Detecting defects in semiconductor specimens using weak labeling

A method of classifying a pattern of interest (POI) on a semiconductor specimen from a high-resolution image of the pattern, the method comprising:

• receiving, by a processor, data informative of a high-resolution image of a POI on the specimen;
• and
• generating, by the processor, data usable for classifying the POI in accordance with a defectiveness-related classification,
• wherein the generating utilizes a machine learning model that has been trained in accordance with, at least a plurality of training samples, each training sample comprising:
• a high-resolution training image captured by scanning a respective training pattern on a specimen, the training pattern being similar to the POI,
• wherein the respective training pattern is associated with a label derivative of low-resolution inspection of the respective training pattern;
• wherein the machine learning model comprises a neural network, the neural network comprising:
• a first series of one or more neural network layers configured to, given input data informative of the high-resolution image of the POI, output a feature map, and
• a second series of neural network layers configured to, given an input of the feature map, generate data indicative of at least one per-pixel-block classification score,
• wherein each per-pixel-block classification score pertains to a region of the POI that is represented by pixels of a respective pixel-block of the high-resolution image, the per-pixel-block classification score being indicative of a likelihood of defectiveness of the respective region,
• thereby giving rise to data usable for classifying the POI in accordance with the defectiveness-related classification.

A computer program product stored on a non-transitory computer-readable medium containing program instructions, which program instructions when read by a processor, cause the processor to perform a method of training a machine learning model to, given input data informative of a high-resolution image of a pattern of interest (POI), generate data usable for classifying the POI in accordance with a defectiveness-related classification, the method comprising:

• a) obtaining a plurality of training samples, each training sample comprising:
• a high-resolution training image captured by scanning a respective training pattern on a specimen, the training pattern being similar to the POI,
• wherein the respective training pattern is associated with a label derivative of low-resolution inspection of the respective training pattern;
• and
• b) training the machine learning model in accordance with, at least, the plurality of training samples;
• wherein the machine learning model comprises a neural network, the neural network comprising:
• a first series of one or more neural network layers configured to, given input data informative of the high-resolution image of the POI, output a feature map, and
• a second series of neural network layers configured to, given an input of the feature map, generate data indicative of at least one per-pixel-block classification score,
• wherein each per-pixel-block classification score pertains to a region of the POI that is represented by pixels of a respective pixel-block of the high-resolution image, the per-pixel-block classification score being indicative of a likelihood of defectiveness of the respective region,
• thereby giving rise to data usable for classifying the POI in accordance with the defectiveness-related classification.