Defect Inspection Of Flip Chip Based On Characteristic Features In Time And Frequency Domain

The flip chip package, with its small size, high I/O density, excellent electrical properties,and high reliability, has become the mainstream technology in microelectronics packagingindustry. However, internal defects of flip chip package happen easily because of highpacking density and thermal expansion coefficient mismatch of chip and substrate. Now theinspection of solder bumps has been an important process in the electronics manufacturingindustry to reduce cost and ensure product quality and reliability. The primary works of thedissertation include:Firstly, the vibration model of flip chip was deduced and the principle of defectsinspection base on characteristic features of vibration signals in time and frequency domainwas presented. An experimental system for solder bump inspection was constructed, in whichFC was excited with air-coupled ultrasound, and then the vibration signal was measured by alaser vibrometer.Secondly, an inspection method of flip chips with arrayed solder bumps was proposed,based on natural frequencies. The influence of defects on the natural frequencies wasdescribed. The natural frequencies of flip chips were calculated, simulated and measured inthe experiments. The results validates that the natural frequencies of defective chips weresmaller than the ones of non-defective chips. The sensitive degree of natural frequency amongdifferent modes was not equal. The mode shapes were obtained to disclose the differences offrequency shifts in different modes.At last, for flip chip with peripheral solder bumps, an inspection base on characteristicfeatures of vibration signals in time and frequency domain was presented. The coefficients intime domain and the energy features in frequency domain were extracted for fault diagnosis.Then the features were normalized and fed to the BP neural network for classification andrecognition. The results showed the relative importance of signal features in the inspection offlip chip. Experimental results revealed the effectiveness of the BP neural network indiagnosis of flip chip and number of missing solder bumps.