Chip micro-crack visual detection method and system based on multi-angle polarization imaging

Abstract

The application discloses a chip micro-crack visual detection method and system based on multi-angle polarization imaging, relates to the technical field of visual detection, and comprises the following steps: acquiring polarization reflection images of a chip under illumination of different polarization angles, and extracting light intensity response values of the polarization reflection images under different polarization angles to obtain a polarization response curve; identifying polarization response direction difference characteristics of each pixel point in the image belonging to a metal interconnection layer, and comparing the polarization response direction difference characteristics with the polarization response curve to obtain a first resonance zone; continuously adjusting the illumination polarization angle corresponding to the first resonance zone to obtain a polarization rotation path of the first resonance zone; extracting a path mutation section, marking a crack feature of the first resonance zone based on the path mutation section, obtaining a micro-crack candidate area, and generating a chip micro-crack detection result according to spatial distribution, so that the problem that metal lines of an advanced process node chip cause polarization direction rotation when producing anisotropic reflection under different angles is solved.

Description

Technical Field

[0001] This invention relates to the field of visual inspection technology, and more specifically, to a method and system for visual inspection of microcracks in chips based on multi-angle polarization imaging. Background Technology

[0002] In existing semiconductor chip manufacturing and testing technologies, microcracks are a significant defect affecting chip reliability and performance. Their size is typically only on the order of micrometers or even smaller, and their directional distribution is highly random. They may form along crystal boundaries, between thin film layers, or in stress concentration areas, and their morphology is complex and diverse, including linear, branching, or network cracks. These microcracks easily propagate under the influence of wafer processing, packaging, thermal cycling, or mechanical stress, leading to abnormal chip function, reduced lifespan, or premature failure.

[0003] However, chips at advanced process nodes typically contain multi-layered metal interconnect structures. These interconnect layers consist of densely packed wires, and the orientation of these wires may differ in each layer. When light strikes these metal lines at different incident angles, the reflection of light from the metal surface exhibits anisotropy, meaning the intensity of the reflected light varies in different polarization directions. This anisotropy causes the polarization direction of the reflected light to rotate with the incident angle and the orientation of the metal wires, resulting in different optical responses at the same location under different polarization angles. This polarization rotation phenomenon is more pronounced in areas with multiple superimposed or interleaved interconnects, and it can also amplify the local effects of microstructural defects or microcracks on optical reflection. To address these issues, this invention proposes a solution. Summary of the Invention

[0004] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a chip microcrack visual inspection method and system based on multi-angle polarization imaging. By combining the polarization response curve and polarization rotation path with the orientation information of the metal interconnect layer, the microcracks and their distribution can be accurately identified. This solves the problem of polarization direction rotation caused by anisotropic reflection of metal lines at different angles in the multi-layer metal interconnect structure of advanced process node chips.

[0005] To achieve the above objectives, the present invention provides the following technical solution: A chip microcrack visual detection method based on multi-angle polarization imaging includes the following steps: acquiring polarization reflection images of the chip under illumination at different polarization angles, and extracting the light intensity response values ​​of the polarization reflection images at different polarization angles to obtain polarization response curves; identifying the polarization response direction difference features of the metal interconnect layer to which each pixel belongs in the image, and comparing the polarization response curves with the polarization response direction difference features to obtain the first resonance region; continuously adjusting the illumination polarization angle corresponding to the first resonance region according to a preset scanning step size to obtain the polarization rotation path of the first resonance region; extracting the path abrupt change segments in the polarization rotation path and marking the crack features of the first resonance region based on the path abrupt change segments to obtain microcrack candidate regions; and generating chip microcrack detection results based on the spatial distribution of the microcrack candidate regions.

[0006] In a preferred embodiment, the step of extracting the light intensity response values ​​of the polarized reflection image at different polarization angles to obtain the polarization response curve specifically involves: reading the grayscale value of each polarized reflection image pixel by pixel and using the grayscale value as the initial light intensity response value of that pixel at the current illumination polarization angle; arranging the initial light intensity response values ​​obtained at the same pixel location at different polarization angles in ascending order of polarization angle to obtain an initial light intensity sequence; mapping the initial light intensity sequence onto a polarized reflection image coordinate system with the illumination polarization angle as the horizontal axis and the light intensity response value as the vertical axis; and performing curve fitting on the mapped discrete points using a piecewise cubic spline interpolation algorithm to obtain the polarization response curve.

[0007] In a preferred embodiment, the identification of the polarization response direction difference features of the metal interconnect layer to which each pixel belongs in the image is identified, and the polarization response curve is compared with the polarization response direction difference features to obtain the first resonance region, specifically: acquiring chip layout design data, extracting the layout direction information of each metal interconnect layer and mapping the layout direction information to a coordinate system, marking the extension direction of the conductor of the metal interconnect layer to which each pixel belongs, and obtaining a direction label layer; based on the direction label layer, traversing each pixel, calculating the correlation between the polarization response curve of the pixel and the polarization response curves of all pixels in the same layer neighborhood, and identifying the deviating pixels according to the results; obtaining the illumination polarization angle corresponding to the peak position of the polarization response curve of the deviating pixel and comparing it with the extension direction of the conductor of the metal interconnect layer to which the deviating pixel belongs, and calculating the angle difference between the two; when the angle difference is less than a preset angle matching threshold, marking the deviating pixel as a resonance point; clustering all resonance points to form several connected regions as the first resonance region.

[0008] In a preferred embodiment, the step of extracting the layout direction information of each metal interconnect layer and mapping the layout direction information to a coordinate system, and marking the extension direction of the wires of the metal interconnect layer to which each pixel belongs to obtain a direction label layer, specifically involves: parsing the chip layout design data, extracting the graphic data of each metal interconnect layer, and vectorizing the polygonal lines in the graphic data of each metal interconnect layer to obtain vectorized metal interconnect layer graphic data; calculating the direction angle of each vector line segment based on the vectorized metal interconnect layer graphic data, quantizing the direction angle to several preset main direction intervals to obtain the main direction interval value of the vector line segment; and obtaining... The resolution and field of view of the polarized reflection image are determined, and based on the resolution and field of view of the polarized reflection image, the vectorized metal interconnect layer graphic data is mapped to a coordinate system through affine transformation. In the coordinate system, the graphic data of each metal interconnect layer after mapping is rasterized, and each pixel is assigned the metal interconnect layer identifier of its location and the main direction interval value of the vector line segment to which it belongs. All pixels are traversed. If a pixel is located in the overlapping area of ​​multiple metal interconnect layers, the layout direction of the top metal interconnect layer is selected as the direction label of the pixel according to the vertical projection order of each layer at the pixel, thus obtaining the direction label layer.

[0009] In a preferred embodiment, the step of continuously adjusting the illumination polarization angle corresponding to the first resonance region according to a preset scanning step size to obtain the polarization rotation path of the first resonance region specifically involves: extracting the geometric center coordinates of each first resonance region and determining the metal interconnect layer to which the geometric center coordinates belong in the direction label layer; obtaining the standard wire extension direction of the metal interconnect layer, and using the standard wire extension direction as a reference, expanding to both sides by a preset angle range to obtain the scanning interval of the illumination polarization angle; within the scanning interval, sequentially changing the illumination polarization angle with a preset scanning step size to obtain several adjusted illumination polarization angles; re-acquiring the polarization reflection image at the location of the first resonance region under each adjusted illumination polarization angle, and extracting the average light intensity response value of all pixels in the first resonance region; plotting the resonance response curve of the first resonance region with the adjusted illumination polarization angle as the abscissa and the corresponding average light intensity response value as the ordinate, and defining the change trajectory of the resonance response curve as the polarization rotation path of the first resonance region.

