Effect of Dispensing Type on Void Formation Using Convolutional Neural Network

Abstract

Underfilling in flip chip packages is a critical component of reliability. This study utilized I-type, L-type, and U-type dispensing methods to address the issue, namely, voiding that creates empty spaces, which compromises reliability. An automated solution using convolutional neural network (CNN) is proposed for void detection in chip images to replace the conventional manual inspection approach. The CNN model built on MobileNetV2 attains a mean average precision of 0.533. This method calculates void percentage, adhering to Institute for Interconnecting and Packaging Electronic Circuits (IPC) standards, to determine product acceptance or rejection, offering an efficient solution for quality control in flip-chip package manufacturing.