Wafer detection method and device, electronic equipment and storage medium

Abstract

The application discloses a wafer detection method and device, an electronic device and a storage medium. The wafer detection method obtains three-dimensional point cloud data by performing field-of-view scanning on a region of interest of a wafer, uses common microstructures in an overlapping region as a height alignment basis to construct a dynamic reference surface with a selected initial reference field of view, calculates and corrects a relative offset in a height direction field by field, and finally generates a three-dimensional point cloud model corresponding to the region of interest and having global height consistency. Compared with related technologies, the application does not require external calibration marks or absolute accuracy of a motion platform, does not rely on two-dimensional image features or general three-dimensional registration methods, is specialized in height direction consistency optimization, significantly improves the accuracy and reliability of a global three-dimensional model, and effectively overcomes the height deviation problem between fields of view caused by factors such as sensor drift and wafer warping.

Description

Technical Field

[0001] This application relates to the field of semiconductor manufacturing technology, specifically to a wafer inspection method, a wafer inspection device, an electronic device, a computer-readable storage medium, and a computer program product. Background Technology

[0002] In semiconductor manufacturing and packaging processes, wafer surfaces are formed with numerous three-dimensional microstructures, such as bumps, copper pillars, and metal pads. High-precision three-dimensional measurement of these microstructures (including their height, shape, and coplanarity) is crucial for defect detection, process monitoring, and reliability assessment. To achieve high-precision monitoring of these microstructure morphologies, three-dimensional (3D) optical inspection technology has become a key component of integrated circuit manufacturing.

[0003] In advanced packaging microbump 3D optical metrology, errors introduced by the metrology technology itself are becoming a bottleneck restricting the improvement of metrology accuracy. As the diameter and height of microbumps decrease to the submicron or even nanometer level, tiny deviations in the metrology system can lead to relative errors on the order of magnitude of the measured object, thus rendering the metrology results worthless for process control.

[0004] Suppressing errors caused by the measurement principle and the optical characteristics of the system itself has become the key to breakthroughs in micro-bump 3D inspection technology. Summary of the Invention

[0005] In view of the lack of the above-mentioned related technologies, the purpose of this application is to disclose a wafer inspection method, a wafer inspection device, an electronic device, a computer-readable storage medium, and a computer program product to solve various problems in the related technologies.

[0006] This application discloses a wafer inspection method, including the following steps:

[0007] A field of view is determined as an initial reference field of view, and a dynamic reference surface is constructed with the height direction of the initial reference field of view. The initial reference field of view is a set of fields of view obtained by performing field-by-field scanning on the region of interest of the wafer according to a preset scanning rule. Each field of view in the set of fields of view has corresponding three-dimensional point cloud data, and there is an overlapping region between adjacent fields of view in the set of fields of view. The overlapping region contains a common microstructure formed by the wafer manufacturing process.

[0008] According to the preset processing order, other fields of view in the field of view set are processed sequentially starting from the initial reference field of view until all fields of view in the field of view set are included in the dynamic reference surface, generating a three-dimensional point cloud model corresponding to the region of interest and having global high consistency.

[0009] For the current field of view to be processed, the following steps are performed: in the overlapping area between the current field of view and the processed adjacent field of view, at least one common microstructure is matched; based on the height measurement values ​​of the at least one common microstructure in the current field of view and the processed adjacent field of view, the relative offset between the two in the height direction is calculated; according to the relative offset, the height direction of the three-dimensional point cloud data of the current field of view is corrected so that its height reference is consistent with the dynamic reference plane; the corrected current field of view is incorporated into the dynamic reference plane and the dynamic reference plane is updated.

[0010] In some embodiments of this application, the wafer inspection method adopts an online real-time processing mode and the preset processing order is the scanning order of the preset scanning rules: during the scanning process, the field of view obtained by the first scan is used as the initial reference field of view, and a dynamic reference plane is constructed; according to the scanning order, each time the scanning of another field of view in the set of fields of view is completed, it is used as the current field of view, and the height direction is corrected based on the dynamic reference plane and the dynamic reference plane is updated.

[0011] In some embodiments of this application, the wafer inspection method includes: performing field-of-view scanning on the wafer to acquire and store data of all fields of view in the field of view set; selecting one field of view from all acquired fields of view as the initial reference field of view and constructing a dynamic reference plane; sequentially reading and processing data of other fields of view in the field of view set from the storage according to a preset processing order until the correction of all fields of view is completed; wherein, the preset processing order is determined based on the scanning order of the preset scanning rules.

[0012] In some embodiments of this application, the initial reference field of view is selected from any of the following: a first field of view acquired according to scanning rules; a field of view located at the geometric center of multiple fields of view; or a field of view located in the central region of the wafer.

[0013] In some embodiments of this application, calculating the relative offset of the two in the height direction includes: matching at least three non-collinear common microstructures in the overlapping region; calculating the height difference of each corresponding common microstructure in two different fields of view; and taking the statistical mean of each height difference as the relative offset in the height direction.

[0014] In some embodiments of this application, before taking the statistical mean, the method further includes: performing a consistency assessment on the height differences of the corresponding common microstructures obtained by calculation, and removing abnormal differences that deviate from the preset height difference threshold; and taking the statistical mean of the remaining height differences as the relative offset in the height direction.

[0015] In some embodiments of this application, the preset scanning rule is determined based on a preset path, which includes at least one of a serpentine path, a spiral path, and a jump path optimized according to the wafer chip layout.

[0016] In some embodiments of this application, the overlapping area between adjacent fields of view is dynamically adjusted during scanning based on the size and / or density of the shared microstructure; when the density of the shared microstructure in a local area is lower than a preset density threshold and / or the size of the shared microstructure is smaller than a preset size threshold, the overlapping area is increased; when the density of the shared microstructure in a local area is higher than a preset density threshold and the size of the shared microstructure is larger than a preset size threshold, the overlapping area is decreased.

[0017] In some embodiments of this application, the wafer inspection method further includes: performing relative alignment of the current field of view with the processed adjacent field of view based on the two-dimensional image features and / or the planar projection features of the three-dimensional point cloud in the overlapping region, so as to improve the matching accuracy of the common microstructure in the overlapping region.

[0018] In some embodiments of this application, the relative alignment in the planar direction includes only translation and / or rotation transformations.

[0019] In some embodiments of this application, the calculation of the relative alignment in the planar direction and the relative offset in the height direction is performed based on the same at least one common microstructure.

[0020] In some embodiments of this application, during the processing, after each predetermined number of fields of view are completed, a local closed-loop verification is performed: at least one closed path is constructed based on the overlap relationship between the multiple fields of view included in the dynamic reference plane, and the cumulative height deviation derived along the closed path is calculated; if the cumulative height deviation exceeds a preset deviation threshold, the height direction offset of each field of view involved in the closed path is optimized and adjusted as a whole.

[0021] In some embodiments of this application, the width of the overlapping region is greater than or equal to twice the maximum feature size of the common microstructure.

[0022] In some embodiments of this application, the common microstructure is a three-dimensional structure formed on the wafer during packaging or front-end processing, including at least one of bumps, copper pillars, metal pads, and test solder joints.

[0023] This application also discloses a wafer inspection device, comprising:

[0024] The three-dimensional measurement module is configured to perform field-of-interest scanning on the wafer according to a preset scanning rule to acquire three-dimensional point cloud data corresponding to multiple fields of interest; the multiple fields of interest constitute a field of interest set, and there are overlapping regions between adjacent fields of interest in the field of interest set, and the overlapping regions contain common microstructures formed by the wafer manufacturing process;

[0025] The reference plane construction module is configured to determine a field of view from the set of fields of view as an initial reference field of view, and to construct a dynamic reference plane with the height direction of the initial reference field of view.

[0026] A field-of-view processing module is configured to process other fields of view in the field-of-view set sequentially, starting from the initial reference field of view, according to a preset processing order. The field-of-view processing module includes: a matching unit configured to match at least one common microstructure within the overlapping area of ​​the current field of view and the processed adjacent fields of view; an offset calculation unit configured to calculate the relative offset in the height direction between the current field of view and the processed adjacent fields of view based on the height measurements of the at least one common microstructure in both fields of view; a correction unit configured to perform height direction correction on the 3D point cloud data of the current field of view according to the relative offset, so that its height reference is consistent with the dynamic reference plane; and a reference plane update unit configured to incorporate the corrected current field of view into the dynamic reference plane and update the dynamic reference plane.

[0027] The model generation module is configured to generate a 3D point cloud model corresponding to the region of interest and having global high consistency after all fields of view in the field of view set are included in the dynamic reference plane.

[0028] In some embodiments, the three-dimensional measurement module includes: a two-dimensional image acquisition unit for acquiring a two-dimensional image of the wafer surface; and a three-dimensional topography scanning unit for acquiring three-dimensional height data corresponding to the two-dimensional image space.

[0029] In some embodiments, the three-dimensional topography scanning unit employs a white light interferometer, a confocal microscope, or a structured light scanning system.

