Improved Measurement of Thin Film Thickness in Spectroscopic Reflectometer Using Convolutional Neural Networks

This research introduces a novel method of ensuring more reliable measurement of thin film thickness in spectroscopic reflectometer. Nonlinear fitting is the method most commonly used for measuring thin film thickness; however, it runs into the problem of a local minimum, which entails ambiguity. To improve measurement, prior to analysis of spectral reflectance profiles using nonlinear fitting, initial thickness value is estimated based on convolutional neural networks. Due to the supportive role of convolutional neural networks, thin film thickness can be determined without ambiguity.