Auto defect screening using adaptive machine learning in semiconductor device manufacturing flow

A method for auto defect screening in manufacturing a semiconductor device using a system having one or more computing processors and one or more memory devices configured and programmed to function as an adaptive model controller, a data modeling analytics executor and an auto defect screener, said method comprising the steps of:

• preparing a feed-forward input for said adaptive model controller to receive a plurality of defect candidates acquired from inspecting one or more wafers of said semiconductor device and a feedback input for said adaptive model controller to receive defects of interest already screened by using one or more existing defect screening models;
• selecting data samples from said plurality of defect candidates and said defects of interest in said adaptive model controller and interfacing with a scanning electron microscope (SEM) review/inspection tool to acquire corresponding SEM results of said data samples, each of said data samples being validated and labelled as being a real defect or nuisance in said corresponding SEM results;
• compiling model training data and model validation data from accumulated data samples in said adaptive model controller according to the corresponding SEM results of the accumulated data samples;
• using said data modeling analytics executor to perform data modeling analytics to generate one or more updated defect screening models from said model training data to satisfy a target specification validated with said model validation data;
• and
• using said one or more updated defect screening models in said auto defect screener to predict if a defect candidate is a real defect or nuisance and filter out the predicted nuisance;
• wherein said method further uses said adaptive model controller to control when to generate said one or more updated defect screening models according to a pre-set criteria.