A Method and System for Optimizing Fault Grouping and Disconnection Matching Process Based on FGPM Fault Grouping Algorithm
By constructing a two-layer cache architecture based on topological feature quintuples and hash identifiers, the fault group unmatching process of the FGPM algorithm is optimized, solving the problems of unmatching failure and high computational complexity of the FGPM algorithm in large-scale storage arrays, and achieving efficient fault group repair and chip yield improvement.
Patent Information
- Application Number
- CN202611306287.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-26
- Publication Date
- 2026-11-17
AI Technical Summary
The existing FGPM algorithm fails to match fault groups with more than 5 fault points, has high computational complexity, serious repetitive calculations, and cannot meet the real-time requirements of mass production testing. Furthermore, complex fault groups are easily misjudged as unrepairable, leading to a decrease in chip yield.
A two-layer cache architecture is constructed using topological feature quintuples and unique topological hash identifiers. Through coarse filtering of topological features and fine matching of hashes, a unique mapping from any complex fault group to historical solutions is achieved, generating the optimal repair solution, reducing computational complexity and improving solution generation efficiency.
It achieves a unique mapping for any complex fault group, avoids repetitive deduction, reduces computational complexity, improves solution generation efficiency, adapts to high-speed automated testing equipment, and ensures chip repair success rate and yield.
Smart Images

Figure CN123116074A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of semiconductor memory manufacturing and testing technology, specifically to an optimization method for the fault group matching process of memory arrays based on the FGPM (FaultGroup Pattern Matching) algorithm, which is particularly suitable for the mass production testing and repair process of high-capacity, high-bandwidth memory chips such as DDR5 and HBM3. Background Technology
[0002] As semiconductor process nodes continue to evolve to 3nm and below, the storage density of modern memory chips has increased exponentially, with a single DDR5 chip reaching 64Gb and HBM3 chips even exceeding 1TB. However, memory cell failures caused by process defects, material inhomogeneity, and lithography errors have become increasingly serious, with the number of failure points increasing from dozens in the early days to hundreds or even thousands.
[0003] To improve chip yield, the industry commonly employs redundancy repair technology, which involves pre-designing a certain number of spare rows and spare columns in the memory array. During mass production testing, an automated test equipment (ATE) performs a full address traversal test on the memory array, collecting the row and column addresses of all faulty cells and constructing a two-dimensional fault matrix. The fault matrix is then input into a redundancy repair algorithm for decision-making, and the repair circuit maps the row or column containing the faulty cell to the corresponding spare resource. However, as the array size increases and the number of fault points grows, the problem of optimizing the combination of redundant resources and fault points has evolved into an NP-hard combinatorial coverage problem.
[0004] Existing repair techniques can be mainly divided into the following categories: Traditional restoration techniques Early repair techniques primarily employed simple repair strategies, such as row-first repair, column-first repair, and combined row-column repair. While these methods were simple to implement, they had significant limitations: when the fault distribution was complex, they easily led to a waste of backup resources, had a low repair success rate, and could not cope with the complex fault scenarios of large-scale storage arrays.
[0005] FGPM (Fault Group Pattern Matching) algorithm To overcome the shortcomings of traditional methods, the FGPM algorithm was proposed and widely applied. The core idea of this algorithm is to decompose a large-scale problem into multiple smaller sub-problems through fault grouping: first, interconnected fault points are divided into independent fault groups; then, candidate repair schemes are generated for each fault group; finally, the optimal global repair combination is searched under resource constraints. This algorithm significantly reduces computational complexity and has become the mainstream repair algorithm in mass production testing.
