Memory chip failed cell location method based on fault group pattern matching

Abstract

The application relates to the chip repair technical field, and particularly discloses a memory chip failure unit positioning method based on fault group mode matching, which comprises the following steps: obtaining a failure bitmap of a memory chip to be repaired; iterating the failure bitmap, aggregating adjacent failure units in spatial positions, and forming at least one fault group; extracting geometric morphological features and distribution statistical features of the fault group, combining the geometric morphological features and the distribution statistical features to generate a feature vector of the fault group; matching the feature vector with historical feature vectors in a repair scheme feature library generated in advance, selecting a recommended repair scheme from the repair scheme feature library according to a matching result; and integrating recommended repair schemes of all the fault groups to form a global repair scheme for the memory chip to be repaired. The application can improve the overall efficiency and yield of the memory chip manufacturing and testing links under the premise of ensuring repair reliability.

Description

Technical Field

[0001] This invention relates to the field of chip repair technology, and more specifically to a method for locating faulty cells in memory chips based on fault group pattern matching. Background Technology

[0002] In the manufacturing and testing of dynamic random access memory (DRAM) chips, as semiconductor process nodes continue to shrink, the integration and storage density of memory arrays continue to increase, and the number of memory cells contained in a single chip grows exponentially. Under this trend, affected by factors such as lithography deviations, material defects, and process fluctuations, failed cells inevitably occur in the memory array, and the number and distribution patterns of failed cells are becoming increasingly complex.

[0003] After completing chip testing, automated test equipment typically outputs a large amount of location information for failed cells in the form of a failure bitmap. These failed cells may exhibit localized clustering or dispersed distribution in space, resulting in significant uncertainty and diversity in the overall failure mode. In existing memory repair analysis workflows, faced with massive and complexly distributed data on failed cells, it is difficult to systematically identify and summarize the spatial distribution characteristics of these failed cells in a timely and effective manner, thereby accurately locating the potential relationships between different failed cells.

[0004] Furthermore, the failure distribution patterns vary significantly among different chips, and there is a lack of an effective way to uniformly characterize and distinguish the distribution characteristics of failed cells. This leads to the repair analysis process relying heavily on computation or manual experience, and the repair analysis time increases significantly with the scale of the failure. Therefore, how to effectively describe and locate the distribution pattern of failed cells in memory chips, while ensuring the accuracy of failed cell location, in order to reduce the complexity and time cost of repair analysis, has become an urgent technical problem to be solved in the manufacturing and testing of memory chips. Summary of the Invention

[0005] The purpose of this invention is to provide a method for locating memory chip failure cells based on fault group pattern matching, thereby solving the above-mentioned technical problems.

[0006] The objective of this invention can be achieved through the following technical solutions: The memory chip failure cell location method based on fault group pattern matching includes the following steps: Obtain a failure bitmap of the memory chip to be repaired, wherein the failure bitmap records the physical coordinates of the failed cell in the memory array; Traverse the failure bitmap and aggregate adjacent failure units in spatial location to form at least one failure group; Extract the geometric morphological features and distribution statistical features of the fault group, and combine the geometric morphological features and distribution statistical features to generate the feature vector of the fault group; The feature vector is matched with historical feature vectors in a pre-generated repair scheme feature library, and a suggested repair scheme is selected from the repair scheme feature library based on the matching result. The suggested repair solutions for all fault groups are integrated to form a global repair solution for the memory chip to be repaired.

[0007] As a further aspect of the present invention: generating a fault group includes: Starting from any one of the failed units, search for all other failed units within its preset neighborhood. If other failed units are found within the preset neighborhood, the starting point and the other failed units are merged into the same set. Using each failed unit in the set as a new starting point, the search and merging process is repeated until there are no failed units not included in the set within the preset neighborhood of any failed unit in the set. At this point, the final set is marked as a fault group. Traverse all failed units in the failure bitmap that are not assigned to any failure group, and repeat the above process until all failed units have been processed.

[0008] As a further aspect of the present invention: extracting geometric morphological features includes: Obtain the coordinate set of all failed units that constitute the fault group; Calculate the length, width, and aspect ratio of the smallest bounding rectangle of the coordinate set; Calculate the centroid coordinates of the coordinate set, and calculate the average and maximum distances from all failed unit coordinates to the centroid; The length, width, and aspect ratio of the minimum bounding rectangle, along with the average and maximum distances, are used together as the geometric features of the fault group.

