Energy-Efficient Nanophotonic Neuromorphic Computing
Abstract
This proposed project pursues energy-efficient computing through a combination ofneuromorphic algorithms, architectures, and nanotechnologies integrated into a 3D systemdesigned as a reconfigurable hierarchical neural network. We will design, architect, and prototypethe nanophotonic neurons, hierarchical photonic neural networks, and the 3D computing system.We will compare its performance against computing systems based on IBM TrueNorth and von-Neumann machines. The project will involve the following tasks:• Develop the interconnect architecture and the routing algorithm to implement different deepneural networks,• Develop on-line training algorithms for the proposed neural networks,• Design and fabricate nanophotonic neurons and self-reconfiguring synaptic interconnectionfabrics using nanophotonic, nanoelectronic, and nanomems (NEMS) technologies,• Develop compact models for the proposed neurons to be used in an architecture-levelsimulator to evaluate the functional correctness of the neural networks,• Conduct comparative studies to evaluate energy efficiency and throughput of the computingsystem.Through these tasks, this project aims to advance the field of bio-inspired neuromorphiccomputing by means of nanophotonic, nanoelectronic, and nanomems (NEMS) technologies tobuild reconfigurable 2D-3D integrated circuits as hierarchical neural networks. The goal is todemonstrate underlying technologies to realize a computing system with the potential of achieving1000× improvements in energy-per-operation compared to state-of-the-art systems through acombination of novel approaches.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- May 30, 2018
- Source ID
- FA95501810186
Entities
People
- S.j. Ben Yoo
Organizations
- Air Force Office of Scientific Research
- United States Air Force
- University of California, Davis