Wafer surface defect recognition method, storage medium and electronic device based on defocused image

Abstract

The present invention discloses a method for identifying wafer surface defects based on defocused images, comprising the following steps: first, reading the original image of the defocused wafer surface defects; second, using discrete Fourier transform to obtain a frequency domain image; third, using a Gaussian low-pass filter to obtain an unevenly illuminated background to obtain a wafer illumination background frequency domain image; fourth, using inverse Fourier transform to obtain a background image in the spatial domain; fifth, subtracting the background image from the original image to obtain a wafer image; sixth, selecting median filtering for denoising; seventh, using a watershed algorithm to perform image segmentation and perform region merging; eighth, performing regional analysis and employing a coarse screening plus fine screening method to identify defective areas. The present invention uses an adaptive background estimation method to remove the image illumination background for wafer images under different illumination scenes, thereby facilitating subsequent processing; employs a watershed algorithm for segmentation to reduce similar interference areas; and employs a joint screening method to increase recognition effect.

Description

Technical Field

[0001] The present invention relates to a wafer surface defect recognition method, in particular to a wafer surface defect recognition method based on defocused images. Background Art

[0002] The semiconductor chip industry is extremely important to information security and the national economy. Wafers are the carriers of chips, and their quality determines chip production. Therefore, wafer defect detection is a crucial step in semiconductor chip production.

[0003] Currently, wafer defect detection and inspection mostly rely on machine vision methods. For example, a two-dimensional Fourier transform is used to analyze two-dimensional images containing key parameters to identify wafer defects. Alternatively, a multi-layer threshold segmentation algorithm is used to automatically segment wafer defect images, which can accurately and effectively segment defects. Alternatively, a triple median filter algorithm is used to reduce wafer image noise, and normalized template matching is used to detect wafer defects, which is robust to lighting conditions. All of the above are wafer defect detection and inspection methods. Conventional wafer defect detection and inspection methods first focus the wafer and then perform defect detection on the acquired, in-focus image.

[0004] Shortcomings of traditional wafer identification and testing processes:

[0005] During the process of automatically focusing the wafer, the entire wafer area will be focused, resulting in low detection efficiency.

[0006] Regarding the shortcomings of existing wafer surface defect recognition and detection methods:

[0007] 1. During wafer surface defect detection, wafer lighting conditions are critical. Good lighting conditions are required to capture wafer images for defect detection. Defect detection on wafer images under poor lighting conditions is ineffective. Existing wafer defect detection methods can only detect defects on wafer images under fixed lighting conditions. However, in actual production, the lighting environment varies, requiring defect recognition methods to adapt to changing lighting conditions.

[0008] 2. Existing watershed image segmentation algorithms respond well to image noise and subtle grayscale variations on the wafer surface. However, this can lead to over-segmentation, making it impossible to correctly segment the wafer image. Defective areas on the wafer surface are over-segmented into several fragments, preventing them from being transformed into complete defect targets, making subsequent defect extraction difficult.

[0009] The wafer surface defect recognition method based on out-of-focus images can detect wafer defects in advance even when the wafer is out of focus. This allows for pre-detection of wafer defects. After the defect location is detected, autofocus is used to obtain a clear image for subsequent processing. For images where no defect location is detected on the wafer, autofocus is not used and that area is skipped, improving detection efficiency.

[0010] Therefore, there is an urgent need for a wafer surface defect recognition method based on defocused images, which can first pre-process the original image of the wafer surface defects and then perform defect recognition on the processed image to improve the detection efficiency. Summary of the Invention

[0011] In response to the deficiencies in the prior art, the present invention provides a wafer surface defect recognition method based on defocused images to solve the problem of low detection efficiency caused by not first identifying whether the wafer contains defects during wafer surface defect detection.

