Photoetching machine calibration method, device and equipment based on multi-view vision

Abstract

The invention relates to the technical field of photoetching machine calibration, and discloses a photoetching machine calibration method, device and equipment based on multi-view vision, and the method comprises the following steps: carrying out imaging and phase sensitive detection analysis on a mask plane and a wafer plane through a four-path optical beam splitting system to obtain four groups of mask-wafer initial alignment position information; performing scanning white light interference edge enhancement processing to obtain three-dimensional surface contour data; performing wavelet transform processing on the three-dimensional surface contour data, and performing dynamic registration on the four-path optical beam splitting system to obtain multi-view visual feature registration data; a comprehensive error model including mechanical errors, optical errors and environmental errors is established, real-time correction and error compensation are carried out on a six-degree-of-freedom motion platform of the photoetching machine, a multi-view vision calibration result of the photoetching machine is obtained, the influence of environmental vibration and thermal drift on calibration precision is effectively eliminated, and the calibration accuracy of the photoetching machine is improved. The optical system is ensured to be always in the optimal imaging state, and the calibration precision is improved.

Description

Technical Field

[0001] This application relates to the technical field of lithography machine calibration, and particularly relates to a lithography machine calibration method, device and equipment based on multi - vision. Background Technique

[0002] Traditional lithography machine calibration methods generally use monocular vision systems and rely on specific alignment marks for overlay accuracy control. This method faces physical constraints of the diffraction limit during nano - scale processing and cannot meet the accuracy requirements of advanced processes. In addition, traditional calibration techniques are mostly offline calibrations, resulting in a large difference between the calibration state and the actual exposure state. As the process progresses, the overlay error accumulates continuously, ultimately affecting the chip yield.

[0003] In extreme ultraviolet lithography (EUV) systems, the influence of environmental factors on calibration accuracy is significantly enhanced. Mechanical vibration, thermal drift, and air - flow disturbance can all cause a significant decrease in calibration accuracy. Traditional through - the - lens alignment (TTL) and through - the - grating alignment (TTR) technologies show great instability when facing these environmental disturbances and are difficult to provide continuous and stable high - precision calibration results. At the same time, when traditional technologies are used for calibration on complex wafers with multi - layer structures, due to material differences and optical property changes between layers, systematic errors are likely to occur, affecting the final overlay accuracy. Summary of the Invention

[0004] This application provides a lithography machine calibration method, device and equipment based on multi - vision, thereby effectively eliminating the influence of environmental vibration and thermal drift on calibration accuracy, ensuring that the optical system is always in the best imaging state, and improving the calibration accuracy.

[0005] In the first aspect of this application, a lithography machine calibration method based on multi - vision is provided. The lithography machine calibration method based on multi - vision includes: Performing imaging and phase - sensitive detection and analysis on the mask plane and the wafer plane through a four - way optical beam - splitting system to obtain four groups of mask - wafer initial alignment position information; Performing scanning white - light interference edge enhancement processing on the four groups of mask - wafer initial alignment position information to obtain three - dimensional surface profile data; Performing wavelet transform processing on the three - dimensional surface profile data to extract target feature points, and at the same time dynamically registering the four - way optical beam - splitting system according to the target feature points to obtain multi - vision feature registration data; Establishing a comprehensive error model including mechanical error, optical error, and environmental error based on the multi - vision feature registration data, and performing real - time correction and error compensation on the six - degree - of - freedom motion platform of the lithography machine based on the comprehensive error model to obtain the multi - vision calibration result of the lithography machine.

[0006] The second aspect of the present application provides a lithography machine calibration device based on multi - vision, and the lithography machine calibration device based on multi - vision includes: An imaging module, configured to perform imaging and phase - sensitive detection and analysis on the mask plane and the wafer plane through a four - way optical beam splitting system, and obtain four sets of mask - wafer initial alignment position information; A processing module, configured to perform scanning white - light interference edge enhancement processing on the four sets of mask - wafer initial alignment position information to obtain three - dimensional surface profile data; A dynamic registration module, configured to perform wavelet transform processing on the three - dimensional surface profile data, extract target feature points, and simultaneously perform dynamic registration on the four - way optical beam splitting system according to the target feature points to obtain multi - vision feature registration data; An error compensation module, configured to establish a comprehensive error model including mechanical error, optical error, and environmental error based on the multi - vision feature registration data, and perform real - time correction and error compensation on the six - degree - of - freedom motion platform of the lithography machine based on the comprehensive error model to obtain the multi - vision calibration result of the lithography machine.

[0007] The third aspect of the present application provides an electronic device, including: a memory and at least one processor, wherein instructions are stored in the memory; the at least one processor invokes the instructions in the memory so that the electronic device executes the above - mentioned lithography machine calibration method based on multi - vision.

[0008] Compared with the prior art, the present application has the following beneficial effects: By simultaneously imaging and performing phase-sensitive detection on the mask plane and the wafer plane through a four-channel optical beam splitting system, comprehensive monitoring of the microstructures on the mask and the wafer surfaces is achieved, overcoming the drawback of the traditional monocular vision system being vulnerable to local interference and improving the robustness and reliability of the calibration system. The mask-wafer markless alignment technology is adopted, using the features of the product structure itself as natural marks, avoiding the systematic deviation caused by specially making alignment marks, reducing the production cost, enhancing the process flexibility, and at the same time, significantly improving the alignment accuracy through the sub-wavelength resolution phase-sensitive detection algorithm. By performing multi-scale feature extraction on the three-dimensional surface profile data through wavelet transform and compensating the wavefront aberration of the optical system in real time in combination with a wavefront sensor, the influence of environmental vibration and thermal drift on the calibration accuracy is effectively eliminated, ensuring that the optical system is always in the best imaging state. A comprehensive error model including mechanical error, optical error, and environmental error is established to achieve precise control of the six-degree-of-freedom motion platform of the lithography machine, significantly reducing the deterministic error and random error of the system and improving the stability and reliability of the calibration. Through the dual-precision strategy of coarse alignment with a 633-nm wavelength light source and fine calibration with a 13.5-nm EUV light source, combined with the dual-frequency heterodyne interference technology, the influence of the DC noise in the traditional single-frequency interference is effectively avoided, greatly improving the calibration accuracy, shortening the calibration time, and enhancing the production efficiency. The technical bottleneck of the inconsistency between the traditional off-line calibration and the actual exposure state is broken through, and real-time calibration is achieved without affecting normal exposure, effectively solving the problem of the accumulation of overlay errors and significantly improving the yield and performance of the final chips. Description of the Drawings

[0009] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0010] The structures, ratios, sizes, etc. shown in the drawings of this specification are only used to cooperate with the content disclosed in the specification for those skilled in this technology to understand and read, and are not used to limit the limiting conditions under which the present invention can be implemented. Therefore, they do not have a substantial technical meaning. Any modification of the structure, change of the proportional relationship, or adjustment of the size, without affecting the effects that the present invention can produce and the purposes that can be achieved, should still fall within the scope that can be covered by the technical content disclosed in the present invention.

[0011] Figure 1 It is a schematic flowchart of the lithography machine calibration method based on multi-view vision provided by the embodiment of the present invention; Figure 2It is a schematic block diagram of the structure of a lithography machine calibration device based on multi - vision provided by an embodiment of the present invention; Figure 3 It is a schematic block diagram of the structure of an electronic device provided by an embodiment of the present invention. Detailed implementation manners

[0012] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative work belong to the scope of protection of the present invention.

[0013] The flowcharts shown in the drawings are only illustrative examples, which do not necessarily include all the content and operations / steps, nor do they necessarily need to be executed in the described order. For example, some operations / steps can also be decomposed, combined or partially merged. Therefore, the actual execution order may be changed according to the actual situation.

[0014] It should also be understood that the terms used in the specification of this application are only for the purpose of describing specific embodiments and are not intended to limit this application. As used in the specification of this application and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include the plural forms.

[0015] It should be further understood that the term "and / or" used in the specification of this application and the appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations. Please refer to Figure 1 An embodiment of the multi - vision - based lithography machine calibration method in the embodiments of this application includes: Step 100: Image and perform phase - sensitive detection and analysis on the mask plane and the wafer plane through a four - way optical splitting system to obtain four groups of initial mask - wafer alignment position information; It can be understood that the execution subject of this application can be a lithography machine calibration device based on multi - vision, or a terminal or a server. Specifically, it is not limited here. This application takes the server as the execution subject for illustration.

