A method and system for diagnosing early potential defects of a solid state drive

Abstract

The present application relates to the technical field of solid state disk, especially relates to a kind of solid state disk early potential defect diagnosis method and system, the method first obtains the charge holding level of each flash channel under pressure aging and reading error rate timing data;Second, construct time domain threshold shift feature and filter sensitive data holding window;Then, synchronous analysis error rate gradient change and window level fluctuation, identify abnormal leakage area caused by oxide layer defect and electron tunneling effect;Subsequently, establish the mapping relationship of leakage position, attenuation rate and leakage defect evolution dynamics model, quantitatively evaluate early leakage position and risk tendency value;Finally generate health status diagnosis report.The present application realizes the accurate positioning and risk quantization of early charge leakage defect of solid state disk through multi-source signal fusion and physical model mapping, significantly improves the predictability and operation safety of storage system.

Description

Technical Field

[0001] This invention relates to the field of solid-state drive (SSD) technology, and in particular to a method and system for diagnosing early potential defects in SSDs. Background Technology

[0002] With the widespread application of solid-state drives (SSDs) in data centers, industrial control, and consumer electronics, their long-term operational reliability has become a crucial factor affecting system stability. Among various failure modes, charge leakage is one of the key mechanisms leading to data loss in NAND flash memory. Especially after the device undergoes numerous programming / erasing cycles and long periods of data retention, the charge in the floating gate or charge trapping layer of the memory cell can be unexpectedly lost due to defects in the tunnel oxide layer, interface state traps, or electron tunneling effects, leading to threshold voltage drift and an increase in read error rate. This problem is particularly prominent in 3D NAND devices with continuously shrinking process dimensions, potentially causing early data integrity degradation or even functional failure.

[0003] Traditional SSD health monitoring primarily relies on global metrics such as bad block count, average write cycles, remaining lifetime percentage, and static bit error rate (BER) statistics. These methods struggle to pinpoint and quantify weak anomalies in the early stages of charge leakage. Especially for non-uniform charge retention degradation across channels caused by manufacturing process variations or localized material defects, conventional methods lack sufficient spatial resolution and sensitivity. Furthermore, existing diagnostic techniques often analyze isolated level changes or BER metrics, failing to effectively integrate the temporal coupling characteristics of multi-source signals. This results in a delay in identifying early leakage defects, often only being detected after significant data distortion.

[0004] The above content is only used to help understand the technical solution of the present invention and does not represent an admission that the above content is prior art. Summary of the Invention

[0005] The main objective of this invention is to provide a diagnostic method and system for early potential defects in solid-state drives (SSDs), aiming to solve the technical problem that existing SSD health monitoring technologies rely on global statistical indicators and lack the ability to fuse and analyze multi-source signals, making it difficult to accurately identify cross-channel non-uniform early charge leakage caused by process fluctuations or local defects.

[0006] To achieve the above objectives, the present invention provides a method for diagnosing early potential defects in solid-state drives, the method comprising: Acquire a multi-source timing monitoring dataset of the target solid-state drive under stress aging conditions. The multi-source timing monitoring dataset includes charge retention level signals and read error rate signals deployed at preset test units in each flash channel. Based on the charge retention level signal, a time-domain threshold drift characteristic reflecting the charge decay characteristics of the channel unit is constructed, and sensitive data retention windows related to physical erasure and programming cycles of flash memory blocks are screened according to pressure aging conditions. Synchronous analysis of the gradient distribution changes of the bit error rate signal, combined with the level fluctuations within the window of sensitive data, identifies the abnormal leakage area of ​​the channel caused by oxide layer defect trap capture and electron tunneling effect; Establish a mapping relationship between the logical address of the abnormal leakage area, the decay rate of the time domain threshold drift characteristic, and the preset leakage defect evolution dynamic model, and generate a quantitative assessment result of the early charge leakage location and risk tendency value of the target solid-state drive. Based on the location of early charge leakage and risk propensity value in the quantitative assessment results, a health status diagnostic report of the target solid-state drive is generated.

[0007] Optionally, the settings of the flash channel preset test unit include: Based on the physical architecture and stress aging conditions of flash memory chips, key lines and bit line regions that are prone to charge loss are identified, thus obtaining the critical regions. Within the critical area, the test unit allocation positions are selected by cross-selection based on the location of the maximum electric field intensity on the surface of the flash memory block and the location of the programming interference mode node, thus obtaining the initial allocation positions; Based on the conduction attenuation relationship between the charge holding level signal and the read error rate signal in the flash array, the relative spacing and logical mapping relationship of the test cells at the initial allocation position are adjusted to obtain the optimized layout of the preset test cells of the flash channel.

[0008] Optionally, constructing a time-domain threshold drift feature reflecting the charge decay characteristics of the channel unit based on the charge retention level signal includes: The time-domain voltage distribution of the charge-holding level signal is obtained by performing a time-frequency transformation on the charge-holding level signal; In the time-domain voltage distribution, based on the continuity of the voltage distribution, the evolution path of the voltage level in the concentrated voltage region over time is traced to obtain the initial drift ridge. By combining the known charge loss response mode of the target solid-state drive under pressure aging conditions, the level value of the initial drift ridge is trend-fitted to obtain the time-domain threshold drift characteristics that reflect the charge decay characteristics of the channel cell.

[0009] Optionally, the sensitive data retention window, which is selected based on pressure aging conditions and associated with the physical erasure and programming cycles of the flash memory block, includes: By analyzing the pressure aging conditions, typical stress components corresponding to the erase and programming operations of the flash memory block were separated, and typical stress time series were obtained. By mapping the time phase of a typical stress time series to the level change phase of the time domain threshold drift characteristic, a time synchronization relationship between stress excitation and charge level response is established. Based on the time synchronization relationship, the fluctuation period of each window level and the change period of typical stress components in the time domain threshold drift characteristics are analyzed; windows with stable period correspondence are identified as sensitive data holding windows.

[0010] Optionally, the synchronous analysis of the gradient distribution changes of the bit error rate signal, combined with the level fluctuations within the sensitive data maintenance window, identifies abnormal channel leakage regions caused by oxide layer defect trapping and electron tunneling effects, including: Based on the read error rate signals deployed in preset test units of different flash memory channels, an error field reflecting the spatial distribution of flash disk errors is constructed; The error field on a continuous time series is analyzed, and the rate of change of the error difference between adjacent logic cells of the flash disk over time is calculated to obtain the gradient distribution change. From the gradient distribution changes, regional features that exceed the rate of change caused by the background read interference of the flash disk are extracted to obtain candidate abnormal leakage regions. Acquire the time series of level fluctuations within the same time period in the sensitive data retention window; correlate the time phase of the candidate abnormal leakage area with the changing phase of the level fluctuation time series, analyze the time correspondence and intensity ratio between the changes of the candidate abnormal leakage area and the level fluctuation time series, and obtain the electrical-miscoupling correlation relationship; Based on the electrical-miscoupling correlation, sub-regions in the candidate abnormal leakage region whose changes have a stable intensity ratio with the time series of level fluctuations are identified as abnormal leakage regions of the target solid-state drive.

[0011] Optionally, the leakage defect evolution dynamics model is constructed through the following steps: Based on data retention test data of the same flash memory medium as the target solid-state drive under pressure aging conditions, the constitutive relationship between charge leakage rate and stress intensity factor range was determined. In the constitutive relation, a damping attenuation coefficient related to level energy dissipation is introduced as a correction term to establish a two-dimensional parameter space with the stress intensity factor range and the damping attenuation coefficient as coordinates, and the boundary between the stable development region and the unstable development region of leakage is delineated in this space. By integrating the two-dimensional parameter space and boundary, a criterion system for quantitatively assessing the risk of charge leakage is constructed, resulting in a dynamic model of leakage defect evolution.

[0012] Optionally, the mapping relationship between establishing the logical address of the abnormal leakage region, the decay rate of the time-domain threshold drift characteristic, and the preset leakage defect evolution dynamics model includes: For abnormal leakage areas, based on their logical addresses and the principal stress trajectory of the target solid-state drive under pressure aging conditions, the potential leakage initiation direction and local stress concentration factor are determined, and local stress state parameters are obtained. The amplitude of the time-domain threshold drift feature within the sensitive data holding window is fitted over time, and the percentage decrease in amplitude per unit time is calculated to obtain the energy decay rate. Based on the local stress state parameters and energy decay rate, a feature point characterizing the current flash memory medium state is determined in the parameter space defined by the leakage defect evolution dynamics model. Based on the relative relationship between the position of feature points in the parameter space and the boundary of the stable development region of leakage defined by the model, the tendency of abnormal leakage regions to develop into severe charge leakage defects is quantified, and the mapping relationship is obtained.

[0013] Optionally, the quantitative assessment results of the early charge leakage locations and risk propensity values ​​of the generated target solid-state drive include: From the abnormal leakage area, select the area with the highest degree of overlap between its logical address and the potential leakage initiation direction indicated by the local stress state parameter as the early charge leakage location; Obtain the charge leakage risk propensity value quantified by the mapping relationship corresponding to the early charge leakage location; The charge leakage risk propensity value is dynamically corrected based on the energy decay rate to obtain the corrected risk propensity value. The early charge leakage locations are combined with the revised risk propensity values ​​to form a quantitative assessment result.

[0014] Optionally, the step of generating a health status diagnostic report for the target solid-state drive based on the early charge leakage locations and risk propensity values ​​in the quantitative assessment results includes: The urgency level of maintenance actions is determined based on the magnitude of the risk propensity value and the local stress state parameters at the location of early charge leakage. Based on the correlation between energy decay rate and electrical-miscoupling, maintenance strategy recommendations are generated for early charge leakage locations; The logical address of the early charge leakage location, risk propensity value, urgency level of maintenance action, and maintenance strategy recommendations are structurally integrated to generate a health status diagnostic report for the target solid-state drive.

[0015] Furthermore, to achieve the above objectives, the present invention also provides a diagnostic system for early potential defects in solid-state drives, the system comprising: The multi-source acquisition module is used to acquire cross-channel multi-source timing monitoring datasets of the target solid-state drive under stress aging conditions. The multi-source timing monitoring datasets include charge retention level signals and read error rate signals deployed at preset test units in each flash channel. The feature construction module is used to construct a time-domain threshold drift feature that reflects the charge decay characteristics of the channel unit based on the charge retention level signal, and to filter sensitive data retention windows associated with physical erasure and programming cycles of flash memory blocks according to pressure aging conditions. The anomaly identification module is used to synchronously analyze the gradient distribution changes of the read bit error rate signal, combine sensitive data to maintain the level fluctuation within the window, and identify the channel abnormal leakage area caused by oxide layer defect trap capture and electron tunneling effect. The quantitative evaluation module is used to establish the mapping relationship between the logical address of the abnormal leakage area, the decay rate of the time domain threshold drift characteristics and the preset leakage defect evolution dynamic model, and generate quantitative evaluation results of the early charge leakage location and risk tendency value of the target solid-state drive. The report generation module is used to generate a health status diagnostic report for the target solid-state drive based on the location of early charge leakage and risk tendency value in the quantitative assessment results.

[0016] This invention provides a diagnostic method for early potential defects in solid-state drives (SSDs). This method integrates multi-source timing data of cross-channel charge retention levels and read error rates to construct time-domain threshold drift characteristics and combines them with sensitive data retention window analysis, achieving highly sensitive detection of early charge leakage anomalies in flash memory cells. By synchronously identifying the coupling relationship between level fluctuations and error gradient changes, it accurately locates abnormal leakage regions caused by oxide layer defects and electron tunneling effects. Furthermore, it introduces a leakage defect evolution dynamic model based on physical degradation mechanisms, establishing a mapping relationship between logical address, decay rate, and risk tendency, thereby achieving a quantitative assessment of the location and development trend of early charge leakage. Finally, it generates a health status diagnostic report containing risk levels and maintenance recommendations, significantly improving the predictability, location accuracy, and risk assessment capability of SSDs for early charge leakage defects under complex stress conditions. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating an embodiment of the diagnostic method for early potential defects in solid-state drives according to the present invention. Figure 2 This is a flowchart illustrating another embodiment of the diagnostic method for early potential defects in solid-state drives according to the present invention. Figure 3 This is a flowchart illustrating another embodiment of the diagnostic method for early potential defects in solid-state drives according to the present invention. Figure 4This is a structural block diagram of an embodiment of the diagnostic system for early potential defects in solid-state drives according to the present invention.