[0012] To achieve the above object, the present invention adopts the following technical solutions:

[0013] A method for identifying wafer surface defects based on defocused images, comprising the following steps:

[0014] Step 1: read the original image f(x,y) of the defocused wafer surface defect, whose pixel size is M×N, where M is the width of the original image and N is the length of the original image;

[0015] Step 2: Use discrete Fourier transform on the original image to obtain the frequency domain image F(u,v);

[0016] Step 3: Use a Gaussian low-pass filter H(u,v) to obtain the uneven illumination background in the frequency domain image F(u,v); retain the low-frequency information in the frequency domain image F(u,v) to obtain the wafer illumination background frequency domain image G(u,v);

[0017] Step 4: Use inverse Fourier transform to obtain the wafer illumination background image g(x,y) in the spatial domain of the wafer illumination background frequency domain image G(u,v);

[0018] Step 5: In the airspace, subtract the wafer illumination background image g(x,y) from the original image f(x,y) to obtain a background-removed wafer image h(x,y); the formula is: h(x,y) = f(x,y) - g(x,y);

[0019] Step 6: Use median filtering as the image denoising method to denoise the wafer image;

[0020] Step 7: Use the watershed algorithm to segment the denoised wafer image to obtain a pre-segmentation result of the image, and then perform region merging on the result;

[0021] Step eight, performing regional analysis on the segmented and merged wafer images, and using a combined screening method of coarse screening and fine screening to identify defective areas.

[0022] To optimize the above technical solutions, specific measures taken also include:

[0023] Furthermore, the calculation formula of the discrete Fourier transform in step 2 is:

[0024]

[0025] Among them, j is an imaginary unit; e is a constant in mathematics.

[0026] Furthermore, the calculation formula of the Gaussian low-pass filter H(u,v) in step 3 is:

[0027]

[0028] Where D(u,v) is the distance from point (u,v) to the center of the frequency domain; D0 is the cutoff frequency.

[0029] Furthermore, D0 is obtained by the following steps:

[0030] Step 1, calculate the total energy P of the frequency domain image F(u,v) T ;

[0031]

[0032] Among them, P(u,v) is the energy modulus of the frequency domain image. The calculation formula of P(u,v) is as follows, where R(u,v) is the real part of F(u,v) and I(u,v) is the imaginary part of F(u,v);

[0033] P(u,v)=|F(u,v)| 2 =R 2 (u,v)+I 2 (u,v)

[0034] Step 2: For the frequency domain image F(u,v), construct filters with different cutoff frequencies D0=1, 2, 3, ... 30; and calculate the cutoff frequency retention energy P of each filter. n , the calculation formula is as follows;

[0035]

[0036] Step 3: construct P for the frequency domain image F(u,v)n Lookup table with relationship to D0;

[0037] Step 4, cutoff frequency retains energy P n The total energy P of the frequency domain image F(u,v) T The energy proportion is β, and the relationship is as follows:

[0038] β=P n / P T

[0039] Among them, in the same frequency domain image F(u,v), there are 30 values ​​of D0, that is, β also corresponds to 30, and the solved β value corresponds to the D0 value;

[0040] Step 5: Select 100 wafer images f(x,y) under different illumination conditions as random samples. For each of these images, use the methods in Steps 2 to 4 to obtain the energy fraction β corresponding to different cutoff frequencies D0. Obtain 100 sets of β values ​​corresponding to D0 = 1, 2, 3, ... 30. Use this data to construct a scatter plot of different cutoff frequencies D0 and corresponding energy fraction β. Use the least squares method to perform curve fitting, and obtain the relationship between cutoff frequency D0 and energy fraction β as follows:

[0041] β=-6.748*D0^-1.116+99.11

[0042] Step 6: Calculate the second-order derivative of the relationship between the cutoff frequency D0 and the energy proportion β to obtain the inflection point. Analyzing the second-order derivative graph, it can be seen that the curve is always less than zero, and when the cutoff frequency D0 = 10, the curve begins to approach zero infinitely. Therefore, D0 = 10 is taken as the inflection point, and the corresponding β is 98.6%;

[0043] Step 7, for wafer images f(x,y) under different lighting conditions, the total energy P T The cutoff frequency D0 of the required filter is different. According to the inflection point value β=98.6% obtained in Step 6, the cutoff frequency D0 of the required filter is also different. n / P T The cutoff frequency retention energy P of the wafer image f(x,y) under different lighting conditions is obtained by reverse calculation n ; Then according to Step 3 P n The relationship lookup table with D0 is used to construct the P of wafer image f(x,y) under different lighting conditions. n The relationship between P and D0 can be found in the table. n To solve the corresponding cutoff frequency D0.

