Semiconductor wafer surface defect detection method, device and equipment and storage medium

Abstract

The invention provides a semiconductor wafer surface defect detection method, device and equipment and a storage medium, and the method comprises the steps: carrying out the spatial filtering of a detection light beam emitted by an LED detection light source array, and irradiating the surface of a to-be-detected wafer at a preset angle. The reflected light is divided into a direct light path and a filtering light path, the direct light path obtains an original image through a second detector, and the filtering light path obtains a low-pass filtering image through a third detector after being subjected to spatial frequency filtering through a 4f optical system. A high-pass filtering image is obtained through differential operation of an original image and a low-pass filtering image, and self-adaptive threshold segmentation and morphological optimization processing are carried out on the high-pass filtering image to extract image feature data. And finally, inputting the image feature data into a defect detection model, and determining a defect detection result of the wafer surface. According to the method, interference of background noise is effectively reduced through spatial filtering, the contrast ratio of defect edge features is enhanced, and the detection precision and classification accuracy of tiny defects are improved.

Description

Technical Field

[0001] The present invention relates to the field of semiconductor technology, and in particular to a method, device, equipment and storage medium for detecting surface defects of semiconductor wafers. Background Art

[0002] Semiconductor wafer surface defect detection is a key process in the semiconductor device manufacturing process, and its detection accuracy directly affects product yield and device reliability. With the continuous advancement of semiconductor technology, the size of wafer surface defects is getting smaller and smaller, and the requirements for detection accuracy are getting higher and higher. Traditional semiconductor wafer surface defect detection methods mainly rely on direct optical imaging analysis, that is, obtaining defect images by directly illuminating the wafer surface with a light source. The defect images obtained by this method are easily interfered by background noise. The contrast between the defect area and the background area in the image is low, and it is difficult to highlight the edge features of the defect. Especially for tiny defects, the imaging quality is often poor, which seriously affects the accuracy of subsequent defect detection and classification. Summary of the invention

[0003] The main purpose of the present invention is to solve the technical problem that the existing semiconductor wafer surface defect detection relies on direct optical imaging analysis and is difficult to highlight the edge features of the defects; A first aspect of the present invention provides a method for detecting surface defects of a semiconductor wafer, the method comprising: Performing spatial filtering on the detection beam emitted by the LED detection light source array, and irradiating the detection beam after spatial filtering at a preset angle to the surface of the wafer to be tested through a preset reflector group; Splitting the detection light path reflected from the surface of the wafer to be tested into a direct light path and a filtered light path, receiving the direct light path through a preset second detector to obtain an original image, performing spatial frequency filtering on the filtered light path through a preset 4f optical system, and then receiving the low-pass filtered image through a third detector; Performing a difference operation on the original image and the low-pass filtered image to obtain a high-pass filtered image, and performing adaptive threshold segmentation and morphological optimization processing on the high-pass filtered image to obtain image feature data; The image feature data is input into a preset defect detection model, and a defect detection result of the surface of the wafer to be tested is determined by the defect detection model.

[0004] Optionally, in a first implementation of the first aspect of the present invention, the spatial filtering of the detection light beam emitted by the LED detection light source array and irradiating the surface of the wafer to be tested at a preset angle with the detection light beam after the spatial filtering through a preset reflector group comprises: The preset scanning sequence and driving current control each LED in the LED detection light source array row by row to emit a detection light beam; The detection light beam is collimated and spatially filtered by a preset collimating lens device and a pinhole aperture to obtain a detection light beam after spatial filtering; The reflection angle of the detection light beam after the spatial filtering process is adjusted by using a plurality of preset groups of reflection mirrors, so that the detection light beam after the filtering is incident on the surface of the wafer to be tested at an angle of 45°.

[0005] Optionally, in a second implementation of the first aspect of the present invention, the collimating lens device includes a first converging lens and a second converging lens; The detection light beam is collimated and spatially filtered by a preset collimating lens device and a pinhole aperture respectively, and the detection light beam after spatial filtering is obtained includes: Converging the detection light beam through the first converging lens to obtain a converged light beam, and spatially adjusting the convergence focus of the converged light beam so that the convergence focus is coaxial with the aperture of the pinhole aperture and is located on the incident surface of the pinhole aperture; Performing inter-filter processing on the detection light beam through the pinhole aperture to remove scattered light components in the convergent light beam to obtain a filtered divergent light beam; The filtered divergent light beam is collimated by the second converging lens to transform the filtered divergent light beam into a parallel light beam, thereby obtaining a detection light beam after spatial filtering.

[0006] Optionally, in a third implementation of the first aspect of the present invention, the 4f optical system includes a first lens and a second lens; The method of splitting the detection light path reflected from the surface of the wafer to be tested into a direct light path and a filtered light path, receiving the direct light path by a preset second detector to obtain an original image, and performing spatial frequency filtering on the filtered light path by a preset 4f optical system, and then receiving the low-pass filtered image by a third detector includes: The detection light path reflected from the surface of the wafer to be tested is split by a preset beam splitter to obtain a direct light path and a filtered light path, wherein the light intensity ratio of the direct light path to the filtered light path is 1:1; Performing photoelectric conversion processing on the direct light path through a preset second detector, converting the light intensity information of the direct light path into an electrical signal, and obtaining an original image; Focusing the filtering light path through the first lens, and obtaining a Fourier spectrum containing spatial frequency information of the surface of the wafer to be measured on the back focal plane of the first lens; Performing spatial frequency filtering on the Fourier spectrum through an aperture provided on the back focal plane of the first lens, so as to retain low-frequency components in the central region of the Fourier spectrum; The Fourier spectrum after spatial frequency filtering is imaged by the second lens, and the light intensity information after imaging is received by the third detector to obtain the low-pass filtered image.

[0007] Optionally, in a fourth implementation manner of the first aspect of the present invention, performing spatial frequency filtering on the Fourier spectrum by a stop provided on a rear focal plane of the first lens, and retaining a low-frequency component in a central area of ​​the Fourier spectrum comprises: The position of the diaphragm on the back focal plane of the first lens is adjusted by a three-axis translation stage, so that the aperture center of the diaphragm on the back focal plane of the first lens coincides with the optical axis of the first lens, and the diaphragm is located on the back focal plane of the first lens; The aperture size of the diaphragm is adjusted by a preset adjustment motor so that the aperture size of the diaphragm matches the size of the bright spot at the center of the Fourier spectrum, thereby obtaining an adjusted aperture size; The frequency of the Fourier spectrum is selected by the adjusted aperture, and only the low-frequency components in the central area of ​​the Fourier spectrum are allowed to pass through, so as to obtain a spectrum image after spatial frequency filtering.

