NAND Flash bad block management method and system based on artificial intelligence

Abstract

The invention discloses an NAND Flash bad block management method and system based on artificial intelligence, and belongs to the technical field of solid-state storage, the method comprises the following steps: S1, carrying out multi-dimensional data acquisition and preprocessing on a bad Page based on preset parameters and preset indexes; s2, inputting the preprocessed multi-dimensional data into a space-time diagram convolutional network model, and outputting the failure probability of each Page; and S3, according to the failure probability obtained in the step S2, a hierarchical isolation strategy is determined, and then error correction code dynamic loading and preventive data migration are carried out. Actual measurement shows that more than 85% of normal Pages can be recycled in a Block with a single bad page, and the effective capacity is improved by 10 times compared with that of a traditional scheme; the overall service life of the chip is prolonged by about 30% by reducing the discarding of the whole chip caused by local damaged pages. The method also has a real-time risk management and control function, the prediction accuracy of the AI model to a potential risk area reaches 92%, and the secondary failure rate is reduced by 70% compared with a static isolation scheme.

Description

Technical Field

[0001] The present invention relates to the technical field of solid-state storage, and particularly relates to a method and system for NAND Flash bad block management based on artificial intelligence. Background Art

[0002] NAND Flash memories are widely used in various electronic devices. With the development of 3D NAND technology, their storage density and capacity have been greatly improved. However, this has also brought a series of new challenges, especially in the aspect of bad block management. In traditional NAND Flash memories, bad block management is a key issue. With the introduction of 3D NAND technology, the number of Pages within a single Block has increased significantly. For example, it has evolved from 64-layer stacking to more than 256 layers. This high-density storage structure has led to the following problems: 1) Bad Page diffusion effect. After a single Page is damaged due to reasons such as programming failure, read interference, or charge leakage, its physically adjacent Pages are affected by crosstalk or charge migration and may quickly fail in subsequent operations. This bad Page diffusion effect is particularly significant in high-density 3D NAND structures because the distance between Pages is closer and the coupling effect is stronger.

[0003] 2) Severe resource waste. Traditional bad block management strategies mark the entire Block as unavailable, resulting in thousands of normal Pages being discarded together. For example, in a certain 512GB chip, a single Block contains 1024 Pages, and due to just 1 bad Page, 2MB of effective space is lost, resulting in a significant loss of yield.

[0004] 3) Limitations of static isolation. Existing technologies use fixed-range isolation (such as isolating N Pages before and after a bad Page), and cannot dynamically adapt to the actual influence ranges of different processes, data loads, or aging stages, resulting in insufficient isolation (residual risk) or over-isolation (resource waste). Summary of the Invention

[0005] The purpose of the present invention is to overcome the deficiencies of the prior art and provide a method and system for NAND Flash bad block management based on artificial intelligence.

[0006] The purpose of the present invention is achieved through the following technical solutions: In the first aspect, the present invention discloses a method for NAND Flash bad block management based on artificial intelligence, including the following steps: S1. Perform multi-dimensional data collection on bad Pages based on preset parameters and preset metrics. A Page represents a page, which is the smallest unit for performing a read operation. Then, preprocess the collected multi-dimensional data; S2. Train a spatio-temporal graph convolutional network model, load the trained spatio-temporal graph convolutional network model, input the preprocessed multi-dimensional data into the spatio-temporal graph convolutional network model. The spatio-temporal graph convolutional network model extracts and transforms features from the preprocessed multi-dimensional data through graph convolutional layers and temporal convolutional layers, and outputs the failure probability of each Page. S3. According to the failure probability obtained in step S2, determine the directly affected area and the potential risk area. For the Pages in the directly affected area, take isolation measures to prohibit read and write operations; for the Pages in the potential risk area, allow reading but restrict writing, and start real-time health monitoring; at the same time, reorganize the remaining normal Pages across blocks in space to form virtual blocks for maximizing the utilization rate of the storage space; then perform dynamic loading of error correction codes and preventive data migration.