[0044] Furthermore, the step three includes the following steps:

[0045] A Gaussian low-pass filter H(u,v) is constructed using a determined cutoff frequency D0=10, and the obtained Gaussian low-pass filter H(u,v) is used to obtain the background of uneven illumination in the frequency domain image F(u,v), retaining the low-frequency information in the frequency domain image F(u,v); the frequency domain image F(u,v) is multiplied by the Gaussian low-pass filter H(u,v) to obtain the wafer illumination background frequency domain image G(u,v).

[0046] Furthermore, the calculation formula of the inverse Fourier transform in step 4 is:

[0047]

[0048] Furthermore, the process of region merging in step seven is as follows:

[0049] Step 1: Select the point with the minimum gray value on the watershed line as the candidate point P. i , the adjacent areas on both sides of the watershed line are divided into R i and R j , the minimum gray value in the adjacent areas on both sides is and Calculate P i and and The maximum value of the difference is W max , W max The calculation formula is as follows,

[0050]

[0051] Step 2: When W is satisfied max <T, where T is the region merging threshold; the region R i With R j Carry out a merger;

[0052] Step 3, enter the next candidate point P i+1 , repeat Step 1 to 2 above until all regions are traversed and the region merging is completed;

[0053] Step 4: For wafer images f(x,y) under different lighting conditions, the merging criteria used are consistent;

[0054] Step 5: Set the adaptive threshold T x =(T+W max ) / 2 is the termination condition of regional merging; where T x is the adaptive threshold for different lighting conditions, and T is the region merging threshold in Step 2.

[0055] Furthermore, the step eight includes the following steps:

[0056] Step 1: Calculate the grayscale mean value of each area in the segmented and merged wafer image, and select the grayscale mean value of a defect area as a coarse screening condition; after the coarse screening, obtain areas with a grayscale mean value similar to the defect area; wherein, the grayscale mean value is calculated as follows;

[0057]

[0058] Where R is the region of the wafer image after segmentation and merging, P is a pixel in region R, and g(p) is the grayscale value of the region;

[0059] Step 2: For the roughly screened areas, the maximum inter-class variance method is used to perform threshold segmentation on each individual area to achieve local adaptive threshold segmentation of each area, and each area is finely screened to extract defects in each area;

[0060] Step 3: Using the area threshold method, the maximum area of ​​the defective region after the image is finally identified is set as the judgment standard to remove the interference caused by the wafer background morphology to defect detection.

[0061] Furthermore, a computer-readable storage medium stores a computer program, characterized in that: the computer program enables a computer to execute the above-mentioned wafer surface defect identification method based on defocused images.

[0062] Furthermore, an electronic device is provided, characterized in that it comprises: a memory, a processor, and a computer program stored in the memory and runnable on the processor, wherein when the processor executes the computer program, it implements a wafer surface defect recognition method based on defocused images as described above.

[0063] The beneficial effects of the present invention are:

[0064] The present invention uses an adaptive background estimation method to remove the image illumination background for wafer images under different lighting scenes, providing a more convenient basis for the subsequent recognition and detection of image defects; uses an optimized watershed algorithm to segment the wafer defect area, reduces similar interference areas, and has strong anti-interference ability; uses a joint screening method for regional screening, which can further ensure the reliability of the recognition effect. The present invention identifies wafer surface defects through defocused images. When this method is used in a wafer detection system, it can identify wafer defects when the system is defocused, automatically focus on the defective position, and then use conventional detection methods; for positions where defects are not identified, no automatic focus is required and no subsequent detection operations are required. This method is executed before wafer focus detection, and is a pre-processing method before wafer focusing, to solve the problem of low efficiency due to the need to automatically focus the entire wafer during wafer detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 This is a flow chart of a wafer surface defect recognition method based on defocused images proposed by the present invention;

[0066] Figure 2 A schematic diagram of an image background removal process of a wafer surface defect recognition method based on defocused images proposed by the present invention;

[0067] Figure 3 This is a comparison diagram of the image segmentation effects of a wafer surface defect recognition method based on defocused images proposed by the present invention;

[0068] Figure 4 This is a recognition effect diagram of the defect area of ​​a wafer surface defect recognition method based on defocused images proposed by the present invention;

