A Multiplexer-Based High-Capacity Spintronic Synapse

Abstract

In recent years, there have been significant advancements in the manufacturing of emerging technologies, especially in the areas of in-memory computing and neural networks, which are currently some of the most actively researched topics. With the increasing need to process complex tasks, the development of intelligent processors has become more crucial than ever. This letter advances a high-capacity spintronic synapse using magnetic tunnel junctions (MTJs) and carbon nanotube field-effect transistors (CNTFETs) to implement associative memory. The choice of MTJ devices is due to their remarkable features, including reliable reconfiguration and nonvolatility. Moreover, CNTFETs have overcome traditional complementary metal–oxide semiconductor limitations, such as the short-channel effect and suboptimal hole mobility. The design seeks to improve accuracy and memory capacity by increasing the number of weights. Simulation results indicate that the design offers a 19%–73% higher number of weights and a lower error rate than the state-of-the-art counterparts.