[0018] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0019] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0020] Reference Figure 1 , Figure 1 This is a flowchart illustrating an embodiment of the method for diagnosing early potential defects in a solid-state drive according to the present invention. An embodiment of the method for diagnosing early potential defects in a solid-state drive according to the present invention is presented.

[0021] In one embodiment, the method for diagnosing early potential defects in the solid-state drive includes: Step S100: Obtain the cross-channel multi-source timing monitoring dataset of the target solid-state drive under stress aging conditions. The multi-source timing monitoring dataset includes the charge holding level signal and read error rate signal deployed at the preset test unit of each flash channel.

[0022] The target solid-state drive (SSD) can be a solid-state storage device undergoing early potential defect diagnosis. It can serve as the object of diagnostic methods, housing flash memory channels and test units for data acquisition and analysis. Stress aging conditions can simulate environmental or operational stress combinations that accelerate device degradation, stimulating potential charge leakage behavior and allowing early defects to be observed within a limited time. In an exemplary embodiment, stress aging conditions can be applied to the SSD using high temperature, high humidity, high voltage, and frequent write / erase operations to induce early failure mechanisms. Flash memory channels can be data transmission paths between the SSD controller and NAND flash memory chips, serving as physical acquisition dimensions for charge retention levels and bit error rate signals, supporting cross-channel comparative analysis. Pre-selected test units can be representative flash memory blocks or pages pre-selected for monitoring in each flash memory channel, providing stable and repeatable signal acquisition points for constructing multi-source timing monitoring datasets. Furthermore, the pre-selected test units can be configured by firmware during the initialization phase and distributed across different physical locations to cover areas of process variation. The charge hold level signal can be an analog or digital representation of the remaining charge in the floating gate or charge trapping layer of a flash memory cell. It can be used to characterize charge decay dynamics and to construct time-domain threshold drift characteristics. For example, the charge hold level signal can be indirectly obtained by reading the cell's on-state or threshold voltage offset at a reference voltage.

[0023] The read error rate (BER) signal can be a sequence of the proportion of bit errors occurring per unit amount of data under specific read conditions. It can be used to reflect the degree of data integrity degradation, and its gradient change can be used to identify abnormal leakage areas. In a specific embodiment, the BER signal can be obtained by statistically analyzing and normalizing the original number of bit errors using an ECC decoder, and recorded by timestamp. The BER signal can include, but is not limited to, random BER, burst BER, and uncorrectable BER. The cross-channel multi-source timing monitoring dataset can be a collection of synchronously acquired results of charge hold level signals and read BER signals in multiple flash memory channels under a unified time axis. It can be used to provide high-dimensional input for subsequent feature construction and coupling analysis, breaking through the limitations of traditional global indicators. Furthermore, the cross-channel multi-source timing monitoring dataset can be integrated by scheduling test unit read operations through firmware and aligning timestamps. In an exemplary embodiment, the cross-channel multi-source timing monitoring dataset relies on the synchronous acquisition of charge hold level signals and read BER signals, which together constitute its data dimension.

[0024] Obtaining a multi-channel, multi-source timing monitoring dataset of a target solid-state drive under stress aging conditions can be achieved by scheduling preset test cells in each flash memory channel to periodically read charge maintenance levels and bit error rates during the stress aging process, and then aligning the data by timestamps. Furthermore, this operation can be implemented by inserting lightweight read tasks during background garbage collection intervals to reduce performance interference, or by utilizing dedicated diagnostic firmware to acquire data at full speed in offline mode. This achieves spatiotemporal synchronous acquisition of multi-source signals, providing high-resolution input for early anomaly detection. For example, under the condition of high temperature and pressure aging of the target solid-state drive at 85 degrees Celsius, the system inserts lightweight read tasks into the preset test units of each channel with a millisecond time difference during the background garbage collection operation interval. It successfully collected a set of reference data in which the initial charge holding level of channels 0 to 3 was 2450 mV and the read error rate was maintained at a low level of 1.0 x 10^-6, and stored it aligned with the precise timestamp. Subsequently, in the maintenance window at dawn every day, the dedicated diagnostic firmware was enabled to enter offline mode. Full-speed scanning revealed that after 24 hours of aging, the charge holding level of channel 3 dropped sharply to 2410 mV, and its read error rate surged to 1.2 x 10^-4. This online and offline acquisition method not only avoids interference with normal business performance, but also accurately captures the nonlinear level decay and error mutation of channel 3 at a specific time point through high time resolution cross-channel data comparison, providing a high-quality spatiotemporal synchronization signal input for subsequent identification of early charge leakage defects.

[0025] Step S200: Based on the charge retention level signal, construct a time-domain threshold drift feature that reflects the charge decay characteristics of the channel cell, and screen sensitive data retention windows associated with physical erase and programming cycles of flash memory blocks according to pressure aging conditions.

[0026] The channel cell charge decay characteristic can be described as the dynamic law of charge reduction over time in a specific flash memory cell during its retention period. This can serve as the physical basis for time-domain threshold drift characteristics, reflecting local material and interface quality. The time-domain threshold drift characteristic can be a quantized sequence based on the charge retention level signal, characterizing the cell threshold voltage shift over time. This can be used to characterize the charge leakage rate and for subsequent decay rate extraction and model mapping. In one specific embodiment, the time-domain threshold drift characteristic can map the level signal to an equivalent threshold voltage and calculate the offset or slope over a time window. The time-domain threshold drift characteristic can include, but is not limited to, linear drift slope, nonlinear decay exponent, and piecewise fitting residuals. The physical erase and programming cycle can be the actual number of erase and write operations performed on the flash memory block and its historical records. This can be used as a key correlation variable for screening sensitive data retention windows and identifying high-stress areas. The sensitive data retention window can be a data retention period strongly correlated with high erase / write cycles or specific stress conditions during stress aging. This can be used to focus on analyzing the level and error behavior during high-risk periods, improving the signal-to-noise ratio of anomaly detection. Furthermore, the sensitive data retention window can be used to filter out time intervals where charge retention performance is prone to degradation based on historical data from physical erasure and programming cycles. In one exemplary embodiment, the sensitive data retention window is used in conjunction with temporal threshold drift features, and extracting features within this window is more diagnostically valuable.

[0027] Constructing a time-domain threshold drift feature reflecting the charge decay characteristics of channel units can be achieved by converting the charge retention level signal into an equivalent threshold voltage sequence and calculating its time derivative or fitting parameters. Furthermore, this operation can be implemented by using sliding window linear regression to extract local slopes or by using an exponential decay function to fit the overall drift trend, thus achieving the technical effect of transforming the original level signal into a decay dynamic feature with clear physical meaning. Screening sensitive data retention windows related to the physical erasure and programming cycles of flash memory blocks based on pressure aging conditions can be achieved by combining erase / write count logs with aging stress intensity to identify the time period when charge retention performance is most prone to degradation. Furthermore, this operation can be achieved by setting the retention period of blocks above the erase / write count threshold as the sensitive window, or by dynamically calculating the window boundary based on a temperature-voltage joint stress model, thereby narrowing the analysis scope to high-risk periods and improving the efficiency and accuracy of anomaly detection. For example, under high-temperature aging stress of 85 degrees Celsius, for high-risk blocks with 3000 write / erase counts, the system sets this as a sensitive data retention window to focus the analysis range. Within this window, the monitored charge retention level signal is converted. For example, the 2450 mV level collected at hour 0 is equivalent to a 3.5 volt threshold voltage, and the 24-hour level drops to 2420 mV, equivalent to 3.45 volts. The local slope is calculated using sliding window linear regression to obtain an hourly decay of approximately 2.08 mV, or the overall decay constant is obtained by fitting an exponential decay function. Thus, the original voltage reading is successfully converted into a time-domain threshold drift decay rate feature with clear physical meaning. This operation, which combines write / erase threshold screening with mathematical fitting, not only narrows the analysis range to the high-risk period where charge retention performance is most prone to degradation, but also effectively eliminates environmental noise, realizing the accurate conversion of the original level signal into a quantitative indicator reflecting the dynamic characteristics of the physical degradation of the channel unit.

[0028] Step S300: Synchronously analyze the gradient distribution changes of the read bit error rate signal, combine the level fluctuations within the sensitive data maintenance window, and identify the abnormal leakage area of ​​the channel caused by oxide layer defect trap capture and electron tunneling effect.

[0029] The gradient distribution change of the bit error rate (BER) signal can be used to read the rate of change of the BER signal in the time dimension or the spatial distribution pattern of the difference sequence. This can be used to reveal the accelerating trend of BER growth and assist in identifying sudden leakage events. In a specific embodiment, the gradient distribution change of the BER signal can be performed on the BER time series using first- or second-order difference and expanded along the channel or address dimension. Level fluctuations can be non-monotonic or abnormal fluctuations in the charge retention level signal within the sensitive data retention window. This can be used as an indirect characterization of oxide layer defects or interface state activity, coupled with the BER gradient to determine anomalies. Oxide layer defect trap capture can be the unintended capture behavior of charge by electron traps formed by manufacturing defects in the tunnel oxide layer. This can be used as one of the microscopic physical mechanisms of charge leakage, leading to threshold voltage instability and increased BER. Electron tunneling effect can be the physical process by which electrons lose charge through the thin oxide layer via quantum tunneling. This can be used to exacerbate charge leakage in miniaturization processes and is a major cause of early data retention failure. Channel anomaly leakage regions can be physical areas within a specific flash memory channel where abnormal charge loss occurs due to oxide layer defect trapping or electron tunneling effects. These regions can serve as spatial benchmarks for subsequent logic address mapping and risk assessment. Furthermore, channel anomaly leakage regions can be located by synchronously identifying the coupling relationship between level fluctuations and bit error gradient changes. Channel anomaly leakage regions may include, but are not limited to, interface state-dominated regions, volume defect-dominated regions, and edge tunneling enhancement regions.

[0030] Synchronous analysis of the gradient distribution changes in the bit error rate (BER) signal can be achieved by differential processing of the BER time series and observing its spatial gradient pattern in the channel dimension. Further, this operation can calculate the first-order difference of the BER at adjacent time points, or use the Sobel operator to detect edge abrupt changes in the channel-time two-dimensional plane, thereby capturing the accelerated behavior of BER growth and identifying non-uniform degradation regions. Combining the level fluctuations within the sensitive data holding window, identifying channel abnormal leakage regions caused by oxide layer defect trapping and electron tunneling effects can be achieved by finding channel locations within the sensitive window where level fluctuations and BER gradients increase synchronously, and determining these as abnormal leakage regions. Further, this operation can be achieved by setting a threshold for the product of the level standard deviation and the BER gradient for region labeling, or by using clustering algorithms to group multi-dimensional feature points to identify abnormal clusters, thereby improving positioning accuracy through multi-source signal coupling and distinguishing between normal attenuation and defect-induced abnormal leakage. For example, within a set sensitive data holding window, the system performs differential processing on the read bit error rate time series. For instance, at the 48th hour, it detects a sudden increase in the bit error rate of channel 2 from 5.0 x 10⁻⁵ to 8.5 x 10⁻⁵. The system calculates that its first-order differential value is positive and significantly increased. Simultaneously, the Sobel operator is used to detect a significant gradient edge abrupt change at this location on the channel-time two-dimensional plane. Combined with the synchronously monitored charge holding level fluctuation signal, it is found that the standard deviation of the channel level surges to 15 mV and shows an increasing trend in the same direction as the bit error gradient. Based on this, the product of the standard deviation of the level and the bit error gradient exceeds a preset threshold, and channel 2 is determined to be an abnormal leakage region. This operation, which analyzes the changes in the bit error rate gradient distribution and couples it with the level fluctuation characteristics in a multidimensional way, successfully captures the non-uniform degradation caused by oxide layer defect traps and electron tunneling effects, effectively distinguishes between normal aging decay and abnormal leakage caused by physical defects, and significantly improves the accuracy of channel anomaly location.

[0031] Step S400: Establish the mapping relationship between the logical address of the abnormal leakage area, the decay rate of the time domain threshold drift characteristic and the preset leakage defect evolution dynamic model, and generate a quantitative assessment result of the early charge leakage location and risk tendency value of the target solid-state drive.

