Machine Learning-Based Damage Predicion Method for the Micro/Nano Structures Fabricated by Helium Focused Ion Beam

The helium focused ion beam (He-FIB) is a highly efficient method for micro/nano structure fabrication, while it also causes significant damages to the substrates. Through experiments, this paper summarizes the damage characteristic as a function of fabrication parameters of the He-FIB on the Si substrate. A model based on deep neural network (DNN) is proposed, which can predict the contours of four kinds of substrate damages: low damage, amorphization, small bubbles and large bubbles. The proposed model achieves >95% precision and recall under limited experimental data. Additionally, the model generalizes excellently to other energies and doses that are not presented in the training set, and the predictions are consistent with the experiments.