粗铝丝超声引线键合强度在线监测方法研究

【作者】 刘少华；

【导师】 王福亮；

【作者基本信息】 中南大学 ， 机械电子工程， 2010， 硕士

【摘要】 粗铝丝超声引线键合强度对超声功率、键合力、键合时间等因素敏感,实现键合强度的在线监测对于提高键合可靠性具有重要的工程价值。针对当前缺乏有效的在线键合强度预测方法的现状,本文提出了通过实时在线分析换能器驱动电流、建立电流信号与强度之间的关联规律,来监测并估计键合强度的方法。主要工作包括：1.基于U3000型粗铝丝引线键合机,建立了粗铝丝引线键合强度在线监测实验平台,搭建了换能器驱动电流信号传感电路,开发了基于NI数据采集卡PCI6110及LabView的超声引线键合过程信号监测系统；比较分析了反映键合过程的四种信号,确定并选择换能器驱动电流信号为超声引线键合过程监测信号；通过键合点的抗剪切试验,获得了反映键合点强度的抗剪切力数据。2.分析比较了主要时频分析方法,根据换能器驱动电流信号变化快的特点,选择小波包变换和Wigner-Ville变换作为主要的信号分析方法。获得了驱动电流信号反映的键合过程特征,包括键合过程系统消耗的平均能量、键合过程系统消耗能量的变化程度、键合中后期的系统消耗能量的变化等12个特征,并初步分析了上述特征与强度间的关联关系。3.建立了基于人工神经网络的粗铝丝超声引线键合强度在线监测系统,针对典型的键合条件,选择了键合过程系统消耗的平均能量、键合过程系统消耗能量的变化程度、键合中后期的系统消耗能量的变化等8种特征作为输入,键合强度作为输出,采集了120组实际键合过程数据,对人工神经网络进行训练,获得了信号特征与键合强度之间的关联关系。并将上述系统应用于U3000粗铝丝超声引线键合机,实现了键合失败的实时在线识别和键合强度的实时在线估测。更多还原

【Abstract】 The bond quality of the aluminum wire bond is sensitive to ultrasonic power, bond force, bond time etc. Realizing the bond quality on-line monitoring is very valuable to improve the bond reliability. Contraposing the absence of the methods of effective on-line bond quality forecast, a method was proposed that is obtaining the relations between the current signals and bond quality through the analysis of the transducer drive current.1. An on-line quality monitoring system of the aluminum wire bond was constructed based on the aluminum wire bonder U3000.The sensing circuit was constructed.The signals monitoring system of ultrasonic wire bond was developed based on PCI6110 and LabView. The bonding quality data were gathered by the offline destructive tests.2. The main time-frequency analysis methods were contrasted. According to the characteristics of levity of transducer drive current, the analysis method was found based on wavelet and wavelet package,12 features were selected, such as consumed average energy, changing extent of consumed average energy of bond process, change of consumed energy of upper bond process etc. The relations between the current signals and bond quality were analyzed briefly.3. Contraposing the typical bond conditions,8 features were selected, such as consumed average energy, changing extent of consumed average energy of bond process, change of consumed energy of upper bond process etc. An artificial neural network was set up using these features qua input, bond quality qua output. The relations between the current signals and bond quality were acquired by neural network training. The above methods were applied in the aluminum wire bonder U3000. The recognition of bond failure and estimate of bond quality were realized.更多还原