[0032] The logical address of the abnormal leakage region can be the corresponding identifier of the abnormal leakage region in the file system or logical block address space. This can be used to map physical defects to the logical layer, facilitating data migration or early warning in the upper-layer system. The decay rate can be the rate of change of the time-domain threshold drift feature within the sensitive data retention window. This can be used to quantify the severity of charge leakage and serve as an input variable for the dynamic model. In a specific embodiment, the decay rate can be obtained by fitting the slope or exponential decay parameter to the threshold drift curve. The leakage defect evolution dynamic model can be a mathematical model based on the physical degradation mechanism that describes the evolution of charge leakage over time. This can be used to correlate the decay rate with future risk tendencies, supporting trend prediction. Furthermore, the leakage defect evolution dynamic model can be constructed by combining parameters such as oxide layer defect density, tunneling probability, and interface state distribution to build differential equations or empirical formulas. The leakage defect evolution dynamic model can include, but is not limited to, the Arrhenius thermal activation model, the Weibull failure distribution model, and the physical substrate tunneling model. The early charge leakage location can be the logical or physical coordinates of the flash memory cell in the early stage of charge leakage, determined by quantitative evaluation. This can be used to provide precise location information for generating risk location descriptions in diagnostic reports. A risk propensity value can be an indicator of the probability or severity of an early charge leakage area developing into functional failure within a specific future timeframe, and can be used to support the risk level classification in a health status diagnostic report. In an exemplary embodiment, the risk propensity value can be calculated by a leakage defect evolution kinetic model based on the decay rate and the current state. The quantitative assessment result can be structured output data containing the location of early charge leakage and the corresponding risk propensity value, which can be used as direct input for generating a health status diagnostic report.

[0033] Establishing a mapping relationship between the logical address of an abnormal leakage region, the decay rate of the time-domain threshold drift characteristic, and a pre-defined dynamic model of leakage defect evolution can be achieved by inputting the logical address and measured decay rate of each abnormal region into the dynamic model and outputting the corresponding risk propensity value. Furthermore, this operation can be implemented by matching pre-stored model parameters using a lookup table or by solving differential equations online to obtain real-time risk values, thereby achieving a quantitative extrapolation from the current state to future risks and supporting trend prediction. Generating a quantitative assessment result of the early charge leakage location and risk propensity value of the target solid-state drive can be achieved by integrating the logical addresses and corresponding risk propensity values ​​of all abnormal regions to form a structured data record. Further, this operation can be implemented by storing the address-risk mapping in key-value pair form or by generating a probability distribution list with confidence intervals, thereby providing standardized input for diagnostic reports and ensuring the traceability and resolvability of the assessment results. For example, for the two identified abnormal leakage channel areas, the system matches the measured threshold decay rate of 0.12mV / PE cycle for channel 2 with logical address LBA0x12345678 and the measured decay rate of 0.07mV / PE cycle for channel 3 with logical address LBA0x87654321 with the corresponding parameters in the pre-existing leakage defect evolution dynamics model library. The risk propensity values ​​for the two areas are 0.87 and 0.32, respectively. The system then integrates the output into structured data records in the form of key-value pairs, storing the address, corresponding risk, and 95% confidence interval error range to generate quantitative evaluation results. This not only realizes the quantitative deduction from the current state to the failure probability after 1000 subsequent PE cycles through the mapping relationship between the measured decay rate and the dynamics model, but also provides a standardized and traceable input for solid-state drive fault diagnosis reports.

[0034] Step S500: Based on the location of early charge leakage and risk tendency value in the quantitative assessment results, generate a health status diagnosis report for the target solid-state drive.

[0035] The health status diagnostic report can be a structured document summarizing early charge leakage locations, risk propensity values, and maintenance recommendations, providing actionable reliability management data for the operation and maintenance system. The health status diagnostic report can include, but is not limited to, JSON format diagnostic summaries, visual heatmap reports, and API response bodies. Based on the early charge leakage locations and risk propensity values ​​from the quantitative assessment results, a health status diagnostic report for the target solid-state drive is generated. This can be done by classifying risk propensity values ​​and combining them with location information to generate text or structured output containing maintenance recommendations. Furthermore, this operation can automatically generate data migration, redundancy reservation, or replacement early warning recommendations based on risk levels, or output extended attributes compatible with the SMART standard, thereby achieving the technical effect of transforming technical analysis results into actionable decision-making data for operation and maintenance.

[0036] Taking data center SSD reliability monitoring as an example, the diagnostic method for early potential defects in solid-state drives in this embodiment can be as follows: In SSDs deployed in batches in the data center, the operation and maintenance system periodically triggers stress aging tests, and the firmware automatically collects the charge retention level and bit error rate timing data of preset test units in each channel; the background analysis engine constructs time-domain threshold drift characteristics and filters sensitive retention windows of high-write blocks; through synchronous analysis, it is found that a certain channel exhibits drastic level fluctuations and a sharp increase in bit error gradient after high-temperature aging, which is determined to be an abnormal leakage area caused by oxide layer defects; the system inputs the logical address of this area and the measured decay rate into the leakage defect evolution dynamics model to predict that its risk tendency value will reach the high-risk threshold within three months; finally, a diagnostic report is generated to recommend that the data in this area be migrated in advance and marked as a high-risk block to avoid subsequent business writes.

[0037] In one embodiment, the settings for the flash channel preset test unit include: Based on the physical architecture and stress aging conditions of flash memory chips, key lines and bit line regions that are prone to charge loss are identified, thus obtaining the critical regions. Within the critical area, the test unit allocation positions are selected by cross-selection based on the location of the maximum electric field intensity on the surface of the flash memory block and the location of the programming interference mode node, thus obtaining the initial allocation positions; Based on the conduction attenuation relationship between the charge holding level signal and the read error rate signal in the flash array, the relative spacing and logical mapping relationship of the test cells at the initial allocation position are adjusted to obtain the optimized layout of the preset test cells of the flash channel.

[0038] The physical architecture of the flash memory chip can be the physical design features such as the three-dimensional stacking structure, word line / bit line arrangement, and memory cell array organization within the NAND flash memory chip. These features can serve as the basis for identifying high-risk areas for charge loss and guide the spatial deployment of test units. In an exemplary embodiment, the physical architecture of the flash memory chip can be combined with stress aging conditions such as high temperature and high voltage to locate structurally sensitive areas. Key word lines and bit lines can be the intersection areas of word lines and bit lines that are prone to charge leakage under specific stress aging conditions due to electric field concentration or interference effects. These can be used to constitute the spatial boundaries of critical areas for focused test unit deployment. Furthermore, key word lines and bit lines can be used to identify high-stress word line layers and susceptible bit line segments through simulation or historical failure data analysis. For example, key word lines and bit lines can include, but are not limited to, high-electric-field regions on edge word lines, crosstalk-sensitive regions on center bit lines, and hot carrier injection regions on the top layer of the memory layer. Critical areas can be the set of physical spaces most prone to charge retention performance degradation, determined by the intersection of key word lines and bit lines. These can be used as candidate ranges for initial allocation of test units, improving the targeting of monitoring. Based on the physical architecture and stress aging conditions of flash memory chips, key word lines and bit line regions prone to charge loss are identified to obtain critical regions. This can be achieved by combining device structural parameters and aging stress types to locate word line layers and bit line segments with high electric field or high interference risk, and taking their intersection to form critical regions. Furthermore, this operation can be achieved by using TCAD simulation to predict the electric field distribution and mark high-risk word lines, and by statistically inferring easily failed bit line regions based on historical bad block distribution. This allows the monitoring range to be narrowed from the entire array to a high-risk subset, improving resource utilization efficiency.

[0039] The location of the maximum electric field intensity on the flash memory block surface can be the physical location where the local electric field intensity on the flash memory block surface reaches its peak during programming or reading operations. This location can be used to identify hotspots with the highest probability of charge tunneling, guiding test cell positioning. In one embodiment, the location of the maximum electric field intensity on the flash memory block surface can be obtained through device-level electric field simulation or back-calculation of process parameters. The location of the programming interference mode node can be a typical location where a threshold voltage shift occurs due to coupling interference from neighboring cells under a specific programming mode. This location can be used to reflect dynamic stress concentration points during data writing, and its intersection with the electric field peak point determines the initial allocation position. Furthermore, the programming interference mode node location identifies extreme points of interference energy distribution based on a programming interference model. The initial allocation position can be the candidate coordinates of test cells determined by the intersection of the location of the maximum electric field intensity and the programming interference mode node location within a critical region. This can be used as the starting point for optimized layout, ensuring coverage of high-risk points where multiple physical mechanisms are superimposed. Within the critical area, test unit allocation positions are selected by cross-referencing the location of the maximum electric field intensity on the flash memory block surface with the location of the programming interference mode node. This initial allocation position can be obtained by overlaying the electric field peak point map and the interference mode node map within the critical area, selecting overlapping or adjacent locations as initial test points. For example, this operation can be achieved by using a gridded scan to match the minimum Euclidean distance between two sets of positions and using Boolean operations to extract the intersection of the binarized hotspot map. This ensures that the test units cover hotspots affected by multiple physical degradation mechanisms, enhancing the ability to detect early anomalies.

[0040] The conduction attenuation relationship in a flash memory array can be understood as the charge holding level signal and read error rate signal weaken with increasing physical distance during transmission from the storage cell to the sensing circuit. This can be used to correct the impact of test cell spacing on signal fidelity and guide layout optimization. The relative spacing of test cells can be understood as the spatial interval between adjacent preset test cells in the physical array. This can affect the spatial independence and local representativeness of multi-source signals and needs to be adjusted according to the conduction attenuation relationship. The logical mapping relationship can be understood as the correspondence between the physical location of the test cells and their logical address space. This ensures that diagnostic results can be parsed and responded to by the upper-layer system and needs to be adjusted synchronously in layout optimization to maintain address continuity or traceability. Based on the conduction attenuation relationship of the charge holding level signal and read error rate signal in the flash memory array, the relative spacing and logical mapping relationship of the test cells at the initially allocated positions are adjusted to obtain the optimized layout of the preset test cells for the flash memory channel. This can be achieved by calculating the effective sensing radius based on the signal attenuation model, adjusting the test cell spacing to avoid signal aliasing, and remapping the logical address to maintain addressability. In one specific embodiment, this operation can set the minimum spacing to the distance corresponding to a signal attenuation of 90% threshold, and use hash mapping to map physical clusters into contiguous logic blocks to simplify firmware management. This allows for maintaining logical consistency while ensuring signal fidelity, improving the comparability and diagnostic reliability of cross-channel data. The optimized layout of the flash memory channel's preset test units can be the final test unit space configuration determined after key area screening, initial position cross-selection, and adjustment of spacing and mapping relationships. This can provide a high signal-to-noise ratio and highly representative monitoring point distribution, supporting the accuracy and sensitivity of subsequent multi-source timing analysis. Furthermore, the optimized layout of the flash memory channel's preset test units can include, but is not limited to, hotspot-dense layouts, uniform coverage layouts, and gradient sampling layouts.

[0041] Taking the pre-shipment reliability screening of 3D NAND SSDs as an example, the diagnostic method for early potential defects of solid-state drives in this embodiment can be as follows: before the 3D NAND chips are packaged, the manufacturer identifies the key area formed by the top word line and the edge bit line based on its 128-layer stacked architecture and high-temperature and high-voltage aging conditions. Within this area, the maximum surface electric field intensity is determined to be near WL127 through electric field simulation, and at the same time, the programming interference model shows that BL0-BL3 are modal nodes. The intersection of WL127 and BL2 is selected as the initial allocation position. Then, based on the signal attenuation characteristics of the sensor amplifier input, the spacing between adjacent test units is set to more than 4 physical pages, and their logical addresses are mapped to continuous diagnostic blocks. The resulting optimized layout enables the clear separation of local charge leakage signals in subsequent aging tests, significantly improving the early defect detection rate.

[0042] In one embodiment, a time-domain threshold drift feature reflecting the charge decay characteristics of the channel unit is constructed based on the charge retention level signal, including: Step 201: Perform time-frequency transformation on the charge holding level signal to obtain the time-domain voltage distribution of the charge holding level signal.

