Method for reducing influence on flash memory data retention
By extracting the optimal read voltage in the flash block group and processing it using machine learning algorithms, the problem that flash particles cannot measure the impact of temperature and placement time on data retention when they are not powered on is solved, and the reliability of data retention on flash particles is achieved. and the effect of improving performance.
Patent Information
- Application Number
- CN202411202007.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-29
- Publication Date
- 2024-12-20
AI Technical Summary
In the prior art, the reliability of flash memory particles is affected by temperature and placement time, but it cannot be quantified directly through these two factors without power on, resulting in a lack of effective data retention measurement indicators and the inability to effectively reduce data retention. Effect on flash memory particles.
By selecting a group of flash blocks, the number of erases is 100, and after writing data at room temperature, place it in the thermostat to extract the optimal read voltage, and use machine learning algorithms such as random forests and simulated annealing algorithms to divide according to the placement time Data retention category, use read voltage as a measure of data retention to determine whether data needs to be moved.
Reduces the impact of data retention on the reliability of flash memory particles, filters out redundant or irrelevant features, reduces the read voltage acquisition time, reduces the impact on performance, and improves the reliability of data retention.
Smart Images
Figure CN119170085A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of data storage, and in particular is a method for reducing the influence on flash memory data retention. Background Art
[0002] As the market demand for increased capacity of flash memory particles continues to increase, the number of bits in each cell gradually increases, resulting in more and more attention being paid to the reliability of flash memory particles. The main indicators of flash memory particle reliability are wear times, read disturbance and data retention, and wear times and read disturbance can be quantitatively measured.
[0003] A Chinese patent with publication number CN116682484A discloses a flash memory particle testing method, device and electronic equipment, including collecting flash memory particle testing information corresponding to each temperature holding time at each set temperature holding time; the flash memory particle testing information at least includes: the real temperature of different flash memory particles set at different positions in the temperature box, and the real temperature of the temperature box; for each flash memory particle, sampling the characteristic quantity of at least one physical block in the flash memory particle in the temperature box, determining the characteristic quantity to be compensated corresponding to the flash memory particle based on the characteristic quantities of each sampled physical block, and determining the characteristic quantity compensation value corresponding to the flash memory particle based on the flash memory particle testing information of the flash memory particle collected at each temperature holding time; determining the target characteristic quantity corresponding to the flash memory particle based on the characteristic quantity compensation value corresponding to the flash memory particle and the characteristic quantity to be compensated corresponding to the flash memory particle; the target characteristic quantity is used as the basis for flash memory particle screening; by compensating the characteristic quantities of flash memory particles at different positions in the temperature box, the influence of temperature box differences or temperature unevenness can be offset, thereby improving the accuracy of flash memory particle screening and reliability testing.
[0004] In the above technology, the reliability of flash memory particles is improved by offsetting the impact of temperature differences or uneven temperatures. However, the influencing factors usually also include temperature and the time placed at the temperature. When the flash memory particles are not powered on, the current temperature and the time placed at the temperature cannot be obtained, and these two factors cannot be directly used to quantitatively measure them. Due to the lack of effective data retention measurement indicators, the impact of data retention cannot be effectively reduced, thereby affecting the reliability of the flash memory particles.
[0005] To this end, the present invention provides a method for reducing the impact on flash memory data retention. Summary of the invention
[0006] In order to make up for the deficiencies of the prior art and solve the problem that the factors affecting the reliability of flash memory particles proposed in the background technology usually also include temperature and the time they are placed at the temperature, but when the flash memory particles are not powered on, the current temperature and the time they are placed at the temperature cannot be obtained, and they cannot be directly quantitatively measured by these two factors. Due to the lack of effective data retention measurement indicators, the impact of data retention cannot be effectively reduced, thereby affecting the reliability of the flash memory particles.
[0007] The technical solution adopted by the present invention to solve the technical problem is: a method for reducing the influence on flash memory data retention described in the present invention comprises the following steps:
[0008] S1: Select a group of flash memory block groups, the flash memory block group includes multiple flash memory blocks, and the number of erasure and write times of the selected flash memory blocks is 100;
[0009] S2: After writing data at room temperature, place it in a temperature box at a temperature of T c After the storage time t1, each flash memory block extracts a set of optimal read voltages at room temperature; after the storage time t2 in the temperature box, each flash memory block extracts a set of optimal read voltages at room temperature;
[0010] S3: The most relevant features are selected through machine learning algorithms. Different placement times in the incubator indicate different degrees of data retention. Data retention is divided into different categories according to the placement time, i.e., placement time t1, t2, …, t N Corresponding to categories 1, 2, ..., N respectively;
[0011] S4: The selected read voltage is used as a measure of data retention. The technician uses this measure to judge the degree of data retention and thus determines whether to move the data in the flash memory block.
