Search

CN-121994703-A - Wafer defect detection system based on hyperspectrum

CN121994703ACN 121994703 ACN121994703 ACN 121994703ACN-121994703-A

Abstract

The invention relates to the technical field of semiconductor detection and discloses a wafer defect detection system based on hyperspectrum, which comprises a light source control module, a hyperspectral imaging module, a data preprocessing module and a defect identification module. The system ensures uniform and stable illumination through precise light source distance calibration and optimization, acquires complete spectrum information of the surface of a wafer through hyperspectral imaging, performs pixel level anomaly detection and spectrum anomaly pixel identification through a spectrum angle matching method after preprocessing of denoising, correction and registration, aggregates the anomaly pixels into a defect area through spatial clustering, extracts spectrum and morphological characteristics of the anomaly pixels, and finally accurately classifies and judges defects through a preset rule base. The invention realizes high-precision, automatic and non-contact detection of the wafer defects, and remarkably improves the detection efficiency and reliability.

Inventors

  • YANG MEIHUI
  • XU HAITONG
  • CHEN YIXIN

Assignees

  • 无锡彩鸿芯宇科技有限公司

Dates

Publication Date
20260508
Application Date
20260203

Claims (8)

  1. 1. The wafer defect detection system based on hyperspectrum is characterized by comprising a light source control module, a hyperspectral imaging module, a data preprocessing module and a defect identification module; The light source control module is used for providing stable and uniform irradiation light for the hyperspectral imaging module, and the spectrum range of the light source module is matched with the detection spectrum range of the hyperspectral imaging module; The hyperspectral imaging module is used for collecting hyperspectral image data of the surface of the wafer and transmitting the hyperspectral image data to the data preprocessing module; The data preprocessing module is connected with the hyperspectral imaging module and is used for carrying out noise reduction, spectrum correction and spatial registration processing on hyperspectral image data and outputting preprocessed hyperspectral data; The defect identification module is connected with the data preprocessing module and is used for extracting characteristics of the preprocessed hyperspectral data and judging whether the wafer has defects or not according to the characteristics.
  2. 2. The hyperspectral based wafer defect detection system of claim 1, wherein the light source control module calibrates the distance between the light source and the wafer before the detection starts, so as to ensure the uniformity of the light source, and the working process of calibrating the distance between the light source and the wafer comprises: Step one, a light source is placed at a distance, the power of the light source is adjusted to enable the brightness of an image of a camera to be moderate under reasonable exposure, and the power is locked; Step two, slowly moving the light source to be close to the camera, simultaneously observing the real-time image, and finding out two critical points of the light source which are too close to each other: A critical point, when the image is continuously close to the light source, the average gray value of the image edge area is obviously lower than that of the central area, and the relative difference exceeds the position of a preset threshold value (such as 5 percent); The critical point B is a position when the gray value of any pixel in the central area of the image reaches the full range of the camera for the first time when the gray value is continuously close to the light source; recording the approximate positions of the light sources A and B, wherein the optimal working distance is required to be A, B; Step three, a series of distance points are arranged at equal intervals in the interval between A, B 、 、...、 ; Step four, at each distance point I is n, the light source is moved to the position, and the same hyperspectral imaging parameters as the formal detection are used for acquiring a piece of complete white board hyperspectral cube data; Step five, for each distance point And processing and analyzing the acquired imaging parameters, and determining the optimal distance according to the analysis result.
  3. 3. The hyperspectral based wafer defect inspection system of claim 2, wherein the operation of step five comprises: At a distance point Defining a rectangular region ROI in the middle of the acquired gray level image, wherein N pixels are shared in the rectangular region ROI, and calculating the average gray level value of the image , wherein, The gray value of the Kth pixel, K belongs to N, and an average gray value change curve along with the distance is constructed; firstly, calculating standard deviation of gray values of all pixels in a rectangular region ROI Calculating the relative standard deviation of gray values of all pixels in the rectangular region ROI Constructing a curve of the relative standard deviation of the gray value of the rectangular region ROI along with the distance; drawing a straight line representing an upper limit on a curve of the relative standard deviation of gray values of the rectangular region ROI along with the distance , Setting according to the detection process requirements; finding out that the relative standard deviation of the gray value of the rectangular region ROI is lower than or equal to the distance change curve Corresponding distance interval , wherein, , All belong to n; checking interval on average gray value change curve with distance Whether the gray value of all points in the system is larger than or equal to the lowest gray value set by the system If yes, the whole interval is a feasible region, if no, the distance points with the intensity not reaching the standard are removed, and the feasible region is reduced; In the feasible region, the point with the highest average gray value is selected as the optimal light source position.
