Optimal design and fatigue life prediction for QFN solder joints by BP Artificial Neural Networks and Genetic Algorithm

Based on the theories of Back-Propagation (BP) Artificial Neural Networks (ANN) and Genetic Algorithm (GA), combining with multiple statistical analysis method, fatigue life prediction and technological parameter optimization of QFN solder joints were studied in this paper. Firstly, correlation coefficient matrix of the swatch was gained by factor analysis; taking length and width of the pad, stand-off and the solder volume as input parameters, fatigue life as output parameter, Levenberg-Marquardt (LM) algorithm was used to train the BP ANN and network topology was determined through the experiment, then nonlinear relationship between the input and output parameters was established. Network performance was assessed by linear regression method. Finally, the trained ANN was selected as the objective function solver; GA is used to optimize the QFN solder joints in the software of Matlab. The result shows that, the network obtains precision forecast capability when network topology is in the state of 4-6-1, while the experimental error is within 5%, SSE is 0.0054 and MSE is 0.0011.The optimal combination parameters were gained the pad length of 0.8 m, the pad width of 0.3283 m, the stand-off of 0.1022 m, the volume of 0.014 m³, the fatigue life of the QFN solder joints increase of 20% while the error is 2.8143%.