Knowledge augmented broad learning system for computer vision based mixed-type defect detection in semiconductor manufacturing

•A knowledge augmented broad learning system featured by knowledge module and broad selective sampling module is proposed for mixed-type defects on wafer map.•Broad selective sampling module accomplishes multichannel selective sampling of basic wafer map defects.•Decision function for activation of sampling channel helps network decouple the mixed-type defects.•Sampling area visualization and defective points sampling ratio reveal the effectiveness of prior knowledge.