[0012] Preferably, each word line of each flash memory block in S2 has a set of optimal read voltages, and the optimal read voltage of a characteristic word line is selected as the optimal read voltage of the flash memory block; or a set of voltages suitable for most word lines of the flash memory block are extracted as the optimal voltages through an intelligent optimization algorithm.
[0013] Preferably, the intelligent optimization algorithm adopts one of a simulated annealing algorithm or a particle swarm optimization algorithm.
[0014] Preferably, the algorithm in S3 adopts a forest algorithm, takes the optimal read voltage as the input of the random forest, and the category as the output of the random forest. After the model is trained, the importance evaluation parameters of the random forest model are obtained, the largest read voltage among the importance evaluation parameters is selected, and the read voltage whose importance evaluation parameters exceed the threshold is selected.
[0015] Preferably, each of the flash memory blocks performs N operations, where Tc ,t1,t2,…,t N Determined based on technician's experience.
[0016] Preferably, the optimal voltage refers to the read voltage corresponding to the position of the trough of the threshold voltage distribution, which is related to the state of the flash memory block and no clear value is set.
[0017] Preferably, the flash memory particles of the flash memory block adopt an operation mode including one or more of SLC, MLC, QLC and PLC.
[0018] The beneficial effects of the present invention are as follows:
[0019] 1. A method for reducing the influence on flash memory data retention according to the present invention comprises extracting a read voltage and an optimal read voltage; selecting a group of flash memory block groups, the flash memory block group comprising a plurality of flash memory blocks, the erasing and writing times of the selected flash memory blocks are all 100; after writing data at room temperature, placing the flash memory block in a temperature box at a temperature of T c After being placed for time t1, each flash memory block extracts a set of optimal read voltages at room temperature; after being placed in the temperature box for time t2, each flash memory block extracts a set of optimal read voltages at room temperature; the most relevant features are selected through a machine learning algorithm; the selected read voltage is used as a measurement indicator of data retention. The technician uses this indicator to judge the degree of data retention, thereby determining whether to move the data in the flash memory block, reducing the impact of data retention on the reliability of the flash memory particles, and at the same time screening out redundant or irrelevant features, reducing the time to obtain the read voltage, and reducing the impact on performance.
[0020] 2. A method for reducing the impact on flash memory data retention described in the present invention adopts a simulated annealing algorithm, by heating a solid to a sufficiently high temperature and then slowly cooling it. When heated, the particles inside the solid become disordered as the temperature rises, and the internal energy increases, while the particles gradually tend to be ordered when slowly cooled, reaching an equilibrium state at each temperature, and finally reaching a ground state at room temperature, reducing the internal energy to a minimum, and extracting a set of voltages suitable for most word lines of the flash memory block as the optimal voltage, wherein the optimal voltage refers to the read voltage corresponding to the position of the trough of the threshold voltage distribution, which is related to the state of the flash memory block and has no clear value. BRIEF DESCRIPTION OF THE DRAWINGS
[0021] The present invention will be further described below in conjunction with the accompanying drawings.
[0022] Figure 1 is an overall block diagram of a method according to an embodiment of the present invention;
[0023] Figure 2 It is a block diagram for extracting the optimal read voltage of a flash memory block group in the present invention;
[0024] Figure 3is a threshold voltage distribution diagram of the MLC operation mode in the present invention;
[0025] Figure 4 is a flow chart of the data retention method in the present invention; DETAILED DESCRIPTION
[0026] In order to make the technical means, creative features, objectives and effects achieved by the present invention easy to understand, the present invention is further explained below in conjunction with specific implementation methods.
