Chip non-destructive testing system based on embedded current scanning

Abstract

This invention relates to the field of chip inspection technology, and more particularly to a chip non-destructive testing system based on embedded current scanning. The system includes a current scanning module, an information acquisition module, a data analysis module, an abnormal scanning region generation module, and a scanning path generation module. Instead of simply performing a higher-density secondary scan of the entire chip, this invention intelligently filters out abnormal scanning regions requiring special attention based on the initial scan results and calculates their scanning priority. This allows valuable inspection time and computing resources to be concentrated on the most suspicious and severe areas, avoiding unnecessary repeated scanning in normal areas. The system dynamically selects the optimal scanning path based on the regularity of the region shape; for example, parallel lines are used for regular regions, and radial lines are used for irregular regions. This makes the scanning path more closely match the geometric features of the defect, obtaining a more complete region characterization with fewer scanning points, resulting in higher path planning efficiency.

Description

Technical Field

[0001] This invention relates to the field of chip testing technology, and in particular to a chip non-destructive testing system based on embedded current scanning. Background Technology

[0002] As semiconductor process nodes continue to evolve (e.g., into the nanoscale), the integration, complexity, and fragility of chips are increasing dramatically. Traditional physical destructive analysis (such as opening packages and cross-sectioning) is costly and can damage samples, while many electrical testing methods struggle to accurately locate microscopic or potential defects. Traditional techniques such as current scanning or voltage contrast imaging typically require uniform, high-density scanning of the entire chip or a large area, resulting in large data volumes, time-consuming analysis, and difficulty in quickly focusing on truly abnormal areas. Existing methods often treat each detected abnormal scan point as an independent event, lacking the ability to intelligently identify and differentiate defect clusters (such as sheet defects caused by process variations) or discrete defects (such as point defects caused by random particle injection) from a global spatial distribution perspective.

[0003] Chinese Patent Application Publication No. CN114942378A discloses a non-destructive testing system and method for detecting micro- and nano-magnetic features inside a chip. The system includes a combined vibration and displacement device, a high permeability probe, a voltage amplifier, a lock-in amplifier, and a processor. The method includes the following steps: powering on the chip under test; driving the high permeability probe to vibrate using the combined vibration and displacement device; collecting magnetic information inside the chip using the high permeability probe and transmitting it to the processor via the lock-in amplifier; imaging the current and comparing it with the chip circuit diagram to evaluate defects; and visualizing the test results. This invention provides a non-destructive testing technology for detecting micro- and nano-magnetic features inside a chip by collecting magnetic field information generated by weak currents inside the chip, inverting the current image inside the chip, and processing it. The non-destructive testing process has the advantages of high precision and high practicality.

[0004] However, existing technologies still have the following problems:

[0005] Traditional techniques such as current scanning or voltage contrast imaging typically require uniform, high-density scanning of the entire chip or a large area, resulting in large amounts of data, time-consuming analysis, and difficulty in quickly focusing on the real abnormal areas. Summary of the Invention

[0006] To address this, the present invention provides a chip non-destructive testing system based on embedded current scanning, which overcomes the problems of traditional current scanning or voltage comparison imaging techniques in the prior art, which usually require uniform and high-density scanning of the entire chip or a large area, resulting in large data volume, time-consuming analysis, and difficulty in quickly focusing on the real abnormal area.

[0007] To achieve the above objectives, the present invention provides a chip non-destructive testing system based on embedded current scanning, comprising:

[0008] A current scanning module is used to inject current into the chip under test and control the current to scan the chip under test along a preset test path.

[0009] An information acquisition module, which is connected to the current scanning module, is used to acquire the scanning data of the chip under test;

[0010] The data analysis module, which is connected to the information acquisition module, is used to determine the abnormal scanning points of the chip under test based on the scanning data, calculate the aggregation feature value based on the spatial location of the abnormal scanning points, calculate the point anomaly degree based on the scanning data, and determine the distribution type of the abnormal scanning points based on the aggregation feature value. The distribution type includes aggregation type and dispersion type.

[0011] An abnormal scanning region generation module, connected to the data analysis module, is used to determine the merging method of the abnormal scanning points based on the distribution type of the abnormal scanning points, including:

[0012] The first generation module is used to merge abnormal scan points with an aggregation type distribution based on the aggregation kernel density of abnormal scan points to generate an abnormal scan region.

[0013] The second generation module is used to merge abnormal scan points of the dispersed type based on the dispersed connection radius of the abnormal scan points to generate an abnormal scan region.

[0014] The scanning path generation module, which is connected to the abnormal scanning region generation module and the data analysis module, is used to calculate the region scanning priority of each region based on the regularity of the shape features of the abnormal scanning region and the point anomaly degree of the abnormal scanning points in the region, and to determine the secondary scanning path for each abnormal scanning region based on the region scanning priority and the regularity of the shape features.

[0015] Furthermore, the data analysis module is used to calculate clustering feature values ​​based on the deviation between the local density and global distribution of the abnormal scan points, including:

[0016] Obtain the spatial coordinates of all abnormal scan points on the chip under test;

[0017] Calculate the local density of each abnormal scan point, where the local density is represented by the number of other abnormal scan points within a preset neighborhood radius centered on that point.

[0018] Based on the spatial location of all abnormal scan points, calculate the global distribution center and the overall dispersion.

[0019] For each abnormal scan point, the global distribution deviation of that point is calculated based on its distance from the global distribution center and the overall dispersion.

[0020] By weighted fusion of the local density and global distribution deviation of the abnormal scan points, the clustering feature value is obtained.

