A multi-light-source-like wafer defect detection system and method

Abstract

The application discloses a kind of multi-light source-like wafer defect detection system and method, it is related to wafer defect detection technical field, including: light source module;Light modulation module is set in the light path of multiple illumination light beams, the light modulation module is used to respectively apply each road illumination light beam with different spatial frequency and / or different direction carrier modulation, to convert each road illumination light beam into multiple structured light and with different preset incident angle irradiation to the detection area on the wafer surface to be measured;Imaging acquisition module is used to receive the detection light signal generated from the detection area different structured light and obtain first detection image;Signal processing module is connected with the imaging acquisition module communication connection.The above scheme is used, while improving detection efficiency, can fundamentally avoid the light path crosstalk generated by multi-light path detection.

Description

Technical Field

[0001] This invention relates to the field of wafer defect detection technology, and in particular to a multi-source wafer defect detection system and method. Background Technology

[0002] As semiconductor manufacturing processes continue to evolve towards smaller nodes, the microstructure of wafer surfaces is becoming increasingly complex. High aspect ratio structures, low-contrast defects, and various morphologies of flaws pose significant challenges to optical inspection technologies. To address these challenges, modern inspection equipment widely employs multi-source illumination techniques, such as multi-angle ring lights and multi-wavelength spectral light sources. This utilizes the differentiated sensitivity of light with different incident angles and wavelengths to specific defect morphologies to achieve comprehensive defect detection.

[0003] However, while pursuing high imaging quality, existing multi-light source implementation methods face bottlenecks in detection efficiency. Currently, mainstream multi-light source detection technologies mostly employ a time-division multiplexing strategy, controlling different angles or types of light sources to flash alternately, and coordinating with a camera for multiple exposures. While this alternating operation mode avoids direct interference between optical paths to some extent, it inevitably leads to a significant increase in the imaging cycle, severely limiting the throughput of the detection equipment and failing to meet the urgent needs of modern wafer fabs for high-speed online inspection. Attempting to use a scheme that triggers multiple light sources simultaneously to improve speed not only places extremely high technical requirements on the sensitivity of the image sensor and the intensity of the illumination, but more problematic is that the complex reflections and scattering generated by multiple light sources on the wafer surface easily cause optical path crosstalk, resulting in blurred images or artifacts, severely affecting the accuracy of defect identification.

[0004] Therefore, in the current field of wafer defect detection, how to improve detection efficiency while ensuring the high detection coverage and high imaging quality brought by combined illumination, and at the same time fundamentally avoid the optical path crosstalk problem that is very easy to occur when multiple optical paths are imaged simultaneously. Summary of the Invention

[0005] The purpose of this invention is to provide a multi-source wafer defect detection system and method to solve the technical problem in the prior art of how to improve detection efficiency while ensuring high detection coverage and high imaging quality brought about by combined illumination, and at the same time fundamentally avoiding optical path crosstalk that is easily generated when multiple optical paths are imaged simultaneously.

[0006] To achieve the above objectives, the technical solution of the present invention is as follows: Firstly, a multi-source wafer defect detection system is provided, comprising: The light source module is used to provide multiple independent illumination beams; An optical modulation module is disposed in the optical path of the multiple illumination beams. The optical modulation module is used to apply carrier modulation with different spatial frequencies and / or different directions to each illumination beam to convert each illumination beam into multiple structured light and irradiate the detection area on the surface of the wafer under test with different preset incident angles. The imaging acquisition module is used to receive detection light signals generated by different structured lights from the detection area and acquire a first detection image; The signal processing module is communicatively connected to the imaging acquisition module; The signal processing module is used to perform frequency domain transformation on the first detection image, and according to the different spatial frequencies and / or different directions, separate independent frequency band signals corresponding to each preset incident angle in the frequency domain, and demodulate to generate different second detection images.

[0007] Furthermore, this includes bright-field detection mode and dark-field detection mode, wherein: In the dark field detection mode, the imaging acquisition module is positioned above the wafer to be inspected to receive scattered light. In the bright field detection mode, the imaging acquisition module includes multiple sub-detection acquisition units, each corresponding to a different preset incident angle setting, for receiving the detection light signal after each path of structured light is reflected from the surface of the wafer under test.