[0007] Based on the first aspect, step S1 specifically includes the following steps: S11. Physical feature acquisition, obtain and store the three-dimensional coordinates of the bad page and the measured values of the coupling capacitance of adjacent Pages to obtain physical feature data. S12. Electrical feature acquisition, measure and record the distortion of the threshold voltage distribution of the bad page and the voltage offset when reading fails to obtain electrical feature data. S13. Environmental feature acquisition: Record and save the temperature and power supply voltage fluctuations during operation to obtain environmental feature data. S14. Temporal correlation analysis, collect the operation history of the Blocks before and after the bad page, including the number of erase and write operations, read frequencies, and the programming order of adjacent Pages, to obtain temporal correlation data; A Block represents a block, which consists of multiple pages and is the smallest addressable unit for erase operations. S15. Preprocess the collected physical feature data, electrical feature data, environmental feature data, and temporal correlation data, including cleaning and normalization, to remove outliers and noise data in the data.

[0008] Based on the first aspect, the training of the spatio-temporal graph convolutional network model in step S2 specifically includes the following steps: S211. Training data acquisition, in a controllable experimental environment, manufacture preset types of bad pages, including programming failure type bad pages, read interference type bad pages, charge leakage type bad pages, multiple erase and write aging type bad pages, temperature sensitive type bad pages, record the diffusion paths and influence ranges of the preset types of bad pages as the first training sample data. S212. Supplement the first training sample data by collecting data on the spread of bad pages caused by natural aging of terminal devices in the actual usage environment to obtain the second training sample data. The training loss function of the spatio-temporal graph convolutional network model is the mean square error loss function , where N is the number of samples of the second training sample data, represents the true failure situation, represents the failure probability predicted by the model, which is used to measure the difference between the predicted failure probability and the true failure situation; S214. The spatio-temporal graph convolutional network model includes an input layer, a graph convolutional layer, a temporal convolutional layer, and an output layer. The parameters of the spatio-temporal graph convolutional network model are updated and optimized through the Adam optimization algorithm.

[0009] Based on the first aspect, step S2 further includes the following steps: S221. Model the preprocessed multi-dimensional data into a three-dimensional grid graph, take each Page as a node in the three-dimensional grid graph, and the node attributes include the health status and electrical parameters of the Page, and the coupling strength between Pages is reflected by the edge weights; S222. Receive the three-dimensional grid graph through the input layer, including the attribute information of all nodes and the edge weight matrix between nodes; S223. Perform graph convolution operations on the received three-dimensional grid graph through the graph convolutional layer to aggregate and update the node features to obtain spatial features, where represents the feature matrix of the nodes in the l-th layer, represents the adjacency matrix after adding self-connections, represents the corresponding degree matrix, represents the weight matrix of the l-th layer, and σ is the activation function; S224. Model the state changes of Pages at different time steps through the temporal convolutional layer to further improve the accuracy of the model's prediction of the impact range of bad pages; S225. According to the prediction target of the model, the output layer outputs the failure probability of each Page within a preset time, and divides the Pages into a direct impact area and a potential risk area.

[0010] Based on the first aspect, step S3 includes the following steps: S31. For the direct impact area, immediately isolate it and prohibit reading and writing; for the potential risk area, allow reading but restrict writing, and enable real-time health monitoring; S32. Construct a secondary mapping layer from logical address to physical address, combine the green area Pages from different Blocks into new virtual blocks, and during the recombination process, first select the Pages with similar programming / erasing cycles for combination; S33. Each virtual block reserves 5% - 10% of spare Pages. When a Page in the potential risk area is upgraded to the direct impact area, the replacement mechanism is automatically triggered, and the spare Pages are used to replace it.

[0011] Based on the first aspect, the dynamic loading of the error correction code in step S3 includes: using the default BCH code for the green area Page, upgrading the potential risk area Page to the LDPC code, and enabling the read voltage gradual adjustment algorithm.

[0012] Based on the first aspect, the preventive data migration described in step S3 includes: automatically migrating data to the green area when the monthly failure rate of Page in the potential risk area is predicted to exceed 15%.