[0069] Figure 5 A schematic diagram of a defect area image recognition process of a wafer surface defect recognition method based on defocused images proposed in the present invention;

[0070] Figure 6 This is a schematic diagram of the defect-free area image recognition process of a wafer surface defect recognition method based on defocused images proposed by the present invention. DETAILED DESCRIPTION

[0071] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0072] As attached Figure 1 As shown, a wafer surface defect recognition method based on defocused images according to an embodiment of the present invention is characterized by comprising the following steps:

[0073] Step 1: read the original image f(x,y) of the defocused wafer surface defect, whose pixel size is M×N, where M is the width of the original image and N is the length of the original image;

[0074] Step 2: Use discrete Fourier transform on the original image to obtain the frequency domain image F(u,v);

[0075] Step 3: Analyze the frequency domain image F(u,v). The energy of the spectrum image F(u,v) is mostly concentrated in the low-frequency part, which is also the spectrum area where background information exists. The uneven illumination background can be obtained by designing a spectrum filter to eliminate the influence of uneven illumination. Therefore, a Gaussian low-pass filter H(u,v) is selected to obtain the uneven illumination background in the frequency domain image F(u,v). The low-frequency information in the frequency domain image F(u,v) is retained to obtain the wafer illumination background frequency domain image G(u,v).

[0076] Step 4: Use inverse Fourier transform to obtain the wafer illumination background image g(x,y) in the spatial domain of the wafer illumination background frequency domain image G(u,v);

[0077] Step 5: In the airspace, subtract the wafer illumination background image g(x,y) from the original image f(x,y) to obtain a background-removed wafer image h(x,y); the formula is: h(x,y) = f(x,y) - g(x,y);

[0078] Step 6: Use median filtering as the image denoising method to denoise the wafer image, smooth the image background morphology, and avoid blurring of details;

[0079] Step 7: Use the watershed algorithm to segment the denoised wafer image to obtain a pre-segmentation result of the image, and then perform region merging on the result;

[0080] Step eight, performing regional analysis on the segmented and merged wafer images, and using a combined screening method of coarse screening and fine screening to identify defective areas.

[0081] In this embodiment, the calculation formula of the discrete Fourier transform in step 2 is:

[0082]

[0083] Wherein, j is an imaginary unit (ie, j^2=-1); e is a constant in mathematics.

[0084] In this embodiment, the calculation formula of the Gaussian low-pass filter H(u,v) in step 3 is:

[0085]

[0086] Where D(u,v) is the distance between point (u,v) and the center of the frequency domain; D0 is the cutoff frequency, which is the distance between the cutoff frequency point (u0,v0) and the frequency center on the circular radius of the Gaussian low-pass filter.

[0087] In this embodiment, D0 is obtained by the following steps:

[0088] Step 1, calculate the total energy P of the frequency domain image F(u,v) T ;

[0089]

[0090] Among them, P(u,v) is the energy modulus of the frequency domain image. The calculation formula of P(u,v) is as follows, where R(u,v) is the real part of F(u,v) and I(u,v) is the imaginary part of F(u,v);

[0091] P(u,v)=|F(u,v)| 2 =R 2 (u,v)+I 2 (u,v)

[0092] Step 2: For the frequency domain image F(u,v), construct filters with different cutoff frequencies D0=1, 2, 3, ... 30; and calculate the cutoff frequency retention energy P of each filter. n (n=1,2,3,…30, the value of n corresponds to the value of D0), the calculation formula is as follows;

[0093]

[0094] Step 3: construct P for the frequency domain image F(u,v) n Look up the table with the relationship between D0 to reflect P n The corresponding relationship with D0; used in Step 7 to solve the cutoff frequency D0 of the wafer image f(x,y) under different lighting conditions, which can be obtained by P n To inversely calculate the D0 value;

[0095]

[0096]

[0097] Step 4, cutoff frequency retains energy P n The total energy P of the frequency domain image F(u,v) T The energy proportion is β, and the relationship is as follows:

[0098] β=P n / P T

[0099] Among them, in the same frequency domain image F(u,v), there are 30 values ​​of D0, that is, β also corresponds to 30, and the solved β value corresponds to the D0 value;