[0008] Furthermore, the light source module includes a broadband light source and a beam splitter group. The beam splitter group is disposed in the output light path of the broadband light source and is used to split the light beam emitted by the broadband light source into multiple independent illumination beams. The beam splitter group includes at least one beam splitter prism or beam splitter lens.

[0009] Furthermore, the optical modulation module includes multiple transmissive gratings, which are respectively disposed in the optical paths of each illumination beam output by the beam splitter group. Each transmissive grating has a different grating period and / or grating direction, and is used to apply carrier modulation with different spatial frequencies and / or different directions to each illumination beam.

[0010] Furthermore, the beam splitter group splits the beam emitted by the broadband light source into four independent illumination beams; the optical modulation module includes four transmissive gratings, the grating directions of the four transmissive gratings are arranged in pairs orthogonal, so that the carriers corresponding to each structured light are distributed in mutually orthogonal directions in the frequency domain.

[0011] Furthermore, after performing frequency domain transformation on the first detection image, the signal processing module obtains four carrier signal regions in the frequency domain plane. The four carrier signal regions are respectively located in a direction perpendicular to the grating stripe direction of each of the transmissive gratings. The signal processing module performs bandpass filtering and inverse Fourier transform on the four carrier signal regions respectively to demodulate and generate four second detection images.

[0012] Furthermore, a relay module is provided in the imaging optical path of the imaging acquisition module, and a dynamic spatial light modulator is provided at the Fourier plane of the relay module; the dynamic spatial light modulator is used to load a physical mask to physically block the background diffraction noise generated by the periodic texture of the wafer surface before photoelectric conversion, while allowing the defect scattered light signal and the carrier signal of the structured light to pass through.

[0013] Furthermore, the signal processing module also includes a deep learning unit, which is used to generate a background diffraction spectrum prediction map based on the process parameters of the wafer under test, and then invert the background diffraction spectrum prediction map and load it onto the dynamic spatial light modulator to form the physical mask.

[0014] Secondly, a multi-source wafer defect detection method is provided, including the following steps: Provides multiple independent illumination beams; Each illumination beam is modulated with a carrier wave of different spatial frequency and / or different direction to convert the illumination beam into multi-path structured light; Control the structured light beams to illuminate the detection area on the surface of the wafer under test at different preset incident angles; Receive detection light signals generated by different structured lights from the detection area to obtain a first detection image; The first detection image is subjected to frequency domain transformation. Based on the different spatial frequencies and / or different directions, independent frequency band signals corresponding to each preset incident angle are separated in the frequency domain and demodulated to generate different second detection images.

[0015] Furthermore, before receiving the detection light signals generated by different structured light sources from the detection area, the following steps are also included: A relay module is set in the imaging optical path, and a dynamic spatial light modulator is set at the Fourier plane of the relay module. A physical mask is applied to the dynamic spatial light modulator to physically block the light beam carrying background diffraction noise generated by the periodic texture of the wafer surface before photoelectric conversion, while allowing the defect scattered light signal and the carrier signal of the structured light to pass through.

[0016] Further, loading a physical mask onto the dynamic spatial light modulator includes the following steps: Obtain the process parameters of the wafer under test; The process parameters are input into a deep learning model to generate a background diffraction spectrum prediction map. Invert the background diffraction spectrum prediction image to generate a physical masking pattern; The physical mask pattern is loaded onto the dynamic spatial light modulator.

