WO2021135045A1 - Wafer detection device, data processing method, and storage medium - Google Patents

Abstract

A wafer detection device, a data processing method, and a storage medium. The device comprises a detection target image generation unit and a data processing unit; the data processing unit comprises a preprocessing module and a noise reduction module; the preprocessing module comprises a lithography simulator (502) and a noise power spectrum extraction module; the lithography simulator (502) performs analog simulation calculation on wafer layout data (501) that serves as a design result of a detection object, so as to obtain a power spectrum Ps(u,v) (504); the noise power spectrum extraction module obtains a full signal power spectrum Ps+n(u,v) of a detection target image comprising a noise signal, and obtains a noise power spectrum Pn(u,v) (508) using the power spectrum Ps(u,v) according to the full signal power spectrum Ps+n(u,v); the noise reduction module determines the type of a noise reduction filter (513) and a characteristic function value (511) thereof according to a signal-noise ratio requirement, obtains a system parameter of the noise reduction filter (513) by using the characteristic function (511), the noise power spectrum Pn(u,v) (508), and the power spectrum Ps(u,v) (504) as input, and constitutes the noise reduction filter (513) by means of the system parameter to obtain a detection target image from which noise information is removed.

Description

Wafer inspection device, data processing method and storage medium

cross reference

This application claims the priority of the Chinese patent application with the application number 201911388824.X filed on December 30, 2019. The content of the above application is included here by reference.

Technical field

The present invention relates to the technical field of semiconductor manufacturing, and more specifically, to a wafer inspection device, a data processing method and a storage medium.

technical background

Wafer inspection devices are commonly used and necessary equipment in the semiconductor manufacturing process. The wafer inspection devices usually include a detection target image generation unit and a data processing unit. The detection target image generation unit is used to obtain information related to the inspected object (such as a wafer). The detection target image, the data processing unit is used to process and determine the defect extraction of the detection target image; the key core work of the data processing unit is to distinguish between effective defects (true defects) and suspected defects (noise signals), and suspected defects are detected Defects caused by small non-essential differences in the target image, among which suspected defects occur randomly during the process.

The detection target image generation unit may generally be an optical microscope or a detection device using a scanning electron microscope, or the like. For example, Bright Field Inspector (BFI), Dark Field Inspector (Dark Field Inspector), Scanning Electron Microscope (SEM) inspection devices, etc. Please refer to FIG. 1. FIG. 1 is a schematic structural diagram of a wafer inspection device commonly used in the prior art. As shown in Fig. 1, reference numeral 100 in Fig. 1 denotes a bright-field optical detection device. Brightfield microscope (brightfield microscope) is the most versatile optical microscope. Using light illumination, each point in the specimen is imaged in a bright background according to its light absorption and reflection.

As shown in FIG. 1, in order to optically detect defects of the wafer 106 placed on the stage 107, the light generated by the light source 102 illuminates the wafer 106 through the condenser lens 103, the beam splitter 104 and the objective lens 105. The fine structure of the large-scale integrated circuit (LSI) manufactured on the wafer 106 is obtained by the sensor 101 through the objective lens 105 and the beam splitter 104 to form a detection target image. The detection target image captured by the sensor 101 is read into the computer system 110 (data processing unit) as data. The computer system 110 processes the data read from the sensor 101 according to the execution program 112 recorded in the computer readable medium 111. There are various programs executed in the computer system 110. Among them, one of the most important functions is a defect extraction program, which extracts a part of the detection target image corresponding to the defect of the LSI.

Please refer to FIG. 2. FIG. 2 is an explanatory diagram of the principle of the defect extraction procedure in the prior art. As shown in FIG. 2, reference numeral 200 is an example diagram of the defect extraction program processing the inspection target image, the reference image and the difference extraction as the main output obtained in the wafer inspection apparatus 100. The processing steps are as follows: First, the test target image 202 (test) obtained from the wafer inspection device is compared with the reference image 201 (reference) to create a difference image 203 (difference); then, the difference image 203 is observed, if there is no defect , Then nothing appears on the difference image 203. If there is a defect shown in 204 on the detection target image, the difference image 203 appears as a difference defect 205, and the difference image 203 may have a larger size as shown by the difference defect 205. , It also shows the slight difference defect that does not affect the operation of the LSI as shown in 206. From the above-mentioned difference defects, for example, the difference defect 205 and the difference defect 206, it is necessary to select the difference image element actually related to the defect of the LSI.

That is, if there is no difference between the detection target image 202 and the reference image 201, the difference defect should become zero. However, in reality, the difference cannot be zero due to variations in the manufacturing process of the two images and fine dust. Therefore, the above-mentioned small difference signal that affects the image elements caused by the LSI manufacturing process and conditions is called a noise signal.

Please refer to FIG. 3, which shows a flowchart of a procedure for extracting defects based on the difference between the detection target image and the reference image. As shown in FIG. 3, reference number 300 is a schematic diagram of the process, which includes the following steps:

Step 301: The detection target image obtained by the sensor is temporarily stored in the image buffer;

Step 302: Obtain a reference image library;

Step 303: Receive the detection target image of the detected object;

Step 304: Determine the corresponding reference image, and retrieve the reference image corresponding to the detection target image of the detected object from the reference image library;

Step 305: align the image size and image position of the detection target image and the reference image;

Step 306: Compare and judge the difference between the detection target image and the reference image;

Step 307: Obtain effective difference defects according to the difference calculated in step 306;

Step 308: Determine whether the processing of all the detection target images is finished, if not, repeat the processing from step 303 to step 308 for the new detection target image.

The above method of calculating the difference between the detected target image and the reference image is based on morphology (Mathematical Morphology) image processing. Morphology is usually used to extract image components that are effective for expressing and describing the shape of the region from the image to enable subsequent recognition work It can grasp the most essential shape features of the target object, such as the processing of boundaries and connected areas. In the detection field, according to the processing purpose, the processing of the detection target image requires the steps of applying structural elements, that is, selecting the most suitable structural element, combining multiple morphological processing, and realizing the processing corresponding to the target.

Referring to FIG. 3, in the processing in step 307, it is necessary to discuss in detail how to divide the difference information into valid defects (true defects) and meaningless defects (false defects). In the prior art, the point where the intensity of the difference information exceeds a certain threshold is regarded as a bad point (effective defect).

Please refer to FIG. 4 again. The reference number 400 in FIG. 4 is a schematic diagram showing the difference between the detection target image and the reference image and the difference intensity of the passing defect on a certain straight line. In the figure shown by reference number 401, the intensity of the difference between the detection target image and the reference image is represented by the shades of the X-axis 405 and Y-axis 406 coordinate points. The dark spot indicated by reference numeral 402 is a part where the difference signal between the detected object and the reference image is large. The portion indicated by reference number 403 that is brighter than reference number 402 is a portion where the difference between the detected object and the reference image is small. Reference numeral 410 denotes a schematic diagram of the intensity of the difference on the straight line A-A' in the figure 401. Reference numeral 411 represents a curve of the difference signal intensity. Reference numeral 412 is an axis representing position coordinates on the straight line A-A'. Reference numeral 413 is an axis indicating the intensity of the difference. The reference number 414 is one of the threshold values of the difference intensity judged as a defect. The reference number 415 is also the threshold value of the difference intensity, which is lower than the reference number 414.

