Silicon photonic convolution operator exploiting on-chip nonlinear activation function.

Area of Science:

• Photonics and Optical Computing
• Artificial Intelligence and Machine Learning

Background:

• Nonlinear activation functions (NAFs) are crucial for artificial neural networks (ANNs) but often require separate optoelectronic or digital components, hindering optical computing efficiency.
• Existing NAF implementations limit the advantages of optical computing by introducing latency and system complexity.

Purpose of the Study:

• To propose and demonstrate the first implementation of a novel NAF using an electro-optic IQ modulator integrated onto a silicon photonic chip.
• To validate the effectiveness of this integrated NAF in a real-world machine learning application.

Main Methods:

• Developed a novel nonlinear modulation process using an electro-optic IQ modulator on a silicon photonic convolution operator chip.
• Constructed a convolutional neural network (CNN) incorporating the integrated NAF for radio machine learning classification tasks.

Main Results:

• Achieved 92.5% accuracy in radio machine learning classification, a 27% improvement compared to a CNN without the NAF.
• Demonstrated near-synchronous execution of the NAF with the convolution operation, significantly reducing latency.
• Showcased reduced peripheral control system complexity due to the fully integrated silicon photonic chip.

Conclusions:

• The integrated silicon photonic NAF offers a significant advancement for optical neural network computation.
• This approach enables high-accuracy, low-latency machine learning classification on-chip, paving the way for large-scale optical neural networks.