[0010] In a preferred embodiment, obtaining the standard conductor extension direction of the metal interconnect layer and expanding a preset angle range to both sides based on the standard conductor extension direction to obtain the scanning interval of the illumination polarization angle specifically involves: obtaining the standard conductor extension direction of the metal interconnect layer to which the first resonant region belongs, defining the angle of this direction as a reference angle and calculating the vertical angle of the reference angle; expanding a first angle range to the side closer to the vertical angle with the reference angle as the center, and simultaneously expanding a second angle range to the side farther from the vertical angle, wherein the first angle range is larger than the second angle range; merging the reference angle, the first angle range, and the second angle range to form an asymmetric continuous angle interval; and arranging all angle values ​​within this asymmetric continuous angle interval in ascending order to obtain the scanning interval of the illumination polarization angle.

[0011] In a preferred embodiment, the step of extracting path abrupt change segments in the polarization rotation path and marking crack features in the first resonance region based on these abrupt change segments to obtain microcrack candidate regions specifically involves: calculating the second derivative of the polarization rotation path to obtain the curvature change value at each point on the polarization rotation path, and marking continuous intervals where the curvature change value exceeds a preset curvature abrupt change threshold as initial abrupt change segments; extracting the illumination polarization angle interval corresponding to each initial abrupt change segment, and acquiring a polarization reflection image sequence of the first resonance region within the illumination polarization angle interval; calculating the light intensity fluctuation variance of each pixel in the polarization reflection image sequence to obtain a light intensity fluctuation distribution map; extracting pixels with variance values ​​exceeding a preset fluctuation threshold in the light intensity fluctuation distribution map as fluctuation points, and performing morphological closing operations on the fluctuation points to form several fluctuation clusters; matching the fluctuation clusters with the initial abrupt change segments in spatial location and angle interval, and identifying microcrack candidate regions based on the matching results.

[0012] In a preferred embodiment, generating chip microcrack detection results based on the spatial distribution of microcrack candidate regions specifically involves: numbering all microcrack candidate regions and extracting the contour line of each microcrack candidate region; calculating the minimum bounding rectangle of the contour line and obtaining the major axis direction and size information of the minimum bounding rectangle; calculating the angle between the major axis direction and the extension direction of the conductor in the metal interconnect layer to which the microcrack candidate region belongs; identifying the type of microcrack based on the angle to obtain the determination type of the microcrack; counting the total area and number of all microcrack candidate regions, and calculating the microcrack surface density based on the detection area of ​​the chip, while recording the geometric size information of each microcrack candidate region; and integrating the determination type, microcrack surface density, number, and geometric size information of the microcracks to generate chip microcrack detection results.

[0013] The technical effects and advantages of the chip microcrack visual inspection method and system based on multi-angle polarization imaging of this invention are as follows: This invention acquires polarization reflection images of a chip under illumination at different polarization angles and extracts the light intensity response value of each pixel at each polarization angle to generate a polarization response curve, effectively reflecting the polarization direction rotation characteristics of metal lines caused by anisotropic reflection. Subsequently, by analyzing the polarization response direction differences of the metal interconnect layer to which each pixel belongs, the polarization response curve is compared with the direction difference characteristics to identify the first resonance region, thereby accurately locating potential crack areas in complex stacked structures. Furthermore, continuous scanning of the illumination polarization angle corresponding to the first resonance region generates a polarization rotation path, which can capture local optical anomalies caused by microcracks. Then, by extracting abrupt path segments and marking crack features in the first resonance region based on these segments, microcrack candidate regions are obtained, achieving pixel-level crack identification. Finally, based on the spatial distribution of the microcrack candidate regions, chip microcrack detection results are generated, providing an intuitive presentation of the number, distribution, and location of cracks. Through the above steps, this invention not only effectively solves the problem of polarization direction rotation caused by anisotropic reflection of metal lines in multilayer interconnect structures, but also has the technical advantages of high sensitivity, accurate positioning, quantifiable analysis and full-process automation, significantly improving the reliability and engineering applicability of chip microcrack detection. Attached Figure Description

[0014] Figure 1 This is a flowchart illustrating the chip microcrack visual inspection method based on multi-angle polarization imaging according to the present invention.

[0015] Figure 2 This is a schematic diagram of the chip microcrack visual inspection system based on multi-angle polarization imaging according to the present invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0017] Example 1, Figure 1 The present invention provides a chip microcrack visual detection method based on multi-angle polarization imaging, comprising the following steps: S1, acquire polarization reflection images of the chip under illumination at different polarization angles, and extract the light intensity response values ​​of the polarization reflection images at different polarization angles to obtain polarization response curves; In this embodiment, the light intensity response values ​​of the polarization reflection image at different polarization angles are extracted to obtain the polarization response curve, specifically: The grayscale value of each polarized reflection image is read pixel by pixel, and the grayscale value is used as the initial light intensity response value of the pixel under the current illumination polarization angle. The initial light intensity response values ​​obtained at the same pixel location under different polarization angles are arranged in ascending order of polarization angle to obtain the initial light intensity sequence. The initial light intensity sequence is mapped onto a polarization reflection image coordinate system with the illumination polarization angle as the horizontal axis and the light intensity response value as the vertical axis. The polarization response curve is obtained by fitting the mapped discrete points using a piecewise cubic spline interpolation algorithm.

[0018] It should be noted that the polarization reflection images of a chip under illumination at different polarization angles refer to the process of performing microscopic imaging of a chip sample. A rotatable linear polarizer is placed in the incident light path, and the polarization direction of the incident light is changed, for example, sequentially set to multiple angles such as 0°, 15°, 30°, 45° up to 180°. A reflection image of the chip surface is acquired at each polarization angle, thus obtaining a sequence of images corresponding to different polarization directions. Because the internal metal interconnect layers of the chip have a significant anisotropic structure, their reflection intensity exhibits directional selectivity differences for incident light with different polarization directions. Therefore, the same location will show different brightness characteristics at different polarization angles. For example, a metal line area may appear as a grayscale value of 180 at a 0° polarization angle, while it may only be 95 at a 90° polarization angle. This brightness variation with the polarization angle constitutes the polarization reflection characteristic, and the images acquired at each angle are the polarization reflection images under illumination at different polarization angles. They correspond completely in spatial position, differing only in the polarization direction of the incident light.

[0019] Furthermore, reading the grayscale value of each polarized reflection image pixel by pixel means that after obtaining an image at a certain polarization angle, the image is treated as a regularly arranged matrix of pixels. For example, a resolution of 2048×2048 contains approximately four million pixels. The grayscale value of each pixel is read in row-major or column-major order. If it is an 8-bit grayscale image, the grayscale value of each pixel ranges from 0 to 255. For example, if the grayscale value of a pixel in an image with a 0° polarization angle is 172, this value is recorded as the initial light intensity response value of that pixel at a 0° polarization angle. When the polarization angle is adjusted to 15° and the image is re-acquired, the grayscale value is read at the same coordinate position, for example, 165, which is recorded as the initial light intensity response value of that pixel at a 15° polarization angle. By performing the same pixel-by-pixel reading operation on each image with a different polarization angle, a set of light intensity response data corresponding one-to-one with the polarization angle can be established for each pixel. These grayscale values ​​directly reflect the reflection intensity of the pixel under incident light in the corresponding polarization direction.

[0020] Furthermore, arranging the initial light intensity response values ​​acquired at the same pixel location under different polarization angles in ascending order of polarization angle means that for a pixel at a fixed coordinate position, such as (100, 250), its corresponding grayscale values ​​at all acquisition angles are collected and sorted according to the corresponding polarization angle. For example, if the grayscale values ​​of this pixel at 0°, 30°, 60°, 90°, 120°, and 150° are 172, 168, 140, 102, 130, and 160 respectively, then after arranging them in ascending order of angle, an ordered data sequence is formed: [172, 168, 140, 102, 130, 160], and a correspondence is established with the angle sequence [0°, 30°, 60°, 90°, 120°, 150°]. If the original acquisition order is not in ascending order of angle, then the angle values ​​need to be sorted first, and then the corresponding grayscale values ​​need to be rearranged according to the sorting result. The resulting set of grayscale values ​​arranged in ascending order of angle is the initial light intensity sequence of the pixel, which fully reflects the discrete light intensity response of the pixel as the polarization angle changes.