As shown in Fig. 4, the curve representing the intensity of the difference on the straight line A-A' in Fig. 401 is indicated by reference numeral 411. In the graph 410, it can be determined that the number of defective points is different by setting the threshold to IT1 (shown by reference numeral 414) or IT0 (shown by reference numeral 415). Specifically, if the threshold is set to IT1 (indicated by reference numeral 414), one defect is detected, and if the threshold is set to IT0 (indicated by reference numeral 415), then 3 defects are detected. In other words, whether the detected suspected defect is a valid defect or a meaningless defect depends on the threshold IT.

It should be noted that if the threshold is set low, then the possibility of effective defects becomes higher. Conversely, if the threshold is set high, then the possibility of ignoring effective defects becomes higher. Although the above technology proposes a method to find the optimal threshold, the improvement is limited.

In other words, if the threshold IT is not simply adjusted, there are many elements that should be adjusted, such as the combination of processing programs and the setting of various parameters, in image processing such as morphological reduction of meaningless defect processing. Therefore, the above-mentioned image processing method of reducing meaningless defect processing using morphology may become a means to improve the accuracy of judging valid defects and meaningless defects.

Summary of the invention

The object of the present invention is to provide a means for accurately estimating parameters such as the power spectrum of the noise signal required for the structure of the noise signal reduction filter. As a result, combined with filtering based on image processing such as morphological transformation and image feature extraction, it is possible to improve SNR (Signal to Noise Ratio) and reduce meaningless defect determination.

In order to achieve the above objective, the technical solution of the present invention is as follows:

A wafer inspection device includes a detection target image generation unit and a data processing unit. The detection target image generation unit is used to obtain a detection target image related to a detected object. The data processing unit is used to The image is processed and judged for defect extraction; the data processing unit includes a preprocessing module and a noise reduction module, the preprocessing module includes a lithography simulator and a noise signal extraction module, the lithography simulator is used for the design before inspection The resultant tested wafer layout data is simulated and calculated to obtain the power spectrum Ps(u,v); wherein the tested wafer layout data does not contain noise information; the noise signal extraction module is used to calculate The full signal power spectrum Ps+n(u,v) of the noise of the detection target image on which the noise signal is superimposed, and the full signal power spectrum Ps+n(u,v) is subtracted from the power spectrum Ps( u,v) to obtain the noise power spectrum Pn(u,v); the noise reduction module is used to determine the type of noise reduction filter and its characteristic function according to the signal-to-noise ratio requirements, and combine the characteristic function and the noise power spectrum Pn(u,v) and the power spectrum Ps(u,v) are used as input to obtain the system parameters of the noise reduction filter, and the noise reduction filter is formed by the system parameters; and the detection target image is passed through all After filtering by the noise reduction filter, the detection target image after the noise information is removed is obtained. Among them, u, v represent the spatial frequency.

Preferably, the data processing unit further includes a defect extraction module and a restoration module; after obtaining the detection target image, the defect extraction module extracts the detection target image in the detection target image based on morphological image processing technology Elements to obtain the noise image of the detection target image; the restoration module adds the detection target image elements in the detection target image to the detection target image after removing the noise information in the detection target image to form The detection target image after noise is removed.

Preferably, the type of the noise reduction filter is a two-dimensional Wiener filter.

In order to achieve the above objective, another technical solution of the present invention is as follows: A data processing method using the above wafer inspection device includes the following steps:

Step S11: Perform simulation calculation on the layout data of the tested wafer as the design result before testing to obtain the power spectrum Ps(u,v); wherein the layout data of the tested wafer does not contain noise information ；

Step S12: Calculate the full signal power spectrum Ps+n(u,v) of the detection target image superimposed with the noise signal, and subtract the power spectrum Ps from the full signal power spectrum Ps+n(u,v) (u,v), get the noise power spectrum Pn(u,v);

Step S13: Determine the noise reduction filter type and its characteristic function according to the signal-to-noise ratio requirement, and use the characteristic function, the noise power spectrum Pn(u,v) and the power spectrum Ps(u,v) as inputs, Obtain the system parameters of the noise reduction filter, and form the noise reduction filter through the system parameters; and filter the detection target image by the noise reduction filter to obtain the detection target image after the noise information is removed.

Preferably, the wafer inspection device further includes: step S10 and step S14:

Step S10: After the detection target image is obtained, based on morphological image processing technology, extract the detection target image elements in the detection target image to obtain a noise image of the detection target image;

Step S14: The restoration module adds the detection target image elements in the detection target image to the detection target image after removing the noise information in the detection target image to form the detection target image after removing the noise .

In order to achieve the above objective, another technical solution of the present invention is as follows:

A computer-readable medium that stores a computer-executable program, and the executable program is used to execute the data processing method of the wafer inspection apparatus; it includes the following programs:

Use the tested wafer layout data as the design result before testing to perform simulation simulation to obtain the power spectrum Ps(u,v); wherein, the tested wafer layout data does not contain noise information;

Obtain the full signal power spectrum Ps+n(u,v) of the detection target image superimposed with the noise signal, and subtract the power spectrum Ps(u) from the full signal power spectrum Ps+n(u,v) ,v), get the noise power spectrum Pn(u,v);

Determine the type of noise reduction filter and its characteristic function according to the signal-to-noise ratio requirements, and use the characteristic function, the noise power spectrum Pn(u,v) and the power spectrum Ps(u,v) as inputs to obtain noise reduction The system parameters of the filter are used to form a noise reduction filter through the system parameters; and after the detection target image is filtered by the noise reduction filter, the detection target image after the noise information is removed is obtained.

In order to achieve the above objective, another technical solution of the present invention is as follows:

A wafer inspection device includes a detection target image generation unit and a data processing unit. The detection target image generation unit is used to detect and generate a detection target image of a detected object, and the data processing unit is used to compare the detection target image. Perform processing and judgment of defect extraction; wherein, the data processing unit includes a preprocessing module and a noise reduction module, and the preprocessing module includes:

The noise region determination sub-module divides the detection target image formed by the detection target region into M subregions, and selects N subregions smaller than M that contain noise and are used for the estimation of the noise power spectrum to participate in the noise reduction process ；

The estimation processing sub-module is used to obtain the full signal power spectrum P ^N s+n(u,v) of the detection target image of the N sub-regions,

The power spectrum estimation stator module obtains the full signal power spectrum Ps+n(u,v) of the detection target image, and uses the full signal power spectrum P ^N s+n(u,v) of the N sub-regions as the noise power Spectrum Pn(u,v), subtract the noise power spectrum Pn(u,v) from the full signal power spectrum Ps+n(u,v) composed of the M sub-regions to obtain the power spectrum Ps(u,v) );

The noise reduction module determines the type of noise reduction filter and its characteristic function according to the signal-to-noise ratio requirement, and combines the characteristic function, the noise power spectrum Pn(u,v) and the power spectrum Ps(u,v) As input, the system parameters of the noise reduction filter are obtained, and the noise reduction filter is formed by the system parameters; and the detection target image is filtered by the noise reduction filter to obtain the detection after the noise information is removed. Target image.

Preferably, the M sub-regions have the same shape, and are preferably rectangular, square, diamond or honeycomb. In addition, sometimes the M sub-regions are preferably seamlessly connected.

