SSD full-automatic aging test system and method

Abstract

The invention discloses an SSD full-automatic aging test system and method, and the system comprises a physical test cluster layer which comprises one or more test bins, a central control and data processing unit, a task scheduling and resource management engine, a self-adaptive stress field generator, a data lake, and a digital twinborn model. The system comprises a data processing module for establishing a corresponding virtual instance and a model calibration and synchronization interface for each SSD body, a predictive analysis module for predicting the residual life and the potential fault mode of the SSD body based on accelerated operation of a digital twin model, and a reinforcement learning module for establishing a model calibration and synchronization interface for each SSD body, and keeping the state of the virtual instance consistent with the state of the physical SSD body by using information of a data lake. The self-adaptive stress field generator works cooperatively with the self-adaptive stress field generator; and the feedback and evolution layer is used for calculating a test efficiency index TEI and periodically optimizing a self-adaptive stress field generation algorithm and a strategy network of a reinforcement learning agent by utilizing the accumulated TEI and data.

Description

Technical Field

[0001] This invention belongs to the field of electronic device technology, specifically, it relates to a fully automated SSD aging test system and method. Background Technology

[0002] Existing SSD aging test technologies are essentially a crude, fixed-condition fatigue test, and mainly suffer from the following problems: 1. Traditional testing uses fixed, high-intensity stress (such as high temperature and continuous full-load writing) to attempt to simulate years of aging in a short period of time. This method is blind and inefficient, with most of the time and energy wasted on ineffective aging, and it cannot accurately trigger deep defects.

[0003] 2. Fixed-pattern testing can only stimulate known and common failure modes. Traditional testing is ineffective for "corner case" defects caused by complex factors (such as slight voltage fluctuations, specific GC stress, temperature cycling, etc.).

[0004] 3. Traditional testing is a "black box" process. We only know that the SSD "failed," but it's difficult to answer "why it failed?" or "how it gradually failed?" Test data is mainly used for pass / fail determination, lacking in-depth analysis.

[0005] 4. Traditional tests can only give the conclusion that "it survived for X hours under specific conditions", and cannot make accurate and personalized predictions about the remaining lifespan of a single SSD.

[0006] 5. The testing process and stress parameters are pre-set by engineers, and one solution is used for all SSDs. The testing system itself does not learn from historical tests and cannot become "smarter" with use. Summary of the Invention

[0007] To address the shortcomings of existing technologies, the present invention aims to provide a fully automated SSD aging test system and method.

[0008] To achieve the aforementioned objectives, the technical solution adopted by this invention includes: a fully automated SSD aging test system, comprising an SSD body, and: The physical test cluster layer includes one or more test chambers, each containing: A multi-channel programmable DC power supply is used to provide precise voltage Vcc to the SSD body. Semiconductor coolers and temperature sensors are used to precisely control the ambient temperature T of the SSD. Vibration suppression platforms are used to isolate external vibration interference. The FPGA board communicates with the SSD via an interface such as PCIe or SATA. The FPGA board performs microsecond-level precise control and monitoring of the SSD's physical pin levels and timings, and performs high-precision data load application and directional perturbation testing. The physical test cluster layer communicates with the host computer through a high-speed internal network (such as 10 Gigabit Ethernet) to achieve ultra-low latency transmission of test commands and test data streams; The central control and data processing unit includes a task scheduling and resource management engine, an adaptive stress field generator, a data lake, and a digital twin model. The task scheduling and resource management engine is used to receive test tasks and dynamically allocate them to idle test warehouses. The adaptive stress field generator receives the analysis results from its internal digital twin model and uses the adaptive stress field generation algorithm to generate real-time test commands, which are then sent to the physical test cluster layer. The data lake storage SSD ontology and digital twin model generate all data throughout their entire lifecycle; The digital twin model includes: A high-fidelity model library creates a corresponding virtual instance for each SSD entity. The model calibration and synchronization interface utilizes information from the data lake to maintain consistency between the virtual instance and the physical SSD. Predictive analytics module: Based on accelerated operation using a digital twin model, it predicts the remaining lifespan and potential failure modes of the SSD. The reinforcement learning module works in conjunction with the adaptive stress field generator. The feedback and evolution layer calculates the Test Performance Index (TEI) and uses the accumulated TEI and data to periodically optimize the adaptive stress field generation algorithm and the policy network of the reinforcement learning agent.

[0009] This invention deeply integrates digital twins, reinforcement learning, physical failure models, and multi-objective optimization into a fully automated SSD aging test system. It is an intelligent entity with sensing, decision-making, prediction, and evolutionary capabilities. By constructing a high-fidelity digital twin, applying an adaptive stress field, employing an autonomously explored test path, executing FTL-oriented perturbations, and achieving closed-loop optimization of system performance, this solution can discover deeper and more unpredictable SSD defects and failure modes in a shorter time and at a lower cost.

