Overshoot‐Suppressed Memristor Crossbar Array with High Yield by AlOx Oxidation for Neuromorphic System

• DOI: 10.1002/admt.202400063
• Publication Date: 2024-04-08T17:10:11Z
• Authors: Sungjoon Kim; Kyungchul Park; Kyungho Hong; Tae‐Hyeon Kim; Jinwoo Park; Sangwook Youn; Hyungjin Kim; Woo Young Choi
• Keywords: Neuromorphic engineering; MNIST database; Memristor; Synaptic weight; Overshoot (microwave communication)

Abstract There is a need to design hardware synapse array appropriate for enhancing the efficiency of neuromorphic computing systems while minimizing energy consumption. This study introduces memristor device with an AlO x overshoot suppression layer (A‐OSL) achieve self‐compliance effect. By optimizing each cell within 16 × crossbar array, synaptic devices are successfully fabricated reliable characteristics and 3‐bit multilevel capabilities. In addition, oxygen composition TiO annealing conditions optimized reduce forming voltage minimize variation in switching voltage. As result, stable forming‐free obtained through A‐OSL insertion, reduction voltage, optimization. Also, target weights accurately transferred conducted inference process by applying spike signals following designated time step. The spiking neural network (SNN) demonstrated measuring vector‐matrix multiplication (VMM) array. VMM results exhibit classification accuracy 90.80% MNIST dataset, which close achieved software‐based approaches, amounting 91.85%.