A wafer defect detection system based on dual-channel optical imaging

Abstract

This invention belongs to the field of optical precision measurement technology, specifically a wafer defect detection system based on dual-channel optical imaging. This invention combines dark-field scattering and diffraction phase microscopy as two optical imaging channels; it ensures optical path compatibility and synchronization between the two imaging modes; through optical system calibration and optimization, it achieves high-precision acquisition and alignment of multi-modal images; it uses standard samples to calibrate the system's sensitivity and resolution, ensuring that the detection performance meets preset requirements; it performs dual-channel synchronous imaging on the wafer sample under test, acquiring dark-field scattering images and diffraction phase images respectively; it uses image algorithms to jointly analyze the multi-modal data, extracting the size, depth, and structural information of defects, and generating a high-precision defect detection report; specifically, it includes a dark-field scattering imaging module, a diffraction phase microscopy imaging module, and an image processing unit; this invention effectively overcomes the shortcomings of a single imaging mode, significantly improving the system's detection accuracy and robustness.

Description

Technical Field

[0001] This invention belongs to the field of optical precision measurement technology, and specifically relates to a wafer defect detection system. Background Technology

[0002] With the rapid development of the semiconductor industry, the size of integrated circuit chips is continuously miniaturizing according to Moore's Law, and process nodes are gradually moving towards the nanometer scale. This technological advancement has not only significantly improved the performance of semiconductor devices but also placed higher demands on defect detection technology. Wafer defects, including particle contamination, scratches, and residues, can not only affect the electrical performance of devices but also lead to device failure and even reduce the yield of the entire production line. Therefore, accurate and reliable defect detection technology plays a crucial role in semiconductor manufacturing, becoming a key link in ensuring product quality and production efficiency.

[0003] Optical inspection technology has become the mainstream method for wafer defect detection due to its advantages of non-contact, high-speed, and large-area inspection. Traditional bright-field optical inspection technology mainly detects surface defects by utilizing the reflection or transmission characteristics of light. However, as process nodes continue to shrink, defect sizes are gradually approaching or even falling below the wavelength scale of light, causing significant bottlenecks in resolution and sensitivity for bright-field inspection technology. Specifically, bright-field inspection struggles to effectively distinguish between minute defects and background noise, especially under low contrast or complex backgrounds, where its detection performance is severely limited.

[0004] To overcome the aforementioned problems, dark-field inspection technology has gradually become a core method for wafer defect detection due to its high sensitivity to minute defects. Unlike bright-field inspection, dark-field technology significantly enhances the signal response of subwavelength defects by collecting the scattered light signals caused by defects, thereby achieving efficient detection of defects such as tiny particles and scratches. In recent years, dark-field scattering inspection technology has been widely used in wafer surface defect detection, especially in the field of semiconductor metrology equipment, where it has made significant progress. For example, KLA, a leading semiconductor metrology equipment company, has developed the Puma series and Surfscan SP systems using dark-field optics technology, which have greatly improved the detection capability of wafer defects, providing advanced inspection solutions for semiconductor manufacturers worldwide.

[0005] However, despite its excellent performance in detecting minute defects, dark-field scattering (DFS) technology still faces several technical bottlenecks in high-precision detection. First, the defect signal intensity in DFS is closely related to the direction of the light source; for defects with specific orientations, weak scattered light can make detection difficult. Second, interference between background light and scattered light in low-contrast or complex backgrounds can reduce detection accuracy. Furthermore, DFS lacks the ability to acquire defect depth information, making it difficult to accurately characterize deep defects. These limitations restrict the application potential of DFS technology in high-end semiconductor manufacturing.

[0006] Patent CN116359249A discloses a line-scanning dark-field scattering wafer surface defect detection device and method based on TDI (Transient Direction Identification). The device includes an illumination source, a beam shaping component, a motion control component, a microscope imaging component, a CMOS image sensor, and a computer. By utilizing the principles of line scanning and dark-field scattering, it can effectively collect scattered light from wafer surface defects and perform rapid imaging using a CMOS image sensor, followed by image enhancement and defect extraction via computer. However, the limited depth of field of its microscope system, the increased edge field aberration due to its large field of view, and the instability caused by mechanical vibration can still lead to signal annihilation in the image edge regions, affecting the detection of deep subwavelength defects.

[0007] Patent CN118883578A discloses an automatic optical inspection system and defect detection method, which can quickly detect minute defects on the surface of semiconductor wafers by using multi-angle light source illumination and high-resolution camera imaging. The system can adjust the light source angle and imaging parameters according to different defect types to improve detection accuracy. However, the detection accuracy of this technology is still limited when dealing with low-contrast defects, and it lacks the ability to acquire defect depth information.

[0008] Patent CN109991233A proposes an optical detection device and method. This method utilizes multiple light sources positioned relative to the sample under test, along with a strobe controller that time-divisionally controls the emission of detection beams from the light sources. A camera then acquires images under different light sources, achieving high-precision detection. However, this technology primarily relies on a single imaging mode, and its detection accuracy needs improvement for small defects in low-contrast or complex backgrounds. Furthermore, this technology lacks the ability to acquire defect depth information, making it difficult to accurately characterize deep defects. Summary of the Invention

[0009] The purpose of this invention is to propose a wafer defect detection system based on dual-channel optical imaging with high detection accuracy and strong anti-interference capability, so as to meet the requirements of advanced process nodes for high-precision defect detection.