[0008] Optionally, in a fifth implementation of the first aspect of the present invention, performing a differential operation on the original image and the low-pass filtered image to obtain a high-pass filtered image, and performing adaptive threshold segmentation and morphological optimization processing on the high-pass filtered image to obtain image feature data includes: Performing pixel-corresponding differential operation on the original image and the low-pass filtered image by an image processor to obtain a high-pass filtered image reflecting edge and detail information of the wafer surface; Performing multi-level threshold segmentation processing on the high-pass filtered image according to the grayscale histogram distribution of the high-pass filtered image to obtain an initial segmented image; The initial segmented image is subjected to morphological processing of opening and closing operations to eliminate noise points and burrs in the initial segmented image, so as to obtain a segmented image with a smooth contour; The segmented image with smooth contour is subjected to connected domain analysis by using a region marking algorithm, and feature parameters of each connected domain are extracted to obtain image feature data.

[0009] Optionally, in a sixth implementation of the first aspect of the present invention, inputting the image feature data into a preset defect detection model, and determining the defect detection result of the surface of the wafer to be tested by the defect detection model includes: Performing dimensionality reduction processing on the image feature data by using a principal component analysis algorithm to obtain a principal feature vector representing the geometric shape of the feature region; Analyzing the main feature vector by using a preset multi-dimensional anomaly detection algorithm based on Mahalanobis distance to obtain a confidence level, and determining whether there is a feature abnormal area on the surface of the wafer to be tested according to a preset confidence interval and the confidence level; If there is no characteristic abnormal area, setting the defect detection result as the surface of the wafer to be tested is defect-free; If there is a feature abnormal region, a nonlinear feature transformation is performed on the feature abnormal region through a feature mapping algorithm based on a kernel function to obtain a high-dimensional feature representation containing regional morphology and topological structure information; The high-dimensional feature representation is analyzed by a nearest neighbor classification algorithm and a defect assessment criterion to obtain a defect type determination result and a severity determination result of the feature abnormality area, and the defect type determination result and the severity determination result are used as the defect detection result.

[0010] A second aspect of the present invention provides a semiconductor wafer surface defect detection device, the semiconductor wafer surface defect detection device comprising: The spatial filtering module is used to perform spatial filtering on the detection beam emitted by the LED detection light source array, and irradiate the detection beam after spatial filtering at a preset angle to the surface of the wafer to be tested through a preset reflector group; A spectroscopic module is used to split the detection light path reflected from the surface of the wafer to be tested into a direct light path and a filtered light path, receive the direct light path through a preset second detector to obtain an original image, perform spatial frequency filtering on the filtered light path through a preset 4f optical system, and receive the low-pass filtered image through a third detector; An image processing module, used for performing a differential operation on the original image and the low-pass filtered image to obtain a high-pass filtered image, and performing adaptive threshold segmentation and morphological optimization processing on the high-pass filtered image to obtain image feature data; The detection module is used to input the image feature data into a preset defect detection model, and determine the defect detection result of the surface of the wafer to be tested through the defect detection model.

[0011] The third aspect of the present invention provides a semiconductor wafer surface defect detection device, comprising: a memory and at least one processor, wherein instructions are stored in the memory, and the memory and the at least one processor are interconnected via lines; the at least one processor calls the instructions in the memory so that the semiconductor wafer surface defect detection device performs the steps of the above-mentioned semiconductor wafer surface defect detection method.

[0012] A fourth aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the steps of the above-mentioned semiconductor wafer surface defect detection method.

[0013] The above-mentioned semiconductor chip surface defect detection method, device, equipment and storage medium detect the detection light beam emitted by the LED detection light source array, and after spatial filtering, irradiate the surface of the chip to be tested at a preset angle. The reflected light is divided into a direct light path and a filtered light path. The direct light path obtains the original image through the second detector, and the filtered light path obtains the low-pass filtered image through the third detector after spatial frequency filtering through the 4f optical system. The high-pass filtered image is obtained by differential operation of the original image and the low-pass filtered image, and adaptive threshold segmentation and morphological optimization processing are performed on it to extract image feature data. Finally, the image feature data is input into the defect detection model to determine the defect detection result of the chip surface. The present invention effectively reduces the interference of background noise through spatial filtering, enhances the contrast of defect edge features, and improves the detection accuracy and classification accuracy of tiny defects.

[0014] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention are realized and obtained by the structures particularly pointed out in the description, claims and drawings.

[0015] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A schematic diagram of a first embodiment of a semiconductor wafer surface defect detection method according to an embodiment of the present invention; Figure 2 A schematic diagram of an embodiment of a semiconductor wafer surface defect detection device according to an embodiment of the present invention; Figure 3 Schematic diagram of an embodiment of a semiconductor wafer surface defect detection device in an embodiment of the present invention. DETAILED DESCRIPTION

[0017] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0018] The terms "including" and "having" and any variations thereof mentioned in the embodiments of the present invention are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device end including a series of steps or units is not limited to the listed steps or units, but may optionally include other steps or units that are not listed, or may optionally include other steps or units that are inherent to these processes, methods, products or device ends.

[0019] To facilitate understanding of this embodiment, a semiconductor wafer surface defect detection method disclosed in an embodiment of the present invention is first described in detail. Figure 1 As shown, the method comprises the following steps: 101. Perform spatial filtering on the detection light beam emitted by the LED detection light source array, and irradiate the detection light beam after the spatial filtering on the surface of the wafer to be tested at a preset angle through a preset reflector group; In one embodiment of the present invention, the detection light beam emitted by the LED detection light source array is spatially filtered, and the detection light beam after spatial filtering is irradiated on the surface of the chip to be tested at a preset angle through a preset reflector group, including: each LED in the LED detection light source array is controlled row by row according to a preset scanning sequence and driving current to emit a detection light beam; the detection light beam is collimated and spatially filtered respectively through a preset collimating lens device and a pinhole aperture to obtain a detection light beam after spatial filtering; and the reflection angle of the detection light beam after spatial filtering is adjusted through a preset plurality of groups of reflectors so that the filtered detection light beam is incident on the surface of the chip to be tested at an angle of 45°.

[0020] Specifically, the LED array control system consists of a main controller, a driving current control module and a timing generation circuit. The main controller adopts a multi-bit microprocessor with a built-in digital-to-analog converter and a timer to generate accurate control signals. The driving current control module includes a constant current source circuit and a current feedback regulation circuit. The constant current source circuit provides an adjustable DC driving current for each LED. The current feedback regulation circuit samples the working current of the LED in real time, generates an error signal through a comparator circuit composed of an operational amplifier, and then adjusts the conduction state of the power device to ensure the stability of the LED working current. The timing generation circuit can be implemented using a programmable logic device, including a multi-stage shift register and a programmable frequency divider, which can generate accurate scanning timing. The scanning process is carried out according to a preset program, and each row of LEDs is lit in turn according to a preset time interval. A protection time is set between adjacent LEDs to eliminate crosstalk. When a row is scanned, it switches to the next row after a row blanking period. The clock frequency of the shift register can be adjusted within a certain range to meet different detection speed requirements. The brightness uniformity of each LED is guaranteed by a photoelectric feedback circuit, and the feedback circuit can compensate for the temperature drift and aging effect of the LED in real time. The system also includes overvoltage protection, overcurrent protection and temperature protection circuits to ensure that the LED array operates within a safe range.