[0100] Step 5. Select 100 wafer images f(x, y) under different illumination conditions as random samples. For each of these images, use the methods in Steps 2 to 4 to obtain the energy fraction β corresponding to different cutoff frequencies D0. Obtain 100 sets of corresponding β values ​​for D0 = 1, 2, 3, ... 30, where each set has 30 β values, for a total of 100 sets. Use this data to construct a scatter plot of different cutoff frequencies D0 and corresponding energy fraction β. Use the least squares method to perform curve fitting, and obtain the relationship between cutoff frequency D0 and energy fraction β as follows:

[0101] β=-6.748*D0^-1.116+99.11

[0102] Step 6: Calculate the second-order derivative of the relationship between the cutoff frequency D0 and the energy fraction β to obtain the inflection point. Analyzing the second-order derivative image, it can be seen that its curve is always less than zero, and when the cutoff frequency D0 = 10, the curve begins to approach zero infinitely. Therefore, D0 = 10 is taken as the inflection point, and the corresponding β is 98.6%. β = 98.6% is a fixed value and can be used for the same wafer sample image under different lighting conditions.

[0103] Step 7, for wafer images f(x,y) under different lighting conditions, the total energy P T The cutoff frequency D0 of the required filter is different. According to the inflection point value β=98.6% obtained in Step 6, the cutoff frequency D0 of the required filter is also different. n / P T The cutoff frequency retention energy P of the wafer image f(x,y) under different lighting conditions is obtained by reverse calculation n ; Then according to Step 3 P n The relationship lookup table with D0 is used to construct the P of wafer image f(x,y) under different lighting conditions. n The relationship between D0 and the table can be found through P n To solve the corresponding cutoff frequency D0; to achieve the adaptation of the cutoff frequency D0 under different lighting conditions.

[0104] Among them, Steps 1 to 5 are to obtain the relationship between the cutoff frequency D0 and the energy proportion β under moderate lighting conditions; D0 = 10 obtained in Step 6 is used to return to the calculation formula of the Gaussian low-pass filter H(u,v) to solve the Gaussian low-pass filter H(u,v); Step 7 is to realize the self-adaptation of the filter cutoff frequency D0 for the wafer image f(x,y) under different lighting conditions.

[0105] In this embodiment, step three includes the following steps:

[0106] A Gaussian low-pass filter H(u,v) is constructed using the determined cutoff frequency D0=10, and the obtained Gaussian low-pass filter H(u,v) is used to obtain the uneven illumination background in the frequency domain image F(u,v), retaining the low-frequency information in the frequency domain image F(u,v). The frequency domain image F(u,v) is multiplied by the Gaussian low-pass filter H(u,v) to obtain the wafer illumination background frequency domain image G(u,v). The calculation formula is as follows:

[0107] G(u,v)=F(u,v)×H(u,v).

[0108] In this embodiment, the calculation formula for the inverse Fourier transform in step 4 is:

[0109]

[0110] In this embodiment, the key to region merging lies in determining the merging criteria and the merging termination condition. The grayscale difference between the watershed line and its neighborhood is selected as the similarity criterion for merging regions. The termination condition for region merging is set by setting a threshold. Therefore, the process of region merging in step 7 is as follows:

[0111] Step 1: Under moderate light conditions, select the point with the minimum gray value on the watershed line as the candidate point P. i , the adjacent areas on both sides of the watershed line are divided into R i and R j , the minimum gray value in the adjacent areas on both sides is and Calculate candidate point P i The minimum grayscale value in the area adjacent to both sides and The maximum value of the difference is W max , W max The calculation formula is as follows,

[0112]

[0113] Step 2: When W is satisfied max <T, where T is the region merging threshold, and T is a fixed value, which is selected based on the sample; the region R i With R j Merge, that is, update region R j The size of the area;

[0114] Step 3, enter the next candidate point P i+1 , repeat Step 1 to 2 above until all regions are traversed and the region merging is completed;

[0115] Step 4: For wafer images f(x,y) under different lighting conditions, the merging criteria used are consistent and the candidate points P are calculated. i The minimum grayscale value in the area adjacent to both sides and The maximum value of the difference is W max ;

[0116] Step 5: The maximum value W of the difference between the wafer image f(x,y) under different lighting conditions max will change, resulting in non-unique termination conditions for region merging, which need to change with the lighting conditions; set the adaptive threshold T x =(T+W max ) / 2 is the termination condition of regional merging; where T x is the adaptive threshold for different lighting conditions, which is obtained based on the fixed threshold T of the sample under moderate lighting conditions. x The appropriate merging threshold can be adaptively obtained according to the different illumination conditions of the sample wafer; T is the regional merging threshold in Step 2.