[0009] As a further aspect of the present invention: extracting distribution statistical features includes: Count the total number of failed units contained in the fault group; The smallest bounding rectangle region where the fault group is located is divided into multiple equally spaced intervals in both the row and column directions. The number of failure units distributed in each interval along the row direction is counted, and the variance of their distribution is calculated and denoted as the row distribution density variance. The distribution of failure units in each interval along the column direction is statistically analyzed, and the variance of their distribution is calculated and denoted as the column distribution density variance. The total number of failed units, the variance of the row distribution density, and the variance of the column distribution density are used together as the distribution statistical characteristics of the fault group.

[0010] As a further aspect of the present invention: the recommended repair solution includes: The repair scheme feature library stores the feature vectors of historical fault groups and their corresponding replacement schemes containing redundant rows and columns. Calculate the Euclidean distance between the feature vector of the current fault group and each feature vector in the feature library of the repair plan; Sort all the calculated Euclidean distances in ascending order of value, select the Euclidean distance A that is ranked first, and compare it with the preset distance threshold. If the Euclidean distance A ≤ the distance threshold, then the matching is determined to be successful, and the replacement scheme corresponding to the historical feature vector is obtained as a suggested repair scheme; If the Euclidean distance A is greater than the distance threshold, then the matching is deemed to have failed.

[0011] As a further aspect of the present invention, the recommended repair solution also includes: If the matching fails, the feature vectors in the feature library of the repair schemes corresponding to the first K Euclidean distances in the sorting are recorded as candidate vectors, and the replacement schemes corresponding to the candidate vectors are taken as candidate schemes. Determine whether the redundant rows and columns contained in the candidate scheme can cover all the failed units in the fault group. If so, the candidate scheme is selected as a candidate scheme. The candidate solution that ranks first among the corresponding candidate vectors is taken as the suggested repair solution for the fault group.

[0012] The beneficial effects of this invention compared to the prior art are as follows: This invention addresses the challenges of numerous and complexly distributed failure cells in memory chips, which limit the efficiency of repair analysis. By systematically characterizing and locating the spatial distribution characteristics of failure cells, it achieves effective identification and summarization of failure cell distribution patterns, helping to maintain the stability and controllability of the repair analysis process even with large-scale failure data. By analyzing the clustering relationships formed by failure cells in the memory array, this invention transforms the originally discrete and disordered failure cell information into failure distribution cells with clear spatial characteristics, thereby reducing the reliance on point-by-point processing during failure cell location. Based on a comprehensive characterization of failure distribution patterns and statistical features, this invention enhances the ability to identify differences between different failure distribution patterns, making the repair analysis process more targeted and consistent. Furthermore, by introducing a matching mechanism based on historical features, this invention can fully utilize existing repair experience, improving the accuracy of failure distribution identification and location, and reducing the uncertainty caused by differences in failure modes. Based on this, this invention helps to shorten the overall process time of memory chip repair analysis and reduce the complexity of the repair analysis process, thereby improving the overall efficiency and yield of memory chip manufacturing and testing while ensuring repair reliability, demonstrating significant engineering application value and practical significance. Attached Figure Description

[0013] The invention will now be further described with reference to the accompanying drawings.

[0014] Figure 1 This is a flowchart illustrating the memory chip failure cell location method based on fault group pattern matching of the present invention. Figure 2 This is a flowchart illustrating the process of selecting a suggested repair solution according to the present invention; Figure 3 This is a schematic diagram illustrating how the fault unit is divided into four independent fault groups according to the present invention. Figure 4 This is a schematic diagram illustrating all possibilities of the five characteristic information of the fault group when the number of fault units in the fault group is 1-5. Figure 5 This invention is for Figure 1 A schematic diagram illustrating the generation of fault group pattern diagrams for the three multi-fault groups is provided. Figure 6 This is a schematic diagram of a repair case according to the present invention; Figure 7 This is a schematic diagram of a repair scheme with the same fault group characteristic information of the present invention. Detailed Implementation

[0015] 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 skilled in the art without creative effort are within the scope of protection of the present invention.

[0016] Please see Figures 1-2 As shown, this invention is a method for locating faulty cells in a memory chip based on fault group pattern matching, comprising the following steps: Obtain a failure bitmap of the memory chip to be repaired, wherein the failure bitmap records the physical coordinates of the failed cell in the memory array; In one optional implementation, the coordinates of the failed cell recorded in the failure bitmap are physical coordinates after address descrambling. The physical coordinates include at least identification information for uniquely determining the storage array level where the failed cell is located. The level identification information includes, but is not limited to, the memory bank, memory block, subarray, or equivalent physical partition identifier in the memory chip.