[0016] Specifically, a multi - vision optical platform that supports multi - wavelength, multi - path optical separation, and high - precision phase modulation is constructed. A reference laser with a wavelength of 633 nm and an extreme ultraviolet (EUV) light source with a wavelength of 13.5 nm are introduced as the core light sources into a four - path optical splitting system. Using a precision optical path separator, it is ensured that the light beams of two different wavelengths can be effectively separated in physical space. Through a specially designed semi - transparent and semi - reflective beam splitter array, the separated multi - wavelength light beams are split, thereby simultaneously constructing four independent high - stability optical paths between the mask plane and the wafer plane. A virtual calibration point is set in space for each path, and these dot matrices serve as the spatial reference for subsequent multi - vision detection. A piezoelectric - driven phase modulator is equipped on each optical path. These modulators have extremely high phase adjustment accuracy and can control the phase states of each path, forming four phase - controllable optical channels, enabling the optical calibration information of each path to be independently modulated and collected. Through this design, the system can achieve spatial multi - point and high - synchrony optical interference measurement of the mask and the wafer surface under the multi - vision framework. Based on the four phase - controllable optical paths, the mask imaging system on the mask plane and the wafer imaging system on the wafer plane are controlled to perform synchronous motion, constructing a multi - vision imaging basic platform. Based on the multi - vision imaging basic platform, images of the mask plane and the wafer plane are collected to obtain the original multi - vision images. The original multi - vision images are pre - processed, including operations such as noise suppression, background normalization, and brightness equalization, to obtain the multi - vision image data of four fields of view. Phase - sensitive detection analysis is performed on the multi - vision image data of the four fields of view. Using an embedded high - speed image - processing unit, a two - dimensional phase - sensitive algorithm is executed on the interference images of the four fields of view. This algorithm combines techniques such as the fast Fourier transform and the Hilbert transform to convert the interference fringe information into a sub - pixel - level phase distribution map, and then extracts the relative displacement amount between the mask and the wafer in their respective fields of view. Through the four - step phase - shift method of the piezoelectric - driven phase modulator, four sets of complementary phase data are obtained, and phase unwrapping and weighted averaging are performed on them to suppress the influence of local noise and environmental disturbances, thereby improving the robustness and accuracy of the initial alignment, and finally obtaining four sets of mask - wafer initial alignment position information.

[0017] In this embodiment, based on the multi-eye visual imaging platform, phase detection and data processing operations are performed on the multi-eye visual image data collected synchronously from four independent fields of view. The reflected or transmitted signals from the mask and the planar wafer are respectively converged on each high-resolution detector array through a precise optical path to obtain four sets of optical image data under the original field of view. Using the principle of interferometric measurement, optical interference processing is performed on each collected image, and corresponding interference fringe images are formed in these four sets of fields of view with the help of the microstructures naturally existing on the mask and wafer surfaces. Through high-speed data acquisition and synchronous control, the interference image is input into a dedicated image processing unit, and the spatial information of the interference image is efficiently converted into frequency domain features using fast Fourier transform, and then the phase information is extracted in combination with Hilbert transform to obtain phase detection data representing the phase change of the microstructure at each spatial position. Based on the phase detection data, the natural distribution of the microstructures of the mask and wafer surface itself is used as a natural marker for feature recognition. By comparing the phase response of the same microstructure in each field of view, the feature extraction and correlation analysis algorithm is used to establish the feature correspondence between the mask and the wafer. The feature correspondence is subjected to a four-step phase shift operation by a piezoelectric-driven phase modulator. By controlling the phase modulator to continuously collect four interference images with a step length of 90°, the four-step phase shift method is used to solve the phase change, and then the relative displacement information of each microstructure point is calculated. In order to ensure the spatiotemporal stability and anti-environmental interference ability of the data, the adaptive frequency window division technology is introduced to dynamically analyze the signal fluctuations during the acquisition period, automatically screen out the time window with the smallest environmental disturbance, and use the signal averaging processing algorithm to fuse the multiple frames of data in the window, effectively improve the signal-to-noise ratio, reduce the error caused by occasional interference, and obtain the target position data. The target position data is calculated at the sub-pixel level through the sub-wavelength resolution algorithm. The algorithm comprehensively utilizes multi-field of view, multi-scale features and interference fringe phase changes to achieve a resolution far superior to that of a single-wavelength system, and can track tiny spatial changes between the mask and the wafer. Four sets of mask-wafer initial alignment position information are obtained.

[0018] Step 200, performing scanning white light interference edge enhancement processing on four sets of mask-wafer initial alignment position information to obtain three-dimensional surface profile data; Specifically, a piezoelectric vertical scanning device is used to vertically scan the initial alignment position information of four groups of mask-wafer. This piezoelectric scanning device has sub-nanometer displacement control ability and can achieve continuous and controllable height changes in an extremely narrow space. By placing the mask and the wafer surface respectively under the scanning path, high-density sampling of the surface structure is completed within a specified vertical interval, and the scanning signals at each corresponding spatial position are obtained in real time. The scanning data of the mask and the wafer surface is processed with continuously changing wavelengths, that is, using the principle of white light interference, the wavelength of the incident light source is continuously changed from 380 nanometers in the visible light region to 780 nanometers, and stepped scanning is carried out in small steps (such as 5 nanometers). Each wavelength point corresponds to different interference conditions, and the system synchronously acquires a large number of multi-view interference images with controllable wavelengths. Noise suppression processing is performed on the interference images of different wavelengths. Spatial domain filtering algorithms such as 9×9 Gaussian filtering are used to effectively suppress high-frequency noise and background disturbances in the interference images, ensuring the clarity and contrast of the interference fringes themselves. After noise suppression, edge extraction processing is performed on the filtered interference images. Through edge detection algorithms such as the Sobel operator, the bright and dark boundaries caused by the surface microstructure in the interference images are captured, and these edge features are mapped into a spatial point set with extremely high precision. Based on the surface microstructure edge information, the spatial height data of each edge point is deduced by the position of the maximum contrast of the interference fringes. In the multi-wavelength interference image sequence, the point of maximum contrast of each interference fringe corresponds to the spatial position where the optical path difference is an integer multiple of the wavelength. By analyzing the contrast change of the interference fringes of different edge points at each wavelength, the maximum contrast position of each feature point is automatically located using relevant algorithms, and the absolute height of this point is deduced based on this and the known optical parameters. Based on the above steps, the microstructure height information of the entire mask and wafer surface is reconstructed point by point, and finally high-resolution and high-precision three-dimensional surface profile data is obtained.

[0019] Step 300: Perform wavelet transform processing on the three-dimensional surface profile data, extract target feature points, and at the same time perform dynamic registration on the four-way optical splitting system according to the target feature points to obtain multi-view visual feature registration data; It should be noted that multi-level wavelet decomposition processing is performed on the three-dimensional surface contour data. Wavelet basis functions with good locality and resolution ability, such as Daubechies D4, are used to perform multi-scale decomposition on the surface data. Through multi-level wavelet decomposition, the complex surface structure is disassembled into several levels, and each level respectively reflects the detailed features at different spatial scales. An adaptive threshold strategy is used to suppress the background noise of the hierarchical data of each layer of features. The threshold standard is dynamically set according to the root mean square value of each layer of feature signals (such as set to 1.5 times the root mean square value), automatically shielding the background noise and invalid data below the threshold, and obtaining the purified multi-scale features. The purified multi-scale features are screened, and considering the contrast of the feature points (required to be significantly higher than the background), the edge stability (the change rate of the edge sharpness needs to be small enough), and the recognition repeatability across multiple fields of view, the feature points that can be stably tracked in different fields of view are identified as target feature points. While extracting the target feature points, the wavefront aberration changes of the four-channel optical spectroscopic system are monitored in real time. A high-precision Shack-Hartmann wavefront sensor is deployed, which consists of a microlens array and a high-resolution detector, and can measure the wavefront aberration in the real-time imaging path. The sensor feeds back the change amount of the wavefront aberration of each optical path to the system control unit in the form of a data stream, and can be decomposed into multiple Zernike polynomial components. Based on the real-time wavefront aberration data, the system dynamically adjusts the deformable mirror through an adaptive control algorithm. A large number of independently controlled piezoelectric actuators are integrated inside the deformable mirror, and each actuator independently adjusts the local mirror shape according to the feedback, thereby instantaneously eliminating or compensating the wavefront aberration introduced by the environment, machinery, and optical devices themselves, so that the imaging quality of each optical path always maintains the optimal state. When the optical imaging path is stable, the extracted target feature points are registered and matched with the four optimized imaging paths to obtain multi-view visual feature registration data. This process includes spatial reprojection under multi-view vision, fitting of the feature point space transformation matrix, and iteration of the optimization algorithm. The system calculates the coordinate differences of each target feature point in the four fields of view, and through mathematical tools such as the least squares method, fits the optimal spatial registration matrix to unify all feature points into the same coordinate system. At the same time, dynamic registration can continuously perform adaptive fine-tuning according to the wavefront aberration feedback of the optical system to ensure high-precision feature coincidence and spatial consistency even under environmental disturbances or system drifts.

