Nonlinear Computation with Linear Optics via Source-Position Encoding

2025-04-29

AI Summary

Optical neural networks (ONNs) face a fundamental challenge in realizing energy-efficient nonlinear activation functions without resorting to power-hungry or bandwidth-limited nonlinear optical materials. Method: We propose a novel paradigm that eliminates the need for nonlinear media by encoding nonlinearity differentiably through the spatial position of coherent light sources within a purely linear photonic medium. Our approach integrates source-position encoding with continuous-domain topology optimization, differentiable photonic simulation, and hardware-algorithm co-training into an end-to-end photonic neural network design framework. Contribution/Results: Experimental evaluation on image classification demonstrates that our method significantly outperforms conventional linear ONNs and achieves accuracy comparable to standard electronic artificial neural networks—while maintaining full optical implementation and ultra-low power consumption. This work establishes a new pathway toward high-efficiency photonic computing by redefining nonlinearity as a geometric, differentiable degree of freedom rather than a material property.

Technology Category

• Search and Optimization: Non-convex Optimization
• Computer Vision: Learning & Optimization for CV
• Machine Learning: Optimization

Application Category

• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• 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

Optical computing systems provide an alternate hardware model which appears to be aligned with the demands of neural network workloads. However, the challenge of implementing energy efficient nonlinearities in optics -- a key requirement for realizing neural networks -- is a conspicuous missing link. In this work we introduce a novel method to achieve nonlinear computation in fully linear media. Our method can operate at low power and requires only the ability to drive the optical system at a data-dependent spatial position. Leveraging this positional encoding, we formulate a fully automated, topology-optimization-based hardware design framework for extremely specialized optical neural networks, drawing on modern advancements in optimization and machine learning. We evaluate our optical designs on machine learning classification tasks: demonstrating significant improvements over linear methods, and competitive performance when compared to standard artificial neural networks.

Problem

• Implementing energy-efficient optical nonlinearities for neural networks
• Achieving nonlinear computation in fully linear optical media
• Designing specialized optical neural networks via topology optimization

Innovation

• Nonlinear computation via source-position encoding
• Low-power operation in linear optical media
• Topology-optimized automated optical network design