[0010] Furthermore, the central control and data processing unit receives the analysis results from the digital twin model and uses an adaptive stress field generation algorithm to generate real-time test commands. The adaptive stress field generation algorithm includes: The objective function F(S) is defined as: F(S) = α · (dRUL / dt) + β · (Fault_Entropy) - γ· (Power_Consumption), Where S is the stress vector, dRUL / dt is the remaining lifetime decay rate predicted based on digital twin, (Fault_Entropy is the fault entropy, Power_Consumption is the system power consumption, the metaheuristic algorithm is used to solve the optimal stress vector S_optimal that maximizes F(S) in real time, and apply it to the SSD body, the objective function is used to determine how to test, and generate the test strategy problem.

[0011] α · (dRUL / dt) addresses the issue of "stress blindness" and achieves "precise and efficient aging" by introducing the remaining service life degradation rate predicted based on digital twins as the core feedback. This is equivalent to installing an "aging rate meter" in the testing system. β · (Fault_Entropy) addresses the issue of "single fault" and enables "broad spectrum defect discovery." It innovatively proposes the indicator of fault entropy to quantify the potential or possibility of exposing diverse and unpredictable fault modes under current stress. γ · (Power_Consumption) solves the problem of "uncontrollable cost" and realizes "green and intelligent testing". It introduces system power consumption as a penalty term into the objective function, so that the system must consider energy cost while pursuing high efficiency and in-depth testing. The remaining useful life degradation rate, predicted based on digital twins, is introduced as the core feedback. This is equivalent to installing an "aging rate meter" for the testing system.

[0012] Furthermore, S is a stress vector that includes voltage Vcc, temperature T, data load mode L, and read / write ratio R / W.

[0013] Furthermore, the twin model of the digital twin model includes a time-varying degradation equation describing the threshold voltage drift of the NAND cell: ΔV_th(t) = A · [ln(1 + B · t)]^C · exp(-E_a / (k · T(t))) · (1+ D · P / E_Cycles) Wherein, ΔV_th(t): threshold voltage drift after time t, A, B, C, D: fitting parameters related to process technology and material properties, obtained through prior characterization tests, E_a: activation energy of the degradation process, k: Boltzmann constant, T(t): real-time temperature, which is time-varying, P / E_Cycles: number of erase / write cycles, realistically simulating data retention errors and read interference errors, which determines why the model can make accurate predictions and solves the problems of model accuracy and physical reliability.

[0014] The simulation [ln(1 + B · t)]^C describes the decay of data retention capability, depicting the rate at which charge leaks through the tunnel oxide layer in a floating gate at high temperatures—a process that is sublinear with time. This allows digital twins to dynamically predict how data retention capability changes with time and temperature. The twin model can calculate how much the threshold voltage of a memory cell has drifted after a specific time t and temperature T(t), thus quantitatively predicting which data might be read incorrectly due to charge leakage, rather than simply giving a general conclusion that "it might be wrong."

[0015] The expression exp(-E_a / (k · T(t)) precisely quantifies the accelerating effect of temperature, accurately describing how chemical reaction rates (here, charge leakage and oxide layer damage) accelerate exponentially with increasing temperature. It gives digital twins the ability to precisely quantify the impact of temperature stress. The model knows exactly how much the aging acceleration factor is at 125°C relative to 85°C, making lifetime prediction (RUL) more accurate.

[0016] The equation (1 + D · P / E_Cycles) simulates the coupling effect of "durability" wear, introducing the wear amplification effect of write / erase cycles on data retention. A block that has undergone tens of thousands of P / E cycles has a damaged tunnel oxide layer, and charge leakage will be faster. This equation couples the two key failure modes of durability wear and data retention. This allows digital twins to predict that a nearly fully written, highly worn SSD will have a much higher risk of data loss when stored at high temperatures than a brand new SSD.

[0017] Furthermore, the central control and data processing unit models the testing process as a Markov decision process, where the state is a comprehensive health index of the SSD ontology and its twin model, the action is the adjustment of the stress field, the reward is the value of the F(S) function, and the deep reinforcement learning agent is trained to learn the optimal testing strategy.

[0018] Furthermore, it also includes: A test performance index is introduced to measure the return on investment of a single test: Test Performance Index (TEI) = (Σ W_i· F_i) / (T · E), Where F_i: the severity weight of the i-th type of fault discovered, T: the total test time, and E: the total test energy consumption. W_i: The weight of the fault type. It continuously records the TEI of all test tasks and uses this data to back-optimize the adaptive stress field generation algorithm and the policy network of the reinforcement learning agent, forming a self-evolving test system that gets smarter the more it is used.

[0019] A fully automated SSD aging test method, using an SSD fully automated aging test system, is described below: Step 1: Initialization and Digital Twin Construction Step 1.1: Enter the basic information of the SSD under test (model, capacity, controller, NAND type, etc.) into the central control and data processing unit, set the test target (such as: target aging degree, specific failure mode screening, extreme life assessment, etc.), and automatically initialize the weight coefficients α, β, γ of the objective function F(S) in the adaptive stress field generation algorithm according to the test target. Once the task is assigned, the test chamber performs a self-test, the SSD is installed and powered on, and the central control unit drives the FPGA to perform characteristic testing on the SSD, collect data, and build and calibrate its corresponding digital twin model. Step 1.2: The digital twin model creates a unique virtual instance of the current SSD and loads a high-fidelity digital twin model that matches the configuration of the SSD. The initial parameters of the model (such as A, B, C, D, E_a) are preset according to the NAND chip characteristics of the SSD. Before the test begins, the system performs a brief characterization process on the physical SSD, and the data is used to perform the first calibration of the digital twin model to ensure that the logical state (such as initial bad block information) and performance baseline of the two are consistent.