[0020] Step 400: Establish a comprehensive error model including mechanical error, optical error, and environmental error based on the multi-view visual feature registration data, and perform real-time correction and error compensation on the six-degree-of-freedom motion platform of the lithography machine based on the comprehensive error model to obtain the multi-view visual calibration result of the lithography machine.

[0021] Specifically, based on the registered multi-view feature points, all geometric position information of the six-degree-of-freedom motion platform in the three linear translation directions of X, Y, and Z and the three rotation directions of Rx, Ry, and Rz is extracted from continuous motion sampling through a high-precision spatial fitting algorithm. Combining with the target path data under the real working conditions, the geometric error parameters in each degree-of-freedom direction are deduced by statistical optimization methods such as the least squares method. These parameters reflect the spatial deviation caused by the superposition of multiple factors such as structure manufacturing, tolerance accumulation, installation error, and micro-displacement during the execution of the motion platform, forming a mechanical error model. At the same time, standard test pattern acquisition data is introduced to systematically model the aberration, distortion, and magnification error of the imaging optical system. By separately exposing and detecting a set of high-precision standard patterns on the mask and the wafer surface, the system can accurately measure the spatial aberration, local distortion, and overall magnification change caused by defects in the optical elements themselves, misalignment, or focal length change, and parametrically extract the above optical errors through a series of back-projection algorithms. To characterize the influence of the environment on the lithography accuracy, high-sensitivity temperature sensors, air flow velocity sensors, and vibration sensors are arranged at key nodes to collect high-frequency data on temperature changes, air flow disturbances, and mechanical vibrations respectively. Using statistical modeling means to quantify the actual disturbance effects of these environmental factors on the motion platform and the optical system, an environmental error model is established. The mechanical error model, optical error model, and environmental error model are integrated to obtain a system comprehensive error model. During the specific operation process, the spatial distribution and correlation of each type of error are modeled, and the coupling and superposition effects of various error sources are probabilistically modeled and dynamically predicted through methods such as Markov chain Monte Carlo to achieve the global optimal estimation of the error distribution. Based on the system comprehensive error model, the motion control system of the lithography machine adopts a feedforward control compensation strategy to real-time correct the target position and action trajectory of the motion platform according to the output of the error model, actively eliminating the deterministic part of the mechanical, optical, and environmental error sources before the command is issued. At the same time, continuously collect multi-view visual feedback data and environmental sensing data during the motion process to make minor corrections to the residual errors, realizing the deep integration of feedforward compensation and real-time feedback, and obtaining the multi-view visual calibration result of the lithography machine.

[0022] In this embodiment, the multi-source error data in the system comprehensive error model is subjected to eigen-decomposition. Through statistical methods such as principal component analysis, the eigenvectors and eigenvalues of the high-dimensional error matrix in the model are decomposed, effectively dividing the original error into two major parts: one is the deterministic component matrix with obvious physical laws or structural causal relationships, representing the main control trends of traceable errors such as mechanical, optical, and environmental errors in the system; the other is the random component matrix, mainly reflecting non-structural errors such as external disturbances, transient random fluctuations, and microscopic noises. Based on the data of these two types of components, the error principal component distribution data is generated, and these principal components reflect the distribution weights and sensitive directions of multi-source errors in each degree of freedom and each time period. Based on the error principal component distribution data, independent compensation controllers are designed for each degree of freedom of the six-degree-of-freedom motion platform of the lithography machine, that is, the translations of X, Y, and Z and the rotations of Rx, Ry, and Rz. Each compensation controller establishes an independent prediction and compensation model according to the error change law of the corresponding degree of freedom based on the principal component distribution data, forming a six-channel fully decoupled error compensation control system. In the specific parameter optimization stage, an orthogonal experimental design is carried out for the feedforward control parameters in the decoupling compensation strategy. By systematically changing the compensation gain coefficients of each degree of freedom, the motion accuracy and response performance under different parameter combinations are measured respectively, and the best parameter configuration is selected by means of orthogonal analysis to find the optimal compensation gain coefficients for each degree of freedom. Based on this set of gain coefficients, the spatial mapping relationship of the six-dimensional error vector is constructed, and the error data actually collected by the motion platform and the target pose requirements are corresponded through the spatial mapping algorithm, and the compensation control instructions are generated in real time accordingly. The six-degree-of-freedom motion platform of the lithography machine is subjected to feedforward control compensation based on the compensation control instructions, and at the same time, high-frequency position information from four independent fields of view of the multi-camera vision system is collected. These feedback data are processed by the complementary filtering algorithm to fuse the response speed of the high-bandwidth field of view and the noise suppression ability of the low-bandwidth field of view. Through the deep coupling of feedforward compensation and real-time complementary filtering feedback, the system dynamically monitors and adaptively adjusts the six-degree-of-freedom compensation control effect, so that the final output multi-camera vision calibration result of the lithography machine can maintain sub-nanometer-level spatial alignment and extremely high long-term stability in the full-time and full-space range.

[0023] In the embodiments of the present application, by using a four-channel optical spectroscopic system to simultaneously image and perform phase-sensitive detection on the mask plane and the wafer plane, comprehensive monitoring of the microstructures on the mask and the wafer surfaces is achieved, overcoming the drawback of the traditional monocular vision system being vulnerable to local interference, and improving the robustness and reliability of the calibration system. The mask-wafer unmarked alignment technology is adopted, using the features of the product structure itself as natural marks, avoiding the systematic deviation caused by specially making alignment marks, reducing the production cost, enhancing the process flexibility, and at the same time, significantly improving the alignment accuracy through the sub-wavelength resolution phase-sensitive detection algorithm. By performing multi-scale feature extraction on the three-dimensional surface profile data through wavelet transform and combining with a wavefront sensor to compensate for the wave aberration of the optical system in real time, the influence of environmental vibration and thermal drift on the calibration accuracy is effectively eliminated, ensuring that the optical system is always in the best imaging state. A comprehensive error model including mechanical error, optical error, and environmental error is established to achieve precise control of the six-degree-of-freedom motion platform of the lithography machine, significantly reducing the deterministic error and random error of the system, and improving the stability and reliability of the calibration. Through the dual-precision strategy of rough alignment with a 633 nm wavelength light source and fine calibration with a 13.5 nm EUV light source, combined with the dual-frequency heterodyne interferometry technology, the influence of DC noise in traditional single-frequency interferometry is effectively avoided, greatly improving the calibration accuracy, shortening the calibration time, and enhancing the production efficiency. The technical bottleneck of the inconsistency between traditional offline calibration and the actual exposure state is broken through, and real-time calibration is achieved without affecting normal exposure, effectively solving the problem of the accumulation of overlay errors, and significantly improving the yield and performance of the final chip.

[0024] In a specific embodiment, the process of executing step 100 may specifically include the following steps: Introduce a reference laser with a wavelength of 633 nm and an EUV light source with a wavelength of 13.5 nm into the four-channel optical spectroscopic system, and perform optical path separation through the four-channel optical spectroscopic system to obtain separated multi-wavelength light beams; Divide the separated multi-wavelength light beams through a semi-transparent and semi-reflective beam splitter array to construct four independent optical paths between the mask plane and the wafer plane, and obtain four groups of virtual fiducial dot arrays; Control the phase of the four groups of virtual fiducial dot arrays through a piezoelectric-driven phase modulator to obtain four phase-controllable optical paths; Based on the four phase-controllable optical paths, control the mask imaging system on the mask plane and the wafer imaging system on the wafer plane to perform synchronous motion to obtain a multi-view vision imaging basic platform; Based on the multi-view vision imaging basic platform, collect images of the mask plane and the wafer plane to obtain original multi-view vision images, and preprocess the original multi-view vision images to obtain multi-view vision image data of four fields of view; Perform phase-sensitive detection analysis on the multi-view visual image data of four fields of view to obtain four sets of mask-wafer initial alignment position information.