[0017] Before generating fault groups, the logical addresses obtained from the test are mapped to their corresponding physical coordinates to ensure that the adjacency judgment based on spatial location conforms to the actual physical layout of the storage array.

[0018] Traverse the failure bitmap and aggregate adjacent failure units in spatial location to form at least one failure group; In a preferred embodiment of the present invention, generating a fault group includes: The failure bitmap records the physical coordinate information of each failed cell in the memory array. The physical coordinate information has undergone the necessary address mapping process before entering the failure group generation process, so that the proximity relationship of coordinates within the same physical partition can reflect the actual spatial adjacency relationship of the memory array. At the same time, the physical coordinate information contains partition identifiers used to define physical boundaries, so that the subsequent merging process will not cross memory banks, subarrays or equivalent physical partitions and produce unrealizable aggregation results.

[0019] After the failure bitmap is read, a set of unprocessed failure units is established to mark failure units that have not yet been assigned to any fault group. Then, a failure unit is taken from this set as the starting point. The input data corresponding to the starting point includes its physical coordinates and the physical partition identifier to which it belongs.

[0020] When conducting a neighborhood search around the starting point, the preset neighborhood range is determined based on the proximity relationship in the physical coordinate space, and physical boundary constraints are applied during the determination process to ensure that the search is only carried out within the same physical partition to which the starting point belongs, thereby avoiding the misclassification of failure units located in different physical partitions into the same fault group.

[0021] The neighborhood search process retrieves the physical coordinates of other failed units in the failure bitmap within a preset neighborhood, centered on the starting point. The search is limited to failed units in the currently unprocessed set. The output is a candidate failed unit list containing all failed units that meet the neighborhood condition and are located in the same physical partition as the starting point. When the candidate failed unit list is not empty, the starting point and all failed units in the candidate list are merged into a single set. This set is maintained as a temporary set for the current fault group. Elements in this set use the physical coordinates and partition identifier of the failed unit as basic record items. Simultaneously, failed units merged into the temporary set are removed from the unprocessed set to ensure that the same failed unit is not processed repeatedly during subsequent traversals.

[0022] After the temporary set is formed, each failed unit in the set is sequentially taken as a new starting point and the neighborhood search and merging process is repeated. The input of the new starting point also includes its physical coordinates and partition identifier. The neighborhood search is still carried out under the same preset neighborhood range and the same physical boundary constraints. The search range is limited to the same physical partition to which the new starting point belongs, and the search objects are limited to failed units that have not yet been merged into the temporary set, so as to gradually absorb failed units that have spatial adjacency with the temporary set. The list of candidate failed units obtained from each search is merged into the temporary set and is not removed from the unprocessed set. The temporary set is continuously expanded in this process until any failed unit in the temporary set can not find any failed unit still in the unprocessed set within the preset neighborhood range when performing a neighborhood search as a starting point. At this time, the temporary set satisfies the closure condition, and the output is a final set and is marked as a fault group. The output of the fault group includes the set of physical coordinates of all failed units in the fault group and the common physical partition belonging information, so that subsequent feature extraction and repair scheme matching based on the fault group can be carried out under the premise of consistent physical boundaries.

[0023] After marking a fault group, the fault units still in the unprocessed set in the fault bitmap are traversed again, and new fault units are selected as new starting points. The above-mentioned neighborhood search, merging expansion and closure judgment process is repeated until the unprocessed set is empty. At this time, all fault units in the fault bitmap have been assigned to a fault group or form a fault group in the form of a separate set. The output is a set of at least one fault group. This set covers all fault units in the fault bitmap and maintains the consistency of physical partition boundaries, providing structured input for the subsequent generation and matching of fault group feature vectors.

[0024] Extract the geometric morphological features and distribution statistical features of the fault group, and combine the geometric morphological features and distribution statistical features to generate the feature vector of the fault group; In a preferred embodiment of the present invention, the extraction of geometric morphological features includes: Obtain the coordinate set of all failed units that constitute the fault group; Calculate the length, width, and aspect ratio of the smallest bounding rectangle of the coordinate set; Calculate the centroid coordinates of the coordinate set, and calculate the average and maximum distances from all failed unit coordinates to the centroid; The length, width, and aspect ratio of the minimum bounding rectangle, along with the average and maximum distances, are used together as the geometric features of the fault group.