[0020] Step 2: Closed-loop test of the core loop Step 2.1 The test chamber receives the current optimal stress vector S_current = [Vcc,T, L, R / W, ...] from the adaptive stress field. Multi-channel programmable DC power supply with adjustable output voltage Vcc. The semiconductor cooler precisely adjusts the temperature of the test chamber to the target temperature T. The FPGA board receives instructions from the adaptive stress field generator of the central control and data processing unit via a high-speed network (such as PCIe or 10 Gigabit Ethernet). The instruction packet contains a specific stress vector S (including the data load mode L (such as sequential, random, or specific data mode) and read / write ratio R / W, applying a data flow to the SSD body), and applies a data flow to the SSD body. According to the testing strategy, the FPGA substrate can simultaneously perform targeted perturbation tests on the physical SSD, injecting abnormal command sequences such as "atomic write paradox" or "garbage collection stress wave" into the SSD. These perturbation modes can first be verified for security and effectiveness in a digital twin model before being sent to the FPGA for execution; Step 2.2: The system collects real-time data of the SSD at a high frequency, including: Performance data: read / write bandwidth, IOPS, latency. Health data: Raw SMART values ​​(number of bad blocks, ECC error rate, wear level, etc.). Physical data: real-time chip temperature and operating current. Logical data: Obtain the status information of the internal FTL (Flash TL converter) of the SSD (such as garbage collection activity, number of free blocks, mapping table cache hit rate, etc.) through the SSD's debugging interface, manufacturer-specific commands, or internal log pages for calibration of the digital twin model; All collected data is timestamped, stored in a massive data lake, and synchronized to the digital twin model in real time; Step 2.3: Using the new data (especially logical and health data) collected in Step 2.2, the digital twin model performs state calibration and parameter updates to ensure consistency with the SSD in key states. After calibration, the digital twin high-fidelity model runs forward in accelerated mode (e.g., 100 times faster than physical time). During the simulation, the physical degradation equation calculates the expected degradation of the NAND chip based on the historical stress S_history, and the predictive analytics module generates a preview of the future state accordingly. Key predictive metrics output during the simulation include: Remaining service life decay rate, potential failure modes and their probability of occurrence, and failure entropy under current stress; Step 2.4: The adaptive stress field receives predicted metrics (dRUL / dt, Fault_Entropy) and real-time power consumption data from the digital twin model. Generate a maximizing objective function F(S) = α · (dRUL / dt) + β · (Fault_Entropy) - γ· (Power_Consumption), and run a metaheuristic algorithm to solve for the optimal stress vector S_next for the next test cycle. Step 3: Termination, Analysis, and System Evolution Step 3.1: Fault Diagnosis and Test Termination The system monitors the following termination conditions in real time, and will automatically terminate the test if any condition is met: The SSD experiences functional failures (such as being unrecognizable, data read / write errors exceeding thresholds, or complete damage); the digital twin high-fidelity model predicts the SSD's lifespan has expired (RUL ≤ 0); the preset test duration or number of cycles is reached; or the test is manually interrupted by the operator. Once a fault is detected, the system records detailed fault context (pre-fault stress state, SMART log, twin model snapshot) and automatically saves all relevant data for subsequent root cause analysis. Step 3.2: Test Performance Evaluation and Feedback Optimization After the test, the system calculates the test performance index: TEI = (Σ W_i · F_i) / (T· E), The TEI value, along with detailed test logs, is stored in the knowledge base.

[0021] Furthermore, in step 1.2, the digital twin high-fidelity model is continuously calibrated through step 2.3. The digital twin high-fidelity model and the SSD body are synchronized for the first time to ensure that the logical state (such as LBA mapping table, bad block information) is consistent.

[0022] Furthermore, the system also regularly uses accumulated Test Performance Index (TEI) data and test records to conduct offline training and optimization of the parameters of the adaptive stress field generation algorithm and the policy network of the reinforcement learning agent, making the entire system more and more efficient and intelligent as test experience accumulates.

[0023] A computer-readable medium having a computer program stored thereon, which, when executed by a processor, implements the steps of a fully automated SSD aging test method.

[0024] Compared with the prior art, the advantages of the present invention include: (1) The present invention provides an SSD fully automatic aging test system and method, which uses a digital twin model for real-time prediction and adaptive stress field for dynamic adjustment. The system can always apply the "most effective" stress, avoiding wasting resources on already saturated or ineffective stress, thereby greatly shortening the time and energy required to expose defects.

[0025] (2) The present invention provides an SSD fully automated aging test system and method, which introduces the concept of fault entropy and a multi-objective optimization algorithm. The system not only pursues rapid aging (dRUL / dt), but also actively explores stress combinations that can cause diverse and unpredictable failures, like a "security vulnerability detection expert", actively looking for design weaknesses and significantly improving the defect detection rate.

