Methods, apparatus, readable storage media and electronic devices for optimizing memory read count tables
By using neural network models in storage devices to predict the optimal readable voltage and update the voltage value and threshold of the read re-meter, the existing read re-meter cannot adapt to complex environments, and data stability and reliability are improved.
Patent Information
- Application Number
- CN202410954695.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-17
- Publication Date
- 2024-11-05
- Estimated Expiration
- 2044-07-17
AI Technical Summary
The existing read-repeat table cannot fully cover all read error scenarios, resulting in the inability to efficiently complete data rereading in complex usage environments, affecting the data stability and reliability of the storage device.
By obtaining the working environment data of the storage device, input it to the neural network model that has completed the training to obtain the best readable voltage, and update the voltage value and readable threshold in the read table based on the best readable voltage and reference voltage, dynamically adjusting Read the table to adapt to the current environment.
It improves the adaptability of storage devices to complex environments, reduces the risk of data loss, and enhances the data stability and reliability of NAND storage devices.
Smart Images

Figure CN118502677B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of solid-state storage technology, and in particular to a method and device for optimizing a memory read-repeat table, a readable storage medium, and an electronic device. Background Art
[0002] SSD (Solid State Disk or Solid State Drive) is a widely used storage device whose storage medium is NAND FLASH. During long-term use and storage of flash memory, due to factors such as self-erase and write wear, read interference, write interference, and data retention, the data stored in the flash memory is prone to bit flipping, resulting in data errors.
[0003] To address this problem, flash memory manufacturers usually provide a read retry table to handle read errors. However, since there are many situations in which read errors occur, the existing read retry table cannot fully cover all scenarios, making it impossible to efficiently complete the retry operation when rereading data based on the read retry table. Summary of the invention
[0004] The technical problem to be solved by the present invention is to provide a method, device, readable storage medium and electronic device for optimizing a memory read-repeat table, so as to enhance the data stability and reliability of a hard disk in a complex usage environment.
[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is:
[0006] A memory read-repeated table optimization method, comprising:
[0007] Get the working environment data of the current storage device;
[0008] Inputting the working environment data into the trained neural network model to obtain the best readable voltage;
[0009] Obtaining a current weight reading meter and a reference voltage of the current weight reading meter;
[0010] A read weight voltage value is obtained according to the optimal readable voltage and the reference voltage, and the voltage value and the readable threshold value in the current read weight table are updated according to the read weight voltage value.
[0011] In order to solve the above technical problems, another technical solution adopted by the present invention is:
[0012] A memory read-repeated table optimization device, comprising:
[0013] A first acquisition module is used to acquire the working environment data of the current storage device;
[0014] A prediction module, used for inputting the working environment data into the trained neural network model to obtain the best readable voltage;
[0015] A second acquisition module, used for acquiring a current weight reading table and a reference voltage of the current weight reading table;
[0016] The updating module is used to obtain a read weight voltage value according to the optimal readable voltage and the reference voltage, and to update the voltage value and the readable threshold value in the current read weight table according to the read weight voltage value.
[0017] In order to solve the above technical problems, another technical solution adopted by the present invention is:
[0018] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the various steps in the above-mentioned method for optimizing a memory read-repetition table.
[0019] In order to solve the above technical problems, another technical solution adopted by the present invention is:
[0020] An electronic device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor executes each step in a method for optimizing a memory read-repeat table.
[0021] The beneficial effects of the present invention are as follows: by collecting the current working environment data of the storage device, and predicting the collected working environment data based on the upgraded network model that has completed training, the optimal readable voltage corresponding to the current working environment is output, and then the voltage value and the readable threshold in the read weight table are updated through the optimal readable voltage and the reference voltage of the current read weight table, so that the voltage value and the readable threshold in the read weight table can be changed according to the current working environment of the storage device, thereby improving the adaptability of the storage device to complex environments; and the optimized read weight table effectively reduces the risk of data loss under specific environmental conditions, and improves the data stability and reliability of the NAND storage device. BRIEF DESCRIPTION OF THE DRAWINGS
[0022] Figure 1 A flowchart of a method for optimizing a memory read-repeat table according to an embodiment of the present invention;
[0023] Figure 2 A training flow chart of a neural network model in a memory read weight table optimization method in an embodiment of the present invention;
[0024] Figure 3 A schematic diagram of the structure of a memory read-repeat table optimization device according to an embodiment of the present invention;
[0025] Figure 4 FIG. 4 is a schematic diagram of the structure of an electronic device in an embodiment of the present invention. DETAILED DESCRIPTION
[0026] In order to explain the technical content, achieved objectives and effects of the present invention in detail, the following is an explanation in combination with the implementation modes and the accompanying drawings.
