost to AI and making future requirements more possible.