[0043] Time-frequency transformation (TF-F) can be a mathematical processing method that converts a one-dimensional time-series signal into a joint time- and frequency (or voltage) domain representation. It can be used to reveal the voltage distribution structure of the level signal in the time dimension, supporting subsequent ridgeline tracing. In this embodiment, TF-F can apply time-frequency analysis tools such as short-time Fourier transform, wavelet transform, or Wigner-Ville distribution to the charge-holding level signal. The time-domain voltage distribution can be a two-dimensional representation obtained after TF-F, characterizing the energy or probability density distribution of the charge-holding level in the time-voltage plane. It can be used to visualize and quantify the evolution trajectory of the level signal, providing a data basis for extracting the main drift trend. For example, the time-domain voltage distribution can be composed of the amplitude or power spectrum output by the TF-F, with time on the horizontal axis and equivalent voltage value on the vertical axis. In an exemplary embodiment, the time-domain voltage distribution can include spectral representation, rearranged time-frequency plot, sparse time-frequency representation, etc. Performing TF-F to the charge-holding level signal to obtain its time-domain voltage distribution can be achieved by inputting the original one-dimensional level time series into a time-frequency analysis algorithm and outputting a two-dimensional time-voltage distribution plot. Furthermore, this operation can be achieved by using Morlet wavelet transform to obtain high time resolution voltage distribution and using synchronous compressed wavelet transform to enhance ridge clarity, thereby making the implicit dynamic drift structure explicit and improving the observability of subtle decay trends.

[0044] Step 202: In the time-domain voltage distribution, based on the continuity of the voltage distribution, trace the evolution path of the voltage level in the concentrated voltage region over time to obtain the initial drift ridge.

[0045] The voltage concentration region can be a local area in the time-domain voltage distribution where the energy or density is significantly higher than that of the neighboring region. It can represent the dominant level state of most cells at a specific time, reflecting the group charge maintenance behavior. In one specific embodiment, the continuous evolution of the voltage concentration region over time forms the basis of the initial drift ridge. The evolution path of the level over time can be a continuous trajectory of the voltage concentration region along the time axis in the time-domain voltage distribution. It can be used to characterize the macroscopic dynamic trend of charge decay and to construct the initial drift ridge. Further, the evolution path of the level over time can connect the centers of the voltage concentration regions at adjacent time points through connectivity analysis or path tracing algorithms. The initial drift ridge can be a continuous curve extracted from the time-domain voltage distribution, characterizing the main decay trend of the channel cell group level. It can be used as a rough estimate of the original drift characteristics, preserving the dominant physical degradation component and suppressing random noise. Exemplarily, the initial drift ridge can be formed as a continuous ridge based on the evolution path of the voltage concentration region through smoothing or interpolation. In an exemplary embodiment, the initial drift ridge may include a maximum energy ridge, a connected component centerline, a probability density peak trajectory, etc. In time-domain voltage distribution, based on the continuity of voltage distribution, the evolution path of the voltage concentration region over time is traced to obtain the initial drift ridge. This can be achieved by identifying the voltage concentration regions in each time slice and connecting them along the time dimension to form a continuous path. Furthermore, this operation can be implemented by tracing the path based on morphological connected components and using a dynamic programming algorithm to find the optimal continuous trajectory. This allows the main drift trend dominated by physical degradation to be separated from the noisy background, thereby improving the signal-to-noise ratio.

[0046] Step 203: Combining the known charge loss response mode of the target solid-state drive under pressure aging conditions, the level value of the initial drift ridge is trend-fitted to obtain the time-domain threshold drift characteristics that reflect the charge decay characteristics of the channel unit.

[0047] The known charge loss response pattern can be an empirical or physical correlation between flash memory cell charge loss and stress parameters (such as erase / write cycles, temperature, and hold time) under specific pressure aging conditions. This pattern can provide physical constraints for trend fitting, ensuring that the generated time-domain threshold drift characteristics conform to the actual degradation mechanism. In this embodiment, the known charge loss response pattern can be obtained through modeling based on historical aging experimental data or by pre-setting a function form based on the device's physical mechanism. Further, the known charge loss response pattern can include thermal activation exponential models, tunneling current accumulation models, interface state trapping saturation models, etc. Trend fitting can be a process of parametrically approximating the initial drift ridge under the constraints of the known charge loss response pattern. This can be used to generate time-domain threshold drift characteristics with clear physical meaning and predictive ability. For example, trend fitting can use least squares, Bayesian regression, or physical equation inversion to fit the ridge level value. In a specific embodiment, trend fitting can use a pre-set function family (such as exponential decay) for parameter fitting and a data-driven neural network to learn the mapping relationship under physical constraints.

[0048] By combining the known charge loss response patterns of the target solid-state drive under stress aging conditions, trend fitting is performed on the level values ​​of the initial drift ridge to obtain time-domain threshold drift characteristics reflecting the charge decay characteristics of channel cells. This can be achieved by using the initial drift ridge as observation data and performing parameter optimization fitting within the function space of the known charge loss response patterns. Furthermore, this operation can be implemented by fixing the response pattern function form and fitting only the amplitude and time constant, and jointly optimizing the pattern parameters and the ridge matching degree to adapt to individual differences. This allows the generation of time-domain threshold drift characteristics that conform to physical laws and have extrapolation capabilities, supporting subsequent risk assessment.

[0049] Taking the aging screening of a three-dimensional NAND SSD production line as an example, the diagnostic method for early potential defects of solid-state drives in this embodiment can be as follows: During the aging test stage before the SSD leaves the factory, the system applies high temperature and high voltage stress to each chip and collects the charge holding level signal of each channel test unit; the background processing module performs continuous wavelet time-frequency transformation on the signal to generate a time-domain voltage distribution; the algorithm automatically tracks the continuous trajectory of the high-energy region in the distribution to form an initial drift ridge; then, combined with the known charge loss response mode of the process node (such as the exponential decay law related to the P / E cycle number), the ridge is physically constrained and fitted to output standardized time-domain threshold drift characteristics; if the decay rate fitted by a certain channel deviates significantly from the mean, it is marked as an early leakage risk unit to avoid entering the high-end storage market.

[0050] In one embodiment, a sensitive data retention window associated with physical erase and programming cycles of flash memory blocks is selected based on stress aging conditions, including: Step 204: Analyze the pressure aging conditions, separate the typical stress components corresponding to the erase and programming operations of the flash memory block, and obtain the typical stress time series; Step 205: Correspond the time phase of the typical stress time series with the level change phase of the time domain threshold drift characteristic to establish the time synchronization relationship between stress excitation and charge level response; Step 206: Based on the time synchronization relationship, analyze the fluctuation period of each window level and the change period of typical stress components in the time domain threshold drift characteristics; identify the windows with stable period correspondence as sensitive data holding windows.

[0051] In this context, a flash memory block can be the smallest physical unit in NAND flash memory capable of independently performing an erase operation. It can serve as the fundamental object of physical erase and programming cycles, and its operational history is used to correlate stress components. Typical stress components can be stress elements with clear temporal characteristics directly related to the flash memory block's erase or programming operations under pressure aging conditions. They can be used to characterize operational excitation sources that induce specific charge decay behavior. In an exemplary embodiment, typical stress components can be extracted by temporal decomposition of operational logs, voltage waveforms, or temperature records under pressure aging conditions. Further, typical stress components can include, but are not limited to, programming pulse voltage sequences, erase pulse width sequences, and read interference cumulative count sequences. A typical stress time series can be a discrete or continuous signal sequence formed by arranging typical stress components in chronological order. It can be used to provide a temporal reference for stress excitation and to establish a synchronization relationship with charge response. Exemplarily, a typical stress time series can be obtained by mapping the timestamps and intensity parameters of erase / programming operations to a sequence of excitation events on a time axis. In a specific embodiment, a causal relationship is established between the typical stress time series and time-domain threshold drift characteristics through time phase alignment. By analyzing the pressure aging conditions, typical stress components corresponding to erase and programming operations on the flash memory blocks are separated, resulting in typical stress time series. This can be achieved by analyzing firmware operation logs or electrical monitoring data during pressure aging, extracting time-series events strongly correlated with P / E operations, and constructing a sequence. Furthermore, this operation can be achieved by constructing a binary excitation sequence based on the timestamps and command types of the operation logs, and by identifying programming / erase pulses from the power supply current waveform and quantifying their intensity to form a continuous sequence. This decouples the composite aging stress into an interpretable operational excitation source, improving the physical relevance of subsequent analyses.

[0052] The phase of charge level change can be the periodic or event-triggered temporal position of the charge level change over time in the time-domain threshold drift feature. It can be used as a time marker for the charge level response and for phase matching with stress excitation. Stress excitation can be an external operational perturbation applied to the flash memory cell, represented by a typical stress time series. It can be used as an inducement for charge leakage mechanisms and as an input variable in time synchronization analysis. The charge level response can be the dynamic change behavior of the charge holding level exhibited by the flash memory cell under stress excitation. It can be used as the system output response, reflecting the cell's sensitivity to operational stress. The time synchronization relationship can be a time-series alignment mapping established between the typical stress time series and the phase of charge level change. It can be used to reveal the physical causal relationship between stress excitation and charge response, supporting window filtering. In a specific embodiment, the time synchronization relationship can achieve phase matching through cross-correlation analysis, event-triggered alignment, or dynamic time warping. Furthermore, the time synchronization relationship connects the typical stress time series and the time-domain threshold drift feature, forming the basis for periodic comparison.

[0053] By mapping the time phase of a typical stress time series to the phase of the level change characteristic of the time-domain threshold drift, a time synchronization relationship between stress excitation and charge level response can be established. This can be achieved by performing phase alignment processing on the two time series, matching each stress event with the level change it induces on the time axis. Furthermore, this operation can be achieved by aligning the level response after multiple P / E cycles using an event-triggered averaging method and using a cross-correlation function to find the maximum similarity delay to calibrate the phase. This allows the establishment of a causal time chain between operational stress and charge decay, enhancing the interpretability of the mechanism.

[0054] The fluctuation period of the window level can be the time interval during which the level signal exhibits repetitive fluctuations within a local time window of the time-domain threshold drift characteristic. This can be used to characterize the response characteristics of charge decay to periodic stress. For example, the fluctuation period of the window level can include, but is not limited to, quasi-periodic oscillations, damped decaying oscillations, and random walk-type fluctuations. The variation period of the typical stress component can be the time interval between adjacent similar operational events in a typical stress time series or the repetitive pattern period. This can be used to reflect the rhythmic characteristics of external stress and to couple with the level response period for judgment. In an exemplary embodiment, the variation period of the typical stress component can include, but is not limited to, fixed-interval programming cycles, burst erase / write group cycles, and mixed read / write alternation cycles. Based on the time synchronization relationship, analyzing the fluctuation period of each window level and the variation period of the typical stress component in the time-domain threshold drift characteristic can be achieved by calculating the dominant periods of level fluctuations and stress changes on the synchronized time axis and comparing their consistency. Furthermore, this operation can be implemented by extracting the dominant frequency component through Fourier transform for comparison and using an autocorrelation function to detect the repetitive pattern period within the window, thereby identifying the following behavior of the charge response to operational stress and distinguishing between normal decay and abnormal leakage. Windows exhibiting stable periodic correspondences are designated as sensitive data retention windows. This can be achieved by marking a time window as sensitive when the period of level fluctuations and the period of stress change are statistically significantly consistent. Furthermore, this operation can be implemented by setting a period ratio between 0.9 and 1.1 and a correlation coefficient greater than a threshold as a stable correspondence, and by using clustering methods to identify co-occurring periodic coupling windows across multiple channels. This allows for precise location of critical periods when the charge leakage mechanism is activated by operational stress, improving the signal-to-noise ratio of subsequent anomaly detection.

[0055] For example, in the scenario of accelerated life testing of industrial-grade SSDs, the method for diagnosing early potential defects of solid-state drives in this embodiment can be to apply mixed read and write loads to industrial SSDs under high temperature and high humidity conditions. The system records the time and voltage parameters of each programming and erasing operation to form a typical stress time series. At the same time, the charge retention level of the preset test unit is collected to construct time-domain threshold drift characteristics. Through event-triggered alignment, it is found that the level of a certain block shows a regular decrease within 48 hours after each large block erase, and its fluctuation period is highly consistent with the erase operation interval. Based on this, the system determines the 48-hour window as a sensitive data retention window and focuses on monitoring the bit error gradient changes within this window in subsequent analysis, successfully identifying the early leakage area caused by the interface state trap before the bit error rate exceeds the standard.

