Double barrier memristive devices for neuromorphic computing

Abstract

The intensified development of memristive devices for memory applications in recent years lead to the advancement of several other possible fields of operation. Among them, the use of memristive devices for neuromorphic computing is one of the most promising applications. Here, we present an especially for neuromorphic computing attractive quantum mechanical memristive device which offers the benefits of an intrinsic current compliance, a gradual resistance change, and no need for an initial electric forming procedure. Our findings indicate that a homogenous interfacial effect is responsible for the observed memristive I-V curves rather than locally confined filaments. The layer sequence of the investigated device is Nb/Al/Al2O3/NbxOy/Au. The layer thickness of the Al2O3 tunnel barrier and the adjacent NbxOy solid state electrolyte layer are 1.3 nm and 2.5 nm, respectively. Thus it is possible to mutually affect the probability of electron tunneling through Al2O3 and the height of the Schottky NbxOy/Au barrier by oxygen migration (drift-diffusion). For this purpose the important issues of the respective energy barriers and ultra-thin memristive layers are investigated. Experimental findings are supported by an equivalent circuit model which furthermore provides a better understanding of the underlying physical mechanisms and allows the identification of essential fabrication parameters. Electrical characteristics are investigated and discussed in the framework of their capability to emulate synaptic functionality. Finally, the pros and cons of the double-barrier devices are discussed with respect to their possible applications in novel neuromorphic circuits.