Wafer defect image rapid screening method

Abstract

The application discloses a wafer defect image fast screening method, comprising the following steps: reading in an image, pre-processing the read-in image, segmenting into multiple sub-images, calculating the gray value standard deviation of the sub-image, obtaining a standard deviation sequence, constructing a single image statistic p according to the standard deviation sequence, constructing a statistic sequence composed of all image statistics, calculating an initial threshold value ζ by using the maximum inter-class variance method and realizing the initial screening of the image, threshold refinement and defect image screening. The wafer defect image fast screening method adopts a coarse-to-fine threshold calculation method, can achieve high screening efficiency, can greatly reduce the storage space required for saving images in the equipment, and can greatly reduce the operation amount in the subsequent analysis process, improve the analysis efficiency, and the calculation of the screening process is simple, the high fullness rate can be ensured, meanwhile, the method is not sensitive to the image channel, the method can be used for screening of multiple channel images, and the application range is wide.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of semiconductor wafer detection, in particular to a wafer defect image rapid screening method. BACKGROUND

[0002] With the development of semiconductor and material science, the requirements for material property characterization and defect detection are becoming more and more unique, showing new characteristics such as high resolution, high precision, specialization and multi-parameter correlation. The detection objects of semiconductor wafers include single crystal silicon and compound semiconductors such as silicon carbide, sapphire and gallium nitride substrates, epitaxial wafers and etched wafers, and the defect types cover surface defects and lattice defects.

[0003] High-resolution optical imaging inevitably brings a huge amount of image data. In particular, wafer defect detection requires full detection, i.e. all production wafers are detected, and wafer defect detection is to cover the whole wafer for detection, which brings great challenges to image data processing and data storage.

[0004] Defects are classified into systematic defects and random defects. In the real production process, high-purity semiconductor wafers have high single crystal degree and low defect density. Only part of the images contain defect targets, and the remaining images only contain background, which are defect-free information data sets. The current defect detection algorithm uses traditional digital image processing method or deep learning method to process each image, including images without defect information, which has a huge amount of calculation. Due to the needs of industrial process, multiple post-processing analysis can be performed after optical detection to exclude the missed detection rate of online analysis method, and all original image data need to be saved. For example, using 10 times optical magnification, a 6-inch wafer may need to occupy 5-10 GB of storage space. SUMMARY

[0005] The technical problem to be solved by the present application is that the present application proposes a wafer defect image rapid screening method to quickly and accurately eliminate images without defect targets, reduce the amount of calculation and analysis of image information, improve the screening efficiency, ensure a high recall rate, and effectively reduce the pixel size of the original image, thereby reducing the storage space required for saving the original image.

[0006] To solve the above technical problems, one technical solution adopted by the present application is:

[0007] A wafer defect image rapid screening method comprises the following steps:

[0008] S10, reading in a wafer scanning image to be screened;

[0009] S20, pre-processing the read-in image to reduce the influence of background noise;

[0010] S30, dividing the pre-processed image into a plurality of sub-images according to a preset two-dimensional scale;

[0011] S40, obtaining a gray value distribution of each sub-image, and calculating a gray value standard deviation of each sub-image respectively to obtain a standard deviation sequence composed of the gray value standard deviations of the sub-images;

[0012] S50, taking the gray value standard deviation of the sub-image as a sample value, calculating a mean μ and a standard deviation σ of the standard deviation sequence, and constructing a statistic p to represent a possibility of the image containing a defect, wherein p = (μ + 3σ) / σ max , σ max is a maximum value in the standard deviation sequence of the sub-image;

[0013] S60, repeating steps S10 to S50 to construct statistics corresponding to all wafer scanning images to be screened to obtain a statistic sequence composed of all the statistics;

[0014] S70, calculating an initial threshold ζ by using the maximum inter-class variance method, and comparing each statistic in the statistic sequence with the initial threshold ζ to divide all wafer scanning images to be screened into two categories, i.e., a category containing a defect target and a category not containing a defect target;

[0015] S80, taking each statistic corresponding to the category not containing a defect target distinguished in step S70 as a sample value of a refined statistic sequence, calculating a mean μ' and a standard deviation σ' of the refined statistic sequence, taking ζ' = μ' - 2σ' as a final threshold, comparing each statistic in the refined statistic sequence with the final threshold ζ' to divide the wafer scanning images corresponding to each statistic in the refined statistic sequence into two categories, i.e., a category containing a defect target and a category not containing a defect target.