[0010] This invention provides a wafer defect detection system based on dual-channel optical imaging, combining dark-field scattering and diffraction phase microscopy as two optical imaging channels. The optical paths of the two imaging channels are compatible and synchronized. High-precision acquisition and alignment of multi-modal images are achieved through optical system calibration and optimization. The system's sensitivity and resolution are calibrated using standard samples to ensure that the detection performance meets preset requirements. The system performs dual-channel synchronous imaging on the wafer sample under test, acquiring dark-field scattering and diffraction phase images respectively. Multi-modal data are jointly analyzed using image algorithms to extract the size, depth, and structural information of defects, and generate a high-precision defect detection report. Specifically, it includes: a dark-field scattering imaging module, a diffraction phase microscopy imaging module, and an image processing unit; wherein:

[0011] The dark field scattering imaging module is used to acquire the scattered light signals from defects on the wafer surface;

[0012] The diffraction phase microscopy imaging module is used to acquire the phase information and depth structure of the defect;

[0013] The image processing unit is used to perform fusion analysis on the acquired multimodal images and extract the size, depth and structural information of the defects.

[0014] Furthermore:

[0015] The dark field scattering imaging module includes: an oblique incident light source, an imaging unit, and a high-sensitivity detector; wherein:

[0016] An obliquely incident light source is used to provide a non-perpendicular illumination light field;

[0017] An imaging unit is used to collect light signals scattered from the sample surface and to create an image;

[0018] A high-sensitivity detector is used to receive defect-scattered light signals and generate dark-field scattering images.

[0019] The diffraction phase microscopy imaging module includes: a coherent light source, an imaging unit, a phase resolution unit, and a high-sensitivity detector; wherein:

[0020] A coherent light source is used to provide the coherent light field required for phase imaging;

[0021] An imaging unit is used to collect light signals reflected from the sample surface and to create an image;

[0022] The phase analysis unit is used to analyze the phase information of the defect and generate a diffraction phase image;

[0023] High-sensitivity detector: used to record interference images.

[0024] Through optical path design and calibration, the optical path compatibility and synchronization between the dark field scattering imaging module and the diffraction phase microscopy imaging module are ensured (by separating the wavelengths of the two bands using a dichroic mirror and sharing a microscope objective).

[0025] The image processing unit includes: an image registration module and a phase demodulation module for the carrier interferogram; wherein:

[0026] The image registration module is used to spatially align the dark field scattering image with the diffraction phase image;

[0027] The phase demodulation module of the carrier interferogram is used to perform phase recovery processing on the diffraction phase image, thereby extracting more detailed feature information of the sample.

[0028] Furthermore, the tilted incident light source of the dark field scattering imaging module is based on dark field scattering detection technology, the principle of which is as follows:

[0029] By illuminating the wafer surface with oblique incident light, the scattering effect of the incident light by the tiny defects or irregularities on the surface is utilized to collect the scattered light signal at a specific angle to achieve defect imaging. The optical path design of the oblique incident light source ensures that the background light does not participate in the imaging, and only the defect scattered light is received by the detector, thereby significantly improving the image contrast and achieving high-sensitivity detection of tiny defects.

[0030] The scattering of particles on a wafer surface is a complex problem, difficult to solve rigorously. However, it can be qualitatively expressed as:

[0031]

[0032] In the formula, I pLet I0 be the scattering intensity of the surface particles, I0 be the incident light intensity, d be the particle diameter, and λ be the incident wavelength (Reference: Germer T A. Angular dependence and polarization of out-of-plane optical scattering from particulate contamination, subsurface defects, and surface microroughness[J]. Applied Optics, 1997, 36(33): 8798-8805). According to this formula, there is a proportional relationship between the scattering intensity of the surface particles and the sixth power of the particle diameter and the fourth power of the incident wavelength. Therefore, in order to improve the sensitivity of the scattering signal, in this invention, to meet the system's requirements for light source wavelength, adjustability, spot stability, and other optical performance, a semiconductor laser with a center wavelength of 405nm is selected as the light source. This laser has a collimated output circular or elliptical beam, an output power of about 50mW, an integrated heat dissipation module, and good power stability and anti-interference performance.

[0033] Furthermore,

[0034] The laser beam is first emitted from a 455nm wavelength laser. After passing through a beam expander system consisting of a polarizer, lenses L1 and L2 (with a pinhole between lenses L1 and L2 to improve spatial coherence), the beam is expanded and collimated. The expanded beam then passes through lens L5 (e.g., ...). Figure 1 As shown, two dichroic mirrors are installed between lenses L2 and L5. Dichroic mirror 1 reflects light with a wavelength of 455nm, while dichroic mirror 2 allows light with a wavelength of 455nm to pass through. The purpose is to separate light beams of different wavelengths, thereby achieving directional transmission of the light beam and ensuring that the light with a wavelength of 455nm can accurately illuminate lens L5 and be focused at the focal point of the objective lens (e.g., Figure 1 As shown, there is a reflecting mirror between L5 and the objective lens (the purpose of which is to fold the optical path). The focused beam enters the microscope system through the objective lens, becomes parallel light, and is incident perpendicularly on the sample surface. The microscope system consists of a microscope objective lens and a matching tube lens L7 (e.g., Figure 1 As shown, a dichroic mirror 3 is located between the objective lens and the tube lens L7. This dichroic mirror 3 can reflect light with a wavelength of 455nm (the purpose of which is to separate beams of different wavelengths). The sample is imaged after passing through the aforementioned optical system. The beam is focused onto the grating, generating diffraction sub-beams in multiple directions at the grating, thereby achieving spatial replication of the beam. Next, the first-order diffracted light is made to interfere with the zero-order light through a spatial filter, and the interference image is recorded on the camera CMOS2 (e.g., ...). Figure 1As shown, there are two mirrors, 3 and 4, between the spatial filter and the camera CMOS2 (also for folding the optical path), thereby effectively extracting the phase information of the sample and performing imaging. To prevent the fringes at the carrier frequency interferogram from being too dense and affecting phase demodulation, a 4f system consisting of lenses L8 and L9 is inserted between the image plane of the microscope system and the camera.

