A Method for Optimizing Flash Memory Operation Strategies Based on Reliability Awareness and a Flash Memory System

Abstract

The present invention discloses a method for optimizing flash memory operation strategies based on reliability perception and a flash memory system, belonging to the field of chip storage technology. The method can perceive factors affecting the reliability of storage units and determine corresponding agents of different levels by obtaining characteristic parameters of current storage units to determine the reliability level of current storage units and their corresponding failure probabilities. The selected agent generates corresponding operation strategies according to the current state-action value function and the characteristic parameters of the storage unit, and can adjust the operation fault tolerance strategy in real time according to the error characteristics of the storage unit during the storage process, so as to adapt to the reliability changes of flash memory chips under different application scenarios with low overhead, prevent data failure and ensure data storage reliability. After this task is executed, the initial state-action value function for the next task execution is also determined to optimize the flash memory operation of the entire chip during its life cycle, improving the reliability of the flash memory while extending the chip life.

Description

Technical Field

[0001] The present invention belongs to the technical field of chip storage, and more specifically, relates to a method for optimizing flash memory operation strategies based on reliability perception and a flash memory system. Background Art

[0002] Flash memory is a form of electronically erasable programmable read-only memory that allows memory to be erased or written multiple times during operation. It is mainly used for general data storage and for exchanging and transferring data between computers and other digital products, such as memory cards and USB flash drives. Flash memory is a special type of erasable programmable read-only memory that erases in blocks, i.e., data is not lost when the power is off.

[0003] During the operation of the storage system, as the number of write / erase cycles increases, the reliability of the flash memory chip decreases. If storage units with low reliability are continuously operated, problems such as data loss or even damage may occur. It is necessary to design corresponding operation strategies to detect and correct errors in the memory to avoid storage data failure. To avoid data failure problems in the memory, the controller module usually uses methods such as error correction codes, read reference voltage optimization, and bad block management to ensure that the storage units can correctly store data information.

[0004] However, during data storage, the reliability of the storage units is affected by various factors. For example, when judging the reliability of the storage units based on the number of erasure times, it is impossible to optimize the operation strategy according to the actual reliability change of the flash memory chip. When the growth trend of the bit error rate of the storage units exceeds the expected value, data failure problems may also occur during the chip lifetime. Summary of the Invention

[0005] In view of the above-mentioned defects or improvement requirements of the prior art, the present invention provides a method for optimizing flash memory operation strategies based on reliability perception and a flash memory system, aiming to solve the technical problem that data failure may occur during the chip lifetime when the growth trend of the bit error rate of the storage units exceeds the expected value.

[0006] To achieve the above object, according to one aspect of the present invention, a method for optimizing flash memory operation strategies based on reliability perception is provided, including:

[0007] S1: When receiving the current task instruction, initialize the state-action value function;

[0008] S2: Select at least one storage unit from the storage module as the current storage unit; obtain the characteristic parameters of the current storage unit, and determine the reliability level of the current storage unit and its corresponding failure probability according to the characteristic parameters of the current storage unit;

[0009] S3: If the failure possibility indicates that the current storage unit has failed, go to S2; if the failure possibility indicates that the current storage unit is valid, select agents applicable to each level according to the reliability level corresponding to the current storage unit;

[0010] S4: Control the selected agents to generate the current operation policy corresponding to the current task instruction according to the initialized state-action value function and the characteristic parameters of the current storage unit;

[0011] S5: Execute the current operation policy on the current storage unit to obtain the corresponding current execution result;

[0012] S6: Calculate the operation policy return value corresponding to the current execution result, and determine one of the state-action value function corresponding to the current operation policy and the state-action value function initialized for this task as the state-action value function initialized when executing the next task instruction by using the operation policy return value.

[0013] In one embodiment, if the current task instruction is a write operation, then S4 includes:

[0014] Control the selected agents to update the default write mode and the default storage unit mapping table in the default operation policy into the target write mode and the target storage unit mapping table respectively according to the initialized state-action value function and the characteristic parameters of the current storage unit, so as to obtain the current operation policy;

[0015] The current operation policy includes: the target write mode, the target storage unit mapping table, the default read mode, the default configuration offset of the read reference voltage, and the default refresh frequency.