[0030] This application discloses another electronic device, including:

[0031] processor;

[0032] Memory, which stores computer programs;

[0033] The computer program is executed by the processor to perform the wafer inspection method as described above.

[0034] This application further discloses a computer-readable storage medium storing program instructions, which, when executed, perform the wafer inspection method as described above.

[0035] This application also discloses a computer program product, including a computer program that, when executed by a processor, implements the wafer inspection method as described above.

[0036] This application discloses a wafer inspection method, a wafer inspection device, an electronic device, a computer-readable storage medium, and a computer program product. The wafer inspection method acquires 3D point cloud data of the region of interest (ROI) on the wafer through field-of-view scanning. It utilizes common microstructures formed by the manufacturing process within the overlapping region as a height alignment basis, constructs a dynamic reference plane using a selected initial reference field of view, calculates and corrects the relative offset in the height direction for each field of view, incorporates the corrected field of view into the dynamic reference plane, and continuously updates it, ultimately generating a 3D point cloud model corresponding to the ROI and possessing global height consistency. Furthermore, it supports flexible scanning paths during the inspection process, introduces local closed-loop verification to suppress accumulated errors, and can dynamically adjust the overlapping region according to the microstructure density and size. Compared to related technologies, this application does not require external calibration markers or absolute accuracy of the motion platform, does not rely on 2D image features or general 3D registration methods, specializes in height direction consistency optimization, significantly improves the accuracy and reliability of the global 3D model, and effectively overcomes the problem of height deviation between fields of view caused by sensor drift, wafer warping, and other factors. Attached Figure Description

[0037] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0038] Figure 1 The diagram shown is a flowchart of one embodiment of the wafer inspection method of this application.

[0039] Figure 2 The diagram shown is a layout plan view of the three-dimensional measurement module provided in this application in one embodiment.

[0040] Figure 3 The diagram shown is a schematic diagram of the optical path principle of the three-dimensional measurement module provided in this application in one embodiment.

[0041] Figure 4 The diagram shown illustrates a wafer surface scanning process performed by the three-dimensional measurement module provided in this application in one embodiment.

[0042] Figure 5 Displayed as Figure 4 A schematic diagram of the layout of the three-dimensional measurement module scanning the wafer surface to form multiple fields of view.

[0043] Figure 6 and Figure 7 The diagram shows a field-view set obtained by performing field-view scanning on a wafer according to a preset scanning rule. Figure 6 The image shown is a schematic diagram of a full wafer scan. Figure 7 This is a schematic diagram of a partition scan.

[0044] Figure 8 Displayed as Figure 1 A further flowchart of step S13.

[0045] Figure 9 The diagram shown is a schematic representation of the wafer inspection device of this application in one embodiment.

[0046] Figure 10 The diagram shown is a structural schematic of the electronic device of this application in one embodiment. Detailed Implementation

[0047] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0048] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a particular order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented, for example, in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “corresponding,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0049] In this embodiment, the technical solution for realizing three-dimensional measurement can refer to the wafer surface three-dimensional measurement device disclosed in Chinese patent application CN202311016168.7 (application date: October 13, 2023). The acquisition of three-dimensional optical measurement data (three-dimensional point cloud data) is performed by a multi-dimensional structured light system driven by a mechanical displacement stage, acquiring data from the wafer along a preset path in a field-of-view (FOV) manner, and then synthesizing the FOV data through a data processing center to obtain the three-dimensional point cloud data of the field of view. Since the field of view range of a single acquisition is limited, it is necessary to combine multiple fields of view to cover the entire region of interest (ROI).

[0050] However, in actual measurement processes, when the same wafer sample is repeatedly sampled from multiple locations and angles, significant height differences are observed in the measurement data of the same three-dimensional microstructure (such as a bump) in different regions of the same field of view. Even with only a small displacement (e.g., 1 mm) between adjacent FOVs, a perceptible deviation in the height measurement value of the three-dimensional microstructure is observed; when the displacement increases to 5 mm or more, the height difference becomes even more pronounced. It should be clarified that this height inconsistency does not originate from changes in the actual morphology of the wafer sample being measured, but rather from the accumulation of inherent errors in the measurement system (such as optical distortion, illumination non-uniformity, etc.). Essentially, the system lacks a unified and stable Z-axis height reference across the field of view, leading to inconsistent readings for three-dimensional microstructures in the same physical location due to differences in acquisition conditions.

[0051] Through careful research, the inventors of this application discovered that due to the combined effects of factors such as minute vibrations or deflections of the mechanical displacement stage during movement, field curvature errors of the optical system, and the non-absolute parallelism between the camera optical axis and the stage, a systematic inconsistency in the Z-axis height reference generally exists between different fields of view. This inconsistency in the Z-axis height reference directly leads to significant height deviations in the original measurement data of physically identical feature points at the same location (e.g., three-dimensional microstructures) in different fields of view, fundamentally disrupting the local consistency of the measurement results.

[0052] Furthermore, due to the non-ideal characteristics of optical imaging systems, such as lens distortion, light intensity attenuation, depth-of-field variations, and uneven distribution of structured light projection, significant differences in imaging quality exist between 3D microstructures located in the central and peripheral regions of a single field of view. This results in systematic deviations in the Z-axis height values ​​measured for the same 3D microstructure at different field-of-view locations. Such deviations not only disrupt the spatial continuity and geometric fidelity of the 3D reconstruction results but also easily lead to misjudgments of the height of critical 3D microstructures (such as bumps), severely weakening the reliability and practicality of wafer 3D inspection in defect identification, process monitoring, and yield analysis.

[0053] To address the aforementioned issues, some existing technologies attempt to improve the situation in the following ways: some solutions rely on external calibration components or manually set reference height structures to uniformly calibrate the height data of different fields of view; some solutions use global plane fitting, global least squares optimization, or three-dimensional registration algorithms (e.g., iterative nearest point ICP algorithm) to perform one-time alignment processing on the data of all fields of view.

[0054] However, the above solutions still have shortcomings: for example, external calibration components or artificial reference structures increase the complexity of the inspection process and are difficult to truly reflect the height distribution of the wafer surface under actual process conditions; while methods based on global optimization or three-dimensional registration are often computationally complex, sensitive to initial conditions, and usually introduce coordinate transformations in multiple X, Y, and Z directions at the same time, which can easily mask or amplify local height errors, making it difficult to accurately measure the absolute height of three-dimensional microstructures.

[0055] In view of this, this application discloses a wafer inspection method, a wafer inspection device, an electronic device, a computer-readable storage medium, and a computer program product. The wafer inspection method acquires three-dimensional point cloud data of each region of the wafer by scanning in sections. It uses the common microstructures in the overlapping areas of adjacent fields of view as a height alignment reference to calculate the relative height offset between adjacent fields of view. Based on this, it performs field-by-field correction on the three-dimensional point cloud data of subsequent fields of view, and finally unifies all fields of view to a globally consistent height reference plane, thereby eliminating errors between fields of view and improving the accuracy and reliability of wafer three-dimensional topography measurement.

[0056] The wafer inspection method provided in this application will be described in detail below with reference to specific embodiments.

[0057] Please see Figure 1 The diagram shows a flowchart of the wafer inspection method of this application in one embodiment.

[0058] like Figure 1 As shown, the wafer inspection method includes the following steps:

[0059] Step S11: Determine a field of view as the initial reference field of view, and construct a dynamic reference surface with the height direction of the initial reference field of view.

[0060] In some embodiments, the initial reference field of view is a set of fields of view obtained by performing field-by-field scanning on the wafer according to a preset scanning rule.

[0061] In the field of semiconductor manufacturing and metrology, wafers are relatively large (e.g., the mainstream 8-inch and 12-inch wafers), and their surfaces typically contain numerous intricate microstructures. Because high-resolution cameras have limited single-shot imaging range, they cannot cover the entire wafer or a large area of ​​the wafer at once while simultaneously meeting micrometer- or nanometer-level resolution requirements. Therefore, a "field of view (FOV) scanning" metrology method must be employed. This method systematically divides the entire wafer or a designated region of interest on it into multiple ordered and partially overlapping local regions, i.e., fields of view (FOVs), and sequentially acquires three-dimensional topographic data for each FOV.

[0062] In this embodiment, the data acquisition process is performed by an integrated 3D measurement module. Specifically, in some embodiments, the 3D measurement module generally includes a 2D image acquisition unit and a 3D topography scanning unit, which work together to acquire high-precision 3D point cloud data within each field of view.

[0063] Please see Figure 2 and Figure 3 , Figure 2 The diagram shown is a layout plan view of the three-dimensional measurement module provided in this application in one embodiment. Figure 3 The diagram shown is a schematic representation of the optical path principle of the three-dimensional measurement module provided in this application in one embodiment. Wherein, Figure 2 The image shows the positional relationship between the 2D image acquisition unit and the 3D topography scanning unit relative to the wafer under test, the coverage of a single field of view, and the gridded scanning path planning for the region of interest in the wafer. Figure 3 The diagram illustrates how the two-dimensional image acquisition unit and the three-dimensional shape scanning unit work together to acquire two-dimensional image data and three-dimensional height data.