[0021] Secondly, mapping the initial light intensity sequence to a polarization reflection image coordinate system with the illumination polarization angle as the horizontal axis and the light intensity response value as the vertical axis refers to constructing a two-dimensional coordinate representation, where the horizontal axis represents the polarization angle and the vertical axis represents the corresponding grayscale value. Each set of "angle-light intensity" data is then plotted as a point in this coordinate space. Using the example above, the horizontal axis can be marked with scales such as 0°, 30°, 60°, 90°, 120°, and 150°, and the vertical axis can be marked with the grayscale range from 0 to 255. Then, the points (0°, 172), (30°, 168), (60°, 140), (90°, 102), (120°, 130), and (150°, 160) are sequentially marked. These points form a set of discrete distribution points in the coordinate plane, collectively describing the trend of reflection intensity variation of the pixel under different polarization directions. Through this mapping method, the grayscale information originally scattered in different images is transformed into an angular response relationship in the same coordinate space, providing a data foundation for subsequent curve fitting.

[0022] Finally, a piecewise cubic spline interpolation algorithm is used to fit the mapped discrete points to a curve. This involves constructing a continuous and smooth response curve between the aforementioned angle-intensity discrete points, ensuring that the curve passes through every known data point and forms a smooth transition between adjacent data points. In practice, the area between two adjacent angles can be divided into intervals, for example, 0° to 30° and 30° to 60°. A smooth curve is constructed within each interval, ensuring smooth connections and no abrupt changes at the boundaries of each interval. Using the example data, a smooth downward trend curve is formed in the 0° to 30° interval, a curve segment with a larger downward amplitude is formed in the 60° to 90° interval, and an upward trend curve is formed in the 90° to 150° interval. By constructing and stitching these segments together, a complete and continuous polarization response curve is obtained. This curve more accurately reflects the true trend of pixel intensity changing with the polarization angle, and also facilitates subsequent identification of peak positions, response directions, and resonance characteristics, thus providing a stable directional analysis basis for microcrack identification.

[0023] S2, identify the polarization response direction difference characteristics of the metal interconnect layer to which each pixel belongs in the image, and compare the polarization response curve with the polarization response direction difference characteristics to obtain the first resonance region; In this embodiment, the polarization response direction difference characteristics of the metal interconnect layer to which each pixel belongs in the image are identified, and the polarization response curve is compared with the polarization response direction difference characteristics to obtain the first resonance region, specifically: Obtain chip layout design data, extract the layout direction information of each metal interconnect layer and map the layout direction information to the coordinate system, mark the extension direction of the wires of the metal interconnect layer to which each pixel belongs, and obtain the direction label layer; Based on the orientation label layer, each pixel is traversed, and the correlation between the polarization response curve of the pixel and the polarization response curves of all pixels in the same neighboring layer is calculated. Based on the results, the off-target pixels are identified. Obtain the illumination polarization angle corresponding to the peak position of the polarization response curve that deviates from the pixel and compare it with the extension direction of the wire in the metal interconnect layer to which the deviated pixel belongs, and calculate the angle difference between the two. When the angle difference is less than the preset angle matching threshold, the deviating pixel is marked as a resonance point; All resonance points are clustered to form several connected regions, which are then used as the first resonance region.

[0024] It should be noted that when traversing based on the direction label layer, the entire image is first divided into several regions of the same layer, based on the metal interconnect layers and their main direction intervals marked in the direction label layer. Then, the image is traversed point by point according to pixel coordinates, for example, scanning row by row starting from the top left corner. When a pixel is traversed, the identifier of its metal interconnect layer and the value of its main direction interval are read first. Then, a neighborhood window of a preset size is selected within that layer, for example, a 7×7 or 9×9 local region is constructed with the pixel as the center, and only neighboring pixels belonging to the same metal interconnect layer are retained. Subsequently, the polarization response curve of the current pixel is compared with the polarization response curve of each pixel in the neighborhood for similarity calculation. For example, the overall trend of the curve is consistent, the peak positions are close, and the curve shape fluctuates synchronously. All similarity results are then averaged or the consistency ratio is statistically calculated. If the curve trends of most neighboring pixels are highly consistent with those of the current pixel, but the current pixel's curve shows a significant deviation—for example, the peaks of neighboring pixels are concentrated around 90°, while the peak of this pixel appears at 60°; or the curves of the neighboring pixels are generally smooth while the curve of this pixel shows abnormal fluctuations—then this pixel is identified as an off-target pixel. An off-target pixel refers to a pixel whose polarization response behavior is significantly inconsistent with the background of the same layer structure. This usually indicates an abnormal change in the optical reflection characteristics at that location, possibly caused by microcracks, stress concentration, or material structural damage; therefore, it is a key focus of subsequent analysis.

[0025] For identified off-target pixels, the process first locates the position with the highest light intensity response on its fitted polarization response curve, i.e., finding the angle with the maximum response value among all angles. For example, if the response curve of an off-target pixel in the range of 0° to 180° reaches its maximum light intensity value at 75°, then 75° is recorded as the peak illumination polarization angle of that pixel. Next, the extension direction of the conductor in the metal interconnect layer to which the pixel belongs is read from the orientation label layer; for example, if the main direction of this layer is 80°. Then, the peak illumination polarization angle is compared with the conductor extension direction, and the difference between the two angles is calculated; for example, the difference between 75° and 80° is 5°. If the peak angle is 100° and the conductor direction is 80°, the difference is 20°. In this way, it is possible to quantify whether the optical resonance direction of the off-target pixel matches the physical extension direction of the metal conductor, thereby determining whether it has structural orientation consistency characteristics.

[0026] Secondly, when the calculated angle difference is less than a preset angle matching threshold, for example, a threshold set to 10°, if the angle difference of a certain deviating pixel is 5°, its peak response direction is considered to be highly consistent with the extension direction of the metal conductor, and the pixel is marked as a resonance point; if the difference is 18°, it is not marked. A resonance point refers to a pixel whose maximum reflection response direction is basically consistent with the physical direction of the conductor in the metal interconnect layer during polarized light scanning. This indicates that the optical response at this location may be affected by structural disturbances along the conductor direction, such as cracks propagating along the conductor direction causing local stress concentration, which enhances the polarized reflection characteristics in that direction. Therefore, resonance points not only have abnormal curve characteristics but also exhibit consistency with the conductor extension direction in direction, making them more structurally significant candidate pixels selected from a large number of deviating pixels.

[0027] Finally, after obtaining all resonance points, they are clustered according to spatial adjacency. This involves determining whether the resonance points are adjacent in the image or within a preset distance range. For example, using the 8-neighborhood connectivity rule, resonance points that are in contact horizontally, vertically, or diagonally are merged into the same connected set. Resonance points that are slightly farther apart but still within a certain threshold range can also be merged based on distance conditions. After traversal and merging, several spatially continuous sets of resonance points are formed, each of which constitutes a connected region. These connected regions are defined as the first resonance region. The first resonance region refers to a spatially clustered region within the same metal interconnect layer, consisting of multiple resonance points with matched directions and abnormal responses. It represents the region where a consistent and anomalous enhanced response occurs during polarization scanning, typically corresponding to areas where microcracks or local structural damage may exist. Compared to a single pixel, the first resonance region has spatial continuity and consistent direction characteristics, better reflecting the true distribution of physical defects and providing a stable spatial basis for subsequent polarization rotation path analysis and crack feature extraction.