[0016] In one embodiment, the target storage unit mapping table is generated by the selected agents according to the storage unit characteristic parameters to determine the write priority and then in accordance with the write priority.

[0017] In one embodiment, S5 includes:

[0018] If the current task instruction is a write operation, re-select the current storage unit, and execute the write operation instruction corresponding to the target write mode according to the target storage unit mapping table for the re-selected current storage unit.

[0019] In one embodiment, if the current task instruction is a refresh operation, then S4 includes: Control the selected agents to update the default refresh frequency in the default operation policy into the refresh frequency according to the initialized state-action value function and the characteristic parameters of the current storage unit, so as to obtain the current operation policy;

[0020] The current operation strategy includes: a default write mode, a default storage unit mapping table, a default read mode, a default configuration offset of the read reference voltage, and a target refresh frequency.

[0021] In one embodiment, if the current task instruction is a refresh operation, then S5 includes:

[0022] Select an address according to the default storage unit mapping table and execute the write operation instruction corresponding to the default write mode on the current storage unit;

[0023] Execute a refresh operation on the current storage unit according to the target refresh frequency;

[0024] Execute a read operation instruction according to the default read mode and perform error correction; when the error correction code capacity is not exceeded, record the characteristic parameters of the current storage unit after error correction; when the error correction code capacity is exceeded, perform a reread operation on the current storage unit according to the default configuration offset of the read reference voltage until there is no error data, and record the characteristic parameters of the current storage unit after error correction.

[0025] In one embodiment, if the current task instruction is a refresh operation, then S4 includes: controlling the selected agent to update the default read mode and the default configuration offset of the read reference voltage in the default operation strategy to a target read mode and a target configuration offset of the read reference voltage respectively according to the initialized state-action value function and the characteristic parameters of the current storage unit, so as to obtain the current operation strategy;

[0026] The current operation strategy includes: a default write mode, a default storage unit mapping table, a target read mode, a target configuration offset of the read reference voltage, and a default refresh frequency.

[0027] In one embodiment, if the current task instruction is a refresh operation, then S5 includes:

[0028] Execute a read operation instruction according to the target read mode and perform error correction;

[0029] When the error correction code capacity is not exceeded, record the characteristic parameters of the current storage unit after error correction;

[0030] When the error correction code capacity is exceeded, perform a reread operation on the current storage unit according to the target configuration offset of the read reference voltage until there is no error data, and record the characteristic parameters of the current storage unit after error correction.

[0031] In one embodiment, S6 includes: using the formula Calculate the operation strategy return value corresponding to the current execution result R i; where Er i represents the original bit error rate under the state - action value function corresponding to the current operation strategy, Er i- 1 represents the original bit error rate under the state - action value function obtained by initializing the current task;

[0032] When R i = K1, the state - action value function corresponding to the current operation strategy is used as the state - action value function initialized when executing the next task instruction;

[0033] When R i = K2, the state - action value function initialized for this task is used as the state - action value function initialized when executing the next task instruction.

[0034] According to another aspect of the present invention, a flash memory system is provided, including: a storage module and a processor;

[0035] The storage module includes a plurality of memory cells;

[0036] The processor is connected to the storage module and includes, connected in sequence: a controller module, a feature parameter collection module, a reliability perception module, an agent cluster, and a value function update module;

[0037] The controller module is configured to initialize the state - action value function when receiving a current task instruction; and select at least one storage cell from the storage module as the current storage cell;

[0038] The feature parameter collection module is configured to obtain the feature parameters of the current storage cell selected by the controller module and transmit them to the reliability perception module;

[0039] The reliability perception module is configured to determine the reliability level of the current storage cell and its corresponding failure probability according to the feature parameters of the current storage cell; if the failure probability indicates that the current storage cell fails, it feeds back to the controller module to re - select the current storage cell and determine its failure probability until the selected current storage cell is valid, and select the agents applicable to each level according to the reliability level corresponding to the current storage cell;