[0064] Combination Figure 2 and Figure 3 The three-dimensional measurement module is mainly composed of a two-dimensional image acquisition unit and a three-dimensional topography scanning unit, which work together on the surface of the wafer 100 being measured.

[0065] The 2D image acquisition unit can employ a high-resolution 2D area scan camera 201, positioned directly above wafer 100, and, in conjunction with a high numerical aperture objective lens and illumination system, efficiently capture high-contrast 2D images of the wafer surface. The field of view of the 2D area scan camera 201 covers a local area of ​​the wafer, typically using a grid-like scanning method. The illumination system can support, for example, ring lighting or coaxial lighting, significantly enhancing the edge contrast between bumps and the background, and improving the reliability of feature extraction.

[0066] The three-dimensional topography scanning unit can employ non-contact optical three-dimensional measurement technology, such as a white light interferometer, a laser confocal microscope, or a structured light scanning system. Taking a structured light scanning system as an example, the system can include a structured light projection module 203. In practical applications, to overcome problems such as abrupt changes in bump height, dense arrangement, and local occlusion and measurement failures caused by specular reflection from the metal surface, multiple structured light projection modules 203 can be configured in the structured light scanning system. These modules are distributed around the outer periphery of the wafer 100 under test at an angle to the XY plane. Here, we can refer to multiple structured light projection modules at different locations as multi-dimensional structured light projection modules.

[0067] In practical applications, to meet the inspection requirements of large-size wafers, 3D metrology modules are typically equipped with high-precision motion platforms that support multi-field-of-view scanning. Specifically, the 2D image acquisition unit (e.g., a 2D area array camera) and the 3D topography scanning unit (e.g., a multi-dimensional structured light projection module) in the 3D metrology module are mounted on the motion platform at fixed relative positions and angles. Their optical axes are precisely calibrated to ensure a high degree of overlap between the imaging and projection areas. By controlling the movement of the motion platform through a program, the 2D area array camera and the multi-dimensional structured light projection module are driven to move position by position along a predetermined path to complete the imaging of the region of interest in the wafer under test.

[0068] It should be noted that, by Figure 2 and Figure 3 It can be seen that the projection areas of the two-dimensional area array camera 201 in the two-dimensional image acquisition unit and the structured light projection modules 203 in the three-dimensional topography scanning unit overlap in space and their focal planes are also consistent. They share the same field of view and have strict consistency in space, thus ensuring that each pixel in the two-dimensional image can accurately establish a one-to-one mapping relationship with the corresponding point in the three-dimensional height data.

[0069] The 3D measurement module, through the collaborative operation of a 2D area array camera and a multi-dimensional structured light projection module, simultaneously acquires strictly spatially corresponding 2D image data and 3D height data during the measurement process. Under uniform illumination (e.g., white LED light source), the 2D area array camera acquires high-resolution 2D image data, clearly reflecting the planar contour, position, and texture features of the bumps. Simultaneously, the multi-dimensional structured light projection module projects an coded structured light pattern (e.g., phase-shifted sinusoidal fringe light, Gray code pattern, etc.) onto the surface of the wafer under test. After modulation by the 3D topography of the bumps, the deformed fringe image is simultaneously acquired by a camera at an angle to the structured light projection module. Based on algorithms such as phase measurement profilometry, combined with pre-calibrated system parameters, the fringe phase information is calculated to obtain the 3D coordinates (X, Y, Z) corresponding to each pixel, thereby generating a dense 3D point cloud (X, Y) corresponding to the pixel level of the 2D image, where the Z coordinate is the 3D height data. This mechanism ensures an inherent and deterministic spatial mapping relationship between the 2D image data and the 3D height data, laying a solid foundation for subsequent processing.

[0070] Thus, by utilizing the collaborative integration of the two-dimensional image acquisition unit and the three-dimensional shape scanning unit in the three-dimensional measurement module, synchronous acquisition of two-dimensional texture and three-dimensional shape with strict spatial correspondence is achieved.

[0071] The entire 3D measurement module is mounted on a high-precision motion platform. The motion platform is controlled by a program to drive the 2D area array camera and the multi-dimensional structured light projection module to move position by position along a predetermined path to complete the imaging of the region of interest.

[0072] The scanning process follows preset scanning rules determined by a preset path, which plans the order and spatial relationships of field-of-view acquisition. Common preset paths include serpentine paths, spiral paths, or jump paths optimized based on wafer chip layout.

[0073] Taking a serpentine path as an example, the field of view is acquired sequentially along the row direction starting from one side of the wafer. After reaching the boundary of the other side, it turns back and acquires the field of view sequentially along the row direction from the other side until it reaches the boundary of one side again, and so on, row by row, until the entire area of ​​the wafer or the specified region of interest is covered. The serpentine path has the advantages of simple motion control logic, relatively optimal total path length, and easy achievement of full coverage, and is suitable for the detection of most regular-shaped areas.

[0074] Taking a spiral path as an example, it starts from the center of the wafer or a designated area and gradually expands outward along a spiral trajectory. Using a spiral path naturally achieves continuous scanning from the center outward, which helps to control the transmission of the measurement reference from low-deformation areas (e.g., the wafer center is usually more stable) outward, reduces accumulated errors, and is suitable for measurements that are highly dependent on the center reference, or for optimized scanning of circular areas.

[0075] Taking the jump path optimized based on wafer chip layout as an example, it no longer mechanically traverses all possible positions. Instead, it directly "jumps" to the area with chips for field acquisition based on the actual chip (die) distribution map on the wafer or a predefined area of ​​interest, skipping blank areas without chips (such as edges and scribe lines). This scanning method can significantly improve scanning efficiency and reduce empty scan time, making it particularly suitable for sampling inspection, key area detection, or re-inspection of known patterns.

[0076] In practical applications, the selection of scanning paths is not static but rather involves comprehensive judgment and flexible configuration based on various factors such as the specific inspection target, efficiency requirements, wafer process status, and equipment platform characteristics. Systems typically support software parameterization settings or automatic decision-making by more advanced inspection strategy modules to select the optimal single or combined path strategy.

[0077] For example, if 100% wafer-wide inspection is required, a serpentine path is typically chosen to ensure no omissions and for its simple and reliable logic. If only monitoring specific process areas or performing sampling inspections is needed, a jump path based on chip layout can significantly improve efficiency. Alternatively, for example, for critical measurements sensitive to deformation in the central region, a spiral path can be used to establish a stable central reference, while a serpentine path can be switched to for the peripheral areas to increase scanning speed. This combination balances reference accuracy and overall efficiency. Or, for example, if high warpage or concentrated defects are known at the wafer edges, a serpentine path with greater adaptive overlap can be used at the edges to ensure inspection quality, while a jump path can be used for rapid passage through known good areas.

[0078] This configurable and combinable path strategy enables the system to intelligently adapt to diverse production scenarios and stringent process control requirements, maximizing equipment output efficiency while ensuring 3D data quality and detection accuracy.

[0079] In practical applications, to ensure that a complete, continuous and high-precision 3D topographic image of the wafer can be generated, a sufficiently wide overlap area must be set between adjacent fields of view when data is acquired according to the preset scanning rules.

[0080] Please see Figure 4 The image shows a schematic diagram of a wafer surface scanning process performed by the three-dimensional measurement module provided in this application in one embodiment. Figure 4As shown, a 2D area array camera 201 is positioned directly above wafer 100 to acquire 2D images of the wafer surface. A structured light projection module 203 is aligned with the wafer surface and projects a specifically coded structured light pattern to achieve 3D topography acquisition. Wafer 100 can be divided into multiple regularly arranged fields of view, represented by a dashed grid, with each field of view corresponding to one scanning acquisition process. Overlapping regions are provided between adjacent fields of view, containing common microstructures formed by the wafer manufacturing process. This facilitates height correction between fields of view and the construction of dynamic reference surfaces based on these common microstructures. The preset scanning rules followed by the scanning are determined based on a preset path, ensuring coverage of the entire region of interest while maintaining continuity and consistency between fields of view.

[0081] The width of the overlapping region is typically set based on the size of the target feature on the wafer. For example, it needs to be larger than the maximum feature size of the 3D microstructure under test to ensure that the same physical feature can appear completely within the imaging range of two adjacent fields of view. For instance, the width of the overlapping region is greater than or equal to twice the maximum feature size of the shared microstructure. For example, for bumps with sizes from 3 micrometers (μm) to 10 micrometers, the overlapping region width can be set to 20 to 30 micrometers; for bumps with sizes from 50 to 100 micrometers, the overlapping region width can be set to 200 micrometers. In short, the width of the overlapping region can be flexibly set according to the shared microstructure.