[0013] In a second aspect, the present invention discloses an artificial intelligence-based NAND Flash bad block management system, which is used in the artificial intelligence-based NAND Flash bad block management method described in the first aspect, comprising: The bad page feature collection module is used to collect multi-dimensional data and pre-process the collected multi-dimensional data; The AI dynamic impact prediction model module is used to train and optimize the spatiotemporal graph convolutional network model. The trained spatiotemporal graph convolutional network model is then used to extract and transform features from preprocessed multi-dimensional data, outputting the failure probability of each page. The local isolation and virtual block reorganization module determines the hierarchical isolation strategy based on the failure probability output by the model. At the same time, it reorganizes the remaining normal pages across the block space to form virtual blocks. Adaptive error correction enhancement module, used for dynamic loading of error correction codes and preventive data migration.

[0014] The beneficial effects of the present invention are: 1) This invention improves storage utilization. Actual measurements show that in a block with a single bad page, more than 85% of normal pages can be recycled, increasing effective capacity by 10 times compared to traditional solutions.

[0015] 2) The present invention can extend the overall life of the chip by about 30% by reducing the waste of the entire block due to local bad pages.

[0016] 3) The present invention has the function of real-time risk management and control. The AI model has a 92% accuracy rate in predicting potential risk areas, which reduces the secondary failure rate by 70% compared with the static isolation solution. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 Schematic diagram of the steps of a NAND Flash bad block management method based on artificial intelligence according to an embodiment of the present invention; Figure 2 The figure is a schematic structural diagram of an artificial intelligence-based NAND Flash bad block management system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0018] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative efforts belong to the scope of protection of the present invention.

[0019] In the bad block management of NAND Flash, the marking and isolation of bad blocks are key steps. The prior art usually adopts the following two methods: 1) Boot area recording: Record bad block information in the boot area (Boot Block) of NAND Flash or a dedicated bad block marking area. Each block may have a bad block flag bit, which is used to indicate whether the block is a bad block. For example, store some status bytes at the head of the block to indicate whether the block has been marked as a bad block. 2) Bad block table: Maintain a bad block table in the external memory of NAND Flash or the memory of the controller. This table records the addresses or block indexes of all marked bad blocks. When the controller performs read and write operations on NAND Flash each time, it will check this table to avoid accessing bad blocks.

[0020] Bad block isolation and replacement. Once a bad block is identified, measures must be taken to isolate and replace it. Common bad block replacement strategies include: 1) Block replacement: Whenever a bad block is found, the system will select a new free block to replace the bad block. During the replacement process, the valid data in the bad block can be migrated to the new block (data migration). Specifically, when implementing, the system needs to maintain a block mapping table to record the mapping relationship between each logical block address (LBA) and the physical block address. 2) Write log or write mapping table: To ensure that data loss will not occur after bad block replacement, the system should use a write log or mapping table to record the mapping between logical addresses and physical addresses. The logical address is mapped to the physical address. If the physical address points to a bad block, the mapping table can be searched to find the new location. 3) Wear leveling: To avoid premature damage caused by frequent writing to certain blocks, the bad block management strategy should include a wear leveling algorithm (such as Wear Leveling). Wear leveling extends the service life of the entire Flash by balancing the number of erase and write cycles of different blocks.

[0021] Bad Block Management Strategy: 1) Bad Block Detection Frequency: The detection frequency of bad blocks can be adjusted according to the specific requirements of the system. For some applications with high reliability, the detection frequency of bad blocks can be increased. For general consumer-grade storage, stable operation can be maintained through regular scans and online detection. 2) Free Block Management: To improve performance, the system needs to maintain a list of free blocks. This list is used to record all free blocks that can be used for newly written data. When a bad block is discovered and replaced, a new free block should be selected from this list. 3) Adaptive Bad Block Management: Modern NAND Flash controllers usually adopt an adaptive bad block management strategy according to the trend of bad block occurrence. For example: 1. Adopt a more strict isolation strategy for older bad blocks. 2. Increase the isolation effect of newly emerging bad blocks through encryption technology or other means to avoid their impact on other parts of the system.