[0006] However, existing FGPM algorithms still face key technical bottlenecks in practical applications: when the number of fault points N>5 within a fault group, the number of combinations of fault topologies increases explosively, and the solutions stored by the algorithm cannot correspond one-to-one with the fault group patterns, causing the matching process to fail. At this point, the algorithm can only generate repair solutions through real-time exhaustive search, which significantly increases computational overhead and cannot meet the millisecond-level time window requirements in mass production testing; at the same time, some complex topologies are easily misjudged as unrepairable due to the lack of matching solutions, causing an "overkill" problem and reducing chip yield. Summary of the Invention
[0007] The purpose of this invention is to overcome the shortcomings of the prior art and provide a fault group unmatching process optimization method and system based on the FGPM fault grouping algorithm. It aims to solve the problems of unmatching failure, high computational complexity and serious repetitive calculation in the existing FGPM algorithm when dealing with fault groups with more than 5 fault points. While ensuring the repair success rate, it can significantly improve the efficiency of solution generation, thereby meeting the real-time requirements of mass production testing of large-scale storage arrays.
[0008] To achieve the above-mentioned objectives, the present invention adopts the following technical solution: The present invention provides an optimization method for fault group resolution matching based on the FGPM fault grouping algorithm, characterized by the following steps: Step 1: Obtain the memory fault matrix and make a judgment to obtain the fault rows and columns that must be repaired first, and repair them to obtain the repaired fault rows and columns as well as the remaining fault rows and columns. Step 2: Use a breadth-first search method based on row-column connectivity to group the fault points in the remaining fault rows and columns to obtain single-fault fault groups and multi-fault fault groups, and calculate the topological statistical feature quintuple for each fault group. Step 3: Generate a unique topology hash identifier for each fault group based on the topology statistical feature quintuple of each fault group and the row-column adjacency relationship of all fault points within each fault group; Step 4: Build a solution cache library, which stores the repair solutions for all fault groups; Step 5: Using the topological statistical feature quintuple F_k of the kth fault group G_k as the index key of the first level, traverse the solution cache library and search for all fault groups and their repair solution sets that are the same as F_k. Using the topological hash identifier H_k of G_k as the index key of the second layer, search for the set of fault groups and their repair schemes that have the same topological hash identifier as the k-th fault group G_k from all fault groups and their repair schemes, and use it as the optimal repair scheme set S_k of the k-th fault group G_k; if no such set is found, directly generate the optimal repair scheme set S_k of the k-th fault group G_k and update it in the two-layer cache library; Step 6: Summarize the set of optimal repair solutions for all fault groups to construct the global repair solution space corresponding to the fault matrix. Combined with the constraints on the number of remaining redundant rows and columns, search for the globally optimal repair combination from the global repair solution space to perform the repair operation.
[0009] The characteristic of the fault group unmatching process optimization method of the present invention is that, in step 2, the topological statistical feature quintuple of any k-th fault group G_k is denoted as F_k={Nfr_k, Nfc_k, Nmr_k, Nmc_k, Nf_k}, where Nfr_k is the number of fault rows contained in G_k, Nfc_k is the number of fault columns contained in G_k, Nmr_k is the number of multi-fault rows contained in G_k, Nmc_k is the number of multi-fault columns contained in G_k, and Nf_k is the total number of fault points contained in G_k; k=1,2,...,M, where M represents the total number of fault groups.
[0010] Furthermore, step 3 involves generating a unique topological hash identifier as follows: Step 3.1: Normalize the fault groups to remove rows and columns that do not contain fault points in the k-th fault group G_k, so as to form a compact fault group G'_k. Step 3.2: Extract the row set Rows_k and column set Cols_k of all fault points in the k-th fault group G'_k; Step 3.3: Construct a row-to-column adjacency dictionary R2C_k, with each row number in R2C_k as the key and the set of column numbers where all fault points in each row are located as the value; Construct a column-to-row adjacency dictionary C2R_k, with each column number in R2C_k as the key and the set of row numbers where all fault points in the column are located as the value; Step 3.4: Sort the elements in Rows_k and Cols_k in ascending order to obtain the sorted row set Rows'_k and the sorted column set Cols'_k; Based on the sorted row set Rows'_k, the adjacency relationships in R2C_k are converted into a normalized string representation from row to column; Based on the sorted column set Cols'_k, the adjacency relationships in C2R_k are converted into a normalized string representation from column to row; Combine the row-to-column normalized string representation and the column-to-row normalized string representation into a whole normalized string representation; Step 3.5: Apply the MD5 hash algorithm to process the overall normalized string representation to generate a unique topological hash identifier H_k for G_k.
