Rapid configuring method for a programmable photonic integrated circuit based on a tandem neural network.

Area of Science:

• Photonics
• Artificial Intelligence
• Optical Computing

Background:

• Programmable photonic integrated circuits (PPICs) are promising for optical analog matrix multiplication.
• Efficient voltage configuration for PPICs remains a significant developmental challenge.

Purpose of the Study:

• To propose a rapid, black-box method for predicting voltage configurations of arbitrary matrices in PPICs.
• To demonstrate the practical application of this method in optical-electric hybrid models.

Main Methods:

• A tandem neural network approach was employed for black-box prediction of voltage configurations.
• Experimental validation was performed on a 4x4 PPIC.
• An optical-electric hybrid model was implemented using the developed method.

Main Results:

• The method achieved an average fidelity of 0.989 for 10,000 matrices on a 4x4 PPIC.
• The optical-electric hybrid model attained a 97.59% training accuracy on the MNIST dataset.

Conclusions:

• The proposed black-box method effectively predicts voltage configurations for PPICs, overcoming a key challenge.
• This approach enables efficient optical matrix multiplication and facilitates the development of advanced optical-electric hybrid systems.