[0022] The main limitations of traditional bad block management schemes are as follows: 1) Resource Waste: Once a bad page is detected, the entire Block is marked as unavailable, resulting in a large number of normal Pages being discarded. 2) Static Isolation: Adopt a fixed-range isolation strategy and cannot dynamically adapt to the actual impact range of different processes, data loads, or aging stages. 3) Poor Adaptability: Unable to effectively cope with the bad page diffusion effect in high-density 3D NAND structures, resulting in a decline in storage reliability. Existing technologies cannot finely isolate local bad pages and lack the ability to predict potential failures. Aiming at the shortcomings of traditional bad block management schemes, the present invention aims to significantly improve storage utilization, extend the chip life, and reduce the storage space waste caused by local bad pages while ensuring data reliability.

[0023] The present invention provides an NAND Flash bad block management method based on artificial intelligence. The NAND Flash bad page management scheme based on artificial intelligence dynamically predicts the impact range of bad pages through the ST-GCN (Spatio-Temporal Graph Convolutional Network Model), adopts a hierarchical isolation strategy to accurately divide the risk area, and breaks through the physical Block limit to reorganize virtual storage blocks. Combined with adaptive error correction and preventive data migration, it can reduce the storage space waste caused by local bad pages by more than 80% while ensuring data reliability, and significantly improve the economic benefits of high-density NAND Flash chips. The schematic diagram of its steps is as Figure 1 shown and specifically includes the following steps: S1. Conduct comprehensive multi-dimensional data collection on bad Pages based on preset parameters and preset metrics. Here, "Page" represents a page, which is the smallest unit for performing read operations. After that, preprocess the collected multi-dimensional data. The preset parameters and preset metrics are not limited to specific numerical values but are values preset according to the actual working environment. Meanwhile, before conducting multi-dimensional data collection, it is also necessary to initialize the collection devices and sensors to ensure the accuracy and stability of data collection; S2. Train a spatio-temporal graph convolutional network model, load the trained spatio-temporal graph convolutional network model, and input the preprocessed multi-dimensional data into the spatio-temporal graph convolutional network model. The spatio-temporal graph convolutional network model extracts and transforms features from the preprocessed multi-dimensional data through graph convolutional layers and time convolutional layers, and outputs the failure probability of each Page; S3. Based on the failure probability obtained in step S2, determine the directly affected area and potential risk area. For the Pages in the directly affected area, take isolation measures to prohibit read and write operations; for the Pages in the potential risk area, allow reading but restrict writing, and start real-time health monitoring; at the same time, reorganize the remaining normal Pages across blocks in space to form virtual blocks for maximizing the utilization rate of storage space; then perform dynamic loading of error correction codes and preventive data migration.

[0024] Specifically, step S1 specifically includes the following steps: S11. Physical feature collection: Collect the three-dimensional coordinates (layer / row / column) of the bad page within the Block and the measured values of the coupling capacitance of adjacent Pages to obtain physical feature data; the three-dimensional coordinates are used to accurately locate the physical position of the bad page, and the measured values of the coupling capacitance reflect the capacitive coupling situation between pages, which helps to understand the potential impact of the bad page on other pages; S12. Electrical feature collection: Obtain the distortion of the threshold voltage distribution of the bad page and the voltage offset when the read fails to obtain electrical feature data. The distortion of the threshold voltage distribution can reflect the change in the electrical performance of the page, while the voltage offset during read failure directly indicates the abnormal situation that occurs in the page during the read operation. This information is crucial for evaluating the health status of the page; S13. Environmental feature collection: Record and save the temperature and power supply voltage fluctuations during operation to obtain environmental feature data; temperature changes may affect the performance and reliability of the page, and power supply voltage fluctuations may interfere with the read and write operations of the page. These environmental factors are of great significance for analyzing the formation and spread of bad pages; S14. Temporal correlation analysis: Collect the operation history of the Blocks before and after the bad page, including the number of erase cycles, read frequencies, and the programming order of adjacent Pages, to obtain temporal correlation data. A Block, which consists of multiple Pages, is the smallest addressable unit for erase operations. By analyzing this temporal correlation data, potential associations between bad pages and operations can be discovered, providing strong support for predicting the spread trend of bad pages. S15. Preprocess the collected physical feature data, electrical feature data, environmental feature data, and temporal correlation data, including operations such as cleaning and normalization, to remove outliers and noise data in the data, ensuring the accuracy and consistency of the data and preparing for subsequent input of the data into the model.