[0040] The selected agent is configured to generate the current operation strategy corresponding to the current task instruction according to the initialized state - action value function and the feature parameters of the current storage cell, and transmit it to the controller module;

[0041] The controller module is further configured to execute the current operation policy on the current storage unit, obtain the current execution result corresponding to the current task instruction, and transmit it to the value function update module;

[0042] The value function update module is configured to calculate the operation policy return value according to the current execution result corresponding to the current task instruction, determine the state-action value function initialized when the flash memory system executes the next task instruction by using the operation policy return value, and store it in the controller module

[0043] Another embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the above method are implemented.

[0044] Generally speaking, compared with the prior art through the above technical solutions conceived by the present invention, the following beneficial effects can be achieved:

[0045] (1) The present invention provides a method for optimizing flash memory operation policies based on reliability perception, which obtains the characteristic parameters of the current storage unit, and determines the reliability level of the current storage unit and its corresponding failure probability according to the characteristic parameters of the current storage unit, can perceive the factors affecting the reliability of the storage unit, and then determine the corresponding intelligent agent; the intelligent agent generates the corresponding operation policy according to the current state-action value function and the characteristic parameters of the storage unit, and can adjust the operation fault tolerance policy in real time according to the error characteristics of the storage unit during the storage process, so as to adapt to the reliability change of the flash memory chip under different application scenarios with lower overhead, prevent data failure and ensure data storage reliability; after the execution of this task, the state-action value function used for initialization when the next task is executed is also determined, which can extend the chip life while improving the reliability of the flash memory. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 It is a flowchart of a method for optimizing flash memory operation policies based on reliability perception provided by an embodiment of the present invention.

[0047] Figure 2 It is a schematic structural diagram of a reliability perception module provided by an embodiment of the present invention.

[0048] Figure 3 It is a correspondence diagram between storage units and intelligent agents provided by an embodiment of the present invention.

[0049] Figure 4 It is a schematic structural diagram of an operation policy generation intelligent agent provided by an embodiment of the present invention.

[0050] Figure 5 It is a flowchart of the update of an intelligent agent provided by an embodiment of the present invention.

[0051] Figure 6 The structural schematic diagram of the flash memory system provided by an embodiment of the present invention. Detailed implementation manners

[0052] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0053] This embodiment provides an optimization method for flash memory operation strategies based on reliability perception, as Figure 1 shown, including: S1 - S6. Among them, S1: When receiving the current task instruction, initialize the state - action value function. S2: Select at least one storage unit from the storage module as the current storage unit; obtain the characteristic parameters of the current storage unit, where the characteristic parameters of the storage unit are the number of erase - write cycles and the original bit error rate experienced by the storage unit, and determine the reliability level of the current storage unit and its corresponding failure probability according to the characteristic parameters of the current storage unit; among them, the storage units can be grouped according to the characteristic parameters, and the corresponding characteristic parameter values of each group are different, and the grouping method is shown in the table. S3: If the failure probability indicates that the current storage unit fails, then transfer to S2; if the failure probability indicates that the current storage unit is effective, then select the agents applicable to each level according to the reliability level corresponding to the current storage unit. S4: Control the selected agent to generate the current operation strategy corresponding to the current task instruction according to the initialized state - action value function and the characteristic parameters of the current storage unit. S5: Execute the current operation strategy on the current storage unit to obtain the corresponding current execution result. S6: Calculate the operation strategy return value corresponding to the current execution result, and determine one from the state - action value function corresponding to the current operation strategy and the state - action value function initialized for this task as the state - action value function initialized when executing the next task instruction.

[0054] Among them, in S2, the characteristic parameters of the selected storage unit are read, and the original bit error rate is taken as an example for description, and the failure probability is judged by using the read original bit error rate. The structure of the reliability perception module is as Figure 2 shown. The process of the reliability perception module judging the failure probability is as follows:

[0055] (1) The reliability perception module receives the original bit error rate, and the reliability level classifier determines the reliability level of the storage unit: if the original bit error rate value is in the range [0.001*(n - 1), 0.001*n) (n = 1, 2, 3, 4, 5), the reliability level of the target storage unit is i; if the original bit error rate is in the range [0.005, ∞), the reliability level of the target storage unit is 6.

