MODEL BASED APPROACHES FOR SELECTING RELIABLE UNDERFILL-FLUX COMBINATIONS FOR FLIP CHIP PACKAGES
Authors: Satyanarayan Iyer, Nagendra Nagarur and Purushotha Company: SMART Modular Technologies, Inc./State University Date Published: 4/1/2006
Abstract: Selecting an appropriate underfill and a compatible flux is crucial to establish a reliable flip chip assembly. Underfill, being an epoxy material, tends to absorb moisture. The moisture absorbed by underfill has a detrimental effect on reliability of flip chip packages. Researchers in the past have conducted experiments to qualify different underfill-flux combinations based on their moisture sensitivity. The limitations of these experiments are that each study was restricted to only one underfill-flux combination. The current research focuses on developing models that relate the assembly reliability to the properties of underfills and fluxes. The important properties of underfill materials and flux chemistries and their significance on the failures were studied using statistical techniques. The properties that were found to be significantly affecting the reliability of the assembly were used for subsequent modeling. These models were developed using experimental data from JEDEC Level 3 testing of flip chip packages that used different underfill-flux combinations. Results for 95 different underfill-flux combinations were available from various studies for modeling. All other parameters except for the underfill-flux combinations were the same across all the data. The objective of this paper was to propose mathematical models for estimating the defects for any given underfill-flux combination without actually conducting any experiments. In this study, we have proposed regression and neural network models. Both regression and neural network models performed well in explaining the reliability in terms of the selected important properties of the underfill and flux. Comparatively, neural network models performed better than the corresponding regression models in relating the failures with the selected properties of the underfill and flux. The neural network approach of modeling was recommended based on the results obtained as a part of this research effort and also due to the ability of neural networks to better capture non-linearaties. Keywords: Flip chip, underfill, flux, regression, neural networks.