[0028] In this embodiment, the layout direction information of each metal interconnect layer is extracted and mapped to a coordinate system. Each pixel is labeled with the extension direction of the wires in its respective metal interconnect layer, resulting in a direction label layer. Specifically: The chip layout design data is analyzed, the graphic data of each metal interconnect layer is extracted, and the polygon lines in the graphic data of each metal interconnect layer are vectorized to obtain the vectorized metal interconnect layer graphic data. The orientation angle of each vector line segment is calculated based on the vectorized metal interconnect layer graphic data. The orientation angle is then quantized into several preset main orientation intervals to obtain the main orientation interval values ​​of the vector line segments. The resolution and field of view of the polarization reflection image are obtained, and the vectorized metal interconnect layer graphic data are mapped to the coordinate system through affine transformation based on the resolution and field of view of the polarization reflection image. In the coordinate system, the graphic data of each mapped metal interconnect layer is rasterized and each pixel is assigned the metal interconnect layer identifier of its location and the main direction interval value of the vector line segment to which it belongs in that layer. Traverse all pixels. If a pixel is located in the overlapping area of ​​multiple metal interconnect layers, select the layout direction of the top metal interconnect layer as the direction label of the pixel according to the vertical projection order of each layer at the pixel, and obtain the direction label layer.

[0029] It's important to note that chip layout design data refers to the geometric layout files generated during the integrated circuit design phase. These files are typically stored in formats such as GDSII or OASIS, and record the geometric information of each process layer of the chip, including metal interconnect layers, via layers, and polysilicon layers. Each layer consists of numerous polygonal graphics, which describe the routing, width, spacing, and interlayer connections of the conductors. For example, a metal interconnect layer might contain several horizontally extending rectangular polygons and several vertically connected rectangular lines. Parsing this layout data first involves reading the file structure and filtering out all graphic data belonging to the metal interconnect layers according to their layer numbers. Then, the vertex coordinates of each polygon are extracted; for example, a rectangle might be composed of four vertex coordinates. Next, the polygon boundaries are decomposed into several line segments, each defined by two endpoint coordinates. This transforms the original filled area graphic into a geometric representation composed of a set of line segments—a process known as vectorization. After this processing, each metal interconnect layer is no longer just a collection of closed regions, but consists of a large number of vector line segments with clear start and end coordinates, which facilitates subsequent direction analysis and mapping processing.

[0030] After obtaining the vectorized metal interconnect layer pattern data, each line segment has a clear start and end coordinate. For example, a line segment extending from (10,20) to (110,20) indicates a horizontal extension; another line segment extending from (50,30) to (50,130) indicates a vertical extension. The extension direction can be determined based on the coordinate relationship between the two endpoints of the line segment, and its direction angle can be calculated. For example, a horizontal line segment corresponds to 0° or 180°, a vertical line segment corresponds to 90°, and for an oblique line segment, its tilt angle is determined based on the coordinate difference. After obtaining the direction angle, it is divided into several preset main direction intervals. For example, 0°±15° is designated as the horizontal main direction interval, 90°±15° as the vertical main direction interval, and 45°±15° as the first diagonal main direction interval, etc. If a line segment has a direction angle of 88°, it is assigned to the vertical main direction interval; if it is 12°, it is assigned to the horizontal main direction interval. In this way, each vector line segment is assigned a primary direction interval value, which represents its primary extension direction category, providing a basis for subsequent pixel-level direction label assignment.

[0031] Secondly, the polarization reflection image has a fixed resolution after acquisition, such as 2048×2048 pixels, and corresponds to the actual physical size within the microscope's field of view, such as a 500-micrometer × 500-micrometer area covering the chip. The resolution can be obtained by reading the size information of the image file, and the actual field of view can be determined using the microscopic imaging parameters. Subsequently, it is necessary to establish a correspondence between the geometric coordinates in the layout and the image pixel coordinates. For example, if the layout unit is micrometers and the image unit is pixels, the layout coordinates can be scaled proportionally to match the image coordinates in numerical range. Simultaneously, translation adjustments are made based on the chip's actual position in the image. For example, if the layout origin corresponds to pixel (200, 150) in the image, coordinate translation alignment is required. If the image has slight rotation, angle correction can also be performed. After scaling, translation, and necessary rotation processing, the vectorized metal interconnect layer graphic data can be mapped to a two-dimensional coordinate space consistent with the polarization reflection image, ensuring that the layout geometry accurately corresponds to the image pixels in spatial position.

[0032] Furthermore, after completing the coordinate mapping, the vector graphics need to be converted into a pixel-level representation consistent with the image resolution, i.e., rasterization. Specifically, this involves iterating through each pixel in the image, for example, from (0,0) to (2047,2047), and determining whether the pixel's coordinates fall within a certain metal interconnect layer's graphic area. If a pixel is located inside a polygon of a layer, it is assigned the layer identifier of that metal layer; simultaneously, based on the main direction interval value of the vector line segment corresponding to the pixel's location, it is assigned the direction category. For example, if pixel (500,800) is located within the area covered by a horizontal line segment of the second metal layer, its metal layer number is M2, and its main direction interval is horizontal. By judging and assigning values ​​to each pixel, a two-dimensional label layer with the same size as the original polarized reflection image is finally formed, where each pixel carries a metal layer identifier and a direction interval value.

[0033] Finally, in actual chip structures, different metal interconnect layers may overlap under two-dimensional projection; for example, an upper metal line may cover a lower metal line. When traversing all pixels, if a pixel is found to fall within the mapping area of ​​two or more metal layers simultaneously, its actual visible layer needs to be determined based on the process stacking order. Typically, metal layers in chip processes have clear layer numbers; for example, M3 is above M2, and M2 is above M1. Therefore, the vertical relationship can be determined based on the layer number or a preset vertical order table. If a pixel belongs to the regions of both layers M2 and M3, the uppermost layer M3 is selected as the effective direction source for that pixel, and its principal direction interval value is used as the direction label for that pixel. By performing similar judgment processing on all overlapping pixels, a direction label layer containing only a single direction label is finally formed. This layer spatially corresponds completely to the polarization reflection image and provides unique metal layer direction information for each pixel, providing an accurate structural reference for subsequent polarization response direction comparison and resonance region identification.

[0034] S3, continuously adjust the illumination polarization angle corresponding to the first resonance region according to the preset scanning step size to obtain the polarization rotation path of the first resonance region; In this embodiment, the illumination polarization angle corresponding to the first resonance region is continuously adjusted according to a preset scanning step size to obtain the polarization rotation path of the first resonance region, specifically as follows: Extract the geometric center coordinates of each first resonance region and determine the metal interconnect layer to which the geometric center coordinates belong in the direction label layer; Obtain the standard wire extension direction of the metal interconnect layer, and expand the preset angle range to both sides based on the standard wire extension direction to obtain the scanning range of the illumination polarization angle. Within the scanning range, the illumination polarization angle is changed sequentially with a preset scanning step size to obtain several adjusted illumination polarization angles. Under each adjusted illumination polarization angle, the polarization reflection image of the location of the first resonance region is re-acquired, and the average light intensity response value of all pixels in the first resonance region is extracted. Using the adjusted illumination polarization angle as the abscissa and the corresponding average light intensity response value as the ordinate, the resonance response curve of the first resonance region is plotted, and the trajectory of the change of the resonance response curve is defined as the polarization rotation path of the first resonance region.

[0035] It should be noted that after obtaining the first resonance region, each first resonance region is essentially a connected region composed of several spatially adjacent resonant pixels. Therefore, the geometric center coordinates can be extracted by statistically analyzing the coordinates of all pixels within this connected region. Specifically, this involves traversing all pixels within the first resonance region. For example, if a region contains pixels (520, 800), (521, 800), (522, 801), and so on, totaling 300 pixels, the average of the x-coordinates of all pixels is used to obtain the center x-coordinate, and the average of the y-coordinates is used to obtain the center y-coordinate. For example, the final geometric center might be (535, 812). This coordinate represents the center position of the first resonance region in image space. Subsequently, the metal interconnect layer identifier corresponding to the geometric center coordinates is read from the direction label layer. For example, if the label corresponding to this coordinate is layer M3, then the metal interconnect layer to which the first resonance region belongs is determined to be layer M3. If there is a boundary at the geometric center, the layer to which the majority of pixels in the region belong can also be statistically analyzed, and the metal layer with the largest proportion is selected as the layer to which the first resonance region belongs, thus ensuring the stability of the layer determination.