[0011] Furthermore, in step 5, if no solution can be found, the optimal repair scheme set S_k for the k-th fault group G_k is generated as follows: Step 5.1: Compare the number of faulty rows Nfr_k with the number of faulty columns Nfc_k in G_k: If Nfr_k ≤ Nfc_k, then the row-first repair strategy is adopted and steps 5.2-5.3 are executed; otherwise, the column-first repair strategy is adopted; otherwise, steps 5.4-5.5 are executed. Step 5.2: Select the row with the fewest fault points from the k-th fault group G_k, and change its repair strategy to a column-first repair strategy to generate a candidate repair scheme; Step 5.3: Traverse all rows in the k-th fault group G_k in the manner described in Step 5.2 to generate Nfr_k candidate repair schemes; Step 5.4: Select the column with the fewest fault points from the k-th fault group G_k, and change its repair strategy to the row-first repair strategy to generate a candidate repair scheme; Step 5.5: Traverse all columns in the k-th fault group G_k in the manner described in Step 5.3 to generate Nfc_k candidate repair schemes; Step 5.6: Perform deduplication on the Nfr_k+Nfc_k candidate repair schemes to obtain several deduplicated candidate repair schemes; Step 5.7: Sort the deduplicated candidate repair schemes in ascending order according to the total cost r_i+c_i, and select the top N schemes with the smallest total cost to form the optimal repair scheme set S_k of G_k, where r_i and c_i represent the number of redundant rows and columns required for any candidate repair scheme, respectively.
[0012] The present invention provides a fault group resolution matching process optimization system based on the FGPM fault grouping algorithm, characterized by including: The fault priority determination module is used to determine the fault matrix of memory, obtain the fault rows and columns that must be repaired first, and then repair them to obtain the repaired fault rows and columns as well as the remaining fault rows and columns. The fault feature extraction module uses a breadth-first search method based on row-column connectivity to group the fault points in the remaining fault rows and columns, and calculates the topological statistical feature quintuple for the grouped fault groups to generate a unique topological hash identifier. The dual-layer cache matching module is used to perform coarse screening and precise matching of repair solutions for fault groups from the solution cache library, thereby generating the optimal set of repair solutions; The real-time solution generation module is used to generate candidate repair solutions for unmatched fault groups, and after deduplication, sorting and optimization, generate the optimal set of repair solutions. The global repair solution module is used to summarize the set of optimal repair solutions for all fault groups, construct a global repair solution space, and search for the globally optimal repair combination by combining the constraints of the number of remaining redundant rows and columns. The results output module is used to perform repair operations on the fault matrix according to the globally optimal repair combination, verify the repair results, and output relevant statistical information.
[0013] The fault group matching process optimization system of this invention is also characterized by the fact that the dual-layer cache matching module includes: The first-level index unit uses the topological statistical feature quintuple as the first-level index key, and searches the solution cache library to find all fault groups and their repair solution sets that are the same as the current fault group, thereby completing the coarse-grained screening of fault groups. The second-level index unit uses the topology hash identifier as the index key to search for the set of fault groups and their repair schemes that have the same topology hash identifier as the current fault group from all fault groups and their repair schemes, thereby completing the fine-grained matching of fault groups. The cache update unit is used to write the set of optimal repair schemes generated by the unmatched fault groups, along with their corresponding topological statistical feature quintuples and topological hash identifiers, into the scheme cache library to achieve dynamic updating and reuse of the cache library.
