Read voltage adjustment method and storage device
By identifying the target block read for the first time after idle recovery, and combining state attributes to predict the probability of temporary read errors, the reading voltage is dynamically adjusted, thus solving the problem of temporary read errors in 3D NAND Flash memory cells and improving reading efficiency and resource utilization efficiency.
Patent Information
- Application Number
- CN202611357322.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-03
- Publication Date
- 2026-11-10
- Estimated Expiration
- 2046-09-03
AI Technical Summary
In existing technologies, there is a problem of temporary read errors when 3D NAND Flash storage cells recover after a long period of idle time. Existing precharge operations lack effective risk assessment and parameter configuration adaptability, resulting in decreased read efficiency and wasted resources.
By identifying the target block for the first read after idle recovery, and combining the block's state attributes to predict the probability of temporary read errors, the read voltage is dynamically adjusted to optimize the pre-charge parameters. The optimal read voltage adjustment level is adopted and the adjustments within the plane group are merged to achieve targeted and efficient read operations.
It improves the read efficiency of storage devices, reduces temporary read errors, optimizes energy consumption and latency costs, adapts to different block characteristics and operating conditions, and avoids resource waste.
Smart Images

Figure CN122842639B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of memory control, and more particularly to a read voltage adjustment method and a memory device. Background Technology
[0002] As the number of stacked layers in 3D NAND Flash increases, the coupling effect between memory cells and the impact of polysilicon grain boundary defects in the channel on data reliability becomes increasingly significant. In practice, it has been found that when a 3D NAND Flash performs its first read after recovering from a long period of idle time, the number of error bits increases abnormally; this phenomenon is known as temporary read errors. The physical root cause lies in the continuous leakage of trap charges in the memory cells during the idle period, leading to a temporary shift in the threshold voltage distribution, making it impossible for the normal read reference voltage to accurately determine the data status.
[0003] To suppress temporary read errors, existing technologies typically employ a pre-charge operation before the actual read command, accelerating the recovery of the trap charge by extending the pre-charge time or adjusting the pre-charge amplitude. However, the above approach still has shortcomings in practical applications.
[0004] On the one hand, the triggering of precharge operations lacks effective risk assessment criteria, making it difficult to evaluate the likelihood of temporary read errors based on the actual state of the block, resulting in a certain degree of blindness in precharge triggering. On the other hand, the configuration method of precharge parameters is relatively fixed, failing to adapt to the differences in characteristics between different blocks and changes in operating conditions.
[0005] Furthermore, existing solutions primarily focus on the success rate of read operations when determining precharge intensity, neglecting energy consumption and potential latency costs. For memory architectures supporting multi-plane concurrent operations, a lack of coordination mechanisms among the precharge operations of each plane can also lead to a decrease in overall read efficiency, or even over-tuning or resource waste. Summary of the Invention
[0006] In view of this, this application provides a reading voltage adjustment method and a storage device to solve the problem of temporary read errors in the storage device.
[0007] Embodiments of this application provide a read voltage adjustment method applied to a storage device including a memory module and a memory controller. The memory module includes multiple planes, each plane includes multiple blocks, and each block contains multiple pages. The method includes: In response to a read request for multiple pages, the block to which each of the multiple pages belongs is determined, and the block that is read for the first time after being in an idle state is taken as the target block, and the page belonging to the read request in the target block is taken as the target page, wherein being read for the first time after being in an idle state means that the corresponding block in the memory module is read for the first time after being restored from an idle state. Obtain a set of status attributes for each target block, and based on the status attributes, obtain the probability of a temporary read error occurring in each target block; Based on the probability of temporary read errors for each target block and the operational costs of each of the preset multiple read voltage adjustment levels, the optimal read voltage adjustment level for the plane where each target block is located is determined from the multiple read voltage adjustment levels. If there are multiple target blocks in the same plane, the level with the highest intensity among the optimal read voltage adjustment levels corresponding to each target block in the plane is taken as the optimal read voltage adjustment level for the plane. Each of the multiple read voltage adjustment levels corresponds to a different pre-charge voltage amplitude and / or pre-charge duration. If the target block is distributed across multiple planes, then when the preset gear merging condition is met, the multiple planes are divided into at least one plane group according to the gear merging condition, and the optimal read voltage adjustment gears of each plane in the same plane group are merged into a unified read voltage adjustment gear, which is then used as the read voltage adjustment gear to be executed for the plane group. Otherwise, the optimal read voltage adjustment level for each plane will be used as the read voltage adjustment level to be executed. Based on the read voltage adjustment level to be executed, a read command carrying the read voltage adjustment level is sent to the target page in the corresponding plane or plane group, so that the read voltage adjustment operation is performed according to the read voltage adjustment level before the read operation on the target page is performed.
[0008] Embodiments of this application also provide a storage device including a connection interface for electrically connecting to a host computer; A memory module comprising multiple planes, each plane comprising multiple blocks, and each block comprising multiple pages; and The memory controller includes an error checking and correction circuit, a buffer memory, and a memory control circuit, and is electrically connected to the connection interface and the memory module; The memory controller is configured to perform the read voltage adjustment method described above.
[0009] The beneficial effects of this application are: by identifying the target block read for the first time after idle recovery and combining the block's state attributes to predict the probability of temporary read errors, targeted adjustment of the read voltage can be achieved.
[0010] To make the above-mentioned objectives, technical solutions and beneficial effects of this application more apparent and understandable, specific embodiments are described below in conjunction with the accompanying drawings for detailed explanation. Attached Figure Description
[0011] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the embodiments of this application will be briefly introduced below. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort, and these are all within the protection scope of this application.
[0012] Figure 1 This is a schematic diagram of the data storage system shown in the embodiments of this application; Figure 2 This is a schematic diagram of the memory controller shown in the embodiments of this application; Figure 3 This is a schematic diagram of the storage hierarchy structure of the memory module shown in the embodiments of this application; Figure 4 This is a flowchart illustrating the reading voltage adjustment method in an embodiment of this application; Figure 5 This is a schematic diagram of the error bit number update process shown in the embodiments of this application; Figure 6 This is a schematic diagram of the model parameter update process shown in the embodiments of this application; Figure 7 This is a schematic diagram of the optimal reading voltage adjustment range selection process shown in the embodiments of this application; Figure 8 This is a schematic diagram of the gear merging process shown in the embodiments of this application. Detailed Implementation
[0013] Reference will now be made in detail to exemplary embodiments of this application, examples of which are illustrated in the accompanying drawings. Wherever possible, the same element symbols are used in the drawings and description to denote the same or similar parts.
[0014] Figure 1 This is a schematic diagram of a data storage system according to an embodiment of this application. Please refer to... Figure 1The data storage system 10 includes a host 11 and a storage device 12. The storage device 12 can be connected to the host 11 and can be used to store data from the host 11. For example, the host 11 can be a smartphone, tablet computer, laptop computer, desktop computer, industrial computer, automotive infotainment system, advanced driver assistance system (ADAS), game console, server, or computer system installed in a specific carrier (e.g., vehicle, aircraft, or ship), and the type of host 11 is not limited to these. Furthermore, the storage device 12 can include a solid-state drive (SSD), universal flash storage (UFS), embedded multi-media card (eMMC), USB flash drive, memory card, or other types of non-volatile storage devices. It is particularly suitable for fields with stringent requirements for reliability, data integrity, environmental tolerance, and long-term stability, such as industrial control, edge computing, and automotive electronics.