[0016] Further, in step S20, the pre-processing manner of the background noise of the image is Gaussian filtering.

[0017] Further, in step S30, the image is divided by using a longitudinal and lateral bidirectional equal-interval division manner.

[0018] Further, in step S50, the smaller the value of the statistic p is, the greater the possibility of the image containing a defect target is, and vice versa.

[0019] Further, in step S70, the image corresponding to the statistic p smaller than the initial threshold ζ is regarded as containing a defect target.

[0020] Further, in step S80, the image corresponding to the statistic p smaller than the final threshold ζ' is regarded as containing a defect target.

[0021] The application further provides a wafer defect image rapid screening device, which applies the wafer defect image rapid screening method to realize the detection and screening of semiconductor wafers.

[0022] The application further provides an electronic device, which comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor realizes the wafer defect image rapid screening method when executing the program.

[0023] The application further provides a computer readable storage medium, which stores a computer program, and the computer program is executable on a processor to realize the wafer defect image rapid screening method.

[0024] Compared with the prior art, the application has the following advantages:

[0025] The application obtains a large number of sub-images by dividing the image, and constructs a statistical quantity for representing the possibility of containing defects in the image by using the gray standard deviation of the sub-images. An initial threshold is calculated by using the maximum inter-class variance method. The images containing defect targets can be preliminarily screened by comparing each statistical quantity in the statistical quantity sequence with the initial threshold. The mean value and the standard deviation of the statistical quantity sequence corresponding to the remaining images are calculated, and the final threshold is calculated based on the mean value and the standard deviation. The images corresponding to the statistical quantities less than the threshold can be regarded as containing defect targets, and the images are stored based on the result. Therefore, the threshold calculation method from coarse to fine achieves high screening efficiency, and the calculation of the screening process is simple, which can ensure high recall rate. Meanwhile, the method is not sensitive to the image channel, and can be used for screening the images of multiple channels, and has wide adaptability. BRIEF DESCRIPTION OF DRAWINGS

[0026] Figure 1 FIG. 1 is a flowchart of the defect image rapid screening method of the application;

[0027] Figure 2 FIG. 3 is a schematic diagram of the image division method of the application;

[0028] Figure 3 FIG. 5 is a sample diagram of the sub-images obtained by the image division of the application;

[0029] Figure 4 FIG. 7 is a gray distribution histogram corresponding to the sub-image in FIG. 6; Figure 3

[0030] Figure 5 FIG. 10 is a schematic diagram of the screening threshold calculation principle of the application. DETAILED DESCRIPTION

[0031] ​The preferred embodiments of the present invention are described in detail below with reference to the accompanying drawings so that the advantages and features of the present invention can be more easily understood by those skilled in the art, thereby making a clearer and more precise definition of the protection scope of the present invention.

[0032] The present invention provides a rapid wafer defect image screening method, applicable to existing semiconductor wafer inspection equipment. By inspecting wafer scan images, the method screens images containing defective targets to ensure semiconductor wafer quality. Semiconductor wafers inspected include substrates, epitaxial wafers, and etched wafers based on single-crystal silicon and compound semiconductors, such as silicon carbide, sapphire, and gallium nitride. Defect types range from surface defects to lattice defects.