[0021] Furthermore, the data analysis module is used to determine the distribution type of abnormal scan points based on the clustering feature values, including:

[0022] If the clustering feature value is greater than or equal to the preset clustering feature value, then the distribution type of the abnormal scan points is determined to be clustering.

[0023] If the clustering feature value is less than the preset clustering feature value, then the distribution type of the abnormal scan points is determined to be the dispersed type.

[0024] Furthermore, the scan path generation module is used to determine the regularity of the shape features of the abnormal scan region, including:

[0025] Extract the edge contour of the abnormal scan area to obtain its geometry;

[0026] Calculate the matching degree between the geometric shape and the preset standard rule shape, which includes a rectangle, a circle, or a polygon;

[0027] The regularity of the shape features of the abnormal scanned region is determined based on the matching degree, wherein the regularity of the shape features is positively correlated with the matching degree.

[0028] Furthermore, the data analysis module is also used to calculate the point anomaly degree of abnormal scan points, including:

[0029] Obtain the current response data corresponding to each abnormal scan point in the first scan;

[0030] The current response data is compared with a preset normal response threshold range, and its deviation in one or multiple dimensions is calculated.

[0031] After normalizing the calculated deviation, it is used as the point anomaly degree of the abnormal scan point.

[0032] Among them, the point anomaly degree is positively correlated with the deviation degree.

[0033] Furthermore, the scan path generation module is used to determine the secondary scan path of the abnormal scan region, including:

[0034] If the regularity of the shape features of the abnormal scanning region is higher than a preset regularity threshold, then the basic scanning path is determined based on the geometric circumscribed shape of the abnormal scanning region.

[0035] If the regularity of the shape features of the abnormal scanning area is lower than or equal to the preset regularity threshold, then the abnormal scanning point with the highest abnormality within the abnormal scanning area is used as the starting focus to determine the radial basic scanning path.

[0036] Furthermore, the scanning path generation module is also used to dynamically adjust the density of scanning points based on the point anomaly degree of each point within the path coverage area of ​​the determined basic scanning path, wherein the density of scanning points is positively correlated with the point anomaly degree of the covered points.

[0037] Furthermore, the scan path generation module is used to determine the basic scan path based on the geometric circumscribed shape of the abnormal scan region, including:

[0038] Identify the type of geometric circumference shape of the abnormal scan region;

[0039] If the circumscribed shape is rectangular, then a parallel line scan path is used;

[0040] If the external shape is circular or elliptical, then a concentric circle scan path is used;

[0041] If the circumscribed shape is a polygon, then a parallel scan line path is generated along its longest diagonal.

[0042] Furthermore, the scan path generation module uses the abnormal scan point with the highest degree of abnormality within the abnormal scan area as the starting focus to determine the radial basic scan path, including:

[0043] Determine the coordinate position of the initial focus;

[0044] Multiple radial scanning rays are generated at preset angular intervals, centered on the initial focal point.

[0045] Based on the edge contour of the abnormal scanning area, the scanning rays that extend beyond the area are truncated to obtain the final basic scanning path.

[0046] Furthermore, the scan path generation module dynamically adjusts the density of scan points based on the anomaly level of each point within the path coverage area of ​​the determined basic scan path, including:

[0047] The basic scanning path is divided into several path segments;

[0048] Obtain the average point anomaly degree of all abnormal scan points within the coverage area of ​​each path segment;

[0049] The target scan point spacing of the path segment is determined based on the preset interval where the average point anomaly is located.

[0050] Based on the target scan point spacing, a final scan point sequence for secondary scanning is generated on the path segment.

[0051] Compared with existing technologies, the beneficial effects of this invention are as follows: the system in this invention does not simply perform a higher-density secondary scan of the entire chip, but intelligently filters out abnormal scanning areas that require special attention based on the results of the first scan, and calculates the scanning priority for these areas. This allows valuable detection time and computing resources to be concentrated on the most suspicious and serious areas, avoiding unnecessary repeated scanning in normal areas. The optimal scanning path is dynamically selected according to the regularity of the area shape, such as using parallel lines for regular areas and radial lines for irregular areas. This makes the scanning path itself more closely match the geometric features of the defect, obtaining a more complete area representation with fewer scanning points, and achieving higher path planning efficiency.

[0052] Furthermore, this invention introduces two orthogonal and complementary dimensions—local density and global distribution deviation—to achieve a more precise characterization. Local density reflects the microscopic aggregation intensity within the vicinity of a point; the higher the density, the denser the anomalies near that point. Global distribution deviation reflects the marginality of a point relative to the macroscopic distribution of the entire set of anomalies; the greater the deviation, the farther the point is from the centroid of all anomalies, and the more likely it is to be an outlier or a member of another independent cluster. This clearly distinguishes core clusters (i.e., high local density, low global deviation), peripheral clusters (i.e., medium-high local density, medium-high global deviation), and discrete isolated points (i.e., low local density, high global deviation), laying the foundation for more refined classification and processing in the future.

[0053] Furthermore, in the anomaly scanning region generation module of this invention, the point set is naturally split into two sub-modules for parallel or serial processing. The logic is clear, avoiding the trade-offs in logical complexity and effectiveness caused by processing mixed-type data in a single algorithm. The system can allocate different computing resources or adopt different optimization strategies for different types of point sets. For example, it is worthwhile to perform relatively time-consuming kernel density calculations on clustered point sets because it can generate meaningful regions; while for dispersed point sets, a lighter-weight connection algorithm can be used. This achieves modularization, specialization, and efficiency of the processing flow. Based on clearly classified data splitting, each downstream module can handle the data types it is good at, improving the efficiency and elegance of the entire system's data processing.

