Deep Learning for Thin Film Thickness Measurement in Spectroscopic Reflectometry
Xiaolong Cheng, Yan Tang, Kejun Yang, Chenhaolei Han
Abstract
Spectroscopic reflectometry is widely used in industry to measure film thickness. However, it is still a challenge to quickly obtain accurate thickness information from films with less than 1 micron thickness. In order to achieve fast demodulation of thin-film signals, this letter proposes a signal demodulation method based on deep learning techniques for thin-film measurements with optical path differences of 100–2000 <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$nm$ </tex-math></inline-formula> . Here, a reflectivity generation model is used to generate a training dataset. The trained neural network can directly recover thickness information from the reflectance spectrum and the speed is 10 times faster than the traditional method. Simulation results show that this method has strong noise robustness. Experimental results indicate that the deep learning method is reliable in different situations and the average error is less than 3 <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$nm$ </tex-math></inline-formula> .