Training method for learning apparatus, and image generation system
A training method that performs learning using appropriate low-quality image and high-quality image is provided. The invention is directed to a training method including executing learning by inputting a first image generated under a first image generation condition and a second image generated under a second image generation condition different from the first image generation condition to a learning apparatus that adjusts parameters so as to suppress an error between an input image and a converted image, in which the second image is selected such that an index value extracted from the second image is the same as or has a predetermined relationship with an index value extracted from the first image, or the second image is output from a second image generation tool different from a first image generation tool for generating the first image.
1. A training method for a learning apparatus that generates a converted image based on input of an input image acquired by an image generation tool, comprising:
executing learning by inputting a first image generated under a first image generation condition and a second image generated under a second image generation condition different from the first image generation condition to the learning apparatus that adjusts parameters so as to suppress an error between the input image and the converted image one or more times, wherein
the second image is selected such that an index value extracted from the second image is the same as or has a predetermined relationship with an index value extracted from the first image, or
the second image is output from a second image generation tool different from a first image generation tool for generating the first image.
2. The training method according to claim 1 , wherein
the first image is of higher quality than the second image.
3. The training method according to claim 1 , wherein
the image generation tool is a charged particle beam device, and the first image is acquired under the condition that a dose amount per unit area of the charged particle beam device is larger than that of the second image.
4. The training method according to claim 3 , wherein
the first image is acquired in a state in which a number of integrated frames of the charged particle beam device or a beam current is larger, a scanning speed of a charged particle beam is slower, or a field of view of the charged particle beam device is smaller than that of the second image.
5. The training method according to claim 1 , wherein
the first image and a plurality of second images are acquired, a plurality of second index values are extracted from the plurality of second images, and the plurality of extracted second index values are compared with a first index value extracted from the first image.
6. The training method according to claim 1 , wherein
the index value is a dimension value or an area value of a pattern included in the first image and the second image.
7. The training method according to claim 1 , wherein
the first image generation tool is configured to generate a higher quality image than the second image generation tool.
8. The training method according to claim 1 , wherein
the first image generation tool is equipped with an aberration corrector.
9. The training method according to claim 8 , wherein
the image generation tool generates an image of a semiconductor pattern,
the first image is generated under the first image generation condition including aberration correction,
the second image is generated under the second image generation condition that does not include aberration correction, and
in the second image, the index value extracted from the second image is one or more of resolution of the semiconductor pattern, line edge roughness of a line pattern, and line width roughness of the line pattern.
10. The training method according to claim 9 , wherein
the image generation tool is a charged particle beam device,
the semiconductor pattern has a predetermined height, and
the height of the pattern is less than an imaging depth of focus of the first image generation tool and greater than an imaging depth of focus of the second image generation tool.
11. The training method according to claim 10 , wherein
in the semiconductor pattern, a variation in the index value extracted from the first image caused by non-chargeability or a potential change due to irradiation with a charged particle beam is less than or equal to a threshold.
12. The training method according to claim 9 , wherein
line edge roughness or line width roughness including predetermined amplitude or frequency information is built into the semiconductor pattern, or the semiconductor pattern is known.
13. A system configured to generate a converted image based on input of an input image, comprising:
one or more computer subsystems that train a learning apparatus configured to generate a converted image based on the input of the input image and one or more components executed by the one or more computer subsystems, wherein
the one or more components execute learning by inputting a first image generated under a first image generation condition and a second image generated under a second image generation condition different from the first image generation condition to the learning apparatus that adjusts parameters so as to suppress an error between the input image and the converted image one or more times during learning from an input device,
the second image is selected such that an index value extracted from the second image is the same as or has a predetermined relationship with an index value extracted from the first image, or
the second image is output from a second image generation tool different from a first image generation tool for generating the first image.
14. The system according to claim 13 , wherein
the one or more components train the learning apparatus with teacher data including the first image and the selected second image.
15. The system according to claim 14 , wherein
the one or more components select a second image that satisfies a predetermined condition based on an execution of dimension measurement or area measurement of a pattern included in a plurality of second images generated under the second image generation condition, compare a measurement value of the selected second image with a measurement value of the first image, and select the second image based on the comparison.
16. The system according to claim 14 , wherein
the first image generation tool and the second image generation tool include the component,
a first component and a second component are connected through a network,
the first component trains the learning apparatus based on input of images output from the first image generation tool and the second image generation tool, and
the trained model is introduced into the second component via the network.
17. A non-transitory computer-readable medium storing a program configured to instruct a processor to train a learning apparatus configured to generate a converted image based on input of an image, wherein
the program causes learning to be executed by inputting a first image generated under a first image generation condition and a second image generated under a second image generation condition different from the first image generation condition to the learning apparatus that adjusts parameters so as to suppress an error between the input image and the converted image one or more times during learning from an input device,
the second image is selected such that an index value extracted from the second image is the same as or has a predetermined relationship with an index value extracted from the first image, or
the second image is output from a second image generation tool different from a first image generation tool for generating the first image.