NANOPHOTONIC NEURAL NETWORKS WITH NONLINEAR, RECONFIGURABLE MMETASURFACES
Abstract
The highly interconnected nature of the brain enables robust and unintuitive computation with extremely low power consumption. Mimicking this network in engineered hardware requires both highly nonlinear interactions (akin to neuron spiking) and highly interconnected artificial neurons - the two hallmarks of neuromorphic computing. Nonlinear interactions can be achieved in electronic signals, but the required dense interconnection fundamentally limits their speed. Integrated nanophotonics, on the other hand, promises enormous speed and data capacity, provided that nonlinearity, reconfigurability, and high interconnectivity can be achieved. The objective of this proposal is to create a suite of nonlinear, reconfigurable and highly interconnected nanophotonic materials for optical neuromorphic processing. In particular, we will design and fabricate arrays of CMOS-compatible optical nanoantennas, or metasurfaces, that can precisely control light absorption and directional scattering, adapting to varying environmental inputs to learn from their environment. These neuromorphic nanophotonic networks will be, like neurons, highly nonlinear; they can be taught rather than programmed; and they can be electrically reconfigured to execute a desired task. Because our approach relies on photonics, it promises processing speeds that are nearly 100 million times faster than biological neurons (approaching the speed of light), with an estimated processing efficiency eight orders of magnitude beyond that of current microelectronic supercomputers. Our technical approach combines state-of-the-art nonlinear and multiscale electromagnetic simulations with advanced nanofabrication and characterization. We will design metasurface ‘neurons’ that exhibit high quality-factors; demonstrate their highly-efficient nonlinear response with continuous-wave illumination; and dynamically reconfigure metasurfaces through electro-optic and gain-based media.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Aug 12, 2021
- Source ID
- FA95502010120
Entities
People
- Jennifer Dionne
Organizations
- Air Force Office of Scientific Research
- Stanford University
- United States Air Force