Improvement of volatile switching in scaled silicon nanofin memristor for high performance and efficient reservoir computing.

Abstract

Conductive-bridge random access memory can be used as a physical reservoir for temporal learning in reservoir computing owing to its volatile nature. Herein, a scaled Cu/HfOx/n+-Si memristor was fabricated and characterized for reservoir computing. The scaled, silicon nanofin bottom electrode formation is verified by scanning electron and transmission electron microscopy. The scaled device shows better cycle-to-cycle switching variability characteristics compared with those of large-sized cells. In addition, synaptic characteristics such as conductance changes due to pulses, paired-pulse facilitation, and excitatory postsynaptic currents are confirmed in the scaled memristor. High-pattern accuracy is demonstrated by deep neural networks applied in neuromorphic systems in conjunction with the use of the Modified National Institute of Standards and Technology database. Furthermore, a reservoir computing system is introduced with six different states attained by adjusting the amplitude of the input pulse. Finally, high-performance and efficient volatile reservoir computing in the scaled device is demonstrated by conductance control and system-level reservoir computing simulations.