Optimization design for packaging device QFN using a prediction model of the neural-genetic algorithm

This study presents an optimal method to select the material and dimension parameters for designing microelectronics packaging devices loading hygro-thermal and vapor pressure. The failure mechanism for delamination of actual packaging devices is often a complex nonlinear function, which is a shortcoming of traditional methods. The proposed approach is a combination of Error back-propagation neural network (BPNN), principal component analysis (PCA) and genetic algorithms (GAs). First of all, PCA is employed to reduce the dimension and de-noise for the learning matrix of BPNN model. And then GAs is combined with the BPNN model to find the most appropriate linking weight with its global search feature. Secondly, the well-trained network model, which included a nonlinear function between the input parameters and corresponding outputs, is seen as a prediction tool to select optimal parameter size in order to reduce the J-integral value of interface cracking in the packaging device. Finally, optimal parameter groups can be achieved for the device after verification. The optimization results show the well-trained PCA-GA-BPNN model used the proposed approach, can be used well in the optimizing design of the microelectronics packaging device loading hygro-thermal and vapor pressure. Meanwhile, the model is available to reduce the fracture reliability problems, and is of much practical value.