Synaptic Plasticity Mimicked In Phase Change Memristors Based On Nb:AlNO Film

Currently,brain inspired computing needs to find elements resembling either neuron or synapse in micro kinetics.Modulating synaptic behaviors with memory devices has attracted considerable attention,especially through designing a new devices as elementary cells to construct artificial networks and achieve neuromorphic computing.In this paper,dilute Nb-doped Al ON is fabricated for this purpose,further Pd/Nb:Al NO/Pd memristor and Pd/Nb:Al NO/Al NO/Nb:Al NO/Pd memristor.We studied their electrical performances and mechanism of resistance switching.The Ⅰ-Ⅴ performances of Pd/Nb:Al NO/Pd memristor show that the positive The external stimulations prompt the growth of nano-filament composed of Nb oxide nano-clusters towards the top electrode and conduction increase,the negative stimulations prompt the withdraw of the filament and conduction decrease.Periodic undulations are found to be loaded on either the conduction increase or the conduction decrease trend with time constants of micro-second.The microstructure study of Nb:Al NO film confirms that the filaments are controlled by both oxygen and heat diffusion.The local heat induced by current flow let the formation of Nb oxide nano-crystallite firstly and enhance the overall conduction subsequently.However,the consecutive positive stimulations exhaust the neighbored oxygen so that the growth of filaments is restrained for the moment.The locally accumulated heat transforms the crystalline phase to amorphous phase to reduce the conduction.On the contrary,the consecutive negative stimulations decrease the size of nano-clusters and phase change kinetics still works.Therefore,the interaction of these two types of kinetics,which resembles the kinetics of neuron transmitter and voltage gated ionic channels,undulate the long-term trends of conduction.Pd/Nb:Al NO/Pd memristor successfully simulate conventional learning protocols including spike-rate dependent plasticity and spike-time dependent plasticity.While Pd/Nb:Al NO/Al NO/Nb:Al NO/Pd memristor successfully simulate pike-rate dependent plasticity and frequency selectivity.In addition,we demonstrate that two advantages of our memristor system.One is that the very small current density reduces the power consumption significantly.The other is that the interaction of two kinetics establish a possibility of synthesizing signals with different frequencies,including both encodingand decoding methods.We trust that our artificial synapse will contribute to neuromorphic computation greatly.