[0056] In one embodiment, the gradient distribution changes of the read bit error rate signal are analyzed synchronously, and the level fluctuations within the sensitive data maintenance window are combined to identify abnormal channel leakage regions caused by oxide layer defect trapping and electron tunneling effects, including: Step 301: Based on the read error rate signals deployed in preset test units of different flash memory channels, construct an error field that reflects the spatial distribution of error on the flash disk. Step 302: Analyze the bit error field on the continuous time series, calculate the rate of change of the bit error difference between adjacent logic cells of the flash disk over time, and obtain the gradient distribution change; Step 303: Extract the regional features from the gradient distribution changes where the error rate exceeds the rate of change caused by the background read interference of the flash disk, and obtain the candidate abnormal leakage regions; Step 304: Obtain the time series of level fluctuations within the same time period within the sensitive data holding window; correlate the time phase of the candidate abnormal leakage area with the changing phase of the level fluctuation time series, analyze the time correspondence and intensity ratio between the changes of the candidate abnormal leakage area and the level fluctuation time series, and obtain the electrical-miscoupling correlation relationship; Step 305: Based on the electrical-miscoupling correlation, the sub-regions in the candidate abnormal leakage region whose changes have a stable intensity ratio relationship with the time series of level fluctuations are identified as abnormal leakage regions of the target solid-state drive.

[0057] The bit error rate field (BER) can be a two-dimensional or three-dimensional spatial distribution representation constructed by reading the bit error rate signals from each preset test unit in the logical address space of the flash disk. It can be used to achieve spatial visualization modeling of bit errors at the disk level and support the identification of local abnormal regions. In this embodiment, the BER can map the read bit error rates of preset test units in different flash channels to a unified coordinate system according to their logical positions, forming a spatially continuous bit error intensity map. Furthermore, the BER can include, but is not limited to, page-level bit error heatmaps, block-level bit error gradient fields, and channel-address joint bit error tensors. The continuous time series can be a time-series set composed of multiple frames of BER data collected continuously in time. It can be used to support the analysis of the dynamic evolution of the spatial distribution of bit errors and to extract time derivative features. Adjacent logical units on the flash disk can be two or more flash memory storage units that are physically or logically adjacent in the logical address space. They can be used as basic unit pairs for calculating bit error differences and for constructing spatial gradients.

[0058] The rate of change of the bit error rate difference over time can represent the speed at which the difference in bit error rate between adjacent logic units changes at consecutive time points. It can reflect the development speed of local bit error rate non-uniformity and is a core component of gradient distribution changes. In an exemplary embodiment, the rate of change of the bit error rate difference over time can be calculated by first calculating the bit error rate difference between spatially adjacent units, and then performing time differentiation or differential processing on the difference sequence. Background read interference can be non-defective bit error fluctuations caused by voltage noise, reference level drift, or ECC boundary effects during normal read operations. It can be used as a benchmark threshold for the rate of change of the bit error rate, distinguishing between normal fluctuations and abnormal degradation. The regional characteristics of the rate of change can be local spatial patterns where the rate of change of the bit error rate is significantly higher than the background interference level in the gradient distribution changes. This can be used to identify candidate locations of potential abnormal leaks, serving as input for subsequent coupling verification. Candidate abnormal leak regions can be suspected charge leak regions initially screened based on the bit error rate gradient change rate exceeding the background interference threshold. This can be used to narrow down the anomaly search range and provide a high-confidence candidate set for electrical-error coupling analysis. For example, candidate abnormal leakage regions need to be phase- and intensity-aligned with the time series of level fluctuations to verify their physical authenticity.

[0059] The level fluctuation time series can be a sequence of the charge holding level signal of a preset test unit changing over time within a sensitive data holding window. It can provide dynamic information on charge decay, which can be used for coupling and correlation analysis with bit error rate changes. In one specific embodiment, the phase and intensity ratio of the level fluctuation time series are used to verify the physical consistency of candidate anomaly leakage regions. The time phase of the candidate anomaly leakage region can be the time point or period when the peak or inflection point of the bit error rate change rate occurs within the candidate region, which can be used as a time reference for alignment with the level fluctuation time series. The phase change of the level fluctuation time series can be the time point or period when significant fluctuations (such as sharp drops or oscillations) occur in the level fluctuation time series, which can be used to determine whether they occur synchronously with bit error anomalies. The time correspondence can be the coincidence or lag matching relationship between the time phase of the candidate anomaly leakage region and the phase change of the level fluctuation time series on the time axis, which can be used to verify whether bit error anomalies and charge loss are causally synchronized. The intensity ratio can be the quantified ratio or correlation coefficient between the bit error rate change amplitude and the level fluctuation amplitude, which can be used to measure the linear or nonlinear coupling strength of the electrical-error response.

[0060] Electrical-error coupling correlation can be a comprehensive correlation model between candidate abnormal leakage regions and the time series of level fluctuations in terms of temporal correspondence and intensity ratio. This can be used as a criterion for determining genuine abnormal leakage, eliminating false positives that only show bit error anomalies without charge loss support. Furthermore, electrical-error coupling correlation can include, but is not limited to, phase-locked coupling, constant-ratio coupling, and delay-response coupling. A sub-region with a stable intensity ratio can be a local area within the candidate abnormal leakage region where the bit error rate change and level fluctuation maintain a consistent intensity ratio across multiple time points. This can be used as the basis for final determination of an abnormal leakage region, ensuring the consistency of the physical mechanism. Based on the read bit error rate signals deployed in preset test units of different flash memory channels, an error field reflecting the spatial distribution of flash disk errors can be constructed. This can be achieved by mapping the read bit error rate of each preset test unit to a unified spatial grid according to its logical address, forming a spatially continuous bit error intensity distribution map. Furthermore, this operation can be achieved by using interpolation algorithms to fill in the bit error values ​​of unmonitored units to form a continuous field, or by directly constructing a non-uniform grid bit error field using sparse sampling points. This allows for the conversion of bit errors from discrete channel data to a disk-level spatial field, supporting spatial gradient calculation. Analyzing the bit error field over a continuous time series and calculating the rate of change of the bit error difference between adjacent logic units on the flash disk over time yields the gradient distribution change. This can be achieved by performing time differentiation on the bit error difference between adjacent logic units in the bit error field of a continuous time frame, generating a spatiotemporal gradient tensor. Further, this operation can be achieved by using the central difference method to calculate the time rate of change and employing Sobel or Laplacian operators to simultaneously extract spatial and temporal gradients. This allows for capturing the dynamic development of local bit error non-uniformity and highlighting abnormal degradation regions.

[0061] From the gradient distribution changes, regional features are extracted where the bit error rate exceeds the rate of change caused by background read interference on the flash disk, resulting in candidate abnormal leakage regions. This can be achieved by setting a dynamic threshold based on background interference statistics to screen regions with gradient amplitude exceeding the limit as candidates. Furthermore, this operation can be implemented by estimating the background interference distribution using historical stress-free data and setting a 3σ threshold, then adaptively adjusting the threshold online to adapt to different aging stages. This effectively suppresses normal read noise interference and improves the reliability of candidate regions. Obtaining the sensitive data and maintaining the time series of level fluctuations within the same time period can be achieved by extracting level signals aligned with the time window of the candidate abnormal regions from the collected cross-channel multi-source time-series monitoring dataset. Further, this operation can be achieved by low-pass filtering the original level signals to suppress high-frequency noise and using a sliding window averaging to improve the signal-to-noise ratio. This ensures that the bit error rate analysis is performed on the same time scale, guaranteeing the effectiveness of phase alignment.

[0062] The temporal phase of candidate abnormal leakage regions is correlated with the phase changes in the time series of level fluctuations. The temporal correspondence and intensity ratio between these two relationships are analyzed to obtain the electrical-to-error coupling correlation. This can be achieved through cross-correlation analysis or dynamic time warping of the temporal characteristics of both regions, calculating phase shift and amplitude ratio. Further, this operation assesses the stability of the intensity ratio by calculating the Pearson correlation coefficient and uses the peak value of the cross-correlation function to locate the optimal time alignment point. This establishes physical evidence of the correlation between bit error anomalies and charge loss, excluding isolated bit error events. Based on the electrical-to-error coupling correlation, sub-regions in the candidate abnormal leakage regions that exhibit a stable intensity ratio relationship with the time series of level fluctuations are identified as abnormal leakage regions of the target SSD. This can be achieved by selecting sub-regions that meet preset phase tolerance and proportional stability conditions, outputting the final abnormal leakage region. Further, this operation can determine stability by setting a standard deviation threshold for proportional changes and verify the repeatability of the coupling relationship by combining data from multiple aging cycles. This ensures that the identification results are supported by the physical mechanism of charge leakage, improving the specificity and accuracy of the location.

[0063] Taking early failure warning of industrial control SSD as an example, the diagnostic method for early potential defects of solid-state drives in this embodiment can be that after the SSD operating in the industrial field undergoes high-temperature cycling, the system automatically triggers the diagnostic process: First, a disk-level bit error field is constructed based on the bit error rate of each channel test unit; then, the bit error field of five consecutive aging cycles is analyzed, the time change rate of the bit error difference between adjacent logic blocks is calculated, and it is found that the gradient of a certain region is continuously more than 3 times higher than the background interference, and it is marked as a candidate abnormal leakage region; at the same time, the time series of level fluctuations in the corresponding time period of this region is extracted, and it is found that it drops sharply in the third cycle, which is completely aligned with the peak time of the bit error gradient, and the bit error increase and level decrease are in a stable ratio of 1.8:1; after electrical-error coupling verification, this sub-region is determined to be a real abnormal leakage region, and the system reports its logical address to the health diagnosis module to generate a high-risk warning.

[0064] In one embodiment, the dynamic model of leakage defect evolution is constructed through the following steps: Based on data retention test data of the same flash memory media as the target solid-state drive under pressure aging conditions, the constitutive relationship between charge leakage rate and stress intensity factor range was determined.

[0065] The same flash memory medium can be a NAND flash memory chip or wafer using the same process, structure, and material system as the target solid-state drive. It can serve as the physical basis for constructing a dynamic model of leakage defect evolution, ensuring that the experimental data is consistent with the actual product degradation behavior. Data retention test data can be records of charge decay and bit error rate evolution obtained from long-term data retention tests on the flash memory medium under pressure aging conditions. This can be used to calibrate the quantitative relationship between charge leakage rate and external stress, supporting the establishment of constitutive relations. Furthermore, data retention test data can be obtained by storing known data under stress conditions such as high temperature and high voltage, and periodically reading it to measure threshold drift or bit error rate changes. The charge leakage rate can be the rate at which the stored charge in the flash memory cell is lost per unit time. It can be used as the dependent variable in the constitutive relation, reflecting the degradation intensity of the device under specific stress. The stress intensity factor range can be a quantitative range characterizing the comprehensive aging stress level applied to the flash memory device, covering the coupling effect of multi-dimensional parameters such as temperature, voltage, and erase / write cycles. It can be used as the independent variable in the constitutive relation to correlate external stress conditions with internal charge leakage behavior. Constitutive relations can be physical empirical formulas describing the functional dependence between charge leakage rate and the range of stress intensity factors. They can be used to lay the physical interpretability foundation for dynamic models of leakage defect evolution, bridging macroscopic observations and microscopic mechanisms. Furthermore, constitutive relations can be obtained through regression fitting or mechanistic modeling of data-preserving experimental data.

[0066] Based on data retention test data of the same flash memory media as the target SSD under pressure aging conditions, the constitutive relationship between charge leakage rate and stress intensity factor range is determined. This can be achieved by applying multiple levels of stress aging to flash memory using the same process, measuring the charge leakage rate under different conditions, and fitting its functional relationship with the stress intensity factor range. Furthermore, this operation can be achieved by using an Arrhenius-Eyring composite model to fit the temperature-voltage coupling effect and using machine learning regression to replace the analytical expression, thereby establishing a physically consistent fundamental degradation model and avoiding cross-process extrapolation errors.

[0067] In the constitutive relation, a damping attenuation coefficient related to the energy dissipation of the level is introduced as a correction term to establish a two-dimensional parameter space with the stress intensity factor range and the damping attenuation coefficient as coordinates, and the boundary between the stable development region and the unstable development region of leakage is delineated in this space.