[0035] Based on this, we only need to consider the light fields of the 0th-order and 1st-order diffracted beams after passing through the grating:

[0036] U(x,y)=U o (x,y)+U1(x,y)exp(iβx), (2)

[0037] Where U0 is the complex amplitude of the 0th-order diffracted light, U1 is the complex amplitude of the 1st-order light multiplied by an exponential phase tilt factor with diffraction angle, and i represents an imaginary number. The magnitude of the 1st-order diffraction angle is... Where Λ is the grating constant.

[0038] Since passing through the 4f system composed of lenses L8 and L9 is equivalent to undergoing two Fourier transforms, its spatial coordinates are flipped in both the x and y directions. For ease of representation, let:

[0039]

[0040] Among them, M 4f Let A0 be the magnification of the 4f system, A1 be the intensity of the 0th order light, A1 be the intensity of the 1st order light, and α be the normalization factor. Then the interference intensity captured at the camera target surface is obtained as:

[0041]

[0042] In the formula, U c (x,y) represents the light field at the camera, U c * (x, y) is its conjugate. The interferogram I can be recovered using the phase adjustment method. c (x',y'), and remove the carrier frequency β. φ1 is a constant, which can be removed by setting the background height to 0. The resulting φ0(x',y') implicitly contains information such as the surface morphology or object structure of the sample.

[0043] Furthermore, the principle of the dual-channel optical imaging wafer defect detection system is as follows:

[0044] This invention integrates dark-field scattering imaging and diffraction phase microscopy into a single optical system, with the two complementing each other to achieve comprehensive characterization of wafer defects. Dark-field scattering imaging, by enhancing the signal response of minute defects, is suitable for detecting surface particles and cracks at subwavelength scales. The scattering intensity of surface particles in dark-field scattering is proportional to the sixth power of the particle diameter and the fourth power of the incident wavelength. For interferometric detection, the phase change caused by the surface ridge height can be approximately expressed as… Therefore, in depth detection, phase information has higher sensitivity in the vertical direction compared to scattering intensity information. Thus, diffraction phase microscopy can reveal the three-dimensional structure and depth distribution of defects through precise phase information. The combination of these two techniques achieves a complementary advantage in spatial resolution and depth information.

[0045] Furthermore, the image registration module's functions include:

[0046] Load the initial image and perform geometric transformation operations based on the image's capture characteristics;

[0047] Select the region of interest (ROI) in the dark field scattering image;

[0048] Feature points were extracted from diffraction phase microscopy images and dark-field scattering images using the SIFT algorithm.

[0049] Matching is performed using the nearest neighbor matching method, and matching points are filtered using a ratio test.

[0050] Calculate the geometric transformation parameters (affine transformation) between images and perform precise image alignment;

[0051] Display the registration results.

[0052] Furthermore, the phase demodulation module of the carrier frequency interferogram operates on the following principle:

[0053] In off-axis interferometric optics systems, based on the principle of two-beam interference, the interferogram can be represented as follows:

[0054]

[0055] In the formula, a(x,y) is the background intensity of the interferogram, and γ(x,y) is the amplitude modulation factor. This represents phase information. ω x and ω yThe carrier frequency component is used. In off-axis interferograms with carrier frequency components, the interference fringes are highly dense due to the large carrier frequency introduced by the phase, and phase changes at different positions cause fringe distortion. The required phase information is hidden in these dense fringes. Overall, the carrier frequency-induced fringes are dense, while the background intensity and amplitude modulation factor change more slowly. Therefore, assuming that the carrier frequency components in the x and y directions are much larger than the background light intensity, amplitude modulation factor, and phase gradient, the following conditions can be approximated:

[0056]

[0057] The first and second partial derivatives of the interferogram in the y-direction are:

[0058]

[0059] The envelope phase of the object light wave:

[0060]

[0061] Here, `unwrap` represents the unwrapping function. Off-axis interference, through the angle between the object wave and the reference wave, makes the interference fringes carrying phase information denser, thus effectively separating the fringe phase information from the background intensity and amplitude modulation factor. Therefore, even a single interferogram with a carrier frequency can recover the phase information.

[0062] Furthermore, the principle of the image fusion analysis is as follows:

[0063] Since dark-field scattering imaging and diffraction phase microscopy acquire sample information from different physical perspectives, each has its own advantages and limitations in providing images. Therefore, after completing the aforementioned image processing module, it is necessary to compare and analyze the dark-field scattering image and the diffraction phase image to comprehensively extract all information about sample defects. This comparative analysis process combines the advantages of both image types, supplementing the shortcomings of a single image mode and improving the accuracy and robustness of defect detection.

[0064] Compared with the prior art, the present invention has at least the following beneficial effects:

[0065] (1) The present invention example fully utilizes the high sensitivity of dark field scattering in the detection of small defects and the accuracy of diffraction phase microscopy in the characterization of structural information by jointly acquiring multimodal images of the same detection location and combining them with signal processing algorithms, thereby achieving effective complementarity between the two and improving detection accuracy and reliability.

[0066] (2) The diffraction phase microscopy technique in the example of the present invention uses bright field illumination, which overcomes the dependence of dark field scattering on the contrast between background light and scattered light, and improves the detection accuracy. The sensitivity of dark field scattering at small scale can also serve as a supplement to the diffraction phase microscopy technique.

[0067] (3) In the examples of this invention, the off-axis interference design effectively suppresses noise caused by optical components or environmental vibrations, enhances system stability and robustness, and meets the requirements for high precision and high reliability testing in wafer manufacturing. Attached Figure Description

[0068] Figure 1 This is a structural diagram of the dual-channel optical imaging wafer defect detection system of the present invention.