[0056] (2) The reliability perception module determines whether the target storage unit is likely to fail after experiencing 1000*m (m = 1, 2, 3) write / erase cycles and a refresh interval of d hours (d = 720, 2160, 3600) according to the reliability level. For example, the reliability perception module is implemented through a decision tree. The input of the decision tree is the reliability level of the target storage unit, the possible write / erase cycles 1000*m, and the possible storage time of d hours; the output is whether it fails under the input write / erase cycles and storage time.

[0057] (3) The reliability perception module outputs the judgment result. The storage system allocates agents according to the reliability level of the storage unit output by the reliability perception module. The correspondence between the agent and the storage unit follows the Figure 3 corresponding relationship shown. For example, the agent allocated to the storage unit with reliability level 1 is the agent belonging to group 1.

[0058] Among them, the structure of the agent is as shown in Figure 4 The agent is implemented in the form of a Q-table. The input of the agent state is the reliability level of the target storage unit; the actions of the agent are the write priority p (p = 1, 2, 3), TLC write mode, SLC write mode, read reference voltage step s (s = 1, 5, 10), refresh interval d, TLC read mode, and SLC read mode. The specific process of controlling the selected agent to generate the current task instruction corresponding to the current operation strategy in S4 is as follows:

[0059] (1) The agent reads the reliability level;

[0060] (2) Select the corresponding action combination in the Q-table as the operation strategy of the target storage unit through the e-greedy algorithm. The process of the e-greedy algorithm is: take a random number between 0 and 1. If the random number is less than or equal to the preset threshold 0.1, randomly select an action combination from the action combinations corresponding to the state as the operation strategy; otherwise, select the action combination with the largest value in the Q-table as the operation strategy.

[0061] (3) Update the storage unit mapping table according to the write priority p, and generate the mapping table rule as follows: sort the physical addresses in ascending order of the p value.

[0062] In one embodiment, if the current task instruction is a write operation, S4 includes: controlling the selected agent to update the default write mode and the default storage unit mapping table in the default operation policy into a target write mode and a target storage unit mapping table respectively according to the initialized state-action value function and the characteristic parameters of the current storage unit, so as to obtain the current operation policy; wherein, the current operation policy includes: the target write mode, the target storage unit mapping table, the default read mode, the default configuration offset of the read reference voltage, and the default refresh frequency.

[0063] In one embodiment, the target storage unit mapping table is generated by the selected agent according to the storage unit characteristic parameters to determine the write priority and then according to the write priority.

[0064] In one embodiment, S5 includes: if the current task instruction is a write operation, reselecting the current storage unit, and executing the write operation instruction corresponding to the target write mode for the reselected current storage unit according to the target storage unit mapping table.

[0065] In one embodiment, if the current task instruction is a refresh operation, S4 includes: controlling the selected agent to update the default refresh frequency in the default operation policy into a refresh frequency according to the initialized state-action value function and the characteristic parameters of the current storage unit, so as to obtain the current operation policy; wherein, the current operation policy includes: the default write mode, the default storage unit mapping table, the default read mode, the default configuration offset of the read reference voltage, and the target refresh frequency.

[0066] In one embodiment, if the current task instruction is a refresh operation, S5 includes: selecting an address according to the default storage unit mapping table to execute the write operation instruction corresponding to the default write mode for the current storage unit; executing a refresh operation on the current storage unit according to the target refresh frequency; executing a read operation instruction according to the default read mode and performing error correction; when not exceeding the error correction code capability, recording the characteristic parameters of the current storage unit after error correction; when exceeding the error correction code capability, performing a reread operation on the current storage unit according to the default configuration offset of the read reference voltage until there is no error data, and recording the characteristic parameters of the current storage unit after error correction.

