Data Science for Delamination Diagnosis in the Semiconductor Assembly Process

Abstract

In the semiconductor assembly process delamination in die-attach layers is a leading cause of defective products Delamination may exist between die and epoxy molding compound (EMC) epoxy and substrate lead frame and EMC etc The troubleshooting is case-by-case and time-consuming without a systematic diagnosis approach In this thesis we propose a data science framework using least absolute shrinkage and selection operator (LASSO) regression and stepwise regression to extract the key variables affecting delamination Then we construct backpropagation neural network (BPN) support vector regression (SVR) partial least squares (PLS) and gradient boosting machine (GBM) to predict the ratio of the delamination area in a die Besides we investigate the imbalance between a false positive rate and false negative rate after the quality classification with BPN and GBM models to improve the tradeoff between the two types of risks We validate the proposed framework with an empirical study of a semiconductor assembly company The results show that the proposed framework provides delamination prediction with high accuracy and gains managerial insights for supporting the practical troubleshooting Furthermore since that the batch size determination of the dataset significantly affects the performance of the inline model retraining process we suggest the cost-oriented method to address the issue

Date of Award	2018 Feb 5
Original language	English
Supervisor	Chia-Yen Lee (Supervisor)