| China is the largest semiconductor consumer market in the world and has the largest packaging and testing industry in the world.In the chip packaging process,the wafer dicing process is a key step to implement the transformation from overall wafer to individual chip.However,various defects are inevitably generated on the wafer surface during wafer sawing,which directly affects the integrity and reliability of the chip die circuit function.Automatic optical inspection has become a mainstream solution in the field of wafer surface quality control due to its advantages of non-contact,non-destructive,high-speed,and high-precision automatic inspection.In response to the existing research difficulties in the complete process of wafer automatic optical inspection system from positioning,acquisition,to detection,on the basis of in-depth analysis of existing methods,the key technologies such as large-scale wafer circle center location,wafer automatic optical inspection system error calibration,and wafer complex background surface defect detection are emphasized and studied.Also,the wafer automatic optical inspection equipment for dicing process is developed(The equipment has been tested and used in the actual production of Jiangsu Changdian Technology).The main research contents are as follows:(1)The wafer automatic optical inspection system is designed and implemented.First,the overall scheme of wafer automatic optical inspection system is designed,and the system module composition is described.Second,according to the requirements of high-quality,high-speed image acquisition and different-scale precision detection of the visual system module,the design and selection of the wafer micro auto-focus visual system are realized.Then,according to the technical requirements of the motion control module,the motion control system scheme is proposed and the selection of key components is completed.Fourth,according to the functional requirements of the software algorithm module,the modular design idea is adopted to develop the system software and the visual detection algorithm flow is established.Finally,according to the design and development of each module,the wafer automatic optical inspection system is fully implemented.(2)A method of circle center location for a large-scale wafer under the small field of view camera is proposed.First,the calculation principle of the system angle in the horizontal plane is analyzed,and a method for measuring the system angle in the horizontal plane is proposed.Then,the circular arc edge extraction algorithm with distance constraint is proposed,and the edge point sets of multi-segment circular arc images are obtained.The combination of multisegment circular arcs and the compensation of the system angle in the horizontal plane are considered,and the coordinate transformation in the reference coordinate system is realized.An improved circle parameter calculation method is proposed,and the initial circle parameters of the wafer are obtained.Taking the initial circle parameters as a priori constraint,a precise estimation method of circle parameters based on hierarchical Bayesian model is proposed,and the accurate location of the wafer circle center is implemented.Finally,a series of experiments are carried out on the developed wafer automatic optical inspection system,which verifies the effectiveness of the proposed method.(3)An error calibration method based on perspective mapping for the wafer automatic optical inspection system is proposed.First,the error cause of the wafer automatic optical inspection system is analyzed,and the high-precision glass checkerboard calibration board is designed.The total error of the wafer automatic optical inspection system is accurately measured by the calibration experiment.Then,the mapping model between the motor coordinate and the checkerboard coordinate in the wafer automatic optical inspection system is established,and the mutual mapping between the arbitrary motor coordinate and the checkerboard coordinate is implemented.The mapping motor coordinates within the system stroke range are obtained by using the glass checkerboard and the mapping model,and the total error calibration of the wafer automatic optical inspection system is implemented based on the perspective mapping principle.Finally,a series of experimental results show that the total error of the system can be controlled within 10 8)after error calibration,and the proposed method can meet the positioning accuracy requirement of the wafer automatic optical inspection system.(4)A three-stage deep learning method for wafer surface defect detection with complex background is proposed.First,the incomplete and small die images are quickly stitched based on the integral image,and then the complete die image is finely segmented based on the edge information,so that an automatic die extraction method is proposed.Then,the image texture feature,brightness feature and template matching problem of the wafer surface with complex background are analyzed.Aiming at the problem of texture layer with different features on the surface of the die image,a die image texture layer processing method based on U-Net model is proposed,and the uniform texture layer and random strong texture layer of the die image are obtained.To solve the problem of random strong texture on the die surface with complex background,a random strong texture removal method based on JESS-Net model is proposed,and the die image with random strong texture removal and defect structure retention is obtained.For the different brightness differences between different dies and the difficulty in accurate alignment of template matching,a defect-free dynamic template image generation method based on CAE model is proposed,and the defect-free dynamic template image is obtained.Finally,compared with the existing methods,the results show that the proposed method can achieve high detection accuracy and efficiency,and can meet the requirement of wafer surface defect detection with complex background.The above key technologies are not only applied to automatic optical inspection system for wafer dicing,but also can be extended to high-speed automatic optical inspection in other fields,so they have great theoretical research significance and engineering application value. |
PDF Full Text Request