[0017] The present invention provides a multi-source wafer defect detection system and method, which, compared with the prior art, has the following core advantages: By applying carrier modulation at different spatial frequencies and / or in different directions to multiple illumination beams using an optical modulation module, each illumination beam is endowed with distinguishable frequency domain coding characteristics before entering the detection area. This allows for simultaneous illumination from multiple light sources without the time-division multiplexing method used in previous techniques, building upon the provision of multiple independent illumination beams by the light source module. Furthermore, the imaging acquisition module receives and superimposes a first detection image containing information from multiple structured light sources under single or few exposure conditions, significantly reducing the number of exposures and improving detection efficiency. Simultaneously, the signal processing module performs frequency modulation on the first detection image... The system performs domain transformation and separates independent frequency band signals corresponding to each preset incident angle in the frequency domain according to the different spatial frequencies and / or different directions. By utilizing the orthogonal or separable characteristics of each light path in the frequency domain, it achieves effective demodulation of multiple signals and then reconstructs the corresponding second detection images through demodulation. Thus, it not only achieves parallel detection of multiple light sources to improve throughput while maintaining the high detection coverage and high imaging quality brought by multi-angle illumination, but also fundamentally avoids the crosstalk problem caused by the simultaneous action of multiple light paths from the signal processing level, avoiding imaging blur and artifacts, thereby balancing detection efficiency and imaging quality. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of a multi-source wafer defect detection system according to the present invention; Figure 2 This is a top-view diagram showing the distribution of the imaging acquisition module and the optical modulation module in bright-field detection mode according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the beam-splitting lens assembly according to an embodiment of the present invention; Figure 4 This is an optical path diagram of the relay module in an embodiment of the present invention; Figure 5 This is a schematic diagram of the frequency domain spatial carrier distribution according to an embodiment of the present invention; Figure 6 This is a schematic diagram illustrating the generation of a second detection image from a first detection image according to an embodiment of the present invention; Figure 7 This is a flowchart of a multi-source wafer defect detection method according to an embodiment of the present invention; Reference numerals: 1. Light source module; 11. Broadband light source; 12. Beam splitter group; 2. Optical modulation module; 21. Transmissive grating; 3. Imaging acquisition module; 31. Sub-detector acquisition unit; 4. Relay module; 41. Objective lens; 42. First Fourier lens; 43. Dynamic spatial light modulator; 44. Second Fourier lens; 5. Wafer surface; 6. Physical mask. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Unless otherwise defined, the technical or scientific terms used herein should have the ordinary meaning understood by those skilled in the art. The terms "comprising" and similar expressions used herein mean that the element or object preceding the word covers the element or object listed after the word and its equivalents, but does not exclude other elements or objects.

[0020] Combined with appendix Figure 1 - Appendix Figure 7 The specific embodiments of the present invention will be further described in detail below.

[0021] Firstly, in some embodiments of the present invention, addressing the low detection throughput caused by time-division multiplexing of multiple light sources in the prior art, and the optical path crosstalk and imaging blurring problems caused by forced simultaneous triggering, this disclosure proposes a multi-light source wafer defect detection system. Overall, the system mainly includes a light source module 1, an optical modulation module 2, an imaging acquisition module 3, and a signal processing module.

[0022] Reference Figure 1The light source module 1 is the illumination starting point of the entire system, used to stably provide multiple independent illumination beams. The light modulation module 2 is located on the transmission optical path of the multiple illumination beams, and its main function is to apply carrier modulation with different spatial frequencies and / or different directions to each illumination beam. After modulation, each conventional illumination beam is converted into multiple high-frequency structured lights carrying specific spatial spectrum characteristics, and simultaneously illuminates the detection area of ​​the wafer surface 5 under test at different preset incident angles. Subsequently, the imaging acquisition module 3 receives the detection light signals generated by the interaction of different structured lights with the wafer in the detection area, and acquires the first detection image under exposure. Due to the introduction of specific carrier modulation in the physical optical path, the signal processing module, which is communicatively connected to the imaging acquisition module 3, can perform frequency domain transformation on the first detection image. Based on the preset prior information of different spatial frequencies and / or different directions, the signal processing module accurately separates the independent frequency band signals corresponding to each preset incident angle in the frequency domain space, and finally demodulates to generate different second detection images. The above technical solution breaks through the time bottleneck of time-sharing exposure, and at the same time, it fundamentally eliminates multi-path crosstalk in the spatial domain by utilizing frequency domain orthogonality.

[0023] Reference Figure 3 In some specific embodiments of the present invention, to ensure the spectral consistency and energy balance of multiple beams, the light source module 1 specifically includes a broadband light source 11 and a beam splitter group 12. The light source module 1 is a broadband light source 11, and the beam splitter group 12 is disposed in the output optical path of the broadband light source 11. It includes at least one beam splitter prism or beam splitter lens, and mirrors at different angles, to efficiently split the broadband beam emitted by a single light source into multiple physically independent illumination beams. Immediately following beam splitting, the optical modulation module 2 employs multiple transmissive gratings 21, which are respectively disposed in the optical paths of each illumination beam output from the beam splitter group 12. To meet the requirement of frequency domain separation, each transmissive grating 21 is designed with a different grating period, which determines the spatial frequency and / or different grating directions, and determines the spectral distribution axis. For example, in a preferred four-way combined illumination embodiment, the beam splitter group 12 equally divides the beam emitted by the broadband light source 11 into four independent illumination beams. Correspondingly, the optical modulation module 2 includes four transmissive gratings 21, and the grating stripe directions of these four transmissive gratings 21 are ingeniously designed to be arranged in pairs orthogonal. This orthogonal arrangement ensures that after each structured light is projected onto the wafer surface 5, its corresponding carrier is strictly distributed along mutually orthogonal directions in the frequency domain space, maximizing the physical distance between the signals of each channel in the frequency domain and avoiding spectral aliasing.

