基于DOE和BP神经网络对Al线键合工艺优化

Process Optimization for Aluminum Wire Bonding Based on DOE and BP Neural Network

Al丝超声引线键合工艺被广泛地应用在大功率器件封装中,以实现大功率芯片与引线框架之间的电互连。Al丝引线键合的质量严重影响功率器件的整体封装水平,对其工艺参数的优化具有重要工业应用意义。利用正交实验设计方法,对Al丝引线键合工艺中的三个最重要影响因数(超声功率P/DAC、键合时间t/ms、键合压力F/g)进行了正交实验设计,实验表明拉力优化后的工艺参数为:键合时间为40 ms,超声功率为25 DAC,键合压力为120 g;剪切推力优化的工艺参数为:键合时间为50 ms,超声功率为40 DAC,键合压力为120 g。基于BP神经网络系统,建立了铝丝超声引线键合工艺的预测模型,揭示了Al丝超声键合工艺参数与键合质量之间的内在联系。网络训练结果表明训练预测值与实验值之间符合很好,检验样本的结果也符合较好,其误差基本控制在10%以内。

Aluminum wire bonding process is widely employed in high power devices packaging to realize electrical connection between chip and lead frame. The quality of aluminum wire bonding seriously influences on the reliability of power device packaging and process parameter optimizations is quite significant to the industry manufacturing. Based on orthogonal experimental design method, the three parameters of ultrasonic power, bonding time and bonding pressure were used to construct the orthogonal experimental table, and the optimal parameters were obtained. Results show that the optimization process parameters for pull strength are time of 40 ms, power of 25 DAC, force of 120 g, respectively, while for shear strength are time of 50 ms, power of 40 DAC, force of 120 g, respectively. Finally, the prediction model for aluminum wire bonding based on back-propagation neural network （BPNN） was also constructed to show the inherent relationship between the process parameters and the bonding quality. Results reveal that the values of network training result are almost the same with values of the experiment, while the values of network examination have nealy 10% errors with values of experiment.