[0027] A memory read-repeated table optimization method, comprising:
[0028] Get the working environment data of the current storage device;
[0029] Inputting the working environment data into the trained neural network model to obtain the best readable voltage;
[0030] Obtaining a current weight reading meter and a reference voltage of the current weight reading meter;
[0031] A read weight voltage value is obtained according to the optimal readable voltage and the reference voltage, and the voltage value and the readable threshold value in the current read weight table are updated according to the read weight voltage value.
[0032] From the above description, it can be seen that the beneficial effects of the present invention are: by collecting the current working environment data of the storage device, and predicting the collected working environment data based on the upgraded network model that has completed training, the optimal readable voltage corresponding to the current working environment is output, and then the voltage value and the readable threshold in the read weight table are updated through the optimal readable voltage and the reference voltage of the current read weight table, so that the voltage value and the readable threshold in the read weight table can be changed according to the current working environment of the storage device, thereby improving the adaptability of the storage device to complex environments; and the optimized read weight table effectively reduces the risk of data loss under specific environmental conditions, and improves the data stability and reliability of the NAND storage device.
[0033] Further, the optimal readable voltage includes an optimal readable threshold and an optimal reading voltage;
[0034] Obtaining a read voltage value according to the optimal readable voltage and the reference voltage includes:
[0035] Obtaining a first offset voltage value according to a difference between the optimal readable threshold and the reference voltage;
[0036] Obtaining a second offset voltage value according to a difference between the optimal read voltage and the reference voltage;
[0037] The first offset voltage value and the second offset voltage value are used as the read weight voltage value.
[0038] From the above description, it can be seen that the first offset voltage value and the second offset voltage value are obtained by calculating the difference between the optimal readable threshold and the reference voltage, and the difference between the optimal read voltage and the reference voltage, respectively, and the read weight voltage value is obtained according to the first offset voltage value and the second offset voltage value, thereby realizing dynamic updating of the read weight table according to the environment.
[0039] Further, the read weight voltage value updates the voltage value and the readable threshold value in the current read weight table, including:
[0040] Obtaining an original readable threshold of the current read-repeat table;
[0041] Adding a second offset voltage value to each voltage value in the current weight reading table to obtain an updated weight reading table;
[0042] The first offset voltage value is added to the original readable threshold value as the readable threshold value of the updated read weight table.
[0043] From the above description, it can be seen that each voltage in the read weight table is updated by the second offset voltage value, so that the voltage in the read weight table is more suitable for data reading in the current working environment, and the readable threshold of the read weight table is updated by the first offset voltage value to ensure that data can be read under marginal conditions.
[0044] Furthermore, before inputting the working environment data into the trained neural network model, the method further comprises:
[0045] Determine whether the data retention time in the storage device is greater than a retention threshold. If so, execute the step of inputting the working environment data into the trained neural network model.
[0046] From the above description, it can be seen that by monitoring the retention time of data in the storage device, when the data retention time exceeds the retention threshold, that is, there is a risk that the read weight voltage of the data has changed, the read weight table is updated according to the working environment data to avoid data loss due to long-term storage.
[0047] Furthermore, before inputting the working environment data into the trained neural network model, the method further comprises:
[0048] Determine whether the working environment data collected at the current moment is consistent with the working environment data collected at the previous moment. If not, execute the step of inputting the working environment data into the trained neural network model.
[0049] From the above description, it can be seen that by real-time monitoring of changes in the current working environment data, when the working environment data changes, the read weight table is updated in time according to the currently collected working environment data, so that the corresponding voltage value in the read weight table can be changed in time to respond to environmental changes, thereby improving the data stability and reliability of the storage device.
