WDM-enabled photonic edge computing with low cost and high performance

Photonic computing enables high bandwidth, low latency and energy efficient processing. This work proposes a photonic edge computing architecture that leverages wavelength-division multiplexing (WDM) to distribute cloud-managed neural network weights through existing optical line terminal (OLT) apparatus, facilitating lightweight deployment and real-time photonic inference. System-level simulations demonstrate that a single optical interference unit suffices to construct the inference module, improving scalability and reducing hardware cost. In the MNIST image classification task, the architecture achieves a recognition accuracy of 98.23%, with single-sample accuracy consistently above 96.6%, validating its efficiency and application potential in future photonic neural computation.