Free-space optical neural networks (FSONNs) overcome the limitations of electronic integrated circuits, bringing speed, efficiency, and versatility to machine learning applications.
Machine learning has become a focus of the recent technological revolution thanks to artificial neural networks (ANNs), information processing systems designed to mimic the human brain. However, ANNs require high computational power that electronic integrated circuits can no longer meet.
Optical neural networks (ONNs) overcome this challenge. Instead of using electrons as the primary information carriers, ONNs use photons, which improves the speed, power, and scalability of the architecture. Inspired by their potential, Montes McNeil et al. compared two major classes of ONNs with traditional machine learning models:
“In our review, we explore optical computing, which we believe offers significant advantages over today's computing technologies,” said author Alex Montes McNeill. “Machine learning, and neural networks in particular, provide an excellent platform to demonstrate the benefits of optical computing architectures.”
The authors are particularly interested in free-space optical neural networks (FSONNs). In operation, information is encoded into FSONN light sources as input to the neural network, ready for passive computation via hidden layers. This allows for faster speeds and greater energy efficiency. Furthermore, various FSONN architectures, such as 3D printed layers, metasurfaces, and spatial light modulators, are promising for a variety of specialized applications.
Going forward, the authors look forward to designing new free-space optical components and extending the FSONN platform beyond machine learning applications.
“Researchers have already demonstrated some exciting alternative applications, including quantitative phase imaging, encryption, and viewing through random diffusers,” said Montes-McNeil. “We hope that researchers will push this platform well beyond these initial use cases and discover all kinds of new free-space optical computing systems for the future.”
sauce: “Fundamentals and Recent Developments of Free-Space Optical Neural Networks,” by Alexander Montez McNeill, Yuxiao Li, Allen Zhang, Michael Morbius, and Yongmin Liu. Journal of Applied Physics (2024). You can find this article here: https://doi.org/10.1063/5.0215752 .
