BRAIN: Brain-Inspired Memristive Nanofiber Neural Networks

University Of Florida, Gainesville FL

Investigators

Abstract

The human brain is currently the most powerful information processor known to man. Recent advances in neural networks and network science indicate that in order to match the power and efficiency of the brain, there is a need for a brand-new type of neuromorphic hardware that is able to connect physically independent neurons with dedicated, modifiable synapses. The PI and coworkers have developed a brain-inspired concept where, preliminary theoretical and experimental results show that a mat of memristive nanofibers is anticipated to yield neural networks with enhanced connectivity, functionality, and overall performance. This project aims at assessing the potential of this concept and it is guided by the overarching fundamental question can these brain-inspired memristive nanofiber neural network (MN3) architectures be effectively used for advanced neuromorphic computing. The intellectual merit of the project stems from its goal to investigate the potential of MN3 architectures as the basis for novel neural network architectures that emulate the brain's computational abilities. This BRAIN project has three main objectives: 1) Manufacture MN3 architectures based on connective matrices of conductive-core, memristive-shell nanofibers and electrically characterize the networks in order to compare their characteristics to theoretical simulations; 2) Develop a simulation framework for modeling the proposed MN3 architectures in order to investigate and predict the signal behavior and computational properties of the networks; and 3) Investigate methods for implementing and training the networks as artificial neural networks, and evaluate the resulting networks on a set of benchmark machine learning tasks to determine performance characteristics. The broader impacts of the project can be summarized in four main areas: a) investigation of a new brain-inspired design paradigm for fabricating neural networks that, if successful, can potentially transform the broad fields of neuromorphic hardware and machine learning; b) advancement of the discovery and understanding of neural network architectures while training undergraduate and graduate students in STEM fields; c) dissemination of the gained scientific and technological understanding of memristive networks through professional conferences, peer-review publications, and online hubs; and d) dissemination of tutorials and workshops on Artificial Intelligence targeting the general population in collaboration with the Cade Museum.