[0015] Storage device 12 includes a connection interface 121, a memory module 122, and a memory controller 123. The connection interface 121 is used to connect storage device 12 to host 11. For example, the connection interface 121 may support embedded multimedia cards, general-purpose flash memory, Peripheral Component Interconnect Express (PCI Express), Non-Volatile Memory Express (NVM express), Serial Advanced Technology Attachment (SATA), Universal Serial Bus (USB), or other types of connection interface standards. Therefore, storage device 12 can communicate with host 11 (e.g., exchange signals, instructions, and / or data) via connection interface 121.
[0016] Memory module 122 is used to store data. For example, memory module 122 may include one or more rewritable non-volatile memory modules. Each rewritable non-volatile memory module may include one or more memory cell arrays. The memory cells in the memory cell array store data in the form of voltage (also known as threshold voltage). For example, memory module 122 may include single-level cell (SLC) NAND flash memory modules, multi-level cell (MLC) NAND flash memory modules, triple-level cell (TLC) NAND flash memory modules, quadruple-level cell (QLC) NAND flash memory modules, and / or other memory modules with the same or similar characteristics. For multi-level memory cells, such as MLC (MultiLevel Cell), TLC (Triple Level Cell), QLC (Quad Level Cell), and PLC (Penta Level Cell), flash memory modules that have undergone special screening and testing can be used to meet the requirements of wide operating temperature range (e.g., -40°C to 105°C or higher), high durability, and data retention.
[0017] Memory controller 123 is connected to connection interface 121 and memory module 122. Memory controller 123 can be considered the control core of storage device 12 and used to control storage device 12. For example, memory controller 123 can be used to control or manage the overall or partial operation of storage device 12. For example, memory controller 123 may include a central processing unit (CPU), or other programmable general-purpose or special-purpose microprocessor, digital signal processor (DSP), programmable controller, application-specific integrated circuit (ASIC), programmable logic device (PLD), or other similar device or combination of these devices. In one embodiment, memory controller 123 may include a flash memory controller. In particular, for automotive functional safety requirements, the architecture of memory controller 123 may include safety islands, dual-core lockstep, or other redundancy and monitoring mechanisms to ensure the reliability of control logic.
[0018] The memory controller 123 can send instruction sequences to the memory module 122 to access the memory module 122. For example, the memory controller 123 can send a write instruction sequence to the memory module 122 to instruct the memory module 122 to store data in a specific memory cell. For example, the memory controller 123 can send a read instruction sequence to the memory module 122 to instruct the memory module 122 to read data from a specific memory cell. For example, the memory controller 123 can send an erase instruction sequence to the memory module 122 to instruct the memory module 122 to erase data stored in a specific memory cell. Furthermore, the memory controller 123 can also send other types of instruction sequences to the memory module 122 to instruct the memory module 122 to perform other types of operations, which are not limited in this application. The memory module 122 can receive instruction sequences from the memory controller 123 and access its internal memory cells according to these instruction sequences.
[0019] Figure 2 This is a schematic diagram of a memory controller according to an embodiment of this application. Please refer to... Figure 1 and Figure 2 The memory controller 123 includes a host interface 21, a memory interface 22, and a memory control circuit 23. The host interface 21 is used to connect to the host 11 via the connection interface 121 for communication with the host 11. The memory interface 22 is used to connect to the memory module 122 for access to the memory module 122.
[0020] The memory control circuit 23 is connected to the host interface 21 and the memory interface 22. The memory control circuit 23 can be used to control or manage the overall or partial operation of the memory controller 123.
[0021] For example, the memory control circuit 23 can communicate with the host 11 via the host interface 21 and access the memory module 122 via the memory interface 22. For example, the memory control circuit 23 may include control circuitry such as an embedded controller or a microcontroller. In the following embodiments, the description of the memory control circuit 23 is equivalent to the description of the memory controller 123.
[0022] In detail, the memory control circuit 23 includes a parallel scheduler, a region configuration table, and a metric calculation unit. The parallel scheduler connects the host interface 21 and the memory interface 22, maintains the ready / busy status of the first and second grain groups, and dispatches garbage collection operation commands and host read / write commands to the corresponding grain groups according to the region configuration table. The metric calculation unit reads metadata such as effective lifespan, migration count, and source classification from the buffer memory 24, and reads necessary fields from the OOB (Out-Of-Band) area to output the data recyclability metric. The region configuration table is stored in the power-loss protected buffer memory 24 and records the region affiliation, operating mode, and cumulative erase / write count for each grain.
[0023] In one embodiment, the memory controller 123 may further include a buffer memory 24. The buffer memory 24 is connected to the memory control circuitry 23 and controlled by the memory controller 123 to cache data. For example, the buffer memory 24 can be used to cache instructions from the host 11, data from the host 11, and / or data from the memory module 122. Specifically, the logical-to-physical mapping table and its higher-level address management unit are typically resident or cached in the buffer memory 24 to support high-speed access and updates. To ensure the reliability and consistency of the mapping table data in the event of sudden events such as abnormal power failures, the buffer memory 24 may employ a storage medium with power loss protection (PLP) characteristics, or incorporate a backup capacitor design to ensure that critical metadata areas have sufficient time to be written into the memory module 122.
[0024] In one embodiment, the memory controller 123 may further include an error checking and correction circuit 25. The error checking and correction circuit 25 is connected to the memory control circuit 23 and is used to encode and decode data to ensure data integrity. For example, the error checking and correction circuit 25 may support various encoding / decoding algorithms such as Low Density Parity Check code (LDPC), BCH code, Reed-solomon code (RS code), and Exclusive OR (XOR) code.
[0025] Figure 3This is a schematic diagram of the storage hierarchy of the memory module in this embodiment. The memory module 122 adopts a multi-level storage array architecture, which is divided into dies, planes, blocks, and pages from the top to the bottom. A die is an independent storage unit that constitutes the memory module 122. A single die can be divided into one or more independent planes. Each plane contains multiple blocks, and a block is the smallest unit for performing erase operations. Each block contains multiple pages, and a page is the smallest unit for performing read and program operations.
[0026] Figure 3 An example is given, in which a single die contains two planes; each plane contains several blocks arranged in an array, each block contains multiple pages, and each page corresponds to a set of contiguous memory cells.
[0027] Figure 4 This is a general flowchart of a read voltage adjustment method according to an embodiment of this application. The method is applied to a storage device 12 including a memory module 122. A memory control circuit 23 or a memory controller 123 is configured to perform this method. The method begins at step S1.
[0028] S1. In response to read requests for multiple pages, determine the blocks to which each page belongs, and designate the block that is being read for the first time after idle recovery from the determined blocks as the target block, and designate the page belonging to the read request in the target block as the target page. Here, "read for the first time after idle recovery" means that the corresponding block in memory module 122 is read for the first time after recovering from an idle state. Obtain a set of status attributes for each target block, and based on these status attributes, determine the probability of a Temporary Read Error (TRE) occurring in each target block.
[0029] Upon receiving a read request for multiple pages from host 11, the system first parses the request. The purpose of parsing is to determine which block each requested page belongs to. Since in practice, multiple pages involved in a read request may be distributed across completely different blocks, it is necessary to categorize the pages according to their respective blocks to determine which block each page belongs to, and then identify the target block within each page's block.
[0030] The method for determining the target block can be as follows: Each block maintains a timestamp of the last operation's end time. Whenever a new read request arrives, the current system time is read, and the difference between the current system time and the last operation's end time for that block is calculated. This difference represents the idle time experienced by the block from the end of the last operation to the current time. This idle time is compared with a preset idle threshold. If the idle time is greater than the preset idle threshold, the block is determined to be in a state after recovering from idle, and the currently received read request is the first read command received by the block after recovering from the idle state. In this case, the block is marked as a target block. If the idle time is less than or equal to the preset idle threshold, the block is determined not to have experienced a sufficiently long idle period and is not marked as a target block. For blocks not marked as target blocks, regular read operations are performed directly on the block without subsequent read voltage adjustment processing. It should be noted that the aforementioned timestamp can be stored in a buffer memory 24 with power loss protection, or after an abnormal power loss is recovered, blocks with an idle duration exceeding a preset idle determination threshold are regarded by default as candidate target blocks to be read for the first time after the idle recovery.