[0069] Figure 2 This is a flowchart illustrating the image registration module algorithm of the present invention.

[0070] Figure 3 This is an example of registering images taken using a resolution plate by the optical system of this invention.

[0071] Figure 4 This is a flowchart illustrating the algorithm of the phase demodulation module of the carrier interferogram of the present invention.

[0072] Figure 5 These are the test results of the diffraction phase microscopy module in the system of this invention using 10μm polystyrene microspheres.

[0073] Figure 6 The results are from a system test of the dark field scattering module in the system of this invention, which uses 10μm polystyrene microspheres.

[0074] Figure 7 These are the test results for a dual-channel system of 1μm polystyrene microspheres.

[0075] Figure 8 These are the test results for a dual-channel system of 300nm polystyrene microspheres.

[0076] Figure 9 These are the test results for a 200nm polystyrene microsphere dual-channel system.

[0077] Figure 10 These are the test results for a 150nm polystyrene microsphere dual-channel system.

[0078] Figure 11 These are the test results for a 60nm polystyrene microsphere dual-channel system. Detailed Implementation

[0079] The present invention will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present invention, but do not limit the invention in any way. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention. These all fall within the scope of protection of the present invention.

[0080] This embodiment provides a dual-channel optical imaging wafer defect detection system based on dark-field scattering and diffraction phase microscopy, such as... Figure 1 As shown, the system is built on an optical air-bearing platform and is composed of two optical detection systems. The purple beam represents the optical path of the dark field scattering system, and the blue beam represents the optical path of the diffraction phase microscopy system.

[0081] The light source in the dark-field scattering system is a 405nm laser. The beam is first expanded by a Galilean telescope system consisting of lenses L3 and L4, and then passes through dichroic mirror 1. Dichroic mirror 1 is a short-pass, allowing 405nm light to pass through while reflecting 455nm light. Following dichroic mirror 1, dichroic mirror 2 is placed. This dichroic mirror 2 is a long-pass, allowing 455nm light to pass through while reflecting 405nm light, thus separating the two beams of different wavelengths. The 405nm light, after passing through dichroic mirror 2, is reflected and obliquely incident on the sample through an adjustable-angle mirror 1, causing scattering on the sample surface. The sample is placed on a sample stage, which is a displacement stage that allows control of the sample's movement. The scattered light is then received by the objective lens, passes through mirror 2, and then through dichroic mirror 3. Dichroic mirror 3 is a short-pass, allowing 405nm light to pass through while reflecting 455nm light. Finally, the scattered light is imaged onto the camera CMOS1 through the tube lens L6 that is paired with the objective lens, thus achieving image acquisition.

[0082] The optical path of the diffraction phase microscopy system uses a 455nm wavelength laser as the light source. The beam is expanded and collimated by a polarizer and a beam expander system consisting of cemented doublet lenses L1 and L2. A pinhole is added between lenses L1 and L2 to improve the spatial coherence of the beam. The expanded beam is first reflected by dichroic mirror 1, then by dichroic mirror 2, and finally focused by lens L5 to the objective lens focal point. During this process, mirror 2 is used to fold the optical path. The focused beam enters the microscopy system through the objective lens, becoming parallel light and incident perpendicularly on the sample surface. After the objective lens collects the reflected light, it passes through mirror 2 and then by dichroic mirror 3 to the tube lens L7, which is paired with the objective lens. A grating is placed at the focal point of tube lens L7, generating diffraction sub-beams in multiple directions, thus achieving spatial replication of the beam. These diffracted beams then pass through lens L8, with their focal point aligned with the grating. A spatial filter is used to retain only the first-order diffracted light and the zero-order light. Lens L9, along with lenses L8 and L9, forms a 4f system to ensure proper amplification and spatial filtering of the beam. The first-order diffracted light and the zero-order light interfere at this point, and the interference image is recorded on the camera CMOS 2. To further fold the optical path, two mirrors, 3 and 4, are placed between the spatial filter and the camera CMOS 2 to ensure optimized optical path and imaging quality.

[0083] This dual-channel optical imaging wafer defect detection system can perform dark-field scattering imaging and diffraction phase microscopy imaging on the same location on the same wafer. Images acquired by CMOS1 and CMOS2 are uploaded to a computer for further processing and analysis. Specifically, this includes:

[0084] S1, Build a dark field scattering imaging module to collect scattered light signals from defects on the wafer surface;

[0085] S2, Construct a diffraction phase microscopy imaging module to obtain the phase information and depth structure of defects;

[0086] S3 is an image processing unit used to process and analyze the acquired images, extract the size, depth and structural features of defects, and also implements the registration of images acquired by two optical paths to ensure spatial alignment of the images.

[0087] Specifically:

[0088] The dark-field scattering imaging module in S1 includes:

[0089] The oblique incident light source consists of a 405nm wavelength laser, lens L3, lens L4, dichroic mirror 1, dichroic mirror 2, and reflector 1: used to provide a non-perpendicular illumination light field.

[0090] The imaging unit consists of bright-field and dark-field objectives and their matching tube lens L6: it collects light signals scattered from the sample surface and forms an image. In dark-field imaging mode, the objectives collect scattered light at a specific angle, thereby effectively eliminating light directly reflected from the sample surface and enhancing the visibility of defect areas.

[0091] High-sensitivity detector (CMOS1): used to receive images of scattered light in dark fields.

[0092] The diffraction phase microscopy imaging module in S2 includes:

[0093] The coherent light source consists of a 455nm laser, a polarizer, lens L1, a pinhole, lens L2, dichroic mirror 1, dichroic mirror 2, and lens L5. The function of mirror 2 is to fold the optical path to provide the coherent light field required for phase imaging.