[0033] See also Figure 1 The wafer defect image rapid screening method includes the following steps:

[0034] S10. Read in the wafer scan image to be screened. The size of a single semiconductor wafer is relatively small. In actual production and testing, a rectangular array is usually arranged on a chip substrate to form a testing array to achieve batch transfer and testing. Therefore, the wafer scan image includes a larger chip substrate image and multiple array-distributed wafer images. The chip substrate image serves as the image background and is a data set without defect information. Only some wafer images may contain defect targets.

[0035] S20, pre-processing the read image to reduce the influence of background noise; since the background noise level in the wafer scanning image is high, denoising is required. This method uses Gaussian filtering operation to obtain an image with more continuous and smooth grayscale values.

[0036] S30, dividing the pre-processed image into multiple sub-images according to a preset two-dimensional scale. Figure 2 As shown in Figure 2, the image segmentation method used in this method is to segment the image with equal spacing in both vertical and horizontal directions. Assuming the image size is Mpixels×Npixels and the segmentation scale is Spixels, one image can produce Q sub-images, where Q = M×N / S 2 In this embodiment, after the above segmentation method is used, one of the sub-image samples obtained is as follows: Figure 3 shown.

[0037] S40 , obtaining the grayscale value distribution of each sub-image, and respectively calculating the grayscale value standard deviation of each sub-image, to obtain a standard deviation sequence consisting of the grayscale value standard deviation of each sub-image. Figure 3 The grayscale distribution histogram corresponding to the sub-image sample shown in is as follows Figure 4The standard deviation of the gray value of each sub-image can be calculated by the general standard deviation calculation formula, and then the standard deviation sequence with a length of Q can be obtained, that is, the length of the standard deviation sequence is equal to the number of sub-images.

[0038] S50, taking the gray value standard deviation of the sub-image as the sample value, calculating the mean μ and the standard deviation σ of the standard deviation sequence, and constructing the statistic p to represent the possibility of the image containing defects, wherein p = (μ + 3σ) / σ max , σ max is the maximum value in the sub-image standard deviation sequence; the smaller the value of the statistic p, the greater the possibility of the image containing defective targets, and vice versa. Therefore, the statistics corresponding to the two types of images form a bimodal distribution, as shown in Figure 5 .

[0039] S60, repeating steps S10 to S50 to construct the statistics corresponding to all the wafer scan images to be screened, and obtaining a statistic sequence {p1, p2, …, p Q} composed of all the statistics, wherein Q is the total number of wafer scan images to be screened;

[0040] S70, an initial threshold ζ is calculated by the maximum inter-class variance method, and each statistic p i (i = 1, 2, …, Q) in the statistic sequence is compared with the initial threshold ζ to divide all the wafer scan images to be screened into two categories: containing defective targets and not containing defective targets; the statistic p i less than the initial threshold ζ is considered to correspond to an image containing defective targets, and vice versa.

[0041] S80, taking each statistic corresponding to the image not containing defective targets distinguished in step S70 as the sample value of the refined statistic sequence, calculating the mean μ' and the standard deviation σ' of the refined statistic sequence, taking ζ' = μ' - 2σ' as the final threshold (conservative threshold), and comparing each statistic in the refined statistic sequence with the final threshold ζ' to divide the wafer scan images corresponding to each statistic in the refined statistic sequence into two categories: containing defective targets and not containing defective targets. Among them, the image corresponding to the statistic p less than the final threshold ζ' is considered to contain defective targets, and the corresponding original picture is saved at the original scale and together with the previously saved part of the image containing defective targets for subsequent further detection and analysis. The remaining images not containing defective targets do not need to be analyzed and calculated subsequently, but only need to be simply compressed, merged or discarded. In general, images not containing defective targets account for the vast majority of the total number of images, so they can save a lot of computing resources for subsequent analysis and calculation, and can save a lot of storage space for the device.

[0042] The present application is based on a digital image processing method and basic statistical principles, and realizes a wafer defect image fast screening method. A related program part is added in a detection program of an existing semiconductor wafer detection device to realize the image processing and calculation screening method.