  4. 4. The hyperspectral based wafer defect inspection system of claim 2, wherein the operation of shrinking the feasible region comprises: If it is ≥ Then the whole interval Is a feasible region; If it is < Then from Finding the first one satisfying the average gray value ∈r is found to the left, i.e. to the direction of decreasing distance Is recorded as a new right boundary The feasible region is updated as ; If it is The point does not meet the average gray value not less than Very rarely, it is stated that the system luminous flux is severely insufficient, and the system configuration needs to be re-evaluated.
  5. 5. The hyperspectral based wafer defect inspection system of claim 1, wherein the data preprocessing module comprises: carrying out noise reduction treatment on hyperspectral image data by adopting a self-adaptive median filtering algorithm, and removing salt and pepper noise and Gaussian noise in the image; Performing spectrum correction on the hyperspectral image after noise reduction based on the spectrum data of the standard reference plate, and eliminating the influence of light source intensity fluctuation and imaging system errors on the spectrum data; And carrying out spatial registration on hyperspectral images of different wavebands by adopting an image registration algorithm, so as to ensure the accurate correspondence of spectral data of the same spatial position.
  6. 6. The hyperspectral based wafer defect inspection system of claim 4, wherein the defect recognition module operates as follows: Using a well known wafer, collecting its hyperspectral image data, extracting its spectral reflectance data for each pixel in the image, constructing the spectral vector for that pixel ; Scanning a wafer to be detected, and constructing a spectral vector to be evaluated by using spectral reflectivity data of each pixel in a hyperspectral image of the wafer to be detected ; If it is Less than or equal to the set confidence threshold, vector Sum vector The directions are very close, and the spectra are highly similar, indicating that the pixel is normal; If it is Greater than the set confidence threshold, vector Sum vector The difference is large, indicating that the pixel is abnormal; And sequentially judging whether all the pixels are abnormal, performing spatial cluster analysis on all the pixels judged to be spectral abnormal, aggregating the spatially adjacent abnormal pixels into an abnormal region, extracting spectral features and morphological features of the abnormal region, comparing the spectral features and morphological features with a defect feature rule base set by a system, and judging what type of defects exist in the wafer.
  7. 7. The hyperspectral based wafer defect detection system of claim 6, wherein the calculated vector Sum vector Included angle between The process of (1) comprises: By the formula Calculating vectors Sum vector The cosine of the included angle between them, wherein, Is vector quantity Sum vector Is used for the dot product of (a), And Respectively is vector Sum vector Is a mold of (2); Calculating Then, the included angle can be obtained through an inverse cosine function
  8. 8. The hyperspectral based wafer defect detection system of any one of claims 1 to 7, wherein the operation of the system comprises: S1, calibrating uniformity of a light source, selecting an optimal working distance, and fixing the position of the light source; s2, establishing a reference spectrum model, acquiring hyperspectral data of a wafer with known good properties, extracting spectrums of all pixels in the region, calculating average reflectivity of each wave band, and constructing a normal spectrum reference vector; s3, scanning and data acquisition of a wafer to be detected, namely placing the wafer to be detected on a motion platform, and starting a hyperspectral camera to acquire hyperspectral data of the whole wafer surface under the calibrated optimal illumination condition; S4, preprocessing spectrum data; S5, detecting abnormal spectrum at pixel level, extracting spectral reflectance vector of each pixel in the corrected data, and calculating included angle between the corrected data and normal vector Based on included angle Marking spectrum anomaly pixels; And S6, aggregating the abnormal pixels adjacent to each other into independent abnormal areas, extracting spectral features and morphological features of the abnormal areas, and comparing the spectral features and the morphological features with a defect feature rule base set by a system to judge what type of defects exist on the wafer.