[0054] Furthermore, in this invention, the system no longer processes a featureless patch of anomaly points simply connected together. Instead, it extracts edge contours and abstracts them into geometrically meaningful shapes. This is a crucial step in transforming low-level image data into high-level semantic information. By calculating the matching degree with standard shapes (rectangles, circles, etc.), the system can quantify whether the shape of the region is regular or irregular using a continuous numerical value (matching degree). The higher the matching degree, the closer the region is to a certain regular shape. For the first time, the system introduces morphological features as the core judgment criterion in scanning decisions. This enables the system to identify, for example, the essential difference between an approximately rectangular region (which may correspond to a short circuit or open circuit in the metal connection) and a highly irregular star-shaped region (which may correspond to melting and diffusion caused by a single point of penetration), laying the foundation for subsequent customized scanning.

[0055] Furthermore, this invention introduces a crucial continuous quantification dimension by calculating point anomaly degree. By calculating and normalizing the deviation from the normal threshold, each anomaly point is assigned a comparable and standardized severity score (point anomaly degree). The normalization process ensures that the anomaly degrees calculated under different test conditions, different chips, and even different response parameters (such as current and voltage phase) can be compared fairly on the same scale. This is a prerequisite for subsequent advanced operations such as region priority ranking and path density adjustment. It elevates the granularity of detection from the presence or absence of anomalies to the degree of anomaly, providing the system with a rich data layer far exceeding binary information. This is the fundamental reason why all subsequent intelligent decisions can achieve differentiation and refinement.

[0056] Furthermore, in this invention, for rectangular areas, a parallel line path is used, with the scan line direction parallel to the edge, which can completely cover the area with minimal reciprocating motion. For circular or elliptical areas, a concentric circle or spiral path is used, which can naturally conform to its radial symmetry and avoid scanning blind spots or over-scanning at the corners. With the fewest scan points and the shortest path length, complete and uniform sampling of regular areas is obtained, which greatly improves the scanning speed and reduces the mechanical wear of the equipment. For low-regularity areas, the point with the highest point anomaly is used as the starting focus, and a radial path is used to first and concentrate the scanning resources on the most likely fault core, i.e., the point with the highest anomaly, which is conducive to finding the root cause of the problem as quickly as possible. The radial rays can naturally extend in all directions until they reach the boundary of the area. This mode is particularly effective for exploring branching, star-shaped, and irregularly spreading defect areas, which can well delineate their complex contours. Using an exploratory, inside-out approach, the morphological characteristics and internal severity distribution of irregular areas are efficiently captured, avoiding the problems of a large number of invalid external scans or incomplete internal coverage that may occur when using regular paths.

[0057] Furthermore, this invention employs a parallel line path for rectangular regions. The parallel line path is the theoretically optimal solution for covering rectangular regions. The scan line is parallel to the rectangular sides, resulting in the shortest total path length, no turning redundancy, and achieving non-overlapping and non-omission coverage. For most scanning platforms driven by linear motors, parallel line motion is the fastest, most stable, and least loss-prone motion method, completing the full sampling of rectangular defects (commonly seen in wiring short circuits, misalignment, etc.) in the shortest time and with the lowest mechanical loss. For circular or elliptical regions, a concentric circle scanning path is used. Considering that the concentric circle or spiral path perfectly matches the radial symmetry of the circular region, the scanning points are evenly distributed along the radial direction and have a consistent density along the circumference, best conforming to its geometric nature. If a rectangular grid scan is used for a circular region, data oversampling (redundancy) will occur in the four corner areas, while insufficient sampling may occur at the circular boundary. The concentric circle path completely avoids this problem.

[0058] Furthermore, in this invention, the system no longer treats irregular areas as homogeneous surfaces, but instead clearly identifies the points (with the highest anomaly) that are most likely to represent the root cause of the fault or the most dangerous location. Radial scanning starting from the focal point can quickly acquire detailed data of the area surrounding the core point, and can quickly confirm whether the most abnormal point is an isolated critical point or the core of a larger defect area. This provides the most direct information for quickly determining the nature of the fault, transforming a comprehensive survey into a focused breakthrough. When time is tight or the root cause needs to be quickly investigated, this strategy can ensure that the system uses the shortest initial path to explore the most critical location, achieving a qualitative leap in diagnostic efficiency. Attached Figure Description

[0059] Figure 1 This is a structural block diagram of the chip non-destructive testing system based on embedded current scanning according to the present invention;

[0060] Figure 2 This is a flowchart for calculating the clustering feature values ​​according to the present invention. Detailed Implementation

[0061] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0062] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0063] Please see Figures 1-2 As shown, Figure 1 This is a structural block diagram of the chip non-destructive testing system based on embedded current scanning according to the present invention; Figure 2This is a flowchart for calculating the clustering feature values ​​according to the present invention.

[0064] This invention relates to a chip non-destructive testing system based on embedded current scanning, comprising:

[0065] A current scanning module is used to inject current into the chip under test and control the current to scan the chip under test along a preset test path.

[0066] An information acquisition module, which is connected to the current scanning module, is used to acquire the scanning data of the chip under test;

[0067] The data analysis module, which is connected to the information acquisition module, is used to determine the abnormal scanning points of the chip under test based on the scanning data, calculate the aggregation feature value based on the spatial location of the abnormal scanning points, calculate the point anomaly degree based on the scanning data, and determine the distribution type of the abnormal scanning points based on the aggregation feature value. The distribution type includes aggregation type and dispersion type.

