Machine learning based examination for process monitoring

Patent metadata

Field	Value
Publication number	US-12561793-B2
Application number	US-202318113032-A
Country	US
Kind code	B2
Filing date	Feb 22, 2023
Priority date	Feb 22, 2023
Publication date	Feb 24, 2026
Grant date	Feb 24, 2026

Abstract

There is provided a system and method of examination of semiconductor specimens. The method includes generating a sequence of anomaly scores corresponding to a sequence of specimens sequentially fabricated and examined during a fabrication process, comprising, for each given specimen: obtaining an image of the given specimen acquired by an examination tool; using a machine learning (ML) model to process the image and obtaining an anomaly map indicative of pattern variation in the image; and deriving, based on the anomaly map, an anomaly score indicative of level of pattern variation presented in the given specimen, wherein the anomaly score is correlated with a defectivity score related to defect detection in a correlation relationship, and has higher detection sensitivity than the defectivity score; and analyzing the sequence of anomaly scores to monitor on-going process stability, thereby providing defect related prediction along the fabrication process based on the correlation relationship.

First claim

What is claimed is:

1 . A computerized system of examining semiconductor specimens, the system comprising a processor and memory circuitry (PMC) configured to:
generate a sequence of anomaly scores corresponding to a sequence of specimens sequentially fabricated and examined during a fabrication process thereof, comprising, for each given specimen in the sequence: obtaining an image of the given specimen acquired by an examination tool, using a machine learning (ML) model to process the image to generate an anomaly map comprising pixel level probabilities of presence of pattern variation in the image, wherein the ML model is previously trained during a training phase for pattern variation detection, and deriving, based on the anomaly map, an anomaly score indicative of a level of pattern variation presented in at least part of the given specimen, wherein the anomaly score is correlated with a defectivity score related to defect detection in a correlation relationship derived based on defect data collected from a stack of specimens previously examined, and wherein the anomaly score exhibits higher detection sensitivity than the defectivity score; and analyze the sequence of anomaly scores to monitor on-going process stability, thereby providing a defect related prediction along the fabrication process based on the correlation relationship, and generate an alert of yield drop upon detection of a trend of increasing defect presence.

2 . The computerized system according to claim 1 , wherein the pattern variation relates to at least one of: bent lines, edge roughness, surface roughness, critical dimension (CD) variation, missing patterns, and gray level variation.

3 . The computerized system according to claim 1 , wherein the ML model is previously trained using unsupervised learning based on a training set comprising a plurality of training images of which a majority represents normal pattern behaviors without pattern variations.

4 . The computerized system according to claim 1 , wherein the defectivity score is obtained in accordance with a defectivity metric representing a defect characteristic.

5 . The computerized system according to claim 1 , wherein the correlation relationship is derived by: collecting defect data from the stack of specimens previously examined, determining locations on the stack of specimens with a presence of defects based on the defect data, processing images of at least some of the locations using the ML model to obtain anomaly maps and anomaly scores thereof, and correlating the anomaly scores and corresponding defect data to derive the correlation relationship.

6 . The computerized system according to claim 1 , wherein the sequence of specimens is a sequence of dies on a wafer, and the sequence of anomaly scores are analyzed for monitoring pattern variation uniformity across the wafer.

7 . The computerized system according to claim 1 , wherein the sequence of specimens is a sequence of wafers, and the sequence of anomaly scores are analyzed for monitoring process variation along the fabrication process of the sequence of wafers.

8 . The computerized system according to claim 1 , wherein the sequence of specimens comprises a plurality of subsequences of specimens respectively fabricated by a plurality of fabrication tools, and the generating comprises generating a plurality of subsequences of anomaly scores corresponding to the plurality of subsequences of specimens, and the analyzing comprises separately and collectively analyzing the plurality of subsequences of anomaly scores to verify a root cause of one or more pattern variations indicated by one or more anomaly scores.

