Demonstration of cognitive learning, associative learning, and multi-bit reservoir computing using TiOx/HfOx-based volatile memristor with low current

Abstract

In this study, the TiN/TiOx/HfOx/Pt memristor with short-term memory (STM) and self-rectifying properties is characterized for multiple computing functions. The STM properties of the devices are detected by the retention test. The emulation of synaptic memory and forgetfulness by STM effects is demonstrated using paired-pulse facilitation. We also emulate various synaptic behaviors such as several excitatory post-synaptic currents and Pavlovian associative learning. The TiN/TiOx/HfOx/Pt configuration of this device replicates key functions of biological nociceptors for sensory memory. Emulation includes important aspects such as threshold, relaxation, "Hyperalgesia" and "Allodynia." Finally, efficient training reservoir computing is demonstrated in artificial neural network simulations, composed of physical storage memristor devices with 16 and 128 states (4-bit and 7bit), and a readout layer, yielding respective pattern recognition accuracies of 97.73 % and 96.77 % for the dataset.