Preferably, for the determination of the sub-regions selected for the estimation processing of the N noise power spectra, the average value of the sum of the squared differences of each of the M sub-regions is used, and when the average value is less than a predetermined value, It is determined as N sub-regions composed of noise signal components.

In order to achieve the above objective, another technical solution of the present invention is as follows:

A data processing method using the above-mentioned wafer inspection device includes the following steps:

Step S20: dividing the detection target image formed by the detection target area into M sub-areas, and selecting N sub-areas smaller than M containing noise to participate in the estimation processing of the noise power spectrum Pn(u,v);

Step S21: Obtain the full signal power spectrum P ^N s+n(u,v) of the detection target image of N signals containing noise and large difference defects;

Step S22: Obtain the full signal power spectrum Ps+n(u,v) of the detection target image, and use the full signal power spectrum P ^N s+n(u,v) of the N sub-regions as the noise power spectrum Pn( u,v), subtracting the noise power spectrum Pn(u,v) from the full signal power spectrum Ps+n(u,v) formed by the M sub-regions to obtain the power spectrum Ps(u,v);

Step S23: Determine the type of noise reduction filter and its characteristic function according to the signal-to-noise ratio requirement, and use the characteristic function, the noise power spectrum Pn(u,v) and the power spectrum Ps(u,v) as inputs, Obtain the system parameters of the noise reduction filter, and form the noise reduction filter through the system parameters; and filter the detection target image by the noise reduction filter to obtain the detection target image after the noise information is removed. In order to achieve the above objective, another technical solution of the present invention is as follows:

A computer-readable medium that stores a computer-executable program, and the executable program is used to execute the data processing method of the wafer inspection apparatus; it includes the following programs:

Dividing the detection target image formed by the detection target area into M sub-areas, and selecting a noise reduction area determination program of N sub-areas smaller than M and containing noise for the estimation of the noise power spectrum;

^{A procedure for obtaining the full signal power spectrum P N} s+n(u,v) of the detection target image of the N sub-regions containing noise and large difference signals;

Obtain the full signal power spectrum Ps+n(u,v) of the detection target image, and use the full signal power spectrum P ^N s+n(u,v) of the N sub-regions as the noise power spectrum Pn(u,v) ), a procedure for subtracting the noise power spectrum Pn(u,v) from the full signal power spectrum Ps+n(u,v) formed by the M sub-regions to obtain the power spectrum Ps(u,v);

Determine the type of noise reduction filter and its characteristic function according to the signal-to-noise ratio requirements, and use the characteristic function, the noise power spectrum Pn(u,v) and the power spectrum Ps(u,v) as inputs to obtain noise reduction The system parameters of the filter are used to form a noise reduction filter through the system parameters; and after the detection target image is filtered by the noise reduction filter, the detection target image after the noise information is removed is obtained.

In order to achieve the above objective, another technical solution of the present invention is as follows:

A wafer inspection device includes a detection target image generation unit and a data processing unit. The detection target image generation unit is used to obtain a detection target image from a plurality of detection target regions or across a plurality of wafer regions, the data processing The unit is used to perform defect extraction processing and judgment on the detection target image; it is characterized in that the data processing unit includes a preprocessing module and a noise reduction module, and the preprocessing module includes:

The noise reduction area determination sub-module, the detection target image i obtained from the plurality of detection target regions or the plurality of detection target regions across the wafer (i=1, 2... is an index representing each detection target image) ) Is divided into Mi sub-regions, and Ni sub-regions containing noise smaller than Mi are selected as regions for noise estimation to participate in the estimation processing of the noise power spectrum Pn(u,v);

The power spectrum inference stator module estimates the full signal power spectrum P ⁱ s+n(u,v) according to the detection target image;

The noise power spectrum deduces the stator module to obtain the full signal power spectrum P ^Ni s+n(u,v) of ^{the Ni regions, making it the noise power spectrum P i} n(u,v);

The difference signal power spectrum estimation stator module, according to the noise power spectrum P ⁱ n(u,v) and the full signal power spectrum P ⁱ s+n(u,v) to estimate the difference signal power spectrum P ⁱ s(u ,v);

The noise signal extraction sub-module, for the noise power spectrum P ⁱ n(u,v), estimates the average value of the range of i of the multiple detection target regions to obtain the _{average value of} Pn (u,v);

The difference signal extraction sub-module, for the difference signal power spectrum P ⁱ s(u,v), estimates the average of the range of i with respect to multiple detection target regions to obtain the _{average value of} Ps (u,v); and

The noise reduction module is used to determine the type of noise reduction filter and its characteristic function according to the signal-to-noise ratio requirements, and combine the characteristic function, the _{average value of} the noise power spectrum Pn (u, v) and the _{average value of} the power spectrum Ps ( u, v) as input, obtain the system parameters of the noise reduction filter, and form the noise reduction filter through the system parameters; and filter the detection target image by the noise reduction filter to obtain the noise-removed information The detection target image.

In order to achieve the above objective, another technical solution of the present invention is as follows:

A data processing method using the above-mentioned wafer inspection device includes the following steps:

Step S30: Divide the detection target image i obtained from the plurality of detection target regions or the plurality of detection target regions across the wafer (i=1, 2... are indexes representing the respective detection target images) into Mi sub-regions, and Ni sub-regions smaller than Mi are selected to participate in the estimation of the noise power spectrum;

Step S31: Estimate the full signal power spectrum P ⁱ s+n(u,v) according to the detection target image;

Step S32: Obtain the full signal power spectrum P ^Ni s+n(u,v) of ^{the Ni sub-regions, and make it a noise power spectrum P i} n(u,v);

Step S33: According to the noise power spectrum P ⁱ n(u,v) of ^{the selected Ni regions and the full signal power spectrum P i} s+n(u,v), the difference signal power spectrum P ⁱ s( u, v), of P ⁱ s (u, v) to obtain an average value for the range of i to obtain _{the average Zhi} Ps (u, v), the noise power spectrum P ⁱ n (u, v) determined for i The average value of the range of, get the _{average value of} Pn (u,v);

Step S34: used to determine the type of noise reduction filter and its characteristic function according to the signal-to-noise ratio requirement, and combine the characteristic function, the _{average value of} the noise power spectrum Pn (u, v) and the _{average value of} the power spectrum Ps (u , v) As input, obtain the system parameters of the noise reduction filter, and form the noise reduction filter through the system parameters; and filter the detection target image by the noise reduction filter to obtain the noise information removed The detection target image.

