LuxNAS: A Coherent Photonic Neural Network Powered by Neural Architecture Search

2025-03-05

AI Summary

To address the large footprint and high hardware overhead of conventional MZI-based Clements photonic neural networks (PNNs), this paper proposes a compact coherent PNN architecture leveraging phase-change material (PCM)-tunable couplers. We introduce neural architecture search (NAS) to photonic computing for the first time, enabling end-to-end joint optimization of network topology and physical device parameters—including coupling ratios and phase biases—while preserving backpropagation-compatible training and photonic weight encoding. Experimental results demonstrate an 85% reduction in chip area compared to the standard Clements architecture, with no accuracy loss on image classification tasks. The core contribution is a NAS-driven hardware-algorithm co-design paradigm for photonic systems, establishing a new pathway toward high-density, low-power integrated photonic intelligence.

Technology Category

• Machine Learning: Hardware-aware ML
• Computer Vision: Learning & Optimization for CV
• Search and Optimization: Learning to Search

Application Category

• Social Networks and Social Media: Generative AI / large language models and their impact on social systems
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Search and Retrieval-Augmented AI: Efficiency and scalability of Web search engines

Abstract

We demonstrate a novel coherent photonic neural network using tunable phase-change-material-based couplers and neural architecture search. Compared to the MZI-based Clements network, our results indicate 85% reduction in the network footprint while maintaining the accuracy.

Problem

• Develops a coherent photonic neural network
• Utilizes tunable phase-change-material-based couplers
• Reduces network footprint by 85% with maintained accuracy

Innovation

• Coherent photonic neural network
• Tunable phase-change-material couplers
• Neural architecture search optimization