[0068] An abnormal scanning region generation module, connected to the data analysis module, is used to determine the merging method of the abnormal scanning points based on the distribution type of the abnormal scanning points, including:

[0069] The first generation module is used to merge abnormal scan points with an aggregation type distribution based on the aggregation kernel density of abnormal scan points to generate an abnormal scan region.

[0070] The second generation module is used to merge abnormal scan points of the dispersed type based on the dispersed connection radius of the abnormal scan points to generate an abnormal scan region.

[0071] The scanning path generation module, which is connected to the abnormal scanning region generation module and the data analysis module, is used to calculate the region scanning priority of each region based on the regularity of the shape features of the abnormal scanning region and the point anomaly degree of the abnormal scanning points in the region, and to determine the secondary scanning path for each abnormal scanning region based on the region scanning priority and the regularity of the shape features.

[0072] In this invention, the system does not simply perform a higher-density secondary scan of the entire chip. Instead, based on the results of the first scan, it intelligently filters out abnormal scan areas that require special attention and calculates the scan priority for these areas. This allows valuable detection time and computing resources to be concentrated on the most suspicious and serious areas, avoiding unnecessary repeated scans in normal areas. The system also dynamically selects the optimal scan path based on the regularity of the area shape, such as using parallel lines for regular areas and radial lines for irregular areas. This makes the scan path itself more closely match the geometric features of the defect, obtaining a more complete area representation with fewer scan points, and thus improving path planning efficiency.

[0073] Specifically, the data analysis module is used to calculate clustering feature values ​​based on the deviation between the local density and global distribution of the abnormal scan points, including:

[0074] Obtain the spatial coordinates of all abnormal scan points on the chip under test;

[0075] Calculate the local density of each abnormal scan point, where the local density is represented by the number of other abnormal scan points within a preset neighborhood radius centered on that point.

[0076] Based on the spatial location of all abnormal scan points, calculate the global distribution center and the overall dispersion.

[0077] For each abnormal scan point, the global distribution deviation of that point is calculated based on its distance from the global distribution center and the overall dispersion.

[0078] By weighted fusion of the local density and global distribution deviation of the abnormal scan points, the clustering feature value is obtained.

[0079] Specifically, in this embodiment, assuming a current scanning test is performed on a power management chip with a size of 5mm × 5mm, after the initial uniform scan, the system detects 15 abnormal scanning points, whose spatial distribution characteristics are as follows: Abnormal point distribution pattern: In region A (lower left corner), 8 abnormal points are closely clustered, with a distance between them all within 0.1mm, forming a high-density cluster; in region B (upper right corner), 2 abnormal points are relatively close (approximately 0.15mm); the remaining 5 discrete abnormal points are scattered in other locations on the chip, with a distance between them all greater than 0.5mm; the system reads the (x,y) coordinate data of all 15 abnormal points and establishes a coordinate set; the neighborhood radius R is set to 0.2mm (based on chip size and process). (Feature presets); For a point P_A in region A: count the number of other outliers within a 0.2mm radius centered on P_A; since points are dense in region A, this value may be 6-7 (including points other than itself); For a point P_B in region B: there may be 1 adjacent point within a 0.2mm radius; For a discrete point P_C: there are no other outliers within a 0.2mm radius, local density = 0; Global distribution center: calculate the average coordinate position of all 15 points. Since there are 8 points in region A, this center will be significantly biased towards the lower left area of ​​the chip; Overall dispersion: calculate the standard deviation of the distance from all points to this global distribution center, quantifying the overall dispersion of the points; For points in region A: calculate their distance to the global distribution center. Since the global center itself is biased towards region A, this distance is small. Divide this distance by the overall dispersion to obtain a small deviation value (e.g., 0.3); For discrete points: such as isolated points in the upper right corner of the chip, the distance to the global distribution center is very far.Divide this distance by the overall dispersion degree to obtain a relatively large deviation degree value (e.g., 2.5); local density weight = 0.6, global deviation degree weight = 0.4; for the points in region A: high local density score (e.g., 0.8, after normalization); low global deviation degree score (e.g., 0.3); aggregation eigenvalue = 0.6×0.8 + 0.4×(1 - 0.3) = 0.48 + 0.28 = 0.76; local density score = 0; high global deviation degree score (e.g., 2.5, may be 0.9 after normalization); aggregation eigenvalue = 0.6×0 + 0.4×(1 - 0.9) = 0 + 0.04 = 0.04; preset aggregation eigenvalue threshold = 0.5; for the points in region A: aggregation eigenvalue 0.76 > 0.5, determined as the aggregation type; for the discrete points: aggregation eigenvalue 0.04 < 0.5, determined as the dispersion type; for the points in region B: may be near the critical value, classified according to the specific calculation results; enable the kernel density estimation algorithm, the algorithm will generate a smooth density distribution surface, by setting the density threshold (e.g., 50% of the maximum density), automatically outline a continuous abnormal region contour, which can accurately reflect the actual aggregation range formed by 8 points; for the dispersion type points: enable the connection radius algorithm, set the connection radius D = 0.3mm; two points in region B are merged into a small region due to the distance 0.15mm < D; the other 5 discrete points form independent minimum regions because their distances from each other are > D, and the discrete points will not be wrongly connected to the distant aggregation region; the 8 points in region A are correctly identified as the aggregation type; after kernel density merging, the generated region is approximately circular with a diameter of about 0.25mm; engineering inference: may be local process defects (such as uneven ion implantation, local contamination); system response: mark this region as high priority, and perform fine scanning using concentric circle paths subsequently.