[0025] Specifically, the steps for training the spatio-temporal graph convolutional network model described in step S2 specifically include the following steps: S211. Training data acquisition: Under a controllable experimental environment, manufacture preset types of bad pages, including programming failure type bad pages, read interference type bad pages, charge leakage type bad pages, multiple erase aging type bad pages, and temperature sensitive type bad pages, and record the spread paths and influence ranges of the preset type bad pages as the first training sample data. In this embodiment, for example, for programming failure type bad pages, deliberately perform incorrect programming operations on specific Pages and observe their impacts on surrounding Pages; for charge leakage type bad pages, through an accelerated aging experiment, prompt charge leakage in the storage unit and record the corresponding data changes. These experimental data can directly reflect the spread characteristics of bad pages under specific conditions and provide high-quality samples for model training. S212. Supplement the first training sample data by collecting data on the spread of bad pages caused by natural aging of terminal devices in the actual usage environment to obtain the second training sample data. These data cover the real aging processes of NAND Flash memories under different usage scenarios and working conditions, can supplement the deficiencies of the first training sample data, and make the model more general and adaptable. S213. The training loss function of the spatio-temporal graph convolutional network model is the mean square error loss function , where N is the number of samples in the second training sample data, represents the true failure situation (0 represents normal, 1 represents failure), represents the predicted failure probability of the model, which is used to measure the difference between the predicted failure probability and the true failure situation. By minimizing this loss function, the model can accurately predict the influence range of bad pages. S214. The spatio-temporal graph convolutional network model (ST-GCN model) includes an input layer, a graph convolutional layer, a temporal convolutional layer, and an output layer. The parameters of the spatio-temporal graph convolutional network model are updated and optimized through the Adam optimization algorithm. The Adam algorithm combines the advantages of momentum and AdaGrad, can adaptively adjust the learning rate during training, accelerate the convergence speed of the model, and avoid falling into local optimal solutions, which helps the model better learn the complex patterns of bad page diffusion.

[0026] Specifically, step S2 further includes the following steps: S221. Model the preprocessed multi-dimensional data as a three-dimensional grid graph, with each Page as a node in the three-dimensional grid graph. The node attributes include the health status and electrical parameters of the Page. The health status can be obtained by an error detection mechanism, such as whether an uncorrectable error occurs; the electrical parameters cover the threshold voltage distribution, voltage offset when reading fails, etc. These parameters can reflect the current physical state and performance of the Page; the coupling strength between Pages is reflected by the edge weights, which are mainly calculated based on factors such as the physical distance between Pages and the charge coupling effect. The edge weights between adjacent Pages are larger because of the close distance and strong charge coupling effect; for non-adjacent but related Pages, the edge weights are relatively smaller. This weight setting reflects the degree of mutual influence between Pages; S222. Receive the three-dimensional grid graph through the input layer, including the attribute information of all nodes and the edge weight matrix between nodes; the input layer is responsible for integrating and preliminarily processing the complex NAND Flash storage structure data to prepare for subsequent convolution operations; S223. Perform graph convolution operations on the received three-dimensional grid graph through the graph convolutional layer , aggregate and update the node features to obtain spatial features, where represents the feature matrix of the nodes in the l-th layer, represents the adjacency matrix after adding self-connections, represents the corresponding degree matrix, represents the weight matrix of the l-th layer, and σ is the activation function. This operation enables each node to aggregate the information of its neighbor nodes, thereby learning a richer spatial feature representation; during the graph convolution operation, not only the features of the node itself are considered, but also the features of its neighbor nodes and the edge weight information are combined to capture the spatial dependence relationship between Pages; S224. After capturing the spatial features, introduce a temporal convolution operation to model the state changes of the Page at different time steps. Considering that during the long-term use of the NAND Flash memory, the health state of the Page will change over time, the temporal convolution layer can effectively capture the dependencies in this time series, further improving the accuracy of the model's prediction of the impact range of bad pages. S225. According to the prediction target of the model, the output layer outputs the failure probability of each Page within a preset time. In this embodiment, the preset time is 100 programming / erasure cycles. The Pages are divided into a directly affected area (failure probability > 90%) and a potential risk area (failure probability between 30% - 90%), providing a basis for subsequent isolation and reorganization strategies.