[0031] After the target block is determined, a set of status attributes of the target block is obtained. This set of status attributes includes the idle time experienced by the target block from the end of the last operation to the current time, the average temperature, peak temperature, and equivalent temperature during the idle period, the cumulative number of erase / write operations of the target block, and the exponentially weighted moving average of the historical error bits of the target block.
[0032] The idle time is obtained by subtracting the current system timestamp from the last_op_timestamp field in the block context structure; the temperature attribute is sampled and recorded by the temperature sensor inside the storage device 12 at fixed intervals (e.g., 10ms) during the idle period, and the average or peak value is taken; the number of erase / write operations is read directly from the erase_count field in the block context structure; the error bit count EWMA (Exponentially Weighted Moving Average) is updated and persisted in the block context structure after each read is completed, and the initial value can be the FBC (Fail Bit Count) measured when the block is first read.
[0033] The idle time is obtained directly from the difference between the current system time and the timestamp of the last operation's end time. The average temperature (i.e., the average temperature during the idle period), peak temperature (i.e., the peak temperature), or equivalent temperature calculated based on temperature history is periodically sampled and recorded within the idle time, rather than simply reading the current temperature. The erase / write count is obtained from the erase / write count counter maintained by each memory module 122.
[0034] S2. Based on the probability of temporary read errors of each target block and the operation cost of each of the preset multiple read voltage adjustment levels (RVA Level), determine the optimal read voltage adjustment level for the plane where each target block is located from the multiple read voltage adjustment levels.
[0035] Each of these multiple read voltage adjustment levels corresponds to a different precharge voltage amplitude (PVA) and / or precharge duration (PD). Furthermore, if multiple target blocks exist within the same plane, the highest intensity of the optimal read voltage adjustment level among the optimal read voltage adjustment levels corresponding to each target block within that plane is taken as the optimal read voltage adjustment level for that plane.
[0036] The memory module 122 supports multiple read voltage adjustment levels, denoted by k. Level 0 indicates no read voltage adjustment, while higher levels indicate stronger read voltage adjustment. Different levels correspond to different pre-charge voltage amplitudes and pre-charge durations. Specifically, higher levels result in longer pre-charge pulse durations and larger pre-charge voltage amplitudes, leading to better recovery of trap charges, but also increased energy consumption.
[0037] The voltage adjustment level is a unique index number for multiple pre-configured discrete intensity levels. Each level corresponds to a combination of comprehensive voltage parameters determined by the pre-charge duration and pre-charge voltage amplitude. The higher the level number, the greater the adjustment intensity. Level 0 indicates that no adjustment operation is performed.
[0038] S3. If the target block is distributed across multiple planes, then when the preset level merging condition is met, the multiple planes are divided into at least one plane group (PG) according to the preset level merging condition, and the optimal read voltage adjustment level of each plane in the same plane group is merged into a unified read voltage adjustment level (URVAL), which is then used as the read voltage adjustment level to be executed for the plane group.
[0039] The intensity of the aforementioned read voltage adjustment level is determined based on the pre-charge voltage amplitude and / or pre-charge duration corresponding to that level. The core of Level Merging (LM) lies in level compatibility judgment. If the intensity of the optimal read voltage adjustment level in the first plane is less than or equal to the intensity of the optimal read voltage adjustment level in the second plane, then the level in the second plane is considered compatible with the level in the first plane. Performing read voltage adjustment uniformly with a higher intensity level ensures that the threshold voltage replenishment effect is at least no less than that of a lower intensity level, but it may increase energy consumption. Therefore, subsequent cost checks are needed to determine whether to merge. Planes with mutually compatible levels are grouped into the same plane group.
[0040] For each plane group, if the plane group contains multiple planes, the highest intensity of the optimal read voltage adjustment range among all planes in the plane group is used as the unified read voltage adjustment range. The read voltage adjustment requests from all planes in the plane group are then merged into a single Multi-Plane Read Command (MPRC). This MPRC carries the unified read voltage adjustment range and a plane mask (PM) indicating the planes within the plane group in its extended field. If the plane group contains only a single plane, a single-plane read command is generated, and the optimal read voltage adjustment range for that single plane is carried in its extended field.
[0041] The aforementioned level merging is only performed when the expected total cost (ETC) corresponding to the merged unified read voltage adjustment level is less than or equal to the sum of the expected total costs of each plane executing its own optimal read voltage adjustment level; otherwise, level merging is not performed, and each plane executes separately. When dividing the plane group, it is also necessary to check whether the timing and power consumption constraints are met after level merging. In addition, multiple planes within a plane group correspond to planes that support parallel operation within the same die in memory module 122.
[0042] Timing and power constraints include timing constraints and power constraints. The timing constraint means that the total execution time of the merged read voltage adjustment operation does not exceed the maximum precharge time specified in the datasheet of memory module 122. The power constraint means that the peak current of the merged read voltage adjustment operation does not exceed the currently allocable current budget of memory controller 123.
[0043] S4. Otherwise, the optimal read voltage adjustment level for each plane is taken as the read voltage adjustment level to be executed. This "otherwise" includes situations where the target block is distributed on only one plane, or the target block is distributed on multiple planes but does not meet the level merging condition.
[0044] If the target blocks are determined to be distributed on only one plane (i.e., all target blocks are located in the same plane), then the optimal read voltage adjustment level of that plane is directly used as the read voltage adjustment level to be executed. This handling method is a special case of the multi-plane scenario in step S3, and no level merging or plane grouping operations are required. If the target blocks are distributed on multiple planes but the level merging condition is not met, then each plane uses its own optimal read voltage adjustment level as the read voltage adjustment level to be executed.
[0045] S5. Based on the read voltage adjustment level to be executed, send a read command carrying the read voltage adjustment level to the target page in the corresponding plane or plane group, so as to perform the read voltage adjustment operation according to the read voltage adjustment level before performing the read operation on the target page.
[0046] After receiving the command, the memory module 122 performs a read voltage adjustment operation according to the read voltage adjustment level carried in the command before actually performing the read operation on the target page.
[0047] Specifically, based on the gear parameters, a corresponding pre-charge pulse is applied to the selected word line on the specified plane. This pre-charge pulse has a specific duration and pre-charge voltage amplitude. This operation accelerates the recovery of trapped charge in the memory cell by extending the pre-charge duration and applying a pre-charge voltage pulse of specific amplitude and polarity beyond the standard pre-charge phase, causing the drifted threshold voltage distribution to reappear close to the steady-state operating distribution, thereby physically compensating for the threshold voltage offset caused by charge leakage during idle periods. This read voltage adjustment operation is essentially a charge recapture operation.
[0048] After the read voltage adjustment operation is completed, the memory controller 123 controls the memory module 122 to perform normal read operations on each plane that has undergone the read voltage adjustment operation, reading data from the target page. For planes that directly perform read operations without undergoing read voltage adjustment, the normal read operation is also performed.
[0049] In the memory module 122 that supports multi-plane concurrent operation, if multiple planes have been merged into a single multi-plane read command and a unified level is issued through an extended field, then after the read voltage adjustment operation is completed, these planes can simultaneously perform parallel read operations, thereby compressing the total time of the original serial execution into the time of a single operation.
[0050] In one embodiment, Figure 5 This is a schematic diagram illustrating the error bit count update process provided in an embodiment of this application. Figure 5As shown, after this read operation is completed, the actual read result is obtained, and the number of error bits measured in this hard decision read, FBC_current, is read from this result. The old exponentially weighted moving average of the number of error bits, EWMA_FBC_Old, is read from the block context structure.