[0026] (3) The present invention provides a fully automated SSD aging test system and method, which uses a digital twin to record the full data of the physical SSD throughout its entire life cycle and its corresponding state in virtual space. Any failure can be traced back to its complete degradation trajectory and stress history, providing unprecedented data insights for root cause analysis and design improvement.

[0027] (4) The present invention provides an SSD fully automatic aging test system and method. By integrating a high-fidelity digital twin with physical degradation equations, the system can predict the future state of SSD in real time and provide a personalized remaining life prediction based on a physical model, thus realizing the leap from "post-judgment" to "pre-prediction".

[0028] (5) The SSD fully automated aging test system and method provided by this invention, through reinforcement learning and closed-loop feedback of test performance index, enables the system to autonomously explore better test strategies and continuously optimize its own algorithm using historical data. The test system has learning and evolution capabilities and can continuously adapt to new SSD models and new failure modes. Attached Figure Description

[0029] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0030] Figure 1 This is the flowchart of a fully automated SSD aging test system and method according to the present invention. Detailed Implementation

[0031] In view of the shortcomings of the prior art, the inventors of this invention, through long-term research and extensive practice, have proposed the technical solution of this invention. The technical solution, its implementation process, and principles will be further explained below with reference to the accompanying drawings and specific implementation examples in the embodiments of this application.

[0032] It should be noted that the embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention. The described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, the present invention covers any substitutions, modifications, equivalent methods and solutions made on the spirit, principles and scope of the present invention as defined by the claims. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0033] In the description of this application, the terms "first," "second," "third," and similar words do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Similarly, the terms "a" or "one," and similar words, do not indicate a quantity limitation, but rather indicate the presence of at least one. The terms "comprising" or "including," and similar words, mean that the elements or objects preceding "comprising" or "including" encompass the elements or objects listed following "comprising" or "including," and their equivalents, but do not exclude other elements or objects. The terms "connected" or "linked," and similar words, are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect.

[0034] In the description of this application, the terms "center," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used solely for the convenience of describing this application and for simplification, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this application. Furthermore, when using positional terms such as "both sides," "outer side," and "upper and lower," it should be understood that they are used only for ease of understanding and description, taking into account that the structure may be oriented to other positions.

[0035] In the description of this application, unless otherwise expressly specified and limited, the technical or scientific terms used shall have the ordinary meaning understood by a person with ordinary skills in the art to which this application pertains. Terms such as “installation,” “connection,” and “joining” shall be interpreted broadly, for example, as fixed connection, detachable connection, mating connection, or integral connection. For a person skilled in the art, the specific meaning of the above terms in this application can be understood according to the specific circumstances.

[0036] The present invention aims to introduce and explain the structural composition of a fully automated SSD aging test system and method, as well as the cooperation relationship between the various components. Unless otherwise specified, the dimensions, materials, and manufacturing processes of the various components in the fully automated SSD aging test system and method in the present invention can be selected according to specific circumstances, and no special limitations or explanations are made here.

[0037] Furthermore, to provide the public with a better understanding of the present invention, certain specific details are described in detail in the following description of the invention. However, those skilled in the art will fully understand the invention even without these detailed descriptions.

[0038] Example 1 Please see Figure 1A fully automated SSD aging test system, comprising an SSD body, and: The physical test cluster layer includes one or more test chambers, each containing: A multi-channel programmable DC power supply is used to provide precise voltage Vcc to the SSD body. Semiconductor coolers and temperature sensors are used to precisely control the ambient temperature T of the SSD. Vibration suppression platforms are used to isolate external vibration interference. The FPGA board communicates with the SSD via an interface such as PCIe or SATA. The FPGA board performs microsecond-level precise control and monitoring of the SSD's physical pin levels and timings, and performs high-precision data load application and directional perturbation testing. The physical test cluster layer communicates with the host computer through a high-speed internal network (such as 10 Gigabit Ethernet) to achieve ultra-low latency transmission of test commands and test data streams; The central control and data processing unit includes a task scheduling and resource management engine, an adaptive stress field generator, a data lake, and a digital twin model. The task scheduling and resource management engine is used to receive test tasks and dynamically allocate them to idle test warehouses. The adaptive stress field generator receives the analysis results from its internal digital twin model and uses the adaptive stress field generation algorithm to generate real-time test commands, which are then sent to the physical test cluster layer. The data lake storage SSD ontology and digital twin model generate all data throughout their entire lifecycle; The digital twin model includes: A high-fidelity model library creates a corresponding virtual instance for each SSD entity. The model calibration and synchronization interface utilizes information from the data lake to maintain consistency between the virtual instance and the physical SSD. Predictive analytics module: Based on accelerated operation using a digital twin model, it predicts the remaining lifespan and potential failure modes of the SSD. The reinforcement learning module works in conjunction with the adaptive stress field generator. The feedback and evolution layer calculates the Test Performance Index (TEI) and uses the accumulated TEI and data to periodically optimize the adaptive stress field generation algorithm and the policy network of the reinforcement learning agent.