[0036] Secondly, the standard conductor extension direction of a metal interconnect layer refers to the main routing direction of that metal layer in chip design, usually obtained from layout statistics. For example, if a metal layer is mainly used for horizontal routing, its standard conductor extension direction is 0°; if it is mainly used for vertical routing, it is 90°; if the layer has a significant oblique routing structure, the direction with the highest statistical proportion is taken as the standard direction. Specifically, this can be obtained by statistically analyzing the proportion of each main direction interval in the vectorized set of metal layer segments. For example, if horizontal segments account for 70% and vertical segments account for 30%, then the horizontal direction is defined as the standard conductor extension direction of that layer. After determining this standard direction, for example, 80°, a preset angle range is extended to both sides based on this angle, for example, 20° clockwise and 30° counterclockwise, forming a scanning range of 50° to 110°. In this way, the polarization scan is concentrated within the angle range closely related to the conductor structure, which helps to enhance the detection sensitivity of abnormal responses along the conductor direction.

[0037] Furthermore, the polarization reflection image, after acquisition, possesses clear resolution information, such as an image size of 2048×2048 pixels. Combined with the magnification of the microscopic imaging, the actual chip coverage area can be determined to be 400 micrometers × 400 micrometers. The resolution can be obtained by reading the image attributes, and the field of view can be determined using imaging parameters. Subsequently, the vectorized geometric data of the metal interconnect layer is scaled proportionally. For example, if the layout unit is micrometers, the 400-micrometer range is mapped to a 2048-pixel range, resulting in a ratio of approximately 5.12 pixels per micrometer. Simultaneously, translation adjustments are made based on the chip's positioning in the image, ensuring that the layout origin aligns with the actual alignment point in the image; slight rotation errors can be corrected by angle adjustments. Through scaling, translation, and rotation, the vectorized metal interconnect layer pattern and the polarization reflection image are precisely aligned in space, thus ensuring an accurate and reliable correspondence between the center position of the first resonant region and the layout layer information.

[0038] After obtaining the scanning range, for example, if the scanning range is determined to be 50° to 110°, and the preset scanning step size is set to 5°, then starting from 50°, the scan size is increased by 5° sequentially to obtain a series of illumination polarization angles: 50°, 55°, 60°, 65°... up to 110°. If the step size is 2°, a denser sequence of angles is obtained, such as 50°, 52°, 54°...110°. In actual operation, by rotating the incident light polarizer or polarization modulation device, the incident light is made to reach the above angles sequentially. Each adjustment of the angle completes a new polarization state setting, thereby obtaining multiple continuously changing illumination polarization angles for subsequent fine response scanning of the first resonance region.

[0039] Furthermore, at each adjusted illumination polarization angle, the chip is re-imaged, and an image region containing the location of the first resonant region is captured. For example, after acquiring an image at a 60° polarization angle, the pixel range containing the first resonant region is located, such as the region from (500, 780) to (560, 840). Then, the grayscale values ​​of all pixels within this region are read, for example, a total of 350 pixels. The average of these pixel grayscale values ​​is calculated to obtain the average light intensity response value at that angle, for example, 148. The same operation is repeated at a 65° polarization angle, which may yield an average light intensity of 155, and at 70°, 160. By repeating the above steps for each scanning angle, a set of "angle-average light intensity" corresponding data can be obtained, reflecting the overall response change of the first resonant region rather than a single pixel, thereby reducing the influence of random noise.

[0040] Finally, using the adjusted illumination polarization angles obtained above as the horizontal scale and the corresponding average light intensity response values ​​as the vertical values, the data points are sequentially plotted in a two-dimensional coordinate space, for example, (50°, 140), (55°, 145), (60°, 148), (65°, 155), (70°, 160)... These points are then connected sequentially to form a continuous curve, which is the resonance response curve of the first resonance region. This curve fully describes the overall light intensity change trend of the first resonance region as the polarization direction rotates within the scanning angle range. The polarization rotation path of the first resonance region refers to the continuous trajectory formed by the average light intensity response of this region when the illumination polarization direction rotates continuously within a specific interval. It reflects the sensitivity of the structural region to the polarization direction and the shift of the response peak. If the path exhibits a smooth single-peak structure, it indicates that the structural response is stable; if abrupt changes, double peaks, or severe local fluctuations occur, it may indicate the presence of microcracks or localized stress anomalies. Therefore, the polarization rotation path is a dynamic characterization of the directional optical behavior of the first resonance region, providing a basis for the subsequent identification of path abrupt change segments.

[0041] In this embodiment, the standard wire extension direction of the metal interconnect layer is obtained, and based on the standard wire extension direction, a preset angle range is extended to both sides to obtain the scanning range of the illumination polarization angle, specifically: Obtain the standard conductor extension direction of the metal interconnect layer to which the first resonant region belongs, define the angle of this direction as the reference angle, and calculate the angle of the perpendicular direction of the reference angle; Centered on the reference angle, the first angle range is extended to the side closer to the vertical angle, while the second angle range is extended to the side farther away from the vertical angle, wherein the first angle range is greater than the second angle range. The reference angle, the first angle range, and the second angle range are combined to form an asymmetrical continuous angle range. Arrange all angle values ​​within this asymmetric continuous angle range in ascending order to obtain the scanning range of the illumination polarization angle.

[0042] It should be noted that to obtain the standard conductor extension direction of the metal interconnect layer to which the first resonant region belongs, it is first necessary to determine the metal layer number to which the first resonant region belongs based on the aforementioned direction label layer, for example, determining that it belongs to layer M2. Then, in the vectorized graphic data of this metal layer, the distribution of the main direction intervals of all line segments is statistically analyzed. For example, if the statistical results show that horizontal line segments account for 65%, vertical line segments account for 30%, and the remainder are a small number of oblique structures, then the direction angle corresponding to the horizontal direction (e.g., 0° or 180°) is determined as the standard conductor extension direction of this metal layer. If the layer mainly consists of vertical wiring, then the standard direction is 90°. After determining the standard conductor extension direction, its angle value is defined as the reference angle. For example, if the standard direction is 80°, then 80° is used as the reference angle. Next, the vertical angle of this reference angle is calculated, that is, the direction that differs from it geometrically by 90°. For example, the vertical direction of 80° is 170° or -10° (which can be uniformly converted to the range of 0° to 180° and represented as 170°). This vertical direction represents the direction orthogonal to the direction of the conductor, and usually corresponds to different reflection characteristics in polarization response. Therefore, it is necessary to clearly distinguish between the reference direction and its vertical direction.

[0043] Secondly, when performing asymmetric expansion centered on a reference angle, the angular relationship between the reference angle and the vertical direction must first be clarified. For example, if the reference angle is 80° and the vertical direction is 170°, then the side closer to the vertical direction extends from 80° to 170°, while the side farther from the vertical direction extends in the opposite direction. Setting the first angle range to 30° and the second angle range to 15°, expanding 30° from 80° towards the side closer to 170° results in an expansion to 110°; expanding 15° away from 170° results in an expansion to 65°. The first angle range is larger than the second angle range to allow for a wider scan on the side closer to the vertical direction, as microcracks often cause the response direction to shift towards the vertical or produce significant fluctuations, thus requiring more comprehensive coverage of this area. The first angle range refers to the angular span extending from the reference angle towards the vertical direction, and is usually set to a larger value, such as 25° to 40°; the second angle range refers to the angular span extending from the reference angle towards the direction away from the vertical direction, and is usually set to a smaller value, such as 10° to 20°. This asymmetric setting can enhance sensitivity to potential abnormal directions while ensuring scanning efficiency.