[0082] Please see Figure 5 Displayed as Figure 4 A schematic diagram showing the layout of a 3D measurement module scanning the wafer surface to form multiple fields of view. (Example) Figure 5 As shown, an overlapping region 102 is provided between adjacent fields of view 101, which is spatially covered by two or more adjacent fields of view. These overlapping regions are not randomly set, but are carefully designed based on preset scanning rules to ensure that the overlapping region 102 contains a sufficient number and reasonable distribution of common microstructures B (e.g., bumps) formed by wafer manufacturing processes, providing a stable and reliable matching basis for height direction correction between fields of view, thereby achieving high-precision alignment and height consistency correction between fields of view.

[0083] Specifically, each field of view corresponds to the data range acquired in one structured light 3D imaging process. As shown in the figure, the solid box on the left and the dashed box on the right represent two adjacent fields of view, which overlap in the horizontal direction. This overlapping area contains common microstructures formed by wafer manufacturing processes, such as regularly arranged bumps, vias, and pads. These 3D microstructures have stable geometric shapes and positional relationships, and can be repeatedly observed in different fields of view, thus serving as reliable cross-field-of-view matching reference points.

[0084] Since these shared microstructures are physically the same entity, their true three-dimensional coordinates remain unchanged in global space. Therefore, when they are measured in different fields of view, their height values ​​in each field of view should theoretically be consistent.

[0085] However, due to factors such as mechanical positioning errors, light intensity attenuation, optical distortion, and systematic Z-axis reference drift, the original measurement data may exhibit local height deviations. By identifying and matching these common microstructures within overlapping areas, the relative height offset between adjacent fields of view can be calculated. This allows for height orientation correction of the current field of view's 3D point cloud data, ensuring it aligns with the established dynamic reference surface.

[0086] Furthermore, this overlapping design not only supports direct matching between pairs of fields of view, but also provides continuity assurance for constructing a globally consistent model of multi-field fusion. As the scanning sequence progresses, each subsequent field of view is corrected by overlapping with its preceding field of view, gradually incorporating all fields of view within the entire region of interest into a unified height reference system, ultimately generating a 3D point cloud model with global height consistency.

[0087] Furthermore, to improve scanning efficiency and ensure matching reliability, this method also supports dynamically adjusting the overlap area between adjacent fields of view during the scanning process based on the real-time detected distribution of common microstructure features. Specifically, while scanning the current field of view, the size and distribution density of common microstructures (e.g., bumps) in the point cloud data of that current field of view are analyzed simultaneously. Based on this analysis, the position of the next field of view to be scanned is dynamically planned, thereby intelligently adjusting its overlap width with the current field of view. This adjustment is achieved directly by controlling the next moving target position of the motion platform, essentially performing online, adaptive fine-tuning of the preset scanning path.

[0088] In some embodiments, the adjustment strategy includes: when the point cloud density of shared microstructures within a local area is lower than a preset density threshold (e.g., insufficient feature points per unit area), and / or the average size of the microstructures is smaller than a preset size threshold (e.g., the feature size is too small, resulting in a low measurement signal-to-noise ratio), it indicates that the region has few or unstable anchor points available for reliable matching. In this case, the overlapping area of ​​the next field of view will be automatically increased (e.g., from the default 2 times the feature size to 3 times or more) to capture more shared features and ensure the robustness of subsequent registration. Conversely, when the density of shared microstructures within a local area is higher than a preset density threshold and its average size is greater than a preset size threshold, it indicates that the feature points are abundant and stable. Thus, while meeting the minimum matching requirements, the overlapping area of ​​the next field of view can be appropriately reduced, thereby reducing redundant data acquisition and improving overall scanning efficiency. This dynamic adjustment mechanism is achieved by combining a scan path planning algorithm with real-time image analysis feedback, making the size of the overlapping area no longer a fixed value, but a variable that is adaptively optimized based on the actual feature distribution of the wafer surface. This maximizes the intelligence and economy of the scanning process while ensuring the accuracy of 3D reconstruction.

[0089] In some embodiments, the set of fields of view obtained through the above scanning has corresponding 3D point cloud data representing the 3D topography of the local region for each field of view. Two spatially adjacent fields of view within this set of fields of view share an overlapping region, i.e., an overlapping region. This overlapping region is not simply image redundancy, but carries crucial spatial registration information. Specifically, this overlapping region contains physically identical 3D microstructures formed by wafer fabrication processes in adjacent fields of view, such as bumps, copper pillars, metal pads, or test solder joints. These 3D microstructures are represented in the point cloud data as identifiable and matchable 3D feature points or feature clusters. In the following description, the 3D microstructures contained within the overlapping region are also referred to as shared microstructures, indicating that they are shared by two adjacent fields of view within the overlapping region.

[0090] In subsequent processing, by matching the inherent common microstructures within the overlapping region, the relative positions and attitude deviations of adjacent fields of view in three-dimensional space can be accurately calculated, particularly the relative offset in the height direction. These common microstructures act as "anchor points" connecting different fields of view, enabling precise spatial transformation correction of one field of view as a reference, thereby unifying the three-dimensional data of all fields of view into the same global coordinate system. Therefore, the overlapping region and its contained common microstructures provide indispensable geometric constraints and registration basis for achieving global consistent reconstruction from local measurement in this application.

[0091] As mentioned above, in step S11, a field of view is determined as the initial reference field of view, and a dynamic reference surface is constructed with the height direction of the initial reference field of view.

[0092] In the specific embodiments of the wafer inspection method disclosed in this application, the initial reference field of view does not exist independently; it originates from a complete or partitioned data acquisition process. Specifically, the initial reference field of view is one member of a set of fields of view obtained by performing field-by-field scanning on the wafer according to a preset scanning rule.

[0093] It should be noted that the field of view set refers to the entirety of all individual 3D point cloud data acquired according to the selected path rules in a single, ordered scanning task. In practical applications, the field of view set can be flexibly defined according to the detection requirements.

[0094] For example, when performing full-wafer 3D topography inspection, the entire wafer surface is scanned by means of serpentine paths, spiral paths, etc., and the obtained field of view constitutes a global field of view set.

[0095] Taking a serpentine path as an example, in... Figure 6 In the schematic diagram shown, a single field of view 14 is formed by the scanning module (including a 2D area array camera and a multi-dimensional structured light projection module). During data acquisition, the wafer stage carries the wafer 12 to be inspected, and the motion platform drives the scanning module along... Figure 6 The serpentine scanning motion is indicated by the middle arrow line R1 (i.e., the serpentine path) until the entire wafer has been scanned. The obtained field of view constitutes a global field of view set.

[0096] For example, in the scenario of partition detection or key area detection, the wafer can be divided into multiple independent regions of interest. A scan is performed on each region of interest, and all the fields of view corresponding to each region of interest form a local field of view set. For the entire wafer, there are multiple local field of view sets.

[0097] Taking a serpentine path as an example, in... Figure 7 In the schematic diagram shown, the wafer can be divided into an upper region of interest (ROI) and a lower region of interest (ROI), with a single field of view 14 formed by the scanning module (including a 2D area array camera and a multi-dimensional structured light projection module). The wafer stage carries the wafer 12 to be inspected. When data is acquired for the upper inspection region, the motion platform drives the scanning module along... Figure 7 The scanning module moves in a serpentine pattern, as indicated by the middle arrow R2 (i.e., the serpentine path), until it traverses the upper half of the wafer's detection area. The entire field of view obtained constitutes a local field of view set for the upper half of the detection area. When acquiring data for the lower half of the detection area, the motion platform drives the scanning module along... Figure 7The serpentine scanning path, indicated by the middle arrow R3, is used until the lower half of the wafer's detection area is completely scanned. The resulting field of view constitutes a local field of view set for the lower half of the detection area. Of course, in other examples, the wafer can be divided into more regions of interest depending on the actual situation.

[0098] Regardless of the range of the set, the initial reference field of view is selected from within the current set of fields of view to be processed, and a dynamic reference plane is constructed starting from its own height direction. This dynamic reference plane can be dynamically expanded and updated.

[0099] In some embodiments, the initial reference field of view is the first field of view selected from the corresponding set of fields of view; that is, the field of view first scanned in the wafer or region of interest is used as the initial reference field of view.

[0100] In some embodiments, the initial reference field of view is selected from all fields of view in the corresponding set of acquired fields of view.

[0101] Furthermore, regarding the initial reference field of view, in the embodiments of this application, the selection strategy for the initial reference field of view is flexible and can be determined from the set of fields of view obtained by scanning, based on specific detection requirements, algorithm design, or the intention to control error propagation.

[0102] In some embodiments, the first field of view acquired according to the scanning rules can be selected as the initial reference field of view. That is, the starting field of view in the preceding scanning path is used as the source of the height reference. This approach is simple to implement, requires no additional field of view localization calculations, conforms to the pipeline logic of "scan first, process first," and is suitable for scenarios with high requirements for processing timing and algorithm simplicity.

[0103] In some embodiments, a field of view located at the geometric center of multiple fields of view can be selected as the initial reference field of view. That is, by calculating the geometric center of the spatial coordinates of multiple fields of view in the current field of view set, the field of view closest to that center point is selected as the reference. In this way, constructing a height reference starting from the center of the spatial distribution helps to mathematically balance the propagation path of subsequent spatial transformation corrections, thereby optimizing the convergence and consistency of global height data as a whole.

