On the area scalability of valence-change memristors for neuromorphic computing

• Main Authors: Ang, Diing Shenp, Zhou, Yu, Yew, Kwang Sing, Berco, Dan
• Other Authors: School of Electrical and Electronic Engineering
• Format: Article
• Language: English
• Published: 2020
• Subjects: Engineering::Electrical and electronic engineering; Artificial Neural Networks; Atomic Force Microscopy
• Online Access: https://hdl.handle.net/10356/143153
• Institution: Nanyang Technological University

The ability to vary the conductance of a valence-change memristor in a continuous manner makes it a prime choice as an artificial synapse in neuromorphic systems. Because synapses are the most numerous components in the brain, exceeding the neurons by several orders of magnitude, the scalability of artificial synapses is crucial to the development of large scale neuromorphic systems but is an issue which is seldom investigated. Leveraging on the conductive atomic force microscopy method, we found that the conductance switching of nanoscale memristors (∼25 nm2) is abrupt in a majority of the cases examined. This behavior is contrary to the analoglike conductance modulation or plasticity typically observed in larger area memristors. The result therefore implies that plasticity may be lost when the device dimension is scaled down. The contributing factor behind the plasticity behavior of a large-area memristor was investigated by current mapping, and may be ascribed to the disruption of the plurality of conductive filaments happening at different voltages, thus yielding an apparent continuous change in conductance with voltage. The loss of plasticity in scaled memristors may pose a serious constraint to the development of large scale neuromorphic systems.