[0051] For updating the exponentially weighted moving average of the number of error bits, the new value EWMA_FBC_New is calculated according to Formula 1 and written into the block context structure.
[0052] Formula 1: EWMA_FBC_New=α×FBC_current+(1-α)×EWMA_FBC_Old.
[0053] This update process is now complete. The updated EWMA_FBC will be used as a status attribute to calculate the probability of temporary read errors in subsequent read operations.
[0054] Wherein, EWMA_FBC represents the updated exponentially weighted moving average of the number of error bits, FBC_current represents the number of error bits measured in the first hard decision read after the current recovery from idle, EWMA_FBC_Old represents the updated exponentially weighted moving average of the number of error bits, and α is the smoothing coefficient, with a value ranging from 0.05 to 0.3.
[0055] After obtaining the above status attributes, the probability of a temporary read error occurring in the target block is calculated based on this set of status attributes.
[0056] In an alternative implementation, a logistic regression model is used to process the aforementioned state attribute to obtain the probability.
[0057] First, online normalization is performed on each continuous feature in the state attributes. This involves subtracting the historical mean from each continuous feature and then dividing by the standard deviation to eliminate the impact of differences in dimensions on model convergence. The normalized feature vector is represented as follows: x=[1,Δt_idle_norm,T_norm,PE_cycle_norm,EWMA_FBC_norm], Formula 2; Where Δt_idle_norm represents the normalized idle time, T_norm represents the normalized temperature of memory module 122 during idle period, PE_cycle_norm represents the normalized erase / write cycles, EWMA_FBC_norm represents the normalized exponentially weighted moving average of the number of error bits, and the constant 1 in the first term of the vector is used for the bias term in the corresponding logistic regression model.
[0058] Maintain a parameter vector β=[β0,β1,β2,β3,β4] for a logistic regression model, where β0 is the bias term parameter, and β1 to β4 are the weight parameters corresponding to the exponentially weighted moving average of idle time, idle temperature, number of erase / write cycles, and number of error bits, respectively. Calculate the linear combination value z as follows: z = β0 + β1 × Δt_idle_norm + β2 × T_norm + β3 × PE_cycle_norm + β4 × EWMA_FBC_norm, Formula 3; Then, substituting z into the logistic function, we calculate the probability R of a temporary read error: R = 1 / (1 + exp(-z)). exp is the natural exponential function.
[0059] The value of R ranges from 0 to 1. The closer the R value is to 1, the higher the risk of a temporary read error, and the closer the R value is to 0, the lower the risk.
[0060] In one example, suppose a read request is received, requesting the reading of multiple pages located in Block A and Block B. First, the blocks to which these pages belong are determined, resulting in Block A and Block B. For Block A, the last operation ended at time T1, and the current system time is T2. The idle time is calculated as T2 minus T1, assuming this value is 500 milliseconds. This value is greater than the preset idle threshold of 100 milliseconds, therefore Block A is marked as the target block. For Block B, its idle time is 20 milliseconds, less than 100 milliseconds, therefore it is not marked as a target block, and a normal read is performed on Block B.
[0061] For Block A, which is marked as the target block, the current operating temperature of the memory module 122 is further obtained as 25℃, the number of erase / write cycles of Block A is 3000, the exponentially weighted moving average of the number of error bits of Block A is 15, and the sensitivity score of Block A is 0.3. The idle time of Block A (500 milliseconds), the idle temperature (25℃), the number of erase / write cycles (3000), and the exponentially weighted moving average of the number of error bits (15) are normalized to obtain the normalized feature vector [1, 1.2, -0.5, 0.8, 0.3]. The currently maintained model parameter vector is read as [0.1, 0.5, -0.2, 0.3, 0.6], and z=1.22 is calculated, resulting in R=0.772. This means that the probability of a temporary read error occurring in target Block A is 77.2%, which is a high-risk scenario, requiring read voltage adjustment.
[0062] In one embodiment, after the current read operation is completed, the model parameters are updated online based on the actual read results.
[0063] Figure 6 This is a schematic diagram illustrating the model parameter update process provided in an embodiment of this application. For example... Figure 6 As shown, firstly, the normalized feature vector is obtained. This feature vector includes the normalized idle time, the normalized temperature during the idle period, the normalized number of erase / write operations, and the normalized exponentially weighted moving average of the number of error bits. Simultaneously, the currently maintained logistic regression model parameters are read. Based on the aforementioned feature vector and model parameters, a linear combination value z is calculated. Then, z is substituted into the logistic function to calculate and output the probability R of temporary read errors. After this read operation is completed, the actual read result is obtained and the label value y is determined. When the preset update conditions are met, the model parameters are updated using stochastic gradient descent, and the updated model parameters are used for subsequent calculations. If the update conditions are not met, the process ends directly.
[0064] Specifically, the label value y is determined based on whether the read operation triggered a read retry or whether the number of error bits exceeded a preset hard-decision correction capability threshold. If a read retry was triggered or the number of error bits exceeded the hard-decision threshold, y is set to 1; if a read retry was not triggered and the number of error bits did not exceed the hard-decision threshold, y is set to 0. Then, the model parameters are updated using stochastic gradient descent according to the following update rule: βj = βj - λ × (Ry) × xj.
[0065] Where j iterates from 0 to 4, and λ is the learning rate. The initial value of λ can be set to 0.01, and it slowly decays as the number of erases and writes to the target block increases. To prevent abnormal samples from contaminating the model, the model parameters are only updated when one of the following conditions is met: the difference between the probability R of a temporary read error and the label value y exceeds a preset threshold; or the idle time of the target block exceeds a preset minimum threshold. xj represents the j-th component in the normalized feature vector x described in Formula 2.
[0066] In a preferred embodiment, the decay formula for the learning rate λ is: λ = λ0 / (1 + erase_count / E_decay); where λ0 is the initial learning rate, E_decay is the erase / write decay constant, for example, 500. erase_count is the accumulated "erasure / write count" of the target block, i.e., the number of programming / erasing cycles.
[0067] Furthermore, to reduce the impact of read voltage adjustment on label values, the read result is only used to generate label values when the read voltage adjustment level is 0 or lower than a preset threshold; otherwise, model parameter updates are paused, or corrections are performed using historical unadjusted samples. Specifically, a `read_after_adjust` flag can be maintained in the block context structure. This flag is set when an adjustment with an intensity greater than or equal to a certain level (e.g., level 2) was performed before the current read, and the read result is not included in the model update sample set.
[0068] During the cold start phase of the model, the initial value of the parameter vector β is determined as follows: take multiple typical storage blocks produced in the same batch as the target block, measure their temporary read error rate under standard temperature (25℃) and standard idle time (100ms) conditions in a laboratory environment, and use the maximum likelihood estimation method to fit the initial parameter vector. Subsequently, the individual differences of each block are quickly adapted through online learning.
[0069] In one embodiment, Figure 7 This is a schematic diagram illustrating the optimal reading voltage adjustment range selection process provided in an embodiment of this application. Figure 7 As shown, firstly, the temporary read error risk probability R of the target plane is obtained. At the same time, the operation energy consumption weight W_energy, the delay cost weight W_delay, and the soft decoding penalty weight W_penalty under the current power consumption mode are obtained, as well as the operation energy consumption C_energy_k, the read voltage adjustment operation delay cost C_adjust_delay_k, the estimated retry time t_retry_expected, and the conditional soft decoding probability P_soft_k corresponding to each level.
[0070] Furthermore, initialize the gear number k to 0, initialize the minimum cost J_min to infinity, and initialize the optimal gear k to 0.
