Related Experiment Video
Updated: Aug 15, 2026
12:18
Microwave Photonics Systems Based on Whispering-gallery-mode Resonators
Published on: August 5, 2013
Reconfigurable nonlinear activation functions in an optomechanical whispering-gallery-mode microcavity
Optics Express
|August 14, 2026
Summary
Researchers developed reconfigurable all-optical nonlinear activation functions using optomechanics. These functions enhance optical neural network performance, achieving high accuracy on image recognition tasks.
Area of Science:
- Photonics
- Artificial Intelligence
- Quantum Optics
Background:
- Optical neural networks (ONNs) offer parallelism for AI.
- All-optical nonlinear activation functions (ONAFs) are crucial for ONN performance.
- Existing ONAFs face challenges in reconfigurability and efficiency.
Purpose of the Study:
- To theoretically investigate reconfigurable ONAFs using optomechanically induced transparency (OMIT).
- To demonstrate tunable nonlinear activation responses (saturating, ReLU-like) via all-optical control.
- To evaluate the performance of OMIT-based ONAFs in compact neural network simulations.
Main Methods:
- Theoretical investigation of OMIT in a whispering-gallery-mode microcavity.
- All-optical tuning of activation functions with microwatt thresholds.
- Simulation of compact neural networks and convolutional neural networks (CNNs).
Main Results:
- Demonstrated distinct, continuously tunable nonlinear activation responses.
- Achieved improved classification performance on binary tasks with nonlinear boundaries.
- Attained high accuracies on MNIST (98.93%) and Fashion-MNIST (91.16%) using CNNs.
Conclusions:
- The optomechanical effect provides a promising mechanism for on-chip reconfigurable ONAFs.
- OMIT-based ONAFs are suitable for photonic neural computing.
- The proposed system offers efficient and tunable nonlinear activation for AI acceleration.