[0014] The present invention provides an electronic device, including a memory and a processor, characterized in that the memory is used to store a program supporting a method of execution by the processor, and the processor is configured to execute the program stored in the memory.
[0015] The present invention discloses a computer-readable storage medium storing a computer program, characterized in that the computer program is executed by a processor to perform the steps of a method.
[0016] Compared with the prior art, the present invention has significant advantages: 1. This invention addresses the problem of solution matching failure when the original FGPM algorithm handles complex fault groups with multiple fault points. It innovatively introduces a topological feature quintuple and a unique topological hash identifier to construct a two-layer caching architecture of "coarse screening of statistical features and fine matching of topological hash". This achieves a unique mapping from any complex fault group to historical solutions, avoids repeated deduction, and improves solution generation efficiency and scenario coverage.
[0017] 2. Compared to the exponential growth in computational complexity of traditional algorithms, this invention transforms the generation of most solutions into cache lookups through a dual-layer caching mechanism. Combined with an optimization strategy for candidate solutions in missed scenarios, this reduces computational complexity to polynomial level, significantly compressing decision-making time. In scenarios with large-scale DDR5 chip failures, it can achieve microsecond-level solution generation and is compatible with high-speed automated testing equipment. Simultaneously, this method achieves full topology coverage, resolving the "overkill" problem of complex fault groups being erroneously discarded in the original algorithm. It also optimizes global redundant resource scheduling, effectively improving product yield while ensuring optimal chip repair success rate. Attached Figure Description
[0018] Figure 1 This is a schematic flowchart of the method of the present invention; Figure 2 A schematic diagram of topological hash encoding. Detailed Implementation
[0019] In this embodiment, a fault group unmatching process optimization method based on the FGPM fault grouping algorithm is used to solve the problem of unmatching failure when the number of fault points in a fault group is greater than 5 during memory storage array fault repair. In a mass production testing and repair scenario for a 64Gb DDR5 memory chip, the chip adopts an 8-bank storage architecture, with each bank containing 16384 rows and 2048 columns of storage units. Each bank is pre-configured with 64 redundant rows and 32 redundant columns for fault repair. Specifically, for example... Figure 1 As shown, the method includes the following steps: Step 1: Fault Matrix Acquisition and Mandatory Repair Phase: Step 1 involves acquiring the fault matrix and performing mandatory repairs as follows: Obtain the fault matrix of the memory array, identify the fault rows and columns that must be repaired first based on resource constraints, perform repair operations and update the system status to obtain the partially repaired fault matrix and the remaining redundant resources.
[0020] Step 1.1. Perform a full address traversal test on the memory bank using an automated test equipment (ATE), collect the row and column addresses of all faulty units, and construct a two-dimensional fault matrix; initialize the remaining number of redundant rows R_{rem}, the remaining number of redundant columns (C_{rem}), and the row and column fault counts.
[0021] Step 1.2. Determining Items That Must Be Repaired: For any column c, if the number of fault points in the column F_c > R_{rem}, then it is determined to be a column that must be repaired; for any row r, if the number of fault points in the row F_r > C_{rem}, then it is determined to be a row that must be repaired.
[0022] Step 1.3. Use a greedy strategy to prioritize repairing the row or column with the most fault points that must be repaired, and update the fault matrix, redundant resource count, and row and column fault count simultaneously.
[0023] Step 1.4. Iteratively execute the judgment and repair process until there are no necessary repair items, resources are exhausted, or resource types are mismatched.
[0024] In this embodiment, the test equipment obtains the fault matrix of the target bank through ATE testing, which contains a total of 326 fault points; the initial redundancy resource configuration is R_{rem}=64, C_{rem}=32.
[0025] Identification of items that must be repaired: By counting the number of fault points in each row and column, it was found that column 1024 contains 68 fault points, which is greater than the number of remaining redundant rows of 64. Therefore, column 1024 is determined to be a column that must be repaired.