[0021] Specifically, the collimating lens can adopt an achromatic double-cemented lens structure, which has a suitable focal length and numerical aperture, and can effectively correct spherical aberration and chromatic aberration. The collimated light beam has a small divergence angle. After collimation, the light beam enters the 4f system. The first lens is an achromatic lens with a suitable focal length and aperture, and the surface is coated with an anti-reflection film to improve the transmittance. After the light beam passes through the first lens, a Fourier transform surface is formed on its rear focal plane, and a precision pinhole diaphragm is installed on the plane. The pinhole diaphragm is made of precision-machined metal film and has good roundness and edge smoothness. The pinhole is installed on a three-dimensional displacement stage, and the position of the pinhole on the optical axis can be accurately adjusted. The second lens has the same specifications as the first lens, and the distance between the two lenses is controlled at a position of twice the focal length by a precision optical seat. The entire 4f system is installed on an optical platform with a shock-absorbing function. The diffraction pattern at the diaphragm can be monitored in real time by a camera to assist the optical path adjustment of the system. The beam after spatial filtering has good spatial coherence and wavefront quality.

[0022] Specifically, the reflector group is composed of three high-reflectivity reflectors, the base material of the reflectors is quartz glass, and the surface is coated with an enhanced reflection film. The first reflector is mounted on a two-axis angle adjustment seat, and the angle can be adjusted within a certain range. The second reflector is mounted on a two-axis angle adjustment seat and has a translation adjustment mechanism. The adjustment mechanism of the third reflector includes two-axis angle adjustment and three-axis displacement adjustment. The relative positions of the three reflectors can be calibrated by optical detection equipment. The reason for selecting a 45° incident angle is that: first, the reflected light at this angle will not interfere with the incident light path; second, the 45° incident can produce a good light-dark contrast, which is conducive to the detection of defects; third, the scattered light intensity distribution at this angle is uniform, which is convenient for subsequent image processing. The overall structure of the reflector group is made of materials with good temperature stability. The system can be equipped with an active feedback control device to compensate for the angle deviation caused by mechanical vibration and thermal drift to ensure the stability of the incident angle of the light beam.

[0023] Furthermore, the collimating lens device includes a first converging lens and a second converging lens; the detection light beam is collimated and spatially filtered respectively by the preset collimating lens device and the pinhole aperture to obtain the detection light beam after the spatial filtering process, including: converging the detection light beam by the first converging lens to obtain a converged light beam, and spatially adjusting the convergence focus of the converged light beam so that the convergence focus is coaxial with the aperture hole of the pinhole aperture and is located on the incident surface of the pinhole aperture; performing inter-filtering processing on the detection light beam by the pinhole aperture to remove scattered light components in the converged light beam to obtain a filtered divergent light beam; and collimating the filtered divergent light beam by the second converging lens to convert the filtered divergent light beam into a parallel light beam to obtain the detection light beam after the spatial filtering process.

[0024] Specifically, the collimating lens device is composed of two converging lenses, and the two converging lenses are selected with the same focal length to ensure the parallelism of the light beam after passing through the entire system. Both converging lenses adopt an achromatic lens structure, and the lens surface is coated with an anti-reflection film to reduce the reflection loss of light. The two converging lenses are installed on a precision adjustment seat, and the adjustment seat has axial and radial adjustment freedom to adjust the position of the lens and the alignment of the optical axis. The first converging lens is mainly used to converge the divergent detection light beam into a convergent light beam, and the second converging lens is used to re-collimate the divergent light beam after filtering through a pinhole aperture into a parallel light beam. The installation position of the two converging lenses needs to ensure that their optical axes are collinear to reduce aberrations. The material selection of the converging lens needs to consider factors such as the transmission band, chromatic aberration correction and machining performance. The surface accuracy and surface quality of the lens directly affect the beam quality.

[0025] Specifically, after the detection light beam is incident on the first converging lens, the lens converges the divergent light into a converging light beam. The focusing position of the converging light beam is determined by the focal length of the first converging lens, and the converging light beam forms a minimum spot at the focus. In order to enable the converging light beam to be effectively coupled with the pinhole aperture, a three-dimensional displacement adjustment device is required to accurately adjust the converging focus of the converging light beam. The adjustment process includes two aspects: axial position adjustment and lateral position adjustment. The axial adjustment ensures that the converging focus is located on the incident surface of the pinhole aperture, and the lateral adjustment ensures that the converging focus is coaxial with the aperture hole of the pinhole aperture. The position adjustment device adopts a precision displacement platform with sufficient adjustment accuracy. During adjustment, the light intensity distribution behind the aperture can be monitored in real time by a photoelectric detector. When the converging focus is coaxial with the aperture hole and is located on the incident surface, the light intensity passing through the aperture is the largest.

[0026] Specifically, after the convergent light beam is adjusted in space, an Airy disk distribution will be formed at the pinhole aperture. The central bright spot of the Airy disk contains the main optical information, while the outer diffraction ring is mainly produced by the scattered light component in the light beam. The selection of the aperture size of the pinhole aperture requires a balance between the filtering effect and the transmitted light intensity. If the aperture is too small, it will lead to excessive loss of light energy, and if the aperture is too large, the filtering effect will not be obvious. The pinhole aperture performs spatial filtering on the incident convergent light beam, allowing only the light beam in the central area of ​​the Airy disk to pass, thereby removing the scattered light component in the convergent light beam. The light beam passing through the pinhole aperture will form a divergent light beam due to diffraction. The processing quality of the pinhole aperture, such as the roundness of the aperture and the smoothness of the edge, will affect the filtering effect. In order to reduce stray reflections, the surface of the pinhole aperture needs to be treated with an absorption coating.

[0027] Specifically, the divergent light beam after filtering by the pinhole aperture continues to be transmitted to the second converging lens. The function of the second converging lens is to convert the divergent light beam into a parallel light beam, and its focal length needs to be selected the same as that of the first converging lens to ensure that the final output is a parallel light beam. The position of the second converging lens needs to be precisely adjusted so that its front focus coincides with the position of the pinhole aperture. When the divergent light beam passes through the second converging lens, a detection light beam after spatial filtering is obtained. This light beam has good spatial coherence, and its lateral intensity distribution is more uniform, which is suitable for subsequent defect detection. The adjustment mechanism of the second converging lens needs to have angle and position adjustment functions to ensure the accuracy of the outgoing light beam. The entire optical path system needs to take necessary vibration reduction and dust prevention measures to ensure the stability of the system.