[0027] like Figures 1 to 4 As shown, a method for reducing the impact on flash memory data retention according to an embodiment of the present invention includes the following steps:
[0028] S1: Select a group of flash memory block groups, the flash memory block group includes multiple flash memory blocks, and the number of erasure and write times of the selected flash memory blocks is 100;
[0029] S2: After writing data at room temperature, place it in a temperature box at a temperature of T c After the storage time t1, each flash memory block extracts a set of optimal read voltages at room temperature; after the storage time t2 in the temperature box, each flash memory block extracts a set of optimal read voltages at room temperature;
[0030] S3: The most relevant features are selected through machine learning algorithms. Different placement times in the incubator indicate different degrees of data retention. Data retention is divided into different categories according to the placement time, i.e., placement time t1, t2, …, t N Corresponding to categories 1, 2, ..., N respectively;
[0031] S4: The selected read voltage is used as a measure of data retention. The technician uses this measure to judge the degree of data retention and thus determines whether to move the data in the flash memory block.
[0032] In the prior art, the reliability of flash memory particles is improved by offsetting the influence of temperature differences or uneven temperatures. However, the influencing factors usually also include temperature and the time of being placed at the temperature. When the flash memory particles are not powered on, the current temperature and the time of being placed at the temperature cannot be obtained, and these two factors cannot be directly used for quantitative measurement. Due to the lack of effective data retention measurement indicators, the influence of data retention cannot be effectively reduced, thereby affecting the reliability of the flash memory particles.
[0033] When the method for reducing the impact of data retention provided by the present invention is used, firstly a group of flash memory block groups is selected, the flash memory block group includes a plurality of flash memory blocks, and the erasing and writing times of the selected flash memory blocks are all 100;
[0034] Then, after writing the data at room temperature, place it in a temperature box at a temperature of T cAfter the placement time t1, each flash memory block extracts a set of optimal read voltages at room temperature. The extraction block diagram is shown in Figure 2 , each word line of each flash memory block has a set of optimal read voltages. The optimal read voltage of a characteristic word line can be selected as the optimal read voltage of the flash memory block, or a set of voltages suitable for most word lines of the flash memory block can be extracted as the optimal voltage through the intelligent optimization algorithm; after being placed in the temperature box for time t2, each flash memory block extracts a set of optimal read voltages at room temperature; repeat the operation N times, where T c ,t1,t2,…,t N According to the experience of technicians, t1, t2, ..., t N The values of can be the same or different;
[0035] Then, the most relevant features are selected through the machine learning algorithm. Different placement times in the incubator indicate different degrees of data retention. The data retention can be divided into different categories according to the placement time, i.e., placement time t1, t2, …, t N They correspond to categories 1, 2, ..., N respectively; the machine learning algorithm takes the random forest algorithm as an example, takes the optimal read voltage as the input of the random forest, and the category as the output of the random forest. After the model is trained, the importance evaluation parameters of the random forest model are obtained, and the largest read voltage among the importance evaluation parameters is selected, or the read voltage whose importance evaluation parameter exceeds the threshold is selected;
[0036] Finally, the selected read voltage is used as a measure of data retention. Technicians can use this indicator to judge the degree of data retention and thus determine whether to move the data in the flash memory block.
[0037] This technical solution uses a machine learning algorithm to select a read voltage that has a great impact on data retention as a feature value, thereby reducing the impact of data retention on the reliability of flash memory particles. At the same time, it screens out redundant or irrelevant features, reduces the time to obtain the read voltage, and reduces the impact on performance.
[0038] like Figure 2 As shown, each word line of each flash memory block in S2 has a set of optimal read voltages, and the optimal read voltage of a characteristic word line is selected as the optimal read voltage of the flash memory block; or a set of voltages suitable for most word lines of the flash memory block are extracted as the optimal voltages through an intelligent optimization algorithm.
[0039] When the optimal read voltage provided by the present invention is used, the optimal read voltage of a characteristic word line can be selected as the optimal read voltage of the flash memory block; or a group of voltages applicable to most word lines of the flash memory block can be extracted as the optimal voltage through an intelligent optimization algorithm, wherein the voltage applicable to most word lines of the flash memory block as the optimal voltage refers to a read voltage that minimizes the sum of the mean and maximum values of the number of error bits of all flash memory pages in all flash memory blocks.
[0040] like Figure 1 As shown, the intelligent optimization algorithm adopts a simulated annealing algorithm.
[0041] The simulated annealing algorithm provided by the present invention adopts the simulated annealing algorithm when in use, by heating the solid to a sufficiently high temperature and then letting it cool slowly. When heated, the particles inside the solid become disordered as the temperature rises, and the internal energy increases. When cooled slowly, the particles gradually tend to be ordered, reaching an equilibrium state at each temperature, and finally reaching a ground state at room temperature, and the internal energy is reduced to a minimum.
