Deep learning based mode selection for inspection

Patent metadata

Field	Value
Publication number	US-12480890-B2
Application number	US-202418646704-A
Country	US
Kind code	B2
Filing date	Apr 25, 2024
Priority date	Jul 14, 2023
Publication date	Nov 25, 2025
Grant date	Nov 25, 2025

Abstract

Methods and systems for determining information for a specimen are provided. One system includes a computer subsystem and one or more components executed by the computer subsystem that include a setup deep learning (DL) model configured for separately performing defect detection for a specimen based on output generated for the specimen by each of two or more modes of an inspection system, respectively, and separately re-performing defect detection for the specimen based on masked output generated for each of the modes, respectively. The computer subsystem determines a difference between results of separately performing and separately re-performing the defect detections for each of the modes, respectively, and identifies a subset of the modes for which the difference is larger than other modes as candidate mode(s) for inspection of the specimen.

First claim

The invention claimed is:

1 . A system configured for determining information for a specimen, comprising:
a computer subsystem; and one or more components executed by the computer subsystem, wherein the one or more components comprise a setup deep learning model configured for:
separately performing defect detection for a specimen based on output generated for the specimen by each of two or more modes of an inspection system, respectively; and
separately re-performing defect detection for the specimen based on masked output generated for said each of the two or more modes, respectively; and
wherein the computer subsystem is configured for:
separately generating the masked output for said each of the two or more modes, respectively, by determining a single output value from the output generated for at least a portion of the specimen and replacing all of the output generated for the at least the portion of the specimen with the single output value;
determining a difference between results of said separately performing defect detection and said separately re-performing defect detection for said each of the two or more modes, respectively; and
identifying a subset of the two or more modes for which the determined difference is larger than others of the two or more modes as one or more candidate modes for inspection of the specimen.

2 . The system of claim 1 , wherein the computer subsystem is further configured for selecting one or more parameters of the setup deep learning model based on all of the output generated for the specimen by said each of the two or more modes.

3 . The system of claim 1 , wherein the computer subsystem is further configured for separately training the setup deep learning model for said each of the two or more modes of the inspection subsystem, at least one combination of the two or more modes in the subset, and not any other possible combinations of the two or more modes.

4 . The system of claim 1 , wherein the setup deep learning model is further configured as a multi-mode deep learning model.

5 . The system of claim 1 , wherein the single output value is a single median value.

6 . The system of claim 1 , wherein the setup deep learning model is further configured for separately re-performing the defect detection for the specimen based on the masked output generated for at least one combination of the two or more modes.

7 . The system of claim 1 , wherein the two or more modes comprise modes included in a coupling group.

8 . The system of claim 7 , wherein the computer subsystem is further configured for separately training the setup deep learning model as a single mode deep learning model for each of the modes in the coupling group, respectively.

9 . The system of claim 8 , wherein the computer subsystem is further configured for identifying a first portion of the modes in the coupling group that have a higher sensitivity for the inspection than a second portion of the modes in the coupling group.

10 . The system of claim 9 , wherein identifying the first portion comprises performing said separately performing, said separately re-performing, said separately generating, and said determining for said each of the modes in the coupling group and identifying the first portion as the modes in the coupling group for which the determined difference is larger than other of the modes in the coupling group.

11 . The system of claim 9 , wherein the computer subsystem is further configured for re-training the setup deep learning model as a multi-mode deep learning model with the output generated for the specimen by only the first portion of the modes.

12 . The system of claim 11 , wherein the re-trained setup deep learning model is configured for separately performing multi-mode defect detection for the specimen based on different combinations of the masked output generated for each of the modes in the first portion, respectively, with the output generated for each other of the modes in the first portion.

13 . The system of claim 12 , wherein the computer subsystem is further configured for determining if the re-trained setup deep learning model is biased to one or more modes in the first portion based on results of said separately performing the multi-mode defect detection.

14 . The system of claim 12 , wherein the computer subsystem is further configured for determining if the re-trained setup deep learning model is biased to two or more modes in the first portion based on results of said separately performing the multi-mode defect detection, and when the computer subsystem determines that the re-trained setup deep learning model is biased to the two or more modes in the first portion, identifying the two or more modes in the first portion as coupled modes.

15 . The system of claim 14 , wherein the computer subsystem is further configured for eliminating all but one of the coupled modes from the coupling group.

16 . The system of claim 1 , wherein the computer subsystem is further configured for quantitatively ranking said each of the two or more modes based on the determined difference.

17 . The system of claim 1 , wherein the computer subsystem is further configured for training an additional deep leaning model configured for performing the defect detection or another defect detection for the specimen or another specimen based on results generated by the setup deep learning model or the computer subsystem.

18 . The system of claim 1 , wherein the setup deep learning model is not used for defect detection performed during the inspection of the specimen.

19 . A non-transitory computer-readable medium, storing program instructions executable on a computer system for performing a computer-implemented method for determining information for a specimen, wherein the computer-implemented method comprises:
separately performing defect detection for a specimen by inputting output generated for the specimen by each of two or more modes of an inspection system, respectively, into a setup deep learning model included in one or more components executed by the computer system; separately generating masked output for said each of the two or more modes, respectively, by determining a single output value from the output generated for at least a portion of the specimen and replacing all of the output generated for the at least the portion of the specimen with the single output value; separately re-performing defect detection for the specimen by inputting the masked output generated for said each of the two or more modes, respectively, into the setup deep learning model; determining a difference between results of said separately performing defect detection and said separately re-performing defect detection for said each of the two or more modes, respectively; and identifying a subset of the two or more modes for which the determined difference is larger than others of the two or more modes as one or more candidate modes for inspection of the specimen, wherein said inputting the output, separately generating, inputting the masked output, determining, and identifying are performed by the computer system.

20 . A computer-implemented method for determining information for a specimen, comprising:
separately performing defect detection for a specimen by inputting output generated for the specimen by each of two or more modes of an inspection system, respectively, into a setup deep learning model included in one or more components executed by a computer system; separately generating m

Assignees

• Kla Corp

Inventors

• Theagarajan Rajkumar
• Dong Yujie
• Zhang Jing
• Duffy Brian
• Chowdhury Atiqur Rahman
• Bhaskar Kris
• Li Yang
• Jensen Graham

Classifications

• G06T7/0004

  Industrial image inspection · CPC title

• G06T2207/10061

  from scanning electron microscope · CPC title

• G06T2207/20084

  Artificial neural networks [ANN] · CPC title

• G06T2207/30148

  Semiconductor; IC; Wafer · CPC title

• G01N2201/1296

  using neural networks · CPC title