[0028] 102. Splitting the detection light path reflected from the surface of the wafer to be tested into a direct light path and a filtered light path, receiving the direct light path through a preset second detector to obtain an original image, performing spatial frequency filtering on the filtered light path through a preset 4f optical system, and then receiving the low-pass filtered image through a third detector; In one embodiment of the present invention, the 4f optical system includes a first lens and a second lens; the detection light path reflected from the surface of the wafer to be tested is split into a direct light path and a filtered light path, the direct light path is received by a preset second detector to obtain an original image, and the filtered light path is subjected to spatial frequency filtering by a preset 4f optical system, and then a low-pass filtered image is received by a third detector, comprising: performing a splitting process on the detection light path reflected from the surface of the wafer to be tested by a preset spectroscope to obtain a direct light path and a filtered light path, wherein the light intensity ratio between the direct light path and the filtered light path is 1:1; and the filtered light path is subjected to spatial frequency filtering by a preset second detector to obtain a low-pass filtered image. The direct optical path is subjected to photoelectric conversion processing, and the light intensity information of the direct optical path is converted into an electrical signal to obtain an original image; the filtered optical path is subjected to focusing processing by the first lens, and a Fourier spectrum containing the surface spatial frequency information of the wafer to be measured is obtained on the back focal plane of the first lens; the Fourier spectrum is subjected to spatial frequency filtering processing by an aperture arranged on the back focal plane of the first lens, and the low-frequency component in the central area of ​​the Fourier spectrum is retained; the Fourier spectrum after the spatial frequency filtering processing is imaged by the second lens, and the light intensity information after imaging is received by the third detector to obtain the low-pass filtered image.

[0029] Specifically, the 4f optical system is composed of two lenses with the same focal length, and the distance between the two lenses needs to be precisely controlled to be the sum of the two focal lengths. Both lenses adopt an achromatic lens structure, and the selection of lens materials needs to comprehensively consider factors such as dispersion characteristics, mechanical processing performance and service life within the working wavelength range. The surface of the lens is coated with a multi-layer anti-reflection film to reduce reflection loss. The design of the anti-reflection film needs to consider the incident angle range and the working wavelength range. The two lenses are respectively installed on a six-degree-of-freedom precision adjustment seat. The adjustment seat has three orthogonal translation adjustments and three rotation axis angle adjustment functions to adjust the spatial position of the lens and the direction of the optical axis. The first lens is mainly used to transform the image information of the plane of the object to be measured into the frequency domain, and form a Fourier spectrum on its rear focal plane. The second lens is used to transform the frequency-filtered spectrum back to the spatial domain to obtain the filtered image. The installation of the two lenses needs to ensure the collinearity of the optical axis height accuracy, which needs to be achieved through optical detection equipment such as precision alignment tooling and interferometers. The surface accuracy and surface quality of the lens have an important influence on the imaging quality of the system. Generally, the surface accuracy is required to be better than λ / 10 and the surface roughness is required to be better than 0.5nm.

[0030] Specifically, after the detection light path reflected from the surface of the wafer to be tested enters the beam splitter, the beam splitter needs to split the incident light into two beams of completely equal intensity. The beam splitter adopts a non-polarized beam splitter prism structure, which is composed of two right-angle prisms glued together, and the beam splitting surface is coated with a non-polarized beam splitter film. The design of the beam splitter film needs to comprehensively consider factors such as the polarization state of the incident light, the incident angle and the working wavelength, to ensure that the intensity ratio of the two outgoing light beams is strictly maintained at 1:1 under actual working conditions. The beam splitter is installed on a precise five-degree-of-freedom adjustment seat, including two orthogonal inclination adjustments, one azimuth adjustment and two lateral displacement adjustments, which are used to adjust the spatial posture of the beam splitter so that the two beams after the splitting are strictly propagated in the predetermined direction. The direct optical path is used to obtain the original image information, and the filtering optical path is used for spatial frequency domain filtering. The surface flatness of the beam splitter directly affects the wavefront quality, and is generally required to be better than λ / 8. The thickness uniformity and interface flatness of the beam splitter film have an important influence on the splitting effect, and a high-precision optical thin film process is required. In addition, in order to reduce the influence of stray reflections, the non-working surface of the beam splitter needs to be anti-reflective treated.

[0031] Specifically, the direct light path after the splitting enters the second detector, and the detector converts the optical signal into an electrical signal to obtain the original image. The second detector uses a high-sensitivity surface array photodetector. The selection of the detector needs to comprehensively consider parameters such as resolution, sensitivity, dynamic range, and response speed. The size of the photosensitive surface of the detector must match the diameter of the incident light beam. The pixel size determines the spatial sampling interval of the image and needs to meet the requirements of the sampling theorem. The spectral response range of the detector must completely cover the working wavelength range of the detection light source, and the quantum efficiency must meet the detection sensitivity requirements. The analog signal output by the detector is first amplified by the preamplifier, and then the high-frequency noise is filtered out by the bandpass filter. Finally, it is digitally sampled by a high-precision analog-to-digital converter to obtain a digital original image. The installation position of the detector needs to be precisely controlled so that the chip surface is strictly conjugated with the photosensitive surface of the detector to ensure that a clear original image is obtained. The temperature control system of the detector is used to reduce dark current noise and improve the signal-to-noise ratio. In addition, the linearity and uniformity of the detector have an important impact on the image quality, and the response curve needs to be obtained and corrected through calibration.

[0032] Furthermore, the spatial frequency filtering processing is performed on the Fourier spectrum by means of an aperture arranged on the rear focal plane of the first lens to retain the low-frequency components in the central area of ​​the Fourier spectrum, including: adjusting the position of the aperture on the rear focal plane of the first lens by means of a three-axis translation stage so that the aperture center of the aperture on the rear focal plane of the first lens coincides with the optical axis of the first lens, and the aperture is located on the rear focal plane of the first lens; adjusting the aperture size of the aperture by means of a preset adjustment motor so that the aperture of the aperture matches the size of the bright spot at the center of the Fourier spectrum to obtain an adjusted aperture; and frequency selection is performed on the Fourier spectrum by means of the adjusted aperture to allow only the low-frequency components in the central area of ​​the Fourier spectrum to pass through to obtain a spectrum image after spatial frequency filtering.

