The Design And Implementation Of QFN Inspection Software Base On Machine Vision

The inspection of integrated circuit products is one of the most critical process in the back end of semiconductor assembly. As the final process of assembly department, it acts as a quality control gate. At the same time, it gives timely quality feedbacks to all process. With the feedbacks, production issues could be found in the first time and corresponding corrective action would be token immediately.As one packaging type of the semiconductor assembly, QFN (Quad Flat Non-Lead) is young. Currently, few commercial inspection systems are capable to inspect this new package type. The major study of this paper is focused on the design and implementation of QFN inspection software. The paper analyzed both happened and potential critical defects. The defects were categorized into some subclasses, such as body size defect, leads/pads pattern center offset, foreign matter, scratch, mold flash, void and so on. Aimming at the defects which are going to be inspected, the algorithm of teaching before inspecting was created. That means, first of all, teaching the system with the information of package outline, package center and so on. The model image of the type of the package which is going to be inspected was also calculated and saved into computer. During inspection phase, the information was got previously will be compared with each package image which is inspecting. A judgment will be came out accordingly. The system was design in modularized thinking, which make source code simplified and more effective.This QFN inspection system was evaluated by strict procedure. It fulfills the design expectation and the requirement of production line in both dimension measurement and surface detection aspects. The speed of 289ms/unit exceeds the average level of same products of industry. In the condition of no under-kill, the system achieved the overkill rate of 1.4% which is obviously lower than the industrial average rate of 3%.