Development of a Light-Speed Neural Network Using a Plasma Metamaterial Device
Department of Mechanical Engineering, Stanford University
Neural networks have long fascinated scientific communities, serving as an elegant approach to machine learning modelled on our understanding of the human brain. Their increasing sophistication has led to the near ubiquitous role of machine learning – from transportation management to medical diagnostics to the voice recognition powering virtual assistants. The expansive reliance on machine learning now calls for radical thinking to address the ever-greater demand for computational power and speed. This research project with the Stanford Plasma Physics Lab (SPPL) seeks to develop a novel, physical neural network using plasma rods with the ability to process information at the unprecedented speed of light, eclipsing any current computer powered neural network. Simulations previously conducted by the SPPL confirmed that plasma rods could efficiently guide electromagnetic waves. Our project continues this work with the experimental implementation of these simulations; practically this means finding a way to individually tune each plasma rod from a single power source. A light-speed neural network using plasma holds the potential to improve the speed and power-efficiency of machine learning with transformative implications across multiple fields. This project marks another exciting collaboration between Stanford’s SPPL and CentraleSupélec, two institutions at the vanguard of plasma research.