[0033] Specifically, after the filtered light path after the splitting is incident on the first lens, the light field undergoes Fourier transform during the propagation process, and forms a spectrum distribution on the back focal plane of the first lens. This spectrum distribution is actually the spatial frequency decomposition of the object space image. The center of the spectrum corresponds to the zero-frequency component, which contains the average brightness information of the image. The area farther from the center corresponds to the component with higher spatial frequency. These high-frequency components contain information such as edges and details in the image. The focal length selection of the first lens needs to weigh multiple factors: the longer the focal length, the larger the spatial expansion range of the spectrum, the higher the frequency resolution, but at the same time the larger the system volume; the shorter the focal length, the more compact the system structure, but the spectrum is prone to aliasing. The formation of the spectrum follows the principle of Fourier optics, and the complex amplitude distribution of the incident light field on the back focal plane is the Fourier transform of the object light field. The first lens needs to have good aberration correction performance, especially spherical aberration and field curvature need to be strictly controlled, because these aberrations will cause spectrum distortion. The aperture of the lens should be large enough to receive all spatial frequency components. In order to obtain an ideal spectral distribution, the wavefront quality of the incident beam must also be guaranteed, which requires the incident beam to have good collimation and coherence. The installation tilt error of the first lens must be strictly controlled, because tilting will introduce additional aberrations and cause spectral distortion.

[0034] Specifically, an aperture of a specific size is set at the spectrum forming position of the focal plane behind the first lens, and the frequency components to be retained are selected by controlling the aperture size of the aperture. The aperture size of the aperture is directly related to the spatial frequency cutoff characteristics of the system: the smaller the aperture, the lower the cutoff frequency, and the more blurred the filtered image; the larger the aperture, the more high-frequency components are retained, and the richer the image details. The shape of the aperture is generally circular, because the circular aperture has the same cutoff characteristics for frequency components in all directions. The processing quality of the aperture is very important, and the aperture edge should be smooth to avoid diffraction effects. The aperture needs to be precisely positioned on the spectrum plane, which requires a precise three-dimensional displacement adjustment mechanism. The center of the aperture must strictly coincide with the optical axis, and eccentricity will lead to asymmetric filtering characteristics. In order to reduce the influence of stray light, the surface of the aperture needs to be treated with an absorption coating. The thickness of the aperture should be moderate. Too thick will cause aperture effect, and too thin will not be strong enough. In practical applications, multiple apertures with different apertures can be prepared, and the appropriate one can be selected according to specific needs. The diffraction effect and edge effect of the aperture will affect the filtering effect, which needs to be fully considered when designing the aperture.

[0035] Specifically, the frequency-filtered light field continues to propagate to the second lens, which performs an inverse Fourier transform on the filtered spectrum and reconstructs the frequency domain information into an image in the spatial domain. This reconstruction process is actually an inverse transformation process, which requires that the parameters of the second lens and the first lens are completely consistent, and the distance between the two lenses must be precisely controlled. The reconstructed image appears smoother because the high-frequency components are removed, and the noise and fast-changing edge details in the image are suppressed. The reconstructed image is received by the third detector and converted into an electrical signal. The performance index requirements of the third detector are similar to those of the second detector, but considering that the filtered image is received, its dynamic range requirements can be appropriately reduced. The distance from the second lens to the third detector needs to be precisely adjusted to obtain a clear reconstructed image. All components of the entire 4f system need to be fixed on the same stable base to avoid the influence of mechanical vibration and thermal drift. All reflection and transmission surfaces in the optical path need to be anti-reflected to improve the light energy utilization rate of the system. The contrast and resolution of the reconstructed image are directly affected by the aperture size of the aperture, and appropriate parameters need to be selected according to the detection requirements in practical applications. The system assembly process requires the use of precision optical instruments such as interferometers to ensure the precise alignment of the entire optical path.

[0036] 103. Performing a differential operation on the original image and the low-pass filtered image to obtain a high-pass filtered image, and performing adaptive threshold segmentation and morphological optimization processing on the high-pass filtered image to obtain image feature data; In one embodiment of the present invention, the step of performing a differential operation on the original image and the low-pass filtered image to obtain a high-pass filtered image, and performing adaptive threshold segmentation and morphological optimization processing on the high-pass filtered image to obtain image feature data comprises: performing a pixel-corresponding differential operation on the original image and the low-pass filtered image through an image processor to obtain a high-pass filtered image reflecting the edge and detail information of the chip surface; performing multi-level threshold segmentation processing on the high-pass filtered image according to the grayscale histogram distribution of the high-pass filtered image to obtain an initial segmented image; performing morphological processing of opening and closing operations on the initial segmented image to eliminate noise points and burrs in the initial segmented image to obtain a segmented image with a smooth contour; performing a connected domain analysis on the segmented image with a smooth contour through a region marking algorithm to extract feature parameters of each connected domain to obtain image feature data.

[0037] Specifically, the original image and the low-pass filtered image are first converted into grayscale images of the same size to ensure that the pixels of the two images can correspond one to one. The image processor reads the grayscale value of each corresponding pixel of the two images and performs a subtraction operation on these grayscale values. Since the low-pass filtered image retains the low-frequency information of the image and represents the background and slowly changing area of ​​the image, while the original image contains complete frequency information, the difference between the two reflects the high-frequency information in the image, that is, the edge and detail area. This difference process actually achieves the effect of high-pass filtering in the spatial domain. In order to avoid negative values ​​in the difference results, the grayscale values ​​after difference need to be offset, usually by adding a fixed offset value so that all difference results fall within the valid grayscale value range. For each pixel, overflow processing is also required to ensure that the final grayscale value does not exceed the dynamic range of the image. The image processor uses a pipeline structure to perform these operations, which can process multiple pixels in parallel and improve the processing speed. The obtained high-pass filtered image highlights the edge and detail information of the wafer surface, which is of great value for subsequent defect detection.

[0038] Specifically, the gray value distribution of all pixels in the high-pass filtered image is statistically analyzed to establish a gray histogram. The histogram reflects the frequency of occurrence of different gray values ​​in the image, and the peaks correspond to the main gray levels in the image. The statistical process of the gray histogram needs to traverse each pixel of the image, take the gray value of the pixel as an index, and accumulate the number of occurrences of each gray value. After obtaining the gray histogram, the dynamic programming method is used to determine the optimal segmentation threshold. This method regards the histogram as a one-dimensional data sequence, and calculates multiple segmentation thresholds by minimizing the inter-class variance and maximizing the intra-class variance. The number of thresholds can be dynamically determined according to the complexity of the image, and the iterative optimization method is usually used to find the best threshold combination. After determining the threshold, each pixel in the high-pass filtered image is compared with these thresholds, and the corresponding label value is assigned according to the interval in which the gray value of the pixel falls, so that the initial segmented image is obtained. Each region in the initial segmented image represents a set of pixels with similar gray features, and these regions preliminarily reflect the possible targets or defects in the image.

