A High-Accuracy and Low-Power Emerging Technology-Based Associative Memory

Abstract

Associative memory (AM) is a subcategory of neural networks (NNs) inspired by human memory. Over time, the need to process complex tasks has increased, leading to the development of intelligent processors. Most NN circuits have been implemented using complementary metal-oxide-semiconductor (CMOS) technologies. However, some adverse effects have become more apparent with the scaling of transistors. Several emerging technologies, such as magnetic tunnel junctions (MTJ) and carbon nanotube field-effect transistors (CNTFET), have been introduced to address these issues. This paper proposes a novel, robust AM design based on CNTFETs and MTJs. The use of MTJs in the proposed design is motivated by their reliable reconfigurability and nonvolatility. Moreover, CNTFETs overcome the limitations of conventional CMOS technology. The main goal of the proposed method is to increase the voltage swing of the synapse output, reducing the impact of process variations and increasing accuracy. Simulation results indicate that the proposed design offers up to 50% fewer recall attempts and at least 15% and 9% lower average and static energy consumption than the state-of-the-art counterparts.