Automatic optimization of an examination recipe

A computerized method of automatic optimization of an examination recipe usable for examining a sequence of semiconductor specimens in runtime, the method performed by a processing and memory circuitry (PMC), the method comprising:

• obtaining one or more inspection images acquired by an inspection tool during runtime examination of a first semiconductor specimen in the sequence, each inspection image being representative of a portion of the first semiconductor specimen corresponding to a respective defect candidate of a list of defect candidates, the list of defect candidates being selected from a defect map of the first semiconductor specimen using a first classifier included in the examination recipe;
• obtaining label data informative of types of respective defect candidates in the list of defect candidates, the label data being obtained from a review tool and being respectively associated with the one or more inspection images corresponding to the respective defect candidates;
• and
• in parallel with the runtime examination:
• extracting, using at least an unsupervised model previously trained using a set of training inspection images, inspection features characterizing the one or more inspection images;
• associating the inspection features with the label data of the one or more inspection images to obtain a training set;
• retraining the first classifier using the training set to obtain a second classifier;
• and
• in response to determining to optimize the examination recipe, replacing the first classifier with the second classifier to obtain an optimized examination recipe;
• wherein the optimized examination recipe is usable for examining a subsequent semiconductor specimen in the sequence.

A non-transitory computer readable storage medium tangibly embodying a program of instructions that, when executed by a computer, cause the computer to perform a method of automatic optimization of an examination recipe usable for examining a sequence of semiconductor specimens in runtime, the method comprising:

• obtaining one or more inspection images acquired by an inspection tool during runtime examination of a first semiconductor specimen in the sequence, each inspection image being representative of a portion of the first semiconductor specimen corresponding to a respective defect candidate of a list of defect candidates, the list of defect candidates being selected from a defect map of the first semiconductor specimen using a first classifier included in the examination recipe;
• obtaining label data informative of types of respective defect candidates in the list of defect candidates, the label data being obtained from a review tool and being respectively associated with the one or more inspection images corresponding to the respective defect candidates;
• and
• in parallel with the runtime examination:
• extracting, using at least an unsupervised model previously trained using a set of training inspection images, inspection features characterizing the one or more inspection images;
• associating the inspection features with the label data of the one or more inspection images to obtain a training set;
• retraining the first classifier using the training set to obtain a second classifier;
• and
• in response to determining to optimize the examination recipe, replacing the first classifier with the second classifier to obtain an optimized examination recipe;
• wherein the optimized examination recipe is usable for examining a subsequent semiconductor specimen in the sequence.