Abstract

New image recognition approach are proposed for common grating coupler alignment issues. This proposal visualizes the grating coupler by observing the diffraction of visible light on grating surface when it is blocked by fiber/fiber array. Then use machine learning to achieve current position prediction. In this paper, we successfully verified that visible light can be used for grating coupler alignment. And use GoogLeNet as a learning network for machine learning. The results show that the learned network can successfully predict the distance range between the current position and the optimal coupling point position with accuracy of 85%.

© 2022 IEEE