[0050] Furthermore, the step of inputting the working environment data into the trained neural network model to obtain the optimal readable voltage includes:
[0051] Obtaining a state voltage indicator according to the working environment data;
[0052] The state voltage indicator is input into the trained neural network model to obtain the optimal readable voltage.
[0053] From the above description, it can be seen that after the state voltage index is obtained through the working environment data, the input state voltage index is predicted through the neural network to output the optimal readable voltage. Since the state voltage index and the optimal readable voltage are parameters with the same attributes, prediction based on the state voltage index can more accurately predict the optimal readable voltage.
[0054] Furthermore, before obtaining the state voltage indicator according to the working environment data, the method further comprises:
[0055] Perform accelerated life tests on the memory in different working environments, and record the state voltage indicators corresponding to the target working environment data;
[0056] The optimal readable voltage is marked according to the state voltage indicator for use in training the neural network model.
[0057] Furthermore, the training of the neural network model includes:
[0058] Obtain different types of storage devices as test samples;
[0059] Performing accelerated life tests on the test samples under different test conditions, and recording the test results corresponding to each of the test samples; constructing a data set according to the test samples, test conditions and test results;
[0060] The data set is used for training the neural network model to obtain the trained neural network model.
[0061] From the above description, it can be seen that by obtaining the corresponding state voltage indicators under accelerated life testing, not only can the test cycle be shortened, but also the comprehensive performance of the test samples under extreme conditions can be obtained, reflecting the characteristics of the test samples throughout the life cycle of the storage medium, and constructing a data set of test samples, test conditions and test results, thereby realizing the training of the neural network model.
[0062] Furthermore, the recording of the test result corresponding to each of the test samples includes:
[0063] Determining an acceptable maximum read error rate based on the error correction capability of the test sample;
[0064] The lowest voltage value in the test result that can be kept below the acceptable maximum error rate is used as the optimal readable threshold value.
[0065] It can be seen from the above description that by determining the optimal readable threshold corresponding to the test sample, the trained neural network model can effectively predict the optimal readable voltage.
[0066] Furthermore, the recording of the test result corresponding to each of the test samples includes:
[0067] By comparing the changes in the read error rate, data retention capability and erase / write cycle number of the test sample under different read voltages, the voltage value with the smallest impact on the life of the test sample is obtained, and the voltage value is used as the optimal read voltage.
[0068] It can be seen from the above description that by determining the optimal reading voltage corresponding to the test sample, the trained neural network model can effectively predict the optimal reading voltage.
[0069] Furthermore, obtaining the trained neural network model also includes:
[0070] Get verification data;
[0071] Inputting the verification data into the trained neural network model to obtain a verification result;
[0072] Calculating the verification result by using a loss function, and updating the parameters of the neural network model according to the calculation result;
[0073] The parameters of the neural network model are updated by combining the stochastic gradient descent algorithm with the momentum parameter. If the calculation result is that the model converges, the training is terminated; if the model does not converge, the step of obtaining verification data is executed again, and calculations are performed until the model converges.
[0074] From the above description, it can be seen that the trained neural network model is verified by combining the loss function and the stochastic gradient descent algorithm with the momentum parameter, thereby improving the prediction accuracy of the model.
[0075] Another embodiment of the present invention provides a memory read-repeat table optimization device, comprising:
[0076] A first acquisition module is used to acquire the working environment data of the current storage device;
[0077] A prediction module, used for inputting the working environment data into the trained neural network model to obtain the best readable voltage;
[0078] A second acquisition module, used for acquiring a current weight reading table and a reference voltage of the current weight reading table;
[0079] The updating module is used to obtain a read weight voltage value according to the optimal readable voltage and the reference voltage, and to update the voltage value and the readable threshold value in the current read weight table according to the read weight voltage value.
[0080] Another embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the computer program implements the steps in the above-mentioned method for optimizing a memory read-repetition table.
[0081] Another embodiment of the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps in the above-mentioned memory read-repetition table optimization method when executing the computer program.