In order to achieve the above objective, another technical solution of the present invention is as follows:

A computer-readable medium that stores a computer-executable program, and the executable program is used to execute the data processing method of the wafer inspection apparatus; it includes the following programs:

Divide the inspection target image i (i=1, 2...is the index representing each inspection target image) formed by multiple inspection target regions or regions spanning multiple wafers into Mi sub-regions, and select a ratio Mi The small Ni sub-regions are used as regions for noise estimation to participate in ^{the estimation process of the noise power spectrum P i} n(u,v);

Infer the full signal power spectrum P ⁱ s+n(u,v) according to the detection target image i;

Obtain the full signal power spectrum P ^Ni s+n(u,v) from the Ni sub-regions, and make it the noise power spectrum P ⁱ n(u,v), according to the noise power spectrum P ⁱ n(u, v) and the full signal power spectrum P ⁱ s+n(u,v) of ^{the detected object image to estimate the difference signal power spectrum P i} s(u,v);

^{Subtract the corresponding noise power spectrum P i} n(u,v) from each of the full signal power spectrum P ⁱ s+n(u,v) to ^{obtain the difference signal power spectrum P i} s(u,v) for the noise power spectrum P ⁱ n (u, v) determined for the average of _{the average} Pn i _Zhi range (u, v), for the difference signal power spectrum P ⁱ s (u, v) is obtained _{For the average value} of the range of i Ps average value (u, v) program;

Determine the noise reduction filter type and its characteristic function according to the signal-to-noise ratio requirements, and use the characteristic function, the _{average value of} the noise power spectrum Pn (u, v) and the _{average value of} the power spectrum Ps (u, v) as input , Obtain the system parameters of the noise reduction filter, and form the noise reduction filter through the system parameters; and filter the detection target image by the noise reduction filter to obtain the detection target image after the noise information is removed .

It can be seen from the above technical solutions that in order to reduce the possibility of meaningless defects in the wafer inspection device, the present invention provides a noise signal reduction technology that converts the difference between the inspection target image and the reference image into defect evaluation data The ratio of effective defects to insignificant defects (noise signal) in, that is, the signal to noise ratio (SNR). Since the present invention estimates the correct noise power spectrum, various noise signal reduction filters can be used to further improve the SNR. For this reason, the parameters of the noise signal reduction filter, that is, the correct estimation of the power spectrum of the signal and the noise signal, etc. The change is particularly important.

Description of the drawings

Figure 1 shows a schematic diagram of a wafer inspection device in the prior art

2 is a schematic diagram showing defect extraction based on the detection target image and the reference image and the difference image of the wafer inspection device in the prior art

Figure 3 shows a flow chart of image data processing using a wafer inspection device in the prior art

FIG. 4 is a schematic diagram showing the difference signal between the detection target image and the reference image and the difference intensity of the defect on a straight line in the prior art

Fig. 5 is a flowchart showing the determination of the parameters of the noise signal reduction system in an embodiment of the present invention

Figure 6 is a schematic diagram of comparing the effects of the original image and the noise signal reduction filter through simulation in an embodiment of the present invention

FIG. 7 is an explanatory diagram showing the correspondence between the difference information and the position of the detected object in the embodiment of the present invention

Fig. 8 is a flowchart of a method for calculating a power spectrum of a noise signal in an embodiment of the present invention

Figure 9 shows a flowchart of image processing using morphology in an embodiment of the present invention

Figure 10 is a schematic diagram of the parameterized morphological structural elements in the embodiment of the present invention

Summary of the invention

The specific embodiments of the present invention will be further described in detail below in conjunction with the accompanying drawings. It is desirable to describe in order from the overall structure to the detailed structure.

In the embodiment of the present invention, the wafer inspection device includes an inspection target image generation unit and a data processing unit. The inspection target image generation unit may be an optical inspection device or an inspection device using a scanning electron microscope, which is not limited herein. When the data processing unit processes the obtained wafer inspection target image signal, it is the object of the present invention to reduce the meaningless defects contained in the inspection result, that is, to reduce the noise signal contained in the image information.

That is, the object of the present invention is to provide an optimal method for reducing the noise signal contained in the image information. It should be noted that in the embodiment of the present invention, the noise signal (suspected defect) in the image information can be reduced in multiple modes, and the present invention takes the difference between the detection target image and the reference image as an effective defect in the defect evaluation data. The method of considering and dealing with the ratio of meaningless defects (noise signals) can also be used in other fields.

In the embodiment of the present invention, the data processing unit is used to process and determine the defect extraction of the detection target image; the data processing unit may include a preprocessing module and a noise reduction module, and the preprocessing module includes a lithography simulator and noise signal extraction Module. The lithography simulator is used to simulate and calculate the layout data of the inspected wafer as the design result before the inspection to obtain the power spectrum Ps(u,v); wherein the layout data of the inspected wafer does not contain noise Information; The noise signal extraction module is used to obtain the full signal power spectrum Ps+n(u,v) of the detection target image overlapped with the noise signal, and subtract the full signal power spectrum Ps+n(u,v) Remove the power spectrum Ps(u,v) to obtain the noise power spectrum Pn(u,v); the noise reduction module is used to determine the type of noise reduction filter and its characteristic function according to the signal-to-noise ratio requirements, and combine the characteristics Function, the noise power spectrum Pn(u,v) and the power spectrum Ps(u,v) as input to obtain the system parameters of the noise reduction filter, and the noise reduction filter is formed by the system parameters; and After the detection target image is filtered by the noise reduction filter, the detection target image with reduced noise information is obtained.

That is, in the present invention, when the optical simulation is performed, the power spectrum of the noise signal can be set to zero and then the calculation can be performed. Therefore, only the power spectrum Ps of the signal can be calculated. Using the Ps obtained through simulation as the basis, the power spectrum of the noise signal can be calculated by calculating Pn(u,v)=Ps+n(u,v)−Ps(u,v).

Specifically, in order to reduce the possibility of suspected defects in the processing results of the wafer inspection device, it uses the difference between the inspection target image and the reference image as the ratio of the effective defects to the suspected defects in the defect evaluation data (SNR: signal to noise ratio). ) Becomes larger. It is clear to those skilled in the art that by using various noise reduction filters, the SNR can be improved. For this reason, the correct estimation of the parameters of the noise reduction filter, that is, the signal and noise power spectrum, etc. is particularly important.

Generally, there are many methods for correct estimation of signal and noise power spectra in the prior art. Take a two-dimensional Wiener Filer as an example for illustration. Wiener filtering is a method of filtering noise-mixed signals using the correlation and frequency characteristics of a stationary random process. Under certain constraints, the square of the difference between its output and a given function (usually called the expected output) Reaching the minimum, through mathematical operations, it can finally be transformed into a Toblitz equation solving problem. The Wiener filter is also called the least square filter or least square filter.

In the two-dimensional Wiener filter, its system characteristic M(u,v) is as follows:

Among them, H(u,v) is a function representing system characteristics, and H(u,v) is 1 when there is no influence other than noise factors. * Represents conjugate, Pn(u,v) represents the power spectrum of noise (Noise), Ps(u,v) represents the power spectrum of the signal, u,v represents the spatial frequency.

In the detection process of the actual system, Pn and Ps cannot be detected separately, but are observed in the form of Ps+n(u,v). Normally, Pn(u,v) can be estimated based on some assumptions. However, the results of the estimated Pn(u,v) often contain errors. In the embodiment of the present invention, Pn and Ps are separated for estimation, so a high-performance noise signal reduction system can be provided.

Example one

Hereinafter, an embodiment of the present invention will be described with reference to FIG. 5. It should be noted that the reference numeral 500 in FIG. 5 may be a system, a circuit, and a device. The working process will be described below in conjunction with the reference numerals in FIG. 5. If it is a manufacturing method, the steps of the method will be described in conjunction with the accompanying drawings. In the description, since it is difficult to separately describe the structure and operation, etc., it will be explained together with the structure.