[0071] For the current gear k, the conditional retry probability (CRP) is calculated using the formula P_retry_k = R × exp(-γ × k). Here, γ is the decay parameter, and exp is an exponential function with base e. Then, according to the formula: J_k=W_energy×C_energy_k+C_adjust_delay_k+P_retry_k×W_delay×t_retry_expected+P_soft_k×W_penalty×C_penalty_base, formula 4; Calculate the Expected Total Cost (ETC) J_k for this gear position. Here, W_energy is the operation energy consumption weight, C_energy(k) is the operation energy consumption for gear position k, C_adjust_delay(k) is the read voltage adjustment operation delay cost for gear position k, W_delay is the delay cost weight, t_retry_expected is the expected retry time, P_soft(k) is the conditional soft decoding probability corresponding to gear position k, W_penalty is the soft decoding penalty weight, and C_penalty_base is the basic soft decoding cost.
[0072] The initial value of the attenuation parameter γ is set to 0.3, and the value ranges from 0.1 to 0.5.
[0073] The calculated J_k is compared with the currently recorded minimum cost J_min. If J_k is less than J_min, J_min is updated to J_k, and the optimal gear k is updated to the current gear k. If J_k is not less than J_min, the process proceeds directly to the next gear. The gear number k is incremented by 1, and then it is checked whether k is greater than the maximum gear number K. If k is less than or equal to K, the process returns to calculate the conditional retry probability for the next gear and continues execution. If k is greater than K, it indicates that all gears have been calculated, and the optimal voltage adjustment gear k is output.
[0074] Furthermore, the operational cost is the expected total cost. For each read voltage adjustment level k, the expected total cost J(k) for that level is determined based on the probability of a temporary read error, the operational energy consumption of that read voltage adjustment level, the read voltage adjustment operation delay cost, the expected retry delay cost, and the expected soft decoding cost, as well as the operational energy consumption weight, delay cost weight, and soft decoding penalty weight.
[0075] The delay cost C_adjust_delay(k) for reading the voltage adjustment operation is calculated in the following way: C_adjust_delay(k) = T_precharge(k) × Cost_per_us, Formula 5; T_precharge(k) is the precharge pulse duration corresponding to gear k, which can be found in the parameter table; Cost_per_us is the unit time delay cost, which is dynamically set by the memory controller 123 according to the current read delay budget, the service quality requirements of the host 11, or the power consumption mode of the storage device 12. It can be stored in a global variable for all blocks to share.
[0076] In the expected retry delay cost P_retry(k)×W_delay×t_retry_expected, t_retry_expected includes not only the hard-decision retry time, but also the time of possible voltage offset retry sequences and the read retry attempt time before soft decoding. The specific values are derived from the statistical average maintained by the firmware. P_soft(k) is updated online based on the frequency of final soft decoding triggered under the same gear in historical reads. The initial value can be set to zero or written through offline calibration.
[0077] Where J(k) represents the expected total cost of gear k, and C_energy(k) represents the operating energy consumption of gear k, which is obtained by looking up a table.
[0078] For example, C_energy(k) = k × E_unit, where E_unit is the baseline energy consumption value per unit. C_adjust_delay(k) represents the read voltage adjustment operation delay cost for level k, obtained by multiplying the precharge duration corresponding to that level by the unit time delay cost. In one embodiment, the unit time delay cost is determined based on the current read latency budget of the memory controller 123, the quality of service requirements of the host 11, or the power consumption mode of the storage device 12. A higher value is taken in high-performance mode, and a lower value is taken in low-power mode.
[0079] P_retry(k) represents the conditional probability that at least one read retry is still required after the read voltage adjustment for gear k. W_delay is the delay cost weight, and t_retry_expected represents the estimated total time required for retry, which includes the time for possible multiple rounds of hard-decision retries and voltage offset sequences. P_soft(k) represents the conditional probability that soft decoding is still required after the read voltage adjustment for gear k. This probability is obtained based on historical data statistics or determined by a preset decay function. W_penalty is the soft decoding penalty weight, and C_penalty_base is the base penalty cost for triggering soft decoding. E_unit is the energy consumption baseline value for a single gear unit, in microjoules, which is pre-calibrated by the specifications of memory module 122. C_energy(k) is obtained by querying the gear-energy consumption index table pre-stored in memory module 122, which records the measured energy consumption value of the pre-charge operation corresponding to each gear number k.
[0080] The conditional probability P_retry(k) is determined based on the probability R of a temporary read error and the gear k as follows: P_retry(k) = R × exp(-γ × k).
[0081] Here, γ is an attenuation parameter, updated online statistically to reflect the suppression effect of different voltage levels on temporary read errors. In one embodiment, the attenuation parameter γ is obtained statistically based on the suppression effect of different voltage levels on temporary read errors in historical read results; after each read voltage adjustment operation is completed and the actual read result is obtained, γ is updated using the online gradient descent method based on the difference between the predicted value of the conditional retry probability corresponding to that voltage level and whether a retry actually occurs.
[0082] Specifically, after each voltage adjustment operation is completed, the selected gear k, the predicted retry probability P_retry(k), and the actual retry flag value retry_flag are recorded in the circular buffer; every N samples (e.g., N=16), a batch gradient descent update is performed: γ_New=γ_Old-η_γ×Σ_i[(P_retry_pred_i-retry_flag_i)×(-k_i×exp(-γ_Old×k_i)×R_i)]; Where η_γ is the learning rate of γ, which can be set to 0.001 to 0.01 to avoid excessive update amplitude; after the update, γ is limited to between 0.1 and 0.5. This update is only performed when the read voltage adjustment level is non-zero; level 0 does not generate valid samples. P_retry_pred_i is the predicted conditional retry probability in the i-th sample. retry_flag_i is the label value of whether a retry was actually triggered in the i-th sample (1 for triggered retry, 0 for no trigger). exp is the natural exponential function. γ_Old represents the currently stored attenuation parameter value before this update. k_i is the read voltage adjustment level number corresponding to the i-th historical sample. R_i is the probability of a "Temporary Read Error (TRE)" corresponding to the i-th historical sample.
[0083] Calculate the expected total cost J(k) for each gear from k=0 to K, and select the gear with the minimum expected total cost as the optimal voltage adjustment gear: k=argmin_kJ(k). argmin_k is the gear number k with the minimum expected total cost J(k).
[0084] When k=0, it means that the read voltage adjustment is skipped directly, and only normal reading is performed. In this application, the operation energy consumption weight W_energy, the delay cost weight W_delay, and the soft decoding penalty weight W_penalty are dynamically set according to the power consumption mode of the memory controller 123.
[0085] In one example, the probability R of a temporary read error in target Block A is 0.772. Assume there are 8 read voltage adjustment levels, from level 0 to level 7.
[0086] For gear 0, C_energy(0)=0, P_retry(0)=0.772×exp(0)=0.772. For gear 3, assuming C_energy(3)=3×E_unit, E_unit=1, γ=0.3, then P_retry(3)=0.772×exp(-0.9)=0.772×0.407=0.314. Assuming W_energy=1, W_delay=1, W_penalty=1, C_adjust_delay(3)=10, t_retry_expected=50 microseconds, C_penalty_base=100, P_soft(3)=0.05, then J(3)=1×3+10+0.314×1×50+0.05×1×100=3+10+15.7+5=33.7.
[0087] After calculating the expected total cost for each level in the same manner, if the expected total cost for level 3 is the smallest, then level 3 is determined as the optimal reading voltage adjustment level.
[0088] In low-power mode, the operating energy consumption weight W_energy is increased relative to W_delay and W_penalty, causing the expected total cost function to tend to select the read voltage adjustment level with lower operating energy consumption, thereby achieving a dynamic balance between power consumption and read performance of the storage device 12.