[0094] The imaging unit consists of a bright-field objective lens, a dark-field objective lens, a matching tube lens L7, and a dichroic mirror 3: it is used to collect light signals reflected from the sample surface and to form an image.

[0095] The phase resolution unit consists of a grating, lens L8, a spatial filter, and lens L9. The function of mirrors 3 and 4 is to fold the optical path: to resolve the phase information of the defect and generate a diffraction phase image.

[0096] High-sensitivity detector (CMOS2): Used to record interference images.

[0097] Further optical path design and calibration ensure the optical path compatibility and synchronization between the dark field scattering imaging module and the diffraction phase microscopy imaging module (by separating the two light wavelengths with a dichroic mirror and sharing a single objective lens).

[0098] The image processing unit in S3 includes:

[0099] The image registration module is used to spatially align the dark field scattering image with the diffraction phase image;

[0100] The phase demodulation module of the carrier interferogram is used to perform phase recovery processing on the diffraction phase image, thereby extracting more detailed feature information of the sample.

[0101] Because the field of view of the dark-field scattering optical path differs significantly from that of the diffraction phase microscopy system, in a dual-channel optical system, the dark-field scattering optical path is primarily used to roughly locate the initial position and morphological features of sample defects, while the diffraction phase microscopy system precisely locates and further characterizes the detailed structure and phase information of the target defects, thus enabling depth analysis and high-resolution imaging of the defects. Simultaneously, the dark-field scattering imaging system exhibits higher sensitivity in lateral resolution compared to the diffraction phase microscopy system, effectively detecting smaller-scale surface defects. This is because dark-field scattering imaging significantly enhances the detectability of minute scattering signals, thereby improving the detection capability for small defects.

[0102] S31. To ensure the accuracy and consistency of images under the two imaging modes, thorough image calibration and comparison must be performed before imaging. Therefore, an image registration module is designed for calibration, ensuring that images from different optical paths can be accurately matched on a spatial scale. Its flowchart is as follows: Figure 2As shown. First, the diffraction phase microscopy image is a mirror image of the dark-field scattering image in the vertical and horizontal directions. Without proper flipping correction, direct matching will lead to incorrect correspondences of feature points. Therefore, before matching, the diffraction phase microscopy image needs to be flipped horizontally and vertically to ensure its orientation is consistent with the dark-field scattering image. Furthermore, since the dark-field scattering image (low magnification) has a large field of view and a wide coverage area, while the diffraction phase microscopy image (high magnification) has a smaller field of view, covering only a local area, directly performing feature matching globally will significantly increase computational complexity and may reduce matching accuracy. Therefore, during registration, a preliminary Region of Interest (ROI) should first be manually or automatically selected in the dark-field scattering image. This region should correspond to the field of view of the diffraction phase microscopy image. Within this selected ROI, feature point matching methods can be used for precise registration to improve matching accuracy and computational efficiency. Regarding the specific method selection for image registration, based on the characteristics of the optical path and the features of dual-channel images, the following methods were selected and operated sequentially:

[0103] (1) In the feature extraction stage, the SIFT algorithm is used to extract feature points from dark-field scattering images and diffraction phase microscopy images, respectively. The SIFT algorithm has excellent scale invariance and can adapt to images of different resolutions, ensuring that feature points can be stably extracted at different scales of the image. In addition, the SIFT algorithm has strong robustness to illumination changes and noise, and can extract stable feature points under complex conditions, so its performance in image matching is particularly outstanding. Due to the high computational efficiency of the SIFT algorithm, it is particularly suitable for real-time image processing tasks.

[0104] (2) In the feature matching stage, matching point pairs are screened using the nearest neighbor matching method and the ratio test. Nearest neighbor matching is an intuitive and computationally efficient matching method, suitable for large-scale data processing, and provides preliminary matching point pairs for subsequent screening and optimization. Multiple similar feature points may exist in an image, leading to matching errors, while the ratio test can effectively reduce the occurrence of mismatches. The ratio test is robust to changes in illumination, noise interference, and geometric deformation, and can adapt to various complex application scenarios.

[0105] (3) In the geometric transformation estimation stage, based on matching point pairs, an affine transformation model is used to estimate the geometric relationship between images. The affine transformation model can accurately describe geometric deformations such as rotation, scaling, translation, and shearing between images, and is suitable for registration between different fields of view. Through the affine transformation model, the geometric relationship between two images can be accurately estimated and corrected, providing a reliable basis for subsequent image fusion and further processing. Finally, the diffraction phase microscopy image is transformed into the coordinate system of the dark field scattering image to achieve image alignment. Through this calibration process, errors caused by optical path configuration, imaging magnification, and field of view differences can be eliminated, thereby improving the accuracy and reliability of the final detection results. The system was tested using a USAF-1951 resolution plate (positive test target), and the results are as follows. Figure 3 As shown. In Figure 3 In the four pictures, Figure 3 (a) The actual images captured by the dark field scattering system and the image registration and positioning performed. Figure 3 (b) is an actual image taken using diffraction phase microscopy; Figure 3 (c) Figure 3 (d) shows the image registration results of the dual-channel imaging optical path. The results show that the two images were well registered.

