An Artificial-Intelligence-Driven Predictive Model for Surface Defect Detections in Medical MEMS

With the advancement of miniaturization in electronics and the ubiquity of micro-electro-mechanical systems (MEMS) in different applications including computing, sensing and medical apparatus, the importance of increasing production yields and ensuring the quality standard of products has become an important focus in manufacturing. Hence, the need for high-accuracy and automatic defect detection in the early phases of MEMS production has been recognized. This not only eliminates human interaction in the defect detection process, but also saves raw material and labor required. This research developed an automated defects recognition (ADR) system using a unique plenoptic camera capable of detecting surface defects of MEMS wafers using a machine-learning approach. The developed algorithm could be applied at any stage of the production process detecting defects at both entire MEMS wafer and single component scale. The developed system showed an F1 score of 0.81 U on average for true positive defect detection, with a processing time of 18 s for each image based on 6 validation sample images including 371 labels.