Feature selection and automated process window monitoring through outlier detection

1. A method for generating a classifier comprising:

generating at least one wafer image using an imaging device that produces a beam, the wafer image comprising a set of die images;

generating, using a processor, a median die of a wafer and a standard deviation of the wafer based on the set of die images;

generating, using the processor, a segment mask based on the median die of the wafer and based on the standard deviation of the wafer;

applying, using the processor, the segment mask to each of the die images to create a plurality of segmented die images for each of the die images;

representing, using the processor, each of the dies with a vector of feature values, wherein the feature values are of wafer-level features of the wafer image;

calculating, using the processor, die metrics for each of the die images based on the vector; and

identifying, using the processor, one or more statistically significant combinations of the die metrics, the segment mask, and the wafer image to generate a classifier.

2. The method of claim 1 , wherein the die metrics include mean and standard deviation for each of the segmented die images.

3. The method of claim 1 , wherein the die metrics include mean and standard deviation for the segmented die image of a gradient image of the die.

4. The method of claim 3 , wherein the mean and the standard deviation for each of the segmented die images include at least one of a Y-gradient or an X-gradient.

5. The method of claim 1 , wherein the classifier is a one-class Mahalanobis classifier.

6. The method of claim 1 , wherein the classifier is a one-class Support Vector Machine (SVM) classifier.

7. A process control method comprising:

receiving a pre-determined classifier capable of detecting a process non-compliance, wherein the classifier is based on combinations of metrics of masked die images;

generating a wafer image comprising a set of die images using an imaging device that produces a beam;

masking the wafer image using a segment mask;

representing, using a processor, each of the dies in the set of die images with a vector of feature values, wherein the feature values are of wafer-level features of the wafer image;

calculating, using the processor, die metrics for each of the dies;

calculating, using the processor, a metric of the wafer image based on a corresponding combination of the segment mask and the die metrics from the classifier; and

determining, using the processor, a process status using the classifier based on the calculated metrics.

8. The method of claim 7 , wherein the classifier is a one-class Mahalanobis classifier.

9. The method of claim 7 , wherein the classifier is a one-class Support Vector Machine (SVM) classifier.

10. The method of claim 7 , wherein the die metrics include mean and standard deviation for a segmented die image.

11. The method of claim 7 , wherein the die metrics include mean and standard deviation for the segmented die image of a gradient image of the die.

12. The method of claim 11 , wherein the mean and the standard deviation include at least one of a Y-gradient or an X-gradient.

13. The method of claim 7 , further comprising:

training a predictive model with the process status;

calculating an estimated parameter of a fabrication tool based on the process status; and

communicating the estimated parameter to the fabrication tool.

14. The method of claim 7 , further comprising training an auto-encoder using the die images, wherein the auto-encoder is configured to determine focus and exposure skews of a fabrication tool.

15. A process control system comprising:

an imaging device that produces a beam, wherein the imaging device is configured to generate a plurality of images of a surface of a wafer;

a processor;

a storage device in electronic communication with the processor; and

a communication port in electronic communication with the processor and the imaging device;

wherein the processor is programmed to:

receive, at the communication port, the plurality of wafer images of the wafer, each wafer image having a plurality of dies;

mask each of the wafer images using a segment mask to create a set of segmented die images;

represent each of the dies in the set of segmented die images with a vector of feature values, wherein the feature values are of wafer-level features of the wafer image;

calculate die metrics for each of the dies;

calculate a metric of each of the wafer images based on a corresponding combination of the segment mask and the die metrics from a classifier; and

determine a process status using the classifier based on the calculated metrics.

16. The process control system of claim 15 , wherein the classifier is one of a one-class Mahalanobis classifier or a one-class Support Vector Machine (SVM) classifier.

17. The process control system of claim 15 , wherein the die metrics include mean and standard deviation from each of the segmented die images.

18. The process control system of claim 15 , wherein the die metrics include mean and standard deviation for the segmented die image of a gradient image of the die.

19. The process control system of claim 18 , wherein the mean and the standard deviation include at least one of a Y-gradient or an X-gradient.

20. The process control system of claim 15 , wherein the processor is further programmed to:

repeat each program step for additional wafers; and

refine the classifier according to the statistically significant combinations for each wafer.