Research On Mechanism And Sparse Diagnosis Method Of High-frequency Ultrasonic Detection For Chip Micro-defects

As chips become smaller and highly integrated,the packaging speed and capacity of chips have been greatly improved.The demand for miniaturization of chip packaging has promoted the rapid development of packaging technology.However,as the size of chip becomes smaller,the manufacturing process of chip packaging becomes more and more challenging.The defects in the packaging process are also more likely to be generated and difficult to detect.Therefore,the reliable and stable detection of micro defects is very important.Acoustic micro imaging technology based on high-frequency ultrasound has been widely and effectively applied to the detection of micro defects in microelectronics packaging,but the noise and edge diffraction phenomenon occurs during high-frequency ultrasound testing,which seriously affects the accuracy of ultrasound testing and the signal-to-noise ratio.In this dissertation,based on the mechanism of ultrasonic propagation,the model of micro defect in ultrasonic testing chip is simulated by finite element method.The time domain echo signal and ultrasonic image of micro defect are further analyzed and processed based on sparse reconstruction algorithm.In addition,the intelligent classification algorithm of semi-supervised hierarchical extreme learning machine based on local linear embedding is used to recognize the defects of ball missing which are difficult to recognize in flip chip.The specific research contents are as follows,(1)Based on the theory of high-frequency ultrasonic propagation,the whole process transient finite element simulation of the ultrasonic testing model of flip chip which has been widely used is carried out.By extracting and analyzing the ultrasonic echo signal,the change of energy in the process of ultrasonic propagation is clarified,and three typical defects(ball missing,crack,voids)in the flip chip solder ball are analyzed.The influence of different defects on the energy transfer in the process of ultrasonic wave propagation and the phenomenon of edge diffraction in high frequency ultrasonic testing technology are analyzed.(2)The sparse reconstruction model of one-dimensional ultrasonic testing signal is established.The samples with micro defects are prepared by inductively coupled plasma etching process.The high-frequency ultrasound is used to detect the experimental samples,and the detected one-dimensional ultrasonic echo signal is analyzed to further verify the edge diffraction phenomenon of high-frequency ultrasonic testing.A sparse reconstruction algorithm is proposed for noise effect.Firstly,the over complete dictionary is constructed based on Gabor function,and then the one-dimensional sparse reconstruction of ultrasonic signal is carried out by using the over complete dictionary and the least square orthogonal matching pursuit algorithm.The reconstruction results indicate that this method can alleviate the effect of noise in the ultrasonic testing process,so as to more accurately identify and evaluate the micro defects in the chip.(3)The sparse reconstruction model of two-dimensional ultrasonic testing image is established,and a sparse reconstruction method of ultrasonic micro defect image based on blind estimation is proposed.Firstly,BM3 D denoising method is used to enhance and denoise the original ultrasonic image of micro defect.Then,the blur kernel(point spread function)of ultrasonic image is estimated based on the maximum a posteriori algorithm.Finally,an iterative threshold shrinkage algorithm is used to recover the image.The results show that the method can improve the resolution of high frequency ultrasonic detection,and expand the applicability and practicability of sparse reconstruction method.(4)Aiming at the typical defect of ball missing in flip chip,the ultrasonic image of flip chip is obtained by using acoustic micro imaging technology.Firstly,the ultrasonic image is segmented into sub images with only one solder ball based on correlation coefficient method.A model of semi-supervised hierarchical extreme learning machine based on local linear embedding is established to extract features and classify features of the ultrasonic image.The classification results show that the intelligent classification algorithm can quickly and accurately identify the defects of missing ball in flip chip.In this dissertation,the acoustic micro imaging technology is used to realize the effective detection of defects in chip,and the effective methods from the evaluation of defects in chip packaging to the intelligent identification of defects are provided through simulation and experimental methods,which provides theoretical basis and technical methods for the rapid development of chip packaging industry in China.