[0039] In this invention, the central control and data processing unit receives analysis results from the digital twin model and uses an adaptive stress field generation algorithm to generate real-time test commands. The adaptive stress field generation algorithm includes: The objective function F(S) is defined as: F(S) = α · (dRUL / dt) + β · (Fault_Entropy) -γ · (Power_Consumption), Where S is the stress vector, dRUL / dt is the remaining lifetime degradation rate predicted based on digital twin, Fault_Entropy is the fault entropy, and Power_Consumption is the system power consumption. A metaheuristic algorithm is used to solve for the optimal stress vector S_optimal that maximizes F(S) in real time, and this is applied to the SSD itself. The objective function determines how to test, generating a testing strategy problem. The F(S) objective function is the system's "intelligent brain," determining how to test, solving the testing strategy problem, and achieving efficient, comprehensive, and economical testing results.

[0040] Furthermore, S is a stress vector that includes voltage Vcc, temperature T, data load mode L, and read / write ratio R / W.

[0041] Furthermore, the twin model of the digital twin model includes a time-varying degradation equation describing the threshold voltage drift of the NAND cell: ΔV_th(t) = A · [ln(1 + B · t)]^C · exp(-E_a / (k · T(t))) · (1+ D · P / E_Cycles) Where, ΔV_th(t): threshold voltage drift after time t, A, B, C, D: fitting parameters related to process technology and material properties, obtained through prior characterization tests, E_a: activation energy of the degradation process, k: Boltzmann constant, T(t): real-time temperature, which is time-varying, P / E_Cycles: number of erase / write cycles, By realistically simulating data retention errors and read interference errors, the NAND degradation equation is the "physical heart" of the digital twin. It determines why the model can make accurate predictions and solves the problems of model accuracy and physical reliability.

[0042] Traditional testing systems simply apply high temperatures, high pressures, and high loads relentlessly, without understanding whether the applied stress effectively accelerates the aging process. The SSD may already be immune to the current stress, and continuing to apply it is merely a waste of energy and time. Fixed-mode stress can only trigger a limited number of common failure modes. Traditional testing is almost entirely ineffective in covering hidden "corner case" defects that can only be triggered by abnormal stress combinations (such as instantaneous high-voltage surges accompanied by specific GC pressures), leading to missed defects that end up on the market. Traditional brute-force testing is extremely energy-intensive and, due to its inefficiency, requires long testing cycles, resulting in high testing costs.

[0043] The remaining service life degradation rate predicted based on digital twins is introduced as the core feedback. This is equivalent to installing an "aging rate meter" for the testing system; Specifically, α · (dRUL / dt) addresses the issue of "stress blindness" and achieves "precise and efficient aging" by introducing the remaining lifespan degradation rate predicted based on digital twins as the core feedback. This is equivalent to installing an "aging rate meter" on the testing system; instead of blindly applying pressure, it aims to maximize the instantaneous aging rate. It can intelligently identify the "sensitive points" of the SSD under test—for example, the data retention capability may degrade the fastest under a specific combination of voltage and temperature. By optimizing the stress S in real time to chase this highest degradation rate, it ensures that every minute of the test is used effectively, greatly improving testing efficiency.

[0044] β· (Fault_Entropy) addresses the issue of "single-fault nature" and enables "broad-spectrum defect discovery." It innovatively proposes the metric of fault entropy to quantify the potential or likelihood of revealing diverse and unpredictable fault modes under current stress. The system no longer merely accelerates "routine aging" but actively seeks stresses that can trigger system malfunctions and abnormal states. For example, a stress that simultaneously causes a surge in read / write latency, abnormal ECC error rates, and rare changes in SMART parameters has a high fault entropy. By maximizing fault entropy, the system acts like a "hacker," constantly trying various bizarre "attack vectors," thereby discovering deep-seated and complex coupled faults that would never be triggered in traditional testing, significantly improving the coverage and depth of testing.

[0045] γ · (Power_Consumption) solves the problem of "uncontrollable cost" and realizes "green and intelligent testing". It introduces system power consumption as a penalty term into the objective function, so that the system must consider energy cost while pursuing high efficiency and in-depth testing. The remaining lifespan degradation rate, predicted based on digital twins, is introduced as the core feedback. This is equivalent to installing an "aging rate meter" in the testing system. This avoids the system adopting an extremely high-energy-consumption strategy—a "kill a thousand enemies, lose eight hundred of your own" approach—in pursuit of ultimate performance. It automatically seeks the test path with the highest energy efficiency. For example, it might find that lowering the temperature from 125°C to 110°C, although slightly reducing the aging rate, significantly reduces power consumption, resulting in a higher overall objective function F(S). This achieves energy consumption optimization while ensuring test effectiveness, reducing testing costs.

[0046] [ln(1 + B · t)]^C simulates the decay of "data retention" capability, describing the rate at which charge leaks through the tunnel oxide layer in a floating gate at high temperatures—a process that is sublinear with time. This allows digital twins to dynamically predict how data retention capability changes with time and temperature. The twin model can calculate how much the threshold voltage of a memory cell has drifted after a specific time t and temperature T(t), thus quantitatively predicting which data might be read incorrectly due to charge leakage, rather than simply giving a general conclusion that "it might be wrong."

