Methodologies For Thermal Analysis In Single Die And Stacked Dies Electronic Packaging
Loading...
Date
2012-03
Authors
Law, Ruen Ching
Journal Title
Journal ISSN
Volume Title
Publisher
Universiti Sains Malaysia
Abstract
Thermal analysis in single die and stacked dies electronic packaging for portable communication devices is very important due to lack of real estate for active cooling. Recent research had focused on active cooling and neglected the low cost passive cooling by optimizing the architecture of package structure and material selection. Stacked dies electronic package is an economical and good electrical performance innovation but inherent thermal problems which caused by thermal crosstalk. Recent methodology for numerical method and measurement method for thermal analysis in QFN and stacked dies LBGA is labor intensive, needs huge amount of investment and requires expert’s knowledge.
To fill the gap above issue, 3D Finite Element method (FEM) using ANSYS ® with simplified and automated method by APDL language available in ANSYS ® had been established. This method had reduced 3D model building, meshing and computation time tremendously especially for stacked die electronic packaging. Two types of popular electronic packaging namely Quad Flat No-lead (QFN) and stacked dies Low-profile Ball Grid Array (LBGA) had been chosen to demonstrate the capability of the new methodology. The results from the FEM were compared with experimental result from open literature. The thermal performance and temperature predicted by this simplified pre-processing method in FEM have highest relative error of 8% for QFN compared to experiment result carried out. The highest relative error of simplified pre-processing method in FEM for stacked dies LBGA is 12.4%. After the result been verified, the simplified pre-processing method in FEM was used to generate huge amount of data to train the back propagation feed forward artificial neural networks (ANN).
Artificial Neural Networks (ANN) had been introduced in this study. The ANN was built using Neural Network Toolbox in Matlab 6.5. The simplified pre-processing method in FEM was used to generate data to train the ANN. The highest relative error of prediction of thermal performance and temperature by ANN was less than 5.9% and 9.25% for QFN and stacked dies LBGA respectively. ANN can simplify the way package designers to evaluate their design’s thermal performance with very minimum computational power, cost and time by eliminating the error prone model building, complexity of meshing and boundary condition settings.
Description
Keywords
Thermal analysis in single die , stacked dies electronic packaging