Diffraction-Based Overlay Metrology With Optical Convolution Layer

Abstract

Overlay is a crucial indicator of manufacturing processing between layers. Currently, diffraction-based overlay (DBO) is widely adopted in overlay metrology. In response to existing challenges in DBO metrology, this study applied the concept of optical computing to establish a DBO metrology with an optical convolution layer that can predict the overlay of a DBO target, thereby providing a promising measurement method that increases the intuitiveness and efficiency of overlay metrology. In this study, pupil images from DBO metrology were directly output to the optical convolution layer and integrated with subsequent CNN layers in electronic hardware. A total of 630 pupil images with 21 types of designed overlay were generated for network training and testing. The experimental testing results revealed that the root mean square error of the predicted values of 90 pieces of testing pupil image data was 0.17 nm. The completed optical convolution layer is the fundamental unit for establishing an optical interconnection with the functions of weighting and matrix summation, which will contribute to further exploration of the DBO technique.

Keywords

• Convolution neural network (CNN)
• diffraction-based overlay (DBO)
• optical convolution
• optical neural network