[0089] For example, in normal power consumption mode, level 3 is selected as the optimal level. When the storage device 12 switches to low power consumption mode, W_energy is increased from 1 to 5, while W_delay and W_penalty remain unchanged or decrease. After recalculating the expected total cost of each level, the expected total cost of level 2 is lower than that of level 3, so level 2 is selected as the optimal level. Through the above mechanism, this application achieves a dynamic balance between storage device power consumption and read performance, enabling the read voltage adjustment strategy to adaptively adjust according to the current power consumption requirements of the storage device 12.
[0090] In practice, the memory controller 123 can maintain a power mode state machine, including normal mode, low power mode and high performance mode; the switching conditions can be triggered by the host 11 command, IO queue depth or power budget of the storage device 12.
[0091] Each power consumption mode corresponds to a set of weight parameters. For example: Normal mode: W_energy=1, W_delay=1, W_penalty=1; Low power mode: W_energy=5, W_delay=1, W_penalty=1; High performance mode: W_energy=0.5, W_delay=2, W_penalty=1. The memory controller 123 reads the weight table corresponding to the current power consumption mode before each power level selection, without needing online calculation.
[0092] In one embodiment, the correspondence between the read voltage adjustment level and the precharge voltage parameter is pre-fixed in the parameter table of the memory module 122.
[0093] Specifically, each gear number k uniquely corresponds to a set of pre-charge voltage parameters, including the duration of the pre-charge pulse and the pre-charge voltage amplitude. Gear 0 corresponds to a duration of zero and a pre-charge voltage amplitude of zero, indicating that no read voltage adjustment operation is performed. The duration and pre-charge voltage amplitude of gears 1 to 7 increase monotonically with the gear number, that is, the higher the gear number, the longer the duration of the pre-charge pulse and the larger the pre-charge voltage amplitude. The specific values of the above duration and pre-charge voltage amplitude are determined by the design and manufacturing process of the memory module 122, and are calibrated and fixed at the chip factory.
[0094] During read voltage adjustment, the system retrieves the corresponding duration and precharge voltage amplitude from the parameter table based on the determined optimal read voltage adjustment level. Then, configuration parameters containing this duration and precharge voltage amplitude are written to the mode register in the memory module 122. According to the configuration parameters in the mode register, a precharge pulse with a specified duration and precharge voltage amplitude is applied to the selected word line before the start of a standard read operation. This precharge pulse compensates for threshold voltage shifts caused by trap charge leakage during idle periods by pulling the threshold voltage distribution of the memory cell towards a steady-state operating distribution.
[0095] The aforementioned read voltage adjustment operation is essentially a charge recapture operation. Charge recapture refers to the read voltage adjustment operation that accelerates the recovery of trapped charges in the memory cell by extending the pre-charge duration and / or increasing the pre-charge voltage amplitude, based on the pre-charge phase of the read operation of memory module 122. Compared to the standard pre-charge phase, the charge recapture operation is longer in duration and has a stronger pre-charge voltage amplitude, aiming to accelerate the recovery process of trapped charges and allow the memory cell to return to a near-steady-state physical state before the actual read operation.
[0096] In one embodiment, after determining that multiple planes contain the target block, the optimal read voltage adjustment level for each of these planes is obtained, and then a level merging judgment is performed. The core of the level merging judgment lies in comparing the optimal level strength of each plane. The strength of the read voltage adjustment level is determined based on the pre-charge voltage amplitude and / or pre-charge duration corresponding to that level.
[0097] Figure 8 This is a schematic diagram of the gear merging process provided in an embodiment of this application. Figure 8 As shown, after determining that there are multiple planes containing the target block, the optimal reading voltage adjustment level for each of these planes is first obtained.
[0098] Sort all planes containing the target block in ascending order of the intensity of the optimal read voltage adjustment range. Use the plane with the lowest intensity as the reference plane and assign it to the current plane group. Select subsequent planes as comparison planes in turn. If the intensity of the comparison plane is less than or equal to the intensity of the reference plane, assign the comparison plane to the current plane group. If the intensity of the comparison plane is greater than the intensity of the reference plane, use the comparison plane as the new reference plane and start a new plane group. Repeat the above process until all planes are grouped.
[0099] The optimal reading voltage adjustment level for each plane in the current plane group is sorted by intensity, and the level with the highest intensity in the group is selected as the unified reading voltage adjustment level for the plane group.
[0100] In one embodiment, the specific grouping method described above includes: sorting all planes containing the target block from low to high according to their optimal strength level; taking the plane with the lowest strength as a reference, checking the subsequent planes in turn, and if their strength level does not exceed the reference strength, adding them to the same group; otherwise, ending the current group and using that plane as the reference for the new group; repeating until all planes are grouped.
[0101] After grouping, the memory controller 123 generates a multi-plane read command for each plane group. When the memory controller 123 supports multi-plane parallel processing on the same die, these commands can be issued simultaneously; if the plane groups are distributed across different dies, they are issued in parallel and interleaved according to the ready / busy status of the die to avoid single-bus blocking.
[0102] Furthermore, the cost of merging is checked using the same expected total cost formula as for a single plane, but C_adjust_delay and C_energy (operational energy consumption) are calculated at the same level. Considering that the adjustment operations are executed simultaneously under multi-plane parallelism, the actual total energy consumption is the sum of the energy consumption of each plane, while the adjustment delay is only counted once, thus reflecting the parallel benefits in the cost comparison.
[0103] In one embodiment, if there are multiple plane groups, the above sorting and selection operations are performed on each plane group respectively, and finally the unified reading voltage adjustment level of each plane group is output.
[0104] The process involves comparing the intensity of different voltage adjustment levels on a first and second plane. If the intensity value of the optimal reading voltage adjustment level on the first plane is less than or equal to that on the second plane, then the second plane's adjustment level is considered compatible with the first plane's. Level compatibility means that when a higher intensity level is used for uniform reading voltage adjustment, it will not cause any damage to the plane that originally required a lower intensity level; it will only increase the energy consumption of that plane. Following this rule, all planes containing the target block are compared pairwise, and planes with compatible levels are grouped into the same plane group. If a plane's adjustment level is not compatible with the levels of other planes in the group, then that plane becomes a separate plane group or is grouped into another group with planes whose intensity levels are close to its own.
[0105] For each plane group that has been grouped, the intensity value of the optimal read voltage adjustment level for each plane within that plane group is read, and the level with the highest intensity is selected as the unified read voltage adjustment level for that plane group. The selected unified level will be used for the read voltage adjustment operation of all planes within that plane group.
[0106] Furthermore, after determining the unified reading voltage adjustment level, the expected total cost corresponding to the unified level is calculated and compared with the sum of the expected total costs of each plane executing its own optimal level. If the merged cost is less than or equal to the sum of the individual execution costs, the plane group is retained; otherwise, the merge is canceled and the relevant planes are processed independently.
[0107] In one example, assume there are three planes containing the target block: Plane0, Plane1, and Plane2. The optimal gear for Plane0 is gear 2, for Plane1 it's gear 5, and for Plane2 it's gear 3. Comparing Plane0 and Plane1, gear 2 is less intense than gear 5, therefore gear 5 is compatible with gear 2. Comparing Plane1 and Plane2, gear 3 is less intense than gear 5, therefore gear 5 is compatible with gear 3. Since the gears of the three planes are mutually compatible, they are grouped into the same plane group. The highest intensity gear within this group is gear 5, which is designated as the unified voltage adjustment gear. If subsequent timing or power constraint checks fail, planes with incompatible gears or whose merging costs are too high are separated and processed separately.
