Machine learning-based off-line electrical characteristic prediction through in-line pattern integrity inspection

Abstract

Abstract In this study, an image inspection method was introduced to two-arm Archimedean spiral antenna patterns to quantify and qualify their in-line integrity, which was linked to their off-line electrical characteristics in terms of the capacitance values through machine learning. The pattern was intentionally deteriorated in shape to imitate potential fabrication variations existing in the microelectronic production line, and six physical features including the inner line edge roughness (LER), outer LER, integrated LER, inner arm length, outer arm length, and arm area were collected. Two groups of training and testing samples were simulated and fabricated. Based on Gaussian process regression with the covariance function in the form of a squared exponential, a model was developed to predict the capacitance values from the performances of the six features. The accuracy of the developed model was evaluated using the coefficient of determination and root-mean-square error. The results indicate that the developed model is capable of predicting the off-line electrical characteristics of microelectronic components based on their in-line pattern integrities. Advanced studies also reveal that although all LER values and arm lengths contribute to the electrical characteristics, the arm area is decisive.