CN-121981947-A - Method for detecting wafer abnormality and electronic equipment

CN121981947ACN 121981947 ACN121981947 ACN 121981947ACN-121981947-A

Abstract

The application provides a method for detecting wafer abnormality and electronic equipment, the method comprises the steps of obtaining images of core particles in a wafer and position information of the core particles, determining a category corresponding to the core particles and a first abnormality score corresponding to the category according to the images of the core particles, determining position information of target abnormal core particles in a searching range with the target suspicious core particles as the center, and determining second abnormality score of the target suspicious core particles according to the position information of the target suspicious core particles and the position relation of a range corresponding to the defect trend distribution formed by the target abnormal core particles under the condition that the position information of the target abnormal core particles can form defect trend distribution corresponding to the defect type of the target suspicious core particles, and determining the target suspicious core particles to be normal core particles or abnormal core particles according to the second abnormality score and the first abnormality score of the target suspicious core particles. The suspicious core particle fusion self-feature and trend distribution comprehensive judgment is realized, so that the judgment basis is more comprehensive and the dimension is higher, and the occurrence of the phenomena of missing detection or over-detection is reduced.

Inventors

LI ZHENGYI
LIU SHUAI
ZHAO QIDONG
LI JUNYAN
YANG SHANSONG

Assignees

聚好看科技股份有限公司

Dates

Publication Date: 20260505
Application Date: 20251215

Claims (10)

1. A method for detecting wafer anomalies, comprising: acquiring an image of a core particle in a wafer and position information of the core particle; determining a category corresponding to the core particle and a first abnormal score corresponding to the category according to the image of the core particle, wherein the category comprises normal core particles, abnormal core particles of at least one defect type and suspicious core particles; Determining position information of a target abnormal core particle in a search range taking the target suspicious core particle as a center, wherein the target suspicious core particle is any one of the suspicious core particles of at least one defect type; determining a second anomaly score for the target suspicious core particle according to a positional relationship between the positional information of the target suspicious core particle and a range corresponding to the defect trend distribution formed by the target abnormal core particle under the condition that the positional information of the target abnormal core particle can form a defect trend distribution corresponding to the defect type of the target suspicious core particle; and determining the target suspicious core particle as the normal core particle or the abnormal core particle according to the second abnormal score and the first abnormal score of the target suspicious core particle.
2. The method for detecting wafer anomalies according to claim 1, wherein said determining a second anomaly score for the target suspicious core particle based on a positional relationship of a range of positional information of the target suspicious core particle corresponding to the defect trend distribution of the target abnormal core particle comprises: Determining that the second anomaly score of the target suspicious core particle is a first value when the target suspicious core particle is located in a range corresponding to the defect trend distribution formed by the target abnormal core particle; and under the condition that the target suspicious core particle is positioned outside the range corresponding to the defect trend distribution formed by the target abnormal core particle, determining the second abnormal score of the target suspicious core particle as a second value, wherein the second value is smaller than the first value.
3. The method of claim 2, wherein the defect type comprises at least one of a scratch, a cut, an ITO anomaly, a smudge, a crack, wherein the defect trend distribution corresponding to the scratch, the crack, and the cut is a line segment, and the defect trend distribution corresponding to the ITO anomaly and the smudge is a sheet, the method further comprising: When the defect type is the scratch, the bias cutting or the collapse, and the distance between the target suspicious core particle and the line segment is smaller than or equal to a preset distance, determining that the target suspicious core particle is positioned in a range corresponding to the defect trend distribution formed by the target abnormal core particle; when the defect type is the scratch, the bias cutting or the collapse, and the distance between the target suspicious core particle and the line segment is larger than the preset distance, determining that the target suspicious core particle is positioned outside a range corresponding to the defect trend distribution formed by the target abnormal core particle; and under the condition that the defect type is the ITO abnormality or the dirt, determining that the target suspicious core particle is positioned in a range corresponding to the defect trend distribution formed by the target abnormal core particle.
4. The method of detecting wafer anomalies according to claim 1, further comprising: The mapping table is used for reflecting the mapping relation between the defect type and the defect trend distribution; and determining defect trend distribution corresponding to the defect type of the target suspicious core particle according to the mapping table.
5. The method of detecting wafer anomalies according to claim 4, further comprising: and determining the size of the search range according to the defect trend distribution corresponding to the defect type of the target suspicious core particle.
6. The method of detecting wafer anomalies according to claim 1, further comprising: determining that the target suspicious core particle is a third anomaly score if the location information of the target abnormal core particle cannot constitute the defect trend distribution corresponding to the defect type of the target suspicious core particle; And determining that the target suspicious core particle is the normal core particle or the abnormal core particle according to the third abnormal score and the first abnormal score of the target suspicious core particle.
7. The method of any one of claims 1-6, wherein determining the target suspicious core particle as the normal core particle or the abnormal core particle according to the second abnormality score and the first abnormality score of the target suspicious core particle comprises: determining a final anomaly score for the target suspicious core according to the first anomaly score and the second anomaly score for the target suspicious core; and determining that the target suspicious core particle is the normal core particle or the abnormal core particle according to the final abnormal score of the target suspicious core particle.
8. The method for detecting wafer anomalies according to any one of claims 1-6, wherein determining a category corresponding to the core particle and a first anomaly score corresponding to the category according to the image of the core particle comprises: Inputting the images of the core particles into a classification model to obtain the category corresponding to the core particles and the confidence coefficient corresponding to the category output by the classification model; Determining that the core particle is the suspicious core particle under the condition that the class corresponding to the core particle output by the classification model is the abnormal core particle and the confidence is in a first threshold interval; And under the condition that the class corresponding to the core particle output by the classification model is the normal core particle and the confidence is in a second threshold interval, determining the core particle as the suspicious core particle, and taking the class with the second highest confidence as the class of the core particle.
9. The method for detecting wafer anomalies according to any one of claims 1-6, wherein determining a category corresponding to the core particle and a first anomaly score corresponding to the category according to the image of the core particle comprises: Inputting the image of the core particle into a segmentation model to obtain defect areas of each defect type in the image of the core particle output by the segmentation model; and determining the core particle to be the abnormal core particle under the condition that the defect areas of all the defect types do not exceed a preset threshold value and the defect area of any one defect type is in a third threshold value interval.
10. An electronic device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that the processor implements the method according to any of claims 1 to 9 when executing the computer program.