[0082] The memory read-repetition table optimization method, device, readable storage medium and electronic device provided by the present invention can be applied to different types of memories to optimize the read-repetition table in the memory, which is described below through specific implementation methods:
[0083] Embodiment 1
[0084] Please refer to Figure 1 , a memory read-repeated table optimization method, comprising:
[0085] S1. Obtain the working environment data of the current storage device; wherein the working environment data includes the collected physical quantities, such as the temperature of the storage device, the number of read and write interferences, the number of erase and write wears, and the data retention time and other related parameters. Each data point contains the current status information of the storage device.
[0086] S2. Input the working environment data into the trained neural network model to obtain the best readable voltage, including:
[0087] S21, obtaining a state voltage index according to the working environment data; that is, mapping the parameters such as temperature and read / write interference times collected in step S1 to specific indicators of voltage state distribution, and the state voltage index includes indicators such as average voltage, standard deviation, and voltage distribution curve shape; for example, in a specific embodiment, each group of working environment data includes a data pair formed by a specific stability and read / write interference times, and the data pair is mapped to a group of state voltage indicators formed by the average voltage, the standard deviation of the voltage value, and the curve shape of the voltage distribution characteristics; that is, for a given set of temperatures and read / write interference times, the corresponding average voltage, the standard deviation of the voltage value, and the curve shape describing the voltage distribution characteristics are determined through the mapping process, thereby providing quantized and structured data for the input of the neural network model;
[0088] S22, inputting the state voltage index into the trained neural network model to obtain the best readable voltage; wherein the best readable voltage includes the best readable threshold and the best reading voltage, which are used to update the voltage value and the readable threshold of the weight reading table respectively;
[0089] S3, obtaining the current weight reading table and the reference voltage of the current weight reading table; wherein, the voltage value of the middle offset is selected from the original weight reading table as the reference voltage.
[0090] S4, obtaining a read weight voltage value according to the best readable voltage and the reference voltage, and updating the voltage value and the readable threshold value in the current read weight table according to the read weight voltage value, specifically including the following steps:
[0091] S41, obtaining a first offset voltage value ΔV1 according to a difference between an optimal readable threshold value and a reference voltage, and obtaining a second offset voltage value ΔV2 according to a difference between an optimal read voltage and a reference voltage; and using the first offset voltage value and the second offset voltage value as read voltage values;
[0092] S42, obtaining the original readable threshold of the current read-repeat table;
[0093] S43, adding the second offset voltage value to each voltage value in the current weight reading table to obtain an updated weight reading table; that is, adding the second offset voltage value ΔV2 to each voltage value in the original weight reading table to obtain a new "optimal reading voltage table" suitable for the current state;
[0094] S44, adding the first offset voltage value to the original readable threshold value as the readable threshold value of the updated read weight table; that is, adding the original readable threshold value to the first offset voltage value ΔV1 as the new readable threshold value to obtain a new readable threshold value suitable for the current state; applying the optimized read weight table to the actual reading process to improve the reading effect of the NAND storage device under various environmental conditions.
[0095] In a further embodiment, by real-time tracking of changes in working environment data and data retention time, the "optimal readable threshold" and "optimal reading voltage" are cyclically updated to achieve real-time adjustment of the readable threshold and voltage value of the read weight table.
[0096] As in an optional embodiment, before executing step S2, the following steps are included:
[0097] S01. Determine whether the data retention time in the storage device is greater than a retention threshold. If so, execute step S2.
[0098] In another optional embodiment, before executing step S2, the following steps are included:
[0099] S02, determine whether the working environment data collected at the current moment is consistent with the working environment data collected at the previous moment, if not, execute step S2. Among them, step S01 and step S02 can be executed at the same time, and step S2 is executed when the conditions in any step S01 and step S02 are met. That is, when the NAND storage device needs to be read under a specific environment, the voltage offset value predicted by the neural network model is used to optimize the read weight table provided by the original factory. The voltage point where the data can be read can be located more accurately, reducing the probability of read errors.
[0100] Embodiment 2
[0101] This embodiment specifically defines the training process of the neural network model. The neural network model is constructed using the voltage distribution and actual reading results (including the best readable threshold and the best reading voltage) under different working environments contained in the big data set; the model takes the voltage state distribution index as input and outputs the corresponding best readable voltage, that is, the best readable threshold and the best reading voltage are obtained; when the model is inferred, the working environment data must first be mapped to the state voltage index and then the state voltage index must be input into the neural network model.