Please refer to FIG. 5. FIG. 5 is a schematic diagram of a preferred embodiment for determining the parameters of the noise reduction system of the present invention. As shown in the figure, reference number 509 is a wafer, reference number 510 is a light source for illuminating the wafer, reference number 501 is layout design data, reference number 502 is a lithography simulator (usually a computer simulation program), reference number 503 is a power spectrum Ps (u,v) calculation module, 504 is the calculation result of the signal power spectrum Ps(u,v), 505 is the image sensor, 506 is the noise signal extraction sub-module, and 507 is the noise power spectrum The calculation module of Pn, the reference number 508 is the calculation result of the noise power spectrum Pn(u,v), the reference number 511 represents the characteristic function H(u,v) of the System, the reference number 512 represents the determination of the noise reduction system (Noise Reduction System), the reference number 513 represents a noise reduction filter (Noise Reduction Filter).

The working principle of the above noise reduction system is as follows:

On the one hand, using the wafer layout data 501 as the design result, the lithography simulator 502 performs simulation calculation on the layout data 501 without noise information. Based on the simulation calculation result, the power spectrum 503 can be calculated, and the data 504 of the power spectrum Ps(u,v) can be obtained. On the other hand, the detection target image corresponding to the layout data 501 of the wafer 509 is acquired by the sensor 505, and the detection target image obtained from the sensor and the difference image obtained from the reference image contain noise signals. The light source 510 is used to illuminate the wafer 509.

In addition, the detection target image and the simulation result of the lithography simulator 502 are used as input to complete the step of extracting noise signal information (reference number 506); then, through the calculation step of noise power spectrum Pn (reference number 507), the noise power spectrum can be obtained Pn(u,v), (denoted by reference numeral 508).

After the above data is obtained, the characteristic functions H(u,v), Pn(u,v) and Ps(u,v) of the system (System) shown by the reference number 511 are used as input, and the noise reduction system (Noise The parameters of the Reduction System are determined and further constitute a Noise Reduction Filter (No. 513).

As described above, in the detection target image signal obtained by the sensor from the wafer through the optical detection device, on the one hand, due to manufacturing, optical, electrical, and other factors, noise signals are superimposed on the detection target image obtained. On the other hand, the image information obtained through lithography simulation simulation does not contain noise signals, and the lithography simulation simulation is based on Layout Data, etc., which are the design information of the LSI. Therefore, the power spectrum Ps(u,v) of the detection target image can be obtained from the image of the optical simulation, and the power spectrum Pn(u,v) of the noise signal can be based on the detection target image obtained by the detection device and the image obtained by lithography simulation simulation. The difference is calculated. Therefore, using the above two power spectra, a high-precision noise signal reduction system can be realized.

In addition, the function H(u,v) that represents the system characteristics, when there is no influence other than the noise signal factor, the obtained output change is "1", but when there is an influence factor other than the noise signal factor, it means The function H(u,v) of the system characteristic is not 1.

As shown in Figure 5, the detection target image signal of the wafer illuminated by the light source 510 captured by the sensor 505 is superimposed with the following main defect signals: noise signals related to fine defects caused by particles on the wafer, based on optical The noise signal of system defects and the noise signal related to fine defects caused by the wafer manufacturing process. In most cases, the above-mentioned noise signals cannot be eliminated by data processing steps such as difference calculation in the detection device.

In the embodiment of the present invention, the sensor detection output signal Ps+n(u,v) is a mixture of the noise signal Pn(u,v) and the image signal Ps(u,v) of the detected object. In the detection target image signal Ps+n(u,v) obtained from the detection output result, the extraction of the signal related to the noise signal Pn(u,v) is the basis of the high-performance noise signal reduction filter shown in formula (1) . In addition, through the method shown in the flowchart shown in FIG. 5, a signal Ps(u,v) based only on image information can be obtained, and the image-based signal Ps without a noise signal can be obtained by lithography simulation. Thus, by subtracting Ps(u,v) from the sensor detection output result signal Ps+n(u,v), the noise signal Pn(u,v) can be obtained. After the noise signal Pn(u,v) is obtained, a high-performance noise signal reduction filter can be realized.

Please refer to FIG. 6, the number 600 in the figure is a schematic diagram of the effect of comparing the detection target image obtained by the sensor with the image information processed by the noise reduction filter. Numeral 601 shows a schematic diagram of the difference defect between the detected target image including the noise signal obtained by the wafer inspection device and the reference image, wherein the image shown by the numeral 601 has not been processed to reduce the noise signal. Numerals 602 and 603 indicate X-axis and Y-axis, respectively, and on the plane generated by the X-axis and Y-axis, the intensity of the difference signal is expressed in shades. Among them, the number 604 is a large difference caused by the existence of the defect, and a thin spot can be seen in the area of the number 605, which corresponds to a noise signal. The reference number 610 is the image after the noise is removed. Please refer to the figure on the right. The image numbered 610 is a simulation result of using a noise signal reduction filter. Numerals 612 and 613 indicate the X-axis and the Y-axis, and the intensity of the difference signal is expressed in shades on the plane generated by the X-axis and the Y-axis. The number 614 is the large difference caused by the existence of the defect, and the thinner spot can be seen in the area of the number 615, which corresponds to the situation after the noise signal reduction filter is used.

In order to better represent the status area of the noise signal, the number 605 when the noise signal reduction filter is not used is distinguished from the number 615 when the noise signal reduction filter is used, so that the difference in the status can be clearly seen. In the above embodiment, it is confirmed by actual simulation that in the case of 601 without the noise signal reduction filter, the SNR (Signal to Noise Ratio) is 5.61dB, compared to the case of 610 with the noise signal reduction filter Among them, the SNR is 15.0dB.

Example two

Hereinafter, an embodiment of the present invention will be described with reference to FIGS. 7 and 8.

In this embodiment, first, the detection target area is divided into a plurality of sub-areas, and it is judged whether the information indicating the difference intensity information contained in each sub-area is the information related to the peak of the difference or the information related to the noise signal. One method: Use only the sub-regions related to the noise signal to calculate the power spectrum of the noise signal, and use its power spectrum to form a noise signal reduction filter.

That is to say, in a preferred embodiment of the present invention, it is necessary to first determine the noise reduction area used for the estimation of the power spectrum of the noise signal, that is, to divide the detection target image formed by the detection target area into M seamless Connect the sub-regions, and select the N regions below M to participate in the estimation of the noise power spectrum.

The N regions participate in the selection of the noise power spectrum, and the average value of the sum of the squares of the difference intensities of each subregion can be taken. When the average value is less than a predetermined value, it is determined as a subregion composed of noise signal components.

Please refer to FIG. 7. The reference number 700 in FIG. 7 is a schematic diagram showing the difference information corresponding to the position of the detected object. When there is a defect in the detected target image, a peak based on the difference of the defect as shown by the number 701 appears, and the number 702 and the number 703 are peaks based on the difference of other defects, but the intensity of the latter two peaks is relatively small. In the graph shown by reference numeral 700, the peak value of the defect difference is plotted in a three-dimensional space, where the coordinate axis is composed of the X axis (reference numeral 706), the Y axis (reference numeral 707), and the intensity of the difference (reference numeral 708). Here, the X-Y plane is divided into virtual sub-areas (reference number 704). In this example, the number of virtual sub-regions (also called cells) is set to 6×6, which is composed of sub-regions related to the difference peak and sub-regions (reference number 705) mainly composed of noise signal components. The virtual sub-region does not have to be rectangular, and can be any shape. However, it is desirable that the virtual sub-regions cannot overlap each other, and they cover the entire detected object region without gaps.

