Modeling AOD-driven laser microvia drilling with machine learning approaches

Abstract

Acousto-optic deflectors (AODs) have been used to extend the bandwidth of laser beam scanning frequency in order to meet the growing demand on the quality and throughput of creating micro-size vertical interconnect access or microvia in electronic packaging materials. The implementation of AOD offers unprecedented spatiotemporal flexibilities for the laser drilling process control, meanwhile increases complexity for modeling the process and predicting the process result. Design of experiment and analysis of variance can reveal connections between process parameters and responses, but are incapable of predicting results with sufficient accuracy for some of the highly nonlinear processes. In this article, multiple machine learning techniques, including k-Nearest Neighbor, decision tree, support vector machine, support vector regression and artificial neural networks, are explored and compared in predicting the result of AOD-driven drilling processes. Among these techniques, artificial neural network shows advantage on accuracy. For instance, a mean squared error of as low as 1.21 μm2 is achieved in predicting the top diameter of microvias drilled by a 6-parameter process. Thus the artificial neural network is proved an ideal solution for modeling laser applications controlled by multiple process parameters.