[0047] The expression exp(-E_a / (k · T(t)) precisely quantifies the accelerating effect of temperature, accurately describing how chemical reaction rates (here, charge leakage and oxide layer damage) accelerate exponentially with increasing temperature. It gives digital twins the ability to precisely quantify the impact of temperature stress. The model knows exactly how much the aging acceleration factor is at 125°C relative to 85°C, making lifetime prediction RUL more accurate.

[0048] The equation (1 + D · P / E_Cycles) simulates the coupling effect of "durability" wear, introducing the wear amplification effect of write / erase cycles on data retention. A block that has undergone tens of thousands of P / E cycles has a damaged tunnel oxide layer, and charge leakage will be faster. This equation couples the two key failure modes of durability wear and data retention. This allows digital twins to predict that a nearly fully written, highly worn SSD will have a much higher risk of data loss when stored at high temperatures than a brand new SSD.

[0049] The present invention also includes: A test performance index is introduced to measure the return on investment of a single test: Test Performance Index (TEI) = (Σ W_i· F_i) / (T · E), Where F_i: the severity weight of the i-th type of fault discovered, T: the total test time, and E: the total test energy consumption. W_i: The weight of the fault type. It continuously records the TEI of all test tasks and uses this data to back-optimize the adaptive stress field generation algorithm and the policy network of the reinforcement learning agent, forming a self-evolving test system that gets smarter the more it is used.

[0050] A fully automated SSD aging test method, using an SSD fully automated aging test system, is described below: Step 1: Initialization and Digital Twin Construction Step 1.1: Enter the basic information of the SSD under test (model, capacity, controller, NAND type, etc.) into the central control and data processing unit, set the test target (such as: target aging degree, specific failure mode screening, extreme life assessment, etc.), and automatically initialize the weight coefficients α, β, γ of the objective function F(S) in the adaptive stress field generation algorithm according to the test target. Once the task is assigned, the test chamber performs a self-test, the SSD is installed and powered on, and the central control unit drives the FPGA to perform characteristic testing on the SSD, collect data, and build and calibrate its corresponding digital twin model. Step 1.2: The digital twin model creates a unique virtual instance of the current SSD and loads a high-fidelity digital twin model that matches the configuration of the SSD. The initial parameters of the model (such as A, B, C, D, E_a) are preset according to the NAND chip characteristics of the SSD. Before the test begins, the system performs a brief characterization process on the physical SSD, the data of which is used to perform an initial calibration of the digital twin model to ensure that the logical state (such as initial bad block information) and performance baseline of the two are consistent.

[0051] Step 2: Closed-loop test of the core loop Step 2.1 The test chamber receives the current optimal stress vector S_current = [Vcc,T, L, R / W, ...] from the adaptive stress field. Multi-channel programmable DC power supply with adjustable output voltage Vcc. The semiconductor cooler precisely adjusts the temperature of the test chamber to the target temperature T. The FPGA board receives instructions from the adaptive stress field generator of the central control and data processing unit via a high-speed network (such as PCIe or 10 Gigabit Ethernet). The instruction packet contains a specific stress vector S (including the data load mode L (such as sequential, random, or specific data mode) and read / write ratio R / W, applying a data flow to the SSD body), and applies a data flow to the SSD body. According to the testing strategy, the FPGA substrate can simultaneously perform targeted perturbation tests on the physical SSD, injecting abnormal command sequences such as "atomic write paradox" or "garbage collection stress wave" into the SSD. These perturbation modes can first be verified for security and effectiveness in a digital twin model before being sent to the FPGA for execution; Step 2.2: The system collects real-time data of the SSD at a high frequency, including: Performance data: Read / write bandwidth, IOPS, latency. Health data: Raw SMART values ​​(number of bad blocks, ECC error rate, wear level, etc.). Physical data: Real-time chip temperature and operating current. Logical data: Obtain the status information of the internal FTL (such as garbage collection activity, number of free blocks, mapping table cache hit rate, etc.) through the SSD's debugging interface, vendor-specific commands, or internal log pages for calibration of the digital twin model; All collected data is timestamped, stored in a massive data lake, and synchronized to the digital twin model in real time; Step 2.3: Using the new data (especially logical and health data) collected in Step 2.2, the digital twin model performs state calibration and parameter updates to ensure consistency with the SSD in key states. After calibration, the digital twin high-fidelity model runs forward in accelerated mode (e.g., 100 times faster than physical time). During the simulation, the physical degradation equation calculates the expected degradation of the NAND chip based on the historical stress S_history, and the predictive analytics module generates a preview of the future state accordingly. Key predictive metrics output during the simulation include: Remaining service life decay rate, potential failure modes and their probability of occurrence, and failure entropy under current stress; Step 2.4: The adaptive stress field receives predicted metrics (dRUL / dt, Fault_Entropy) and real-time power consumption data from the digital twin model. Generate a maximizing objective function F(S) = α · (dRUL / dt) + β · (Fault_Entropy) - γ· (Power_Consumption) as the objective, and run a metaheuristic algorithm to solve for the optimal stress vector S_next for the next test cycle. Reinforcement learning agents observe the entire state-action-reward sequence and update their policy network, thereby learning what stress adjustments to take in what states to obtain the maximum long-term reward. Step 3: Termination, Analysis, and System Evolution Step 3.1: Fault Diagnosis and Test Termination The system monitors the following termination conditions in real time, and will automatically terminate the test if any condition is met: The SSD experiences functional failures (such as being unrecognizable, data read / write errors exceeding thresholds, or complete damage); the digital twin high-fidelity model predicts the SSD's lifespan has expired (RUL ≤ 0); the preset test duration or number of cycles is reached; or the test is manually interrupted by the operator. Once a fault is detected, the system records detailed fault context (pre-fault stress state, SMART log, twin model snapshot) and automatically saves all relevant data for subsequent root cause analysis. Step 3.2: Test Performance Evaluation and Feedback Optimization After the test, the system calculates the test performance index: TEI = (Σ W_i · F_i) / (T· E), The TEI value, along with detailed test logs, is stored in the knowledge base.