[0104] In some embodiments, a field of view located in the central region of the wafer can be selected as the initial reference field of view. Generally, during wafer manufacturing, the physical center region of the wafer typically exhibits better morphological stability and less inherent deformation. Selecting the field of view of this central region as the reference essentially establishes the reference on a theoretically more stable and reliable physical reference region. This helps reduce the initial reference plane distortion introduced by wafer warping or stress distribution, providing support for establishing a highly reliable global reference plane.

[0105] Of course, the above embodiments are merely illustrative examples, and the strategy for selecting the initial reference field of view can be further extended and varied to adapt to different detection targets and accuracy control requirements.

[0106] Step S13: Following the preset processing order, process other fields of view in the field of view set sequentially, starting from the initial reference field of view, until all fields of view in the field of view set are incorporated into the dynamic reference plane, generating a three-dimensional point cloud model with global high consistency.

[0107] In practical applications, the processing methods for each field of view in the field of view set also differ.

[0108] In some embodiments, the wafer inspection method employs an online real-time processing mode and the preset processing order is the scanning order of a preset scanning rule.

[0109] Specifically, during the scanning process, the field of view obtained from the first scan is used as the initial reference field of view, and a dynamic reference plane is constructed. Subsequently, according to the scanning order, each time the scan of another field of view in this set of fields of view is completed, that is, it is used as the current field of view, and the height direction is corrected based on the dynamic reference plane and the dynamic reference plane is updated.

[0110] In some embodiments, the wafer inspection method employs an offline processing mode.

[0111] Specifically, the wafer is scanned by field of view to acquire and store data from all fields in the field of view set. One field of view is selected from all acquired fields of view as the initial reference field of view, and a dynamic reference plane is constructed. Subsequently, according to a preset processing order, data from other fields of view in the field of view set are sequentially read from storage and processed until the calibration of all fields of view is completed.

[0112] It is worth noting that in this offline mode, the preset processing order can be determined based on the scanning order of the preset scanning rules, or optimized and adjusted based on the scanning order.

[0113] For example, the preset processing order is consistent with the scanning order of the preset scanning rules. That is, data is read and processed sequentially according to the order in which each field of view is scanned. This processing logic directly maps the spatial traversal relationship during data acquisition, enabling the height correction process to proceed along a clear physical path, making it easy to implement and logically coherent.

[0114] However, thanks to the fact that all data is pre-acquired and stored in offline mode, the preset processing sequence is not limited to strict synchronization with the original scanning sequence. Other processing sequences based on optimization objectives can also be selected. These processing sequences are essentially derived from or related to the spatial relationships established by the preset scanning rules. For example, the preset processing sequence can be optimized into a spatial expansion sequence, that is, expanding the processing from the initial reference field of view to the outer field of view step by step according to the principle of spatial proximity. For example, in order to actively assess and correct potential error accumulation, a processing sequence that can connect multiple processed fields of view to form a local closed loop can be specially arranged. In this way, by calculating the sum of the height correction along the loop (which should theoretically be zero), the closure error can be detected and adjusted, thereby improving the overall accuracy. These processing sequences are essentially derived from the original scanning path, but their specific execution sequences can be adaptively adjusted and defined based on them.

[0115] In step S13, based on the determined initial reference field of view and the preset processing order, the following steps are performed sequentially on other fields of view in the field of view set to unify their height reference to the dynamic reference plane corresponding to the initial reference field of view.

[0116] Please see Figure 8 The diagram shows the process flow of height alignment and dynamic reference plane update for the current field of view in step S13.

[0117] like Figure 8 As shown, the height alignment and dynamic reference plane update process for the current field of view includes the following steps:

[0118] Step S201: For the current field of view, locate the overlapping area between the current field of view and the adjacent fields of view that have been included in the dynamic reference plane.

[0119] Based on the known spatial relationship between the current field of view and one or more adjacent fields of view in the dynamic reference plane, three-dimensional point sets corresponding to the same overlapping area in physical space are extracted from the three-dimensional point cloud data of the two.

[0120] For example, if the current field of view is FOV_i, locate the overlapping area between the current field of view_i and the previous field of view_{i-1} that has been included in the dynamic reference plane.

[0121] Step S203: Match at least one total microstructure within the overlapping region.

[0122] Point cloud analysis is performed on the overlapping regions of the extracted current field of view and the overlapping regions of adjacent fields of view to identify and match the physically identical common microstructures formed by wafer manufacturing processes.

[0123] In some embodiments, the common microstructure is a three-dimensional structure formed on the wafer during packaging or front-end processes, including at least one of bumps, copper pillars, metal pads, and test solder joints.

[0124] A sufficiently wide overlap region must be provided between adjacent fields of view to ensure that the entire common microstructure is covered, rather than only a portion of it. In some embodiments, the width of the overlap region is greater than or equal to twice the largest feature size of the common microstructure.

[0125] In addition, in step S203, in some embodiments, at least three non-collinear common microstructures are matched within the overlapping region.

[0126] Selecting at least three non-collinear shared microstructures is crucial for ensuring the robustness, accuracy, and geometric sufficiency of height offset calculations. Individual microstructure features cannot be statistically verified and are highly susceptible to noise or local defects, leading to erroneous estimates. Two microstructure features can only define a straight line; while height difference calculation is possible, it cannot effectively assess or resist outliers during the matching process. More importantly, using only one or two microstructure features cannot verify whether these matching features themselves are reasonably sampled from the same physical plane. Matching at least three non-collinear shared microstructures, however, geometrically constitutes a tiny plane or spatial triangle, whose spatial distribution contains richer information about geometric constraints.

[0127] In practical applications, more than three candidate shared microstructure matching pairs are identified within the overlapping regions of the current field of view and adjacent fields of view. From these candidate shared microstructure matching pairs, priority is given to screening or ensuring that the point pairs used for final calculation contain at least three non-collinear shared microstructures, or these at least three non-collinear shared microstructures can be formed into a subset.

[0128] In some embodiments, to further improve the accuracy and robustness of subsequent common microstructure matching, the current field of view and the adjacent field of view can be relatively aligned in a plane direction perpendicular to the height direction (i.e., the XY plane) before performing the height direction offset calculation. Specifically, this plane alignment can be based on two-dimensional image features (e.g., grayscale gradient, corner points, edge features) and / or planar projection features of the three-dimensional point cloud (e.g., density distribution, contour shape) within the overlapping area. By aligning these planar features, minute planar displacements or rotations caused by mechanical positioning errors or scanning path deviations can be effectively compensated, thereby making the common microstructures in the two fields of view coincide more accurately in space, significantly improving the success rate and accuracy of subsequent three-dimensional matching.

[0129] In some embodiments, the relative alignment in the planar direction involves only translation and / or rotation transformations, without scaling or non-rigid deformation, in order to maintain the authenticity of the microstructure geometry and thus ensure the accuracy of subsequent height measurement and defect determination, which is different from general elastic registration methods that distort local features.

[0130] Specifically, in some implementations, the relative alignment in the planar direction and the subsequent calculation of the relative offset in the height direction are performed based on the same set of shared microstructures. That is, after completing the planar pre-alignment, the height difference is extracted and statistically analyzed using the shared microstructures. This approach ensures that the geometric constraints used for planar and height correction originate from the same set of physical features, enhancing the consistency and cohesion of the entire correction process.

[0131] The step of performing relative alignment of the current field of view with the processed adjacent fields of view in the planar direction can be performed either before or after step S203.

[0132] Step S205: Based on the height measurements of at least one total microstructure in the current field of view and the processed adjacent fields of view, calculate the relative offset between the two in the height direction.

[0133] As mentioned above, in some embodiments, matching at least three non-collinear shared microstructures within the overlapping region results in at least three shared microstructure matching pairs within the overlapping region of the current field of view and the adjacent field of view. For each successfully matched shared microstructure pair, the height measurement value of the shared microstructure in the point clouds of the two fields of view is obtained, and the difference between the two is calculated, i.e., the relative offset between the two in the height direction is obtained. In this way, multiple relative offsets (i.e., multiple height differences) consistent with the number of shared microstructure matching pairs can be obtained.

[0134] Subsequently, the statistical mean of each height difference can be taken as the relative offset in the height direction.

[0135] Furthermore, when there are multiple non-collinear height differences, before taking the statistical mean, the method includes: performing a consistency assessment on the height differences of each corresponding common microstructure and removing abnormal differences that deviate from the preset height difference threshold.

[0136] In some embodiments, firstly, height difference data corresponding to all successfully matched common microstructures are collected, and the median value of these height difference data is calculated. This median value can represent the central tendency of the height difference data and is not easily affected by extreme values.

[0137] To ensure the high reliability and consistency of the calculated height offset, it is necessary to perform consistency evaluation and outlier removal on the multiple height differences obtained from the matching.