It is necessary to classify the sub-regions related to the peak of the difference defect and the sub-regions mainly composed of noise signal components. There is a classification method: use statistical properties related to the strength of the differences in each subregion to make judgments. For example, taking the average value of the sum of the squares of the difference intensity of each sub-region, and when the average value is less than a predetermined value, it is determined as a sub-region composed of noise signal components. This determination method is only used as an example, and other determination methods are also possible, which are not limited here.

Specifically, in the embodiment of the present invention, please refer to FIG. 8. FIG. 8 is a flowchart of a method for calculating a power spectrum of a noise signal in this embodiment. As shown in Figure 8, the method for calculating the power spectrum of the noise signal includes the following steps:

In step 801, the detected object area is divided into a plurality of sub-areas by, for example, a virtual grid method. The shape of the sub-region does not need to be rectangular, and may be any shape.

In step 802, it is necessary to first determine whether the focused sub-region contains a defect signal. The judgment step can use the method based on statistical properties described above.

In step 803, it is necessary to first determine whether a difference peak (defect) is included. If it is included, skip step 804. If it is not included, in step 804, evaluate and improve the power spectrum Pn(u,v) of the noise signal.

In step 805, the next sub-region is extracted, and in step 806, it is determined whether the processing of all the sub-regions has ended. If it is not over, step 802 to step 806 are repeated.

In the embodiment shown in FIG. 8, the evaluation and improvement of the power spectrum related to the noise signal are sequentially implemented in step 804. However, in the specific embodiment of the present invention, the same processing flow can be used to perform After scanning the entire area, after completing the list of sub-areas related to the noise signal, calculate the noise power spectrum Pn(u,v).

Example three

Generally, in a wafer inspection device, it is impossible to finish its operation only by inspecting an area to be inspected. There are often many inspection target areas on a wafer, and it is sometimes necessary to continuously perform inspections on multiple wafers with multiple integrated circuits with the same layout. In this way, when processing multiple detected object areas, it is not necessary to only estimate the noise signal and/or power spectrum of one detected object area as in the first and second embodiments.

Specifically, the wafer inspection apparatus includes a detection target image generation unit and a data processing unit, the detection target image generation unit is used to obtain a plurality of detection target regions or detection target images across wafer regions, and the data processing unit It is used for processing and determining the defect extraction of the detection target image; its data processing unit includes a preprocessing module and a noise reduction module, and the preprocessing module includes a noise reduction area determination submodule and a noise signal extraction submodule. The noise region determining sub-module divides the plurality of detected object regions or the detection target image i formed across the wafer region (i=1, 2...is the index representing each detection target image) into Mi sub-regions, and Select the area composed of noise signal components, mainly Ni, for the estimation of noise power spectrum, where Ni is less than or equal to Mi; the power spectrum estimation stator module is used to estimate the full signal power spectrum P according to the detection target image i ⁱ s+n(u,v), the noise power spectrum estimation stator module obtains the noise power spectrum P ^Ni n(u,v) of the detection target image of the Ni parts, making it the noise power spectrum P ⁱ n (u,v); The difference signal power spectrum estimation stator module ^{estimates the difference signal power spectrum P i} s( ^{according to the noise power spectrum P i} n(u,v) and the full signal power spectrum P ⁱ s+n(u,v) u,v); the noise signal extraction submodule obtains the average value of the range of i from ^{the noise power spectrum P i} _{n(u,v) to obtain the average value of} Pn (u,v); and the difference signal extraction submodule , Obtaining the average value of the range of i from the difference signal power spectrum Ps(u,v) to obtain the _{average value of} Ps (u,v); the noise reduction module is used to determine the noise reduction according to the signal-to-noise ratio requirement type and features of the filter function, the characteristic function, the _average noise power spectrum Pn (u, v), and the _{average value of} the power spectrum Ps (u, v) as the input, to reduce noise filter system Parameters, and a noise reduction filter is formed by the system parameters; and the detection target image is filtered by the noise reduction filter to obtain the detection target image after the noise information is removed. Here, the operation to find the average value is not only a simple arithmetic average, but should be understood as an operation to find a general best estimated value.

In the embodiment of the present invention, the power spectrum improvement calculation of the noise signal is performed in the plurality of detection target regions or across the wafer, so that the accuracy can be improved.

Example four

Please refer to FIG. 9, which is a schematic diagram of an example of image processing using morphology. In the image processing process of FIG. 9 that uses morphological reduction of meaningless defects, etc., the distinction between the small element signals 906 to 909 that are considered to be noise signals and the detection target image element signals 902 to 905 to be extracted is a comparison. Easy, but distinguishing meaningful defects from the element signal requires complex processing.

Therefore, in the embodiment of the present invention, the wafer inspection apparatus of the present invention further includes a defect extraction module and a restoration module. That is, after the detection target image is obtained, the defect extraction module extracts the detection target image elements in the detection target image based on the morphological image processing technology, and removes the noise image of the detection target image. If the idea of the present invention is used to process only the noise image of the detection target image, the selection of the filter is relatively simple, and the accuracy of the filtering effect is improved. Moreover, after the noise image of the detection target image is obtained, the restoration module adds the detection target image elements in the detection target image to the detection target image after the noise information in the detection target image is removed, and it can be directly formed The detection target image after noise is removed.

Specifically, please refer to FIG. 9, which is a schematic diagram of an example of image processing using morphology. The reference numeral 900 in FIG. 9 details an example of filtering the noise signal composed of fine difference defects and extracting an image of a predetermined size or more. Numeral 901 is an input image, which includes element signals 906 to 909 of the fine difference defect of the noise signal and the detection target image element signals 902 to 905 to be extracted. Reference numeral 910 denotes image processing composed of a combination of morphological calculations, that is, morphological processing. The structural elements used in the morphological image processing with reference number 910 are shown by reference number 911. The output image 920 obtained by the morphological processing shown in this example is an image in which the fine image elements of the noise signal have been removed. The morphological processing used in this example can use the Erosion calculus method.

Specifically, the shape and size of the structural element 911 can be selected according to the characteristics of the image structural element of the fine difference defect of the noise signal to be removed. The pattern structure elements 922 to 925 obtained when the Erosion calculation is used are deformed according to the input relationship between the pattern structure elements 902 to 905. Since the deformation based on the Erosion calculation is different, in order to obtain the same pattern elements as those of 902 to 905, further pattern calculations are required. The filtering of the noise signal based on the morphological processing shown here is an example, and it is a technique that can be applied to a variety of complicated processing such as pattern structure extraction and shape extraction.

In image processing using morphology, it is possible to combine the selection of structural elements suitable for the purpose of processing and the morphological processing as a unit to realize a desired processing program. In order to improve the convenience for users, the processing program is described as a macro and stored in a library.

