Artificial synaptic characteristics of TiO₂/HfO₂ memristor with self-rectifying switching for brain-inspired computing

In this work, a built-in selector synaptic memristor is proposed to minimize sneak current in synapse array. The Ti/TiO₂/HfO₂/Si device exhibits nonlinear and high rectifying current-voltage characteristics compared to Ti/TiO₂/Si device. Under negative bias conditions, the physical model supported by material analysis reveals that the Schottky barrier at the interface between Ti and TiO₂ suppresses the current. Due to nonlinear and self-rectifying characteristics, a bilayer device has a significantly larger array size (> 10⁵ × 10⁵) with a securing read margin of 10%, using the modified half-bias scheme. A multi-level interface-type switching behavior is desired to implement synaptic functions, including potentiation and depression, and spike-timing-dependent plasticity. It was verified that the recognition rate in the bilayer device is significantly more accurate in a neural network model employing the Fashion MNIST dataset.