[0102] Please refer to Figure 2 , is the specific training process of the neural network model:
[0103] S221. Environmental data collection: Obtain different types of storage devices as test samples, that is, obtain NAND chips of different capacities and processes from various manufacturers. It is necessary to cover new products and old products that have been running for a certain period of time, and ensure that a sufficient number of samples are collected to ensure statistical significance.
[0104] S222, constructing a data set: The model data set is constructed mainly by performing accelerated life tests on the memory under different working environments, and recording the state voltage index corresponding to the target working environment; then marking the best readable voltage according to the state voltage index for training the neural network model; at the same time, after the model training, a new data set needs to be constructed for performance verification; the reliability of the neural network model is verified by inputting new physical quantity data to ensure its adaptability to various working environments, which specifically includes the following steps:
[0105] S2221. Accelerated life test: Perform life tests on each sample under high temperature, low temperature, read / write interference and other environmental conditions. By testing different data retention times and accelerated aging processes, the test cycle is shortened to obtain the comprehensive performance of the samples under various extreme conditions. For example, the durability of the storage unit is evaluated by the number of erase and write cycles (P / E Cycles); the data retention capability understands the long-term stability of the data, and the read error rate (Bit Error Rate, BER) measures the reading accuracy.
[0106] S2222. Record voltage distribution (measure voltage distribution simultaneously during accelerated life test): comprehensively measure voltage distribution under different working environments and data retention times; obtain sufficient statistical data by testing each sample in turn, including statistical parameters such as average voltage and standard deviation, as well as various characteristics of the voltage distribution curve.
[0107] S2223. Labeling (labeling the data obtained after the accelerated life test): Based on the performance indicator data such as the number of erase and write cycles, data retention capability and read error rate obtained in the accelerated life test, label each set of parameters with "optimal readable threshold" and "optimal read voltage" to establish a reference standard for the reading performance of each sample under different environments; among them, the optimal readable threshold is used to ensure that data can be read even under marginal conditions; the optimal read voltage guides how to read under normal conditions to ensure the highest data accuracy and the lowest error rate.
[0108] S2224. Determine the readable threshold: Under each set of test conditions, based on the trend of the read error rate changing with the read voltage, obtain the voltage point where the error rate begins to increase significantly. This voltage point is the "critical voltage" under the corresponding working environment and data retention time conditions. Based on the error correction capability (ECC capability) of the device, determine an acceptable maximum read error rate, where different storage devices have different maximum read error rates. For each set of test conditions, obtain the lowest voltage value that can just stay below the acceptable maximum error rate. This value is the optimal readable threshold.
[0109] S2225. Optimal read voltage: In the collected data, compare the read error rates under different read voltages, while considering the changes in data retention capability and the number of erase and write cycles; obtain a voltage value with the lowest read error rate and at which the data retention capability is not significantly affected; if there are multiple voltage values with similar performance, further consider the number of erase and write cycles, select the voltage value with the least impact on the life of the storage device, and use this voltage value as the optimal read voltage.
[0110] S223. Input data to the model: The model structure adopts a fully connected neural network. The input layer contains working environment data parameters such as the working temperature of the storage device, the probability of read misoperation, the probability of write misoperation, and the data retention time; the mapped voltage state distribution statistical characteristics include the average voltage, voltage standard deviation, the skewness and kurtosis of the voltage distribution curve, and the probability that the voltage exceeds the normal working range and other related parameters; the output node is the optimal readable threshold and the optimal read voltage. The model inputs the corresponding prediction results according to the input data, and processes the prediction results by calculating the loss function. Among them, the model also includes data preprocessing: standardizing the input and output data to improve the convergence speed of the neural network, and performing de-standardization when in use.
[0111] S224, model prediction: the model outputs corresponding prediction results based on the input data.
[0112] S225. Calculate the loss function: MSE is used as the loss function. The specific expression is as follows:
[0113]
[0114] in, is the model output, is the label; N is the number of samples; i is the i-th sample; after calculating the loss function, the model parameters are further updated.