[0025] In another preferred embodiment of the present invention, the extraction of distribution statistical features includes: Count the total number of failed units contained in the fault group; The smallest bounding rectangle region where the fault group is located is divided into multiple equally spaced intervals in both the row and column directions. The number of failure units distributed in each interval along the row direction is counted, and the variance of their distribution is calculated and denoted as the row distribution density variance. The distribution of failure units in each interval along the column direction is statistically analyzed, and the variance of their distribution is calculated and denoted as the column distribution density variance. The total number of failed units, the variance of the row distribution density, and the variance of the column distribution density are used together as the distribution statistical characteristics of the fault group.

[0026] Specifically, the fault group contains the physical coordinate information of all failed units in the memory array, and this physical coordinate information has met the physical boundary consistency before entering this step, so that the coordinates in the same fault group are all within the same physical partition of the memory array, thereby ensuring that the morphological description obtained later based on the coordinate space relationship can correspond to the same array area.

[0027] After all the failed units constituting the fault group are read, the coordinate set of the failed units is established. This coordinate set records the position of each failed unit in the fault group in the form of a pair of row coordinates and column coordinates. The coordinate set is used as the common input for extracting geometric morphological features and distribution statistical features.

[0028] The processing of geometric features starts from this coordinate set. The range of values ​​in the set is retrieved in the row coordinate dimension and the column coordinate dimension respectively to determine the smallest bounding rectangle region that can cover all coordinate points. The boundary of this rectangle is jointly defined by the minimum and maximum row coordinates in the row direction and the minimum and maximum column coordinates in the column direction. The length and width of the rectangle are obtained according to the row span and column span respectively. At the same time, the aspect ratio is obtained from the length and width as an intermediate result output representing the overall extension shape of the fault group.

[0029] When further describing the centroid-related morphology around the coordinate set, the row and column coordinates in the coordinate set are used to determine the centroid position of the fault group in the coordinate space. The centroid coordinates are output and temporarily stored as reference points for subsequent distance statistics. Then, the distance between the coordinates of each failure unit in the coordinate set and the centroid coordinates is calculated to obtain a set of distance values ​​from each failure unit to the centroid. Based on this set of distance values, the average distance and the maximum distance are obtained. The average distance is used to reflect the overall dispersion of the failure unit around the centroid, and the maximum distance is used to reflect the farthest deviation of the failure unit from the centroid.

[0030] The output of the geometric morphology features is given in the form of a set of defined feature quantities, including the length, width, aspect ratio of the minimum bounding rectangle, and the average and maximum distances. This output is recorded as the geometric morphology features of the fault group.

[0031] The processing of distribution statistics features shares the same set of input coordinates as geometric morphology features, and the minimum bounding rectangle region obtained is used as the limited area for distribution statistics. The total number of failed units contained in the fault group is obtained by counting the number of elements in the coordinate set, and this total number is output as a statistic reflecting the fault scale.

[0032] To characterize the distribution variation of failed units along the row and column directions within the minimum bounding rectangle region, the rectangular region is divided into multiple equally spaced intervals along both the row and column directions. The interval division is based on the coordinate span of the rectangle in the corresponding direction, with the interval boundaries determined by equal spacing within that span. This ensures that each interval corresponds to a continuous range of row or column coordinates, and the set of intervals serves as the input constraint for subsequent counting statistics. During distribution statistics along the row direction, the coordinate set is iterated sequentially, mapping the row coordinates of each failed unit to its corresponding row direction interval. The number of failed units falling within each row direction interval is cumulatively counted, forming a row direction interval counting sequence. This sequence serves as an intermediate output to further obtain the variance of the row direction distribution quantity. The variance of the row direction distribution quantity is denoted as the row distribution density variance and output as part of the distribution statistical characteristics.

[0033] When performing distribution statistics along the column direction, the same processing logic as the row direction is adopted. The column coordinates of each failure unit are mapped to its corresponding column direction interval. The number of failure units falling into each column direction interval is accumulated and counted to form the interval counting sequence of the column direction. This sequence is used to obtain the variance of the number of distributions in the column direction. The variance of the number of distributions in the column direction is recorded as the column distribution density variance and output as part of the distribution statistical characteristics.