[0025] Specifically, the light source part of the system selects a highly stable 633nm laser and a dedicated 13.5nm EUV light source. Through the optical path guiding mechanism and the high-reflectivity multi-layer film mirror, the light energy of the two wavelengths is efficiently coupled into the same main optical path. Through the four-way optical splitting system designed for multi-wavelengths, the light of the two wavelengths is effectively separated respectively, ensuring that four groups of pure 633nm and 13.5nm multi-wavelength light beams are output under the conditions of no crosstalk and loss, and each light beam has a highly consistent energy distribution and phase stability. After the separation of the multi-wavelength light beams is completed, a special semi-transparent and semi-reflective beam splitting mirror array is used to perform spatial splitting on the separated light beams. This beam splitting mirror array is arranged between the mask plane and the wafer plane according to the spatial geometric relationship, and its core function is to accurately split each input light beam into four mutually independent and optically controllable spatial trajectories. These four paths are constructed by a high-precision optical alignment mechanism inside the system to ensure that each path can be stably projected between the mask surface and the wafer surface, forming a structured virtual calibration dot matrix. A high-precision piezoelectric-driven phase modulator is integrated on each independent optical path. These modulators have an extremely high dynamic response speed and sub-wavelength-level phase control accuracy. By applying an accurate voltage excitation to the piezoelectric actuator unit, the dynamic modulation of the phase of the light beam passing through this path is realized. The system uniformly coordinates the working states of the four phase modulators through the controller to achieve the phase control of the four groups of virtual calibration dot matrices that are completely decoupled in space. Through this step, four optically controllable channels with completely controllable phases are formed, and each path can perform independent phase adjustment and interference optimization for the actual exposure or calibration working conditions. Based on the four optically controllable optical paths, the mask imaging system and the wafer imaging system are rigidly connected to the same six-degree-of-freedom high-precision motion platform. This platform uses a carbon fiber composite material bracket and a high-resolution servo motor group, and has an extremely low coefficient of thermal expansion and high dynamic positioning ability. During operation, through the real-time synchronous motion controller, it is ensured that the imaging systems of the mask and the wafer can achieve linkage in all spatial dimensions, and the spatial structure between the mask surface and the wafer surface can be maintained completely consistent whether in the static or dynamic calibration process, thereby constructing a highly stable and spatially rigid multi-view vision imaging basic platform. Based on the multi-view vision imaging basic platform, image acquisition is performed on the mask plane and the wafer plane. Four groups of high-resolution two-dimensional quantum dot-enhanced detectors corresponding to the four independent optical paths respectively capture imaging data from different perspectives. The collected original image data enters the preprocessing unit, and after a series of operations such as noise removal, background equalization, and brightness normalization, the images of each field of view are standardized into an ideal analysis input state. After preprocessing, the multi-view vision image data of the four fields of view is obtained.In the phase-sensitive detection and analysis stage, four sets of multi-view visual image data are input into a high-performance image processing unit. Through algorithms such as fast Fourier transform and Hilbert transform, the image spatial domain data is efficiently converted into frequency domain features, and then sub-pixel level phase change information is extracted from the frequency domain signal. According to the mask and the microstructure distribution on the wafer surface itself, using natural features as markers, the system automatically identifies and matches the phase response changes at the same spatial position under multiple fields of view. Through the four-step phase-shift method and high-precision phase modulation control, the system can perform sub-wavelength level phase difference acquisition on each set of data. Combining with the adaptive window signal averaging technique, it significantly suppresses the influence of environmental vibration and noise, and improves the overall anti-interference ability and data stability of the system. The phase-sensitive detection results of all four fields of view are finally output as the initial alignment position information of the mask-wafer through weighted fusion and spatial coordinate transformation.

[0026] In a specific embodiment, the process of performing the steps to conduct phase-sensitive detection and analysis on the multi-view visual image data of four fields of view to obtain four sets of initial alignment position information of the mask-wafer may specifically include the following steps: Perform interference processing on the multi-view visual image data of four fields of view to obtain four sets of interference images, and perform fast Fourier transform and Hilbert transform on the four sets of interference images to obtain phase detection data; Use the microstructure of the mask and the wafer surface itself as natural markers to analyze the phase detection data to obtain the feature correspondence; Perform four-step phase shift on the feature correspondence through a piezoelectric phase modulator to obtain displacement information, and perform adaptive frequency window division and signal averaging processing on the displacement information to obtain target position data; Calculate the target position data through a sub-wavelength resolution algorithm to obtain four sets of initial alignment position information of the mask-wafer.

[0027] Specifically, interference processing is performed on the multi-view visual image data of four fields of view. Based on the spatial characteristics of the mask and the wafer microstructure, each path of data forms a typical interference fringe image through multi-wavelength light sources (such as 633 nm and 13.5 nm) in the interferometer structure. These fringes reflect the relative height difference between the mask and the wafer at the nanoscale and the slight changes in the spatial topography. The four groups of interference images respectively represent the optical path phase difference distributions of the mask-wafer pair from four spatial perspectives. Using the fast Fourier transform, the spatial domain information of the interference images is efficiently converted into frequency domain features, effectively extracting the main frequency components of the fringes and enhancing the sensitivity to periodic microstructures and local spatial deformations. Applying the Hilbert transform to further process the frequency domain signal, by constructing an analytic signal, calculating the instantaneous phase and amplitude, the phase change information of each spatial point is restored. The Hilbert transform has a natural advantage in sub-pixel phase extraction and can convert the fringe phase information into high-resolution phase detection data. Four fields of view respectively obtain phase maps with fine spatial distribution and sensitive response, and these images reflect the relative micro-displacement state and the relative height field change of the mask-wafer structure pair in four spatial dimensions. Using the microstructures on the surfaces of the mask and the wafer themselves as natural markers for feature extraction and matching. Through the spatial distribution, shape changes, intensity mutations, and fringe spacings of the feature points in the phase detection data, the natural microstructures distributed on the surfaces of the mask and the wafer are identified and feature points are extracted. Based on the image registration algorithm and the feature descriptor technology, the phase responses of the same microstructure under multi-view fields of view in different spatial projections are automatically established corresponding relationships, obtaining four groups of feature correspondences with consistent space and clear physical meanings. Using a piezoelectric phase modulator, a four-step phase shift operation is performed on the key spatial points during the interference process. That is, by precisely applying an electrical signal with a 90° step to the phase modulator, four interference images with complementary phases are respectively acquired, so as to obtain four groups of signals with different phase states at the same spatial position. With the four-step phase shift method, a set of analytical equations can be used to convert the measured signals into unique spatial phase information, suppressing the influence of system noise and fringe distortion and improving the displacement detection accuracy. During the data processing process, an adaptive frequency window division algorithm is adopted. By real-time monitoring the environmental vibration and the signal fluctuation frequency, a time window with the least interference and the best signal is dynamically selected, and the measurement results of the same spatial point under multiple windows are averaged to obtain the target position data. The sub-wavelength resolution algorithm is used to perform sub-pixel level spatial reconstruction and calculation on the target position data. This algorithm is based on the multi-wavelength interference principle, utilizes the sub-wavelength interference characteristics between the fringe periods and the spatial microstructures at different wavelengths, and combines multi-scale data fusion and interpolation reconstruction technologies to correct the actual position information of each spatial target point. The sub-wavelength resolution method can effectively improve the ability to capture the displacement changes of the mask-wafer, making the obtained spatial resolution far better than the theoretical limit of single-wavelength or single-view systems.After the initial alignment position information of four groups of masks and wafers is calculated with high precision, it is uniformly transformed into the spatial coordinate system of multi-view fusion and becomes the core data basis for subsequent three-dimensional surface reconstruction, error compensation, and six-degree-of-freedom motion platform adaptive calibration in the system control logic.

