Photonic in-memory accelerators for low-latency and efficient computing
Abstract
This proposal aims to advance the frontier of nano-photonics for information processing by developing a modular photonic computing engine which offers transformative advantages in processing speed and throughput over state of the art electronic computing hardware. Existing computing hardware is fundamentally limited by the movement of data between memory and processing cores which limits computing speed and causes inefficient heating of metal interconnects. The combination of optical interconnects which do not experience capacitive losses with process in memory techniques could overcome this fundamental speed and efficiency bottleneck. The overall objective of this proposal is to address these key challenges in the field of computing by developing a modular computing platform comprised of nonvolatile optical memory arrays which can be cascaded to form ultra low latency processing blocks within a neural network. The proposed technical plan builds upon the PI s pioneering work in the field of phase change materials and optical computing, while addressing crucial scientific challenges related to cyclability, efficiency, and scalability of these nanophotonic devices. The PI will conduct research under the following three main thrusts- 1) improving the efficiency, reliability, and repeatability of electrically programmable phase change photonic memory; 2) designing fully analog multilayer photonic networks for fast and efficient computing; and 3) demonstrating a multi layer, fully analog photonic in memory accelerator on chip. The outcomes of this work will advance the development of novel materials for reconfigurable photonic devices and integrate these components into optoelectronic computational systems. The resulting platform is expected to have significant impact for Air and Space Force applications requiring ultra low latency computation, target discrimination, and autonomous navigation where there is an immediate need for extremely high speed information processing.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Feb 05, 2025
- Source ID
- FA95502410064
Entities
People
- Nathan Youngblood
Organizations
- Air Force Office of Scientific Research
- United States Air Force
- University of Pittsburgh