Photonic edge intelligence chip for multi-modal sensing, inference and learning.

Area of Science:

• Photonics
• Edge Computing
• Artificial Intelligence

Background:

• Edge computing demands real-time processing of high-throughput analog signals, a challenge for conventional electronics.
• Integrated photonics provides low-latency processing but struggles with direct raw analog data handling.

Purpose of the Study:

• To develop a photonic edge intelligence chip (PEIC) that fuses multiple analog signal modalities for on-chip processing.
• To demonstrate an end-to-end optical neural network for energy-efficient inference at the edge.

Main Methods:

• Fusing images, spectra, and radio-frequency signals into broad optical spectra for single-fiber input.
• Utilizing an arrayed waveguide grating (AWG) for spectral sensing and energy-efficient convolution (29 fJ/OP).
• Implementing a nonlinear activation layer and a fully connected layer to form an optical neural network.

Main Results:

• Achieved on-chip inference with a measured response time of 1.33 nanoseconds.
• Demonstrated supervised and unsupervised learning on drug spectral recognition, image classification, and radar target classification.
• The PEIC processes fused analog signals with high energy efficiency.

Conclusions:

• The photonic edge intelligence chip unifies analog signal acquisition and optical computation for edge intelligence.
• This work presents a pathway for on-chip solutions enabling faster and more efficient edge AI applications.
• The developed chip addresses the limitations of conventional electronics in handling high-throughput analog data at the edge.