[0068] Among them, level energy dissipation can be the energy loss phenomenon caused by interface states, trap capture, or tunneling path non-ideals during charge retention. It can be used to reflect the modulation effect of local material defects on charge stability, providing a basis for introducing non-ideal corrections into the model. The damping attenuation coefficient can be a dimensionless correction parameter used to quantify the suppression or acceleration effect of level energy dissipation on the charge leakage process. It can be used as a correction term in the constitutive relation to improve the model's adaptability to non-uniform degradation behavior. Furthermore, the damping attenuation coefficient can be obtained by analyzing the amplitude and frequency characteristics of the level signal fluctuation and combining it with the energy dissipation model inversion. For example, the damping attenuation coefficient can include, but is not limited to, interface state damping coefficient, bulk defect damping coefficient, tunneling path damping coefficient, etc. The correction term can be an additional function or parameter introduced into the original constitutive relation to compensate for non-ideal physical effects. It can be used to enhance the model's fitting accuracy to the complex degradation behavior of actual devices. The two-dimensional parameter space can be a planar coordinate system constructed with the stress intensity factor range as the horizontal axis and the damping attenuation coefficient as the vertical axis. It can be used to provide a geometric discrimination framework for classifying leakage behavior and support the visual division of stable and unstable regions. The stress intensity factor range and damping attenuation coefficient can be two independent dimensional variables constituting the two-dimensional parameter space. They can jointly determine the dynamic characteristics of charge leakage evolution and are the core inputs of the criterion system. The stable development region of leakage can be a parameter combination region in the two-dimensional parameter space where the charge leakage rate is slow and predictable, and will not lead to sudden failure. It can be used to correspond to low-risk states and support long-term data retention. The unstable development region can be a parameter combination region in the two-dimensional parameter space where charge leakage exhibits accelerated, abrupt, or irreversible degradation. It can be used to correspond to high-risk states and indicate the possibility of early functional failure. The boundary can be a critical curve or interface separating the stable and unstable development regions of leakage. It can be used as a threshold for risk assessment, determining the jump point of the risk tendency value. Furthermore, the boundary can be determined through clustering of large amounts of experimental data or stability theory analysis.

[0069] In the constitutive relation, a damping attenuation coefficient related to the energy dissipation of the signal level is introduced as a correction term. This can be achieved by analyzing the time-domain fluctuation characteristics of the signal level, extracting energy dissipation indices, constructing the damping attenuation coefficient, and embedding it into the constitutive relation. Furthermore, this operation can be implemented by calculating the damping coefficient based on the power spectral density of the signal level and extracting transient dissipation characteristics using wavelet transform to construct the correction term. This can compensate for model deviations caused by local material inhomogeneities and improve sensitivity to abnormal leaks. A two-dimensional parameter space is established with the stress intensity factor range and the damping attenuation coefficient as coordinates. This can be achieved by using the two parameters as orthogonal coordinate axes to construct a planar mapping framework for subsequent region division. Further, this operation can be achieved by using normalization to ensure the two parameters have consistent dimensions and introducing polar coordinate transformation to optimize the spatial distribution. This allows for dimensionality reduction visualization and structured management of multi-dimensional degradation factors. The boundary between the stable and unstable development regions of the leak is delineated in this space. This can be done by labeling sample point attributes and fitting classification boundaries based on whether sudden bit errors or threshold jumps occur during long-term testing. Furthermore, this operation can be achieved by using support vector machines to determine the maximum margin boundary and setting the critical exponential curve based on phase transition theory, thereby forming a clear risk discrimination criterion and supporting the conversion from continuous parameters to discrete risk levels.

[0070] By integrating the two-dimensional parameter space and boundary, a criterion system for quantitatively assessing the risk of charge leakage is constructed, resulting in a dynamic model of leakage defect evolution.

[0071] The criterion system can be a set of rules, composed of a two-dimensional parameter space and its internal boundaries, used to determine the risk level of charge leakage. This can be used to map measured decay rates and damping characteristics into operable risk propensity values, supporting quantitative assessment. Integrating the two-dimensional parameter space and boundaries to form a criterion system for quantitatively assessing charge leakage risk propensity values ​​can be achieved by gridding the parameter space, assigning a risk propensity value to each region, and forming a lookup table or function mapping. Furthermore, this operation can be achieved by constructing a neural network to replace the lookup table for continuous risk output and by using fuzzy logic to handle uncertainties near the boundaries, thus enabling end-to-end mapping from physical parameters to risk output and supporting automated assessment. Obtaining the leakage defect evolution dynamic model can be achieved by encapsulating the criterion system as a callable module, where the input decay rate and damping characteristics output a risk propensity value. Furthermore, this operation can be achieved by integrating it into an SSD controller as a firmware library, serving as an inference engine for cloud-based diagnostic services, thereby forming a core diagnostic model with physical mechanism support and predictable development trends.

[0072] Taking the reliability modeling of the 3D NAND process verification stage as an example, the diagnostic method for early potential defects in solid-state drives in this embodiment can be as follows: Before the mass production of new 3D NAND flash memory, the R&D team uses wafers from the same batch to fabricate test chips, conducts data retention tests under high temperature and high voltage, and records the charge leakage rate under different stress combinations; by analyzing the fluctuation of the level signal, the damping attenuation coefficient is extracted, a constitutive relationship is constructed, and a correction term is introduced; then, a two-dimensional parameter space is established with the stress intensity factor range and the damping attenuation coefficient as the axis, and the boundaries of stable and unstable regions are delineated according to whether sudden bit errors occur; finally, this space and boundary are integrated into a criterion system to form a leakage defect evolution dynamic model; this model is integrated into the SSD firmware, and after subsequent product deployment, the measured attenuation rate and damping characteristics are mapped to risk tendency values ​​in real time to provide early warning of potential early failure blocks.

[0073] In one embodiment, establishing a mapping relationship between the logical address of the abnormal leakage region, the decay rate of the time-domain threshold drift characteristic, and a preset leakage defect evolution dynamics model includes: For abnormal leakage areas, based on their logical addresses and the principal stress trajectory of the target solid-state drive under pressure aging conditions, the potential leakage initiation direction and local stress concentration factor are determined, and local stress state parameters are obtained. The amplitude of the time-domain threshold drift feature within the sensitive data holding window is fitted over time, and the percentage decrease in amplitude per unit time is calculated to obtain the energy decay rate. Based on the local stress state parameters and energy decay rate, a feature point characterizing the current flash memory medium state is determined in the parameter space defined by the leakage defect evolution dynamics model. Based on the relative relationship between the position of feature points in the parameter space and the boundary of the stable development region of leakage defined by the model, the tendency of abnormal leakage regions to develop into severe charge leakage defects is quantified, and the mapping relationship is obtained.

[0074] The principal stress trajectory can be the path of the main environmental or operational stresses experienced by the target solid-state drive during pressure aging, which changes over time. It can be used to correlate logical addresses with physical stress history, supporting the calculation of local stress state parameters. In an exemplary embodiment, the principal stress trajectory can be composed of time series of stress parameters such as temperature, voltage, and erase / write frequency, and obtained through aging test logs. The potential leakage initiation direction can be the most likely physical location or crystal orientation of charge leakage inferred from the logical address and principal stress trajectory. It can be used to characterize the spatial orientation of defect initiation mechanisms and assist in identifying weak areas of the oxide layer. Furthermore, the potential leakage initiation direction can be obtained by combining the flash memory chip layout and stress distribution model to deduce the physical location and stress direction from the logical address. The local stress concentration factor can be a dimensionless parameter reflecting the degree of stress amplification caused by process fluctuations or structural discontinuities in a specific abnormal leakage area. It can be used to quantify the deteriorating effect of the local microenvironment on charge retention performance. For example, the local stress concentration factor can be estimated based on the chip layout, material elastic modulus, and principal stress trajectory through finite element analysis or empirical formulas. In one specific embodiment, the local stress concentration factor may include, but is not limited to, the thermal stress concentration factor, the electrical stress concentration factor, and the mechanical residual stress concentration factor.

[0075] The local stress state parameter can be a multidimensional feature vector describing the physical microenvironment of the abnormal region, formed by combining the potential leakage initiation direction and the local stress concentration factor. It can be used as one of the inputs to the dynamic model of leakage defect evolution, characterizing the intensity of external inducing factors. Furthermore, the local stress state parameter can encode directional information into a unit vector and combine it with the stress concentration factor in a weighted manner. In an exemplary embodiment, the local stress state parameter, together with the energy decay rate, determines the position of the feature point in the parameter space. Based on the logical address of the abnormal leakage region and the principal stress trajectory of the target solid-state drive under pressure aging conditions, the potential leakage initiation direction and local stress concentration factor are determined. This can be achieved by mapping the logical address to the physical layout and combining the temporal-spatial distribution of the principal stress trajectory to infer the local force direction and amplification effect. Further, this operation can be achieved by using a pre-stored chip layout database for address-location reverse lookup and superimposing stress field simulation, or by using simplified geometric rules to estimate the direction and concentration factor, thereby introducing a mechanical perspective and making the stress state quantifiable. The obtained local stress state parameter can be a structured vector formed by integrating the potential leakage initiation direction and the local stress concentration factor. Furthermore, this operation can be achieved by encoding the direction as a spherical coordinate angle or by using the lumped factor as a weight modulating the direction vector, thereby generating a standardized stress description that can be used as model input.

[0076] The amplitude of the time-domain threshold drift feature can be the absolute value of the signal strength or offset presented by the time-domain threshold drift feature within the sensitive data hold window, and can be used as the basic raw data for fitting the energy decay rate. The percentage decrease in amplitude per unit time can be the relative rate at which the amplitude of the time-domain threshold drift feature decays over time within the sensitive data hold window, and can be used as an intermediate computational quantity to derive the energy decay rate. The energy decay rate can be a quantitative indicator characterizing the speed of energy dissipation during charge loss, derived from the time decay ratio of the amplitude of the time-domain threshold drift feature, and can be used to reflect the dynamic response intensity of charge leakage, serving as a key input to the dynamic model. In one embodiment, the energy decay rate can be obtained by performing exponential or power-law fitting on the amplitude-time curve and extracting the normalized decay coefficient.

[0077] Fitting the amplitude variation of the time-domain threshold drift characteristic within the sensitive data holding window can be achieved by selecting time-amplitude data points within the sensitive window and applying a mathematical function for curve fitting. Furthermore, this operation can be implemented by fitting an exponential function using the least squares method or by approximating non-stationary decay using piecewise linear regression, thereby extracting the regularity parameters of charge decay. Calculating the percentage decrease in amplitude per unit time yields the energy decay rate, which can be obtained by differentiating or differencing the fitted function and normalizing to obtain the relative decay rate. Further, this operation can be achieved by calculating the reciprocal of the half-life as the decay rate or extracting the reciprocal of the time constant of the fitted function, thus transforming the original drift into a physically meaningful energy dissipation index.

[0078] The parameter space can be a multidimensional mathematical space defined by the leakage defect evolution dynamics model, spanned by variables such as local stress state parameters and energy decay rate. It can be used to provide a geometric framework for feature point localization and boundary comparison. In an exemplary embodiment, the parameter space can include the boundary of the stable leakage development region as a subset. Feature points can be coordinate points in the parameter space uniquely determined by the local stress state parameters and energy decay rate, representing the current state of the flash memory medium, and can be used as a benchmark for assessing leakage development trends. Furthermore, the risk tendency value can be jointly determined by the feature point and its position in the parameter space relative to the boundary of the stable leakage development region. Based on the local stress state parameters and energy decay rate, determining a feature point representing the current state of the flash memory medium in the parameter space defined by the leakage defect evolution dynamics model can be achieved by using the two parameters as coordinate axis values ​​and calibrating the corresponding point in the predefined parameter space. Further, this operation can be achieved by directly concatenating them into a two-dimensional vector or by mapping after dimensionality reduction through principal component analysis, thereby enabling a geometrical representation of the multidimensional state.