[0028] In a specific embodiment, the process of executing step 200 may specifically include the following steps: Vertically scan the initial alignment position information of four groups of masks and wafers through a piezoelectric vertical scanning device to obtain the scanning data of the mask and wafer surfaces; Perform wavelength continuous change processing on the scanning data of the mask and wafer surfaces to obtain interference images with different wavelengths; Perform noise suppression processing on the interference images with different wavelengths to obtain filtered interference images, and perform edge extraction on the filtered interference images to obtain the edge information of the surface microstructure; According to the edge information of the surface microstructure, determine the height data of each point through the maximum contrast position of the interference fringes to obtain three-dimensional surface contour data.

[0029] Specifically, at the hardware level, the piezoelectric vertical scanning module uses a nano-scale piezoelectric ceramic driver and integrates a high-sensitivity position feedback circuit to achieve height fine-tuning with a scanning step accuracy better than 0.5 nanometers. The module is installed at the key node of the multi-eye visual imaging platform. Through precise synchronous control, the mask and wafer surface are vertically scanned point by point and layer by layer according to the preset height range. Each scanning movement slowly raises or lowers the mask and wafer surface from the reference plane with a very small step. During the whole process, the data acquisition end of the four sets of fields of view can synchronously capture the reflection or transmission signals at different height levels, thereby constructing a spatial height-reflection relationship database. The scanning data of the mask and wafer surface are processed with continuous wavelength changes through multi-wavelength interference imaging technology. A wide-spectrum white light source is provided in the calibration system, and the wavelength coverage range is from 380 nanometers to 780 nanometers in the visible light region. Through a high-speed tunable filter or a multi-wavelength laser combination, the output wavelength is controlled to change step by step with very small steps. At each spatial point corresponding to the height step, the system repeatedly irradiates and collects data with a predetermined wavelength sequence to ensure that each spatial point corresponds to a set of interference fringe images at different wavelengths. Since the phase, fringe spacing and contrast of interference fringes of different wavelengths are extremely sensitive to changes in the height of the surface microstructure, the superposition of multi-wavelength data improves the resolution of sub-nanometer surface fluctuations. At the same time, the multi-wavelength strategy also enhances the overall adaptability of the system to changes in surface optical constants, microstructure gradients and complex mask-wafer interface effects, so that the three-dimensional surface reconstruction can maintain highly consistent and accurate performance under various materials and process conditions. Noise suppression processing is performed on interference images of different wavelengths. Image processing algorithms such as Gaussian filtering and mean filtering are used in the spatial domain and frequency domain to effectively filter out non-structural interference such as high-frequency random noise, optical speckle and electronic detector background fluctuations. The specific parameter settings are adaptively adjusted according to the signal-to-noise ratio of the acquired image and the distribution of the main frequency of the fringes to ensure that the edge clarity and main frequency integrity of the fringing signal are not destroyed. Under some complex working conditions, multi-scale noise modeling and wavelet denoising technology are used to hierarchically optimize the local noise and global fringing features in the image to achieve more sensitive capture of subtle fringes, weak reflection areas and surface gradient changes. For the interference image that has been processed by noise suppression, the geometric features of the surface microstructure are automatically extracted using the edge detection algorithm. The edge operators such as Sobel and Canny are used, combined with threshold segmentation, gradient enhancement and other technologies, to capture the bright and dark boundary lines in the fringe image. These edges correspond to the local extreme points of the steps, grooves, defects and other geometric features of the surface microstructure. For the edge data under multi-wavelength and multi-viewing angles, spatial consistency and multi-scale adaptive fusion algorithms are introduced to automatically remove false boundaries and isolated noise, retain the feature points that overlap in space and have stable structures under all main views, and obtain the edge information of the surface microstructure. According to the edge information of the surface microstructure, the height data of each point is determined by the position of the maximum contrast of the interference fringes.There is a spatial modulation function relationship between the fringe contrast of multi-wavelength interference imaging and the surface height. At a specific wavelength, the position of the maximum fringe contrast corresponds to the spatial height where the optical path difference is an integer multiple of the wavelength. By gradually scanning the interference fringe sequences of each spatial point at different wavelengths and using the contrast peak detection algorithm, automatically find the maximum contrast wavelength corresponding to each spatial point and the corresponding scanning height. Combine this height with the known spatial calibration parameters and system constants to infer the actual height value of each feature point in three-dimensional space. Finally, summarize and fuse the height data of all spatial points in the four fields of view, and through the multi-view spatial projection and coordinate transformation algorithms, uniformly convert them into the three-dimensional surface contour data of the multi-view vision system of the lithography machine.

[0030] In a specific embodiment, the process of executing step 300 may specifically include the following steps: Perform multi-layer wavelet decomposition processing on the three-dimensional surface contour data to obtain feature hierarchical data at different scales; Perform background noise suppression processing on the feature hierarchical data through an adaptive threshold to obtain purified multi-scale features, and screen the purified multi-scale features to obtain target feature points; Monitor the wavefront aberration of the four-way optical beam splitting system through a Shack-Hartmann wavefront sensor to obtain real-time wavefront aberration data; Based on the real-time wavefront aberration data, control the deformable mirror to perform dynamic correction on the four-way optical beam splitting system to obtain a stable optical imaging path; Register and match the target feature points with the stable optical imaging path to obtain multi-view vision feature registration data.

[0031] Specifically, a multi-level wavelet decomposition algorithm is used to process the three-dimensional surface contour data. Orthogonal wavelet basis functions with excellent locality and high-order orthogonality, such as Daubechies, Symlet, or Coiflet, are selected to decompose the three-dimensional height data at five or deeper levels. Each level of wavelet decomposition decomposes the original surface height field into two major parts: low frequency (large-scale background trend) and high frequency (microstructural fluctuations). As the decomposition level deepens, characteristic hierarchical data at different scales are obtained. An adaptive threshold is introduced to suppress background noise for the characteristic hierarchical data. The root mean square value of the characteristic coefficients of each decomposition level is statistically calculated, and the adaptive threshold of the current level is dynamically set. All coefficients below this threshold are set to zero or forced to be reduced, only retaining the spatial features whose signal energy is significantly higher than the noise background. The adaptive threshold strategy can be flexibly adjusted according to the signal-to-noise ratio under different samples and different imaging conditions to obtain purified multi-scale features. The purified multi-scale features are screened, and the screening criteria include three points: first, the contrast of the feature points within this scale needs to be significantly higher than the surrounding background (generally required to be more than three times); second, the sharpness of the edges of the feature points, that is, the spatial gradient, needs to be stable (identified by local maximum gradient points); third, this feature has a very high spatial repetition rate (more than 95%) under different multi-view fields. Through multi-scale cross-discrimination and spatial consistency analysis, a set of target feature points that can reflect the physical essence of the surface structure are finally locked. At the same time, Shack-Hartmann wavefront sensors are deployed at the key nodes of the four-channel optical beam splitting system. Based on the principle of the microlens array, the sensor can real-time sense the wavefront distortion and aberration distribution passing through each optical channel. The sensor array divides the incident wavefront into several sub-beams, and measures the focal spot position offset of each sub-beam through a two-dimensional detector array, thereby calculating the wavefront topography and all Zernike polynomial components. The system obtains real-time wave aberration data through high-frequency sampling, covering the full-spectrum information of low-order aberrations (such as spherical aberration, coma, astigmatism, etc.) and high-order aberrations (such as local misalignment, asymmetric distortion, etc.). These real-time wave aberration data can not only reflect the imaging state of the system under the current environment, but also quantify the local dynamic instability and global optical drift of each channel. Based on the real-time monitoring results of the Shack-Hartmann wavefront sensor, the system synchronously drives the high-precision deformable mirrors integrated on each optical channel. The deformable mirror adopts MEMS or piezoelectric ceramic array technology, with dozens or even hundreds of independent micro-regions, and can quickly adjust the reflection curvature of each local region by independently applying a small voltage. The system control algorithm real-time analyzes the Shack-Hartmann wavefront data, maps the distribution of the wave aberration into the shape adjustment instructions of each micro-region of the deformable mirror, and through active deformation, makes the reflecting surface always maintain the same height as the theoretical optimal imaging surface throughout the whole period.This dynamic correction process features a millisecond-level response speed, eliminating environmental disturbances, temperature drifts, and non-ideal optical path effects in real time, and adapting to minor optical errors caused by different exposure conditions, platform movements, and material aging, thus enhancing the consistency and stability of the multi-channel imaging path. Based on the extraction of multi-scale target feature points and the dynamic stability of the four optical paths, the system enters the multi-view vision feature registration stage. The positions and spatial structures of all the selected target feature points in the four spatial fields of view are corresponded, and through spatial projection relationships and coordinate transformation matrices (such as rigid transformation, affine transformation, least squares fitting, etc.), they are uniformly transformed into the same global coordinate system. To improve the robustness and accuracy of registration, feature points with uniform spatial distribution and strong structural independence are selected to construct a registration matrix. At the same time, during the registration process, the imaging path parameters after wavefront aberration correction are combined to automatically compensate for structural offsets caused by minor optical path distortions or non-coplanarity of multiple channels. Through the deep matching of multi-scale features and dynamic optical paths, multi-view vision feature registration data is output.