[0026] The following repair procedure must be executed: Prioritize repairing column 1024, consuming 1 redundant column resource, updating C_{rem}=31, and simultaneously clearing all 68 fault points on this column.
[0027] Status Update: After updating the fault matrix, the number of remaining fault points is 258, and the fault counts of all rows and columns are updated synchronously.
[0028] Iterative loop: Based on the updated state, re-identify the items that must be repaired, confirm that there are no longer any rows or columns that meet the conditions for repair, and the repair phase ends.
[0029] Step 2: Use a breadth-first search method based on row-column connectivity to group the fault points in the remaining fault rows and columns to obtain single-fault fault groups and multi-fault fault groups, and calculate the topological statistical feature quintuple for each fault group.
[0030] In this embodiment, the BFS method is used to group the remaining 258 fault points by connectivity, resulting in 31 independent fault groups, including 14 single fault groups (containing only 1 fault point) and 17 multi-fault groups (containing 2 or more fault points); among them, 8 fault groups have more than 5 fault points, and the largest fault group contains 12 fault points.
[0031] Step 3: Based on the topological statistical characteristics quintuple of each fault group and the row-column adjacency relationships of all fault points within each fault group, generate a unique topological hash identifier for each fault group; such as... Figure 2 As shown: Step 3.1: Normalize the fault groups, remove rows and columns that do not contain fault points from the k-th fault group G_k, and form a compact normalized fault group G'_k; Step 3.2: Extract the row set Rows_k and column set Cols_k corresponding to all fault points in the normalized fault group G'_k; Step 3.3: Construct a row-to-column adjacency dictionary R2C_k. This adjacency dictionary R2C_k uses the row number as the key and the set of column numbers corresponding to all fault points in that row as the value. It is used to store the fault distribution of each row in a structured way, providing the baseline data for row-by-row traversal for subsequent normalized string generation. Construct a column-to-row adjacency dictionary C2R_k, which uses column number as key and the set of row numbers corresponding to all fault points in that column as value. This dictionary is used to store the fault distribution of each column in a structured manner, providing baseline data for column-by-column traversal for subsequent normalized string generation.
[0032] Step 3.4: Sort the elements in the row set Rows_k and column set Cols_k in ascending order to obtain the sorted row set Rows'_k and column set Cols'_k; Following the key order of the sorted row set Rows'_k, the column number set values corresponding to the row number in the adjacency dictionary R2C_k are read sequentially to generate a normalized string from row to column; Following the key order of the sorted column set Cols'_k, the row number set values corresponding to the column numbers in the adjacency dictionary C2R_k are read sequentially to generate a normalized string from column to row; Concatenate the row-to-column normalized string with the column-to-row normalized string to obtain the overall normalized string of the fault group; Step 3.5: Apply the MD5 hash algorithm to the entire normalized string to generate a unique topological hash identifier H_k for the k-th fault group G_k.
[0033] In this embodiment, taking the multi-fault group numbered G7 as an example, this group contains 9 fault points, involving 5 rows and 4 columns; 3 rows contain more than 2 fault points, and 2 columns contain more than 2 fault points. Therefore, its topological statistical feature quintuple is F_7=(5,4,3,2,9).
[0034] The topological hash generation process is as follows: extract the row set {256, 512, 1024, 2048, 4096} and column set {128, 256, 512, 1024} of the group, and construct the row adjacency dictionary and column adjacency dictionary respectively; after sorting the row and column numbers in ascending order, a normalized adjacency string is generated, and the unique topological hash identifier H_7 of the group is obtained by MD5 hash calculation.
[0035] Step 4: Build a solution cache library, which stores the repair solutions for all fault groups; Step 5: Using the topological statistical feature quintuple F_k of the kth fault group G_k as the index key of the first level, traverse the solution cache library and search for all fault groups and their repair solution sets that are the same as F_k. Using the topological hash identifier H_k of G_k as the index key of the second layer, search for the set of fault groups and their repair schemes that have the same topological hash identifier as the k-th fault group G_k from all fault groups and their repair schemes, and use it as the optimal repair scheme set S_k of the k-th fault group G_k; if no such set is found, directly generate the optimal repair scheme set S_k of the k-th fault group G_k and update it in the two-layer cache library; In this embodiment: the current cache library has accumulated more than 1,200 fault topology-repair scheme entries.
