IC Package Warpage Reduction Based on Fuzzy Adaptive Particle Swarm Optimization Algorithm and Neural Network

The warpage of ICs in IC packaging manufacturing causes the production of defective ICs that can short-circuit or malfunction, including those in sensor devices. Applicable research results that predict IC warpage using a neural network have not been many, although many technologies have been proposed to prevent the warpage. It is necessary to understand the properties of IC materials as each material has a different coefficient of thermal expansion (CTE) for predicting the occurrence of the warpage. To provide a means to predict the warpage, a neural network with fuzzy adaptive particle swarm optimization (FAPSO) is proposed in this study based on the proposed architecture of the neural network and the defined weights of each layer in the IC. As the three layers of epoxy molding compound (EMC), die, and substrate (SBT) in IC packaging have different CTEs, nine conditional variables, namely, die thickness, glass transition temperature (Tg), CTEs (α1, α2), filler size, filler content, total height, post mold cure (PMC) temperature, and PMC time, are defined for predicting the warpage, and their parameters are found for training the neural network. In the comparison of the actual and predicted data of the neural network with FAPSO, the correlation coefficient is 0.9878, and the similarity between the two data sets is 99.7% in training. After the training, the validation is carried out for six data sets, the result of which shows that the correlation coefficient (R2) is 0.8658 and the mean absolute percentage error (MAPE) is 29.74%, which is acceptable for applying the proposed neural network. The result of this study helps to improve the IC packaging process by preventing the warpage.