[0039] Specifically, morphological processing mainly uses two basic operations, opening and closing, both of which are composed of two basic morphological operators, dilation and erosion. The opening operation is to perform erosion first and then dilation. The erosion operation uses the structural element to scan the image. When the structural element is completely located in the target area, the position is retained, otherwise it is removed, so that noise points and burrs smaller than the structural element can be eliminated. The subsequent dilation operation uses the same structural element. When the structural element overlaps with the target area, the position is filled, so that the target area size reduced during the erosion process can be restored while maintaining the smoothness of the edge. The closing operation is to perform dilation first and then erosion. The dilation operation can fill small holes and cracks in the target area, and the subsequent erosion operation restores the target area to its original size while maintaining the filling effect. The choice of structural element has an important influence on the morphological processing effect. Usually, a circular or square structural element is used, and its size needs to be determined according to the noise characteristics that need to be removed. The order of opening and closing operations also needs to be carefully selected. Usually, the opening operation is performed first to remove external noise, and then the closing operation is performed to fill internal holes. During the processing, attention should be paid to maintaining the main morphological features of the target area and avoiding excessive smoothing that may lead to loss of useful information.

[0040] Specifically, the connected regions of the segmented image with smooth contours are marked by a two-pass scanning method. The first pass scans the image from top to bottom and from left to right, assigns a temporary label to each foreground pixel, and records the equivalent relationship between adjacent labels; the second pass merges the labels with equivalent relationships into the same label according to the equivalent relationship, thereby obtaining a uniquely identified connected region. For each identified connected region, a series of shape feature parameters are calculated, including area (number of pixels), perimeter (number of boundary pixels), roundness (the square of the perimeter is used to represent the regularity of the shape), major axis length, minor axis length, direction angle, etc. In addition, the grayscale statistical features of each connected region, such as the average grayscale value, grayscale standard deviation, grayscale entropy, etc., are calculated. The calculation of feature parameters requires traversing all pixels in each connected region, and some parameters such as roundness require complex mathematical operations. All these feature parameters together constitute the feature vector that describes the defect area. These feature vectors, as image feature data, will be used for subsequent defect classification and identification. In order to ensure the comparability of feature parameters, some parameters need to be normalized so that they are not affected by image resolution.

[0041] 104. Input the image feature data into a preset defect detection model, and determine a defect detection result of the surface of the wafer to be tested by using the defect detection model.

[0042] In one embodiment of the present invention, the inputting of the image feature data into a preset defect detection model, and determining the defect detection result of the surface of the wafer to be tested by the defect detection model includes: performing dimensionality reduction processing on the image feature data by a principal component analysis algorithm to obtain a principal feature vector that characterizes the geometric morphology of the feature region; analyzing the principal feature vector by a preset multidimensional anomaly detection algorithm based on Mahalanobis distance to obtain a confidence level, and determining whether there is a feature abnormal region on the surface of the wafer to be tested according to a preset confidence interval and the confidence level; if there is no feature abnormal region, setting the defect detection result to the surface of the wafer to be tested as defect-free; if there is a feature abnormal region, performing nonlinear feature transformation on the feature abnormal region by a feature mapping algorithm based on a kernel function to obtain a high-dimensional feature representation containing regional morphology and topological structure information; analyzing the high-dimensional feature representation by a nearest neighbor classification algorithm and a defect assessment criterion to obtain a defect type determination result and a severity determination result of the feature abnormal region, and using the defect type determination result and the severity determination result as the defect detection result.

[0043] Specifically, the original feature data needs to be standardized first to make features of different dimensions comparable. After standardization, the covariance matrix of the feature matrix is ​​calculated. The covariance matrix reflects the correlation between different features. The covariance matrix is ​​decomposed by eigenvalue to obtain eigenvalues ​​and corresponding eigenvectors. The size of the eigenvalue reflects the degree of data change in the direction of the corresponding eigenvector. The larger the eigenvalue, the more information is contained in this direction. The eigenvalues ​​are sorted from large to small, and the first few eigenvectors with the largest contribution rate are selected to form a projection matrix. The original feature data is linearly transformed through the projection matrix to obtain the main eigenvectors after dimensionality reduction. These main eigenvectors retain the main information of the original data, while greatly reducing the data dimension and simplifying subsequent processing. The selection of eigenvectors needs to comprehensively consider the cumulative contribution rate and the actual calculation complexity. Generally, eigenvectors with a cumulative contribution rate of more than 85% are selected. The dimensionality reduction process not only compresses the data volume, but also eliminates the correlation between features, making subsequent anomaly detection more reliable.

[0044] Specifically, the mean vector and covariance matrix of the sample are calculated based on the main eigenvector. These two statistics describe the distribution characteristics of normal samples. For each sample to be detected, the Mahalanobis distance between it and the mean vector is calculated. The Mahalanobis distance takes into account the scale differences and correlations of each dimension in the feature space, and can more accurately measure the similarity between samples. The inverse matrix of the covariance matrix is ​​used as the metric matrix in the distance calculation, which can eliminate the influence of the correlation between features. The calculated Mahalanobis distance obeys the chi-square distribution, so the threshold can be determined according to the preset confidence level. The calculated Mahalanobis distance is compared with the threshold. If the distance exceeds the threshold, the sample is considered to be an abnormal sample. The selection of the confidence interval requires a balance between the sensitivity and reliability of the detection. Generally, a confidence level of 95% or 99% is selected. The result of anomaly detection directly determines the subsequent processing flow. If anomalies are detected, further feature extraction and classification are required.

[0045] Specifically, an appropriate kernel function is selected. Common kernel functions include Gaussian kernel, polynomial kernel, etc. The selection of kernel function needs to consider the distribution characteristics and nonlinearity of the data. The samples in the original feature space are mapped to the high-dimensional feature space through the kernel function, and the samples that were originally nonlinearly separable in the high-dimensional space become linearly separable. The feature mapping process is actually performed implicitly. There is no need to explicitly calculate the coordinates in the high-dimensional space, but directly calculate the kernel function value. The parameter selection of the kernel function has an important influence on the mapping effect, and the optimal parameters can be selected by cross-validation and other methods. The high-dimensional features obtained by mapping not only contain the morphological information of the original features, but also extract deeper structural features through nonlinear transformation. This nonlinear feature transformation can better characterize the complex morphological characteristics of the defect area and provide more effective feature representation for subsequent classification. The distance metric in the high-dimensional feature space adopts the kernel-induced metric, which is more in line with the essential structure of the data.