[0034] The final output of the distribution statistics features is given as a set of defined feature quantities, including the total number of failed units, the variance of row distribution density, and the variance of column distribution density. This output, along with the geometric morphology features, enters the feature vector generation process. During feature vector generation, the geometric morphology features and distribution statistics features are combined according to a predetermined feature arrangement order to form a feature vector that can characterize the spatial morphology and distribution differences of the fault group. The feature vector is stored as the output of this step and associated with the fault group identifier, providing a unified input format for subsequent matching with historical feature vectors in the repair scheme feature library.

[0035] The feature vector is matched with historical feature vectors in a pre-generated repair scheme feature library, and a suggested repair scheme is selected from the repair scheme feature library based on the matching result. In a preferred embodiment of the present invention, the recommended repair solution includes: The repair scheme feature library stores the feature vectors of historical fault groups and their corresponding replacement schemes containing redundant rows and columns. Calculate the Euclidean distance between the feature vector of the current fault group and each feature vector in the feature library of the repair plan; Sort all the calculated Euclidean distances in ascending order of value, select the Euclidean distance A that is ranked first, and compare it with the preset distance threshold. If the Euclidean distance A ≤ the distance threshold, then the matching is determined to be successful, and the replacement scheme corresponding to the historical feature vector is obtained as a suggested repair scheme; If the Euclidean distance A is greater than the distance threshold, then the matching is deemed to have failed.

[0036] In a preferred embodiment, the recommended repair solution further includes: If the matching fails, the feature vectors in the feature library of the repair schemes corresponding to the first K Euclidean distances in the sorting are recorded as candidate vectors, and the replacement schemes corresponding to the candidate vectors are taken as candidate schemes. Determine whether the redundant rows and columns contained in the candidate scheme can cover all the failed units in the fault group. If so, the candidate scheme is selected as a candidate scheme. The candidate solution that ranks first among the corresponding candidate vectors is taken as the suggested repair solution for the fault group.

[0037] Specifically, the feature vector is obtained by combining the geometric morphological features and distribution statistics of the fault group in a predetermined order. Each feature component in the feature vector corresponds one-to-one with the spatial morphology or distribution statistics of the fault group, thereby ensuring comparability with historical data. At the same time, a pre-generated repair scheme feature library is used as the matching object. The repair scheme feature library is a structured data set that is built offline and can be called in the production or analysis stage. Each record in the repair scheme feature library contains at least the historical feature vector of the historical fault group and the replacement scheme associated with the historical feature vector. The replacement scheme is the configuration data used to describe the redundant repair action. Its content includes the set of redundant row identifiers and the set of redundant column identifiers that need to be enabled. The redundant row identifier is used to indicate the redundant row resources that will be used for replacement, and the redundant column identifier is used to indicate the redundant column resources that will be used for replacement. The replacement scheme and the corresponding historical fault group have an implementable association within the same storage array physical partition.

[0038] The establishment of the repair scheme feature library uses historical memory chip test and repair analysis data as the data source. During the memory chip production or engineering verification stage, data is collected from chips that have completed testing and successfully implemented redundant repairs. The collected data includes at least the corresponding chip's failure bitmap and the finally confirmed feasible redundant repair configuration. For each historical chip, based on the physical coordinate information of the failed units recorded in its failure bitmap, the failed units are classified to form several fault groups. Each fault group corresponds to a set of failed units with consistent spatial distribution characteristics. For each historical fault group, feature information that can characterize its spatial distribution pattern and statistical characteristics is extracted, and the obtained feature information is combined to form a historical feature vector. The construction method of the historical feature vector is consistent with the construction method of the fault group feature vector in the subsequent chips to be repaired to ensure the consistency of feature dimensions and feature meanings. At the same time, for the historical fault group, the redundant repair configurations adopted and verified to be feasible in the actual repair process are recorded. The redundant repair configuration includes the enabled redundant row identifier set and redundant column identifier set. This redundant repair configuration can cover all failed units in the corresponding historical fault group and meet the redundancy resource constraints of the chip at that time. By establishing a one-to-one correspondence between the historical feature vectors of each historical fault group and its corresponding redundant repair configuration, multiple repair records are formed and stored in a structured manner. All repair records together constitute a repair scheme feature library. After its establishment, the repair scheme feature library can be saved as an independent data resource and called upon during subsequent memory chip repair analysis to provide repair scheme references based on historical experience for new fault groups.

