Thermal Performance Prediction of QFN Packages using Artificial Neural Network (ANN)

The thermal performance of QFN with body sizes ranging from 3 mm times 3 mm to 9 mm times 9 mm with various lead counts were modeled on FR4 printed circuit board using a solid model finite element simulation tools (ANSYS). The thermal performance obtained using FEA agrees well with the experiment data (within 12%). A series of data with different package design parameters, PCB thermal balls, powers and ambient temperatures were obtained to train the artificial neural network (ANN). The trained ANN was then used to predict a set of data to be compared with experiment data. The results of ANN's prediction have good agreement with both experiment results (13%) Hence, ANN can become an alternative tool for package level thermal analysis.