Aspect ratio and height estimation in critical dimension scanning electron microscope metrology using deep learning combined with features of proximity effect

To address the challenges of accurate metrology of the aspect ratio parameter and height of three-dimensional structures with critical dimension scanning electron microscopy, a method based on the features of the proximity effect and deep learning is proposed, and its validity is evaluated as well. Monte Carlo simulations are used to obtain a secondary electron signal, which reflects the morphology of the sample. To improve the efficiency of the method, supplemented functions have been made to the Nebula simulator for the creation of realistic geometrical structures with rounded corners and batch simulation process. Analysis on secondary electron signals from trapezoidal structures with different trench bottom widths, sidewall angles, and heights is undertaken, which shows the relation between geometrical parameters and proximity effects. These features of the proximity effect reflected in the signals are extracted by a fully connected residual neural network for aspect ratio and height prediction. To verify the generality of the method, electron beam illumination including vertical and titling conditions are investigated. The simulation results demonstrate the feasibility of the neural network to learn and predict the corresponding aspect ratio and heights.