The structural elements used in morphological processing enable the description of the model to be parameterized, making it possible to generalize the description of macros. Please refer to Fig. 10, which shows an example of parameterized morphological structural elements. As shown in the figure, an example of parameterization of the structural elements of a circle and a rectangle is shown in 1000, and an example of parameterization of a circle is shown in 1010. The circle 1011 is a model with the radius R1012 as a parameter. Suppose that the radius R is set to 1, and the instantiated example 1013 is the circle shown by 1014. On the other hand, if the radius R is set to 2, the illustrated example 1015 is a circle shown by 1016. Assume that an example of a structural element that parameterizes a rectangle is shown in 1020. At this time, as shown by the rectangle of 1021, two parameters of a (1022) and b (1023) are used in the model. In the example 1024, a=1 and b=1.3 are used as parameters to realize the rectangle shown in 1026. In the example 1025, a=2 and b=1 are used as parameters, and the rectangle shown in 1027 is realized. These parameterized structural elements are just an example, and more complex structural element models can be defined. By using other structural elements, it is possible to extract complex structures and remove noise signals through morphological processing.

Therefore, when morphological processing is used to realize the detection accuracy improvement function, the detection accuracy changes through the selection of the morphological processing program and the selection of the detection accuracy parameters. Therefore, the best choice needs to be made. When the evaluation of detection accuracy is realized by a computer program, the selection of parameters and processing programs can be changed, so that the best program and parameter selection can be explored. In addition, when it is impossible to evaluate the detection accuracy through a computer program, the best choice can be made through human judgment.

The above descriptions are only the preferred embodiments of the present invention, and the described embodiments are not used to limit the scope of patent protection of the present invention. Therefore, any equivalent structural changes made using the contents of the description and drawings of the present invention should be included in the same reasoning. Within the protection scope of the appended claims of the present invention.

Claims (14)

1. A wafer inspection device includes a detection target image generation unit and a data processing unit. The detection target image generation unit is used to detect a detection target image of a detection object, and the data processing unit is used to perform defects on the detection target image. Extraction processing and determination; characterized in that, the data processing unit includes a preprocessing module and a noise reduction module, and the preprocessing module includes:

   The lithography simulator is used to simulate and calculate the layout data as the design result related to the object to obtain the power spectrum Ps(u,v);

   The noise power spectrum extraction module is used to obtain the full signal power spectrum Ps+n(u,v) of the noise signal containing the detection target image, and use all the signals according to the full signal power spectrum Ps+n(u,v) Describe the power spectrum Ps(u,v) to get the noise power spectrum Pn(u,v);

   The noise reduction module is configured to determine the type of the noise reduction filter and its characteristic function value according to the signal-to-noise ratio requirements, and combine the characteristic function, the noise power spectrum Pn(u,v) and the power spectrum Ps(u , v) As input, obtain the system parameters of the noise reduction filter, and form the noise reduction filter through the system parameters; and filter the detection target image by the noise reduction filter to obtain the noise information removed The detection target image.
2. The wafer inspection apparatus according to claim 1, wherein the data processing unit further comprises a defect extraction module and a restoration module; after obtaining the inspection target image, the defect extraction module is based on morphological image processing technology , Extracting the detection target image elements in the detection target image to obtain a noise image of the detection target image; the restoration module adds the detection target image elements in the detection target image to remove the noise in the detection target image In the detection target image after the information, the detection target image after noise removal is formed.
3. The wafer inspection apparatus according to claim 1, wherein the type of the noise reduction filter is a two-dimensional Wiener filter.
4. A data processing method using the wafer inspection device according to claim 1, characterized in that it comprises the following steps:

   Step S11: Perform simulation calculation on the wafer layout data as the design result related to the inspection object to obtain the power spectrum Ps(u,v); wherein the inspected wafer layout data does not contain noise information ；

   Step S12: Obtain the full signal power spectrum Ps+n(u,v) including the noise signal of the detection target image, and use the power spectrum Ps( u,v) to get the noise power spectrum Pn(u,v);

   Step S13: Determine the noise reduction filter type and its characteristic function according to the signal-to-noise ratio requirement, and use the characteristic function, the noise power spectrum Pn(u,v) and the power spectrum Ps(u,v) as inputs, Obtain the system parameters of the noise reduction filter, and form the noise reduction filter through the system parameters; and filter the detection target image by the noise reduction filter to obtain the detection target image after the noise information is removed.
5. 4. The data processing method of a wafer inspection device according to claim 4, further comprising: step S10 and step S14:

   Step S10: After the detection target image is obtained, based on morphological image processing technology, extract the detection target image elements in the detection target image to obtain a noise image of the detection target image;

   Step S14: The restoration module adds the detection target image elements in the detection target image to the detection target image after removing the noise information in the detection target image to form the detection target image after removing the noise .
6. A computer-readable medium for storing a computer-executable program, the executable program being used to execute the data processing method of a wafer inspection device as claimed in claim 4; characterized by comprising the following programs:

   Perform simulation calculation on the wafer layout data that is the design result of the inspection object to obtain the power spectrum Ps(u,v); obtain the full signal power spectrum Ps+n(u, v), and use the power spectrum Ps(u,v) according to the full signal power spectrum Ps+n(u,v) to obtain the noise power spectrum Pn(u,v);

   Determine the type of noise reduction filter and its characteristic function according to the signal-to-noise ratio requirements, and use the characteristic function, the noise power spectrum Pn(u,v) and the power spectrum Ps(u,v) as inputs to obtain noise reduction The system parameters of the filter are used to form a noise reduction filter through the system parameters; and after the detection target image is filtered by the noise reduction filter, the detection target image after the noise information is removed is obtained.
7. A wafer inspection device includes a detection target image generation unit and a data processing unit. The detection target image generation unit is used to detect and generate a detection target image of a detected object, and the data processing unit is used to compare the detection target image. Perform processing and judgment of defect extraction; characterized in that, the data processing unit includes a preprocessing module and a noise reduction module, and the preprocessing module includes:

   The noise region determination sub-module divides the detection target image formed by the detection target region into M subregions, and selects N subregions smaller than M that contain noise to participate in the estimation processing of the noise power spectrum Pn(u,v);

   An estimation processing sub-module for obtaining the full signal power spectrum P ^N s+n(u,v) of the detection target image including the N sub-regions; and

   The power spectrum estimation stator module obtains the full signal power spectrum Ps+n(u,v) of the detection target image, and uses the full signal power spectrum P ^N s+n(u,v) of the N sub-regions as the noise power Spectrum Pn(u,v), subtract the noise power spectrum Pn(u,v) from the full signal power spectrum Ps+n(u,v) composed of the M sub-regions to obtain the power spectrum Ps(u,v) );

   The noise reduction module is used to determine the type of noise reduction filter and its characteristic function according to the signal-to-noise ratio requirement, and combine the characteristic function, the noise power spectrum Pn(u,v) and the power spectrum Ps(u, v) As input, the system parameters of the noise reduction filter are obtained, and the noise reduction filter is formed by the system parameters; and after the detection target image is filtered by the noise reduction filter, all the noise information is obtained after removing the noise information. The detection target image.
8. 8. The wafer inspection device according to claim 7, wherein the M sub-regions are seamlessly connected, have the same shape, and are rectangular, square, diamond or honeycomb.
9. 7. The wafer inspection apparatus according to claim 7, wherein the average value of the sum of the square differences of each cell of the M sub-regions is calculated, and when the average value is less than a predetermined value, it is determined that it is caused by noise The N sub-regions constituted by signal components.
10. A data processing method using the wafer inspection device according to claim 7, characterized in that it comprises the following steps:

    Step S20: Divide the detection target image formed by the detection target area into M sub-areas, and select N sub-areas smaller than M that contain noise and participate in the noise power spectrum Pn(u,v) that are involved in the estimation of the noise power spectrum. ) Presumption;