[0044] Furthermore, merging the reference angle, the first angle range, and the second angle range effectively defines a continuous start and end angle interval. For example, in the example above, the reference angle is 80°, extending vertically to 110° on one side and to 65° on the other, resulting in an overall angle interval of 65° to 110°. This interval includes the reference angle itself and covers the entire range of the extensions on both sides. However, due to the different extension angles on both sides, this interval is asymmetrical relative to the reference angle; that is, the reference angle is located inside the interval but not in the center. If the reference angle is 40°, the vertical angle is 130°, the first angle range is 35°, and the second angle range is 15°, then the final interval might be 25° to 75°. In this way, the reference angle and the two different extension ranges together define a continuous angle interval with directional bias, which is the effective range defined by the subsequent fine scanning.

[0045] Finally, after determining the asymmetric continuous angle range, all specific scanning angle values ​​within this range need to be arranged in ascending order to form the scanning range of the illumination polarization angle. For example, if the range is 65° to 110°, and the scanning step size is set to 5°, then a series of angle values ​​of 65°, 70°, 75°, 80°, 85°, 90°, 95°, 100°, 105°, and 110° are generated, arranged in ascending order of value. If the step size is 2°, a denser sequence is generated, such as 65°, 67°, 69° up to 110°. The scanning range of the illumination polarization angle refers to the set of angles actually used to sequentially adjust the polarization direction of the incident light during polarization detection. It defines the range and resolution of polarization direction changes during detection. By using an asymmetric range, more comprehensive angle coverage can be achieved on the side more likely to exhibit abnormal responses, thereby improving the detection sensitivity for directional shift phenomena caused by microcracks, while avoiding excessive invalid scanning in low-sensitivity areas.

[0046] S4. Extract the path abrupt change segment in the polarization rotation path and mark the crack feature of the first resonance region based on the path abrupt change segment to obtain the microcrack candidate region. In this embodiment, path abrupt change segments in the polarization rotation path are extracted, and crack feature marking is performed on the first resonance region based on the path abrupt change segments to obtain microcrack candidate regions, specifically: The second derivative of the polarization rotation path is calculated to obtain the curvature change value at each point on the polarization rotation path, and the continuous interval where the curvature change value exceeds the preset curvature abruptness threshold is marked as the initial abruptness segment. Extract the illumination polarization angle interval corresponding to each initial mutation segment, and obtain the polarization reflection image sequence of the first resonance region within the illumination polarization angle interval; Calculate the light intensity fluctuation variance of each pixel in the polarization reflection image sequence to obtain the light intensity fluctuation distribution map; Pixels with variance values ​​exceeding a preset fluctuation threshold in the light intensity fluctuation distribution map are extracted as fluctuation points, and morphological closing operations are performed on the fluctuation points to form several fluctuation clusters. The fluctuation cluster region is matched with the initial mutation segment in terms of spatial location and angular range, and candidate regions for microcracks are identified based on the matching results.

[0047] It should be noted that calculating the second derivative of the polarization rotation path essentially involves analyzing the change in curvature of the resonance response curve in the first resonance region during continuous angular variation. In practice, a set of "angle-average light intensity" data arranged in ascending order of angle has been obtained, for example, data is collected every 5° within the range of 65° to 110°, yielding corresponding light intensity values ​​of 140, 145, 152, 160, 158, 130, 128, 150, and 165. First, the rate of change in light intensity between adjacent angles is observed, and then the trend is compared to see if there is a sudden reversal. For example, if the light intensity drops sharply from 160 to 130 between 80° and 85°, and then rapidly rises back to 150 between 90° and 95°, this drastic change from rising to a sharp drop and then back to rising indicates a significant increase in the curvature of the curve. By continuously comparing the trend before and after each angle point, it can be determined whether there is a significant abrupt change in curvature near that point. When the curvature change value within a continuous angular segment is significantly higher than a preset curvature abrupt change threshold, for example, if the threshold is set to "the change in curvature between two adjacent intervals exceeds twice the overall average level", then this continuous angular range is marked as the initial abrupt change segment. The so-called initial abrupt change segment refers to the angular range that exhibits an abnormally sharp turn in the polarization rotation path. It reflects that the optical response behavior of the first resonance region within this angular range is no longer smooth and stable, and may be affected by enhanced scattering caused by local structural damage or microcracks. Therefore, it is the key angular range for subsequent detailed analysis.

[0048] After identifying the initial abrupt change segment, for example, determining that 85° to 95° is an initial abrupt change segment, this angular interval is extracted as the key analysis area. Subsequently, the original acquisition process is traced back, and polarization reflection images of the first resonance region are acquired at specific polarization angles of 85°, 90°, and 95°. In other words, a sub-region containing the pixel range of the first resonance region is extracted from the overall image; for example, this region is 60×60 pixels in size, and corresponding image segments are obtained at the three angles of 85°, 90°, and 95°, forming an image sequence sorted by angle. This sequence reflects the actual image changes of the first resonance region within the curvature abrupt change angle segment, providing a data foundation for further analysis of pixel-level light intensity fluctuations.

[0049] Secondly, after obtaining the image sequence within this angular range, the light intensity fluctuation of each pixel in the sequence is statistically analyzed. For example, for a pixel (520, 810) in the first resonance region, the gray values ​​at 85°, 90°, and 95° are 150, 120, and 155 respectively, indicating that this pixel has a relatively large fluctuation within this angular range; while a neighboring pixel has gray values ​​of 148, 150, and 147 at the three angles respectively, indicating a smaller fluctuation. The dispersion of light intensity change for each pixel in the image sequence is statistically analyzed; the larger the value, the more drastic the change within this angular range. The fluctuation levels of all pixels are redrawn in grayscale or pseudo-color form to form a two-dimensional distribution map, thus obtaining the light intensity fluctuation distribution map. In this map, pixels with drastic fluctuations appear as high-value areas, and areas with stable fluctuations appear as low-value areas, thus visually displaying the location of abnormal changes in space.

[0050] Furthermore, in the light intensity fluctuation distribution map, abnormal pixels are filtered out according to a preset fluctuation threshold. For example, a fluctuation exceeding 1.5 times the overall average level is considered an abnormal fluctuation point. All pixels exceeding this threshold are extracted to form a binary image containing only high-fluctuation pixels; these pixels are the fluctuation points. Since individual fluctuation points may be discretely distributed, morphological closing operations can be performed on the binary image to eliminate small gaps and enhance spatial coherence. This involves first performing a dilation operation to connect adjacent fluctuation points, and then performing an erosion operation to restore the overall contour. After the closing operation, the originally scattered fluctuation points will form several spatially continuous clustered regions. The so-called fluctuation clustered region refers to a spatially concentrated area exhibiting significant light intensity instability within a curvature abrupt change angle segment. It usually corresponds to areas where there is stress concentration or crack disturbance in the local structure, and is more physically significant than a single pixel.

[0051] Finally, the fluctuation cluster region is matched with the initial abrupt change segment using both spatial and angular matching. Angular matching confirms that the image sequence corresponding to the fluctuation cluster region indeed originates from the angular range of the initial abrupt change segment, for example, all within the 85° to 95° range. Spatial matching determines whether the fluctuation cluster region is located within the first resonance region or highly coincides with its boundary. For example, if a fluctuation cluster region is located near the center of the first resonance region and exhibits significant fluctuations only within the 85° to 95° angular range, while remaining stable in other angular ranges, it indicates a correspondence between this spatial region and the curvature abrupt change in the polarization rotation path. Fluctuation cluster regions that satisfy both spatial overlap and angular correspondence can be identified as candidate microcrack regions. In other words, only when a spatial region synchronously exhibits pixel-level abnormal light intensity fluctuations within the angular range where the polarization response curve undergoes a sharp inflection is it identified as a potential microcrack region, thereby improving the accuracy of identification and reducing the probability of false detection.

[0052] S5 generates chip microcrack detection results based on the spatial distribution of microcrack candidate regions.