[0080] In this invention, by introducing two orthogonal and complementary dimensions of local density and global distribution deviation degree, more accurate characterization is achieved. The local density reflects the microscopic aggregation intensity within the adjacent range of this point. The higher the density, the denser the abnormal points near this point; the global distribution deviation degree reflects the marginality of the position of this point relative to the macroscopic distribution of the entire abnormal point set. The greater the deviation degree, the farther this point is from the center of gravity of all abnormal points, and the more likely it is to be an outlier or a member of another independent cluster; clearly distinguish the core aggregation points (i.e., high local density, low global deviation degree), edge aggregation points (i.e., medium - high local density, medium - high global deviation degree) and discrete isolated points (i.e., low local density, high global deviation degree), which lays a foundation for subsequent more refined classification and processing.

[0081] Specifically, the data analysis module is used to determine the distribution type of abnormal scanning points based on the aggregation eigenvalue, including:

[0082] If the clustering feature value is greater than or equal to the preset clustering feature value, then the distribution type of the abnormal scan points is determined to be clustering.

[0083] If the clustering feature value is less than the preset clustering feature value, then the distribution type of the abnormal scan points is determined to be the dispersed type.

[0084] Specifically, in this embodiment, the preset clustering feature value can be determined in the following ways: statistically analyze the distribution of clustering feature values ​​of abnormal points in the analyzed chip samples; find the natural boundary points through cluster analysis (such as K-means); observe the bimodal or multimodal distribution of the clustering feature value, and select the valley value as the preset clustering feature value.

[0085] In this invention, within the anomaly scanning region generation module, point sets are naturally split into two sub-modules for parallel or serial processing. This clear logic avoids the trade-offs in logical complexity and performance associated with processing mixed-type data in a single algorithm. The system can allocate different computing resources or employ different optimization strategies for different types of point sets. For example, performing relatively time-consuming kernel density calculations on clustered point sets is worthwhile because it generates meaningful regions; while for dispersed point sets, a lighter-weight connection algorithm can be used. This achieves modularization, specialization, and efficiency in the processing flow. Based on clearly categorized data splitting, each downstream module can handle its preferred data types, improving the overall efficiency and elegance of the system's data processing.

[0086] Specifically, the scan path generation module is used to determine the regularity of the shape features of the abnormal scan region, including:

[0087] Extract the edge contour of the abnormal scan area to obtain its geometry;

[0088] Calculate the matching degree between the geometric shape and the preset standard rule shape, which includes a rectangle, a circle, or a polygon;

[0089] The regularity of the shape features of the abnormal scanned region is determined based on the matching degree, wherein the regularity of the shape features is positively correlated with the matching degree.

[0090] Specifically, in this embodiment, calculating the matching degree between the geometric shape and the preset standard regular shape includes: extracting key geometric features from the edge contour of the abnormal region, such as center point, size, angle distribution, etc.; comparing the extracted features with the mathematical model of the preset standard shape (rectangle, circle, polygon) and calculating the difference between the two; quantifying the degree of difference by calculating the average distance from the contour point to the corresponding boundary of the standard shape, the smaller the distance, the higher the matching degree; normalizing the difference value to a matching degree percentage of 0-100%, where 100% represents a perfect match; calculating the matching degree for each preset shape template and selecting the highest value as the best matching result for that region.

[0091] Specifically, in this embodiment, determining the regularity of the shape features of the abnormal scanning region based on the matching degree includes: selecting the highest value from the matching degree calculation results of each preset standard shape (rectangle, circle, polygon) as the best matching degree for the region; applying a corresponding weight coefficient for correction according to the numerical range of the best matching degree, with a higher matching degree resulting in a larger weight coefficient (usually 0.5-1.0); multiplying the best matching degree by the weight coefficient to obtain a shape feature regularity value between 0 and 1, where the closer the value is to 1, the more regular the shape of the region; comparing the calculated regularity value with a preset threshold (e.g., 0.6), regions higher than the threshold are determined to be high-regularity regions, and regions lower than the threshold are determined to be low-regularity regions.

[0092] In this invention, the system no longer processes featureless patches formed by simply connecting anomalies. Instead, it extracts edge contours and abstracts them into geometrically meaningful shapes. This is a crucial step in transforming low-level image data into high-level semantic information. By calculating the matching degree with standard shapes (rectangles, circles, etc.), the system can quantify whether the shape of the region is regular or irregular using a continuous numerical value (matching degree). The higher the matching degree, the closer the region is to a certain regular shape. For the first time, the system introduces morphological features as the core judgment criterion in scanning decisions. This enables the system to identify, for example, the essential difference between an approximately rectangular region (which may correspond to a short circuit or open circuit in a metal connection) and a highly irregular star-shaped region (which may correspond to melt diffusion caused by a single point of penetration), laying the foundation for subsequent customized scanning.

[0093] Specifically, in this embodiment, the scanning path generation module is further used to calculate the region scanning priority based on the regularity of the shape features of the abnormal scanning region and the point anomaly degree of the abnormal scanning points within it, including:

[0094] Calculate the statistical value of the point anomaly degree of all abnormal scan points within the abnormal scan area, wherein the statistical value includes the average, maximum or weighted sum;

[0095] The region scanning priority of the abnormal scanning area is obtained by weighted fusion of the regularity of the shape feature and the statistical value of the point anomaly. The region scanning priority is positively correlated with both the statistical value of the point anomaly and the regularity of the shape.

