Measurement system, method for generating learning model to be used when performing image measurement of semiconductor including predetermined structure, and recording medium for storing program for causing computer to execute processing for generating learning model to be used when performing image measurement of semiconductor including predetermined structure
The present invention proposes a technique for enabling the execution of measurement processing without referring to a design drawing for which it is difficult to adjust or obtain parameters for image processing that requires knowhow. This measurement system according to the present disclosure refers to a learning model generated on the basis of teaching data, which is generated from a sample image of a semiconductor, and the sample image, generates a region-segmented image from an input image (measurement subject) of a semiconductor having a predetermined structure, and uses the region-segmented image to perform image measurement. Here, the teaching data is an image in which labels, which include a structure of the semiconductor in the sample image, are assigned to each pixel of the image, and the learning model includes parameters for deducing teaching data from the sample image (see indicator 1).
1. A measurement system that performs image measurement of a semiconductor having a periodical structure, comprising:
at least one processor that performs various processes relating to the image measurement; and
an output device that outputs a result of the image measurement, wherein
the at least one processor performs
a process of generating training data from a sample image of the semiconductor,
a process of generating a learning model based on the sample image and the training data,
a process of generating a region-segmented image from an input image relating to the semiconductor based on the learning model,
a measurement process of performing the image measurement using the region-segmented image, and
a process of outputting a result of the measurement process to the output device,
the training data is an image in which a label including a structure of the semiconductor in the sample image is assigned to each of pixels of the image,
the learning model includes a parameter for inferring the training data or the region-segmented image from the sample image or the input image,
the sample image and the training data are smaller than the input image and include an image region corresponding to the periodical structure, and
the at least one processor generates the parameter of the learning model from the sample image and the training data in the process of generating the learning model.
2. The measurement system according to claim 1 , wherein the learning model is a machine learning model that refers to a region present near each of the pixels in the input image in order to determine the label assigned to each of the pixels.
3. The measurement system according to claim 1 , wherein the learning model is a convolution neural network.
4. The measurement system according to claim 1 , wherein the image region corresponds to a structure of at least one period in the periodical structure.
5. A measurement system that performs image measurement of a semiconductor having a predetermined structure, comprising:
at least one processor that performs various processes relating to the image measurement; and
an output device that outputs a result of the image measurement, wherein
the at least one processor performs
a process of generating training data from a sample image of the semiconductor,
a process of generating a learning model based on the sample image and the training data,
a process of generating a region-segmented image from an input image relating to the semiconductor based on the learning model,
a measurement process of performing the image measurement using the region-segmented image, and
a process of outputting a result of the measurement process to the output device,
the training data is an image in which a label including a structure of the semiconductor in the sample image is assigned to each of pixels of the image,
the learning model includes a parameter for inferring the training data or the region-segmented image from the sample image or the input image,
the at least one processor performs a process of segmenting the region-segmented image into small regions of an image size smaller than the region-segmented image according to the label and grouping the small regions based on types of the small regions, and
the at least one processor performs overlay measurement from the center of gravity of each of the grouped small regions as the measurement process.
6. A measurement system that performs image measurement of a semiconductor having a predetermined structure, comprising:
at least one processor that performs various processes relating to the image measurement; and
an output device that outputs a result of the image measurement, wherein
the at least one processor performs
a process of generating training data from a sample image of the semiconductor,
a process of generating a learning model based on the sample image and the training data,
a process of generating a region-segmented image from an input image relating to the semiconductor based on the learning model,
a measurement process of performing the image measurement using the region-segmented image, and
a process of outputting a result of the measurement process to the output device,
the training data is an image in which a label including a structure of the semiconductor in the sample image is assigned to each of pixels of the image,
the learning model includes a parameter for inferring the training data or the region-segmented image from the sample image or the input image,
the sample image includes a combination of images under different imaging conditions obtained by imaging the same position on the semiconductor under the different imaging conditions a plurality of times, and
the at least one processor generates the training data from the sample image according to the different imaging conditions and generates the learning model based on the training data generated according to the imaging conditions and the sample image.
7. A measurement system that performs image measurement of a semiconductor having a predetermined structure, comprising:
at least one processor that performs various processes relating to the image measurement; and
an output device that outputs a result of the image measurement, wherein
the at least one processor performs
a process of generating training data from a sample image of the semiconductor,
a process of generating a learning model based on the sample image and the training data,
a process of generating a region-segmented image from an input image relating to the semiconductor based on the learning model,
a measurement process of performing the image measurement using the region-segmented image, and
a process of outputting a result of the measurement process to the output device,
the training data is an image in which a label including a structure of the semiconductor in the sample image is assigned to each of pixels of the image,
the learning model includes a parameter for inferring the training data or the region-segmented image from the sample image or the input image,
the imaging under the different imaging conditions includes at least one of imaging with different acceleration voltages, capturing different types of electron images, and changing a synthesis ratio for generation of a synthesized image of different types of electron images.
