Improvement of Multilayer Perceptron Model for Predicting Etch Rates in CF₄ Plasma With Optical Emission Spectrometer and VI Sensor Data

We studied a multilayer perceptron model to predict the etch rates of SiO2 and Si3N4 thin films in CF4 plasma using data obtained from a voltage–current (VI) sensor and an optical emission spectrometer (OES). The monitoring data from two devices and etch rate data were collected while varying process parameters such as source power, bias power, and pressure. The data gathered by separately etching a SiO2 coupon or a Si3N4 coupon were determined as training data, and the data gathered by etching both a SiO2 coupon and a Si3N4 coupon simultaneously were determined as test data. We utilized 27 process conditions to obtain training and test data and also employed data augmentation for the training data to improve the learning capability of the model. The multilayer perceptron model trained with data obtained from both the VI sensor and OES showed higher performance in predicting the etch rate of target material in multiple materials than other regression models trained with only data acquired from the OES. This improvement is attributed to the fact that data obtained from the VI sensor includes the information related to ion flux that is proportional to the etch rate and it is not easily captured by only the OES. These results indicate the potential of combining data obtained from the VI sensor and OES to predict the etch rate of target material in multiple materials.