[0115] S226. Update model parameters: The model uses a stochastic gradient descent algorithm combined with momentum parameters to update parameters.
[0116] S227, model convergence: that is, determine whether the result calculated in step S226 converges. If the model converges, the training ends; if the model does not converge, return to step S223 for calculation.
[0117] Embodiment 3
[0118] Please refer to Figure 3 , a memory read-repeat table optimization device, comprising:
[0119] A first acquisition module is used to acquire the working environment data of the current storage device;
[0120] A prediction module, used for inputting the working environment data into the trained neural network model to obtain the best readable voltage;
[0121] A second acquisition module, used for acquiring a current weight reading table and a reference voltage of the current weight reading table;
[0122] The updating module is used to obtain a read weight voltage value according to the optimal readable voltage and the reference voltage, and to update the voltage value and the readable threshold value in the current read weight table according to the read weight voltage value.
[0123] Embodiment 4
[0124] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the computer program implements the various steps in a memory read-repetition table optimization method in Embodiment 1 and Embodiment 2.
[0125] Embodiment 5
[0126] Please refer to Figure 4 An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, each step in a memory read-repetition table optimization method in Embodiment 1 and Embodiment 2 is implemented.
[0127] In summary, the memory read weight table optimization method, device, readable storage medium and electronic device provided by the present invention comprehensively consider the influence of factors such as temperature, read and write interference times, erase and write wear, Data Retention and the like on the storage device, construct a neural network model, and use the neural model to predict the current working environment data and output the optimal readable voltage to update the read weight table, so that the voltage value and the readable threshold in the read weight table can be changed according to the current working environment of the storage device, so that the NAND storage device can maintain high performance under different conditions, thereby improving the adaptability of the storage device to complex environments; and the optimized read weight table effectively reduces the risk of data loss under specific environmental conditions, thereby improving the data stability and reliability of the NAND storage device.
[0128] In the above embodiments provided in the present application, it should be understood that the disclosed methods, devices, computer-readable storage media, and electronic devices can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the modules is only a logical function division. There may be other division methods in actual implementation, such as multiple components or modules can be combined or integrated into another device, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or components or modules, which can be electrical, mechanical or other forms.
[0129] The components described as separate components may or may not be physically separated, and the components shown as components may or may not be physical modules, that is, they may be located in one place or distributed on multiple network modules. Some or all of the components may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
[0130] In addition, each functional module in each embodiment of the present invention may be integrated into one processing module, or each component may exist physically separately, or two or more modules may be integrated into one module. The above integrated modules may be implemented in the form of hardware or in the form of software functional modules.
[0131] If the integrated module is implemented in the form of a software function module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.
[0132] It should be noted that, for the convenience of description, the aforementioned method embodiments are all described as a series of action combinations, but those skilled in the art should be aware that the present invention is not limited by the described action sequence, because according to the present invention, certain steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present invention.
[0133] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.
[0134] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent transformations made using the contents of the present invention's specification and drawings, or directly or indirectly applied in related technical fields, are also included in the patent protection scope of the present invention.
Claims
1. A method for optimizing a memory read-repeated table, characterized in that: include: Get the working environment data of the current storage device; Inputting the working environment data into the trained neural network model to obtain the best readable voltage; Obtaining a current weight reading meter and a reference voltage of the current weight reading meter; Obtaining a read weight voltage value according to the optimal readable voltage and the reference voltage, and updating the voltage value and the readable threshold value in the current read weight table according to the read weight voltage value; The optimal readable voltage includes an optimal readable threshold and an optimal reading voltage; Obtaining a read voltage value according to the optimal readable voltage and the reference voltage includes: Obtaining a first offset voltage value according to a difference between the optimal readable threshold and the reference voltage; Obtaining a second offset voltage value according to a difference between the optimal read voltage and the reference voltage; Obtaining the read weight voltage value according to the first offset voltage value and the second offset voltage value; The read weight voltage value updates the voltage value and the readable threshold value in the current read weight table, including: Obtaining an original readable threshold of the current read-repeat table; Adding a second offset voltage value to each voltage value in the current weight reading table to obtain an updated weight reading table; Adding the first offset voltage value to the original readable threshold value as the readable threshold value of the updated read weight table; Before inputting the working environment data into the trained neural network model, the process includes: Determine whether the working environment data collected at the current moment is consistent with the working environment data collected at the previous moment. If not, execute the step of inputting the working environment data into the trained neural network model.
