Memristive Concept of a High-Dimensional Neuron

Abstract

We propose a novel concept of a hardware implementation of high-dimensional neurons with the aim of a functional simulation of the human hippocampus. To implement a high-dimensional neuron, we suggest using the arrays of trainable memristive devices as part of the custom-made software-hardware system to provide full neuronal functionality with fine-tuning of synaptic weights according to the self-learning rule. The proposed architecture can enable new efficient approaches to processing big data in artificial intelligence (AI) systems and a new generation of robots.