[0042] like Figure 1 As shown, the algorithm in S3 adopts a forest algorithm, takes the optimal read voltage as the input of the random forest, and the category as the output of the random forest. After the model is trained, the importance evaluation parameters of the random forest model are obtained, the largest read voltage among the importance evaluation parameters is selected, and the read voltage whose importance evaluation parameters exceed the threshold is selected.
[0043] When the random forest algorithm provided by the present invention is used, the random forest algorithm is a flexible and easy-to-use machine learning algorithm, which can obtain good results in most cases even without hyperparameter tuning. It is also one of the most commonly used algorithms because it is simple and can be used for both classification and regression tasks; it can select the largest read voltage among the importance evaluation parameters, and can also select the read voltage whose importance evaluation parameters exceed the threshold.
[0044] like Figure 1 As shown, the flash memory block performs N operations, where T c ,t1,t2,…,t N Determined based on technician's experience.
[0045] The flash memory block provided by the present invention is repeatedly operated N times, wherein T c ,t1,t2,…,t N According to the experience of technicians, t1, t2, ..., t N The values can be the same or different.
[0046] like Figure 3 As shown, the optimal voltage refers to the read voltage corresponding to the position of the trough of the threshold voltage distribution, which is related to the state of the flash memory block and no clear value is set.
[0047] The optimal voltage provided by the present invention has no clear value when used, and is related to the state of the flash memory block. The specific embodiments are:
[0048] Take MLC as an example. Figure 3The solid line in the middle is the threshold voltage distribution of the newly written data. A set of read voltages corresponds to three voltage values, that is, the voltage value of R1 is V a , the voltage value of R2 is V b , the voltage value of R3 is V c ; When the state of the flash block changes, the threshold voltage distribution will also change. Assuming that the threshold voltage distribution changes to the dotted line, use [V a ,V b ,V c ] To read, it is necessary to find the optimal reading voltage, that is, the voltage at several valleys of the dotted line distribution.
[0049] The threshold voltage distribution corresponding to different states of the flash memory is different. It may shift to the left or to the right. The optimal read voltage will also change accordingly. The read voltage at the valley position can be found as the optimal read voltage.
[0050] like Figure 2 As shown, the flash memory particles of the flash memory block adopt one or more operating modes including SLC, MLC, QLC and PLC.
[0051] When the operating mode provided by the present invention is in use, different types of flash memory particles have different numbers of read voltages in each group. For SLC, each group of read voltages has 1 value, for MLC, each group of read voltages has 3 values, for TLC, each group of read voltages has 7 values, for QLC, each group of read voltages has 15 values, and for PLC, each group of read voltages has 31 values. This scheme is applicable to different types of flash memory particles. As the capacity of flash memory particles increases, QLC and PLC particles appear one after another, and the number of bits in each cell of QLC and PLC particles is 4 and 5, respectively.
[0052] Working principle: In the prior art, the reliability of flash memory particles is improved by offsetting the impact of temperature differences or uneven temperatures. However, the influencing factors usually also include temperature and the time the flash memory particles are placed at the temperature. When the flash memory particles are not powered on, the current temperature and the time the flash memory particles are placed at the temperature cannot be obtained, and these two factors cannot be directly used for quantitative measurement. Due to the lack of effective data retention measurement indicators, the impact of data retention cannot be effectively reduced, thereby affecting the reliability of the flash memory particles.
[0053] When the method for reducing the influence of data retention provided by the present invention is used, firstly a group of flash memory block groups is selected, the flash memory block group includes multiple flash memory blocks, and the number of erasure and writing times of the selected flash memory blocks are all 100; then after writing data at room temperature, the flash memory blocks are placed in a temperature box, and the temperature of the temperature box is T c After the placement time t1, each flash memory block extracts a set of optimal read voltages at room temperature. The extraction block diagram is shown in Figure 2, each word line of each flash memory block has a set of optimal read voltages. The optimal read voltage of a characteristic word line can be selected as the optimal read voltage of the flash memory block, or a set of voltages suitable for most word lines of the flash memory block can be extracted as the optimal voltage through the intelligent optimization algorithm; after being placed in the temperature box for time t2, each flash memory block extracts a set of optimal read voltages at room temperature; repeat the operation N times, where T c ,t1,t2,…,t N According to the experience of technicians, t1, t2, ..., t N The values of can be the same or different; then the most relevant features are selected through machine learning algorithms. Different placement times in the incubator indicate different degrees of data retention. Data retention can be divided into different categories according to the placement time, i.e., placement time t1, t2, …, t N They correspond to categories 1, 2, ..., N respectively; the machine learning algorithm takes the random forest algorithm as an example, takes the optimal read voltage as the input of the random forest, and the category as the output of the random forest. After the model is trained, the importance evaluation parameters of the random forest model are obtained, and the largest read voltage in the importance evaluation parameters is selected, or the read voltage whose importance evaluation parameters exceed the threshold is selected; finally, the selected read voltage is used as a measurement indicator of data retention, and technicians can use this indicator to judge the degree of data retention, thereby determining whether to move the data in the flash memory block; this technical solution uses a machine learning algorithm to select a read voltage that has a great impact on data retention as a feature value, thereby reducing the impact of data retention on the reliability of flash memory particles, while screening out redundant or irrelevant features, reducing the time to obtain the read voltage, and reducing the impact on performance.
