A Classifier of AMOLED-TFT Defect Images Based on a DNN Ensemble Framework

With the advanced manufacturing technology of organic light-emitting diode (OLED) displays, yield loss in the manufacturing process became the most important index for display manufacturers. However, the reason that the most direct causes for the yield loss come from different defects on thin film. If the defect type can be successfully classified, it will be easier to find out the cause and effectively reduce the yield loss by adjusting the condition and parameters for the fabrication processes. In general, thin-film defects can be detected by automated optical inspection (AOI). When a thin-film defect is found, the reviewing module will relocate to the corresponding position and record the defect images. Then, the images will be further judged as normal or with specific defect types by experienced operators. Therefore, this research proposes a classifier for defect images based on a deep neural network (DNN) ensemble framework to reach a high accuracy of classification. It can help the display industry to save more cost and time in distinguishing the defect types precisely. Moreover, the proposed scheme is able to accelerate the inspection flow and improve the yield to make more profit.