[0046] Specifically, the process of analyzing high-dimensional features through the nearest neighbor classification algorithm and defect assessment criteria. First, a k-nearest neighbor classifier is constructed in the high-dimensional feature space. The value of the nearest neighbor k needs to be selected through the cross-validation method to select the optimal value. The selection of the k value needs to balance the generalization ability of the classifier and the local feature expression ability. For each sample to be classified, the distance between it and all samples in the training sample set in the kernel space is calculated. This distance calculation process utilizes the properties of the kernel function and does not require explicit calculation of the coordinates in the high-dimensional feature space. The k nearest neighbor samples are selected, and the category of the sample to be classified is determined by weighted voting. The weight is inversely proportional to the distance between the samples, so that the samples with closer distance have greater voting weight. At the same time, corresponding evaluation criteria are established for each defect type, including multiple evaluation indicators such as defect size, shape complexity, and edge sharpness. The evaluation criteria use fuzzy rules to map the values ​​of each indicator to the severity score. In the specific calculation, each indicator is first normalized to unify its value range to the [0,1] interval, and then the score corresponding to each indicator is calculated according to the pre-defined membership function. The final severity determination result is obtained by weighted average of each score, and the weight coefficient reflects the importance of different indicators. The final defect detection result contains information on two dimensions: defect type and severity. This multi-dimensional evaluation result can more comprehensively reflect the quality status of the wafer surface. The confidence of the test result can be measured by the consistency of neighboring samples. The higher the consistency, the more reliable the classification result. The entire analysis process adopts a parallel computing architecture, which can process multiple samples to be tested at the same time and improve the detection efficiency.

[0047] In this embodiment, the detection beam emitted by the LED detection light source array is processed by spatial filtering and then irradiated at a preset angle on the surface of the chip to be tested. The reflected light is divided into a direct light path and a filtered light path. The direct light path obtains the original image through the second detector, and the filtered light path obtains the low-pass filtered image through the third detector after spatial frequency filtering through the 4f optical system. A high-pass filtered image is obtained by differential operation of the original image and the low-pass filtered image, and adaptive threshold segmentation and morphological optimization are performed on it to extract image feature data. Finally, the image feature data is input into the defect detection model to determine the defect detection result of the chip surface. The present invention effectively reduces the interference of background noise through spatial filtering, enhances the contrast of defect edge features, and improves the detection accuracy and classification accuracy of tiny defects.

[0048] The above is a description of the semiconductor wafer surface defect detection method in the embodiment of the present invention. The following is a description of the semiconductor wafer surface defect detection device in the embodiment of the present invention. Figure 2 , an embodiment of a semiconductor wafer surface defect detection device in an embodiment of the present invention comprises: The spatial filtering module 201 is used to perform spatial filtering on the detection beam emitted by the LED detection light source array, and illuminate the surface of the wafer to be tested at a preset angle through a preset reflector group with the detection beam after spatial filtering; A light splitting module 202 is used to split the detection light path reflected from the surface of the wafer to be tested into a direct light path and a filtered light path, receive the direct light path through a preset second detector to obtain an original image, perform spatial frequency filtering on the filtered light path through a preset 4f optical system, and receive the low-pass filtered image through a third detector; The image processing module 203 is used to perform a difference operation on the original image and the low-pass filtered image to obtain a high-pass filtered image, and perform adaptive threshold segmentation and morphological optimization processing on the high-pass filtered image to obtain image feature data; The detection module 204 is used to input the image feature data into a preset defect detection model, and determine the defect detection result of the surface of the wafer to be tested through the defect detection model.

[0049] In an embodiment of the present invention, the semiconductor wafer surface defect detection device runs the above-mentioned semiconductor wafer surface defect detection method. The semiconductor wafer surface defect detection device detects the detection light beam emitted by the LED detection light source array, and after spatial filtering, irradiates the surface of the wafer to be tested at a preset angle. The reflected light is divided into a direct light path and a filtered light path. The direct light path obtains the original image through the second detector, and the filtered light path obtains the low-pass filtered image through the third detector after spatial frequency filtering through the 4f optical system. The high-pass filtered image is obtained by differential operation of the original image and the low-pass filtered image, and adaptive threshold segmentation and morphological optimization are performed on it to extract image feature data. Finally, the image feature data is input into the defect detection model to determine the defect detection result of the wafer surface. The present invention effectively reduces the interference of background noise through spatial filtering, enhances the contrast of defect edge features, and improves the detection accuracy and classification accuracy of tiny defects.

[0050] above Figure 2 The semiconductor wafer surface defect detection device in the embodiment of the present invention is described in detail from the perspective of modular functional entities, and the semiconductor wafer surface defect detection device in the embodiment of the present invention is described in detail from the perspective of hardware processing.

[0051] Figure 3 3 is a schematic diagram of the structure of a semiconductor wafer surface defect detection device provided by an embodiment of the present invention. The semiconductor wafer surface defect detection device 300 may have relatively large differences due to different configurations or performances, and may include one or more processors (central processing units, CPU) 310 (for example, one or more processors) and a memory 320, and one or more storage media 330 (for example, one or more mass storage device terminals) storing application programs 333 or data 332. Among them, the memory 320 and the storage medium 330 can be short-term storage or permanent storage. The program stored in the storage medium 330 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations in the semiconductor wafer surface defect detection device 300. Furthermore, the processor 310 may be configured to communicate with the storage medium 330, and execute a series of instruction operations in the storage medium 330 on the semiconductor wafer surface defect detection device 300 to implement the steps of the above-mentioned semiconductor wafer surface defect detection method.

[0052] The semiconductor wafer surface defect detection device 300 may also include one or more power supplies 340, one or more wired or wireless network interfaces 350, one or more input and output interfaces 360, and / or one or more operating systems 331, such as Windows Serve, Mac OS X, Unix, Linux, FreeBSD, etc. It will be appreciated by those skilled in the art that Figure 3 The structure of the semiconductor wafer surface defect detection device shown does not constitute a limitation on the semiconductor wafer surface defect detection device provided by the present invention, and may include more or less components than shown in the figure, or a combination of certain components, or a different arrangement of components.

[0053] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the semiconductor wafer surface defect detection method.

[0054] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described system, device, or unit can refer to the corresponding process in the aforementioned method embodiment and will not be repeated here.

[0055] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the whole or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program code.

[0056] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for detecting surface defects of a semiconductor wafer, characterized in that: The semiconductor wafer surface defect detection method comprises: Performing spatial filtering on the detection beam emitted by the LED detection light source array, and irradiating the detection beam after spatial filtering at a preset angle to the surface of the wafer to be tested through a preset reflector group; Splitting the detection light path reflected from the surface of the wafer to be tested into a direct light path and a filtered light path, receiving the direct light path through a preset second detector to obtain an original image, performing spatial frequency filtering on the filtered light path through a preset 4f optical system, and then receiving the low-pass filtered image through a third detector; Performing a difference operation on the original image and the low-pass filtered image to obtain a high-pass filtered image, and performing adaptive threshold segmentation and morphological optimization processing on the high-pass filtered image to obtain image feature data; The image feature data is input into a preset defect detection model, and a defect detection result of the surface of the wafer to be tested is determined by the defect detection model.

2. The semiconductor wafer surface defect detection method according to claim 1, characterized in that: The method of performing spatial filtering on the detection light beam emitted by the LED detection light source array and irradiating the detection light beam after the spatial filtering at a preset angle on the surface of the wafer to be tested through a preset reflector group includes: The preset scanning sequence and driving current control each LED in the LED detection light source array row by row to emit a detection light beam; The detection light beam is collimated and spatially filtered by a preset collimating lens device and a pinhole aperture to obtain a detection light beam after spatial filtering; The reflection angle of the detection light beam after the spatial filtering process is adjusted by using a plurality of preset groups of reflection mirrors, so that the detection light beam after the filtering is incident on the surface of the wafer to be tested at an angle of 45°.