[0024] Reference Figure 5 and Figure 6In some specific embodiments of the present invention, for example, in Figure 5, the acquired first detection image, after undergoing a two-dimensional fast Fourier transform (2D-FFT), exhibits a feature distribution in the frequency domain with four sets of symmetrical spike signals, labeled as follows: Among them, the horizontal axis and vertical axis Represents spatial frequency. The center point is the DC component. These four sets of signals correspond to grating modulation in four different directions in the embodiment. The first grating direction is set to 0°, with a spatial frequency of f1; the second grating direction is set to 45°, with a spatial frequency of f2; the third grating direction is set to 135°, with a spatial frequency of f3; and the fourth grating direction is set to 180°, with a spatial frequency of f4. The distribution pattern in the figure proves that the four illumination beams are completely orthogonal and do not overlap in the frequency domain. The signal processing module only needs to set bandpass filters at these specific frequency domain coordinates to extract detection information from a mixed image at four different angles without generating optical path crosstalk. In order to cooperate with the above hardware modulation structure, after the signal processing module acquires the first detection image, it performs frequency domain transformation, such as fast Fourier transform, and can clearly obtain four isolated carrier signal regions in the two-dimensional frequency domain plane. According to the Fourier optical principle, the positions of these four carrier signal regions are located on the direction axis perpendicular to the grating fringe direction of each transmissive grating 21. After accurately locating these regions, refer to Figure 6 ,For example Figure 6 The leftmost column shows the periodic striped background formed by the grating, and the waveform below it shows the high-frequency oscillation characteristics after carrier modulation. The second and third columns show different types of defect signals, such as gradual background changes or sudden defect peaks combined with the carrier wave to change the envelope. The top row represents the original light intensity distribution or carrier envelope under ideal conditions. The bottom row represents the actual received signal carrying defect features after reflection / scattering from the wafer surface. It can be seen that the defect information causes small perturbations in the carrier phase or amplitude. The fourth column shows the modulation effect of edge features.

[0025] Furthermore, the signal processing module applies customized bandpass filtering to each of the four carrier signal regions to remove out-of-band noise and ambient stray light. Finally, it performs inverse Fourier transform on each of the filtered independent frequency band signals to demodulate and generate four high-resolution second detection images. These four images are equivalent to the detection results when illuminated by four different light sources at different angles, but the imaging time is only one-quarter of that of the traditional single-channel alternating shooting mode.

[0026] Reference Figure 1 and Figure 2In some embodiments of the present invention, considering the varying sensitivities of different wafer defects, such as particle contamination, scratches, or pattern defects, to light scattering and reflection characteristics, the system is configured to simultaneously support both bright-field and dark-field detection modes. In dark-field detection mode, the system primarily captures the weak scattered light caused by minute defects on the wafer surface 5. Therefore, the main sensor of the imaging acquisition module 3 is positioned directly above the wafer under test for reception. Conversely, in bright-field detection mode, the system focuses on acquiring the macroscopic morphology and reflectivity changes of the wafer surface 5. In this case, the structure of the imaging acquisition module 3 is expanded to include multiple sub-detection acquisition units 31. These sub-detection acquisition units 31 are spatially positioned at different preset incident angles to receive the high-intensity detection light signals after each path of structured light is directly reflected from the wafer surface 5. Therefore, in dark-field detection mode, the imaging acquisition module 3 only needs to acquire a first detection image containing all modulation carrier characteristics in a single exposure. The signal processing module then uses frequency domain transformation to accurately separate four independent frequency band signals corresponding to each preset incident angle from the single first detection image, and finally demodulates and generates four independent second detection images. This method significantly improves the efficiency of dark-field detection while solving the signal aliasing problem under simultaneous illumination by multiple optical paths through frequency domain orthogonality.