[0052] In the method of the present invention, the digital twin high-fidelity model in step 1.2 is continuously calibrated in step 2.3. The digital twin high-fidelity model and the SSD body are synchronized for the first time to ensure that the logical state (such as LBA mapping table, bad block information) is consistent.

[0053] The method of this invention also includes the system periodically using the accumulated Test Performance Index (TEI) data and test records to perform offline training and optimization on the parameters of the adaptive stress field generation algorithm and the policy network of the reinforcement learning agent, so that the entire system becomes more and more efficient and intelligent as test experience accumulates.

[0054] A computer-readable medium having a computer program stored thereon, which, when executed by a processor, implements the steps of a fully automated SSD aging test method.

[0055] It should be understood that the above embodiments are only for illustrating the technical concept and features of the present invention, and are intended to enable those skilled in the art to understand the content of the present invention and implement it accordingly. It should not be considered that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, several simple deductions or substitutions can be made without departing from the concept of the present invention. All equivalent changes or modifications made in accordance with the spirit and essence of the present invention should be covered within the protection scope of the present invention.

Claims

1. A fully automated SSD aging test system, comprising an SSD body, characterized in that: Also includes: The physical test cluster layer includes one or more test chambers, each containing: A multi-channel programmable DC power supply is used to provide precise voltage Vcc to the SSD body. Semiconductor coolers and temperature sensors are used to precisely control the ambient temperature T of the SSD. Vibration suppression platforms are used to isolate external vibration interference. An FPGA substrate is communicatively connected to the SSD body. The FPGA substrate performs microsecond-level precise control and monitoring of the physical pin levels and timing of the SSD, and performs data load application and directional disturbance testing. The central control and data processing unit includes a task scheduling and resource management engine, an adaptive stress field generator, a data lake, and a digital twin model. The task scheduling and resource management engine is used to receive test tasks and dynamically allocate them to idle test warehouses. The adaptive stress field generator receives the analysis results from its internal digital twin model and uses the adaptive stress field generation algorithm to generate real-time test commands, which are then sent to the physical test cluster layer. The data lake storage SSD ontology and digital twin model generate all data throughout their entire lifecycle; The digital twin model includes: A high-fidelity model library creates a corresponding virtual instance for each SSD entity. The model calibration and synchronization interface utilizes information from the data lake to maintain consistency between the virtual instance and the physical SSD. Predictive analytics module: Based on accelerated operation using a digital twin model, it predicts the remaining lifespan and potential failure modes of the SSD. The reinforcement learning module works in conjunction with the adaptive stress field generator. The feedback and evolution layer calculates the Test Performance Index (TEI) and uses the accumulated TEI and data to periodically optimize the adaptive stress field generation algorithm and the policy network of the reinforcement learning agent.

2. The fully automated SSD aging test system according to claim 1, characterized in that: The central control and data processing unit receives analysis results from the digital twin model and uses an adaptive stress field generation algorithm to generate real-time test commands. The adaptive stress field generation algorithm includes: The objective function F(S) is defined as: F(S) = α · (dRUL / dt) + β · (Fault_Entropy) - γ · (Power_Consumption), Where S is the stress vector, dRUL / dt is the remaining lifetime decay rate predicted based on digital twin, (Fault_Entropy is the fault entropy, Power_Consumption is the system power consumption, the metaheuristic algorithm is used to solve the optimal stress vector S_optimal that maximizes F(S) in real time, and apply it to the SSD body, the objective function is used to determine how to test, and generate the test strategy problem.

3. The fully automated SSD aging test system according to claim 2, characterized in that: The stress vector S includes voltage Vcc, temperature T, data load mode L, and read / write ratio R / W.

4. The fully automated SSD aging test system according to claim 3, characterized in that: The digital twin model includes a time-varying degradation equation describing the threshold voltage drift of the NAND cell: ΔV_th(t) = A · [ln(1 + B · t)]^C · exp(-E_a / (k · T(t))) · (1 + D · P / E_Cycles) Wherein, ΔV_th(t): threshold voltage drift after time t, A, B, C, D: fitting parameters related to process technology and material properties, obtained through prior characterization tests, E_a: activation energy of the degradation process, k: Boltzmann constant, T(t): real-time temperature, which is time-varying, P / E_Cycles: number of erase / write cycles, realistically simulating data retention errors and read interference errors, which determines why the model can make accurate predictions and improves model accuracy and physical reliability.

