• A novel near-sensor edge computing system integrates AlN microrings for photonic feature extraction and Si Mach–Zehnder interferometers for photonic neural network operations, achieving real-time AI processing.
• Demonstrates high classification accuracy (96.77% for gestures, 98.31% for gaits) with low latency (<10 ns) and minimal energy consumption (<0.34 pJ).
• Enables low-power, high-speed AI applications with seamless hybrid photonic-electronic integration on a bilayer AlN/Si waveguide platform.
• Bridges the gap between AI models and real-world applications, enabling efficient, privacy-preserving AI solutions for healthcare, robotics, and next-generation human–machine interfaces.