[0053] In this embodiment, the chip microcrack detection results are generated based on the spatial distribution of the microcrack candidate regions, specifically as follows: All microcrack candidate regions are numbered, and the contour lines of each microcrack candidate region are extracted; Calculate the minimum bounding rectangle of the outline and obtain the major axis direction and size information of the minimum bounding rectangle; Calculate the angle between the major axis direction and the wire extension direction of the metal interconnect layer to which the microcrack candidate region belongs; The type of microcrack is determined by identifying the included angle; The total area and number of all microcrack candidate regions are counted, and the microcrack surface density is calculated based on the detection area of ​​the chip. At the same time, the geometric dimension information of each microcrack candidate region is recorded. By integrating the information on the type of microcrack, surface density, number, and geometric dimensions of microcracks, a chip microcrack detection result is generated.

[0054] It should be noted that after identifying the microcrack candidate regions, several independent connected regions are typically obtained. These regions are represented as sets of non-overlapping pixels in the image. First, these connected regions are sequentially assigned numbers according to the scanning order from left to right and top to bottom, such as C1, C2, C3, etc. The numbering method can be based on the centroid position of the regions in the image, ensuring the uniqueness and traceability of the numbers. Then, for each numbered region, its boundary pixels are extracted, that is, it is determined which pixels in the region are adjacent to the background or other regions. These pixels located at the region edges are then connected sequentially to form a closed boundary curve. For example, a candidate region may contain approximately 120 pixels, and its boundary may be enclosed by 30 consecutive edge pixels. This closed boundary line is the contour line of the microcrack candidate region. The contour line is used to accurately describe the geometric shape of the crack region, providing a basis for subsequent orientation and size analysis.

[0055] Secondly, after obtaining the contour line, it is necessary to calculate its minimum bounding rectangle, that is, to find a rectangle that can completely enclose the contour line with the smallest area. In practice, the contour line can be aligned at different rotation angles, and the area of ​​the bounding rectangle formed in each direction can be compared. The one with the smallest area is selected as the minimum bounding rectangle. For example, if the bounding rectangle formed by a contour line without rotation is 20 pixels × 8 pixels, while the bounding rectangle formed after rotating 15° is 18 pixels × 6 pixels, then the latter is selected as the optimal result. The direction of the longer side of this rectangle is defined as the major axis direction. For example, if the major axis is tilted 30° relative to the horizontal direction of the image, then the major axis direction is 30°. The lengths of the long and short sides of the rectangle are also recorded, for example, 18 pixels for the long side and 6 pixels for the short side, and converted to actual dimensions according to the image resolution, for example, 3.6 micrometers × 1.2 micrometers. This information reflects the main extension direction and scale characteristics of the crack candidate region.

[0056] After obtaining the major axis direction of the smallest bounding rectangle, this angle is compared with the extension direction of the conductors in the metal interconnect layer to which the region belongs. The conductor extension direction can be read from the direction label layer. For example, if the standard conductor direction of this layer is 80°, and the major axis direction of the crack candidate region is 95°, then the difference between the two is 15°. If the major axis direction is 10° and the conductor direction is 0°, then the difference is 10°. By calculating this angle difference, the relationship between the crack propagation direction and the metal conductor structure can be determined. It is usually necessary to uniformly convert the angle difference to a range of 0° to 90° for comparison to avoid ambiguity caused by different direction representation methods. This angle is an important basis for determining the nature of the crack.

[0057] Furthermore, when identifying microcrack types based on the included angle, several judgment intervals can be preset. For example, when the included angle is less than 15°, the crack is judged as a "microcrack along the conductor direction," indicating that the crack mainly propagates along the direction of the metal conductor; when the included angle is close to 90°, for example, greater than 75°, it is judged as a "microcrack perpendicular to the conductor direction," indicating that the crack may be caused by transverse stress; when the included angle is in the middle range, for example, between 30° and 60°, it is judged as an "oblique microcrack." The so-called microcrack classification refers to classifying and identifying the cause and propagation trend of cracks based on the relationship between the crack's geometric direction and the direction of the metal wiring. Different types often correspond to different failure mechanisms. For example, a microcrack along the conductor direction may be related to fatigue of the material inside the conductor, while a microcrack perpendicular to the conductor direction may be related to interlayer stress concentration or thermal cycling. Therefore, the classification results have engineering analysis significance.

[0058] Furthermore, during the statistical phase, the number of pixels in each microcrack candidate region is first calculated. For example, region C1 contains 120 pixels, C2 contains 85 pixels, and C3 contains 200 pixels. The number of pixels in all regions is then summed, for example, a total of 405 pixels. If each pixel in the image corresponds to an actual area of ​​0.04 square micrometers, the total crack area is approximately 16.2 square micrometers. Based on the chip's detection coverage area, for example, a detection area of ​​500 micrometers × 500 micrometers, or 250,000 square micrometers, the crack area ratio can be calculated, thus obtaining the microcrack surface density index. Simultaneously, the number of cracks is counted; for example, three microcrack candidate regions were detected. For each region, the geometric dimensions, such as the long side dimension, short side dimension, and corresponding directional angle of its minimum bounding rectangle, are also recorded. This data can be used to evaluate the crack size distribution.

[0059] Finally, the above information is integrated to form a complete chip microcrack detection result. The integrated content includes: the crack number, the metal layer to which it belongs, the classification type, the major axis direction, and the actual size; overall statistical information includes the total number of cracks, the total area, the areal density index, and the distribution of the number of each type of crack. For example, the detection result can be described as: a total of 3 microcracks were found in this detection area, of which 2 were along the conductor direction and 1 was perpendicular to the conductor direction; the maximum crack size was 3.6 μm × 1.2 μm; the overall crack areal density was 0.006%. This comprehensive expression method reflects the structural integrity status of the chip from both individual characteristics and overall statistics, providing a basis for subsequent reliability assessment or process optimization.

[0060] Example 2, Figure 2 The present invention provides a chip microcrack visual inspection system based on multi-angle polarization imaging, comprising a data acquisition module, a feature extraction module, a path generation module, a feature marking module, and a detection output module. The data acquisition module is used to acquire polarization reflection images of the chip under illumination at different polarization angles, and extract the light intensity response values ​​of the polarization reflection images at different polarization angles to obtain polarization response curves. The feature extraction module is used to identify the polarization response direction difference features of the metal interconnect layer to which each pixel belongs in the image, and compare the polarization response curve with the polarization response direction difference features to obtain the first resonance region; The path generation module is used to continuously adjust the illumination polarization angle corresponding to the first resonance region according to a preset scanning step size to obtain the polarization rotation path of the first resonance region. The feature marking module is used to extract path abrupt change segments in the polarization rotation path and mark the crack features of the first resonance region based on the path abrupt change segments to obtain microcrack candidate regions. The detection output module is used to generate chip microcrack detection results based on the spatial distribution of microcrack candidate regions.

[0061] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0062] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0063] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0064] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0065] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for visual inspection of chip microcracks based on multi-angle polarization imaging, characterized in that, Includes the following steps: The polarization reflection images of the chip under illumination at different polarization angles are acquired, and the light intensity response values ​​of the polarization reflection images at different polarization angles are extracted to obtain the polarization response curves. The polarization response direction difference characteristics of the metal interconnect layer to which each pixel belongs in the image are identified, and the polarization response curve is compared with the polarization response direction difference characteristics to obtain the first resonance region; The illumination polarization angle corresponding to the first resonance region is continuously adjusted according to the preset scanning step size to obtain the polarization rotation path of the first resonance region. Extract the path abrupt change segment in the polarization rotation path and mark the crack feature of the first resonance region based on the path abrupt change segment to obtain the microcrack candidate region; The chip microcrack detection results are generated based on the spatial distribution of the microcrack candidate regions.