5. The fully automated SSD aging test system according to claim 4, characterized in that: The central control and data processing unit models the testing process as a Markov decision process, where the state is a comprehensive health index of the SSD ontology and its twin model, the action is the adjustment of the stress field, the reward is the value of the F(S) function, and the deep reinforcement learning agent is trained to learn the optimal testing strategy.

6. The fully automated SSD aging test system according to claim 5, characterized in that: Also includes: A test performance index is introduced to measure the return on investment of a single test: Test Performance Index (TEI) = (Σ W_i · F_i) / (T · E), Where F_i: the severity weight of the i-th type of fault discovered, T: the total test time, and E: the total test energy consumption. W_i: The weight of the fault type. It continuously records the TEI of all test tasks and uses this data to back-optimize the adaptive stress field generation algorithm and the policy network of the reinforcement learning agent, forming a self-evolving test system that gets smarter the more it is used.

7. A fully automated SSD aging test method, characterized in that: The test method using the fully automated SSD aging test system as described in any one of claims 1-5 is as follows: Step 1: Initialization and Digital Twin Construction Step 1.1: Input the basic information of the SSD under test into the central control and data processing unit, set the test target, and automatically initialize the weight coefficients α, β, γ of the objective function F(S) in the adaptive stress field generation algorithm according to the test target; Once the task is assigned, the test chamber performs a self-test, the SSD is installed and powered on, and the central control unit drives the FPGA to perform characteristic testing on the SSD, collect data, and build and calibrate its corresponding digital twin model. Step 1.2: The digital twin model creates a unique virtual instance of the current SSD and loads a high-fidelity digital twin model that matches the configuration of the SSD. The initial parameters of the model are preset according to the NAND chip characteristics of the SSD. Step 2: Closed-loop test of the core loop Step 2.1 The test chamber receives the current optimal stress vector S_current = [Vcc, T, L, R / W, ...] from the adaptive stress field. Multi-channel programmable DC power supply with adjustable output voltage Vcc. The semiconductor cooler precisely adjusts the temperature of the test chamber to the target temperature T. The FPGA substrate receives instructions from the adaptive stress field generator of the central control and data processing unit. The instruction packet contains a specific stress vector S, and applies data flow to the SSD body. According to the testing strategy, the FPGA substrate simultaneously performs directional perturbation tests on the high-fidelity digital twin model, injecting abnormal command sequences; Step 2.2: The system collects real-time data of the SSD at a high frequency, including: Performance data: read / write bandwidth, IOPS, latency. Health data: Raw SMART values ​​for the solid-state drive. Physical data: real-time chip temperature and operating current. Logical data: Obtain the status information of the internal FTL of the SSD through the SSD's debug interface, manufacturer-specific commands, or internal log pages for the calibration of the digital twin model; Step 2.3: Using the new data acquired in Step 2.2, the digital twin high-fidelity model is calibrated to ensure consistency with the SSD in key states. After calibration, the digital twin high-fidelity model runs forward in accelerated mode to perform simulations. During the simulation, the physical degradation equations calculate the expected degradation of the NAND chip based on the historical stress S_history, and the predictive analytics module generates a preview of the future state accordingly. Key predictive metrics output during the simulation include: Remaining service life decay rate, potential failure modes and their probability of occurrence, and failure entropy under current stress; Step 2.4: The adaptive stress field receives predicted metrics and real-time power consumption data from the digital twin model. Generate a maximizing objective function F(S) = α · (dRUL / dt) + β · (Fault_Entropy) - γ · (Power_Consumption), and run a metaheuristic algorithm to solve for the optimal stress vector S_next for the next test cycle. Reinforcement learning agents observe the entire state-action-reward sequence and update their policy network, thereby learning what stress adjustments to take in what states to obtain the maximum long-term reward. Step 3: Termination, Analysis, and System Evolution Step 3.1: Fault Diagnosis and Test Termination The system monitors the following termination conditions in real time, and will automatically terminate the test if any condition is met: The following events may occur: functional failure of the SSD itself, high-fidelity prediction of the SSD's lifespan by the digital twin model, reaching the preset test duration or number of cycles, or manual interruption by the operator. Once a fault is detected, the system records a detailed fault context and automatically saves all relevant data for subsequent root cause analysis. Step 3.2: Test Performance Evaluation and Feedback Optimization After the test, the system calculates the test performance index: TEI = (Σ W_i · F_i) / (T ·E), The TEI value, along with detailed test logs, is stored in the knowledge base.

8. The fully automated SSD aging test method according to claim 1, characterized in that: The digital twin high-fidelity model in step 1.2 is continuously calibrated in step 2.

3. The digital twin high-fidelity model and the SSD body are synchronized for the first time to ensure that the logical state is consistent.

9. The fully automated SSD aging test method according to claim 1, characterized in that: It also includes the system regularly using accumulated Test Performance Index (TEI) data and test records to conduct offline training and optimization of the parameters of the adaptive stress field generation algorithm and the policy network of the reinforcement learning agent, so that the entire system becomes more and more efficient and intelligent as test experience accumulates.

10. A computer-readable medium having a computer program stored thereon, characterized in that: When the program is executed by the processor, it implements the steps of the method as claimed in any one of claims 7-9.

Images