Hybrid optoelectronic neural networks using a mutually pumped phase-conjugate mirror

Area of Science:

• Optoelectronics
• Neural Network Hardware
• Photorefractive Optics

Background:

• Optical interconnections are crucial for high-performance neural networks.
• Photorefractive crystals offer potential for implementing complex optical systems.
• Bragg degeneracies can cause cross talk in volume holographic systems.

Purpose of the Study:

• To propose and evaluate an optical interconnection method for neural networks.
• To mitigate cross talk issues in photorefractive-based optical neural networks.

Main Methods:

• Utilizing mutually pumped phase conjugation in a photorefractive crystal.
• Implementing angular and spatial multiplexing of volume gratings.
• Storing individual connection weights within a continuum of gratings.

Main Results:

• Demonstrated a method to reduce cross talk caused by Bragg degeneracies.
• Successfully stored connection weights using multiplexed volume gratings.
• Achieved an optical interconnection scheme for neural networks.

Conclusions:

• The proposed method effectively reduces cross talk in optical neural networks.
• Mutually pumped phase conjugation in photorefractive crystals is a viable approach for neural network interconnections.
• Multiplexing techniques enhance the performance and scalability of optical neural network hardware.