[0027] Reference Figure 6 In some specific embodiments of the present invention, in the bright-field detection mode, the system focuses on acquiring the macroscopic morphology and reflectivity changes of the wafer surface 5. Since the bright-field signal mainly follows the specular reflection law, its energy is highly concentrated in the reflection direction corresponding to each optical path. Therefore, the imaging acquisition module 3 is expanded to include four sub-detection acquisition units 31. These four sub-detection acquisition units 31 are spatially positioned to strictly correspond to different preset incident angle settings, and are used to receive the high-intensity detection light signals after each path of structured light is reflected by the wafer surface 5 under test. In this mode, the imaging acquisition module 3 will simultaneously acquire four independent first detection images. In response to the physical optical path crosstalk that may exist between the sub-detection acquisition units 31 under bright-field detection, for example, abnormal deflection or large-angle scattering of light caused by the complex texture of the wafer surface 5 entering the non-corresponding acquisition unit, the signal processing module will perform frequency domain transformation on each first detection image. Since each incident light carries a specific spatial frequency or directional carrier, the signal processing module can extract the target signal corresponding to the current acquisition unit through frequency domain expression and filter out interference features from other incident optical paths. In this way, the signal processing module demodulates the four first detection images respectively, and finally generates four high-purity second detection images that have eliminated optical crosstalk.

[0028] Specifically, by differentiating the quantity relationship between the first and second detection images in bright and dark field modes, the system can flexibly adapt to different detection scenarios: the dark field mode maximizes throughput by utilizing multiple solutions per image, while the bright field mode ensures absolute image fidelity in high-speed detection through multi-image filtering. This approach fundamentally overcomes the technical challenge of balancing efficiency and signal-to-noise ratio in high-speed wafer inspection.

[0029] In some other embodiments of the present invention, the advanced process wafer surface 5, due to its high-density periodic wiring, such as memory array textures, generates strong background diffraction noise under strong light. Therefore, a relay module 4 is incorporated into the imaging optical path of the imaging acquisition module 3. A dynamic spatial light modulator 43 is installed at the optical Fourier plane, i.e., the spectral plane, of the relay module 4. The dynamic spatial light modulator 43 is primarily used to load a physical mask 6, directly and physically blocking the high-energy background diffraction spots generated by the periodic textures of the wafer surface 5 before the optical signal undergoes photoelectric conversion, while allowing random defect-scattered light signals and the aforementioned applied structured light carrier signal to pass smoothly.

[0030] Specifically, the relay module 4 is located between the wafer surface 5 and the imaging acquisition module 3. The relay module 4 includes: an objective lens 41 for collecting scattered light and background diffracted light from the wafer surface 5; a first Fourier lens 42 for converting the light field distribution on the wafer surface 5, i.e., the object plane, into a frequency domain distribution; a dynamic spatial light modulator 43 placed on the back focal plane of the first Fourier lens 42, i.e., the spectral plane; and a second Fourier lens 44 disposed after the dynamic spatial light modulator 43 for performing an inverse Fourier transform to re-image the filtered light field onto the imaging acquisition module 3.

[0031] In some specific embodiments of the present invention, in order to achieve accurate and dynamic generation of the physical mask 6, a deep learning unit is also deeply integrated within the signal processing module. This deep learning unit utilizes a pre-trained neural network model to quickly infer and generate a background diffraction spectrum prediction map corresponding to the height of the wafer pattern based on the process parameters of the wafer under test, such as linewidth, pitch, and stacked materials. The system then logically inverts the background diffraction spectrum prediction map, setting the predicted noise-high-brightness areas as opaque blocking areas and the remaining areas as transparent areas, and converts this into a driving signal loaded onto the dynamic spatial light modulator 43, thereby forming an adaptive physical mask 6. By employing this scheme, the system can maintain an extremely high signal-to-noise ratio and defect detection rate even against backgrounds with extremely complex textures.

[0032] Specifically, the deep learning unit employs a generative adversarial network (GAN) architecture to achieve accurate mapping from low-level process parameters to high-dimensional spectral images. Specifically, the model uses structured parameters such as the film thickness, pattern period, and material refractive index of the wafer under test as input vectors, and utilizes a generator network to reconstruct the corresponding two-dimensional background diffraction spectrum distribution within the latent space. Because GANs possess extremely strong nonlinear fitting capabilities in processing structured data to generate high-resolution images, and have mature applications in optical reverse engineering, they can ensure that the generated spectral prediction map highly matches the actual diffraction spot in both spatial distribution and intensity.