[0138] In practice, firstly, height difference data corresponding to all successfully matched common microstructures are collected, and the median of this set of difference data is calculated. This median is the value in the middle of the sorted values ​​(if the height difference corresponding to a common microstructure is even, the average of the two middle values ​​is taken). Next, the dispersion of this set of difference data is evaluated. Specifically, the absolute deviation of each difference from the median is calculated, and the median of these absolute deviations is obtained, thus yielding a dispersion estimate (i.e., absolute median difference) that is robust to data fluctuations. Based on this dispersion estimate, a reasonable discrimination range is established. Typically, this range is centered on the previously obtained median of the differences, and its upper and lower boundaries are determined by extending the dispersion estimate to both sides by several times (e.g., three times). Next, each original height difference is examined one by one, and data points falling outside the discrimination range are identified as outliers and removed. These outliers are usually caused by characteristic local defects, measurement noise spikes, or accidental matching errors.

[0139] After removing outlier data, the statistical mean of the remaining valid height differences is calculated. This statistical mean is ultimately adopted as the relative offset of the current field of view with respect to the dynamic reference plane in the height direction. Through this robust statistical method based on the median and absolute median, the system can effectively resist abnormal interference without making strict assumptions about the data distribution. It automatically selects a reliable subset of data with inherent consistency, thus ensuring the accuracy and stability of the offset estimation results and providing a solid basis for subsequent height correction steps.

[0140] In addition, in some embodiments, consistency assessment and outlier removal can also be implemented using classical statistical methods.

[0141] Specifically, for all successfully matched common microstructures corresponding to height difference data, the mean (μ) and standard deviation (σ) of these height difference data are first calculated. The mean μ represents an estimate of the central tendency of the data, while the standard deviation σ quantifies the dispersion of the data around the mean. Furthermore, a dynamic outlier discrimination interval [μ-K·σ, μ+K·σ] is set based on these two statistics, where K is a preset scaling factor (e.g., K takes the value of 2 or 3, corresponding to approximately 95% or 99.7% confidence intervals for normal distribution, respectively). Next, all height differences ΔZi are iterated, and values ​​falling outside this outlier discrimination interval are identified as outliers and removed. After removal, the mean of the remaining valid height differences is recalculated, and this mean is used as the final relative offset in the height direction. This method is computationally simple and performs well under ideal conditions where the data quality is high and the noise distribution is close to normal, providing a feasible alternative or simplified implementation scheme for robust statistical methods.

[0142] Step S207: Based on the relative offset, perform height direction correction on the 3D point cloud data of the current field of view to make its height reference consistent with the dynamic reference plane.

[0143] In step S207, the height direction correction of the three-dimensional point cloud data of the current field of view can be performed by mathematical transformation to eliminate the height deviation of the current field of view, so as to achieve the unification of its height reference with the dynamic reference surface.

[0144] In practice, the relative offset represents the average vertical translation of the entire current field of view relative to the dynamic reference plane. This value is derived from a robust statistical estimate of the height differences of multiple shared microstructures within the overlapping region. During correction, each shared microstructure in the current field of view is transformed independently: the original height coordinates of the shared microstructure in the current field of view are used to generate a corrected height value based on the relative offset, while the planar coordinates of the shared microstructure remain unchanged. For example, if the relative offset indicates that the current field of view is higher than the dynamic reference plane, the relative offset is subtracted from the original height coordinates of the shared microstructure in the current field of view. If the relative offset indicates that the current field of view is lower than the dynamic reference plane, the relative offset is added to the original height coordinates of the shared microstructure in the current field of view. Mathematically, this operation is equivalent to performing a global rigid translation of the entire point cloud of the current field of view along the Z-axis, which is a direct and efficient compensation method for systematic errors in the height direction.

[0145] Thus, after completing the above height adjustment for all common microstructures, the corrected three-dimensional point cloud dataset is obtained. Theoretically, this dataset and the dynamic reference plane have a consistent height reference in the overlapping area.

[0146] Step S209: Incorporate the corrected current field of view into the dynamic reference plane and update the dynamic reference plane.

[0147] In practice, incorporating the corrected current field of view into the dynamic reference plane is achieved by integrating the 3D point cloud data of the current field of view into the spatial index structure that maintains the dynamic reference plane. For regions overlapping with existing data, a set strategy (e.g., retaining only a single data set or fusing two sets of data) can be used to ensure a smooth and continuous final result. This operation directly expands the physical extent covered by the dynamic reference plane and reinforces its property as a globally unified reference system. After the update, the dynamic reference plane and its data structure can immediately support the processing of the next field of view, efficiently extracting adjacent overlapping data, thereby continuously driving subsequent processing flows until all field of view data is uniformly integrated.

[0148] As can be seen from the above, through Figure 8The steps shown allow a single current field of view to be incorporated into a dynamic reference plane. By performing closed-loop operations such as locating overlapping regions, matching at least one common microstructure, calculating the relative offset of at least one common microstructure in the height direction in different fields of view, correcting the height direction of the 3D point cloud data of the current field of view, and incorporating and updating the corrected current field of view into the dynamic reference plane, it is ensured that each new field of view can be seamlessly integrated with existing data.

[0149] In addition, during the sequential iterative processing of each field of view, in order to further improve the accuracy and robustness of global high consistency and monitor the error propagation, local closed-loop verification can be performed at specific nodes.

[0150] In some embodiments, during the processing, whenever a predetermined number of fields of view are completed, including but not limited to: processing a fixed number of rows, a fixed number of fields of view, or a fixed scanning time, the system can automatically perform a local loop closure check.

[0151] In practical implementation, this local closed-loop verification may include: constructing at least one spatial closed path based on the overlap relationships between multiple fields of view incorporated into the dynamic reference plane. For example, if fields of view FOV_A overlap with fields of view FOV_B, fields of view FOV_B overlap with fields of view FOV_C, and fields of view FOV_C overlap with fields of view FOV_A, a triangular or circular closed loop can be formed. The system calculates the cumulative height deviation along this closed path, that is, it algebraically sums the previously calculated and applied relative height offsets between each pair of adjacent fields of view on the path. In an ideal case of no error accumulation, the cumulative height deviation along the closed loop should be zero. If the absolute value of the calculated cumulative height deviation exceeds a preset deviation threshold (e.g., 10 nm or 50 nm set according to the system accuracy requirements), it indicates that there is a non-negligible error accumulation or mismatch within the closed loop. At this point, an optimization and adjustment process will be initiated to comprehensively optimize and adjust the height direction offsets of each field of view involved in forming the closed path. This may involve methods such as least squares adjustment to redistribute correction amounts, ensuring the closure condition is satisfied to the greatest extent possible while smoothly correcting the height data of each field of view. After optimization, the data for the corresponding field of view in the dynamic reference plane will be updated according to the new adjustment amounts.

[0152] It is evident that the local closed-loop verification mechanism can actively detect and correct the transmission of systematic deviations during the processing, effectively suppressing the global spread of errors, thereby significantly improving the overall accuracy and reliability of the final 3D point cloud model.

[0153] Thus, following the above processing order, the next field of view is taken as the current field of view, and the process is repeated cyclically. Figure 8The height alignment and dynamic reference plane update process shown involves expanding and consolidating the dynamic reference plane after each field of view is processed. This process iterates until the last field of view in the set is processed and incorporated. Through this chain-like correction method, the height reference of all fields of view is reliably unified to the height reference system defined by the initial reference field of view, thereby generating a globally height-consistent seamless 3D point cloud model.

[0154] The following provides a detailed explanation of the online real-time processing mode and the offline processing mode mentioned above.

[0155] The online real-time processing mode is suitable for scenarios with high requirements for detection efficiency and where hardware computing resources can meet the requirements for real-time processing and analysis.

[0156] First, system initialization and scan path planning are performed. The 3D measurement module, including an area scan camera and a structured light projection module, is configured and mounted on a motion platform. Based on the size of the wafer under test and the required resolution, the scan path is determined, ensuring sufficient overlap between adjacent fields of view. The width of this overlap should be at least twice the largest feature size of the representative common microstructure on the wafer. The system loads the scan path, motion control program, and real-time height correction and dynamic reference plane update algorithm.

[0157] Next, the initial reference field of view is scanned and a dynamic reference surface is established. The scan is initiated, and the 3D point cloud data of the first field of view (denoted as FOV_1) is acquired according to the scan path. The first field of view, FOV_1, is determined as the initial reference field of view. Based on its own height (Z-axis) direction in the 3D coordinate system, an initial dynamic reference surface is constructed. At this point, the dynamic reference surface only contains the 3D point cloud data of FOV_1 after its own correction, and its height reference is the original measurement reference of FOV_1. The 3D point cloud data of FOV_1 is then incorporated into the dynamic reference surface data structure in memory or cache.

[0158] Next, other fields of view are scanned and corrected in real time until all fields of view in the set are included in the dynamic reference plane, generating a 3D point cloud model that corresponds to the region of interest and has global high consistency.

[0159] Following the scanning sequence, the motion platform is controlled to move to the next field of view (e.g., FOV_2) for scanning, acquiring its 3D point cloud data in real time. At this point, FOV_2 is the current field of view to be processed, while the adjacent field of view that has already been processed is FOV_1.

