Comparison and Identification of Optimal Machine Learning Model for Rapid Optimization of Printed Line Characteristics of Aerosol Jet Printing Technology

다국어 초록 (Multilingual Abstract)

Among the various direct-write (DW) techniques, aerosol jet printing (AJP) has the advantages of high resolution (~ 10 μm) and flexible working distance (2-5 mm). On this basis, it has emerged as a promising DW technology to precisely customize complex electrical functional devices. However, the micro-electronic devices fabricated using AJP suffer from low electrical performance because of inferior printed line geometric characteristics. Specifically, high edge roughness lines are detrimental to the uniformity of the formed electrical functional devices. In addition, the low controllability of the printed line width may induce overlap of narrowly spaced circuits or unnecessary intertrack voids, which will hinder the wide application of AJP technology in advanced electronic manufacturing industry. Therefore, ensuring high precision of the line width and low edge roughness is of primary importance for AJP technology. In this research, a machine learning framework is proposed for rapid optimization of printed line characteristics. In the proposed framework, SHGFR and CGFR were considered as input variables, and line width and line roughness were taken as the target responses. Three representative machine learning algorithms, tree-based random forest regression, kernel-based support vector machine, and Bayesian-based Gaussian process regression, were then adopted for model development. Subsequently, the identified optimal machine learning model was integrated with a NSGA-III for rapid optimization of printed line characteristics, and experiments validated the effectiveness of the adopted approach.