[0033] In some specific embodiments of the present invention, to ensure the prediction accuracy and real-time performance of the deep learning unit, its training dataset can be constructed by combining electromagnetic simulation calculations with experimental measurement data. On the one hand, numerical algorithms such as the finite difference time-domain method or rigorous coupled-wave analysis are used to simulate the theoretical diffraction spectrum under different combinations of process parameters; on the other hand, measurement data of known defect-free wafers under actual optical systems are collected to fine-tune the model. Once the model training is complete, its inference process requires only millisecond-level computation.

[0034] In some specific embodiments of the present invention, the generated background diffraction spectrum prediction map is converted into an inverse diffraction mask control signal by logical inversion. Since the background noise generated by the periodic texture of the wafer surface 5 exhibits extremely high-energy narrowband coherent diffraction orders in the frequency domain, with a discrete point distribution, while randomly distributed defects, such as particles or scratches, generate broadband and incoherent scattering signals, showing continuous and widespread distribution characteristics in the frequency domain, the physical mask 6 loaded by the dynamic spatial light modulator 43 can accurately intercept these discrete high-brightness noise points, while most of the defect scattering signals distributed in the mask gaps and the structured light carrier signals carrying modulation characteristics can pass through smoothly, significantly improving the system's ultimate detection sensitivity.

[0035] In some specific embodiments of the present invention, the dynamic spatial light modulator 43 can be a high-resolution liquid crystal-based spatial light modulator or a MEMS micromirror array. These devices support high spatial resolution modulation of 1k×1k pixels or higher, and the refresh rate is typically above 100Hz, sufficient to support the real-time switching requirements of scanning detection equipment. The specific workflow is as follows: first, process parameters are input; then, GANs generate a spectral prediction map in real time; after inversion, it is loaded onto the dynamic spatial light modulator located in the Fourier plane via a driving circuit; finally, the optical signal is optically filtered, and the imaging acquisition module acquires a clear image with a high signal-to-noise ratio. In some other embodiments of the present invention, in addition to using GANs to generate the prediction map, the system can also pre-establish a spectral feature library for standard processes. During detection, the signal processing module can quickly coarsely adjust the mask pattern by rapidly searching the feature library and using simple affine transformations, followed by fine compensation by the deep learning unit.

[0036] In some other embodiments of the present invention, those skilled in the art can replace the broadband light source 11 in the light source module 1 with a multi-wavelength laser array according to specific detection requirements to further improve channel isolation; the static transmissive grating in the optical modulation module 2 can also be replaced with a reflective digital micromirror device (DMD) to achieve spatial carrier frequency reconstruction. In addition, in addition to the classic bandpass filtering, the demodulation algorithm in the signal processing module can also use an end-to-end deep convolutional neural network to directly regress the second detection image from the first detection image from multiple perspectives, in order to further reduce the computational delay of image reconstruction and eliminate the edge ringing effect caused by frequency domain truncation.

[0037] Secondly, referring to Figure 7 A multi-source wafer defect detection method is provided, comprising the following steps: Step 101: Provide multiple independent illumination beams; Step 102: Apply carrier modulation with different spatial frequencies and / or different directions to each illumination beam to convert each illumination beam into multi-path structured light; Step 103: Control the structured light beams to illuminate the detection area on the surface 5 of the wafer under test at different preset incident angles; Step 104: Receive detection light signals generated by different structured lights from the detection area to obtain a first detection image; Steps: Perform frequency domain transformation on the first detection image, and separate independent frequency band signals corresponding to each preset incident angle in the frequency domain according to the different spatial frequencies and / or different directions, and demodulate to generate different second detection images.

[0038] Furthermore, before receiving the detection light signals generated by different structured light sources from the detection area, the following steps are also included: A relay module 4 is set in the imaging optical path, and a dynamic spatial light modulator 43 is set at the Fourier plane of the relay module 4. A physical mask 6 is loaded onto the dynamic spatial light modulator 43 to physically block the light beam carrying background diffraction noise generated by the periodic texture of the wafer surface 5 before photoelectric conversion, while allowing the defect scattered light signal and the carrier signal of the structured light to pass through.