[0160] Within the overlapping area of ​​FOV_2 and FOV_1, at least one common microstructure (e.g., bumps) is identified and matched using a 3D point cloud registration algorithm.

[0161] For each successfully matched common microstructure, the height measurement values ​​of the common microstructure in both the field of view (FOV) 2 and the field of view (FOV) 1 are extracted, and the height difference ΔZ between the two is calculated. In this way, the height difference ΔZ between the multiple common microstructures of a matched pair in the field of view (FOV) 2 and the field of view (FOV) 1 can be calculated. i , i=1,2,…,M, where M is the number of successfully matched common microstructure pairs.

[0162] For the calculated set of height differences {ΔZ1, ΔZ2, ..., ΔZ... M A consistency assessment should be conducted.

[0163] Taking the classic statistical method as an example: calculate the mean μ and standard deviation σ of the group of differences, and regard values ​​that deviate from the mean by more than K times the standard deviation (e.g., K=2 or 3) as outliers and remove them; take the statistical mean of the remaining effective height differences as the relative offset Offset_Z of the field of view FOV_2 relative to the field of view FOV_1 (i.e., relative to the current dynamic reference plane) in the height direction.

[0164] Based on the calculated relative offset Offset_Z, a global translation correction is performed on all 3D point cloud data in the field of view FOV_2 in the height direction. Specifically, the Z coordinates of all points in the field of view FOV_2 are uniformly subtracted from (or added to) Offset_Z, so that their height reference is consistent with the current dynamic reference plane.

[0165] The corrected field of view (FOV)_2 point cloud data is incorporated into the dynamic reference plane and updated synchronously. At this point, the dynamic reference plane is updated to a dynamic reference plane composed of both FOV_1 and FOV_2, with the height deviation between them eliminated.

[0166] Repeat the above steps, processing FOV_3 (aligned with the dynamic reference plane formed by FOV_2 and FOV_1), FOV_4 (aligned with the dynamic reference plane formed by FOV_3, FOV_2, and FOV_1), ..., until the last FOV_N in the set of FOVs. After each other FOV is processed and corrected, the dynamic reference plane is expanded and updated. When the last FOV_N is corrected and incorporated, the dynamic reference plane finally evolves into a globally highly consistent 3D point cloud model covering the entire scanned area. The entire processing is synchronized with the scanning, achieving efficient online processing of "scanning one, correcting and incorporating one".

[0167] Offline processing mode is suitable for scenarios where real-time requirements are not high, or where complex point cloud processing and optimization analysis need to be performed on a high-performance backend workstation.

[0168] First, acquire and store all fields of view in the field of view set: according to the preset scanning rules, complete the scanning of all fields of view in the set area (e.g., the entire wafer, or one or more regions of interest divided within the wafer). Store the original 3D point cloud data corresponding to each field of view completely to a hard disk array or a dedicated database.

[0169] Next, from all stored field-of-view (FOV) data, one FOV is selected as the initial reference FOV. The 3D point cloud data of this initial reference FOV is loaded, and an initial dynamic reference plane is constructed using its height direction. This selection operation can be based on various strategies: for example, to simplify the process, the first FOV in the scan sequence can be directly selected; or, to optimize global accuracy geometrically, a FOV located at the geometric center of all FOVs can be selected by calculating the center coordinates of all FOVs; or, based on process knowledge, a FOV located in the central region of the wafer (which typically has less deformation) can be selected.

[0170] Next, the remaining field of view is read and processed according to the preset processing order.

[0171] Set a preset processing order. This order is usually consistent with the original scanning order to ensure the consistency of the processing logic. However, in offline mode, other optimized orders can also be used, such as starting from the initial reference field of view and expanding the processing to the peripheral field of view using a breadth-first or depth-first strategy. Following this order, data from the next field of view is read from storage as the current field of view.

[0172] For the current field of view, perform the same operations as in the aforementioned online real-time processing mode: locate the overlapping region between the current field of view and the processed adjacent fields of view in the point cloud data; match common microstructures within the overlapping region; calculate the relative offset of the current field of view relative to the processed adjacent fields of view; perform height correction on the current field of view point cloud; and incorporate the corrected data into and update the dynamic reference plane. The difference is that the data here comes from local storage, eliminating the need to wait for scanning and allowing for more time-consuming precision registration algorithm calculations.

[0173] Repeat the above process until all fields of view in the stored field of view set have been processed in the preset processing order, and finally generate a complete 3D point cloud model with globally consistent height.

[0174] This application also discloses a wafer inspection device.

[0175] Please see Figure 9 The diagram shown is a schematic diagram of the wafer inspection device of this application in one embodiment.

[0176] like Figure 9As shown, the wafer inspection device of this application includes: a three-dimensional measurement module 301, a reference plane construction module 303, a field of view processing module 305, and a model generation module 307.

[0177] The 3D measurement module 301 is configured to perform field-of-view scanning on the wafer according to a preset scanning rule to acquire 3D point cloud data corresponding to multiple fields of view; the multiple fields of view constitute a field of view set, and there is an overlapping area between adjacent fields of view in the field of view set, and the overlapping area contains a common microstructure formed by the wafer manufacturing process.

[0178] The reference plane construction module 303 is configured to determine a field of view from the set of fields of view as an initial reference field of view, and to construct a dynamic reference plane with the height direction of the initial reference field of view.

[0179] The field of view processing module 305 is configured to process other fields of view in the field of view set sequentially, starting from the initial reference field of view, according to a preset processing order.

[0180] In some embodiments, the field processing module may further include: a matching unit, an offset calculation unit, a correction unit, and a reference plane update unit.

[0181] The matching unit is configured to match at least one total microstructure within the overlapping region of the current field of view and the processed adjacent fields of view.

[0182] The offset calculation unit is configured to calculate the relative offset in the height direction between at least one common microstructure in the current field of view and the processed adjacent field of view.

[0183] The correction unit is configured to perform height orientation correction on the 3D point cloud data of the current field of view based on the relative offset, so that its height reference is consistent with the dynamic reference plane.

[0184] The datum update unit is configured to incorporate the corrected current field of view into the dynamic datum and update the dynamic datum.

[0185] The model generation module 307 is configured to generate a 3D point cloud model corresponding to the region of interest and with global high consistency after all fields of view in the field of view set are included in the dynamic reference plane.

[0186] It should be noted that the wafer inspection apparatus disclosed in the above embodiments and the wafer inspection method disclosed in the above embodiments belong to the same concept. The specific manner in which each module and unit performs its operation has been described in detail in the method embodiments, and will not be repeated here. In practical applications, the wafer inspection apparatus disclosed in the above embodiments can be assigned to different functional modules as needed, that is, the internal structure of the system can be divided into different functional modules to complete all or part of the functions described above, and this is not a limitation here.

[0187] This application discloses another electronic device; please refer to [link / reference needed]. Figure 10 The diagram shown is a structural schematic of the electronic device of this application in one embodiment.

[0188] like Figure 10 As shown, the electronic device includes a processor 401 and a memory 403. The processor 401 and the memory 403 can communicate via a bus 402. The memory 403 can store program instructions, and the processor 401 implements the steps in the wafer inspection method in the previous embodiment by running the program instructions in the memory 403.

[0189] Bus 402 can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of representation, although only one thick line is used in the diagram, this does not indicate that there is only one bus or one type of bus.

[0190] The processor 401 can be implemented as a central processing unit (CPU), a microprocessor unit (MCU), a system on chip (System on Chip), or a field-programmable logic array (FPGA).

[0191] Memory 403 may include volatile memory for temporary data storage during operation, such as random access memory (RAM).

[0192] The memory 403 may also include non-volatile memory for data storage, such as read-only memory (ROM), flash memory, hard disk drive (HDD), or solid-state disk (SSD).

[0193] In some embodiments, the electronic device may further include a communication interface 404. The communication interface 404 is used for communication with external devices. In specific examples, the communication interface 404 may include one or more wired and / or wireless communication circuit modules. For example, the communication interface 404 may include one or more of, such as a wired network card, a USB module, a serial interface module, etc. The wireless communication protocols followed by the wireless communication module include, for example, Near Field Communication (NFC) technology, Infrared (IR) technology, Global System for Mobile Communications (GSM), General Packet Radio Service (GPRS), Code Division Multiple Access (CDMA), Wideband Code Division Multiple Access (WCDMA), Time-Division Code Division Multiple Access (TD-SCDMA), Long Term Evolution (LTE), Bluetooth (BT), Global Navigation Satellite System (GNSS), etc.

[0194] Electronic devices may also include input units and output units. The input unit can be used to receive input digital or character information, and to generate keyboard, mouse, touchscreen, joystick, optical, or trackball signal inputs related to user settings and function control. The output unit can be used to present or transmit processing results, status information, and interactive feedback to the user, for example, through a display screen (e.g., LCD screen, touchscreen), indicator lights, speakers, vibration modules, printers, or other audible, visual, or tactile output devices, generating text, images, sound, or tactile signals corresponding to the device's operating status, operation prompts, or alarm information.