[0032] Among them, registering and matching the target feature points with the stable optical imaging path to obtain multi-view vision feature registration data includes: encoding the target feature points with multi-dimensional feature vectors, classifying the feature points into structural feature points, edge feature points, and surface feature points through feature attribute analysis to obtain a classified feature data set; constructing a feature importance evaluation function based on the classified feature data set, assigning spatial distribution weights to the feature points to obtain a set of weighted feature points; using a hyper-heuristic optimizer to iteratively screen the set of weighted feature points, evaluating the calibration contribution degree of the feature point combination through an objective function to obtain an optimal subset of feature points; constructing an inter-field feature mapping relationship according to the distribution of the optimal subset of feature points in the four fields of view, and calculating the affine transformation matrix between the fields of view through the least squares method to obtain inter-field registration parameters; using a multi-feature fusion integration estimator to evaluate the inter-field registration parameters, identifying abnormal registration relationships through confidence analysis to obtain a stable inter-field registration model; dynamically adjusting the optical imaging paths of the four fields of view based on the stable inter-field registration model to eliminate parallax errors and obtain a unified calibration coordinate system; transforming the target feature points to the unified calibration coordinate system, integrating the feature point information of the four fields of view through a weighted fusion algorithm to obtain a global feature point registration set; applying an adaptive matching algorithm to the global feature point registration set, and iteratively optimizing the registration in combination with the stable optical imaging path to finally obtain high-precision multi-view vision feature registration data.

[0033] In a specific embodiment, the process of executing step 400 may specifically include the following steps: Extract the geometric error parameters of the six-degree-of-freedom motion platform of the lithography machine in the three translational directions of X, Y, and Z and the three rotational directions of Rx, Ry, and Rz from the multi-view vision feature registration data to obtain a mechanical error model; Collect data based on standard test patterns to calculate aberration, distortion, and magnification error parameters, obtain an optical error model, and collect environmental parameters through temperature sensors, air velocity sensors, and vibration sensors to construct an environmental error model; Integrate the mechanical error model, optical error model, and environmental error model to obtain a system comprehensive error model; Perform feedforward control compensation on the six-degree-of-freedom motion platform of the lithography machine based on the system comprehensive error model to obtain the multi-camera vision calibration result of the lithography machine.

[0034] Specifically, the motion state of the six-degree-of-freedom motion platform is fully analyzed based on the multi-view vision feature registration data. Through a high-resolution multi-view registration algorithm, the multi-view coordinates of all spatial feature points are made to correspond one by one, and based on this, the real-time spatial postures of the mask and the wafer in the global coordinate system are dynamically reconstructed. After the motion platform completes a round of spatial sampling, according to the spatial projections of the same feature point under four independent fields of view, through the rigid body transformation matrix and the least squares fitting algorithm, the actual spatial offsets of the motion platform in the three linear translation directions of X, Y, and Z and the three rotational degrees of freedom of Rx, Ry, and Rz are calculated. Differential analysis is carried out with the pose data of the ideal geometric model to quantitatively analyze all motion deviations of the platform in the six degrees of freedom one by one, including translational errors at the micron and even nanometer levels, rotational errors at the sub-arcsecond level, and coupling terms between various degrees of freedom. Through long-term, multi-batch, and multi-angle collection and statistical optimization, the systematic distribution and change trend of the mechanical errors of the platform are modeled to form a mechanical error model, which can dynamically reflect the actual accuracy and error field distribution of the motion platform under different spatial regions, different operating speeds, and different load states. At the same time, an optical error model reflecting the performance of the entire optical link is established. By repeatedly exposing and detecting high-precision standard test patterns, such as standard checkerboards, dot arrays, or line array patterns, on the mask and the wafer surfaces, the spatial projection data of each characteristic pattern under actual imaging are collected. By comparing with the theoretical template, the aberrations (such as spherical aberration, astigmatism, coma) introduced by the optical lens system, the full-field distortion (such as barrel distortion, pincushion distortion), and the global magnification error are measured and calculated. Using the reverse optical path modeling and fitting algorithm, these error components are parameterized according to the spatial distribution and physical nature to form an optical error model with a complete structure and clear physical mechanism. At the same time, the dynamic responses of the optical link under different temperatures, humidities, airflows, and equipment operating states are collected to improve the dynamic tunability and real-time adaptability of the model. The establishment of the environmental error model depends on high-precision temperature sensors, airflow velocity sensors, and vibration sensors distributed at key nodes of the lithography system. Through high-speed multi-channel data acquisition, the influencing quantities such as spatial temperature gradients, environmental flow field disturbances, and mechanical micro-vibrations are synchronously recorded at the millisecond level, and the coupling relationship between environmental disturbances and platform spatial errors and optical system drifts is modeled using multi-dimensional statistical analysis methods. Through time series analysis and multi-variable correlation regression, the influence weights and dynamic evolution of various environmental parameter changes on actual motion and imaging errors are attributed, and an environmental error model is dynamically generated. Integrate the mechanical error model, the optical error model, and the environmental error model. Map the parameters of various models to the same spatial reference system, and use methods such as Bayesian inference, principal component analysis, or Markov chain Monte Carlo to quantify the independent distribution, interactive coupling, and dynamic time series of the three types of errors, forming a comprehensive probability distribution and an error prediction equation.Model integration not only considers the correlation of errors in space, time, and system structure, but also combines external factors such as production processes, platform dynamics, and system aging to achieve continuous adaptive updates. Finally, the comprehensive error model describes the upper bound of the expected error, mean drift, and lower bound of perturbation of the current system in each degree of freedom in the form of a multi-dimensional Gaussian or piecewise distribution, providing a full-information predictive compensation benchmark for the motion control system. Based on the system comprehensive error model, feedforward control compensation is implemented for the six-degree-of-freedom motion platform. Feedforward compensation can adaptively correct key parameters such as the target motion trajectory, acceleration, and micro-step adjustment of the platform in advance according to the predicted output of the error model before the motion command is issued. The motion controller dynamically adjusts the control weights and real-time gain factors of the six degrees of freedom according to the mechanical error field distribution, optical imaging error parameters, and environmental perturbation data in real time, and performs offset adjustment and error inverse solution compensation for each frame of motion. The entire compensation process collects high-frequency feedback from the multi-camera vision system in real time, eliminates the occasional errors between the model and the actual state through residual fine-tuning, realizes the deep coupling of feedforward compensation and closed-loop feedback, and finally obtains the multi-camera vision calibration result of the lithography machine.

[0035] In a specific embodiment, the process of performing feedforward control compensation on the six-degree-of-freedom motion platform of the lithography machine based on the system comprehensive error model to obtain the multi-camera vision calibration result of the lithography machine may specifically include the following steps: Perform eigen decomposition on the system comprehensive error model to obtain a deterministic component matrix and a random component matrix, and generate error principal component distribution data according to the deterministic component matrix and the random component matrix; Based on the error principal component distribution data, construct independent compensation controllers for the six degrees of freedom in the six-degree-of-freedom motion platform of the lithography machine. Each compensation controller corresponds to the translational degrees of freedom of X, Y, Z and the rotational degrees of freedom of Rx, Ry, Rz, and obtain a decoupled compensation strategy; Perform orthogonal experiments on the feedforward control parameters of the decoupled compensation strategy to determine the compensation gain coefficients for each degree of freedom, and establish a spatial mapping relationship of the six-dimensional error vector based on the compensation gain coefficients for each degree of freedom to obtain a compensation control command; Perform feedforward control compensation on the six-degree-of-freedom motion platform of the lithography machine based on the compensation control command, and at the same time collect position feedback data from four fields of view for complementary filtering and error compensation to obtain the multi-camera vision calibration result of the lithography machine.

