基于BP神经网络的模塑封电子器件优化设计

Optimal design of plastic electronic packaging component based on BP neural network

针对塑封SOT（小外形晶体管）器件的使用失效案例，从芯片设计角度出发，提出一种优化设计方法，该方法利用误差反向传播神经网络（BPNN），结合主成分分析（PCA）、遗传算法（GAs）及均匀设计的针对非线性系统的优化设计，设计了该塑封SOT器件的尺寸参数。结合实验和有限元模拟分析，验证该优化设计结果的有效性。结果表明，优化设计的器件各关键界面点的最大应力约减少了70～180MPa，器件的界面层裂现象得到消除，提高了器件的可靠性。

According to failure cases of plastic electronic packaging component SOT in the use, dimension parameters of plastic electronic packaging component SOT were designed with a proposed optimal design method, which combined error back-propagation neural network （BPNN）, principal component analysis （PCA）, genetic algorithms （GAs） and uniform design, and an optimal design for nonlinear system. The validity of optimal design result was verified by combining with experiments and FEA. The results show that the max stresses of each key interface points are is decreased about 70-180 MPa, which greatly improves the reliability of the component. Meanwhile, after improving by the optimal design, the delamination phenomenon is avoided for the device. The feasibility of proposed optimal design to plastic electronic packaging component has been verified by above.