Machine learning for rapid inference of critical dimensions in optical metrology of nanopatterned surfaces

Abstract

Scatterometry-based metrology has the capability to perform high-throughput inspection of large-area nanopatterned surfaces. It utilizes physics-based dependencies between reflectance spectra of light scattered from nanopatterned surfaces and the geometric parameters, or so-called Critical Dimensions (CDs), of such nanopatterns. This paper proposes a novel method for rapid and accurate inference of CDs of complex nanopatterned surfaces through the use of a Machine Learning (ML)-based inverse problem mapping, coupled with the use of a strategic feature selection algorithm, which also originated from the ML domain. Specifically, Latin Hypercube Sampling was utilized to create a multi-dimensional grid of CDs which are used as inputs for physics-based simulations of scattered light spectra for nanopatterned surfaces with those CDs. Simulation results are used to build the training data for the ML model relating simulated reflectance spectra with the corresponding CDs. The Extreme Gradient Boosting (XGBoost) algorithm was utilized to realize the aforementioned mapping. In addition, most informative spectral wavelengths were selected using Recursive Feature Elimination (RFE) algorithm based on the feature importance scores obtained from XGBoost. Capabilities of the newly proposed approach were evaluated through inspection of a semiconductor wafer sample with hourglass patterns, which are characterized by eight CDs. It was observed that the proposed method is capable of real-time CD metrology of large-area nanostructured surfaces with complex nanopatterns, with accuracy and repeatability comparable to that of Scanning Electron Microscopy.