[0036] Specifically, using multivariate statistical methods such as principal component analysis and singular value decomposition, the feature decomposition of the system comprehensive error model is carried out to obtain a deterministic component matrix and a random component matrix. The deterministic component matrix contains error terms with clear physical mechanisms and strong regularity, such as mechanical system structure errors, optical system systematic distortions, and long-term environmental drifts; the random component matrix contains random errors without specific spatio-temporal laws, such as process perturbations, equipment micro-vibrations, environmental noises, and transient perturbations. Through decomposition, the energy weights, sensitive directions, and dynamic coupling relationships of different error principal components in spatial distribution, time evolution, and each degree-of-freedom channel are identified. Using the principal component contribution rate ranking and covariance matrix analysis, a set of error principal component distribution data is output. Based on the error principal component distribution data, independent compensation controllers for six degrees of freedom in the six-degree-of-freedom motion platform of the lithography machine are constructed, and each compensation controller corresponds to the translational degrees of freedom of X, Y, Z and the rotational degrees of freedom of Rx, Ry, Rz. Each compensation controller separately establishes a predictive compensation model according to the error principal component distribution on its own degree of freedom, taking into account both the long-term trend and deterministic bias of the dominant error component, as well as various high-frequency and low-amplitude random perturbations. Through the matrix decoupling algorithm, the error linkage between multiple degrees of freedom originally caused by mechanical structure, optical coupling, or environmental interference is transformed into six independent and non-interfering channel control logics. Orthogonal experiments are carried out on the feedforward control parameters of the decoupling compensation strategy. By designing a full-factor or partial-factor orthogonal experimental array, in the actual operation of the platform or the digital twin simulation environment, the system gradually adjusts the feedforward gain of each degree-of-freedom compensation channel, and batch measures and multi-dimensional evaluates indicators such as spatial positioning error, response speed, residual convergence, and platform stability under different parameter combinations. Combining principal component sensitivity analysis, residual distribution statistics, and system limit response constraints, the optimal parameter configuration scheme is selected from all experimental combinations. The orthogonal experiment calibration outputs a set of globally optimal six-dimensional compensation gain parameters, and these coefficients are embedded into the core algorithm of the platform motion control. Based on these compensation gain coefficients, the spatial mapping relationship of the six-dimensional error vector is established, and the error prediction and compensation correction of each target motion instruction in six degrees of freedom are clarified. The spatial mapping relationship is realized through matrix transformation, affine adjustment, or non-linear function fitting, etc., so that the actual movement of the platform always coincides with the theoretical optimal trajectory with high precision. The compensation control instructions are sent to the six-degree-of-freedom drivers of the motion platform to achieve feedforward control compensation. In each control cycle, the expected error is compensated in advance, and at the same time, real-time high-frequency position feedback data is obtained from four independent fields of view of the multi-camera vision system. The data of the four fields of view are fused through a complementary filtering algorithm, using the data of the high-response-frequency field of view to achieve fast state tracking, and at the same time using the data of the low-noise and wide-dynamic-range field of view to improve the anti-interference ability and long-term stability of the feedback signal. The complementary filtering integrates the advantages of multi-channel data, enabling the feedback compensation of the motion platform to capture dynamic fluctuations and suppress occasional interference, forming an adaptive, high-bandwidth, and high-precision closed-loop error correction mechanism.The motion control system performs dynamic fine-tuning based on the filtered fusion feedback and the feedforward compensation instruction residual, suppresses the remaining errors caused by environmental mutations, micro-vibrations or device aging, and finally obtains the multi-camera vision calibration result of the lithography machine.

[0037] The above describes the lithography machine calibration method based on multi-camera vision in the embodiments of the present application. Next, the lithography machine calibration device 10 based on multi-camera vision in the embodiments of the present application will be described. Please refer to Figure 2 , an embodiment of the lithography machine calibration device 10 based on multi-camera vision in the embodiments of the present application includes: An imaging module 11, configured to perform imaging and phase-sensitive detection and analysis on the mask plane and the wafer plane through a four-way optical beam splitting system, and obtain four groups of mask-wafer initial alignment position information; A processing module 12, configured to perform scanned white light interference edge enhancement processing on the four groups of mask-wafer initial alignment position information to obtain three-dimensional surface profile data; A dynamic registration module 13, configured to perform wavelet transform processing on the three-dimensional surface profile data, extract target feature points, and simultaneously perform dynamic registration on the four-way optical beam splitting system according to the target feature points to obtain multi-camera vision feature registration data; An error compensation module 14, configured to establish a comprehensive error model including mechanical errors, optical errors and environmental errors based on the multi-camera vision feature registration data, and perform real-time correction and error compensation on the six-degree-of-freedom motion platform of the lithography machine based on the comprehensive error model to obtain the multi-camera vision calibration result of the lithography machine.

[0038] Through the collaborative cooperation of the above-mentioned various components, the mask plane and the wafer plane are simultaneously imaged and phase-sensitive detected through a four-way optical splitting system, achieving a comprehensive monitoring of the microstructures on the mask and the wafer surfaces, overcoming the drawback of the traditional monocular vision system being vulnerable to local interference, and improving the robustness and reliability of the calibration system. The mask-wafer unmarked alignment technology is adopted, using the features of the product structure itself as natural marks, avoiding the systematic deviation caused by specially making alignment marks, reducing the production cost, enhancing the process flexibility. At the same time, through the sub-wavelength resolution phase-sensitive detection algorithm, the alignment accuracy is significantly improved. Through wavelet transform, multi-scale feature extraction is performed on the three-dimensional surface profile data, and combined with the wavefront sensor to compensate the wave aberration of the optical system in real time, effectively eliminating the influence of environmental vibration and thermal drift on the calibration accuracy, and ensuring that the optical system is always in the best imaging state. A comprehensive error model including mechanical error, optical error and environmental error is established, realizing the precise control of the six-degree-of-freedom motion platform of the lithography machine, significantly reducing the deterministic error and random error of the system, and improving the stability and reliability of the calibration. Through the dual-precision strategy of rough alignment with a 633nm wavelength light source and fine calibration with a 13.5nm EUV light source, combined with the dual-frequency heterodyne interference technology, the influence of the DC noise in the traditional single-frequency interference is effectively avoided, the calibration accuracy is greatly improved, the calibration time is shortened at the same time, and the production efficiency is enhanced. The technical bottleneck of the inconsistency between the traditional off-line calibration and the actual exposure state is broken through, and real-time calibration is achieved without affecting normal exposure, effectively solving the problem of the accumulation of overlay errors, and significantly improving the yield and performance of the final chip.

[0039] Please refer to Figure 3 , Figure 3 FIG. is a schematic block diagram of the structure of the electronic device 300 provided by an embodiment of the present application. The electronic device 300 includes a processor 301 and a memory 302. The processor 301 and the memory 302 are connected through a device bus 303. Among them, the memory 302 may include a non-volatile storage medium and an internal memory.

[0040] The non-volatile storage medium can store a computer program. The computer program includes program instructions. When the program instructions are executed by the processor 301, the processor 301 can be enabled to execute any one of the above-mentioned lithography machine calibration methods based on multi-view vision.

[0041] The processor 301 is used to provide computing and control capabilities to support the operation of the entire electronic device 300.

[0042] The internal memory provides an environment for the operation of the computer program in the non-volatile storage medium. When the computer program is executed by the processor 301, the processor 301 can be enabled to execute any one of the above-mentioned lithography machine calibration methods based on multi-view vision.

[0043] Those skilled in the art can understand that Figure 3 The structure shown in Figure 3 is only a block diagram of some structures related to the solution of this application, and does not constitute a limitation on the electronic device 300 involved in the solution of this application. Specifically, the electronic device 300 may include more or fewer components than those shown in the figure, or combine some components, or have a different component arrangement.

[0044] It should be understood that the processor 301 may be a central processing unit (CPU), and the processor 301 may also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0045] It should be noted that those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working process of the above-described electronic device 300 can refer to the corresponding process of the foregoing multi-view vision-based lithography machine calibration method, which will not be elaborated here.

[0046] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the above-described system, device, and unit can refer to the corresponding processes in the foregoing method embodiments, which will not be elaborated here.

[0047] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing an electronic device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical discs that can store program codes.

[0048] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the various embodiments of the present application.