[0036] First-level coarse screening: The first-level cache is searched using the quintuple (5,4,3,2,9), resulting in 7 entries that match the quintuple, forming a candidate matching set; Second-level fine matching: Search the second-level index of the candidate set with the topological hash identifier H_7, and if a completely matching entry is found, directly retrieve the three optimal repair schemes stored under that entry to form the repair scheme set S_7 of fault group G7.
[0037] The above matching process was executed sequentially on 31 fault groups. Ultimately, 24 fault groups were cache hits and 7 fault groups were not hits, with an overall cache hit rate of approximately 77.4%.
[0038] Step 5: Generating solutions for cache misses and updating the cache: For fault groups that miss the cache, a repair plan is generated and optimized in real time to obtain a set of repair plans for the fault group. The topological characteristics of the group and the corresponding plan are then updated to the two-layer cache library.
[0039] If no solution is found, the optimal repair scheme set S_k for the k-th fault group G_k is generated as follows: Step 5.1: Compare the number of faulty rows Nfr_k with the number of faulty columns Nfc_k in G_k: If Nfr_k ≤ Nfc_k, then the row-first repair strategy is adopted and steps 5.2-5.3 are executed; otherwise, the column-first repair strategy is adopted; otherwise, steps 5.4-5.5 are executed.
[0040] Step 5.2: Select the row with the fewest fault points from the k-th fault group G_k, and change its repair strategy to a column-first repair strategy to generate a candidate repair scheme; Step 5.3: Traverse all rows in the k-th fault group G_k in the manner described in Step 5.2 to generate Nfr_k candidate repair schemes.
[0041] Step 5.4: Select the column with the fewest fault points from the k-th fault group G_k, and change its repair strategy to the row-first repair strategy to generate a candidate repair scheme; Step 5.5: Traverse all columns in the k-th fault group G_k in the manner described in Step 5.3 to generate Nfc_k candidate repair schemes; Step 5.6: Perform deduplication on the Nfr_k+Nfc_k candidate repair schemes to obtain several deduplicated candidate repair schemes.
[0042] Step 5.7: Sort the deduplicated candidate repair schemes in ascending order according to the total cost r_i+c_i, and select the top N schemes with the smallest total cost to form the optimal repair scheme set S_k of G_k, where r_i and c_i represent the number of redundant rows and columns required for any candidate repair scheme, respectively.
[0043] In this embodiment: For the 7 cache miss groups, candidate solutions are generated using either a row-first or column-first strategy based on the size of the row and column misses, with 12-18 candidate solutions generated for each group. After deduplication and sorting by total cost, the top 5 optimal solutions for each group are retained to form the corresponding repair solution set. Simultaneously, the topology quintuple, topology hash, and corresponding repair solutions for these 7 fault groups are written into a two-layer cache library to complete the cache update.
[0044] Step 6: Summarize the set of optimal repair solutions for all fault groups to construct the global repair solution space corresponding to the fault matrix. Combined with the constraints on the number of remaining redundant rows and columns, search for the globally optimal repair combination from the global repair solution space to perform the repair operation.
[0045] In this embodiment: all candidate solutions for 31 fault groups are aggregated to construct a global solution space. The optimal solution is then searched using Depth-First Search (DFS) combined with branch-and-bound techniques. The globally optimal repair solution is obtained, consuming 21 redundant rows and 11 redundant columns, with a total resource consumption of 32. After executing all repair operations according to the solution, the fault matrix is completely cleared, and the repair is successful.