[0039] At the start of the matching process, the current feature vector is compared with each historical feature vector in the repair scheme feature library. Euclidean distance is used as the similarity measure. The input to the Euclidean distance is two vectors that are identical in dimension and feature order, and the output is a non-negative distance value. This distance value characterizes the degree of difference between the current fault group and the historical fault group in the feature space. Since different feature components may have different dimensions or value ranges, the corresponding feature components in the current and historical feature vectors are normalized or weighted before the distance calculation. This ensures that the influence of different feature components on the distance result is within a controllable range. The normalization or weighting parameters are obtained statistically from historical data during the repair scheme feature library establishment phase and are stored together with the feature library. During matching, the same parameters are used to perform consistent processing on the current feature vector, ensuring the comparability of the distance calculation.

[0040] After calculating the distances to all historical feature vectors in the feature library, a distance list is obtained. Each item in the distance list contains at least a distance value and the corresponding historical feature vector identifier and replacement scheme index. The distance list is sorted in ascending order of distance values. The first distance value in the sorting result is denoted as Euclidean distance A, and the historical feature vector corresponding to Euclidean distance A is denoted as the closest historical vector.

[0041] The Euclidean distance A is compared with a preset distance threshold. The distance threshold is a parameter used to determine whether the similarity is sufficient. The distance threshold is preset and stored during the feature library establishment or process import stage. The input is the Euclidean distance A, and the output is the matching judgment result. When the Euclidean distance A is less than or equal to the distance threshold, the matching is considered successful. The output of a successful match is a suggested repair solution. The suggested repair solution directly adopts the replacement solution in the record corresponding to the Euclidean distance A and establishes an association with the current fault group identifier for subsequent global repair solution integration.

[0042] When the Euclidean distance A is greater than the distance threshold, the matching is deemed to have failed. The output of the matching failure is the trigger condition for entering the candidate filtering process. The top K historical feature vectors with the smallest distance values ​​in the sorting results are selected as the candidate vector set. K is a preset positive integer and is fixed in the implementation. Each candidate vector in the candidate vector set retains its corresponding replacement scheme as a candidate scheme and forms a candidate scheme list.

[0043] The list of candidate solutions enters the coverage judgment process. The coverage judgment takes the physical coordinate set of all failed units in the current fault group and the redundant row identifier set and redundant column identifier set contained in a candidate solution as input. The row replacement range corresponding to the redundant row identifier set and the column replacement range corresponding to the redundant column identifier set are used to determine the coordinate set that the candidate solution can cover. The coverage judgment is based on whether the coordinates of each failed unit in the fault group coordinate set fall within the row replacement range or column replacement range. When all failed units in the fault group meet the coverage conditions, the judgment result of the candidate solution's successful coverage is output, and the candidate solution is marked as a candidate solution.

[0044] During the coverage assessment process, the physical boundary consistency between the candidate solution and the current fault group is checked simultaneously. The redundant row and column resources involved in the candidate solution are required to be within the same storage array physical partition range as the current fault group, so as to ensure that the replacement solution is physically feasible. Candidate solutions that do not meet the physical boundary consistency requirement are not included in the candidate solution set.

[0045] After completing the coverage judgment of the candidate solution list, a candidate solution set is obtained. Each item in the candidate solution set corresponds to a candidate vector in the candidate vector set and can cover all the failed units of the current fault group. The candidate solution with the highest position in the original distance sorting is selected from the candidate solution set as the suggested repair solution for the current fault group and output. This allows the replacement solution that is closest to the features of the current fault group and has coverage capability to be selected based on the similarity sorting result even if the matching fails. The suggested repair solution is associated with the current fault group identifier, stored and passed to the subsequent global repair solution integration step.

[0046] If no candidate solution is found when a match fails, an error will be reported.

[0047] The suggested repair solutions for all fault groups are integrated to form a global repair solution for the memory chip to be repaired.

[0048] In a preferred embodiment of the present invention, the repair solution may further be: Step 1: Apply test vectors to the memory array using automated testing equipment, and compare the output response with the expected value to locate the failed cell and obtain the location information of the failed cell in the memory chip.

[0049] By applying test vectors to the memory array using automated testing equipment and comparing the output response with the expected value, the location of failed cells is determined, thus obtaining the location information of all failed cells in the memory chip. This location information is represented by a two-dimensional grid coordinate graph, where each failed memory cell has corresponding x and y coordinates, i.e., X address and Y address.