Claims

1. A calibration method for a lithography machine based on multi-view vision, characterized in that Including: Imaging and phase-sensitive detection analysis are performed on the mask plane and the wafer plane through a four-way optical splitting system to obtain four sets of mask-wafer initial alignment position information; Performing scanning white light interference edge enhancement processing on the four sets of mask-wafer initial alignment position information to obtain three-dimensional surface profile data; Performing wavelet transform processing on the three-dimensional surface profile data, extracting target feature points, and simultaneously performing dynamic registration on the four-way optical splitting system according to the target feature points to obtain multi-view vision feature registration data; Establishing a comprehensive error model including mechanical error, optical error, and environmental error based on the multi-view vision feature registration data, and performing real-time correction and error compensation on the six-degree-of-freedom motion platform of the lithography machine based on the comprehensive error model to obtain the multi-view vision calibration result of the lithography machine.

2. The calibration method of the lithography machine based on multi-view vision according to claim 1, wherein The imaging and phase-sensitive detection analysis are performed on the mask plane and the wafer plane through a four-way optical splitting system to obtain four sets of mask-wafer initial alignment position information, including: Introducing a reference laser with a wavelength of 633nm and an EUV light source with a wavelength of 13.5nm into the four-way optical splitting system, and performing optical path separation through the four-way optical splitting system to obtain separated multi-wavelength light beams; Dividing the separated multi-wavelength light beams through a semi-transparent and semi-reflective beam splitter array to construct four independent optical paths between the mask plane and the wafer plane to obtain four sets of virtual calibration dot matrices; Performing phase control on the four sets of virtual calibration dot matrices through a piezoelectric drive phase modulator to obtain four phase-controllable optical paths; Based on the four phase-controllable optical paths, controlling the mask imaging system on the mask plane and the wafer imaging system on the wafer plane to perform synchronous motion to obtain a multi-view vision imaging basic platform; Based on the multi-view vision imaging basic platform, performing image acquisition on the mask plane and the wafer plane to obtain original multi-view vision images, and performing preprocessing on the original multi-view vision images to obtain multi-view vision image data of four fields of view; Performing phase-sensitive detection analysis on the multi-view vision image data of the four fields of view to obtain four sets of mask-wafer initial alignment position information.

3. The calibration method of the lithography machine based on multi-view vision according to claim 2, characterized in that The phase-sensitive detection analysis is performed on the multi-view vision image data of the four fields of view to obtain four sets of mask-wafer initial alignment position information, including: Performing interference processing on the multi-view vision image data of the four fields of view to obtain four sets of interference images, and performing fast Fourier transform and Hilbert transform on the four sets of interference images to obtain phase detection data; Using the microstructures on the surfaces of the mask and the wafer themselves as natural markers to analyze the phase detection data to obtain a feature correspondence relationship; Performing four-step phase shift on the feature correspondence relationship through a piezoelectric phase modulator to obtain displacement information, and performing adaptive frequency window division and signal averaging processing on the displacement information to obtain target position data; Calculating the target position data through a sub-wavelength resolution algorithm to obtain four sets of mask-wafer initial alignment position information.

4. The calibration method of the lithography machine based on multi-view vision according to claim 1, wherein Performing scanning white light interference edge enhancement processing on the initial alignment position information of the four groups of mask-wafer to obtain three-dimensional surface profile data, including: Vertically scanning the initial alignment position information of the four groups of mask-wafer by a piezoelectric vertical scanning device to obtain scanning data of the mask and wafer surfaces; Performing wavelength continuous change processing on the scanning data of the mask and wafer surfaces to obtain interference images of different wavelengths; Performing noise suppression processing on the interference images of different wavelengths to obtain filtered interference images, and performing edge extraction on the filtered interference images to obtain edge information of the surface microstructure; According to the edge information of the surface microstructure, determining the height data of each point through the maximum contrast position of the interference fringes to obtain three-dimensional surface profile data.

5. The calibration method of a lithography machine based on multi-camera vision according to claim 1, wherein Performing wavelet transform processing on the three-dimensional surface profile data to extract target feature points, and dynamically registering the four-way optical splitting system according to the target feature points to obtain multi-view vision feature registration data, including: Performing multi-layer wavelet decomposition processing on the three-dimensional surface profile data to obtain feature hierarchical data at different scales; Performing background noise suppression processing on the feature hierarchical data through an adaptive threshold to obtain purified multi-scale features, and screening the purified multi-scale features to obtain target feature points; Monitoring the wavefront aberration of the four-way optical splitting system by a Shack-Hartmann wavefront sensor to obtain real-time wavefront aberration data; Based on the real-time wavefront aberration data, controlling a deformable mirror to perform dynamic correction on the four-way optical splitting system to obtain a stable optical imaging path; Registering and matching the target feature points with the stable optical imaging path to obtain multi-view vision feature registration data.

6. The calibration method of a lithography machine based on multi-view vision according to claim 1, wherein Establishing a comprehensive error model including mechanical error, optical error and environmental error according to the multi-view vision feature registration data, and performing real-time correction and error compensation on the six-degree-of-freedom motion platform of the lithography machine based on the comprehensive error model to obtain the multi-view vision calibration result of the lithography machine, including: Extracting the geometric error parameters of the six-degree-of-freedom motion platform of the lithography machine in the three translation directions of X, Y, Z and the three rotation directions of Rx, Ry, Rz from the multi-view vision feature registration data to obtain a mechanical error model; Calculating aberration, distortion and magnification error parameters based on standard test pattern acquisition data to obtain an optical error model, and collecting environmental parameters through a temperature sensor, an air flow velocity sensor and a vibration sensor to construct an environmental error model; Integrating the mechanical error model, the optical error model and the environmental error model to obtain a system comprehensive error model; Performing feedforward control compensation on the six-degree-of-freedom motion platform of the lithography machine based on the system comprehensive error model to obtain the multi-view vision calibration result of the lithography machine.

7. The calibration method of the lithography machine based on multi-view vision according to claim 6, characterized in that Performing feedforward control compensation on the six-degree-of-freedom motion platform of the lithography machine based on the system comprehensive error model to obtain the multi-view vision calibration result of the lithography machine, including: Perform eigen-decomposition on the comprehensive error model of the system to obtain a deterministic component matrix and a stochastic component matrix, and generate error principal component distribution data based on the deterministic component matrix and the stochastic component matrix; Based on the error principal component distribution data, construct independent compensation controllers for six degrees of freedom in the six-degree-of-freedom motion platform of the lithography machine. Each compensation controller corresponds to the translational degrees of freedom of X, Y, and Z and the rotational degrees of freedom of Rx, Ry, and Rz to obtain a decoupling compensation strategy; Perform an orthogonal experiment on the feedforward control parameters of the decoupling compensation strategy to determine the compensation gain coefficients for each degree of freedom, and establish a spatial mapping relationship of the six-dimensional error vector based on the compensation gain coefficients for each degree of freedom to obtain a compensation control command; Perform feedforward control compensation on the six-degree-of-freedom motion platform of the lithography machine based on the compensation control command, and simultaneously collect position feedback data from four fields of view for complementary filtering and error compensation to obtain the multi-machine vision calibration result of the lithography machine.

8. A calibration device for a lithography machine based on multi-view vision, characterized in that, For executing the multi-machine vision-based lithography machine calibration method according to any one of claims 1-7, the multi-machine vision-based lithography machine calibration device includes: An imaging module for imaging and phase-sensitive detection analysis of the mask plane and the wafer plane through a four-way optical beam splitting system to obtain four sets of mask-wafer initial alignment position information; A processing module for performing scanning white light interference edge enhancement processing on the four sets of mask-wafer initial alignment position information to obtain three-dimensional surface profile data; A dynamic registration module for performing wavelet transform processing on the three-dimensional surface profile data to extract target feature points, and simultaneously performing dynamic registration on the four-way optical beam splitting system according to the target feature points to obtain multi-machine vision feature registration data; An error compensation module for establishing a comprehensive error model including mechanical error, optical error, and environmental error based on the multi-machine vision feature registration data, and performing real-time correction and error compensation on the six-degree-of-freedom motion platform of the lithography machine based on the comprehensive error model to obtain the multi-machine vision calibration result of the lithography machine.

9. An electronic device, characterized in that, The electronic device includes: a memory and at least one processor, and instructions are stored in the memory; The at least one processor invokes the instructions in the memory so that the electronic device executes the multi-machine vision-based lithography machine calibration method according to any one of claims 1-7.

Images