[0106] S32, the image captured by the diffraction phase microscopy system undergoes phase recovery processing via the phase demodulation module of the carrier interferogram. In the frequency domain, dense interference fringes correspond to the high-frequency portion, while background intensity and other information reside in the low-frequency portion. Using a suitable frequency domain filter, background interference can be effectively removed, and phase information extracted. The spectrum of the carrier interferogram consists of three parts: the 0th-order fundamental frequency, the +1st-order high-frequency (containing object wave phase information), and its -1st-order conjugate term. The +1st-order part extends along k in the spectrum... x and k y Direction offset ω x and ω yThe +1 level component is extracted using a spatial filter, and the complex amplitude is recovered using the inverse Fourier transform (IFT). The phase is further recovered using the arctangent function. However, the Fourier transform, as a global operation, performs poorly in processing local signal feature variations, especially signals with large frequency changes. In carrier frequency interferograms, regions with large phase gradients have dense fringes, while regions with small phase gradients have sparse fringes, thus requiring the algorithm to have strong local analysis capabilities. For this purpose, the fractional Fourier transform (FrFT) provides an effective solution. Unlike the traditional Fourier transform, the fractional Fourier transform achieves multi-scale local analysis by adjusting the transform order, enhancing its ability to capture local signal features. Its flexible parameter control and time-frequency domain switching characteristics enable it to more accurately recover the local phase information of interferograms with large frequency variations, outperforming the Fourier transform. The fractional Fourier transform is a "transformation" of the Fourier transform, using the parameter α to control the "angle" of the transform. Its transformation formula is as follows:

[0107]

[0108] Among them, K α (x,x') is the kernel of the fractional Fourier transform. The complete process of the diffraction domain defect detection algorithm based on the fractional Fourier transform is as follows: Figure 4 As shown, firstly, the interference patterns of the reference wafer and the wafer under test are simultaneously separated using dual-exposure interferometry. The WFF algorithm is then used to reconstruct the complex amplitude of the interference fringes, effectively suppressing speckle noise and improving reconstruction accuracy. Based on this, multi-distance diffraction propagation calculations are performed on the complex amplitude fields of the reference and test samples to construct a complex amplitude sequence containing information about different axial positions, thereby enhancing the response characteristics of defects in the spatial and frequency domains. To improve the phase sensitivity of the defect region, the algorithm performs point-by-point complex conjugate multiplication of the complex amplitude fields of the test and reference samples, converting the phase perturbation caused by micro-defects into an amplitude modulation signal. Subsequently, FrFT is used to extract fractional-domain features from the modulated complex amplitude field. By optimizing the transform order and fractional-domain parameters, the directional enhancement of the subwavelength defect phase gradient features is achieved.

[0109] To evaluate the system's ability to detect defects of different sizes, wafer samples containing polystyrene microspheres of varying sizes were prepared to simulate particulate defects on patternless wafers. The concentration of the microsphere suspension used was 2.5%, and for example, 1 ml of suspension contained approximately 40 billion microspheres with a particle size of 1 μm.

[0110] To prevent microsphere aggregation, the suspension was diluted 500 times with ultrapure water and then uniformly dispersed onto a 2-inch single-sided polished silicon wafer using a spray bottle. The wafer had a roughness of less than 0.5 nm, a flatness TIR of less than 3 μm, and a warpage TTV of less than 10 μm. All sample preparation and testing were performed in a cleanroom.

[0111] First, the sample was examined using a diffraction phase microscopy system to evaluate the feasibility of the algorithm. The test results for 10 μm particle size are as follows: Figure 5 As shown. Figure 5 (a) is the original recorded interferogram, which was adjusted to achieve the best contrast in the interferogram fringes; Figure 5 (b) is for Figure 5 (a) Calibration image after flipping; Figure 5 (c) is the Fourier spectrum corresponding to the interferogram. The three lobes in the spectrum are completely separated, which meets the phase recovery condition of the carrier frequency interferogram, verifying that the constructed diffraction phase microscopy system has the ability to detect the phase in the vertical direction of the sample. Figure 5 (d) The phase recovery results of the microsphere sample using the fractional Fourier transform method. Figure 5 (a) Background noise and diffraction bands can be seen in the interferogram, while... Figure 5 In the phase recovery result in (d), the noise effect of the original image is almost invisible, and the phase result and its lateral width are consistent with the characteristics of 10μm microspheres.

[0112] At the same time, comparing the results captured by the dark field scattering system, such as Figure 6 As shown. Figure 5 and Figure 6 A comparison of the two images shows that it is feasible to accurately align the images obtained from the dark-field scattering system and the diffraction phase microscopy system using image registration techniques. Specifically, in a dark-field scattering image with a large field of view, the defect is first initially located. Subsequently, by combining the diffraction phase microscopy image, detailed morphological information of the defect can be obtained.

[0113] Subsequently, the system was further tested on even smaller defects, gradually decreasing to the subwavelength scale, to verify its detection capabilities in high-precision defect identification. Figures 7 to 11As shown, microsphere samples with sizes of 1 μm, 300 nm, 200 nm, and 150 nm were tested. In each figure, (a) represents the original image captured by the scattering imaging system and localized using an image registration algorithm; (b) represents the original image captured by the diffraction phase microscopy system; (c) and (d) are the registered dark-field scattering system image and diffraction phase microscopy image, respectively; (e) and (f) are the 3D and 2D views of the phase recovery results of the microsphere sample, respectively. The figures show that the system has the capability to detect subwavelength scale defects, as demonstrated by the analysis of the detection results for samples of different sizes. Specifically, for larger polystyrene microspheres (especially those above 200 nm), the diffraction phase microscopy system exhibits superior performance in defect detection, clearly presenting the size, morphology, and other structural features of the defects. In contrast, the dark-field scattering system has weaker imaging clarity and particle resolution in this size range, only providing information about the presence of defects without accurately characterizing their specific morphological features. Therefore, compared with a single dark-field scattering system, the dual-channel detection system combined with diffraction phase microscopy significantly improves the accuracy of detection and characterization capabilities.