[0050] Step 2: Based on the location information characteristics of different failed units in the memory chip and the basic principle of fault grouping, all failed units are grouped.

[0051] The grouping principles are as follows: In a memory chip, faulty cells with the same row address or column address are grouped together.

[0052] Based on the location information characteristics of different faulty cells in the memory chip, and according to the basic principle of fault grouping, all faulty cells are divided into several single-fault groups and multiple-fault groups. The grouping is based on faulty cells having the same row address or column address; that is, grouping is based on the correlation of coordinate information between faulty cells, resulting in multiple independent fault groups. A single-fault group refers to a fault group with only one faulty cell, while a multiple-fault group refers to a fault group with multiple faulty cells. For example... Figure 1 The entire fault unit is divided into 4 independent fault groups.

[0053] Step 3: Based on the grouping results of the memory failure units in Step 2, extract five feature information for each failure group to form a failure group pattern, and provide a redundancy repair scheme for each failure group.

[0054] The five characteristic information of each fault group are defined as follows: Number of faulty units Nf, number of faulty rows Nfr, number of faulty columns Nfc, number of rows with multiple faults Nmr, number of columns with multiple faults Nmc The redundant repair scheme for each fault group is not unique; that is, a fault group may have multiple redundant repair schemes.

[0055] Based on the grouping results of memory faulty cells in step 2, five feature pieces of information are extracted for each fault group. These five feature pieces of information include: number of faulty cells Nf, number of faulty rows Nfr, number of faulty columns Nfc, number of multiple faulty rows Nmr, and number of multiple faulty columns Nmc. The number of faulty cells, as the name suggests, is the total number of faulty cells in a fault group; the number of faulty rows and columns refers to the number of rows and columns occupied by faults in a fault group; and the number of multiple faulty rows and columns refers to the number of faulty rows or columns in a fault group where the number of faults is greater than 1. Figure 2 The document lists all the possibilities for the five characteristic information of a fault group when the number of fault units in the fault group is 1-5.

[0056] Step 4: Based on the five characteristic information of the fault group and the redundancy repair method, form a list of solutions for various fault types and store it in the corresponding memory chip failure unit solution search library.

[0057] Create a fault group pattern diagram for each group and find redundant repair solutions for each group. Figure 3 right Figure 1 The invention generates fault group pattern diagrams for three multi-fault groups. Since each fault group in the same memory chip shares the same redundant resources, but the repair methods of each group are independent of each other, there may be inappropriate utilization of redundant resources when searching for redundant repair schemes. Therefore, this invention specifically provides three methods for searching for redundant repair schemes, and each group can make reasonable selections based on the usage of redundant resources to avoid the phenomenon of mutual exclusion of redundant resources.

[0058] The five characteristic information of each fault group collected in step 3 and the corresponding redundancy repair scheme generated in step 4 are combined to form a solution list, which is then placed into the memory chip failure cell solution search library. This involves collecting the fault group types appearing in a specific memory chip and the corresponding solutions.

[0059] Step 5: In subsequent tests, compare the fault groups of the test chip with the fault groups in the solution search scheme to see if they have the same five characteristic information. If they are the same, it means that a repair scheme for that fault group already exists, and it can be directly called from the solution search library. A further technical solution is: if a fault group cannot find a solution in the memory chip failure cell solution search library, that is, the characteristic information is different, then a solution is searched again, and the five information characteristics and the solution are stored in the solution search library.

[0060] Step 6 provides three methods for finding a set of redundancy repair solutions, and the specific method to choose is based on the limitations of redundant resources.

[0061] Based on the continuous accumulation of various fault group types and their solutions in the memory chip failure unit solution search library, when the test machine detects a new memory chip, it first compares the five characteristic information of the fault group in the chip with the five characteristic information of the existing fault group in the solution search library. If a fault group with the same characteristic information as the fault group in the chip exists in the solution search library, it means that a repair solution for that fault group already exists in the test chip, and it can be directly called from the solution search library. Figure 5 As shown, the three listed faults have the same fault group characteristic information, so their repair solutions are the same and can be directly searched in the search library.

[0062] For example, the three redundancy repair schemes for the fault group are as follows: (1) Repairing all redundant rows (2) Repairing all redundant columns (3) Prioritize repairing rows and columns with more fault points until all faulty units are repaired.

[0063] Step 7: According to step 6, if the feature information of the fault group in the test chip fails to match the feature information in the solution search library, then step 4 is executed to find the redundant repair scheme for the group and store it in the memory failure unit solution search library.

