System, method for training and applying defect classifiers in wafers having deeply stacked layers

1. A non-transitory computer readable medium storing a computer program product having code executable by a processor to perform a method comprising:

acquiring a plurality of images generated by an inspection system for a location of a defect detected on a wafer by the inspection system, wherein the location on the wafer is comprised of a plurality of stacked layers, and wherein each image of the plurality of images is generated by the inspection system at the location using a different focus setting;

adding the defect to a training set;

determining a classification of the defect, utilizing the plurality of images,

wherein the classification of the defect is determined using a through-focus signal profile generated for the defect utilizing the plurality of images, the through-focus signal profile being a 1-dimensional (1D) histogram having the focus setting on a first axis and attributes of a corresponding image on a second axis;

labeling the defect in the training set based on the classification determined for the defect; and

training a machine learning classifier of the inspection system, using the training set, to tune performance of the machine learning classifier during classification of defects for subsequent inspections of wafers having stacked layers.

2. The non-transitory computer readable medium of claim 1 , wherein the wafer is a three-dimensional (3D) NAND wafer.

3. The non-transitory computer readable medium of claim 1 , wherein the defect is located in an underlying layer of the plurality of stacked layers.

4. The non-transitory computer readable medium of claim 1 , wherein each of the different focus settings focuses light, at a different depth, into the wafer at the location.

5. The non-transitory computer readable medium of claim 1 , wherein the classification of the defect is determined directly from the plurality of images.

6. The non-transitory computer readable medium of claim 1 , wherein the attributes include brightness.

7. The non-transitory computer readable medium of claim 1 , wherein the attributes include magnitude.

8. The non-transitory computer readable medium of claim 1 , wherein the classification of the defect indicates a layer of the plurality of stacked layers at which the defect exists.

9. The non-transitory computer readable medium of claim 1 , further comprising binning the defects using the trained classifier.

10. The non-transitory computer readable medium of claim 9 , further comprising determining a portion of the defects that are included in a same bin and plotting the portion of the defects to generate a spatial signature for the bin.

11. The non-transitory computer readable medium of claim 10 , further comprising tuning the inspection system using the spatial signature generated for the bin.

12. A method, comprising:

acquiring, by a computer processor, a plurality of images generated by an inspection system for a location of a defect detected on a wafer by the inspection system, wherein the location on the wafer is comprised of a plurality of stacked layers, and wherein each image of the plurality of images is generated by the inspection system at the location using a different focus setting;

adding, by the computer processor, the defect to a training set;

determining, by the computer processor, a classification of the defect, utilizing the plurality of images, wherein the classification of the defect is determined using a through-focus signal profile generated for the defect utilizing the plurality of images, the through-focus signal profile being a 1-dimensional (1D) histogram having the focus setting on a first axis and attributes of a corresponding image on a second axis;

labeling, by the computer processor, the defect in the training set based on the classification determined for the defect; and

training, by the computer processor, a machine learning classifier of the inspection system, using the training set, to tune performance of the machine learning classifier during classification of defects for subsequent inspections of wafers having stacked layers.

13. A system, comprising:

memory storing computer code; and

a processor coupled to the memory and configured to execute the computer code to perform a method comprising:

acquiring a plurality of images generated by an inspection system for a location of a defect detected on a wafer by the inspection system, wherein the location on the wafer is comprised of a plurality of stacked layers, and wherein each image of the plurality of images is generated by the inspection system at the location using a different focus setting;

adding the defect to a training set;

determining a classification of the defect, utilizing the plurality of images wherein the classification of the defect is determined using a through-focus signal profile generated for the defect utilizing the plurality of images, the through-focus signal profile being a 1-dimensional (1D) histogram having the focus setting on a first axis and attributes of a corresponding image on a second axis;

labeling the defect in the training set based on the classification determined for the defect; and

training a machine learning classifier of the inspection system, using the training set, to tune performance of the machine learning classifier during classification of defects for subsequent inspections of wafers having stacked layers.

14. The system of claim 13 , wherein the system comprises the inspection system.

15. The non-transitory computer readable medium of claim 1 , wherein the machine learning classifier is trained using machine learning that is run on the training set.

16. The non-transitory computer readable medium of claim 1 , wherein the training set includes defects detected by the inspection system and having diverse through-focus signal profiles and images.