High-Dimensional Reconfigurable Integrated Photonic Computing
Abstract
With the fast development of artificial intelligence (AI) technologies, we have recently witnessed a rapid growth in possibilities and capabilities of smart machines and devices, especially in the field of intelligent sensing. The increasing demand for automated real-time decision making, self-adaption and self-validation requires an ability to analyze and process big data collected from the networked sensors, presenting a major challenge in high-speed reconfigurable computing to efficiently handle large, non-local sensing datasets with high throughputs. Reconfigurable computing-based accelerators can execute orders of magnitude faster computation-intensive operations such as matrix-vector multiplication or convolutions that are the fundamental operations used in artificial neuralnetworks. However, as the data complexity increases explosively, signal processing techniques and AI algorithms also require a technological explosion with respect to computational force, making it difficultfor the current electronic hardware implementation.The intellectual merit of this proposed activity is based upon our very recent success in the demonstration of the first in-situ photoniclearning using our lithography-free reconfigurable integrated photonic processor. In this project, we will conduct basic theoretical studies of light-matter interactions in the nonlinear regime, extensive numerical modeling and design of nanophotonics, state-of-the-art fabrication, and characterization of the integrated photonic structures, facilitating a holistic investigation to deliver thefirst high-dimensional reconfigurable integrated photonic processor for in-situ signal control, information processing, and generalphotonic computational operations for real-time sensing and decision making. We will target the fundamental mechanisms of light-matter interactions in active semiconductors and develop active strategies to rationally control and dynamically program high-dimensional parallel light pathways on demand for optical connectivity and information processing, identified in two research thrusts: 1) Explore the broadband optical gain associated with III-V semiconductors (e.g., InGaAsP) for hyperspectral information encoding/decodingusing the wavelength degree of freedom to enable high-dimensional reconfigurable photonic computing, enhancing the information density for high-throughput datasets; and 2) Investigate optical field programmable nonlinearity to execute nonlinear optical neural computing with spatially variant nonlinear activation functions, improving the precision and accuracy of neural photonic computing for networked intelligent sensors. If successful, the developed integrated photonic processors can be conveniently carried by soldiers and implemented on vehicles and warships of many navies to help facilitate real-time detection, classification, and identification ofair, sea-surface, and ground targets in all weather conditions. They can also enable large-scale sensor networks relating to satellites, ground vehicles, warships, soldiers, and other maritime and airborne platforms, and thus construct a mobile sensor network in the war-field.Approved for Public Release
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Oct 13, 2023
- Source ID
- N000142312882
Entities
People
- Liang Feng
Organizations
- Office of Naval Research
- United States Navy
- University of Pennsylvania