[0096] Specifically, in this embodiment, the shape feature regularity and the point anomaly statistics are weighted and fused to obtain the region scanning priority of the abnormal scanning area. This includes: normalizing the shape feature regularity and the point anomaly statistics (average, maximum, or weighted sum) to the same dimension (e.g., 0-1); assigning weight coefficients to the two dimensions according to business requirements; typically, shape regularity reflects the typicality / root cause of the defect, and point anomaly reflects the electrical severity, both with positive correlation weights; linearly weighting the normalized shape regularity value and the point anomaly statistics according to their respective weights, using the formula: Priority = W1 × Regularity + W1 × Anomaly Statistics; where W1 + W2 = 1, and both are usually positive to ensure a double positive correlation; sorting the priority values ​​calculated for all abnormal regions in descending order, with higher values ​​representing higher scanning priority.

[0097] This invention enables the system to take a global perspective, comprehensively evaluate and sort all abnormal areas to be scanned a second time, and calculate a unique area scanning priority. Priority scheduling means that the system will prioritize the limited, high-precision secondary scanning resources (time, equipment wear and tear) to the areas with the highest expected benefits and the greatest risks. When the detection time is limited or the main fault needs to be quickly identified, the system can ensure that the deep scan of the most serious and typical defects is completed first, so as to locate the core problem as quickly as possible and avoid wasting time on secondary areas.

[0098] Specifically, the data analysis module is also used to calculate the point anomaly degree of abnormal scan points, including:

[0099] Obtain the current response data corresponding to each abnormal scan point in the first scan;

[0100] The current response data is compared with a preset normal response threshold range, and its deviation in one or multiple dimensions is calculated.

[0101] After normalizing the calculated deviation, it is used as the point anomaly degree of the abnormal scan point.

[0102] Among them, the point anomaly degree is positively correlated with the deviation degree.

[0103] Specifically, in this embodiment, a single electrical parameter (such as current amplitude) is compared with a preset normal threshold range, and the relative magnitude exceeding the threshold is calculated as the deviation. Multiple electrical parameters (such as amplitude, phase, and harmonic components) are combined, and the overall degree of deviation from the normal state is calculated by weighting or distance measurement (such as Mahalanobis distance). The calculated deviation is mapped to a standard range (such as 0-1) through a linear or nonlinear function to ensure that the anomalies under different chips and test conditions are comparable. The normalized value is the point anomaly degree of the scan point, and the higher the value, the more serious the electrical anomaly at that point.

[0104] This invention introduces a crucial continuous quantification dimension by calculating point anomaly scores. By calculating and normalizing the deviation from the normal threshold, each anomaly is assigned a comparable and standardized severity score (point anomaly score). The normalization process ensures that anomalies calculated under different test conditions, different chips, and even different response parameters (such as current and voltage phase) can be compared fairly on the same scale. This is a prerequisite for subsequent advanced operations such as region priority ranking and path density adjustment. It elevates the granularity of detection from the presence or absence of anomalies to the degree of anomaly, providing the system with a rich data layer far exceeding binary information. This is the fundamental reason why all subsequent intelligent decisions can achieve differentiation and refinement.

[0105] Specifically, the scan path generation module is used to determine the secondary scan path of the abnormal scan area, including:

[0106] If the regularity of the shape features of the abnormal scanning region is higher than a preset regularity threshold, then the basic scanning path is determined based on the geometric circumscribed shape of the abnormal scanning region.

[0107] If the regularity of the shape features of the abnormal scanning area is lower than or equal to the preset regularity threshold, then the abnormal scanning point with the highest abnormality within the abnormal scanning area is used as the starting focus to determine the radial basic scanning path.

[0108] Specifically, in this embodiment, the preset regularity threshold can be determined in the following ways: based on a large amount of chip test data with known failure modes, statistical analysis is performed on the distribution of the regularity of the shape of abnormal areas to find the boundary value that can effectively distinguish between systematic defects (usually regular) and random defects (usually irregular); combined with the failure analysis results, the regularity threshold is calibrated and matched with specific physical failure mechanisms (such as process problems often having regular shapes, while random events often have irregular shapes); on the verification dataset, with the optimization objectives of detection efficiency (such as scanning time) and classification accuracy, the threshold that optimizes the overall performance is determined by methods such as grid search.

[0109] In this invention, for rectangular areas, a parallel line path is used, with the scan line direction parallel to the edge, which can completely cover the area with minimal reciprocating motion. For circular or elliptical areas, a concentric circle or spiral path is used, which can naturally conform to its radial symmetry and avoid scanning blind spots or over-scanning at the corners. With the fewest scan points and the shortest path length, complete and uniform sampling of regular areas is obtained, which greatly improves the scanning speed and reduces the mechanical wear of the equipment. For low-regularity areas, the point with the highest point anomaly is used as the starting focus, and a radial path is used to first and concentrate the scanning resources on the most likely fault core, that is, the point with the highest anomaly, which is conducive to finding the root cause of the problem as quickly as possible. The radial rays can naturally extend in all directions until they reach the boundary of the area. This mode is particularly effective for exploring branching, star-shaped, and irregularly spreading defect areas, which can well delineate their complex contours. Using an exploratory, inside-out approach, the morphological characteristics and internal severity distribution of irregular areas are efficiently captured, avoiding the problems of a large number of invalid external scans or incomplete internal coverage that may occur when using regular paths.

[0110] Specifically, the scan path generation module is also used to dynamically adjust the density of scan points based on the point anomaly degree of each point within the path coverage area of ​​the determined basic scan path, wherein the density of scan points is positively correlated with the point anomaly degree of the coverage points.