[0114] However, as the defect size further decreases, the imaging contrast and phase retrieval accuracy of the diffraction phase microscopy system are gradually affected by noise, especially when the defect size drops to 60 nm. In this situation, the dark-field scattering system, with its high sensitivity to minute defects, can still maintain good imaging contrast and effectively detect the presence of tiny particles. When the phase retrieval results of the diffraction phase microscopy system are hampered by noise and the defect features are difficult to discern, the dark-field scattering system can still provide clear detection results, providing strong support for the reliability of defect identification.

Claims

1. A wafer defect detection system based on dual-channel optical imaging, characterized in that, Combine the dark field scattering and the diffraction phase microscopy as two optical imaging channels; the light paths of the two optical imaging channels are compatible and synchronous; through optical system calibration and optimization, realize high-precision acquisition and alignment of multi-modal images; Calibrate the sensitivity and resolution of the system with standard samples to ensure that the detection performance meets the preset requirements; perform double-channel synchronous imaging on the wafer samples to be tested to obtain dark field scattering images and diffraction phase images respectively; through image algorithm, jointly analyze the multi-modal data to extract the size, depth and structure information of the defects, and generate a high-precision defect detection report; specifically including: a dark field scattering imaging module, a diffraction phase microscopic imaging module, and an image processing unit; wherein: The dark field scattering imaging module is used to collect the scattering light signals of the wafer surface defects; The diffraction phase microscopic imaging module is used to obtain the phase information and depth structure of the defects; The image processing unit is used to process and analyze the collected images, extract the size, depth and structure characteristics of the defects, and also realize the registration of the images obtained by the two light paths to ensure the spatial alignment of the images; In the light path: The light beam is first emitted by a 455nm wavelength laser, then expanded and collimated after passing through the beam expansion system composed of a polarizer, double-cemented lenses L1 and L2, a pinhole is also added between lenses L1 and L2 to improve the spatial coherence of the light beam; the expanded light beam is focused on the focal point of the objective lens, in between, two dichroic mirrors are installed between lenses L2 and L5, dichroic mirror 1 is used to reflect 455nm wavelength light, while dichroic mirror 2 allows 455nm wavelength light to pass through, allowing different wavelength light beams to be separated, achieving directional transmission of the light beam, so that the 455nm wavelength light can accurately irradiate onto lens L5; there is also a reflecting mirror between lens L5 and the objective lens to fold the light path; the focused light beam enters the microscopic system through the objective lens, becomes parallel light, and is perpendicular to the sample surface; the microscopic system is composed of a microscopic objective lens and a supporting tube lens L7, and a dichroic mirror 3 is also provided between the objective lens and the tube lens L7, which is used to reflect 455nm wavelength light to separate different wavelength light beams; the sample is imaged after passing through the foregoing optical system; the light beam is focused on the grating, generating multiple direction diffraction beams at the grating, thereby realizing spatial replication of the light beam; then through a spatial filter, the first-order diffraction light and the zero-order light interfere, and the interference image is recorded on the camera CMOS2; wherein, between the spatial filter and the camera CMOS2, two reflecting mirrors 3 and 4 are also provided to realize the folding of the light path, thereby effectively extracting the phase information of the sample and imaging; a 4f system composed of lenses L8 and L9 is inserted between the image plane of the microscopic system and the camera; Based on this, considering the light field of the 0th order diffraction light and the 1st order diffraction light after passing through the grating: U(x, y) = U o (x, y) + Ui(x, y)exp(iβx), (1) Wherein, U0 is the complex amplitude of the 0th order diffraction light, U1 is the complex amplitude of the 1st order light multiplied by the exponential phase tilt factor with the diffraction angle, i is an imaginary unit, and the size of the 1st order diffraction angle is β = 2π / Λ, wherein Λ is the grating constant; After passing through the 4f system composed of lenses L8 and L9, which is equivalent to performing two Fourier transforms, the spatial coordinates are flipped in both the x direction and the y direction. For convenience of representation, let where M 4f is the magnification of the 4f system, A0is the intensity of the 0th order light, A1is the intensity of the 1st order light, and a is a normalization factor. The captured intensity at the camera target plane is then given by: where U c (x,y) is the light field at the camera, U c * (x,y) is its conjugate; the interferogram I is recovered by phase retrieval c (x',y'), and the carrier frequency β is removed; φ1 is a constant constant; by setting the background height to 0, the resulting φ0(x',y') implicitly contains the surface topography or object structure information of the sample.

2. The wafer defect detection system based on dual-channel optical imaging according to claim 1, wherein: The dark-field scattering imaging module comprises: an oblique incidence light source, an imaging unit, and a high-sensitivity detector; wherein: The oblique incidence light source is configured to provide a non-perpendicular light field; The imaging unit is configured to collect and image the light signals scattered from the sample surface; The high-sensitivity detector is configured to receive the dark-field scattering image; The diffraction phase microscopic imaging module comprises: a coherent light source, an imaging unit, a phase analysis unit, and a high-sensitivity detector; wherein: The coherent light source is configured to provide a coherent light field required for phase imaging; The imaging unit is configured to collect and image the light signals reflected from the sample surface; The phase analysis unit is configured to analyze the phase information of the defect and generate a diffraction phase image; The high-sensitivity detector is configured to record the interference image; The light path design and calibration ensure the light path compatibility and synchronization of the dark-field scattering imaging module and the diffraction phase microscopic imaging module; The image processing unit comprises: an image registration module and a phase demodulation module for the carrier frequency interference image; wherein: The image registration module is configured to spatially align the dark-field scattering image and the diffraction phase image; The phase demodulation module for the carrier frequency interference image is configured to perform phase recovery processing on the diffraction phase image, thereby extracting more detailed feature information of the sample.