[0039] Further, loading the physical mask 6 onto the dynamic spatial light modulator 43 includes the following steps: Obtain the process parameters of the wafer under test; The process parameters are input into a deep learning model to generate a background diffraction spectrum prediction map. Invert the background diffraction spectrum prediction image to generate the physical mask 6 pattern; The physical mask 6 pattern is loaded onto the dynamic spatial light modulator 43.

[0040] By adopting the above scheme, although the spatial domain of the simultaneous illumination by multiple structured lights presents an overlapping composite image, in the frequency domain, the information from each path is distributed in an isolated region that does not interfere with each other due to the orthogonality of the carriers. This achieves synchronous acquisition of multi-view images under a single exposure, completely eliminating optical path crosstalk, ensuring the imaging signal-to-noise ratio, and improving detection efficiency. Furthermore, strong illumination and structured light modulation can excite strong background diffraction noise from the high-density periodic wiring of advanced process wafers. This noise easily drowns out weak defect scattering signals. Therefore, this scheme embeds an F-relay in the imaging optical path and uses a generative adversarial network to predict the diffraction spectrum distribution of the wafer surface 5. A dynamic spatial light modulator 43 is set in the Fourier plane, and an adaptive physical mask 6 generated based on the inverse of the predicted image is loaded. Since the periodic texture noise is distributed as discrete points in the frequency domain, while the random defect signal and structured light carrier exhibit a wide-area distribution, the dynamic spatial light modulator 43 can physically block the background noise before photoelectric conversion. Moreover, traditional multi-source detection often suffers from image registration difficulties. Images captured at different times, angles, or lenses often require complex algorithms for alignment due to minute mechanical vibrations, shutter delays, or optical axis deviations. In this solution, raw information from multiple viewpoints is captured by the same camera within a very short time. Mathematically, this shared, identical coordinate system eliminates registration errors caused by time differences or mechanical displacements, significantly reducing the algorithmic complexity of subsequent defect fusion and assessment, and ensuring the accuracy of defect localization.

[0041] In the several embodiments provided in this application, it should be understood that the disclosed methods and apparatus can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0042] Furthermore, the signal processing modules in the various embodiments of this application can be integrated into one processing unit, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in a combination of hardware and software functional units.

[0043] The integrated units implemented as software functional units described above can be stored in a computer-readable storage medium. These software functional units, stored in a storage medium, include several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute some steps of the transmission and reception methods described in the various embodiments of this application. The aforementioned storage medium can be a volatile or non-volatile computer-readable storage medium, including: USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, optical disks, and other media capable of storing program code.

[0044] It is understood that the signal processing modules in the embodiments described in this application can be implemented using hardware, software, firmware, middleware, microcode, or a combination thereof. For hardware implementation, modules, units, submodules, subunits, etc., can be implemented in one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), general-purpose processors, controllers, microcontrollers, microprocessors, other electronic units for performing the functions described in this application, or combinations thereof.

[0045] The above descriptions are some embodiments of this application. It should be noted that, for those skilled in the art, several improvements and modifications can be made without departing from the principles described in this application, and these improvements and modifications should also be considered within the scope of protection of this application.

[0046] For details of this embodiment of the application, please refer to the description of the foregoing method embodiments. To avoid repetition, the description will not be repeated here.

[0047] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0048] While embodiments of the present invention have been described in detail above, it will be apparent to those skilled in the art that various modifications and variations can be made to these embodiments. However, it should be understood that such modifications and variations fall within the scope and spirit of the present invention. Furthermore, the present invention described herein may have other embodiments and can be implemented or carried out in various ways.

Claims

1. A multi-source wafer defect detection system, characterized in that, include: The light source module is used to provide multiple independent illumination beams; An optical modulation module is disposed in the optical path of the multiple illumination beams. The optical modulation module is used to apply carrier modulation with different spatial frequencies and / or different directions to each illumination beam to convert each illumination beam into multiple structured light and irradiate the detection area on the surface of the wafer under test with different preset incident angles. The imaging acquisition module is used to receive detection light signals generated by different structured lights from the detection area and acquire a first detection image; The signal processing module is communicatively connected to the imaging acquisition module; The signal processing module is used to perform frequency domain transformation on the first detection image, and according to the different spatial frequencies and / or different directions, separate independent frequency band signals corresponding to each preset incident angle in the frequency domain, and demodulate to generate different second detection images.