[0067] Among them, the process of executing the read-retry operation is as follows:

[0068] (1) Determine whether the number of original errors in a 2KB codeword exceeds the error correction capability of the error correction code, which is 100 bits. If it exceeds 100 bits, go to step (2); otherwise, after decoding, transmit the data to the system and continue to read the next address;

[0069] (2) Configure the read reference voltage offset step size s;

[0070] (3) Perform a read operation, decode and determine whether the number of original errors exceeds the error correction code capability. If it exceeds, change the value of the step size s and go to step (2). If it does not exceed, end the reread operation and record the value of the current step size s.

[0071] After completing the reread operation, record the number of write / erase cycles Cpe and the original bit error rate Er of the storage cell under the current operation strategy Si.

[0072] In one embodiment, if the current task instruction is a refresh operation, S4 includes: controlling the selected agent to update the default read mode and the default configuration offset of the read reference voltage in the default operation strategy to the target read mode and the target configuration offset of the read reference voltage respectively according to the initialized state-action value function and the characteristic parameters of the current storage cell, so as to obtain the current operation strategy; the current operation strategy includes: the default write mode, the default storage cell mapping table, the target read mode, the target configuration offset of the read reference voltage, and the default refresh frequency.

[0073] In one embodiment, if the current task instruction is a refresh operation, S5 includes: executing a read operation instruction according to the target read mode and performing error correction; when the error correction code capability is not exceeded, record the characteristic parameters of the current storage cell after error correction; when the error correction code capability is exceeded, perform a reread operation on the current storage cell according to the target configuration offset of the read reference voltage until there is no error data, and record the characteristic parameters of the current storage cell after error correction.

[0074] In one embodiment, S6 includes: calculating using the formula the operation strategy return value corresponding to the current execution result R i ; where Er i represents the original bit error rate under the state-action value function corresponding to the current operation strategy, Er i- 1 represents the original bit error rate under the state-action value function obtained by initializing the current task; where, as Figure 5 shown, when R i = K1, use the state-action value function corresponding to the current operation strategy as the state-action value function initialized when executing the next task instruction; when R i = K2, use the state-action value function initialized for this task as the state-action value function initialized when executing the next task instruction.

[0075] Among them, the agent update algorithm adopted is the SARSA algorithm. The update principle of the SARSA algorithm is as follows:

[0076] ;

[0077] ;

[0078] Where Q represents the value of the agent's Q - table, S represents the state set, A represents the action set, α represents the learning rate, R represents the reward value, and γ represents the discount factor. In this embodiment, the value of the learning rate α is 0.2, and the value of the discount factor γ is 0.1.

[0079] Another embodiment of the present invention also provides a flash memory system, as Figure 6 shown, including: a storage module and a processor; wherein, the storage module includes a plurality of memory cells; the processor is connected to the storage module and includes, connected in sequence: a controller module, a feature parameter collection module, a reliability awareness module, an agent cluster, and a value function update module.

[0080] Among them, the controller module is used to initialize the state - action value function when receiving the current task instruction; and select at least one storage cell from the storage module as the current storage cell; the feature parameter collection module is used to obtain the feature parameters of the current storage cell selected by the controller module and transmit them to the reliability awareness module; the reliability awareness module is used to determine the reliability level of the current storage cell and its corresponding failure probability according to the feature parameters of the current storage cell; if the failure probability indicates that the current storage cell fails, it feeds back to the controller module to re - select the current storage cell and determine its failure probability until the selected current storage cell is valid, and select the agents applicable to each level according to the reliability level corresponding to the current storage cell; the selected agent is used to generate the current operation strategy corresponding to the current task instruction according to the initialized state - action value function and the feature parameters of the current storage cell, and transmit it to the controller module; the controller module is further used to execute the current operation strategy on the current storage cell, obtain the current execution result corresponding to the current task instruction, and transmit it to the value function update module; the value function update module is used to calculate the operation strategy reward value according to the current execution result corresponding to the current task instruction, determine the state - action value function initialized when the flash memory system executes the next task instruction by using the operation strategy reward value, and store it in the controller module.