3. The wafer defect detection system based on dual-channel optical imaging according to claim 2, wherein: The oblique incidence light source of the dark-field scattering imaging module is based on dark-field scattering detection technology, which irradiates the wafer surface with oblique incidence light, uses the scattering effect of surface micro defects or irregularities on the incident light, and collects scattering light signals at a specific angle to realize defect imaging; wherein the light path design of the oblique incidence light source makes the background light not participate in imaging, and only the defect scattering light is received by the detector, thereby significantly improving the image contrast and realizing high-sensitivity detection of micro defects; A semiconductor laser with a center wavelength of 405 nm is selected as the light source; the laser has a collimated output circular or elliptical beam, an output power of 50 mW, an integrated heat dissipation module, and good power stability and anti-interference performance.

4. The wafer defect detection system based on dual-channel optical imaging according to claim 2, wherein: The working content of the image registration module includes: Loading the initial image and performing geometric transformation operation according to the shooting characteristics of the image; Extracting feature points in the diffraction phase microscopic image and matching in the dark-field scattering image; Selecting a region of interest in the dark-field scattering image for further processing; Calculating the geometric transformation parameters between the images and accurately aligning the images; Displaying the registration result and performing precision verification.

5. The wafer defect detection system based on dual-channel optical imaging according to claim 2, wherein: In the phase demodulation module for the carrier frequency interference image, for the off-axis interference optical system, the interference image is represented as where a(x,y) is the background intensity of the interferogram, γ(x,y) is the amplitude modulation factor, is the phase information; ω x and ω y is the carrier frequency component; In off-axis interferograms with carrier frequency components, the interference fringes are highly dense due to the large phase-induced carrier frequency, and the phase variation at different positions will cause the fringe deformation; The required phase information is hidden in these dense fringes; from the overall point of view, the fringe density is caused by the carrier frequency, while the background intensity and amplitude modulation factor change slowly; thus, assuming that the carrier frequency components in the x and y directions are much larger than the background light intensity, amplitude modulation factor and phase gradient, the following conditions are met: The first-order and second-order partial derivatives of the interferogram in the y direction are calculated as: The wrapped phase of the object light wave is: Where unwrap represents the unwrapping function; off-axis interferometry separates the phase information of the interference fringes from the background intensity and amplitude modulation factor by making the interference fringes carrying phase information dense through the angle between the object light wave and the reference light; thus, even a single off-axis interferogram with carrier frequency can recover the phase information.

6. The dual-channel optical imaging based wafer defect detection system of claim 2, wherein: Since the dark-field scattering imaging and the diffraction phase microscopy obtain sample information from different physical angles, after completing the aforementioned image processing module, all information of the sample defects is comprehensively extracted by comparing and analyzing the dark-field scattering image and the diffraction phase image; the comparison and analysis process can combine the advantages of the two kinds of images, make up for the shortcomings of a single image mode, and improve the accuracy and robustness of defect detection.

7. The dual-channel optical imaging based wafer defect detection system of claim 2, wherein: The system is integrated by two optical detection systems, wherein the purple light beam represents the light path of the dark-field scattering system, and the blue light beam represents the light path of the diffraction phase microscopy system; The light source of the dark-field scattering system light path is a 405 nm wavelength laser, which first expands the beam through the Galileo telescope system composed of lenses L3 and L4, and then passes through dichroic mirror 1; the dichroic mirror is short-wave-pass, which can pass 405 nm wavelength light and reflect 455 nm wavelength light; after dichroic mirror 1, dichroic mirror 2 is set, which is long-wave-pass, allowing 455 nm wavelength light to pass through while reflecting 405 nm wavelength light, thereby separating the two beams of different wavelengths; The 405 nm wavelength light is reflected after passing through the dichroic mirror 2 and is obliquely incident on the sample through an adjustable angle mirror 1, and the sample surface scatters; the sample is placed on a sample stage, which is a displacement stage capable of controlling the displacement of the sample; the scattered light is then received by the objective lens and passes through mirror 2, then through dichroic mirror 3, which is short-wave-pass, allowing 405 nm wavelength light to pass through and reflecting 455 nm wavelength light; finally, the scattered light is imaged onto the camera CMOS1 through the tube lens L6 matched with the objective lens, realizing image acquisition; The light path of the diffraction phase microscopy system takes a 455 nm wavelength laser as a light source, is expanded and collimated through a polarizer and an expansion system composed of double cemented lenses L1 and L2, and a pinhole is further added between the lenses L1 and L2 to improve the spatial coherence of the light beam; the expanded light beam is first reflected by a dichroic mirror 1, then passes through a dichroic mirror 2, and is then focused to the focal point of an objective lens through a lens L5; in the process, a mirror 2 is used to fold the light path; the focused light beam enters the microscopic system through the objective lens, becomes parallel light, and is perpendicularly incident on the sample surface; after the objective lens collects the reflected light, the light is reflected by a mirror 2, then reflected by a dichroic mirror 3 to a tube lens L7 matched with the objective lens; a grating is arranged at the focal point of the L7, can produce multiple direction diffraction beams, so as to realize spatial replication of the light beam; then, the diffraction beams pass through a lens L8, the focal point position of which is aligned with the grating; Only the first-order diffraction light and the zero-order light are reserved through a spatial filter, and then pass through a lens L9; the lens L8 and the lens L9 together constitute a 4f system, which ensures that the light beam is properly magnified and spatially filtered; the first-order diffraction light and the zero-order light interfere with each other at this point, and an interference image is recorded on a camera CMOS2; in order to further fold the light path, two mirrors 3 and 4 are further arranged between the spatial filter and the camera CMOS2, which ensures the optimization of the light path and the imaging effect.

Images