Memristor-based in-memory computing

Abstract

Memristor is a resistive random-access memory (RRAM) technology that enables the development of low-power and low-cost solutions. It has the potential to be used for storage, computing, security, and sensing applications. One of the critical features of a memristor device is its ability to behave like an analog memory that mimics the synapse behavior in the brain and consequently achieves a bio-inspired system. Neuromorphic systems based on such circuits produce high-density, extremely low-power hardware designs capable of performing many MAC operations in parallel. Additionally, since the memristor combines storage and computing capabilities, it is an ideal building block for in-memory computing (IMC). This chapter introduces the memristor-based IMC technology, along with a review of state-of-the-art memristor accelerator frameworks in different applications. In addition, the different challenges related to IMC with memristor devices and crossbars on the device, circuit, and system levels are discussed. Furthermore, in order to demonstrate the potential of the memristor device, especially its analog memory characteristics, which are essential for many applications, a RRAM-based transformer has been implemented, simulated, and tested. Finally, a discussion of the commercial availability of the memristor design is provided.

Keywords: Data-Centric Computing, Efficient Computing, In-Memory Computing, Machine Learning, Memristor, Near-Memory Computing