[0064] Step 8. In summary, when searching for a redundancy repair solution for a memory chip, solutions should be prioritized from the memory failure cell solution search library. When determining the final memory chip redundancy repair solution, the balance between the memory chip's redundancy resources and the redundancy repair solutions for different fault groups should be comprehensively considered. Each fault group can choose any one of the three redundancy repair solutions, but it must be ensured that the final repair result does not exceed the redundancy resource limit.

[0065] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the present invention should still fall within the scope of the present invention.

Claims

1. A method for locating faulty cells in a memory chip based on fault group pattern matching, characterized in that, Includes the following steps: Obtain a failure bitmap of the memory chip to be repaired, wherein the failure bitmap records the physical coordinates of the failed cell in the memory array; Traverse the failure bitmap and aggregate adjacent failure units in spatial location to form at least one failure group; Extract the geometric morphological features and distribution statistical features of the fault group, and combine the geometric morphological features and distribution statistical features to generate the feature vector of the fault group; The feature vector is matched with historical feature vectors in a pre-generated repair scheme feature library, and a suggested repair scheme is selected from the repair scheme feature library based on the matching result. The suggested repair solutions for all fault groups are integrated to form a global repair solution for the memory chip to be repaired.

2. The memory chip failure cell location method based on fault group pattern matching according to claim 1, characterized in that, The generated fault groups include: Starting from any one of the failed units, search for all other failed units within its preset neighborhood. If other failed units are found within the preset neighborhood, the starting point and the other failed units are merged into the same set. Using each failed unit in the set as a new starting point, the search and merging process is repeated until there are no failed units not included in the set within the preset neighborhood of any failed unit in the set. At this point, the final set is marked as a fault group. Traverse all failed units in the failure bitmap that are not assigned to any failure group, and repeat the above process until all failed units have been processed.

3. The memory chip failure cell location method based on fault group pattern matching according to claim 1, characterized in that, Extracting geometric morphological features includes: Obtain the coordinate set of all failed units that constitute the fault group; Calculate the length, width, and aspect ratio of the smallest bounding rectangle of the coordinate set; Calculate the centroid coordinates of the coordinate set, and calculate the average and maximum distances from all failed unit coordinates to the centroid; The length, width, and aspect ratio of the minimum bounding rectangle, along with the average and maximum distances, are used together as the geometric features of the fault group.

4. The memory chip failure cell location method based on fault group pattern matching according to claim 1, characterized in that, Extracting distributional statistical features includes: Count the total number of failed units contained in the fault group; The smallest bounding rectangle region where the fault group is located is divided into multiple equally spaced intervals in both the row and column directions. The number of failure units distributed in each interval along the row direction is counted, and the variance of their distribution is calculated and denoted as the row distribution density variance. The distribution of failure units in each interval along the column direction is statistically analyzed, and the variance of their distribution is calculated and denoted as the column distribution density variance. The total number of failed units, the variance of the row distribution density, and the variance of the column distribution density are used together as the distribution statistical characteristics of the fault group.

5. The memory chip failure cell location method based on fault group pattern matching according to claim 1, characterized in that, The recommended repair options include: The repair scheme feature library stores the feature vectors of historical fault groups and their corresponding replacement schemes containing redundant rows and columns. Calculate the Euclidean distance between the feature vector of the current fault group and each feature vector in the feature library of the repair plan; Sort all the calculated Euclidean distances in ascending order of value, select the Euclidean distance A that is ranked first, and compare it with the preset distance threshold. If the Euclidean distance A ≤ the distance threshold, then the matching is determined to be successful, and the replacement scheme corresponding to the historical feature vector is obtained as a suggested repair scheme; If the Euclidean distance A is greater than the distance threshold, then the matching is deemed to have failed.

6. The memory chip failure cell location method based on fault group pattern matching according to claim 5, characterized in that, The recommended repair options also include: If the matching fails, the feature vectors in the feature library of the repair schemes corresponding to the first K Euclidean distances in the sorting are recorded as candidate vectors, and the replacement schemes corresponding to the candidate vectors are taken as candidate schemes. Determine whether the redundant rows and columns contained in the candidate scheme can cover all the failed units in the fault group. If so, the candidate scheme is selected as a candidate scheme. The candidate solution that ranks first among the corresponding candidate vectors is taken as the suggested repair solution for the fault group.

Images