[0027] Specifically, step S3 includes the following steps: S31. For the directly affected area (red area), immediately isolate it and prohibit reading and writing to prevent the failure from further spreading and affecting other pages. For the potential risk area (yellow area), allow reading but restrict writing, and enable real-time health monitoring. This can reduce the possibility of further deterioration of the potential risk area without affecting data reading. S32. Construct a secondary mapping layer from logical addresses to physical addresses, breaking the limitation of physical Blocks. Combine the Pages in the green area (healthy or unused areas) from different Blocks into new virtual blocks. This helps optimize the layout of the storage space and improve the utilization rate of storage resources. During the reorganization process, first select Pages with similar programming / erasure cycles for combination to prevent the new virtual blocks from failing again due to excessive local aging. In this way, the service life of the new virtual blocks can be extended and the overall stability of the storage system can be improved. S33. Each virtual block reserves 5% - 10% of spare Pages. When a Page in the potential risk area is upgraded to the directly affected area, the replacement mechanism is automatically triggered, and the spare Pages are used to replace it to ensure data reliability and the continuity of the storage system.

[0028] Specifically, the dynamic loading of the error correction code in step S3 includes: using the default BCH code for the Pages in the green area, upgrading the Pages in the potential risk area to the LDPC code, and enabling the read voltage progressive adjustment algorithm.

[0029] Specifically, the preventive data migration in step S3 includes: automatically migrating data to the green area when the monthly failure rate of the Pages in the potential risk area exceeds 15%.

[0030] The present invention also discloses an NAND Flash bad block management system based on artificial intelligence for the above-mentioned NAND Flash bad block management method based on artificial intelligence. The structural schematic diagram is as Figure 2 shown, including: A bad page feature acquisition module, which is used to collect multi-dimensional data and preprocess the collected multi-dimensional data; An AI dynamic impact prediction model module, which is used to train and optimize the spatio-temporal graph convolutional network model, and then through the trained spatio-temporal graph convolutional network model, extract and transform the features of the preprocessed multi-dimensional data, and output the failure probability of each Page; A local isolation and virtual block recombination module, which determines a hierarchical isolation strategy according to the failure probability output by the model. At the same time, the remaining normal Pages are recombined across blocks in space to form virtual blocks; An adaptive error correction enhancement module, which is used for dynamic loading of error correction codes and preventive data migration.

[0031] The present invention proposes a method and system for local bad page dynamic isolation and space recombination based on artificial intelligence, which solves the limitations of traditional bad block management solutions: 1) Dynamic impact prediction, by using an AI model to analyze bad page features (such as error types, physical coordinates, P / E cycles, etc.) in real time, predicting its potential impact range, rather than relying on fixed rules. This method can dynamically adapt to the actual impact range of different processes, data loads or aging stages, effectively avoiding the problems of insufficient isolation or over-isolation; 2) Cross-block space recombination, recombining the unaffected normal Pages in multiple Blocks into virtual blocks, breaking through the physical Block boundary limit. This method can maximize the utilization of scattered normal Pages and significantly improve storage utilization; 3) Adaptive error correction enhancement, dynamically allocating the strength of error correction codes and the read voltage compensation strategy according to the AI risk assessment results of the area where the Page is located. This method can effectively improve data reliability and reduce the risk of data loss caused by bad pages; 4) A closed-loop management system, constructing a full-chain closed-loop management system of "prediction - isolation - recombination - monitoring" to achieve adaptive optimization of storage resources. It can monitor the storage status in real time and dynamically adjust the management strategy to ensure the long-term reliability of the storage system. On the premise of ensuring data reliability, the present invention greatly improves storage utilization, extends the chip life, and significantly reduces the storage space waste caused by local bad pages, and has important practical application value.

