Fully CMOS Compatible Charge Trap Memory-Based Reservoir Computing System

Abstract

Reservoir computing (RC) systems have gained considerable attention for their effectiveness in temporal data processing. Although extensive research has been conducted on RC systems, studies focusing on complementary metal-oxide semiconductor-compatible flash memory devices remain scarce. In this study, the potential of RC systems based on TiN/Al2O3/Si3N4/SiO2/poly-Si (TANOS) is explored, utilizing the high-pressure annealing (HPA) process to enhance the performance of the device. Specifically, HPA-treated TANOS devices are employed in the readout layer to ensure stable long-term memory characteristics, while untreated TANOS devices are used in the reservoir layer, leveraging their short-term memory properties induced by interfacial traps. This study also investigates the feasibility of TANOS devices for neuromorphic computing. Based on Modified National Institute of Standards and Technology simulations, the complete TANOS-based RC system achieves a recognition rate of 84.48%, demonstrating its potential for temporal pattern recognition tasks.