[0117] Among them, Steps 1 to 3 are the image segmentation process of the wafer image h(x, y) with background removed, and Steps 4 to 5 are the adaptive merging of the termination condition threshold for wafer images f(x, y) under different lighting conditions.

[0118] In this embodiment, as shown in the attached Figure 1 As shown in the defect identification in, the step eight includes the following steps:

[0119] Step 1: Calculate the grayscale mean value of each area in the segmented and merged wafer image, and select the grayscale mean value of a defect area as a coarse screening condition; after the coarse screening, obtain areas with a grayscale mean value similar to the defect area; wherein, the grayscale mean value is calculated as follows;

[0120]

[0121] Where R is the region of the wafer image after segmentation and merging, P is a pixel in region R, and g(p) is the grayscale value of the region;

[0122] Step 2: For the areas after rough screening, the maximum inter-class variance method is used to perform threshold segmentation on each individual area to achieve local adaptive threshold segmentation of each area, and each area is finely screened to extract the defects in each area; Step 3: Due to the detection interference caused by the wafer background morphology, the regional defects extracted in step 2 may be detected as defects due to the wafer background morphology; to address this problem, the area threshold method is used, and the maximum area of ​​the defect area after the final recognition of the image is set as the judgment standard to remove the interference caused by the wafer background morphology to defect detection, so as to ensure that defects can be effectively identified.

[0123] In another embodiment, the present invention provides a computer-readable storage medium storing a computer program, wherein the computer program enables a computer to execute the above-mentioned method for identifying wafer surface defects based on defocused images.

[0124] In another embodiment, the present invention proposes an electronic device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements a wafer surface defect recognition method based on defocused images as described above.

[0125] The process of out-of-focus wafer image defect detection is divided into image preprocessing and defect recognition. The image preprocessing process is mainly to eliminate the unevenness of the image background illumination. The image background removal process is as follows: Figure 2 As shown in Figure 2. Among them, the original image of the wafer is as follows: Figure 2 As shown in (a), there is an uneven brightness phenomenon in the image, and the defect features are not obvious, which is not conducive to defect detection in the defocus state; an adaptive cutoff frequency is used to construct a suitable filter to obtain the wafer illumination background image, such as Figure 2 (b); then the background image is subtracted from the original image to obtain the wafer background removed image, as shown in Figure 2 (c) shows that the background illumination of this image is uniform and the defect features are obvious, which is convenient for subsequent image processing.

[0126] The defect recognition process of wafer images is divided into image segmentation and defect recognition. In image segmentation, Figure 3 (a) The traditional watershed algorithm will produce over-segmentation, resulting in too many interference areas and unable to accurately extract defects; the optimized watershed algorithm solves the image over-segmentation problem to a certain extent. At the same time, the region merging method can reduce the interference areas with consistent similarity, reducing the amount of calculation for subsequent region screening. The image segmentation effect is compared. Figure 3 As shown. The defect recognition process is to identify the defect area of ​​the segmented image area, as shown Figure 4(a) shows that the coarse screening method can extract the area containing defects, but there are problems such as the non-defective area is not removed and the defect area cannot completely cover the defect. To solve this problem, the fine screening method is used. The maximum inter-class variance method is used to perform adaptive threshold segmentation of the area after coarse screening to extract the defects of each area. The defect recognition effect after fine screening is shown in Figure 2. Figure 4 (b) is shown. Figure 4 (a) and Figure 4 (b) From the defect recognition effect diagram, it can be seen that the combined screening method of coarse screening and fine screening can effectively extract defects.

[0127] In the specific experiment, the system primarily performed defect detection comparisons on defocused wafer images acquired under different lighting conditions. All experimental images were acquired on the same type of wafer. The experimental process included comparing background removal, segmentation, and defect detection results in defective and non-defective areas of the wafer.