[0108] In one embodiment, after completing plane grouping and determining the unified read voltage adjustment level, the unified level and plane group information are encapsulated into a multi-plane read command. The extended fields of this read command carry the unified read voltage adjustment level field and the plane mask field. The instruction format of this read command is predefined by the interface protocol of the memory interface 22 between the memory module 122 and the memory module 122.
[0109] The unified read voltage adjustment level field carries the level number that all planes within the plane group execute together. The plane mask field is a binary value with a bit width equal to the total number of planes contained in the memory module, where each bit corresponds to a plane. When constructing the plane mask, the bits corresponding to the planes belonging to the plane group are set to 1, and the bits corresponding to the planes outside the plane group are set to 0. Taking a memory module 122 containing 4 planes as an example, if the plane group consists of Plane0 and Plane2, the plane mask is binary 0101, where the least significant bit corresponds to Plane0 and the second bit corresponds to Plane2. This plane mask, along with the unified level, is filled into the extended field of the read command, and then the command is sent to the memory module 122 through the memory interface 22.
[0110] Upon receiving the command, memory module 122 first parses the plane mask field of the extended field. For each bit in the mask that is 1, the precharge execution circuit of the plane corresponding to that bit is activated. For bits in the mask that are 0, the corresponding plane remains inactive and does not participate in this operation. At the same time, the gear number is read from the uniform gear field in the extended field, and the corresponding precharge duration and precharge voltage amplitude are looked up from the internal parameter table based on the gear number. Then, all activated planes simultaneously begin to execute read voltage adjustment operations with the same duration and the same precharge voltage amplitude. After the read voltage adjustment operation is completed, the planes remain active, and then a parallel read command is issued. Each plane synchronously performs normal read operations, moving data from the memory cell to its respective page cache register.
[0111] In one example, assume Plane0 and Plane2 are grouped into the same plane group and their unified gear is designated as gear 5. A multi-plane read command is constructed, filling the unified gear field with gear 5 and the plane mask field with binary 0101. After the command is issued, the 0th and 2nd bits of the mask are detected to be 1, thus activating the precharge execution circuits of Plane0 and Plane2, simultaneously applying the precharge pulse corresponding to gear 5 to these two planes. Plane1 and Plane3 are not activated because their corresponding mask bits are 0. After precharging is complete, a parallel read command is issued, and Plane0 and Plane2 simultaneously perform data read operations.
[0112] In one embodiment, if it is determined that multiple planes containing target blocks cannot be merged into the same plane group, a priority-based resource scheduling process is then executed. Reasons for failure to merge include, but are not limited to, excessive differences in the optimal power levels of the planes leading to incompatibility issues, combined power consumption exceeding the power consumption constraints of the storage device after merging, or timing of the merging execution not meeting the specifications of the memory module 122. In these cases, read voltage adjustment resources need to be allocated sequentially among the planes in a certain order. In this application, read voltage adjustment resources refer to the power budget and / or command issuance time slot corresponding to the read voltage adjustment operation, including the current budget consumed by the precharge operation and the time slice for the precharge command to be issued on the bus.
[0113] First, the probability of a temporary read error in the current read request for each plane is obtained, which has been calculated in step S1. Then, the estimated read operation time and estimated retry time for each plane are obtained. The estimated read operation time refers to the time required from issuing the read command to the transfer of data from the storage unit to the page cache register, and this time is determined by the specifications of the memory module 122. The estimated retry time refers to the estimated time required to complete one or more rounds of retry operations if the current read triggers a retry. The expected delay cost (EDC) for each plane is calculated as follows: Expected delay cost = probability of temporary read error × (estimated read operation time + estimated retry time).
[0114] The probability of a temporary read error is multiplied using a value between 0 and 1, and the estimated read operation time and estimated retry time are in microseconds. After calculation, each plane corresponds to a value for expected latency cost. This value reflects the expected time loss that might occur if the plane is served with delay. A larger value indicates that the plane should be served with priority.
[0115] Prioritize each plane according to its expected latency cost, from highest to lowest. The plane with the highest expected latency cost receives the highest priority, and the plane with the lowest expected latency cost receives the lowest priority. Read voltage adjustment resources are allocated to each plane in descending order of priority, and corresponding read voltage adjustment commands are issued.
[0116] In specific implementation, the memory controller 123 calculates the expected latency cost for each plane to be processed and inserts it into a scheduling queue arranged in descending order of cost. A plane is retrieved from the head of the queue, allocated a command slot (e.g., t_slot = 20 microseconds), and the total current required for the plane to perform a read voltage adjustment operation is calculated to see if it exceeds the current remaining power budget. If it does not exceed the current, a read voltage adjustment command is issued for the plane, and the power budget is released upon completion. If it does exceed the current, the process waits for the previous plane to complete before attempting the operation again.
[0117] The power consumption budget can be dynamically adjusted by the memory controller 123 according to the overall power consumption management strategy of the storage device 12. For example, if the peak current budget reserved for the read voltage adjustment operation is 200mA and the peak current of the precharge of a single plane is about 50mA, then a maximum of 4 planes can be scheduled in parallel at the same time; when parallelism is not possible, they are executed serially according to priority.
[0118] In one embodiment, resources are released and the next priority plane is processed only after the read voltage adjustment operation and subsequent normal read operation of a plane are completed.
[0119] In one example, suppose three planes containing the target block cannot be merged. Plane0 has a temporary read error probability of 0.9, an estimated read operation time of 50 microseconds, and an estimated retry time of 100 microseconds. Its expected latency cost is 0.9 multiplied by 150 microseconds, equaling 135 microseconds. Plane1 has a temporary read error probability of 0.3, an estimated read operation time of 50 microseconds, and an estimated retry time of 100 microseconds. Its expected latency cost is 0.3 multiplied by 150 microseconds, equaling 45 microseconds. Plane2 has a temporary read error probability of 0.7, an estimated read operation time of 50 microseconds, and an estimated retry time of 100 microseconds. Its expected latency cost is 0.7 multiplied by 150 microseconds, equaling 105 microseconds. Planes are then served in descending order of expected latency cost: Plane0, Plane2, and Plane1.
[0120] In one embodiment, after determining the unified reading voltage adjustment level, the expected total cost corresponding to the unified level is calculated and compared with the sum of the expected total costs of each plane executing its own optimal level; if the combined cost is less than or equal to the sum of the individual execution costs, the plane group is retained; otherwise, the combination is canceled and the relevant planes are processed independently.
[0121] Figures 3 to 8 Each step can be implemented as multiple program codes or circuits, applied to the storage device 12 including the memory module 122, and the memory control circuit 23 is configured to execute the above method. This application does not limit this. Furthermore, Figures 3 to 8 The method can be used in conjunction with the above examples and embodiments, or it can be used alone; this application does not impose any restrictions.
[0122] In summary, this application provides a read voltage adjustment method and a storage device. The storage device includes a memory module and a memory controller. The memory module includes multiple planes, each plane includes multiple blocks, and each block contains multiple pages. The method includes: responding to read requests for multiple pages, determining the block to which each page belongs; selecting the block that is first read after idle recovery from the determined blocks as the target block, and selecting the page belonging to the read request in the target block as the target page, wherein "first read after idle recovery" means the corresponding block in the memory module is read for the first time after recovering from an idle state; obtaining a set of state attributes for each target block, and obtaining the probability of a temporary read error occurring in each target block based on the state attributes; determining the optimal read voltage adjustment level for the plane where each target block is located from the multiple read voltage adjustment levels based on the probability of a temporary read error for each target block and the operation cost of each of the preset multiple read voltage adjustment levels; if multiple target blocks exist in the same plane, then... The highest intensity of the optimal read voltage adjustment level among the optimal read voltage adjustment levels corresponding to each target block within a plane is taken as the optimal read voltage adjustment level for that plane. Each of the multiple read voltage adjustment levels corresponds to a different pre-charge voltage amplitude and / or pre-charge duration. If the target block is distributed across multiple planes, then, when a preset level merging condition is met, the multiple planes are divided into at least one plane group according to the preset level merging condition. The optimal read voltage adjustment levels of each plane within the same plane group are merged into a unified read voltage adjustment level, which is then used as the read voltage adjustment level to be executed for that plane group. Otherwise, the optimal read voltage adjustment level of each plane is used as the read voltage adjustment level to be executed. Based on the read voltage adjustment level to be executed, a read command carrying the read voltage adjustment level is issued to the target page within the corresponding plane or plane group, so that the read voltage adjustment operation is performed according to the read voltage adjustment level before performing the read operation on the target page. This application achieves targeted adjustment of the read voltage by identifying the target block being read for the first time after idle recovery and predicting the probability of temporary read errors by combining the block's state attributes.
