Exploring the synaptic response of reactive Mn electrodes based TiO2 resistive switches

Abstract

Mn/TiO2/Mn devices, prepared by reactive sputtering and photolithography techniques, were characterized by analyzing their current–voltage (I-V) dependence, non-volatile memory properties, and artificial synapse behavior. The detailed study of its I-V characteristics allowed for highlighting the main conduction mechanisms involved in the electrical transport through the Mn-TiO2 junctions and determining an equivalent circuit model. These results show that the oxidation of metallic Mn electrodes and the application of electrical pulses produce a complex scenario associated with a highly inhomogeneous oxygen vacancy distribution. The resistance hysteresis switching loops were determined, as well as the synaptic-like weight depreciation and potentiation, revealing a linear dependence of the reset voltage as a function of the amplitude of the set voltage and a quasi-linear variation of the conductance with the number of applied pulses. Simulations based on spiking neural network architecture, considering different updates of the synaptic weights, were trained to learn handwriting patterns. Notably, those based on the linear learning rule of the Mn/TiO2/Mn devices outperformed others with increasing non-linear behavior, demonstrating both high recognition and noise tolerance factors, further highlighting the robustness of this approach.