[0128] For three typical scenarios, namely, low overall image brightness (Scene 1: image exposure time is 1800us), moderate overall image brightness (Scene 2: image exposure time is 3000us), and high overall image brightness (Scene 3: image exposure time is 4800us), the detection method of the present invention is used to perform image defect detection on the wafer defect area image and the wafer non-defect area image, and the detection results are shown in the attached figure. Figure 5 and attached Figure 6 The figure shows a schematic diagram of out-of-focus wafer image defect detection under different lighting conditions.

[0129] In the embodiments disclosed herein, computer storage media can be tangible media that can contain or store programs for use by or in conjunction with an instruction execution system, device, or apparatus. Computer storage media can include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or equipment, or any suitable combination of the foregoing. More specific examples of computer storage media can include electrical connections based on one or more lines, portable computer disks, hard disks, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memory), optical fibers, portable compact disk read-only memories (CD-ROMs), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0130] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed in this application can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0131] The above are merely preferred embodiments of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions based on the principles of the present invention are within the scope of protection of the present invention. It should be noted that for those skilled in the art, various improvements and modifications that do not depart from the principles of the present invention should be considered within the scope of protection of the present invention.

Claims

1. A wafer surface defect recognition method based on defocused images, characterized in that: The steps include: Step 1: read the original image f(x,y) of the defocused wafer surface defect, whose pixel size is M×N, where M is the width of the original image and N is the length of the original image; Step 2: Use discrete Fourier transform on the original image to obtain the frequency domain image F(u,v); Step 3: Use a Gaussian low-pass filter H(u,v) to obtain the uneven illumination background in the frequency domain image F(u,v); retain the low-frequency information in the frequency domain image F(u,v) to obtain the wafer illumination background frequency domain image G(u,v); Step 4: Use inverse Fourier transform to obtain the wafer illumination background image g(x,y) in the spatial domain of the wafer illumination background frequency domain image G(u,v); Step 5: In the airspace, subtract the wafer illumination background image g(x,y) from the original image f(x,y) to obtain a background-removed wafer image h(x,y); the formula is: h(x,y) = f(x,y) - g(x,y); Step 6: Use median filtering as the image denoising method to denoise the wafer image; Step 7: Use the watershed algorithm to segment the denoised wafer image to obtain a pre-segmentation result of the image, and then perform region merging on the result; Step eight, performing regional analysis on the segmented and merged wafer images, and using a combined screening method of coarse screening and fine screening to identify defective areas.

2. The wafer surface defect recognition method based on defocused images according to claim 1, wherein: The calculation formula of the discrete Fourier transform in step 2 is: Among them, j is an imaginary unit; e is a constant in mathematics.

3. The wafer surface defect recognition method based on defocused images according to claim 1, characterized in that: The calculation formula of the Gaussian low-pass filter H(u,v) in step 3 is: Where D(u,v) is the distance from point (u,v) to the center of the frequency domain; D0 is the cutoff frequency; and e is a mathematical constant.

4. The method for identifying wafer surface defects based on defocused images according to claim 3, wherein: The D0 is obtained by the following steps: Step 1, calculate the total energy P of the frequency domain image F(u,v) T ; Among them, P(u,v) is the energy modulus of the frequency domain image. The calculation formula of P(u,v) is as follows, where R(u,v) is the real part of F(u,v) and I(u,v) is the imaginary part of F(u,v); P(u,v)=|F(u,v)| 2 =R 2 (u,v)+I 2 (u,v) Step 2: For the frequency domain image F(u,v), construct filters with different cutoff frequencies D0=1, 2, 3, ... 30; and calculate the cutoff frequency retention energy P of each filter. n , the calculation formula is as follows; Step 3: construct P for the frequency domain image F(u,v) n Lookup table with relationship to D0; Step 4, cutoff frequency retains energy P n The total energy P of the frequency domain image F(u,v) T The energy proportion is β, and the relationship is as follows: β=P n / P T Among them, in the same frequency domain image F(u,v), there are 30 values ​​of D0, that is, β also corresponds to 30, and the solved β value corresponds to the D0 value; Step 5: Select 100 wafer images f(x,y) under different illumination conditions as random samples. For each of these images, use the methods in Steps 2 to 4 to obtain the energy fraction β corresponding to different cutoff frequencies D0. Obtain 100 sets of β values ​​corresponding to D0 = 1, 2, 3, ...

