Ensemble of deep learning models for defect review in high volume manufacturing

Patent metadata

Field	Value
Publication number	US-12211196-B2
Application number	US-202318305287-A
Country	US
Kind code	B2
Filing date	Apr 21, 2023
Priority date	Apr 21, 2023
Publication date	Jan 28, 2025
Grant date	Jan 28, 2025

Abstract

Methods and systems for detecting defects in images of a specimen are provided. One system includes a computer subsystem configured for training an ensemble of deep learning models by altering one or more parameters of the ensemble until a pseudo-loss function determined based on output of the ensemble is approximately equal to but not greater than 0.5. The computer subsystem is also configured for detecting defects in runtime specimen images by inputting the runtime specimen images into the trained ensemble and generating runtime labels for the runtime specimen images indicating if a defect has been detected in the runtime specimen images based on outputs of the deep learning models in the trained ensemble.

First claim

The invention claimed is:

1. A system configured to detect defects in images of a specimen, comprising:
a computer subsystem; and
one or more components executed by the computer subsystem, wherein the one or more components comprise an ensemble of deep learning models and a pseudo-loss function based on output generated by the ensemble of deep learning models; and
wherein the computer subsystem is configured for:
training the ensemble with a training dataset comprising training specimen images and training labels indicating if a defect is detected in the training specimen images, wherein the training comprises altering one or more parameters of the ensemble until the pseudo-loss function determined based on the output of the ensemble is approximately equal to but not greater than 0.5; and
detecting defects in runtime specimen images by inputting the runtime specimen images into the trained ensemble of deep learning models and generating runtime labels for the runtime specimen images indicating if a defect has been detected in the runtime specimen images based on outputs of the deep learning models in the trained ensemble.

2. The system of claim 1 , wherein the outputs of each of the deep learning models in the trained ensemble are a same type of information for the runtime specimen images.

3. The system of claim 1 , wherein the training specimen images comprise less than 150 positive and negative examples of defects.

4. The system of claim 1 , wherein each of the deep learning models comprises 6 convolutional layers and about 100,000 parameters.

5. The system of claim 1 , wherein a sequence of weight and bias matrices represent layers of each of the deep learning models.

6. The system of claim 1 , wherein the training further comprises determining the pseudo-loss function for each of the deep learning models in the ensemble and each sample in the training dataset.

7. The system of claim 1 , wherein the pseudo-loss function is configured so that a false positive defect detection continuously increases a value of the pseudo-loss function and a true defect detection reduces the value of the pseudo-loss function in proportion to a maximum size of the true defect detection.

8. The system of claim 1 , wherein the pseudo-loss function is configured so that weak true detections and false positive detections affect the pseudo-loss function on a continuous scale.

9. The system of claim 1 , wherein the computer subsystem is further configured for computing a probability weight for each of multiple samples in the training dataset and selecting a portion of the multiple samples used for the training based on the computed probability weight for each of the multiple samples.

10. The system of claim 1 , wherein the outputs of the deep learning models in the trained ensemble comprise two-dimensional probability map outputs, and wherein said generating the runtime labels comprises generating a two-dimensional weighted average map from the two-dimensional probability map outputs and assigning the runtime labels based on whether the two-dimensional weighted average map contains one or more detections that meet predetermined threshold and size criteria.

11. The system of claim 1 , wherein the outputs of the deep learning models in the trained ensemble comprise two-dimensional probability map outputs, and wherein said generating the runtime labels comprises computing a final scalar decision function on a combination of the two-dimensional probability map outputs of each of the deep learning models in the trained ensemble.

12. The system of claim 1 , wherein the training does not comprise optimizer hyper-parameter tuning of any of the deep learning models in the ensemble.

13. The system of claim 1 , wherein at least one of the deep learning models in the ensemble is configured as a weak learning algorithm.

14. The system of claim 1 , wherein the training further comprises hypothesis boosting.

15. The system of claim 1 , wherein the runtime specimen images are generated during a defect review process performed on the specimen in a high volume manufacturing process.

16. The system of claim 1 , wherein the training specimen images and the runtime specimen images are generated by an imaging subsystem of a defect review tool.

17. The system of claim 1 , wherein the training specimen images and the runtime specimen images are generated by an electron beam-based imaging subsystem.

18. The system of claim 1 , wherein the runtime specimen images input into the trained ensemble of deep learning models for any one location on the specimen comprise images generated with multiple detectors of an imaging subsystem.

19. A non-transitory computer-readable medium, storing program instructions executable on a computer system for performing a computer-implemented method for detecting defects in images of a specimen, wherein the computer-implemented method comprises:
training an ensemble of deep learning models with a training dataset comprising training specimen images and training labels indicating if a defect is detected in the training specimen images, wherein the training comprises altering one or more parameters of the ensemble until a pseudo-loss function determined based on output of the ensemble is approximately equal to but not greater than 0.5, and wherein one or more components executed by the computer system comprise the ensemble of deep learning models and the pseudo-loss function; and
detecting defects in runtime specimen images by inputting the runtime specimen images into the trained ensemble of deep learning models and generating runtime labels for the runtime specimen images indicating if a defect has been detected in the runtime specimen images based on outputs of the deep learning models in the trained ensemble.

20. A computer-implemented method for detecting defects in images of a specimen, comprising:
training an ensemble of deep learning models with a training dataset comprising training specimen images and training labels indicating if a defect is detected in the training specimen images, wherein the training comprises altering one or more parameters of the ensemble until a pseudo-loss function determined based on output of the ensemble is approximately equal to but not greater than 0.5, and wherein one or more components executed by a computer system comprise the ensemble of deep learning models and the pseudo-loss function; and
detecting defects in runtime specimen images by inputting the runtime specimen images into the trained ensemble of deep learning models and generating runtime labels for the runtime specimen images indicating if a defect has been detected in the runtime specimen images based on outputs of the deep learning models in the trained ensemble, wherein the training and the detecting are performed by the computer system.

Assignees

• Kla Corp

Inventors

• Virk Kuljit S
• Andreeva Vera
• Muray Lawrence

Classifications

• G06T2207/30148

  Semiconductor; IC; Wafer · CPC title

• G06T2207/10061

  from scanning electron microscope · CPC title

• G06V10/776 Primary

  Validation; Performance evaluation · CPC title

• G06T2207/20081

  Training; Learning · CPC title

• G06V20/70

  Labelling scene content, e.g. deriving syntactic or semantic representations · CPC title