Optical ReLU-like activation function based on a semiconductor laser with optical injection.

Area of Science:

• Optoelectronics
• Photonics
• Artificial Intelligence Hardware

Background:

• Artificial neural networks (ANNs) rely on linear and nonlinear operations.
• Optical neural networks (ONNs) excel at linear operations but struggle with nonlinear activation functions.
• Implementing optical activation functions is a key challenge in advancing ONNs.

Purpose of the Study:

• To present a novel optical Rectified Linear Unit-like (ReLU-like) activation function.
• To demonstrate the feasibility of optical nonlinear activation functions in semiconductor lasers.
• To enable reconfigurable optical neural network architectures.

Main Methods:

• Utilized a semiconductor laser subjected to optical injection.
• Achieved a 180° rotated ReLU-like function above the Hopf bifurcation in the injection-locking diagram.
• Operated the system in continuous-wave mode.

Main Results:

• Demonstrated a functional optical ReLU-like activation function.
• Showcased the reconfigurability of the activation function's slope by tuning laser frequency differences.
• Confirmed operation in a broad regime above the Hopf bifurcation.

Conclusions:

• The developed optical ReLU-like activation function overcomes a major hurdle in optical neural network design.
• This work paves the way for more powerful and efficient all-optical computing systems.
• Reconfigurable optical activation functions are achievable with semiconductor lasers.