[0081] The feature parameter collection module includes a buffer unit, a comparator, and a counter. When collecting feature parameters, the controller performs a write operation on the target storage unit to write data. Meanwhile, the buffer unit caches the written data. Then the controller reads the data in the storage unit. The comparator compares the data in the buffer unit with the read data, and the counter records the number of bits where the read data is inconsistent with the data in the buffer unit. The feature parameter collection module calculates the original bit error rate and updates the number of write / erase cycles. The calculation formula for the original bit error rate is as follows: Original bit error rate = Number of error bits / Number of written data bits.

[0082] The reliability perception module receives the number of write / erase cycles and the original bit error rate collected by the feature parameter collection module, and judges the reliability level of the target storage unit and the failure possibility of the storage unit.

[0083] The intelligent agent cluster includes multiple intelligent agents. Each intelligent agent is responsible for generating operation strategies for storage units with corresponding reliability levels, specifically including: write operation strategies, read operation strategies, and refresh strategies. Among them, for the storage unit mapping table, it reads the write priority p output by the read intelligent agent and generates the mapping relationship between the data address and the physical storage unit in ascending order of the p value. Each group of storage units corresponds to the same mapping table.

[0084] The controller module is responsible for executing write and read instructions according to the operation strategies; when an uncorrectable error occurs, the controller executes the re-read operation instruction until the number of uncorrectable errors is 0 or the user terminates the read data operation; when the data retention time of the storage unit reaches the refresh operation interval d, the controller reads the data saved in the storage unit and rewrites it into the storage unit.

[0085] Another embodiment of the present invention also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the above method are implemented.

[0086] Those skilled in the art can easily understand that the above are only the preferred embodiments of the present invention and are not used to limit the present invention. Any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A flash memory operation strategy optimization method based on reliability perception, characterized in that: include: S1: When receiving the current task instruction, initialize the state-action value function; S2: Select at least one storage unit from the storage module as the current storage unit; Acquiring characteristic parameters of the current storage unit, and determining a reliability level of the current storage unit and a corresponding failure probability thereof according to the characteristic parameters of the current storage unit; S3: If the failure possibility indicates that the current storage unit is failed, then proceed to S2; if the failure possibility indicates that the current storage unit is valid, then select an agent applicable to each level according to the reliability level corresponding to the current storage unit; S4: Control the selected agent to generate the current task instruction corresponding to the current operation strategy according to the initialized state-action value function and the characteristic parameters of the current storage unit; S5: Execute the current operation strategy on the current storage unit to obtain a corresponding current execution result; S6: Calculate the operation strategy reward value corresponding to the current execution result, and use the operation strategy reward value to determine a state-action value function from the state-action value function corresponding to the current operation strategy and the state-action value function initialized for this task as the state-action value function to be initialized when executing the next task instruction.

2. The flash memory operation strategy optimization method based on reliability perception as claimed in claim 1, characterized in that: If the current task instruction is a write operation, then S4 includes: Control the selected agent to update the default write mode and the default storage unit mapping table in the default operation strategy to the target write mode and the target storage unit mapping table respectively according to the initialized state-action value function and the characteristic parameters of the current storage unit, so as to obtain the current operation strategy; The current operation strategy includes: a target write mode, a target storage unit mapping table, a default read mode, a default configuration offset of a read reference voltage, and a default refresh frequency.

3. The flash memory operation strategy optimization method based on reliability perception as claimed in claim 2, characterized in that: The target storage unit mapping table is generated by the selected agent determining the write priority according to the storage unit characteristic parameters and then generating the target storage unit mapping table according to the write priority.

4. The flash memory operation strategy optimization method based on reliability perception as claimed in claim 2, characterized in that: The S5 includes: When the current task instruction is a write operation, the current storage unit is selected for a second time, and an address is selected according to the target storage unit mapping table for the second selected current storage unit to execute a write operation instruction corresponding to a target write mode.