[0032] Exemplarily, taking 256-layer 3D NAND Flash as an example, NAND Flash: A Nand flash memory is a type of flash memory. It adopts a non-linear macrocell mode internally, providing a cheap and effective solution for the implementation of solid-state large-capacity memory. Currently, the most shipped and widely used is 3D TLC NAND.

[0033] Bad page detection. The ECC module (error correction code module) discovers that Page(125,34,78) has an uncorrectable error and marks it as a bad page. Feature extraction. Collect the lateral crosstalk coefficient, longitudinal charge leakage rate of the layer where this Page is located, and the variance of the operating voltage fluctuations in the recent 100 operations AI prediction. The output of the ST-GCN model shows that the directly affected area is Page(124~126,33~35,77~79) of the same layer, and the potential risk area is the adjacent Pages in the upper and lower layers. Isolation and reorganization. In the red area, 24 Pages (directly affected area) are isolated. In the yellow area, 48 Pages are marked as restricted for use. Reorganization: The remaining 952 green Pages in the original Block are merged with the normal Pages of another two Blocks to form 3 virtual blocks.

[0034] Dynamic monitoring. Each time a Page in the yellow area is written, it triggers a health assessment. If the failure rate reaches the threshold, data migration occurs.

[0035] The above are only the preferred embodiments of the present invention. It should be understood that the present invention is not limited to the form disclosed herein, should not be regarded as excluding other embodiments, but can be used in various other combinations, modifications, and environments, and can be changed within the scope of the concept described herein through the above teachings or the technology or knowledge in the relevant field. And any changes and modifications made by those skilled in the art without departing from the spirit and scope of the present invention shall fall within the protection scope of the appended claims of the present invention.

Claims

1. A method for managing bad blocks of NAND Flash based on artificial intelligence, characterized in that, It includes the following steps: S1. Perform multi-dimensional data collection on bad Pages based on preset parameters and preset metrics. A Page represents a page, which is the smallest unit for performing read operations. Then, preprocess the collected multi-dimensional data; S2. Train a spatio-temporal graph convolutional network model, load the trained spatio-temporal graph convolutional network model, and input the preprocessed multi-dimensional data into the spatio-temporal graph convolutional network model. The spatio-temporal graph convolutional network model extracts and transforms the features of the preprocessed multi-dimensional data through graph convolutional layers and time convolutional layers, and outputs the failure probability of each Page; S3. Determine the directly affected area and potential risk area according to the failure probability obtained in step S2. For the Pages in the directly affected area, take isolation measures to prohibit read and write operations; for the Pages in the potential risk area, allow reading but restrict writing, and start real-time health monitoring; at the same time, reorganize the remaining normal Pages across blocks in space to form virtual blocks for maximizing the utilization rate of storage space; After that, perform dynamic loading of error correction codes and preventive data migration.

2. The method for managing NAND Flash bad blocks based on artificial intelligence according to claim 1, wherein, The specific steps of step S1 include the following steps: S11. Physical feature collection, obtain and store the three-dimensional coordinates of the bad page and the measured values of the coupling capacitance of adjacent Pages to obtain physical feature data; S12. Electrical feature collection, measure and record the distortion of the threshold voltage distribution of the bad page and the voltage offset when reading fails to obtain electrical feature data; S13. Environmental feature collection: Record and save the temperature and power supply voltage fluctuations during operation to obtain environmental feature data; S14. Temporal correlation analysis, collect the operation history of the Blocks before and after the bad page, including the number of erase and write cycles, read frequencies, and the programming order of adjacent Pages, to obtain temporal correlation data; A Block represents a block, which consists of multiple pages and is the smallest addressable unit for erase operations; S15. Preprocess the collected physical feature data, electrical feature data, environmental feature data, and temporal correlation data, including cleaning and normalization, to remove outliers and noise data in the data.