[0111] Specifically, in this embodiment, the planned basic scanning path (such as parallel lines, concentric circles, or radial lines) is divided into several continuous path segments. For each path segment, the point anomaly degree of all abnormal scanning points within the coverage area of ​​that path segment is statistically analyzed, and its average (or maximum) value is calculated as the anomaly degree index of that segment. According to a preset mapping rule, the anomaly degree index is converted into target scanning point density (or point spacing). The mapping principle is: the higher the anomaly degree, the denser the scanning points. For example: anomaly degree > 0.8: high-density scanning (small point spacing); 0.5 < anomaly degree ≤ 0.8: medium-density scanning; anomaly degree ≤ 0.5: low-density scanning (large point spacing). Based on the determined target density, the final secondary scanning point sequence is generated at equal intervals on each path segment. Segments with high anomaly degree have small point spacing and many points; segments with low anomaly degree have large point spacing and few points.

[0112] Specifically, the scan path generation module is used to determine the basic scan path based on the geometric circumference shape of the abnormal scan region, including:

[0113] Identify the type of geometric circumference shape of the abnormal scan region;

[0114] If the circumscribed shape is rectangular, then a parallel line scan path is used;

[0115] If the external shape is circular or elliptical, then a concentric circle scan path is used;

[0116] If the circumscribed shape is a polygon, then a parallel scan line path is generated along its longest diagonal.

[0117] In this invention, a parallel line path is used for rectangular areas. The parallel line path is the theoretically optimal solution for covering rectangular areas. The scan line is parallel to the rectangle's edge, resulting in the shortest total path length, no turning redundancy, and achieving non-overlapping and non-omission coverage. For most scanning platforms driven by linear motors, parallel line motion is the fastest, most stable, and least lossy motion mode, completing the complete sampling of rectangular defects (commonly seen in wiring short circuits, misalignment, etc.) in the shortest time and with the lowest mechanical loss. For circular or elliptical areas, a concentric circle scanning path is used. Considering that the concentric circle or spiral path perfectly matches the radial symmetry of the circular area, the scanning points are evenly distributed along the radial direction and have a consistent density in the circumferential direction, which best conforms to its geometric nature. If a rectangular grid scan is used for a circular area, data oversampling (redundancy) will occur in the four corner areas, while insufficient sampling may occur at the circular boundary. The concentric circle path completely avoids this problem.

[0118] Specifically, the scan path generation module uses the abnormal scan point with the highest degree of abnormality within the abnormal scan area as the starting focus to determine a radial basic scan path, including:

[0119] Determine the coordinate position of the initial focus;

[0120] Multiple radial scanning rays are generated at preset angular intervals, centered on the initial focal point.

[0121] Based on the edge contour of the abnormal scanning area, the scanning rays that extend beyond the area are truncated to obtain the final basic scanning path.

[0122] Specifically, in this embodiment, the preset angle interval can be determined in the following ways: The minimum angle interval is determined based on the critical dimensions of the chip process and the required detection accuracy. For example, for advanced processes requiring fine detection, a smaller interval of 5°-10° may be used; for general detection, a larger interval of 15°-30° may be used; for irregularly shaped areas with complex shapes, a smaller angle interval (e.g., 10°-15°) is used to ensure that the rays can fully detect the contour details in all directions; for relatively compact areas, a larger angle interval (e.g., 20°-30°) can be used to improve efficiency while ensuring coverage; the higher the anomaly of the highest point within the area, the smaller the angle interval used to achieve dense detection of high-risk core areas; a mapping relationship can be established: anomaly > 0.8, interval 10°; anomaly 0.5-0.8, interval 15°; anomaly <0.5, interval 20°; set a maximum limit on the number of rays (e.g., no more than 72 rays, i.e., minimum interval 5°) to prevent too many scan points and too long scan time due to too small an interval; dynamically adjust based on the estimated total scan time budget; after the initial generation and truncation of the radiation path, analyze the distribution of the endpoints of each ray; if the distance between the endpoints of adjacent rays differs greatly (indicating drastic contour changes), automatically insert additional rays in the corresponding local area (reducing the local angle interval); the preset angle interval is usually selected between 10° and 30°.

[0123] In this invention, the system no longer treats irregular areas as homogeneous surfaces, but instead clearly identifies the points (with the highest anomaly) that are most likely to represent the root cause of the fault or the most dangerous location. Radial scanning starting from the focal point can quickly acquire detailed data of the area surrounding the core point, and can quickly confirm whether the most abnormal point is an isolated critical point or the core of a larger defect area. This provides the most direct information for quickly determining the nature of the fault, transforming a comprehensive survey into a focused breakthrough. When time is tight or the root cause needs to be quickly investigated, this strategy ensures that the system uses the shortest initial path to explore the most critical location, achieving a qualitative leap in diagnostic efficiency.

[0124] Specifically, the scan path generation module dynamically adjusts the density of scan points based on the anomaly level of each point within the path coverage area of ​​the determined basic scan path, including:

[0125] The basic scanning path is divided into several path segments;

[0126] Obtain the average point anomaly degree of all abnormal scan points within the coverage area of ​​each path segment;

[0127] The target scan point spacing of the path segment is determined based on the preset interval where the average point anomaly is located.

[0128] Based on the target scan point spacing, a final scan point sequence for secondary scanning is generated on the path segment.