Description

Method for detecting wafer abnormality and electronic equipment Technical Field The application belongs to the technical field of wafer detection, and particularly relates to a method for detecting wafer abnormality and electronic equipment. Background The semiconductor wafer core particle anomaly detection is an important link of quality control and yield improvement in semiconductor manufacturing, the industry commonly adopts automatic optical detection equipment (Automated Optical Inspection) to carry out primary channel detection, visual detection is an important technical path in the primary channel detection, and the existing mainstream visual detection scheme comprises the following two types: the traditional image processing method is characterized in that the defect clamping control is realized by frequency domain filtering, morphological operation, edge detection, template matching and the like according to image characteristics and rules defined by experts, and the traditional image processing method is intuitive in principle, small in calculated amount and quick in online, but is sensitive to illumination, rotation, noise and the like according to artificial characteristic engineering and poor in robustness. The deep learning method comprises the main flow scheme including classification, target detection (such as YOLO and SSD), semantic segmentation (such as U-Net and deep Lab) and other networks, wherein the end-to-end learning does not need manual design of features, the detection precision and the robustness are superior to those of the traditional method, but a large amount of marked data are needed, the calculation cost is high, the interpretation is poor, and the problem of missed detection or over detection is easy to occur in critical defect and difficult defect detection. Disclosure of Invention The embodiment of the application provides a method for detecting wafer abnormality and electronic equipment, which can solve the problem that in the prior art, detection omission or overdetection is easy to occur in critical defect and difficult defect detection. In a first aspect, an embodiment of the present application provides a method for detecting wafer anomalies, including: acquiring an image of a core particle in a wafer and position information of the core particle; determining a category corresponding to the core particle and a first abnormal score corresponding to the category according to the image of the core particle, wherein the category comprises normal core particles, abnormal core particles of at least one defect type and suspicious core particles; Determining position information of a target abnormal core particle in a search range taking the target suspicious core particle as a center, wherein the target suspicious core particle is any one of the suspicious core particles of at least one defect type; determining a second anomaly score for the target suspicious core particle according to a positional relationship between the positional information of the target suspicious core particle and a range corresponding to the defect trend distribution formed by the target abnormal core particle under the condition that the positional information of the target abnormal core particle can form a defect trend distribution corresponding to the defect type of the target suspicious core particle; and determining the target suspicious core particle as the normal core particle or the abnormal core particle according to the second abnormal score and the first abnormal score of the target suspicious core particle. In a possible implementation manner of the first aspect, the determining the second anomaly score of the target suspicious core particle according to the positional relationship between the positional information of the target suspicious core particle and the range corresponding to the defect trend distribution formed by the target abnormal core particle includes: Determining that the second anomaly score of the target suspicious core particle is a first value when the target suspicious core particle is located in a range corresponding to the defect trend distribution formed by the target abnormal core particle; and under the condition that the target suspicious core particle is positioned outside the range corresponding to the defect trend distribution formed by the target abnormal core particle, determining the second abnormal score of the target suspicious core particle as a second value, wherein the second value is smaller than the first value. In one possible implementation manner of the first aspect, the defect type includes at least one of scratch, bias cut, ITO abnormality, dirt, and chipping, wherein the defect trend distribution corresponding to the scratch, the chipping, and the bias cut is a line segment, and the defect trend distribution corresponding to the ITO abnormality and the dirt is a sheet, and the method further includes: When the defect type is the scratch, the