5. The flash memory operation strategy optimization method based on reliability perception as claimed in claim 1, characterized in that: If the current task instruction is a refresh operation, S4 includes: controlling the selected agent to update the default refresh frequency in the default operation strategy to the refresh frequency according to the initialized state-action value function and the characteristic parameters of the current storage unit to obtain the current operation strategy; The current operation strategy includes: a default write mode, a default storage unit mapping table, a default read mode, a default configuration offset of a read reference voltage, and a target refresh frequency.

6. The flash memory operation strategy optimization method based on reliability perception as claimed in claim 5, characterized in that: If the current task instruction is a refresh operation, S5 includes: Selecting an address according to the default storage unit mapping table to execute a write operation instruction corresponding to the default write mode on the current storage unit; Performing a refresh operation on the current storage unit according to the target refresh frequency; Execute a read operation instruction and perform error correction according to the default read mode; when the error correction code capability is not exceeded, record the characteristic parameters of the current storage unit after error correction; when the error correction code capability is exceeded, perform a reread operation on the current storage unit according to the default configuration offset of the read reference voltage until there is no error data, and record the characteristic parameters of the current storage unit after error correction.

7. The flash memory operation strategy optimization method based on reliability perception as claimed in claim 1, characterized in that: If the current task instruction is a refresh operation, S4 includes: controlling the selected agent to update the default read mode and the default configuration offset of the read reference voltage in the default operation strategy to the target read mode and the target configuration offset of the read reference voltage according to the initialized state-action value function and the characteristic parameters of the current storage unit, respectively, to obtain the current operation strategy; The current operation strategy includes: a default write mode, a default storage unit mapping table, a target read mode, a target configuration offset of a read reference voltage, and a default refresh frequency.

8. The flash memory operation strategy optimization method based on reliability perception as claimed in claim 7, characterized in that: If the current task instruction is a refresh operation, S5 includes: Executing a read operation instruction and performing error correction according to the target read mode; When the error correction code capability is not exceeded, recording the characteristic parameters of the current storage unit after error correction; When the error correction code capability is exceeded, a reread operation is performed on the current storage unit according to the target configuration offset of the read reference voltage until there is no error data, and characteristic parameters of the current storage unit after error correction are recorded.

9. The flash memory operation strategy optimization method based on reliability perception according to any one of claims 1 to 8, characterized in that: The S6 includes: Using the formula Calculate the return value of the operation strategy corresponding to the current execution result R i ;in, Er i represents the original bit error rate under the state-action value function corresponding to the current operation strategy, Er i- 1 Represents the original bit error rate under the state-action value function obtained by initializing the current task; when R i = K1, the state-action value function corresponding to the current operation strategy is used as the state-action value function initialized when executing the next task instruction; when R i = K2, the state-action value function initialized for this task is used as the state-action value function initialized when executing the next task instruction.

10. A flash memory system, characterized in that: include: A storage module including a plurality of memory cells; A processor connected to the storage module, including: a controller module, a characteristic parameter collection module, a reliability perception module, an intelligent agent cluster and a value function update module connected in sequence; The controller module is used to initialize the state-action value function when receiving the current task instruction; and select at least one storage unit from the storage module as the current storage unit; The characteristic parameter collection module is used to obtain the characteristic parameters of the current storage unit selected by the controller module and transmit them to the reliability perception module; The reliability perception module is used to determine the reliability level of the current storage unit and its corresponding failure probability according to the characteristic parameters of the current storage unit; if the failure probability indicates that the current storage unit is failed, feedback is given to the controller module so that it reselects the current storage unit and determines its failure probability until the selected current storage unit is valid, and selects the intelligent agent applicable to each level according to the reliability level corresponding to the current storage unit; The selected agent is used to generate the current task instruction corresponding to the current operation strategy according to the initialized state-action value function and the characteristic parameters of the current storage unit, and transmit it to the controller module; The controller module is further used to execute the current operation strategy on the current storage unit, obtain the current execution result corresponding to the current task instruction, and transmit it to the value function update module; The value function update module is used to calculate the operation strategy reward value according to the current execution result corresponding to the current task instruction, use the operation strategy reward value to determine the state-action value function initialized when the flash memory system executes the next task instruction, and store it in the controller module.

Images