Method of classifying defects in a semiconductor specimen and system thereof

A method of automated classifying defects in a specimen into a plurality of classes, the method comprising:

• obtaining by a processing and memory circuitry (PMC) data indicative of quality requirements separately defined for each class of the plurality of classes;
• obtaining by the PMC data indicative of assignment of each class of the plurality of classes to a classification group among three or more classification groups with different priorities;
• during training:
• obtaining by the PMC training data comprising a plurality of pre-classified training defects and attribute values thereof;
• providing by the PMC an initial automated classification of the plurality of training defects into the classes of the plurality of classes to yield initially classified training defects;
• processing by the PMC the initially classified training defects to obtain an optimal working point corresponding to the highest possible contribution of automated classification while meeting per-class quality requirements defined for each class of the plurality of classes;
• generating by the PMC classification rule with confidence thresholds corresponding to the optimal working point, wherein the confidence thresholds correspond to prioritized rejection bins, and wherein at least “cannot decide (CND)” rejection bins correspond to the three or more classification group;
• and
• applying by the PMC the generated classification rule when classifying defects in the specimen.

A non-transitory computer readable medium comprising instructions that, when executed by a computer, cause the computer to perform a method of automated classifying defects in a specimen into a plurality of classes, the method comprising:

• obtaining data indicative of quality requirements separately defined for each class of the plurality of classes;
• obtaining data indicative of assignment of each class of the plurality of classes to a classification group among three or more classification groups with different priorities;
• during training:
• obtaining training data comprising a plurality of pre-classified training defects and attribute values thereof,
• providing an initial automated classification of the plurality of training defects into the classes of the plurality of classes to yield initially classified training defects;
• processing the initially classified training defects to obtain an optimal working point corresponding to the highest possible contribution of automated classification while meeting per-class quality requirements defined for each class of the plurality of classes;
• generating classification rule with confidence thresholds corresponding to the optimal working point, wherein the confidence thresholds correspond to prioritized rejection bins, and wherein at least “cannot decide (CND)” rejection bins correspond to the three or more classification group;
• and
• applying the generated classification rule when classifying defects in the specimen.