[0043] Therefore, the embodiment of the present application further provides a specific implementation of an electronic device capable of realizing all steps in the method in the above embodiment, which specifically includes the following contents: a processor, a memory, a communication interface and a bus. The processor, the memory and the communication interface complete mutual communication through the bus. The processor is used to call a computer program in the memory, and the processor realizes all steps in the method in the above embodiment when executing the computer program. The electronic device can be integrated in an existing semiconductor wafer detection device,

[0044] The embodiment of the present application further provides a computer readable storage medium capable of realizing all steps in the method in the above embodiment, and the computer readable storage medium stores a computer program. The computer program is executed by a processor to realize all steps in the method in the above embodiment.

[0045] The above description is only the embodiment of the present application, and does not limit the patent scope of the present application. Any equivalent structure or equivalent flow transformation obtained by using the content of the specification and the drawings, or direct or indirect application in other related technical fields is also included in the patent protection scope of the present application.

Claims

1. A method for rapid screening of wafer defect images, characterized in that: The following steps are involved: S10, reading in the scanned image of the wafer to be screened; S20, pre-processing the read image to reduce the influence of background noise; S30, dividing the pre-processed image into multiple sub-images according to a preset two-dimensional scale; S40, obtaining the grayscale value distribution of each sub-image, and calculating the grayscale value standard deviation of each sub-image respectively, to obtain a standard deviation sequence consisting of the grayscale value standard deviation of each sub-image; S50. Using the grayscale value standard deviation of the sub-image as the sample value, calculate the mean μ and standard deviation σ of the standard deviation sequence, and construct a statistic p to represent the possibility that the image contains defects, where p = (μ + 3σ) / σ max , σ max is the maximum value in the sub-image standard deviation sequence; S60, repeating steps S10 to S50, constructing statistics corresponding to all wafer scan images to be screened, and obtaining a statistical sequence consisting of all statistics; S70, using the maximum inter-class variance method to calculate an initial threshold ζ, and by comparing each statistic in the statistic sequence with the initial threshold ζ, classify all wafer scan images to be screened into two categories: those containing defective targets and those not containing defective targets; S80. Use the statistics corresponding to the images that do not contain defective targets identified in step S70 as sample values ​​of the refined statistical sequence, calculate the mean μ′ and standard deviation σ′ of the refined statistical sequence, use ζ′=μ′-2σ′ as the final threshold, and compare the size of each statistic in the refined statistical sequence with the final threshold ζ′ to divide the wafer scanning images to be screened corresponding to each statistic in the refined statistical sequence into two categories: containing defective targets and not containing defective targets.

2. The method for rapid wafer defect image screening according to claim 1, wherein: In step S20 , the background noise of the image is pre-processed by Gaussian filtering.

3. The method for rapid wafer defect image screening according to claim 1, wherein: In step S30 , the image segmentation is performed in a manner of equal spacing in both vertical and horizontal directions.

4. The method for rapid wafer defect image screening according to any one of claims 1 to 3, wherein: In step S50 , the smaller the value of the statistic p, the greater the possibility that the image contains a defective object, and vice versa.

5. The method for rapid wafer defect image screening according to claim 4, wherein: In step S70 , the image corresponding to the statistic p that is smaller than the initial threshold ζ is considered to contain a defective object.

6. The method for rapid wafer defect image screening according to claim 5, wherein: In step S80 , the image corresponding to the statistic p that is smaller than the final threshold ζ′ is considered to contain a defective object.

7. A wafer defect image rapid screening device, characterized by: The device uses the wafer defect image rapid screening method described in any one of claims 1 to 6 to realize the detection and screening of semiconductor wafers.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the method for rapid screening of wafer defect images according to any one of claims 1 to 6 is implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for rapid screening of wafer defect images according to any one of claims 1 to 6 is implemented.

Images