Research Progress on Memristors for Compute‐In‐Memory Architectures

The era of artificial intelligence and big data has witnessed an explosive growth in computational demands. The von Neumann architecture faces constraints from the “memory wall” and “power wall” due to its separation of storage and computation. Compute‐in‐memory (CIM) technology has emerged as a key pathway to overcome the von Neumann bottleneck. As a novel class of semiconductor devices capable of both storage and computation, memristors offer advantages such as simple structure, low‐power consumption, and great potential for high‐density integration, showing great promise for CIM architectures. This article provides a systematic review of the latest research progress on memristors in terms of their operating mechanisms, array integration, and applications. It mainly focuses on the working principles, performance characteristics, and optimization strategies of four types of memristors based on the conductive filaments, phase‐change, magnetic tunnel junction, and ferroelectric tunnel junction mechanisms. Key performance parameters of various memristive devices are compared. In addition, the paper introduces multiple integrated solutions for overcoming sneak‐path current interference, presents typical application cases of memristors, and provides an outlook on future development, serving as a reference for further research on memristors.