A Method of Predicting the Deposited Film Thickness in IC Fabrication Based on Automatic Machine Learning

The semiconductor IC (Integrated Circuits) industry involves a multitude of fabrication processes, and quality verification at critical stages is crucial for ensuring successful post-processing outcomes. However, conventional physical measurements can be time-consuming, and machine learning methods that predict thickness of deposited films require extensive effort in model selection and hyperparameter optimization. Automatic machine learning, a new intelligent learning approach that uses Bayesian optimization to automate feature extraction from input data, model selection and hyperparameter optimization, has the potential to minimize the time and cost required for training and model selection. This paper presents a novel approach for predicting film thickness of IC deposited films based on automatic machine learning. The study analyzes three key techniques including meta-learning, Bayesian optimization and model integration, and establishes a prediction model using the Auto-Sklearn system. The model's prediction performance is compared with three common machine learning algorithms, namely random forests, support vector machines and multilayer perceptron, and the results show that the automatic machine learning prediction model outperforms these algorithms with a coefficient of determination of 0.759 on the test set and a mean relative error of 0.262% for the prediction results. This paper demonstrates the efficacy of automatic machine learning for predicting film thickness and provides a promising direction for future work in the semiconductor industry.