[0046] After verification in the mass production test environment, the total time for repair decision-making of a single bank of the chip after adopting this optimization method is 2.1ms, which is about 4 times more efficient than the original FGPM algorithm of 8.7ms. At the same time, all repairable fault groups have obtained effective matching solutions and no overkill occurs.
[0047] In this embodiment, an electronic device includes a memory and a processor. The memory stores a program that supports the processor in executing the above-described method, and the processor is configured to execute the program stored in the memory.
[0048] In this embodiment, a computer-readable storage medium stores a computer program, which is executed by a processor to perform the steps of the above method.
Claims
1. A method for optimizing the fault grouping and matching process based on the FGPM fault grouping algorithm, characterized in that, Includes the following steps: Step 1: Obtain the memory fault matrix and make a judgment to obtain the fault rows and columns that must be repaired first, and repair them to obtain the repaired fault rows and columns as well as the remaining fault rows and columns. Step 2: Use a breadth-first search method based on row-column connectivity to group the fault points in the remaining fault rows and columns to obtain single-fault fault groups and multi-fault fault groups, and calculate the topological statistical feature quintuple for each fault group. Step 3: Generate a unique topology hash identifier for each fault group based on the topology statistical feature quintuple of each fault group and the row-column adjacency relationship of all fault points within each fault group. Step 4: Build a solution cache library, which stores the repair solutions for all fault groups; Step 5: Using the topological statistical feature quintuple F_k of the kth fault group G_k as the index key of the first level, traverse the solution cache library and search for all fault groups and their repair solution sets that are the same as F_k. Using the topological hash identifier H_k of G_k as the index key of the second layer, search for the set of fault groups and their repair schemes that have the same topological hash identifier as the k-th fault group G_k from all fault groups and their repair schemes, and use it as the optimal repair scheme set S_k of the k-th fault group G_k; if no such set is found, directly generate the optimal repair scheme set S_k of the k-th fault group G_k and update it in the two-layer cache library; Step 6: Summarize the set of optimal repair solutions for all fault groups to construct the global repair solution space corresponding to the fault matrix. Combined with the constraints on the number of remaining redundant rows and columns, search for the globally optimal repair combination from the global repair solution space to perform the repair operation.
2. The fault group matching process optimization method according to claim 1, characterized in that, In step 2, the topological statistical feature quintuple of any k-th fault group G_k is denoted as F_k={Nfr_k, Nfc_k, Nmr_k, Nmc_k, Nf_k}, where Nfr_k is the number of fault rows contained in G_k, Nfc_k is the number of fault columns contained in G_k, Nmr_k is the number of multi-fault rows contained in G_k, Nmc_k is the number of multi-fault columns contained in G_k, and Nf_k is the total number of fault points contained in G_k; k=1,2,...,M, where M represents the total number of fault groups.
3. The fault group resolution matching process optimization method according to claim 2, characterized in that, Step 3 involves generating a unique topological hash identifier using the following steps: Step 3.1: Normalize the fault groups to remove rows and columns that do not contain fault points in the k-th fault group G_k, so as to form a compact fault group G'_k. Step 3.2: Extract the row set Rows_k and column set Cols_k of all fault points in the k-th fault group G'_k; Step 3.3: Construct a row-to-column adjacency dictionary R2C_k, with each row number in R2C_k as the key and the set of column numbers where all fault points in each row are located as the value; Construct a column-to-row adjacency dictionary C2R_k, with each column number in R2C_k as the key and the set of row numbers where all fault points in the column are located as the value; Step 3.4: Sort the elements in Rows_k and Cols_k in ascending order to obtain the sorted row set Rows'_k and the sorted column set Cols'_k; Based on the sorted row set Rows'_k, the adjacency relationships in R2C_k are converted into a normalized string representation from row to column; Based on the sorted column set Cols'_k, the adjacency relationships in C2R_k are converted into a normalized string representation from column to row; Combine the row-to-column normalized string representation and the column-to-row normalized string representation into a whole normalized string representation; Step 3.5: Apply the MD5 hash algorithm to process the overall normalized string representation to generate a unique topological hash identifier H_k for G_k.
