Neural network using longitudinal modes of an injection laser with external feedback.

Area of Science:

• Optoelectronics
• Computational Neuroscience
• Laser Physics

Background:

• Traditional neural networks face limitations in speed and power consumption.
• Optical computing offers a potential solution for high-performance computation.
• Injection lasers provide a tunable platform for optical signal processing.

Purpose of the Study:

• To introduce a new optical neural network concept based on injection lasers.
• To demonstrate the feasibility of using laser modes as neurons and feedback as weights.
• To explore the potential for high-speed optical computation.

Main Methods:

• A simple laser model was developed to describe the optical neural network.
• Numerical simulations using laser rate equations were performed.
• Experiments were conducted using Gallium Aluminum Arsenide (GaAlAs) injection lasers.
• Stochastic learning was employed to determine optimal feedback masks.
• Winner-take-all and exclusive-or operations were implemented and tested.

Main Results:

• The simple laser model's predictions were validated by numerical simulations and experiments.
• The optical neural network successfully performed winner-take-all and exclusive-or operations.
• A processing speed exceeding 10 giga-operations per second (GCPS) was achieved.
• The potential for teracycles per second (TCPS) processing in integrated circuits was projected.

Conclusions:

• The proposed optical neural network architecture is a viable approach for high-speed computation.
• Injection lasers offer a promising platform for developing advanced optical neural network hardware.
• Further development in optoelectronic integrated circuits could lead to significant advancements in neural network speed and efficiency.