2. A memory read-repeat table optimization method according to claim 1, characterized in that: Before inputting the working environment data into the trained neural network model, the process includes: Determine whether the data retention time in the storage device is greater than a retention threshold. If so, execute the step of inputting the working environment data into the trained neural network model.
3. The method for optimizing a memory read-repeat table according to claim 1, characterized in that: The step of inputting the working environment data into the trained neural network model to obtain the optimal readable voltage comprises: Obtaining a voltage status indicator according to the working environment data; The voltage state indicator is input into the trained neural network model to obtain the optimal readable voltage.
4. The method for optimizing a memory read-repeat table according to claim 1, characterized in that: The training of the neural network model includes: Obtain different types of storage devices as test samples; Performing accelerated life tests on the test samples under different test conditions, and recording the test results corresponding to each of the test samples; constructing a data set according to the test samples, test conditions and test results; The data set is used for training the neural network model to obtain the trained neural network model.
5. A method for optimizing a memory read-repeat table according to claim 4, characterized in that: The recording of the test result corresponding to each test sample includes: Determining an acceptable maximum read error rate based on the error correction capability of the test sample; The lowest voltage value in the test result that can be kept below the acceptable maximum error rate is used as the optimal readable threshold value.
6. A method for optimizing a memory read-repeat table according to claim 4, characterized in that: The recording of the test result corresponding to each test sample includes: By comparing the changes in the read error rate, data retention capability and erase / write cycle number of the test sample under different read voltages, the voltage value with the smallest impact on the life of the test sample is obtained, and the voltage value is used as the optimal read voltage.
7. A method for optimizing a memory read-repeat table according to claim 4, characterized in that: The neural network model that has completed training also includes: Get verification data; Inputting the verification data into the trained neural network model to obtain a verification result; Calculating the verification result by using a loss function, and updating the parameters of the neural network model according to the calculation result; The parameters of the neural network model are updated by combining the stochastic gradient descent algorithm with the momentum parameter. If the calculation result is that the model converges, the training is terminated; if the model does not converge, the step of obtaining verification data is executed again, and calculations are performed until the model converges.
8. A memory read-repeated table optimization device, characterized in that: include: A first acquisition module is used to acquire the working environment data of the current storage device; A prediction module, used for inputting the working environment data into the trained neural network model to obtain the best readable voltage; A second acquisition module, used for acquiring a current weight reading table and a reference voltage of the current weight reading table; An updating module, configured to obtain a read weight voltage value according to the optimal readable voltage and the reference voltage, and to update a voltage value and a readable threshold value in the current read weight table according to the read weight voltage value; The optimal readable voltage includes an optimal readable threshold and an optimal reading voltage; Obtaining a read voltage value according to the optimal readable voltage and the reference voltage includes: Obtaining a first offset voltage value according to a difference between the optimal readable threshold and the reference voltage; Obtaining a second offset voltage value according to a difference between the optimal read voltage and the reference voltage; Obtaining the read weight voltage value according to the first offset voltage value and the second offset voltage value; The read weight voltage value updates the voltage value and the readable threshold value in the current read weight table, including: Obtaining an original readable threshold of the current read-repeat table; Adding a second offset voltage value to each voltage value in the current weight reading table to obtain an updated weight reading table; Adding the first offset voltage value to the original readable threshold value as the readable threshold value of the updated read weight table; Before inputting the working environment data into the trained neural network model, the process includes: Determine whether the working environment data collected at the current moment is consistent with the working environment data collected at the previous moment. If not, execute the step of inputting the working environment data into the trained neural network model.
9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, each step in a memory read-repeat table optimization method as described in any one of claims 1 to 7 is implemented.
10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the processor implements the various steps in the memory read duplicate table optimization method as described in any one of claims 1 to 7.