2. The chip microcrack visual inspection method based on multi-angle polarization imaging according to claim 1, characterized in that, The process of extracting the light intensity response values ​​of the polarization reflection image at different polarization angles to obtain the polarization response curve is as follows: The grayscale value of each polarized reflection image is read pixel by pixel, and the grayscale value is used as the initial light intensity response value of the pixel under the current illumination polarization angle. The initial light intensity response values ​​obtained at the same pixel location under different polarization angles are arranged in ascending order of polarization angle to obtain the initial light intensity sequence. The initial light intensity sequence is mapped onto a polarization reflection image coordinate system with the illumination polarization angle as the horizontal axis and the light intensity response value as the vertical axis. The polarization response curve is obtained by fitting the mapped discrete points using a piecewise cubic spline interpolation algorithm.

3. The chip microcrack visual inspection method based on multi-angle polarization imaging according to claim 2, characterized in that, The method involves identifying the polarization response direction difference characteristics of the metal interconnect layer to which each pixel belongs in the image, and comparing the polarization response curve with the polarization response direction difference characteristics to obtain the first resonance region, specifically: Obtain chip layout design data, extract the layout direction information of each metal interconnect layer and map the layout direction information to the coordinate system, mark the extension direction of the wires of the metal interconnect layer to which each pixel belongs, and obtain the direction label layer; Based on the orientation label layer, each pixel is traversed, and the correlation between the polarization response curve of the pixel and the polarization response curves of all pixels in the same neighboring layer is calculated. Based on the results, the off-target pixels are identified. Obtain the illumination polarization angle corresponding to the peak position of the polarization response curve that deviates from the pixel and compare it with the extension direction of the wire in the metal interconnect layer to which the deviated pixel belongs, and calculate the angle difference between the two. When the angle difference is less than the preset angle matching threshold, the deviating pixel is marked as a resonance point; All resonance points are clustered to form several connected regions, which are then used as the first resonance region.

4. The chip microcrack visual inspection method based on multi-angle polarization imaging according to claim 3, characterized in that, The process involves extracting the layout direction information of each metal interconnect layer and mapping this information to a coordinate system. Each pixel is then labeled with the extension direction of the wires in its respective metal interconnect layer, resulting in a direction label layer. Specifically: The chip layout design data is analyzed, the graphic data of each metal interconnect layer is extracted, and the polygon lines in the graphic data of each metal interconnect layer are vectorized to obtain the vectorized metal interconnect layer graphic data. The orientation angle of each vector line segment is calculated based on the vectorized metal interconnect layer graphic data. The orientation angle is then quantized into several preset main orientation intervals to obtain the main orientation interval values ​​of the vector line segments. The resolution and field of view of the polarization reflection image are obtained, and the vectorized metal interconnect layer graphic data are mapped to the coordinate system through affine transformation based on the resolution and field of view of the polarization reflection image. In the coordinate system, the graphic data of each mapped metal interconnect layer is rasterized and each pixel is assigned the metal interconnect layer identifier of its location and the main direction interval value of the vector line segment to which it belongs in that layer. Traverse all pixels. If a pixel is located in the overlapping area of ​​multiple metal interconnect layers, select the layout direction of the top metal interconnect layer as the direction label of the pixel according to the vertical projection order of each layer at the pixel, and obtain the direction label layer.

5. The chip microcrack visual inspection method based on multi-angle polarization imaging according to claim 4, characterized in that, The step of continuously adjusting the illumination polarization angle corresponding to the first resonance region according to a preset scanning step size to obtain the polarization rotation path of the first resonance region is as follows: Extract the geometric center coordinates of each first resonance region and determine the metal interconnect layer to which the geometric center coordinates belong in the direction label layer; Obtain the standard wire extension direction of the metal interconnect layer, and expand the preset angle range to both sides based on the standard wire extension direction to obtain the scanning range of the illumination polarization angle. Within the scanning range, the illumination polarization angle is changed sequentially with a preset scanning step size to obtain several adjusted illumination polarization angles. Under each adjusted illumination polarization angle, the polarization reflection image of the location of the first resonance region is re-acquired, and the average light intensity response value of all pixels in the first resonance region is extracted. Using the adjusted illumination polarization angle as the abscissa and the corresponding average light intensity response value as the ordinate, the resonance response curve of the first resonance region is plotted, and the trajectory of the change of the resonance response curve is defined as the polarization rotation path of the first resonance region.

6. The chip microcrack visual inspection method based on multi-angle polarization imaging according to claim 5, characterized in that, The process involves obtaining the standard wire extension direction of the metal interconnect layer, and using this standard wire extension direction as a reference, expanding to both sides by a preset angle range to obtain the scanning range of the illumination polarization angle. Specifically: Obtain the standard conductor extension direction of the metal interconnect layer to which the first resonant region belongs, define the angle of this direction as the reference angle, and calculate the angle of the perpendicular direction of the reference angle; Centered on the reference angle, the first angle range is extended to the side closer to the vertical angle, while the second angle range is extended to the side farther away from the vertical angle, wherein the first angle range is greater than the second angle range. The reference angle, the first angle range, and the second angle range are combined to form an asymmetrical continuous angle range. Arrange all angle values ​​within this asymmetric continuous angle range in ascending order to obtain the scanning range of the illumination polarization angle.

7. The chip microcrack visual inspection method based on multi-angle polarization imaging according to claim 6, characterized in that, The process of extracting path abrupt change segments in the polarization rotation path and marking crack features in the first resonance region based on these segments to obtain microcrack candidate regions is as follows: The second derivative of the polarization rotation path is calculated to obtain the curvature change value at each point on the polarization rotation path, and the continuous interval where the curvature change value exceeds the preset curvature abruptness threshold is marked as the initial abruptness segment. Extract the illumination polarization angle interval corresponding to each initial mutation segment, and obtain the polarization reflection image sequence of the first resonance region within the illumination polarization angle interval; Calculate the light intensity fluctuation variance of each pixel in the polarization reflection image sequence to obtain the light intensity fluctuation distribution map; Pixels with variance values ​​exceeding a preset fluctuation threshold in the light intensity fluctuation distribution map are extracted as fluctuation points, and morphological closing operations are performed on the fluctuation points to form several fluctuation clusters. The fluctuation cluster region is matched with the initial mutation segment in terms of spatial location and angular range, and candidate regions for microcracks are identified based on the matching results.

8. The chip microcrack visual inspection method based on multi-angle polarization imaging according to claim 7, characterized in that, The process of generating chip microcrack detection results based on the spatial distribution of microcrack candidate regions specifically involves: All microcrack candidate regions are numbered, and the contour lines of each microcrack candidate region are extracted; Calculate the minimum bounding rectangle of the outline and obtain the major axis direction and size information of the minimum bounding rectangle; Calculate the angle between the major axis direction and the wire extension direction of the metal interconnect layer to which the microcrack candidate region belongs; The type of microcrack is determined by identifying the included angle; The total area and number of all microcrack candidate regions are counted, and the microcrack surface density is calculated based on the detection area of ​​the chip. At the same time, the geometric dimension information of each microcrack candidate region is recorded. By integrating the information on the type of microcrack, surface density, number, and geometric dimensions of microcracks, a chip microcrack detection result is generated.

9. A chip microcrack visual inspection system based on multi-angle polarization imaging, applied to the chip microcrack visual inspection method based on multi-angle polarization imaging as described in any one of claims 1-8, characterized in that, It includes a data acquisition module, a feature extraction module, a path generation module, a feature labeling module, and a detection output module: The data acquisition module is used to acquire polarization reflection images of the chip under illumination at different polarization angles, and extract the light intensity response values ​​of the polarization reflection images at different polarization angles to obtain polarization response curves. The feature extraction module is used to identify the polarization response direction difference features of the metal interconnect layer to which each pixel belongs in the image, and compare the polarization response curve with the polarization response direction difference features to obtain the first resonance region; The path generation module is used to continuously adjust the illumination polarization angle corresponding to the first resonance region according to a preset scanning step size to obtain the polarization rotation path of the first resonance region. The feature marking module is used to extract path abrupt change segments in the polarization rotation path and mark the crack features of the first resonance region based on the path abrupt change segments to obtain microcrack candidate regions. The detection output module is used to generate chip microcrack detection results based on the spatial distribution of microcrack candidate regions.

Images