9 . The computerized system according to claim 1 , wherein the anomaly score is derived based on the anomaly map for one of: one or more pixels in the image, one or more structures represented in the image, or the image as a whole.

10 . The computerized system according to claim 1 , wherein the ML model is a variational auto-encoder (VAE) configured to map the image to one or more latent variables in a latent space each representing a respective feature extracted from the image, and model a probability distribution of each feature, thereby allowing to visualization of different levels of pattern variations impacting the feature.

11 . The computerized system according to claim 10 , wherein at least one latent variable in the latent space represents a specific pattern variation of a structure in the image, and wherein the PMC is further configured to provide an indication of the level of pattern variation of the structure based on a value of the at least one latent variable extracted from the latent space and the probability distribution of the specific pattern variation, and determine the level of pattern variation of the structure based on the indication and the anomaly score derived for the structure.

12 . A computerized method of examining semiconductor specimens, the method comprising:
generating a sequence of anomaly scores corresponding to a sequence of specimens sequentially fabricated and examined during a fabrication process thereof, comprising, for each given specimen in the sequence:
obtaining an image of the given specimen acquired by an examination tool,
using a machine learning (ML) model to process the image to generate an anomaly map comprising pixel level probabilities of a presence of a pattern variation in the image, wherein the ML model is previously trained during a training phase for pattern variation detection, and
deriving, based on the anomaly map, an anomaly score indicative of a level of pattern variation presented in at least part of the given specimen, wherein the anomaly score is correlated with a defectivity score related to defect detection in a correlation relationship derived based on defect data collected from a stack of specimens previously examined, and wherein the anomaly score exhibits higher detection sensitivity than the defectivity score; and
analyzing the sequence of anomaly scores to monitor on-going process stability, thereby providing defect related prediction along the fabrication process based on the correlation relationship, and generate an alert of yield drop upon detection of a trend of increasing defect presence.

13 . The computerized method according to claim 12 , wherein the ML model is previously trained using unsupervised learning based on a training set comprising a plurality of training images of which a majority represents normal pattern behaviors without pattern variations.

14 . The computerized method according to claim 12 , wherein the correlation relationship is derived by: collecting defect data from the stack of specimens previously examined, determining locations on the stack of specimens with a presence of defects based on the defect data, processing images of at least some of the locations using the ML model to obtain anomaly maps and anomaly scores thereof, and correlating the anomaly scores and corresponding defect data to derive the correlation relationship.

15 . The computerized method according to claim 12 , wherein the sequence of specimens is a sequence of dies on a wafer, and the sequence of anomaly scores are analyzed for monitoring pattern variation uniformity across the wafer.

16 . The computerized method according to claim 12 , wherein the sequence of specimens is a sequence of wafers, and the sequence of anomaly scores are analyzed for monitoring process variation along the fabrication process of the sequence of wafers.

17 . The computerized method according to claim 12 , whe

Assignees

• Applied Materials Israel Ltd

Inventors

• Tal Noam
• Levant Boris
• Sinitsa Sergey
• Sturlesi Boaz
• Yogev Shay
• Ariel Assaf
• Choona Lilach
• Pres Shaul

Classifications

• H10P74/203

  Structural properties, e.g. testing or measuring thicknesses, line widths, warpage, bond strengths or physical defects · CPC title

• H10P74/23

  characterised by multiple measurements, corrections, marking or sorting processes · CPC title

• G06V10/70

  using pattern recognition or machine learning (optical pattern recognition or electronic computations therefor G06V10/88) · CPC title

• G06T2207/20081

  Training; Learning · CPC title

• G01N21/8851 Primary

  Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges (G01N21/8806 and G01N21/93 - G01N21/95692 take precedence; optical measurement of dimensions G01B11/00; optical scanning G02B26/10; image transformation G06T3/00; computerised image enhancement G06T5/00; image processing per se for flaw detection G06T7/0002) · CPC title