[0054] A simulated annealing algorithm is used. The solid is heated to a sufficiently high temperature and then slowly cooled. When heated, the particles inside the solid become disordered as the temperature rises, and the internal energy increases. When slowly cooled, the particles gradually become ordered and reach an equilibrium state at each temperature. Finally, the ground state is reached at room temperature, and the internal energy is reduced to a minimum. A set of voltages suitable for most word lines of the flash memory block is extracted as the optimal voltage, where the optimal voltage refers to the read voltage corresponding to the position of the trough of the threshold voltage distribution, which is related to the state of the flash memory block and has no clear value.
[0055] The random forest algorithm is a flexible and easy-to-use machine learning algorithm that can obtain good results in most cases even without hyperparameter tuning. It can select the largest read voltage among the importance evaluation parameters, and can also select the read voltage whose importance evaluation parameters exceed the threshold.
[0056] The above shows and describes the basic principles, main features and advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention. The scope of protection of the present invention is defined by the attached claims and their equivalents.
Claims
1. A method for reducing the impact on flash memory data retention, characterized in that: The following steps are involved: S1: Select a group of flash memory block groups, the flash memory block group includes multiple flash memory blocks, and the number of erasure and write times of the selected flash memory blocks is 100; S2: After writing data at room temperature, place it in a temperature box at a temperature of T c After the storage time t1, each flash memory block extracts a set of optimal read voltages at room temperature; after the storage time t2 in the temperature box, each flash memory block extracts a set of optimal read voltages at room temperature; S3: The most relevant features are selected through machine learning algorithms. Different placement times in the incubator indicate different degrees of data retention. Data retention is divided into different categories according to the placement time, i.e., placement time t1, t2, …, t N Corresponding to categories 1, 2, ..., N respectively; S4: The selected read voltage is used as a measure of data retention. The technician uses this measure to judge the degree of data retention and thus determines whether to move the data in the flash memory block.
2. The method for reducing the impact on flash memory data retention according to claim 1, characterized in that: Each word line of each flash memory block in S2 has a set of optimal read voltages, and the optimal read voltage of a characteristic word line is selected as the optimal read voltage of the flash memory block; or a set of voltages suitable for most word lines of the flash memory block are extracted as the optimal voltages through an intelligent optimization algorithm.
3. The method for reducing the impact on flash memory data retention according to claim 2, characterized in that: The intelligent optimization algorithm adopts one of a simulated annealing algorithm and a particle swarm optimization algorithm.
4. The method for reducing the impact on flash memory data retention according to claim 3, characterized in that: The algorithm in S3 adopts the forest algorithm, takes the optimal read voltage as the input of the random forest, and the category as the output of the random forest. After the model is trained, the importance evaluation parameters of the random forest model are obtained, the largest read voltage among the importance evaluation parameters is selected, and the read voltage whose importance evaluation parameters exceed the threshold is selected.
5. The method for reducing the impact on flash memory data retention according to claim 4, characterized in that: The flash memory blocks all perform N operations, where T c ,t1,t2,…,t N Determined based on technician's experience.
6. The method for reducing the impact on flash memory data retention according to claim 5, characterized in that: The optimal voltage refers to the read voltage corresponding to the position of the valley of the threshold voltage distribution, which is related to the state of the flash memory block and no clear value is set.
7. The method for reducing the impact on flash memory data retention according to claim 6, characterized in that: The flash memory particles of the flash memory block adopt one or more operating modes including SLC, MLC, QLC and PLC.