2. The multi-source wafer defect detection system according to claim 1, characterized in that, This includes bright field detection mode and dark field detection mode, among which: In the dark field detection mode, the imaging acquisition module is positioned above the wafer to be inspected to receive scattered light. In the bright field detection mode, the imaging acquisition module includes multiple sub-detection acquisition units, each corresponding to a different preset incident angle setting, for receiving the detection light signal after each path of structured light is reflected from the surface of the wafer under test.

3. The multi-source wafer defect detection system according to claim 1, characterized in that, The light source module includes a broadband light source and a beam splitter group. The beam splitter group is disposed in the output light path of the broadband light source and is used to split the light beam emitted by the broadband light source into multiple independent illumination beams. The beam splitter group includes at least one beam splitter prism or beam splitter lens.

4. The multi-source wafer defect detection system according to claim 3, characterized in that, The optical modulation module includes multiple transmissive gratings, which are respectively disposed in the optical paths of each illumination beam output by the beam splitter group. Each transmissive grating has a different grating period and / or grating direction, and is used to apply carrier modulation with different spatial frequencies and / or different directions to each illumination beam.

5. The multi-source wafer defect detection system according to claim 3, characterized in that, The beam splitter array splits the beam emitted by the broadband light source into four independent illumination beams; the optical modulation module includes four transmissive gratings, the grating directions of the four transmissive gratings are arranged in pairs orthogonal, so that the carriers corresponding to each structured light are distributed in mutually orthogonal directions in the frequency domain.

6. The multi-source wafer defect detection system according to claim 1, characterized in that, After performing frequency domain transformation on the first detection image, the signal processing module obtains four carrier signal regions in the frequency domain plane. The four carrier signal regions are located in directions perpendicular to the grating stripe direction of each of the transmissive gratings. The signal processing module performs bandpass filtering and inverse Fourier transform on the four carrier signal regions respectively to demodulate and generate four second detection images.

7. The multi-source wafer defect detection system according to claim 2, characterized in that, The imaging acquisition module has a relay module in its imaging optical path, and a dynamic spatial light modulator is provided at the Fourier plane of the relay module. The dynamic spatial light modulator is used to load a physical mask to physically block the background diffraction noise generated by the periodic texture of the wafer surface before photoelectric conversion, while allowing the defect scattered light signal and the carrier signal of the structured light to pass through.

8. The multi-source wafer defect detection system according to claim 7, characterized in that, The signal processing module further includes a deep learning unit, which generates a background diffraction spectrum prediction map based on the process parameters of the wafer under test, and inverts the background diffraction spectrum prediction map and loads it onto the dynamic spatial light modulator to form the physical mask.

9. A method for detecting wafer defects using a multi-source light source, characterized in that, Includes the following steps: Provides multiple independent illumination beams; Each illumination beam is modulated with a carrier wave of different spatial frequency and / or different direction to convert the illumination beam into multi-path structured light; Control the structured light beams to illuminate the detection area on the surface of the wafer under test at different preset incident angles; Receive detection light signals generated by different structured lights from the detection area to obtain a first detection image; The first detection image is subjected to frequency domain transformation. Based on the different spatial frequencies and / or different directions, independent frequency band signals corresponding to each preset incident angle are separated in the frequency domain and demodulated to generate different second detection images.

10. The method for detecting wafer defects using a multi-source light source according to claim 9, characterized in that, Before receiving the detection light signals generated by different structured light sources from the detection area, the following steps are also included: A relay module is set in the imaging optical path, and a dynamic spatial light modulator is set at the Fourier plane of the relay module. A physical mask is applied to the dynamic spatial light modulator to physically block the light beam carrying background diffraction noise generated by the periodic texture of the wafer surface before photoelectric conversion, while allowing the defect scattered light signal and the carrier signal of the structured light to pass through.

11. The method for detecting wafer defects using a multi-source light source according to claim 10, characterized in that, The process of loading a physical mask onto the dynamic spatial light modulator includes the following steps: Obtain the process parameters of the wafer under test; The process parameters are input into a deep learning model to generate a background diffraction spectrum prediction map. Invert the background diffraction spectrum prediction image to generate a physical masking pattern; The physical mask pattern is loaded onto the dynamic spatial light modulator.

Images