[0123] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.
Claims
1. A method for reading voltage adjustment, characterized in that, Applied to a storage device including a memory module and a memory controller, the memory module comprising multiple planes, each plane comprising multiple blocks, each block containing multiple pages, the method comprising: In response to read requests for multiple pages, the blocks to which each of the multiple pages belongs are determined, the block that is read for the first time after the idle period is recovered from the determined block is taken as the target block, and the page in the target block that belongs to the read request is taken as the target page; Obtain a set of status attributes for each target block, and based on the status attributes, obtain the probability of a temporary read error occurring in each target block; Based on the probability of temporary read errors for each target block and the operational costs of each of the preset multiple read voltage adjustment levels, the optimal read voltage adjustment level for the plane where each target block is located is determined from the multiple read voltage adjustment levels, wherein each of the multiple read voltage adjustment levels corresponds to a different pre-charge voltage amplitude and / or pre-charge duration. If the target block is distributed across multiple planes, then when the preset gear merging condition is met, the multiple planes are divided into at least one plane group according to the gear merging condition, and the optimal read voltage adjustment gears of each plane in the same plane group are merged into a unified read voltage adjustment gear, which is then used as the read voltage adjustment gear to be executed for the plane group. Otherwise, the optimal read voltage adjustment level for each plane will be used as the read voltage adjustment level to be executed. Based on the read voltage adjustment level to be executed, a read command carrying the read voltage adjustment level is sent to the target page in the corresponding plane or plane group, so that the read voltage adjustment operation is performed according to the read voltage adjustment level before the read operation on the target page is performed.
2. The reading voltage adjustment method according to claim 1, characterized in that, The set of state attributes includes at least one of the following: idle duration, average temperature during idle periods, peak temperature, equivalent temperature, number of erase / write cycles, and an exponentially weighted moving average of the number of error bits.
3. The reading voltage adjustment method according to claim 1, characterized in that, The method further includes: Based on the results of the read operation, update the parameters of the logistic regression model used to determine the probability of the temporary read error and the attenuation parameters used to calculate the operation cost of each read voltage adjustment level.
4. The reading voltage adjustment method according to claim 3, characterized in that, The step of updating the logistic regression model parameters used to determine the probability of the temporary read error based on the result of the read operation includes: The label value is determined based on whether the read operation triggers a retry or the number of error bits exceeds the hard decision threshold. Based on the difference between the probability of the temporary read error and the label value, the parameters of the logistic regression model are updated using the gradient descent method. The learning rate of the gradient descent method decreases with the number of erases and rewrites of the target block, and the parameter update of the logistic regression model is triggered only when one of the following conditions is met: The difference between the probability of the temporary read error and the tag value exceeds a preset threshold; The idle time of the target block exceeds the minimum threshold.
5. The reading voltage adjustment method according to claim 3, characterized in that, The operational cost is the expected total cost. Based on the probability of the temporary read error and the operational costs of each of the preset multiple read voltage adjustment levels, the optimal read voltage adjustment level is determined from the multiple read voltage adjustment levels, including: For each read voltage adjustment level, based on the probability of the temporary read error, the operating power consumption of the read voltage adjustment level, the expected retry latency cost, the expected soft decoding cost, and the read voltage adjustment operation latency cost; and The expected total cost is determined by weighting the operating energy consumption, latency cost, and soft decoding penalty. The optimal read voltage adjustment level is selected as the one that minimizes the expected total cost.
6. The reading voltage adjustment method according to claim 5, characterized in that, The operation energy consumption weight, the latency cost weight, and the soft decoding penalty weight are dynamically set according to the power consumption mode of the memory controller; In low-power mode, the weight of operation energy consumption is increased relative to the weight of latency cost and the weight of soft decoding penalty, so that the expected total cost tends to select the read voltage adjustment level with low operation energy consumption.
7. The reading voltage adjustment method according to claim 5, characterized in that, According to preset merging conditions, the multiple planes are divided into at least one plane group, and the optimal read voltage adjustment levels of each plane within the same plane group are merged into a unified read voltage adjustment level. This unified read voltage adjustment level is used as the read voltage adjustment level to be executed for that plane group, including: The intensity of the reading voltage adjustment level is determined based on the pre-charge voltage amplitude and / or pre-charge duration corresponding to that level; Sort all planes containing the target block in order of optimal gear strength from low to high. Using the plane with the lowest strength as the reference, successively group planes whose gear strength does not exceed the gear strength of the reference plane into the same plane group. When a plane with gear strength exceeding the gear strength of the reference plane is encountered, use that plane as the new reference to continue grouping until all planes are grouped. If the strength of the optimal reading voltage adjustment level of the first plane is less than or equal to the strength of the optimal reading voltage adjustment level of the second plane, then the level of the second plane is compatible with the level of the first plane, and the planes with compatible levels are divided into the same plane group. The highest intensity level among the optimal reading voltage adjustment levels for each plane within the same plane group is taken as the unified reading voltage adjustment level. Furthermore, the level merging is only performed when the expected total cost corresponding to the merged unified read voltage adjustment level is less than or equal to the sum of the expected total costs of each plane executing its own optimal read voltage adjustment level.
8. The reading voltage adjustment method according to claim 1, characterized in that, The preset multiple read voltage adjustment levels each correspond to different pre-charge durations and / or pre-charge voltage amplitudes, which are used to adjust the threshold voltage distribution of the storage unit before reading to compensate for charge leakage caused by idle time. The read voltage adjustment operation is a charge recapture operation.
9. The reading voltage adjustment method according to claim 1, characterized in that, The unified reading voltage adjustment level is issued through the extended field in the multi-plane reading command. The reading command carries the unified reading voltage adjustment level and a mask for indicating each plane in the plane group. The multiple planes within the plane group correspond to planes within the same die of the memory module that support parallel operations.
10. The reading voltage adjustment method according to claim 1, characterized in that, The method further includes: In response to the inability of multiple planes containing the target block to be merged into the same plane group, the priority of each plane is determined based on the probability of temporary read errors and the expected delay cost of each plane, and the power consumption budget and / or command issuance time slot corresponding to the read voltage adjustment operation are allocated to each plane in order of priority. The expected delay cost is determined based on the probability of temporary read errors in each plane, the estimated read operation time, and the estimated retry time.
11. A storage device, characterized in that, include: A connection interface for electrically connecting to a host computer; The memory module includes multiple planes, each plane includes multiple blocks, and each block includes multiple pages; as well as The memory controller includes an error checking and correction circuit, a buffer memory, and a memory control circuit, and is electrically connected to the connection interface and the memory module; The memory controller is configured to perform the read voltage adjustment method as described in any one of claims 1 to 10.
Citation Information
Patent Citations
Data reading method and memory controller
CN120010785A
Memory management method and storage device
CN121354636A
