Self-measurement of semiconductor image using deep learning

1. A method comprising:

receiving an image of a product associated with a manufacturing process, wherein the product comprises a plurality of structures;

identifying, using a trained machine learning model, a segment of the image that comprises a structure of the plurality of structures, wherein the trained machine learning model having been trained using 1) an original image generated by an image capturing device and 2) a plurality of synthetic images each generated by identifying a feature, in the original image, that is expect to change based on manufacturing parameters of the manufacturing process, and generating a variation in the feature by changing a position of at least one value, associated with the feature, from a first location to a second location;

determining a plurality of image measurements of the segment that comprises the structure; and

storing the plurality of image measurements.

2. The method of claim 1 , wherein identifying the segment comprises performing instance segmentation using the trained machine learning model to identify a plurality of segments, wherein each of the plurality of segments comprises one of the plurality of structures.

3. The method of claim 1 , wherein the trained machine learning model comprises a trained convolutional neural network that comprises a convolution operation that decreases spatial resolution of image data and a deconvolution operations that increases spatial resolution of the image data.

4. The method of claim 3 , wherein the deconvolution operation comprises a convolution operation that uses a stride rate that is less than one.

5. The method of claim 1 , wherein identifying the segment comprises generating a segmentation map that indicates pixels of the image that comprise the structure of the product associated with the manufacturing process.

6. The method of claim 1 , wherein determining the plurality of image measurements comprises:

detecting a first edge of the segment that comprises the structure;

identifying a first point on the first edge and a second point on a second edge, wherein the second edge corresponds an edge of the structure or an edge of another structure of the plurality of structures; and

calculating one of the plurality of image measurements based on a distance between the first point and second point.

7. The method of claim 1 , wherein determining the plurality of image measurements comprises:

determining a first set of image measurements along a first axis, wherein the first set comprises a distance between edges for each pixel row of the segment; and

determining a second set of image measurements along a second axis, wherein the second set comprises a distance between edges for each pixel column of the segment; and

selecting the plurality of image measurements from the first set and the second set.

8. The method of claim 1 , further comprising selecting the plurality of image measurements from a set of image measurements of the plurality of structures, wherein the plurality of image measurements comprise a plurality of horizontal measurements from the set and a plurality of vertical measurements from the set.

9. The method of claim 1 , further comprising updating a manufacturing parameter of the manufacturing process based on an image measurement of the plurality of image measurements.

10. A system comprising:

a memory; and

a processing device, coupled to the memory, to:

access a first set of images, wherein the first set comprises a training image that is generated by an image capturing device and is annotated to indicate a portion of the training image comprising a product structure;

create a second set of images based on the first set of images, wherein one or more image of the second set of images are created by:

identifying a feature, in the training image, that is expect to change based on manufacturing parameters of a manufacturing process; and

generating a variation in the feature by changing a position of at least one value, associated with the feature, from a first location to a second location; and

train a machine learning model using the first set of images and the second set of images.

11. The system of claim 10 , wherein the processing device is further to:

receive an image of a product associated with a manufacturing process, wherein the product comprises a plurality of structures;

identify, using the machine learning model, a segment of the image that comprises a structure of the plurality of structures;

determine a plurality of image measurements of the segment that comprises the structure; and

store the plurality of image measurements.

12. The system of claim 11 , wherein to identify the segment, the processing device performs instance segmentation using the machine learning model to identify a plurality of segments, wherein each of the plurality of segments comprises one of the plurality of structures.

13. The system of claim 10 , wherein the machine learning model comprises a convolutional neural network that comprises a convolution operation that decreases spatial resolution of image data and a deconvolution operations that increases spatial resolution of the image data.

14. The system of claim 13 , wherein the deconvolution operation comprises a convolution operation that uses a stride rate that is less than one.

15. The system of claim 11 , wherein to identify the segment, the processing device is to generate a segmentation map that indicates pixels of the image that comprise the structure of the product associated with the manufacturing process.

16. A method comprising:

accessing, by a processing device, a set of images, wherein the set of images comprises an image associated with an annotation that indicates a portion of the image that comprises a structure;

generating a synthetic image by:

identifying a feature, in the image associated with the annotation, that is expected to change based on manufacturing parameters of a manufacturing process, and

generating a variation in the feature by changing a position of at least one value, associated with the feature, from a first location to a second location;

generating a machine learning model that comprises a resolution contracting layer that is based on a convolution operation and a resolution expanding layer that is based on a deconvolution operation;

updating the machine learning model to change a value for the convolution operation and a value for the deconvolution operation based on the image associated with the annotation and the synthetic image; and

storing, by the processing device to a data store, the machine learning model comprising the value for the convolution operation and the value for the deconvolution operation.

17. The method of claim 16 , wherein the machine learning model comprises a convolutional neural network for analyzing an image of a product associated with a manufacturing process.

18. The method of claim 16 , wherein the machine learning model receives an input image of a product and generates a segmentation map that indicates pixels of the input image that correspond to a structure of the product.

19. The method of claim 16 , wherein the convolution operation decreases a spatial resolution of a feature map and the deconvolution operation increases the spatial resolution of the feature map.

20. The method of claim 16 , wherein the deconvolution operation comprises a convolution operation that uses a stride rate that is less than one.

21. The method of claim 1 , wherein the value comprises a pixel value.