SrTiO3-based memristor with a metal/oxide bilayer electrode for high recognition accuracy neuromorphic computing

Abstract

Memristors can modulate conductance, have multiple levels of storage, and have attracted significant attention in the field of artificial synapses. However, single-metal-electrode memristors are associated with the disadvantages of a low switching ratio and low stability, among others. Thus, herein, a metal/oxide bilayer electrode memristor device with the structure of Pt/La_0.7Sr_0.3MnO₃/SrTiO₃/Nb:SrTiO₃ (Pt/LSMO/STO/NSTO) was fabricated by inserting a transition metal oxide electrode, LSMO, between a Pt metal electrode and the resistive material STO. Oxygen vacancies in the LSMO layer could reduce the barrier height (Φ) and barrier width (W_d) of the STO/NSTO interface, resulting in a higher on/off ratio (1.2 × 10⁵), lower V_set (0.58 V) and higher stability (0.124/0.18) compared with a single-metal-electrode memristor without LSMO (on/off ratio = 9 × 10³, V_set = 0.9 V, and σ/μ = 0.23/0.25). In addition, it effectively simulated the features of artificial synapses and accomplished the function of a D-latch and decimal logic neuron computation. In particular, the convolutional neural network based on the metal/oxide bilayer memristor realized the high-precision recognition of traffic signals, demonstrating a high recognition rate of 95.4% for a traffic dataset, and its recognition accuracy remained above 80% even in 10% Gaussian noise.