30. Use this data to construct a scatter plot of different cutoff frequencies D0 and corresponding energy fraction β. Use the least squares method to perform curve fitting, and obtain the relationship between cutoff frequency D0 and energy fraction β as follows: β＝-6.748*D0^-1.116+99.11 Step 6: Calculate the second-order derivative of the relationship between the cutoff frequency D0 and the energy proportion β to obtain the inflection point. Analyzing the second-order derivative graph, it can be seen that the curve is always less than zero, and when the cutoff frequency D0 = 10, the curve begins to approach zero infinitely. Therefore, D0 = 10 is taken as the inflection point, and the corresponding β is 98.6%; Step 7, for wafer images f(x,y) under different lighting conditions, the total energy P T The cutoff frequency D0 of the required filter is different. According to the inflection point value β=98.6% obtained in Step 6, the cutoff frequency D0 of the required filter is also different. n / P T The cutoff frequency retention energy P of the wafer image f(x,y) under different lighting conditions is obtained by reverse calculation n ; Then according to Step 3 P n The relationship lookup table with D0 is used to construct the P of wafer image f(x,y) under different lighting conditions. n The relationship between D0 and the table can be found through P n To solve the corresponding cutoff frequency D0.

5. The method for identifying wafer surface defects based on defocused images according to claim 4, wherein: The step three includes the following steps: A Gaussian low-pass filter H(u,v) is constructed using a determined cutoff frequency D0=10, and the obtained Gaussian low-pass filter H(u,v) is used to obtain the background of uneven illumination in the frequency domain image F(u,v), retaining the low-frequency information in the frequency domain image F(u,v); the frequency domain image F(u,v) is multiplied by the Gaussian low-pass filter H(u,v) to obtain the wafer illumination background frequency domain image G(u,v).

6. The wafer surface defect recognition method based on defocused images according to claim 1, characterized in that: The calculation formula of the inverse Fourier transform in step 4 is: Among them, j is an imaginary unit; e is a constant in mathematics.

7. The wafer surface defect recognition method based on defocused images according to claim 1, characterized in that: The process of region merging in step 7 is as follows: Step 1: Select the point with the minimum gray value on the watershed line as the candidate point P. i , the adjacent areas on both sides of the watershed line are divided into R i and R j , the minimum gray value in the adjacent areas on both sides is and Calculate P i and and The maximum value of the difference is W max , W max The calculation formula is as follows, Step 2: When W is satisfied max <T, where T is the region merging threshold; the region R i With R j to carry out a merger; Step 3, enter the next candidate point P i+1 , repeat Step 1 to 2 above until all regions are traversed and the region merging is completed; Step 4: For wafer images f(x,y) under different lighting conditions, the merging criteria used are consistent; Step 5: Set the adaptive threshold T x =(T+W max ) / 2 is the termination condition of regional merging; where T x is the adaptive threshold for different lighting conditions, and T is the region merging threshold in Step 2.

8. The wafer surface defect recognition method based on defocused images according to claim 1, characterized in that: The step eight comprises the following steps: Step 1: Calculate the grayscale mean value of each area in the segmented and merged wafer image, and select the grayscale mean value of a defect area as a coarse screening condition; after the coarse screening, obtain areas with a grayscale mean value similar to the defect area; wherein, the grayscale mean value is calculated as follows; Where R is the region of the segmented and merged wafer image, P is a pixel in region R, and g(p) is the grayscale value of the region. Step 2: For the coarsely screened regions, the maximum inter-class variance method is used to perform threshold segmentation on each individual region to achieve local adaptive threshold segmentation for each region. Each region is then finely screened to extract defects in each region. Step 3: Using the area threshold method, the maximum area of ​​the defective region after the image is finally recognized is set as the judgment standard to remove the interference caused by the wafer background morphology to defect detection.

9. A computer-readable storage medium storing a computer program, characterized in that: The computer program enables a computer to execute a wafer surface defect identification method based on defocused images as described in any one of claims 1 to 8.

10. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, a method for identifying wafer surface defects based on defocused images as described in any one of claims 1 to 8 is implemented.

Images