4. The fault group matching process optimization method according to claim 1, characterized in that, In step 5, if no solution can be found, the optimal repair scheme set S_k for the k-th fault group G_k is generated as follows: Step 5.1: Compare the number of faulty rows Nfr_k with the number of faulty columns Nfc_k in G_k: If Nfr_k ≤ Nfc_k, then the row-first repair strategy is adopted and steps 5.2-5.3 are executed; otherwise, the column-first repair strategy is adopted; otherwise, steps 5.4-5.5 are executed. Step 5.2: Select the row with the fewest fault points from the k-th fault group G_k, and change its repair strategy to a column-first repair strategy to generate a candidate repair scheme; Step 5.3: Traverse all rows in the k-th fault group G_k in the manner described in Step 5.2 to generate Nfr_k candidate repair schemes; Step 5.4: Select the column with the fewest fault points from the k-th fault group G_k, and change its repair strategy to the row-first repair strategy to generate a candidate repair scheme; Step 5.5: Traverse all columns in the k-th fault group G_k in the manner described in Step 5.3 to generate Nfc_k candidate repair schemes; Step 5.6: Perform deduplication on the Nfr_k+Nfc_k candidate repair schemes to obtain several deduplicated candidate repair schemes; Step 5.7: Sort the deduplicated candidate repair schemes in ascending order according to the total cost r_i+c_i, and select the top N schemes with the smallest total cost to form the optimal repair scheme set S_k of G_k, where r_i and c_i represent the number of redundant rows and columns required for any candidate repair scheme, respectively.
5. A fault group resolution matching process optimization system based on the FGPM fault grouping algorithm, characterized in that, include: The fault priority determination module is used to determine the fault matrix of memory, obtain the fault rows and columns that must be repaired first, and then repair them to obtain the repaired fault rows and columns as well as the remaining fault rows and columns. The fault feature extraction module uses a breadth-first search method based on row-column connectivity to group the fault points in the remaining fault rows and columns, and calculates the topological statistical feature quintuple for the grouped fault groups to generate a unique topological hash identifier. The dual-layer cache matching module is used to perform coarse screening and precise matching of repair solutions for fault groups from the solution cache library, thereby generating the optimal set of repair solutions; The real-time solution generation module is used to generate candidate repair solutions for unmatched fault groups, and after deduplication, sorting and optimization, generate the optimal set of repair solutions. The global repair solution module is used to summarize the set of optimal repair solutions for all fault groups, construct a global repair solution space, and search for the globally optimal repair combination by combining the constraints of the number of remaining redundant rows and columns. The results output module is used to perform repair operations on the fault matrix according to the globally optimal repair combination, verify the repair results, and output relevant statistical information.
6. The fault group matching process optimization system according to claim 5, characterized in that, The dual-layer cache matching module includes: The first-level index unit uses the topological statistical feature quintuple as the first-level index key, and searches the solution cache library to find all fault groups and their repair solution sets that are the same as the current fault group, thereby completing the coarse-grained screening of fault groups. The second-level index unit uses the topology hash identifier as the index key to search for the set of fault groups and their repair schemes that have the same topology hash identifier as the current fault group from all fault groups and their repair schemes, thereby completing the fine-grained matching of fault groups. The cache update unit is used to write the set of optimal repair schemes generated by the unmatched fault groups, along with their corresponding topological statistical feature quintuples and topological hash identifiers, into the scheme cache library to achieve dynamic updating and reuse of the cache library.
7. An electronic device, comprising a memory and a processor, characterized in that, The memory is used to store a program that enables the processor to perform the method of any one of claims 1-4, and the processor is configured to execute the program stored in the memory.
8. A computer-readable storage medium storing a computer program thereon, characterized in that, The steps of the method of any one of claims 1-4 are executed by a computer program when it is run by a processor.
