Nonlinear Processing with Linear Optics

Abstract: Abstract A mechanically programmable diffractive neural network based on Pancharatnam‐Berry (PB) phase metasurfaces, consisting of rotatable PB phase meta‐atoms as fundamental building blocks is presented. By precisely controlling the rotation angle of each programmable meta‐atom in a mechanical way, modulations of the PB phase distribution are achieved, and flexible programmability of the network to adapt to diverse tasks is enabled. The system seamlessly integrates the low‐power consumption advantage of passive networks with the flexibility of programmable networks, achieving orders‐of‐magnitude reduction in energy consumption while maintaining optimal performance balance. Experimental results demonstrate the system's mechanical programmability with high‐precision classification ability (100% test accuracy) in multi‐task operations, including zodiac sign recognition and handwritten digit classification. The proposed MP‐DNN operates without an external energy supply during the matrix computations and consumes only minimal power during task switching, thus offering an energy‐efficient and low‐power solution for reconfigurable diffractive neural networks.

TL;DR: Researchers develop diffractive processors that can perform large-scale, parallel nonlinear computation using linear optics, enabling universal approximation of nonlinear functions, including those used in neural networks, at unprecedented spatial densities and speeds.

TL;DR: Researchers develop a spatial frequency domain training method to create arbitrary analog convolution kernels using optical metasurfaces, achieving 98.59% accuracy on MNIST and outperforming digital convolution kernels in machine vision tasks.

Abstract: Photonic neural networks (PNNs), which share the inherent benefits of photonic systems, such as high parallelism and low power consumption, could challenge traditional digital neural networks in terms of energy efficiency, latency and throughput. However, producing scalable photonic artificial intelligence (AI) solutions remains challenging. To make photonic AI models viable, the scalability problem needs to be solved. Large optical AI models implemented on PNNs are only commercially feasible if the advantages of optical computation outweigh the cost of their input–output overhead. In this Perspective, we discuss how field-programmable metasurface technology may become a key hardware ingredient in achieving scalable photonic AI accelerators and how it can compete with current digital electronic technologies. Programmability or reconfigurability is a pivotal component for PNN hardware, enabling in situ training and accommodating non-stationary use cases that require fine-tuning or transfer learning. Co-integration with electronics, 3D stacking and large-scale manufacturing of metasurfaces would significantly improve PNN scalability and functionalities. Programmable metasurfaces could address some of the current challenges that PNNs face and enable next-generation photonic AI technology. Programmable metasurfaces may offer a transformative approach to scalable photonic neural networks by overcoming key hardware limitations. This Perspective explores their potential to enhance energy efficiency, computation speed, and adaptability, positioning them as a promising alternative to traditional digital artificial intelligence hardware.

TL;DR: This paper demonstrates how constraints from the task domain can be integrated into a backpropagation network through the architecture of the network, successfully applied to the recognition of handwritten zip code digits provided by the U.S. Postal Service.

TL;DR: Fashion-MNIST is intended to serve as a direct drop-in replacement for the original MNIST dataset for benchmarking machine learning algorithms, as it shares the same image size, data format and the structure of training and testing splits.

TL;DR: Larger models are significantly more sample-efficient, such that optimally compute-efficient training involves training very large models on a relatively modest amount of data and stopping significantly before convergence.

TL;DR: A new architecture for a fully optical neural network is demonstrated that enables a computational speed enhancement of at least two orders of magnitude and three order of magnitude in power efficiency over state-of-the-art electronics.