Nanophotonic Devices for Fast and Efficient LIDAR
Abstract
Light detection and ranging (LIDAR) provides some of the best spatial precision for remote sensing. Current lidar systems, however, are slow and operate on millisecond timescales. One of the main bottleneck to these systems is the need to convert large amounts of data from optical to electronic signals. All-optical methods for signal processing could eliminate this bottleneck enabling ultra-fast lidar systems that can work in real-time applications that require fast response. This proposal aims to develop ultra-fast and efficient lidar systems by processing signals in the optical domain. The PI will investigate the potential for all-optical neural networks that can perform rapid machine learning tasks in real time. Each neural network will be composed of a complex linear optical switch network that can perform fast matrix multiplication, followed by an array of nonlinear nanophotonic cavities. Each nonlinear cavity contains an ensemble of quantum dots that act as saturable absorbers that produce a strong nonlinear optical response. The proposal will address the following scientific objectives: 1) Design, fabricate, and test a low-power optical nonlinear neural network node using quantum dots coupled to a nanophotonic cavity. This node will implement the nonlinear function using saturable absorption or saturable gain. We will characterize the switching power, loss, and robustness of these devices 2) Develop new fabrication methods to integrate III-V materials with integrated photonics based on Si and lithium niobate, the most advanced photonic platforms for large-scale integration. 3) Perform a detailed systems-level numerical analysis and simulation of optical neural network architectures and compare them to software and electronic implementations. The goal of this thrust is to evaluate the anticipated performance improvements and key performance metrics required to achieve a scalable all-optical neural network. This thrust will draw heavily from measurements performed in the other two thrusts to attain practically achievable performance metrics. Success of this project would set the stage for a latency-free neural network that can quickly acquire lidar signal and implement various machine-learning tasks such as classification, targeting, and prediction in real-time. The ability such tasks with ultra-high response times would be a critical enabler for targeting, autonomous vehicles, and other ranging applications that require fast response time. Optical machine learning is rapidly advancing, and a comprehensive understanding of new and emerging technologies and their impact is crucial to provide a realistic analysis of itÕs feasibility. This program will aim to provide a definitive analysis at the systems level which is critically missing and poorly understood. The PI has extensive knowledge in photonics, device integration, nonlinear optics, and materials putting him in an ideal position to carry out the proposed work.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jun 25, 2019
- Source ID
- W911NF1910378
Entities
People
- Edo Waks
Organizations
- Army Contracting Command
- United States Army
- University of Maryland