Subpixel Chip Alignment with Blurred Edge Model Using the Revised Newton Method

• Main Authors: Wei- Chun Hung, 洪暐鈞
• Other Authors: Chin-Hsing Chen
• Format: Others
• Language: en_US
• Published: 2007
• Online Access: http://ndltd.ncl.edu.tw/handle/65583791209033001872

碩士 === 國立成功大學 === 電腦與通信工程研究所 === 95 === The thesis studies the subpixel alignment of BGA and the IR cut filter of the CMOS sensor. BGA is a mature technique of packaging and widely applied in electronic manufactures because of its better electric and heat dissipation property. Under more accurate alignment, we can mark the package accurately. The IR cut filter is used to cut off infrared rays and avoid its disturbance to the CMOS sensor. The CMOS sensor of digital camera responds to the range of optical spectrum from 350nm to 1200nm, it can detect infrared ray but infrared ray will make image to be distorted so digital camera must install the filter in order to reduce the interference of the infrared ray. We can accelerate and stabilize the procedure of packaging if we have accurate alignment. The algorithm employed by the thesis consists of the following steps: project the package on x and y axis to delimit the ROI (region of interest) and find the edge elements by using DOB (difference of boxes) operator, then apply edge following to eliminate defected edge points and improve the edge points to subpixel accuracy using the blurred edge model modified by the revised Newton method, finally, find the center position and the sloped angle of the package by linefitting and geometric calculation. For synthetic images without noise, the detected errors of displacement in horizontal are smaller than 0.1 pixel and the errors of sloped angle are smaller than 0.01 degree. For synthetic images corrupted by Gaussian noise, the standard deviation of the sloped angle error is smaller than 0.01 degree. Finally, we evaluate two types of practical packages. Under the conditions of slight noise, the standard deviation of the center position and the sloped angle are less than 0.07 pixel and 0.02 degree respectively.