| Summary: | 博士 === 國立臺灣海洋大學 === 電機工程學系 === 99 === Light-emitting diodes (LEDs) have become increasingly used in the electronics industry. In addition to increasing wafer dimensions, the yield ratio can be improved to increase revenue. To increase yield, defect inspection before die encapsulation is commonly used. Conventional inspection approaches require dozens of operators or engineers visually checking the dies and marking defective ones with the help of an electron microscope. Manual visual inspection, however, has considerable labor costs and suffers from misjudgments due to human fatigue. Additionally, variability between observers (inter-observer variability) or that between two observations made by an observer (intra-observer variability) also affects the inspection results. In contrast, an automatic inspection system provides consistent inspection results, reducing the variance caused by human error.
In this dissertation, computational intelligence is applied to automatic defect inspection. Applying computational intelligence to automatic defect inspection can effectively prevent potential misjudgments that are caused by inexperience or human error. Automatic inspection is cost efficient. It provides high inspection throughput. Since manual intervention is reduced, errors due to human fatigue and potential manual misjudgments are prevented. For an LED wafer that contains more than eighty thousand dies, automatic defect inspection provides stable inspection accuracy for yield estimation, and reliable inspection results to improve subsequent encapsulation.
|
|---|