[0129] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention can have various modifications and variations. 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 non-destructive testing system for a chip based on embedded current scanning, characterized in that, The application relates to a chip scan test system and method. The system comprises: a current scanning module for injecting current into a chip under test and controlling the current to scan the chip under test along a preset test path; an information acquisition module connected to the current scanning module and used for acquiring scan data of the chip under test; a data analysis module connected to the information acquisition module and used for determining abnormal scan points of the chip under test based on the scan data, calculating an aggregation characteristic value based on the spatial positions of the abnormal scan points, calculating a point abnormality degree based on the scan data, and determining a distribution type of the abnormal scan points based on the aggregation characteristic value, wherein the distribution type comprises an aggregation type and a dispersion type; an abnormal scan region generation module connected to the data analysis module and used for determining a merging method of the abnormal scan points based on the distribution type of the abnormal scan points, comprising: a first generation module for merging abnormal scan points with the aggregation type based on the aggregation kernel density of the abnormal scan points to generate an abnormal scan region; a second generation module for merging abnormal scan points with the dispersion type based on the dispersion connection radius of the abnormal scan points to generate an abnormal scan region; 2. The embedded current scanning based chip non-destructive inspection system of claim 1, wherein, a scan path generation module connected to the abnormal scan region generation module and the data analysis module and used for calculating a region scan priority of each region based on the shape characteristic regularity degree of the abnormal scan region and the point abnormality degree of the abnormal scan points in the region, and determining a secondary scan path for each abnormal scan region based on the region scan priority and the shape characteristic regularity degree. The data analysis module is used to calculate the aggregation characteristic value based on the local density and the global distribution deviation of the abnormal scan points, comprising: acquiring spatial position coordinates of all the abnormal scan points on the chip under test; calculating the local density of each abnormal scan point, wherein the local density is represented as the number of other abnormal scan points within a preset neighborhood radius centered on the point; calculating the global distribution center and the overall dispersion degree based on the spatial positions of all the abnormal scan points; calculating the global distribution deviation of each abnormal scan point according to the distance between the point and the global distribution center and the overall dispersion degree; 3. The embedded current scanning based chip non-destructive inspection system of claim 2, wherein, obtaining the aggregation characteristic value by weightedly fusing the local density and the global distribution deviation of the abnormal scan points. The data analysis module is used to determine the distribution type of the abnormal scan points based on the aggregation characteristic value, comprising: if the aggregation characteristic value is greater than or equal to a preset aggregation characteristic value, determining that the distribution type of the abnormal scan points is the aggregation type; 4. The embedded current scanning based chip non-destructive inspection system of claim 1, wherein, if the aggregation characteristic value is less than the preset aggregation characteristic value, determining that the distribution type of the abnormal scan points is the dispersion type. The scan path generation module is used to determine the shape characteristic regularity degree of the abnormal scan region, comprising: extracting the edge contour of the abnormal scan region to obtain the geometric shape thereof; calculating the matching degree of the geometric shape and a preset standard regular shape, wherein the preset standard regular shape comprises a rectangle, a circle or a polygon; determining the shape characteristic regularity degree of the abnormal scan region based on the matching degree, wherein the shape characteristic regularity degree is positively correlated with the matching degree.

5. The embedded current scanning based chip non-destructive inspection system of claim 4, wherein, The data analysis module is also used to calculate the point anomaly degree of the abnormal scanning point, comprising: Obtaining the current response data corresponding to each abnormal scanning point in the first scanning; Comparing the current response data with the preset normal response threshold range, and calculating the single dimension or multi dimension deviation degree; After normalizing the calculated deviation degree, the point anomaly degree of the abnormal scanning point is obtained. The point anomaly degree is positively correlated with the deviation degree.

6. The embedded current scanning based chip non-destructive inspection system of claim 5, wherein, The scanning path generation module is used to determine the secondary scanning path of the abnormal scanning area, comprising: If the shape feature regularity degree of the abnormal scanning area is higher than the preset regularity degree threshold, the basic scanning path is determined according to the geometric circumscribed shape of the abnormal scanning area; If the shape feature regularity degree of the abnormal scanning area is lower than or equal to the preset regularity degree threshold, the basic scanning path of the abnormal scanning area is determined as the starting focus point, and the basic scanning path of the abnormal scanning area is determined as the starting focus point.

7. The embedded current scanning based chip non-destructive inspection system of claim 6, wherein, The scanning path generation module is also used to dynamically adjust the density of scanning points based on the point anomaly degree of each point in the path coverage range of the determined basic scanning path, wherein the density of scanning points is positively correlated with the point anomaly degree of the coverage point.

8. The embedded current scanning based chip non-destructive inspection system of claim 6, wherein, The scanning path generation module is used to determine the basic scanning path according to the geometric circumscribed shape of the abnormal scanning area, comprising: Identifying the type of the geometric circumscribed shape of the abnormal scanning area; If the circumscribed shape is rectangular, the parallel line scanning path is adopted; If the circumscribed shape is circular or elliptical, the concentric circle scanning path is adopted; If the circumscribed shape is polygon, the parallel scanning line path is generated along the longest diagonal direction.

9. The embedded current scanning based chip non-destructive inspection system of claim 6, wherein, The scanning path generation module determines the basic scanning path of the abnormal scanning area as the starting focus point, comprising: Determining the coordinate position of the starting focus point; Generating a plurality of radial scanning rays at a preset angle interval with the starting focus point as the center; According to the edge profile of the abnormal scanning area, the scanning rays beyond the area are truncated to obtain the final basic scanning path.

10. The embedded current scanning based chip non-destructive inspection system of claim 7, wherein, The scanning path generation module dynamically adjusts the density of scanning points based on the abnormal degree of each point in the path coverage range of the determined basic scanning path, comprising: Dividing the basic scanning path into several path segments; Obtaining the average point anomaly degree of all abnormal scanning points in each path segment coverage range; According to the preset interval of the average point anomaly degree, the target scanning point interval of the path segment is determined; According to the target scanning point interval, the final scanning point sequence for secondary scanning is generated on the path segment.

Images