Description

Wafer defect detection system based on hyperspectrum Technical Field The invention relates to the technical field of semiconductor detection, in particular to a wafer defect detection system based on hyperspectrum. Background Wafers are core substrates for semiconductor fabrication, and surface defects (e.g., scratches, particles, etc.) directly affect the performance and reliability of semiconductor devices, so defect detection is a critical element in wafer fabrication. The existing wafer defect detection system has the following problems that 1, the influence of a light source on a result in the detection process is not considered, tiny fluctuation and nonuniformity of the light source can directly cause image gray level change to generate a large number of false defects or missed detection, 2, only the space information of the wafer surface can be acquired, the defect type cannot be distinguished by utilizing spectrum information, and misjudgment and missed judgment are easy to occur. Therefore, the invention provides a wafer defect detection system based on hyperspectrum. Disclosure of Invention The invention aims to provide a wafer defect detection system based on hyperspectrum, which solves the technical problems. A wafer defect detection system based on hyperspectrum comprises a light source control module, a hyperspectral imaging module, a data preprocessing module and a defect identification module; The light source control module is used for providing stable and uniform irradiation light for the hyperspectral imaging module, and the spectrum range of the light source module is matched with the detection spectrum range of the hyperspectral imaging module; The hyperspectral imaging module is used for collecting hyperspectral image data of the surface of the wafer and transmitting the hyperspectral image data to the data preprocessing module; The data preprocessing module is connected with the hyperspectral imaging module and is used for carrying out noise reduction, spectrum correction and spatial registration processing on hyperspectral image data and outputting preprocessed hyperspectral data; The defect identification module is connected with the data preprocessing module and is used for extracting characteristics of the preprocessed hyperspectral data and judging whether the wafer has defects or not according to the characteristics. As a further description of the technical scheme of the present invention, before the detection starts, the light source control module performs calibration on the distance between the light source and the wafer, so as to ensure uniformity of the light source, and the working process of performing calibration on the distance between the light source and the wafer includes: Step one, a light source is placed at a distance, the power of the light source is adjusted to enable the brightness of an image of a camera to be moderate under reasonable exposure, and the power is locked; Step two, slowly moving the light source to be close to the camera, simultaneously observing the real-time image, and finding out two critical points of the light source which are too close to each other: A critical point, when the image is continuously close to the light source, the average gray value of the image edge area is obviously lower than that of the central area, and the relative difference exceeds the position of a preset threshold value (such as 5 percent); The critical point B is a position when the gray value of any pixel in the central area of the image reaches the full range of the camera for the first time when the gray value is continuously close to the light source; recording the approximate positions of the light sources A and B, wherein the optimal working distance is required to be A, B; Step three, a series of distance points are arranged at equal intervals in the interval between A, B 、、...、; Step four, at each distance pointI is n, the light source is moved to the position, and the same hyperspectral imaging parameters as the formal detection are used for acquiring a piece of complete white board hyperspectral cube data; Step five, for each distance point And processing and analyzing the acquired imaging parameters, and determining the optimal distance according to the analysis result. As a further description of the technical solution of the present invention, the working process of the fifth step includes: At a distance point Defining a rectangular region ROI in the middle of the acquired gray level image, wherein N pixels are shared in the rectangular region ROI, and calculating the average gray level value of the image, wherein,The gray value of the Kth pixel, K belongs to N, and an average gray value change curve along with the distance is constructed; firstly, calculating standard deviation of gray values of all pixels in a rectangular region ROI Calculating the relative standard deviation of gray values of all pixels in the rectangular region ROIConstructing a