3. The semiconductor wafer surface defect detection method according to claim 2, characterized in that: The collimating lens device comprises a first converging lens and a second converging lens; The detection light beam is collimated and spatially filtered by a preset collimating lens device and a pinhole aperture respectively, and the detection light beam after spatial filtering is obtained includes: Converging the detection light beam through the first converging lens to obtain a converged light beam, and spatially adjusting the convergence focus of the converged light beam so that the convergence focus is coaxial with the aperture of the pinhole aperture and is located on the incident surface of the pinhole aperture; Performing inter-filter processing on the detection light beam through the pinhole aperture to remove scattered light components in the convergent light beam to obtain a filtered divergent light beam; The filtered divergent light beam is collimated by the second converging lens to transform the filtered divergent light beam into a parallel light beam, thereby obtaining a detection light beam after spatial filtering.

4. The semiconductor wafer surface defect detection method according to claim 1, characterized in that: The 4f optical system includes a first lens and a second lens; The method of splitting the detection light path reflected from the surface of the wafer to be tested into a direct light path and a filtered light path, receiving the direct light path by a preset second detector to obtain an original image, and performing spatial frequency filtering on the filtered light path by a preset 4f optical system, and then receiving the low-pass filtered image by a third detector includes: The detection light path reflected from the surface of the wafer to be tested is split by a preset beam splitter to obtain a direct light path and a filtered light path, wherein the light intensity ratio of the direct light path to the filtered light path is 1:1; Performing photoelectric conversion processing on the direct light path through a preset second detector, converting the light intensity information of the direct light path into an electrical signal, and obtaining an original image; Focusing the filtering light path through the first lens, and obtaining a Fourier spectrum containing spatial frequency information of the surface of the wafer to be measured on the back focal plane of the first lens; Performing spatial frequency filtering on the Fourier spectrum through an aperture provided on the back focal plane of the first lens, so as to retain low-frequency components in the central region of the Fourier spectrum; The Fourier spectrum after spatial frequency filtering is imaged by the second lens, and the light intensity information after imaging is received by the third detector to obtain the low-pass filtered image.

5. The semiconductor wafer surface defect detection method according to claim 4, characterized in that: The performing spatial frequency filtering on the Fourier spectrum by using a stop disposed on the back focal plane of the first lens to retain the low-frequency components in the central area of ​​the Fourier spectrum comprises: The position of the diaphragm on the back focal plane of the first lens is adjusted by a three-axis translation stage, so that the aperture center of the diaphragm on the back focal plane of the first lens coincides with the optical axis of the first lens, and the diaphragm is located on the back focal plane of the first lens; The aperture size of the diaphragm is adjusted by a preset adjustment motor so that the aperture size of the diaphragm matches the size of the bright spot at the center of the Fourier spectrum, thereby obtaining an adjusted aperture size; The frequency of the Fourier spectrum is selected by the adjusted aperture, and only the low-frequency components in the central area of ​​the Fourier spectrum are allowed to pass through, so as to obtain a spectrum image after spatial frequency filtering.

6. The method for detecting surface defects of a semiconductor wafer according to claim 1, characterized in that: The step of performing a differential operation on the original image and the low-pass filtered image to obtain a high-pass filtered image, and performing adaptive threshold segmentation and morphological optimization processing on the high-pass filtered image to obtain image feature data includes: Performing pixel-corresponding differential operation on the original image and the low-pass filtered image by an image processor to obtain a high-pass filtered image reflecting edge and detail information of the wafer surface; Performing multi-level threshold segmentation processing on the high-pass filtered image according to the grayscale histogram distribution of the high-pass filtered image to obtain an initial segmented image; The initial segmented image is subjected to morphological processing of opening and closing operations to eliminate noise points and burrs in the initial segmented image, so as to obtain a segmented image with a smooth contour; The segmented image with smooth contour is subjected to connected domain analysis by using a region marking algorithm, and feature parameters of each connected domain are extracted to obtain image feature data.

7. The semiconductor wafer surface defect detection method according to claim 1, characterized in that: The step of inputting the image feature data into a preset defect detection model and determining the defect detection result of the surface of the wafer to be tested by using the defect detection model comprises: Performing dimensionality reduction processing on the image feature data by using a principal component analysis algorithm to obtain a principal feature vector representing the geometric shape of the feature region; Analyzing the main feature vector by using a preset multi-dimensional anomaly detection algorithm based on Mahalanobis distance to obtain a confidence level, and determining whether there is a feature abnormal area on the surface of the wafer to be tested according to a preset confidence interval and the confidence level; If there is no characteristic abnormal area, setting the defect detection result as the surface of the wafer to be tested is defect-free; If there is a feature abnormal region, a nonlinear feature transformation is performed on the feature abnormal region through a feature mapping algorithm based on a kernel function to obtain a high-dimensional feature representation containing regional morphology and topological structure information; The high-dimensional feature representation is analyzed by a nearest neighbor classification algorithm and a defect assessment criterion to obtain a defect type determination result and a severity determination result of the feature abnormality area, and the defect type determination result and the severity determination result are used as the defect detection result.

8. A semiconductor wafer surface defect detection device, characterized in that: The semiconductor wafer surface defect detection device comprises: The spatial filtering module is used to perform spatial filtering on the detection beam emitted by the LED detection light source array, and irradiate the detection beam after spatial filtering at a preset angle to the surface of the wafer to be tested through a preset reflector group; A spectroscopic module is used to split the detection light path reflected from the surface of the wafer to be tested into a direct light path and a filtered light path, receive the direct light path through a preset second detector to obtain an original image, perform spatial frequency filtering on the filtered light path through a preset 4f optical system, and receive the low-pass filtered image through a third detector; An image processing module, used for performing a differential operation on the original image and the low-pass filtered image to obtain a high-pass filtered image, and performing adaptive threshold segmentation and morphological optimization processing on the high-pass filtered image to obtain image feature data; The detection module is used to input the image feature data into a preset defect detection model, and determine the defect detection result of the surface of the wafer to be tested through the defect detection model.

9. A semiconductor wafer surface defect detection device, characterized in that: The semiconductor wafer surface defect detection device comprises: a memory and at least one processor, wherein the memory stores instructions; The at least one processor calls the instructions in the memory to enable the semiconductor wafer surface defect detection device to perform the steps of the semiconductor wafer surface defect detection method as described in any one of claims 1-7.

10. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the steps of the semiconductor wafer surface defect detection method as described in any one of claims 1 to 7 are implemented.

Images