A Microwave Photonic Processor for Convolutional Neural Networks With Increased Effective Speed of Convolution
Mahdi Chegini, Yiran Guan, Jianping Yao · Journal of Lightwave Technology · 2025
Due to the strong feature extraction capabilities, convolutional neural networks (CNNs) have been utilized for various tasks, including image recognition, object detection, and natural language processing. The primary computational demand of CNNs stems from the convolution operations. In this paper, we propose a novel microwave photonic processor to accelerate the convolution operations in a CNN by increasing the effective speed of convolution. Thanks to the novel system architecture and the associated serialization approach, the effective speed is increased. Specifically, for a CNN with an M×M kernel size, the effective speed is increased by M times. The proposed processor is experimentally tested in which the MNIST and Fashion MNIST datasets are employed for its performance evaluation. The increase in the effective speed of convolution is experimentally confirmed. A computing speed of 102.4 giga operations per second (GOPS) with a root mean squared error (RMSE) of 0.0110 is demonstrated. In addition, the accuracies for the MNIST and Fashion MNIST image classification tasks are 98% and 88%, respectively.