    Step S21: Obtain the full signal power spectrum P ^N s+n(u,v) of the detection target image of N signals containing noise and large difference defects;

    Step S22: Obtain the full signal power spectrum Ps+n(u,v) of the detection target image, and use the full signal power spectrum P ^N s+n(u,v) of the N sub-regions as the noise power spectrum Pn( u,v), subtracting the noise power spectrum Pn(u,v) from the full signal power spectrum Ps+n(u,v) formed by the M sub-regions to obtain the power spectrum Ps(u,v);

    Step S23: Determine the type of noise reduction filter and its characteristic function according to the signal-to-noise ratio requirement, and use the characteristic function, the noise power spectrum Pn(u,v) and the power spectrum Ps(u,v) as inputs, Obtain the system parameters of the noise reduction filter, and form the noise reduction filter through the system parameters; and filter the detection target image by the noise reduction filter to obtain the detection target image after the noise information is removed.
11. A computer-readable medium for storing a computer-executable program, the executable program being used to execute the data processing method of a wafer inspection apparatus according to claim 10; characterized by comprising the following programs:

    Dividing the detection target image formed by the detection object region into M subregions, and selecting N subregions that are smaller than M and contain noise that participate in the estimation processing of the noise power spectrum;

    Infer the full signal power spectrum P ^N s+n(u,v) of the N sub-regions;

    Obtain the full signal power spectrum Ps+n(u,v) of the detection target image, and use the full signal power spectrum P ^N s+n(u,v) of the N sub-regions as the noise power spectrum Pn(u,v) ), subtracting the noise power spectrum Pn(u,v) from the full signal power spectrum Ps+n(u,v) formed by the M sub-regions to obtain the power spectrum Ps(u,v);

    Determine the type of noise reduction filter and its characteristic function according to the signal-to-noise ratio requirements, and use the characteristic function, the noise power spectrum Pn(u,v) and the power spectrum Ps(u,v) as inputs to obtain noise reduction The system parameters of the filter are used to form a noise reduction filter through the system parameters; and after the detection target image is filtered by the noise reduction filter, the detection target image after the noise information is removed is obtained.
12. A wafer inspection device includes a detection target image generation unit and a data processing unit. The detection target image generation unit is used to obtain inspections from multiple inspection target regions or multiple inspection target regions across multiple wafers. The target image, the data processing unit is used to perform defect extraction processing and judgment on the detection target image; it is characterized in that the data processing unit includes a preprocessing module and a noise reduction module, and the preprocessing module includes:

    The noise reduction area determination sub-module, which divides the detection target image i obtained from the multiple detected object areas or multiple detected object areas across the wafer into Mi sub-areas, and selects less than Mi containing noise Ni sub-regions of are used as noise estimation regions to participate in ^{the estimation processing of the noise power spectrum P i} n(u,v);

    The power spectrum estimation stator module estimates the full signal power spectrum P ⁱ s+n(u,v) according to the detection target image i;

    The noise power spectrum deduces the stator module to obtain the full signal power spectrum P ^Ni s+n(u,v) of ^{the Ni sub-regions, and make it the noise power spectrum P i} n(u,v);

    The difference signal power spectrum estimation stator module, according to the noise power spectrum P ⁱ n(u,v) and the full signal power spectrum P ⁱ s+n(u,v) to estimate the difference signal power spectrum P ⁱ s(u ,v);

    The noise signal extraction sub-module calculates the average value of the range of i from ^{the noise power spectrum P i} _{n (u, v) to obtain the average value of} Pn (u, v); and

    The difference signal extraction submodule calculates the average value of the range of i from the difference signal power spectrum Ps(u,v) to obtain the _{average value of} Ps (u,v);

    The noise reduction module, for determining a noise filter function according to the type and features SNR requirement, the characteristic function, the _mean noise power spectrum Pn _Zhi (u, v), and _average difference signal power spectrum Ps _{The value} (u, v) is used as input to obtain the system parameters of the noise reduction filter, and the noise reduction filter is formed by the system parameters; and the detection target image is filtered by the noise reduction filter to obtain the noise reduction filter The detection target image after the information.
13. A data processing method using the wafer inspection device according to claim 12, characterized in that it comprises the following steps:

    Step S30: Divide the detection target image i obtained from the plurality of detected object areas or the plurality of detected object areas across the wafer into Mi sub-areas, and select Ni sub-areas that are smaller than Mi and contain noise As a noise estimation area, it participates in the estimation processing of the noise power spectrum Pn(u,v);

    Step S31: Estimate the full signal power spectrum P ⁱ s+n(u,v) according to the detection target image i;

    Step S32: Obtain the full signal power spectrum P ^Ni s+n(u,v) of ^{the Ni sub-regions, and make it a noise power spectrum P i} n(u,v);

    Step S33: According to the noise power spectrum P ⁱ n(u,v) of ^{the selected Ni sub-regions and the full signal power spectrum P i} s+n(u,v), the difference signal power spectrum P ⁱ s( u,v), calculate the average value of the range of i for the difference signal power spectrum P ⁱ _{s(u,v) to obtain the average value of} Ps (u,v), and compare the noise power spectrum P ⁱ n(u, v) Calculate the average value of the range of i, and obtain the _{average value of} Pn (u, v);

    Step S34: used to determine the type of noise reduction filter and its characteristic function according to the signal-to-noise ratio requirement, and combine the characteristic function, the _{average value of} the noise power spectrum Pn (u, v) and the _{average value of the} difference signal power spectrum Ps (u , v) As input, obtain the system parameters of the noise reduction filter, and form the noise reduction filter through the system parameters; and filter the detection target image by the noise reduction filter to obtain the noise information removed The detection target image.
14. A computer-readable medium for storing a computer-executable program, the executable program being used to execute the data processing method of a wafer inspection device according to claim 13; characterized by comprising the following programs:

    The detection target image i obtained from the plurality of detection target regions or the plurality of detection target regions across the wafer is divided into Mi subregions, and Ni subregions smaller than Mi containing noise are selected as noise estimation Participate in the estimation processing of noise power spectrum P ⁱ n(u,v) by area;

    Infer the full signal power spectrum P ⁱ s+n(u,v) according to the detection target image i;

    Obtain the full signal power spectrum P ^Ni s+n(u,v) of ^{the Ni sub-regions, and make it a noise power spectrum P i} n(u,v);

    Infer the difference signal power spectrum P ⁱ s(u,v) according to the noise power spectrum P ⁱ n(u,v) and the full signal power spectrum P ^{i s+n(u,v);}

    Calculating the average value of the range of i for the noise power spectrum P ⁱ _{n(u,v) to obtain the average value of} Pn (u,v);

    Calculate the average value of the range of i for the difference signal power spectrum P ⁱ s(u,v) to obtain the _{average value of} Ps (u,v);

    Determine the noise reduction filter type and its characteristic function according to the signal-to-noise ratio requirements, and combine the characteristic function, the _{average value of} the noise power spectrum Pn (u, v) and the _{average value of} the difference signal power spectrum Ps (u, v) As input, the system parameters of the noise reduction filter are obtained, and the noise reduction filter is formed by the system parameters; and the detection target image is filtered by the noise reduction filter to obtain the detection after the noise information is removed. Target image.

Images