[0079] The boundary of the stable leakage development region can be the interface dividing stable leakage behavior and accelerated failure behavior in the parameter space, and can be used as a criterion for judging whether an abnormal region tends to severe failure. In a specific embodiment, the boundary of the stable leakage development region can be obtained based on empirical or theoretical surfaces derived from a large amount of aging experimental data and physical model calibration. For example, the boundary of the stable leakage development region can include, but is not limited to, linear discrimination boundaries, nonlinear manifold boundaries, probability density contour line boundaries, etc. Based on the relative relationship between the position of feature points in the parameter space and the boundary of the stable leakage development region defined by the model, the tendency of an abnormal leakage region to develop into a severe charge leakage defect can be quantified, which can be by calculating the Euclidean distance, Mahalanobis distance, or crossing probability from the feature point to the boundary. Furthermore, this operation can be achieved by setting the tendency value to a high risk level if the feature point is outside the boundary, or by obtaining a continuous tendency value through linear interpolation based on the distance from the boundary, thereby converting the geometric position into a risk value.

[0080] The tendency of severe charge leakage defects can be defined as the probability or trend strength of an anomalous leakage region evolving into functional data loss or uncorrectable errors in the future. This can be used as the output of a mapping relationship for risk quantification. In an exemplary embodiment, the tendency of severe charge leakage defects can be calculated based on the distance, crossing velocity, or projection direction from the feature point to the boundary of the stable leakage development region. Obtaining the mapping relationship can be achieved by encapsulating the above calculation process into a deterministic or probabilistic function from the input to the tendency value. Furthermore, this operation can be implemented by storing discrete mappings in the form of lookup tables, or by implementing continuous mappings using neural networks or analytical formulas, thereby establishing a reusable risk prediction mechanism. The mapping relationship can be a functional correspondence from the logical address and decay rate of the anomalous leakage region to the tendency of severe charge leakage defects, which can be used to achieve a calculable conversion from early defects to future risks.

[0081] Taking industrial control SSD lifespan prediction as an example, the diagnostic method for early potential defects of solid-state drives in this embodiment can be as follows: During a high-temperature and high-humidity cyclic aging test of an industrial SSD, the system identifies an abnormal leak at logical address 0x3A2F; based on the chip's physical location corresponding to this address and the temperature-voltage principal stress trajectory during aging, the potential leak initiation direction is calculated to be dominated by vertical tunneling, with a local stress concentration factor of 1.8; simultaneously, the energy decay rate is obtained by fitting the amplitude of the temporal threshold drift characteristic of this region within the sensitive window to a value of 0.045 per hour; both are input into the parameter space of the leakage defect evolution dynamics model to form a feature point (1.8, 0.045); this point is located outside the boundary of the stable leakage development region and 0.3 units away from the boundary, and the system quantifies its tendency to develop into a serious defect as 0.87; finally, this mapping relationship is used to update the health report, recommending that the data be migrated before the next maintenance cycle.

[0082] In one embodiment, a quantitative assessment of the early charge leakage locations and risk propensity values ​​of the target solid-state drive is generated, including: From the abnormal leakage area, select the area with the highest degree of overlap between its logical address and the potential leakage initiation direction indicated by the local stress state parameter as the early charge leakage location; Obtain the charge leakage risk propensity value quantified by the mapping relationship corresponding to the early charge leakage location; The charge leakage risk propensity value is dynamically corrected based on the energy decay rate to obtain the corrected risk propensity value. The early charge leakage locations are combined with the revised risk propensity values ​​to form a quantitative assessment result.

[0083] The local stress state parameter can be a comprehensive characterization of the multi-dimensional stress intensity (thermal, electrical, mechanical, etc.) in the local physical environment of the flash memory cell, and can be used to indicate the differences in degradation driving forces caused by manufacturing fluctuations or differences in usage conditions in different regions. In this embodiment, the local stress state parameter can be obtained through on-chip sensor arrays, process corner models, or aging history inversion, and is distributed according to spatial location. Further, the local stress state parameter can include, but is not limited to, thermal stress gradient, electric field intensity distribution, lattice strain density, etc. The potential leakage initiation direction can be a physical spatial direction in which charge leakage is most likely to occur, derived from the local stress state parameter, and can be used to provide prior information on leakage evolution driven by physical mechanisms for directional consistency comparison with logical addresses. For example, the potential leakage initiation direction can be a direction vector or region identifier calculated based on a stress concentration region or interface defect distribution model. In an exemplary embodiment, the potential leakage initiation direction can include, but is not limited to, the tunneling dominance direction, the interface state diffusion direction, the bulk defect chain propagation direction, etc.

[0084] Overlap degree can be a quantitative indicator of the spatial alignment between the physical location corresponding to the logical address of an abnormal leakage region and the potential leakage initiation direction. It can be used as a criterion for screening early charge leakage locations with the greatest evolutionary potential. In a specific embodiment, overlap degree can be obtained by coordinate mapping and calculation using the cosine of the vector angle or the region intersection-union ratio. Further, overlap degree can include, but is not limited to, geometric projection overlap degree, probability density matching degree, topological proximity, etc. From the abnormal leakage region, selecting the region with the highest overlap degree between its logical address and the potential leakage initiation direction indicated by the local stress state parameter as the early charge leakage location can be achieved by mapping the logical address of the abnormal leakage region to physical coordinates, calculating its spatial overlap degree with the potential leakage initiation direction, and selecting the region corresponding to the maximum value. Further, this operation can be achieved by calculating the directional alignment degree through vector dot product, generating an initiation hotspot map using thermal-electric coupling simulation and matching it with the logical address, thereby improving the consistency of the physical mechanism for selecting early leakage locations and avoiding mislocation caused by relying solely on statistical anomalies.

[0085] The charge leakage risk tendency value can be an initial risk assessment parameter corresponding to the early charge leakage location, quantified by a mapping relationship. Obtaining the charge leakage risk tendency value quantified by the mapping relationship corresponding to the early charge leakage location can be achieved by querying a previously established mapping relationship and extracting the initial risk tendency value corresponding to that location. Furthermore, this operation can be implemented by directly looking up a pre-stored value in a table or by calling an online leakage defect evolution dynamics model for real-time calculation, thereby obtaining a baseline risk assessment based on a physical degradation model. The energy decay rate can be a dynamic time-series characteristic parameter reflecting the rate of energy dissipation during charge leakage, which can be used to characterize whether the leakage process is in an acceleration phase and to dynamically correct the risk tendency value. In this embodiment, the energy decay rate can be extracted from the time-domain threshold drift characteristics to obtain the energy integral change rate of the nonlinear decay segment. For example, the energy decay rate can include, but is not limited to, barrier penetration energy loss rate, trap release energy flow rate, and interface composite energy dissipation rate. Dynamically correcting the charge leakage risk tendency value based on the energy decay rate to obtain the corrected risk tendency value can be achieved by using the energy decay rate as an adjustment factor to nonlinearly correct the initial risk tendency value. Furthermore, this operation can be achieved by using an exponentially weighted fusion of initial values ​​and decay rates, and constructing a piecewise function to dynamically adjust the weights according to the decay stages. This allows the risk assessment to reflect the time-varying acceleration characteristics of the leakage process and enhances the timeliness of prediction.

[0086] The corrected risk propensity value can be a charge leakage risk propensity value dynamically adjusted by the energy decay rate, which can be used to enable the risk assessment results to respond in real time to accelerated degradation behavior. In one specific embodiment, the corrected risk propensity value can be calculated by combining the initial risk propensity value with a functional relationship (such as weighted or exponential modulation) of the energy decay rate. Furthermore, the corrected risk propensity value relies on the charge leakage risk propensity value as a basic input and is dynamically corrected by the energy decay rate. Combining the early charge leakage location with the corrected risk propensity value to form a quantitative assessment result can be achieved by pairing selected location identifiers with their corresponding corrected risk propensity values ​​to form a structured output. Further, this operation can be achieved by encapsulating and generating timestamped risk event records in key-value pair format, thereby generating assessment data with both high-confidence location and dynamic risk level to support accurate diagnosis.

[0087] Taking industrial control SSD lifespan prediction as an example, the diagnostic method for early potential defects of solid-state drives in this embodiment can be as follows: During high-temperature cycling testing of an industrial SSD, the system identifies multiple channels with abnormal leakage areas; by reversing the local stress state parameters of each block through the process model, it is deduced that the potential leakage initiation direction is concentrated in the high electric field region at the chip edge; the overlap between the physical location corresponding to the logical address of each abnormal area and this direction is calculated, and the block with the highest overlap is selected as the early charge leakage location; the leakage defect evolution dynamics model is called to obtain an initial risk tendency value of medium; further analysis of its time domain threshold drift curve reveals that the energy decay rate increases exponentially, and the risk tendency value is dynamically corrected to high risk accordingly; finally, the quantitative evaluation result marks the logic block as a high-risk early leakage point, triggering the firmware to add it to the migration queue.

[0088] In one embodiment, based on the location of early charge leakage and the risk propensity value in the quantitative assessment results, a health status diagnostic report of the target solid-state drive is generated, including: The urgency level of maintenance actions is determined based on the magnitude of the risk propensity value and the local stress state parameters at the location of early charge leakage. The local stress state parameter can be a set of multi-dimensional stress features describing the physical or operational environment of the early charge leakage location. It can be used to enhance the context of risk propensity values, improving the environmental adaptability of maintenance urgency level determination. In this embodiment, the local stress state parameter can be extracted from firmware logs or sensor data, including local temperature, channel I / O load, erase / write history density, and neighboring block interference intensity. Furthermore, the local stress state parameter can include, but is not limited to, thermal stress parameters, electrical stress parameters, and mechanical stress parameters. The urgency level of maintenance action can be a discretized level characterizing the time window for taking maintenance measures, based on a comprehensive determination of risk propensity values ​​and local stress state parameters. This can be used to guide the operation and maintenance system to implement differentiated response times for different areas. For example, the urgency level of maintenance action can map the risk and stress combination into low, medium, high, or numerical priorities using preset rules or classification models. In an exemplary embodiment, the urgency level of maintenance action can be divided according to response time limits, such as immediate intervention level, planned processing level, and monitoring and observation level. The urgency level of maintenance actions is determined based on the magnitude of the risk propensity value and the local stress state parameters at the location of early charge leakage. This can be achieved by inputting the risk propensity value and local stress state parameters into a preset judgment logic, which outputs a discretized urgency level. Furthermore, this operation can be implemented by combining risk and stress scores using a weighted scoring method and automatically dividing the level range using a decision tree model. This ensures that maintenance priorities not only reflect the current risk level but also incorporate environmental acceleration factors, improving the rationality of decision-making.

[0089] Based on the correlation between energy decay rate and electrical-miscoupling, maintenance strategy recommendations are generated for early charge leakage locations; The energy decay rate can be a parameter representing the rate at which the charge holding level signal decays exponentially or power-law-law-wise within the sensitive data holding window. It can serve as a key input for maintenance strategy recommendations, characterizing the physical acceleration of the leakage process. In one specific embodiment, the energy decay rate is obtained by fitting a physical degradation function to the time-domain threshold drift characteristics, reflecting the energy barrier characteristics of charge loss. Furthermore, the energy decay rate can be classified according to decay models, including but not limited to thermally activated dominant decay rate, tunneling dominant decay rate, and composite mechanism decay rate. The electrical-error coupling correlation can be the statistical dependence or temporal synchronization between charge holding level fluctuations and read error rate gradient changes. It can be used to distinguish between leakage caused by real physical defects and random noise interference, improving the targeting of maintenance strategies. In this embodiment, the electrical-error coupling correlation is quantified by cross-correlation analysis, Granger causality tests, or joint probability modeling. For example, the electrical-error coupling correlation can be classified according to coupling modes, such as synchronous mutation type, hysteresis response type, and weakly correlated background type. Maintenance strategy recommendations can be physically interpretable operational interventions generated for specific early charge leakage locations. These recommendations can provide actionable maintenance guidance, such as prioritizing data migration, adjusting reread verification frequency, or local wear leveling interventions. In one exemplary embodiment, the maintenance strategy recommendations are based on the correlation between energy decay rate and electrical-misalignment coupling, matching the output of a pre-defined strategy library or a generative rule engine. Furthermore, maintenance strategy recommendations can include, but are not limited to, data migration recommendations, read strategy optimization recommendations, and write distribution rescheduling recommendations, categorized by intervention type.