[0195] Specifically, in this embodiment, the processor 401 in the electronic device loads the executable files corresponding to the processes of one or more computer programs into the memory 403 according to the following instructions, and the processor 401 runs the program instructions stored in the memory 403 to realize the various steps of the aforementioned wafer inspection method.

[0196] For details on the implementation of each of the above steps, please refer to the previous examples, which will not be repeated here.

[0197] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be performed by a computer program, or by a computer program controlling related hardware. The computer program can be stored in a computer-readable storage medium and loaded and executed by a processor.

[0198] Therefore, this application further discloses a computer-readable storage medium storing a computer program that can be loaded by a processor to execute the steps of any of the wafer inspection methods disclosed in this application.

[0199] The computer-readable storage medium may include: read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0200] Since the computer program stored in the computer-readable storage medium can execute the steps of any of the wafer inspection methods disclosed in the embodiments of this application, the beneficial effects that any of the wafer inspection methods disclosed in the embodiments of this application can achieve can be realized, as detailed in the preceding embodiments, and will not be repeated here.

[0201] This application also discloses a computer program product, which includes a computer program stored in a computer-readable storage medium. The computer program is executed by a processor using the methods disclosed in various optional implementations of the above-described wafer inspection method.

[0202] The foregoing has provided a detailed description of a wafer inspection method, wafer inspection device, electronic device, computer-readable storage medium, and computer program product disclosed in the embodiments of this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the embodiments above are only for the purpose of helping to understand the method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A wafer inspection method, characterized in that, Includes the following steps: A field of view is determined as the initial reference field of view, and a dynamic reference surface is constructed with the height direction of the initial reference field of view. The initial reference field of view is a set of fields of view obtained by scanning the region of interest of the wafer according to a preset scanning rule. Each field of view in the set of fields of view has corresponding three-dimensional point cloud data, and there is an overlapping region between adjacent fields of view in the set of fields of view. The overlapping region contains a common microstructure formed by the wafer manufacturing process, and the width of the overlapping region is greater than or equal to twice the maximum feature size of the common microstructure. as well as According to the preset processing order, other fields of view in the field of view set are processed sequentially starting from the initial reference field of view until all fields of view in the field of view set are included in the dynamic reference surface, generating a three-dimensional point cloud model corresponding to the region of interest and having global high consistency. For the current field of view to be processed, the following steps are performed: In the overlapping area between the current field of view and the processed adjacent field of view, at least three non-collinear common microstructures are matched; Based on the height measurements of the at least three non-collinear common microstructures in the current field of view and the processed adjacent field of view, the relative offset between the two in the height direction is calculated; According to the relative offset, the height direction of the three-dimensional point cloud data of the current field of view is corrected so that its height reference is consistent with the dynamic reference plane; The corrected current field of view is incorporated into the dynamic reference plane and the dynamic reference plane is updated.

2. The wafer inspection method according to claim 1, characterized in that, The wafer inspection method employs an online real-time processing mode, and the preset processing order is the scanning order of the preset scanning rules: During the scanning process, the field of view obtained from the first scan is used as the initial reference field of view, and a dynamic reference plane is constructed. as well as According to the scanning order, after scanning one other field of view in the field of view set is completed, that is, it is taken as the current field of view, and the height direction is corrected based on the dynamic reference plane and the dynamic reference plane is updated.

3. The wafer inspection method according to claim 1, characterized in that, The wafer inspection method includes: The wafer is scanned by field of view to acquire and store data of all fields of view in the field of view set; Select one field of view from all acquired fields of view as the initial reference field of view, and construct a dynamic reference surface; and According to a preset processing order, data from other fields of view in the field of view set are read from the storage and processed sequentially until the correction of all fields of view is completed; wherein, the preset processing order is determined based on the scanning order of the preset scanning rules.

4. The wafer inspection method according to claim 2 or 3, characterized in that, The initial reference field of view is selected from any of the following: The first field of view acquired according to the scanning rules; A field of view located at the geometric center of multiple fields of view; as well as A field of view located in the central region of the wafer.

5. The wafer inspection method according to claim 1, characterized in that, The calculation of the relative offset between the two in the height direction includes: Calculate the height difference of each corresponding common microstructure in two different fields of view; and The statistical mean of the height differences is taken as the relative offset in the height direction.

6. The wafer inspection method according to claim 5, characterized in that, Before taking the statistical mean, the following steps are also included: The calculated height differences of each corresponding common microstructure are evaluated for consistency, and abnormal differences deviating from the preset height difference threshold are removed; and Based on the remaining height difference, the statistical mean is taken as the relative offset in the height direction.

7. The wafer inspection method according to claim 1, characterized in that, The preset scanning rules are determined based on preset paths, which include at least one of serpentine paths, spiral paths, and jump paths optimized according to wafer chip layout.

8. The wafer inspection method according to claim 1, characterized in that, During the scanning process, the overlapping area between adjacent fields of view is dynamically adjusted according to the size and / or density of the shared microstructure. When the density of the shared microstructure in a local area is lower than a preset density threshold and / or the size of the shared microstructure is smaller than a preset size threshold, the overlapping area is increased. When the density of the shared microstructure in a local area is higher than a preset density threshold and the size of the shared microstructure is larger than a preset size threshold, the overlapping area is decreased.

9. The wafer inspection method according to claim 1, characterized in that, The wafer inspection method further includes: Based on the two-dimensional image features and / or the planar projection features of the three-dimensional point cloud within the overlapping region, relative alignment of the planar direction is performed between the current field of view and the processed adjacent fields of view to improve the matching accuracy of the common microstructure within the overlapping region.

10. The wafer inspection method according to claim 9, characterized in that, The relative alignment in the planar direction includes only translation and / or rotation transformations.

11. The wafer inspection method according to claim 9, characterized in that, The calculation of the relative alignment in the planar direction and the relative offset in the height direction is performed based on the same at least one common microstructure.

12. The wafer inspection method according to claim 1, characterized in that, During processing, a local loop closure check is performed after each predetermined number of fields of view are completed. Construct at least one closed path based on the overlap relationship between multiple fields of view incorporated in the dynamic reference plane, and calculate the cumulative height deviation derived along the closed path. as well as If the cumulative height deviation exceeds a preset deviation threshold, the height direction offset of each field of view involved in the closed path will be optimized and adjusted as a whole.

13. The wafer inspection method according to claim 1, characterized in that, The common microstructure is a three-dimensional structure formed on the wafer during the packaging process, including at least one of bumps, copper pillars, metal pads, and test solder joints.

14. A wafer inspection device, characterized in that, include: The 3D measurement module is configured to perform field-of-interest scanning on the wafer according to preset scanning rules to acquire 3D point cloud data corresponding to multiple fields of interest. The multiple fields of view constitute a field of view set, and there is an overlapping region between adjacent fields of view in the field of view set. The overlapping region contains a common microstructure formed by the wafer manufacturing process, and the width of the overlapping region is greater than or equal to twice the maximum feature size of the common microstructure. The reference plane construction module is configured to determine a field of view from the set of fields of view as an initial reference field of view, and to construct a dynamic reference plane with the height direction of the initial reference field of view. A field-of-view processing module is configured to process other fields of view in the field-of-view set sequentially, starting from the initial reference field of view, according to a preset processing order. The field-of-view processing module includes: a matching unit configured to match at least three non-collinear common microstructures within the overlapping region of the current field of view and the processed adjacent fields of view; an offset calculation unit configured to calculate the relative offset in the height direction between the current field of view and the processed adjacent fields of view based on the height measurements of the at least three non-collinear common microstructures in the current field of view and the processed adjacent fields of view; a correction unit configured to perform height direction correction on the 3D point cloud data of the current field of view according to the relative offset, so that its height reference is consistent with the dynamic reference plane; and a reference plane update unit configured to incorporate the corrected current field of view into the dynamic reference plane and update the dynamic reference plane. The model generation module is configured to generate a 3D point cloud model corresponding to the region of interest and having global high consistency after all fields of view in the field of view set are included in the dynamic reference plane.

15. The wafer inspection apparatus according to claim 14, characterized in that, The three-dimensional measurement module includes: A two-dimensional image acquisition unit is used to acquire two-dimensional images of the wafer surface; and A three-dimensional topography scanning unit is used to acquire three-dimensional height data corresponding to the two-dimensional image space.

16. The wafer inspection apparatus according to claim 15, characterized in that, The three-dimensional topography scanning unit employs a white light interferometer, a confocal microscope, or a structured light scanning system.

17. An electronic device, characterized in that, include: processor; as well as Memory, which stores computer programs; The computer program is executed by the processor to perform the wafer inspection method as described in any one of claims 1 to 13.

18. A computer-readable storage medium, characterized in that, The device stores program instructions that, when executed, perform the wafer inspection method as described in any one of claims 1 to 13.

19. A computer program product, characterized in that, It includes a computer program that, when executed by a processor, implements the wafer inspection method as described in any one of claims 1 to 13.

Images