Ultra-low latency integrated optical sensing-compression-inference system

Distributed acoustic sensing (DAS) enables long-range, continuous monitoring, yet the real-time analysis of its massive spatiotemporal datasets remains a formidable challenge. As electronic processing faces bottlenecks from the slowing of Moore’s law, optical neural networks offer a promising high-speed, energy-efficient alternative. However, their integration with DAS is impeded by a fundamental dimensionality mismatch between the extensive DAS data and the limited input capacity of photonic computing chips. Furthermore, real-world deployment requires adaptive learning to handle dynamic environmental variations. Here, we report a hybrid architecture that synergizes optical sensing, compression and inference. We develop two integrated silicon photonic circuits: a DAS interrogator and a reconfigurable diffractive neural network achieving classification task. Notably, this architecture achieves an overall inference latency of just 0.1 ms—an order of magnitude lower than conventional electronic systems—with significant potential for further reduction. Crucially, we employ a free-space diffraction interface for nonlinear optical compression to bridge the dimensionality gap, improving classification accuracy by 30% over linear amplitude compression. Leveraging system reconfigurability and rapid online training algorithms, this adaptive framework facilitates robust, real-time event analysis, establishing a scalable path for next-generation intelligent DAS.