[0090] Based on the correlation between energy decay rate and electrical-miscoupling, maintenance strategy recommendations for early charge leakage locations are generated. This can be achieved by combining the physical meaning of energy decay rate with electrical-miscoupling strength to match corresponding maintenance action templates. Furthermore, this operation can be implemented by looking up a predefined set of strategy rules and generating customized recommendations through a lightweight neural network, thereby generating strategy recommendations with a mechanistic basis and avoiding resource waste or insufficient protection caused by generalized processing.

[0091] The logical address of the early charge leakage location, risk propensity value, urgency level of maintenance action, and maintenance strategy recommendations are structurally integrated to generate a health status diagnostic report for the target solid-state drive.

[0092] The logical address of early charge leakage locations, risk propensity values, urgency levels of maintenance actions, and maintenance strategy recommendations are structurally integrated to generate a health status diagnostic report for the target SSD. This can be achieved by organizing the aforementioned fields in a standard format (such as JSON or TLV) to form diagnostic output that can be parsed by upper-layer systems. Furthermore, this operation can be implemented by generating reports that conform to the extended attributes of the NVMe management interface and outputting time-stamped versioned diagnostic snapshots, thereby achieving a closed-loop output from assessment to decision-making and supporting direct invocation by automated operation and maintenance systems.

[0093] Taking predictive maintenance of cloud storage SSDs as an example, the diagnostic method for early potential defects of solid-state drives in this embodiment can be as follows: A cloud service provider enables this solution in its SSD cluster. The system detects early charge leakage in a certain logical address area with a risk tendency value of 0.72. At the same time, this area is located in a high I / O channel and has recently undergone intensive erase and write operations. The local stress state parameters show that both thermal load and electrical stress are higher than the average value. Based on this, the maintenance urgency level is determined to be high. Further analysis shows that its energy decay rate conforms to the tunneling-dominated model, and the electrical-miscoupling correlation shows strong synchronous mutation characteristics. The system automatically generates maintenance strategy suggestions: prioritize the migration of data in this area, increase the background scan frequency of this LBA range, and avoid adjacent physical pages in subsequent writes. Finally, the above information is structured and integrated into a health status diagnosis report and pushed to the storage resource scheduling platform for automatic execution.

[0094] In addition, refer to Figure 4 To achieve the above objectives, the present invention also provides a diagnostic system for early potential defects in solid-state drives, the system comprising: The multi-source acquisition module 10 is used to acquire the cross-channel multi-source timing monitoring dataset of the target solid-state drive under pressure aging conditions. The multi-source timing monitoring dataset includes the charge holding level signal and the read error rate signal deployed at the preset test unit of each flash channel. The feature construction module 20 is used to construct a time-domain threshold drift feature that reflects the charge decay characteristics of the channel unit based on the charge retention level signal, and to filter sensitive data retention windows associated with physical erasure and programming cycles of flash memory blocks according to pressure aging conditions. The anomaly identification module 30 is used to synchronously analyze the gradient distribution changes of the read bit error rate signal, combine sensitive data to maintain the level fluctuation within the window, and identify the channel abnormal leakage area caused by oxide layer defect trap capture and electron tunneling effect. The quantitative evaluation module 40 is used to establish the mapping relationship between the logical address of the abnormal leakage area, the decay rate of the time domain threshold drift characteristics and the preset leakage defect evolution dynamic model, and generate the quantitative evaluation results of the early charge leakage location and risk tendency value of the target solid-state drive. The report generation module 50 is used to generate a health status diagnostic report for the target solid-state drive based on the location of early charge leakage and risk tendency value in the quantitative assessment results.

[0095] Other embodiments or specific implementations of the diagnostic system for early potential defects of solid-state drives described in this invention can be referred to the above-described method embodiments, and will not be repeated here.

[0096] The above are merely preferred embodiments of the present invention and do not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for diagnosing early potential defects in solid-state drives, characterized in that, The method includes: Acquire a multi-source timing monitoring dataset of the target solid-state drive under stress aging conditions. The multi-source timing monitoring dataset includes charge retention level signals and read error rate signals deployed at preset test units in each flash channel. Based on the charge retention level signal, a time-domain threshold drift characteristic reflecting the charge decay characteristics of the channel unit is constructed, and sensitive data retention windows related to physical erasure and programming cycles of flash memory blocks are screened according to pressure aging conditions. Synchronous analysis of the gradient distribution changes of the bit error rate signal, combined with the level fluctuations within the window of sensitive data, identifies the abnormal leakage area of ​​the channel caused by oxide layer defect trap capture and electron tunneling effect; Establish a mapping relationship between the logical address of the abnormal leakage area, the decay rate of the time domain threshold drift characteristic, and the preset leakage defect evolution dynamic model, and generate a quantitative assessment result of the early charge leakage location and risk tendency value of the target solid-state drive. Based on the location of early charge leakage and risk propensity value in the quantitative assessment results, a health status diagnostic report of the target solid-state drive is generated.

2. The method for diagnosing early potential defects in a solid-state drive as described in claim 1, characterized in that, The settings for the preset test unit of the flash channel include: Based on the physical architecture and stress aging conditions of flash memory chips, key lines and bit line regions that are prone to charge loss are identified, thus obtaining the critical regions. Within the critical area, the test unit allocation positions are selected by cross-selection based on the location of the maximum electric field intensity on the surface of the flash memory block and the location of the programming interference mode node, thus obtaining the initial allocation positions; Based on the conduction attenuation relationship between the charge holding level signal and the read error rate signal in the flash array, the relative spacing and logical mapping relationship of the test cells at the initial allocation position are adjusted to obtain the optimized layout of the preset test cells of the flash channel.

3. The method for diagnosing early potential defects in a solid-state drive as described in claim 1, characterized in that, The construction of a time-domain threshold drift feature reflecting the charge decay characteristics of a channel unit based on the charge retention level signal includes: The time-domain voltage distribution of the charge-holding level signal is obtained by performing a time-frequency transformation on the charge-holding level signal; In the time-domain voltage distribution, based on the continuity of the voltage distribution, the evolution path of the voltage level in the concentrated voltage region over time is traced to obtain the initial drift ridge. By combining the known charge loss response mode of the target solid-state drive under pressure aging conditions, the level value of the initial drift ridge is trend-fitted to obtain the time-domain threshold drift characteristics that reflect the charge decay characteristics of the channel cell.

4. The method for diagnosing early potential defects in a solid-state drive as described in claim 1, characterized in that, The sensitive data retention window, which is selected based on pressure aging conditions and associated with the physical erasure and programming cycles of flash memory blocks, includes: By analyzing the pressure aging conditions, typical stress components corresponding to the erase and programming operations of the flash memory block were separated, and typical stress time series were obtained. By mapping the time phase of a typical stress time series to the level change phase of the time domain threshold drift characteristic, a time synchronization relationship between stress excitation and charge level response is established. Based on the time synchronization relationship, the fluctuation period of each window level and the change period of typical stress components in the time domain threshold drift characteristics are analyzed; windows with stable period correspondence are identified as sensitive data holding windows.

5. The method for diagnosing early potential defects in a solid-state drive as described in claim 1, characterized in that, The synchronous analysis reads the gradient distribution changes of the bit error rate signal, combines the level fluctuations within the sensitive data maintenance window, and identifies abnormal channel leakage regions caused by oxide layer defect trapping and electron tunneling effects, including: Based on the read error rate signals deployed in preset test units of different flash memory channels, an error field reflecting the spatial distribution of flash disk errors is constructed; The error field on a continuous time series is analyzed, and the rate of change of the error difference between adjacent logic cells of the flash disk over time is calculated to obtain the gradient distribution change. From the gradient distribution changes, regional features that exceed the rate of change caused by the background read interference of the flash disk are extracted to obtain candidate abnormal leakage regions. Acquire the time series of level fluctuations within the same time period in the sensitive data retention window; correlate the time phase of the candidate abnormal leakage area with the changing phase of the level fluctuation time series, analyze the time correspondence and intensity ratio between the changes of the candidate abnormal leakage area and the level fluctuation time series, and obtain the electrical-miscoupling correlation relationship; Based on the electrical-miscoupling correlation, sub-regions in the candidate abnormal leakage region whose changes have a stable intensity ratio with the time series of level fluctuations are identified as abnormal leakage regions of the target solid-state drive.

6. The method for diagnosing early potential defects in a solid-state drive as described in claim 1, characterized in that, The dynamic model of the leakage defect evolution is constructed through the following steps: Based on data retention test data of the same flash memory medium as the target solid-state drive under pressure aging conditions, the constitutive relationship between charge leakage rate and stress intensity factor range was determined. In the constitutive relation, a damping attenuation coefficient related to level energy dissipation is introduced as a correction term to establish a two-dimensional parameter space with the stress intensity factor range and the damping attenuation coefficient as coordinates, and the boundary between the stable development region and the unstable development region of leakage is delineated in this space. By integrating the two-dimensional parameter space and boundary, a criterion system for quantitatively assessing the risk of charge leakage is constructed, resulting in a dynamic model of leakage defect evolution.

7. The method for diagnosing early potential defects in a solid-state drive as described in claim 1, characterized in that, The mapping relationship between the logical address of the abnormal leakage region, the decay rate of the time-domain threshold drift characteristic, and the preset leakage defect evolution dynamics model includes: For abnormal leakage areas, based on their logical addresses and the principal stress trajectory of the target solid-state drive under pressure aging conditions, the potential leakage initiation direction and local stress concentration factor are determined, and local stress state parameters are obtained. The amplitude of the time-domain threshold drift feature within the sensitive data holding window is fitted over time, and the percentage decrease in amplitude per unit time is calculated to obtain the energy decay rate. Based on the local stress state parameters and energy decay rate, a feature point characterizing the current flash memory medium state is determined in the parameter space defined by the leakage defect evolution dynamics model. Based on the relative relationship between the position of feature points in the parameter space and the boundary of the stable development region of leakage defined by the model, the tendency of abnormal leakage regions to develop into severe charge leakage defects is quantified, and the mapping relationship is obtained.

8. The method for diagnosing early potential defects in a solid-state drive as described in claim 1, characterized in that, The quantitative assessment results of the early charge leakage locations and risk propensity values ​​of the generated target solid-state drive include: From the abnormal leakage area, select the area with the highest degree of overlap between its logical address and the potential leakage initiation direction indicated by the local stress state parameter as the early charge leakage location; Obtain the charge leakage risk propensity value quantified by the mapping relationship corresponding to the early charge leakage location; The charge leakage risk propensity value is dynamically corrected based on the energy decay rate to obtain the corrected risk propensity value. The early charge leakage locations are combined with the revised risk propensity values ​​to form a quantitative assessment result.

9. The method for diagnosing early potential defects in a solid-state drive as described in claim 1, characterized in that, Based on the early charge leakage locations and risk propensity values ​​from the quantitative assessment results, a health status diagnostic report for the target solid-state drive is generated, including: The urgency level of maintenance actions is determined based on the magnitude of the risk propensity value and the local stress state parameters at the location of early charge leakage. Based on the correlation between energy decay rate and electrical-miscoupling, maintenance strategy recommendations are generated for early charge leakage locations; The logical address of the early charge leakage location, risk propensity value, urgency level of maintenance action, and maintenance strategy recommendations are structurally integrated to generate a health status diagnostic report for the target solid-state drive.

10. A diagnostic system for early potential defects in solid-state drives, characterized in that, The system includes: The multi-source acquisition module is used to acquire cross-channel multi-source timing monitoring datasets of the target solid-state drive under stress aging conditions. The multi-source timing monitoring datasets include charge retention level signals and read error rate signals deployed at preset test units in each flash channel. The feature construction module is used to construct a time-domain threshold drift feature that reflects the charge decay characteristics of the channel unit based on the charge retention level signal, and to filter sensitive data retention windows associated with physical erasure and programming cycles of flash memory blocks according to pressure aging conditions. The anomaly identification module is used to synchronously analyze the gradient distribution changes of the read bit error rate signal, combine sensitive data to maintain the level fluctuation within the window, and identify the channel abnormal leakage area caused by oxide layer defect trap capture and electron tunneling effect. The quantitative evaluation module is used to establish the mapping relationship between the logical address of the abnormal leakage area, the decay rate of the time domain threshold drift characteristics and the preset leakage defect evolution dynamic model, and generate quantitative evaluation results of the early charge leakage location and risk tendency value of the target solid-state drive. The report generation module is used to generate a health status diagnostic report for the target solid-state drive based on the location of early charge leakage and risk tendency value in the quantitative assessment results.

Images