Classification, search and retrieval of semiconductor processing metrology images using deep learning/convolutional neural networks

1. A method of classifying substrates with a metrology tool comprising:

training a deep learning framework using convolutional neural networks with a training dataset for classifying an image dataset of a substrate;

extracting a new image of the substrate obtained by the metrology tool operating on the substrate;

processing the new image in the trained deep learning framework to find a set of images of the substrate from the image dataset;

classifying the new image from the set of images as a set of analogous features of the substrate; and

identifying at least one root cause for the set of analogous features in the substrate based on a comparison of the new image to the image dataset using a K nearest neighbor (KNN) algorithm.

2. The method of claim 1 , further comprising:

running the image dataset through the K nearest neighbor (KNN) algorithm to extract features at locations on the set of images;

extracting the set of analogous features at the same locations on the new image using the trained deep learning framework;

comparing the set of analogous features of the new image with the features of the image dataset through the KNN algorithm;

providing a set of most similar images from the image dataset based on said comparing; and

identifying the at least one root cause for the set of analogous features from the set of most similar images.

3. The method of claim 2 , further comprising:

writing quality status information to metadata associated with the substrate.

4. The method of claim 3 , wherein each quality status information of the substrate is of a die portion of the substrate.

5. The method of claim 3 , further comprising:

providing a notification of the quality status for the substrate if the quality status is outside a threshold.

6. The method of claim 5 , wherein the notification is an electronic notification provided to a user or a system controller.

7. The method of claim 5 , further comprising:

modifying processing parameters on factory process equipment for processing a next substrate.

8. The method of claim 5 , further comprising:

modifying processing parameters on factory process equipment for processing the substrate.

9. The method of claim 2 , wherein training a deep learning framework comprises:

dividing a dataset of images into the training dataset and the image dataset;

flattening and extracting multiple features into layers in each image of the training dataset;

dropping extraneous feature information for each image; and

classifying each layer in each image, wherein the classification may include a number of categories.

10. A non-transitory computer readable storage medium comprising a program product which, when executed, is configured to perform an operation for operating a network device to classify substrates with a metrology tool, the operation comprising:

training a deep learning framework using convolutional neural networks with a training dataset for classifying an image dataset of a substrate;

extracting a new image of the substrate obtained by the metrology tool operating on the substrate;

processing the new image in the trained deep learning framework to find a set of images of the substrate from the image dataset;

classifying the new image from the set of images as a set of analogous features of the substrate; and

identifying at least one root cause for the set of analogous features in the substrate based on a comparison of the new image to the image dataset using a K nearest neighbor (KNN) algorithm.

11. The non-transitory computer readable storage medium of claim 10 , further comprising:

running the image dataset through the K nearest neighbor (KNN) algorithm to extract features at locations on the set of images;

extracting the set of analogous features at the same locations on the new image using the trained deep learning framework;

comparing the set of analogous features of the new image with the features of the image dataset through the KNN algorithm;

providing a set of most similar images from the image dataset based on said comparing; and

identifying the at least one root cause for the set of analogous features from the set of most similar images.

12. The non-transitory computer readable storage medium of claim 11 , further comprising:

writing quality status information to metadata associated with the substrate.

13. The non-transitory computer readable storage medium of claim 12 , further comprising:

providing an electronic notification of the quality status for the substrate if the quality status is outside a threshold.

14. The non-transitory computer readable storage medium of claim 13 , further comprising:

modifying processing parameters on factory process equipment for processing a next substrate.

15. The non-transitory computer readable storage medium of claim 13 , further comprising:

modifying processing parameters on factory process equipment for processing the substrate.

16. A system, comprising:

a processor; and

a memory, wherein the memory includes an application program configured to perform an operation for classifying substrates with a metrology tool, the operation comprising:

training a deep learning framework using convolutional neural networks with a training dataset for classifying an image dataset of a substrate;

extracting a new image of the substrate obtained by the metrology tool operating on the substrate;

processing the new image in the trained deep learning framework to find a set of images of the substrate from the image dataset;

classifying the new image from the set of images as a set of analogous features of the substrate; and

identifying at least one root cause for the set of analogous features in the substrate based on a comparison of the new image to the image dataset using a K nearest neighbor (KNN) algorithm.

17. The system of claim 16 , further comprising:

running the image dataset through the K nearest neighbor (KNN) algorithm to extract features at locations on the set of images;

extracting the set of analogous features at the same locations on the new image using the trained deep learning framework;

comparing the set of analogous features of the new image with the features of the image dataset through the KNN algorithm;

providing a set of most similar images from the image dataset based on said comparing; and

identifying the at least one root cause for the set of analogous features from the set of most similar images.

18. The system of claim 17 , further comprising:

writing quality status information to metadata associated with the substrate.

19. The system of claim 18 , further comprising:

modifying processing parameters on factory process equipment for processing a next substrate.

20. The system of claim 18 , further comprising:

modifying processing parameters on factory process equipment for processing the substrate.