8. A measurement system that performs image measurement of a semiconductor having a predetermined structure, comprising:
at least one processor that performs various processes relating to the image measurement; and
an output device that outputs a result of the image measurement, wherein
the at least one processor performs
a process of generating training data from a sample image of the semiconductor,
a process of generating a learning model based on the sample image and the training
a process of generating a region-segmented image from an input image relating to the semiconductor based on the learning model,
a measurement process of performing the image measurement using the region-segmented image, and
a process of outputting a result of the measurement process to the output device,
the training data is an image in which a label including a structure of the semiconductor in the sample image is assigned to each of pixels of the image,
the learning model includes a parameter for inferring the training data or the region-segmented image from the sample image or the input image,
the at least one processor segments the sample image into two or more sample image groups, assigns the label to an image included in a first sample image group to generate first training data, generates an intermediate learning model based on the image of the first sample image group and the first training data, adds training data generated by inferring an image included in an image group other than the first sample image group based on the intermediate learning model to the first training data to generate second training data, and generates, based on the sample image and the second training data, the learning model to be applied to the input image.
9. The measurement system according to claim 8 , wherein
the at least one processor performs correction based on a statistical process on the training data generated by inferring the image included in the image group other than the first sample image group.
10. The measurement system according to claim 9 , wherein
the at least one processor performs the correction based on the statistical process on a plurality of images obtained by repeatedly imaging the same position on the semiconductor.
11. The measurement system according to claim 9 , wherein
the at least one processor extracts partial regions having a high similarity among the partial regions in the sample image and performs the correction based on the statistical process on the extracted partial regions.
12. The measurement system according to claim 9 , wherein
the correction based on the statistical process is to move the small regions in parallel in units of the small regions to which the label is assigned in the second training data or geometrically deform the small regions in units of the small regions to which the label is assigned in the second training data.
13. The measurement system according to claim 5 , wherein
the training data includes a positional information image indicating a displacement from each of the pixels to representative positions of the small regions to which the label is assigned, and
the at least one processor generates the region-segmented image of the input image and the positional information image based on the learning model including the positional information image and uses the positional information image to perform the overlay measurement on the grouped small regions.
14. The measurement system according to claim 13 , wherein the positional information image indicates the displacement calculated using the training data subjected to the correction based on the statistical process.
15. A measurement system that performs image measurement of a semiconductor having a predetermined structure, comprising:
at least one processor that performs various processes relating to the image measurement; and
an output device that outputs a result of the image measurement, wherein
the at least one processor performs
a process of generating training data from a sample image of the semiconductor,
a process of generating a learning model based on the sample image and the training
a process of generating a region-segmented image from an input image relating to the semiconductor based on the learning model,
a measurement process of performing the image measurement using the region-segmented image, and
a process of outputting a result of the measurement process to the output device,
the training data is an image in which a label including a structure of the semiconductor in the sample image is assigned to each of pixels of the image,
the learning model includes a parameter for inferring the training data or the region segmented image from the sample image or the input image,
the at least one processor performs a process of changing the layout of the training data to generate changed training data, adding the changed training data to the training data before the layout change to obtain updated training data, and a process of adding an image inferred from the changed training data to the sample image to obtain an updated sample image, and
the at least one processor generates the learning model based on the updated training data and the updated sample image.
16. The measurement system according to claim 15 , wherein
the at least one processor changes the layout of the training data in consideration of occlusion between labels included in the training data.
17. The measurement system according to claim 1 , wherein
the measurement process is an overlay measurement process, a dimension measurement process, a defective pattern detection process, or a pattern matching process that is performed on the semiconductor.
18. A method for generating a learning model to be used when performing image measurement of a semiconductor including a predetermined structure, the method comprising:
causing at least one processor to generate training data by assigning a label including a structure of at least one measurement target to a region-segmented image obtained from a sample image of the semiconductor; and
causing the at least one processor to generate the learning model using the region-segmented image of the sample image and the training data based on a network structure of a plurality of layers, wherein
the learning model includes a parameter for inferring the training data or the region-segmented image from the sample image or the input image related to the semiconductor,
the at least one processor performs a process of segmenting the region-segmented image into small regions of an image size smaller than the region-segmented image according to the label and grouping the small regions based on types of the small regions, and
the at least one processor performs overlay measurement from the center of gravity of each of the grouped small regions as the measurement process.
19. A non-transitory recording medium storing a program for causing a computer to execute processing for generating a learning model to be used when performing image measurement of a semiconductor including a predetermined structure, the program causing the computer to execute:
processing for generating training data by assigning a label including a structure of at least one measurement target to a region-segmented image obtained from a sample image of the semiconductor; and
processing for generating the learning model using the region-segmented image of the sample image and the training data based on a network structure of a plurality of layers by the at least one processor, wherein
the learning model includes a parameter for inferring the training data or the region-segmented image from the sample image or the input image related to the semiconductor, and
the program further causes the computer to execute:
performing a process of segmenting the region-segmented image into small regions of an image size smaller than the region-segmented image according to the label and grouping the small regions based on types of the small regions, and
performing overlay measurement from the center of gravity of each of the grouped small regions as the measurement process.