3. The method for managing bad blocks of NAND Flash based on artificial intelligence according to claim 2, wherein, The specific steps of training the spatio-temporal graph convolutional network model described in step S2 include the following steps: S211. Training data acquisition, in a controllable experimental environment, manufacture preset types of bad pages, including programming failure type bad pages, read interference type bad pages, charge leakage type bad pages, multiple erase and write aging type bad pages, and temperature sensitive type bad pages, and record the diffusion paths and influence ranges of the preset types of bad pages as the first training sample data; S212. Supplement the first training sample data by collecting data on the spread of bad pages caused by natural aging of terminal devices in the actual usage environment to obtain the second training sample data; S213. The training loss function of the spatio-temporal graph convolutional network model is the mean square error loss function , where N is the number of samples of the second training sample data, represents the true failure situation, represents the failure probability predicted by the model, which is used to measure the difference between the predicted failure probability and the true failure situation; S214. The spatio-temporal graph convolutional network model includes an input layer, graph convolutional layers, time convolutional layers, and an output layer. Update and optimize the parameters of the spatio-temporal graph convolutional network model through the Adam optimization algorithm.

4. The artificial intelligence-based NAND Flash bad block management method according to claim 3, characterized in that: Step S2 also includes the following steps: S221. Model the preprocessed multi-dimensional data as a three-dimensional grid graph. Take each Page as a node in the three-dimensional grid graph, and the node attributes include the health status and electrical parameters of the Page. The coupling strength between Pages is reflected by the edge weights. S222. Receive the three-dimensional grid graph through the input layer, including the attribute information of all nodes and the edge weight matrix between nodes. S223. Perform graph convolution operation on the received three-dimensional grid graph through the graph convolution layer , aggregate and update the node features to obtain spatial features, where represents the feature matrix of the nodes in the l-th layer, represents the adjacency matrix after adding self-connections, represents the corresponding degree matrix, represents the weight matrix of the l-th layer, and σ is the activation function; S224. Model the state changes of Pages at different time steps through the temporal convolutional layer to further improve the accuracy of the model's prediction of the impact range of bad pages. S225. According to the prediction target of the model, the output layer outputs the failure probability of each Page within a preset time, and divides the Pages into a directly affected area and a potential risk area.

5. The method for managing bad blocks of NAND Flash based on artificial intelligence according to claim 4, wherein, The step S3 includes the following steps: S31. For the directly affected area, immediately isolate it and prohibit reading and writing. For the potential risk area, allow reading but restrict writing, and enable real-time health monitoring. S32. Construct a secondary mapping layer from logical addresses to physical addresses, and combine the green area Pages from different Blocks into new virtual blocks. During the recombination process, first select the Pages with similar programming / erasing cycles for combination. S33. Each virtual block reserves 5% - 10% of spare Pages. When a Page in the potential risk area is upgraded to the directly affected area, the replacement mechanism is automatically triggered, and the spare Page is used to replace it.

6. The artificial intelligence-based NAND Flash bad block management method according to claim 5, characterized in that: The dynamic loading of error correction codes described in step S3 includes: using the default BCH code for green area Pages, upgrading the Pages in the potential risk area to LDPC codes, and enabling the read voltage progressive adjustment algorithm.

7. The artificial intelligence-based NAND Flash bad block management method according to claim 5, characterized in that: The preventive data migration described in step S3 includes: automatically migrating data to the green area when the monthly failure rate of the Pages in the potential risk area exceeds 15%.

8. An NAND Flash bad block management system based on artificial intelligence, for use in the method for managing NAND Flash bad blocks based on artificial intelligence according to any one of claims 1-7, characterized in that, Including: A bad page feature acquisition module, which is used to collect multi-dimensional data and preprocess the collected multi-dimensional data. An AI dynamic impact prediction model module, which is used to train and optimize the spatio-temporal graph convolutional network model, and then through the trained spatio-temporal graph convolutional network model, extract and transform the features of the preprocessed multi-dimensional data, and output the failure probability of each Page. A local isolation and virtual block recombination module, which determines the hierarchical isolation strategy according to the failure probability output by the model. At the same time, it reorganizes the remaining normal Pages across blocks in space to form virtual blocks. An adaptive error correction enhancement module, which is used for dynamic loading of error correction codes and preventive data migration.

Images