A radiation dose measurement method and system based on double decision level and vertical fusion

Abstract

This invention discloses a radiation dose measurement method and system based on dual-decision level and vertical fusion, solving the technical problem that existing radiation dose measurement methods ultimately lead to poor accuracy of radiation dose measurement results. The method includes configuring a dedicated monitoring area in a 3D NAND flash memory chip, integrating idle physical blocks within it to obtain an independent monitoring area; performing programming writing and multi-reference voltage scanning on the main monitoring block within the area, outputting dual-decision levels; based on the physical page mapping relationship of the main monitoring block, forming monitoring sampling groups according to vertical positions and counting the total number of bits in each sampling group; performing dual-reference voltage reading based on the dual-decision levels and the total number of bits, outputting the truncated cumulative probability; screening valid monitoring sampling groups and extracting their threshold distribution mean, performing baseline alignment and dose inversion, and finally outputting the radiation dose estimate.

Description

Technical Field

[0001] This invention relates to the field of radiation sensing and detection technology, and in particular to a radiation dose measurement method and system based on dual decision levels and vertical fusion. Background Technology

[0002] In extreme application scenarios in key national fields such as aerospace, high-altitude exploration, and nuclear industry, electronic equipment, as the core carrier of system operation, is exposed to a continuous cumulative radiation field environment for a long time, facing severe reliability challenges posed by total ionizing dose (TID). The performance degradation and functional failure of devices caused by cumulative radiation can directly lead to abnormal operation of electronic equipment, thereby threatening the safe and stable operation of the entire critical equipment system. Therefore, achieving accurate and real-time monitoring of total ionizing dose has become a core technical prerequisite for ensuring the reliable operation of electronic equipment in extreme scenarios.

[0003] As a core non-volatile memory device in extreme scenarios, 3D NAND Flash is extremely sensitive to total ionizing radiation. Radiation energy directly causes a large number of charge traps and interface states in its tunneling layer and interface region. These radiation-induced defects trap transported charges, weaken the storage charge control capability, and cause unexpected shifts in the threshold voltage of the memory cell. As the radiation dose accumulates, the threshold voltage shift will exceed the device's error correction capability boundary, ultimately causing data read errors and storage failures. However, traditional radiation dose measurement methods are difficult to adapt to the 3D NAND memory architecture and cannot achieve in-situ dose monitoring at the device level. There is an urgent need for targeted radiation dose measurement methods to solve this technical pain point.

[0004] Existing radiation dose measurement methods typically rely on limited online monitoring samples, using statistical analysis of correlated data to quantify radiation dose. However, these methods are limited by the available sample size of online monitoring, and non-ideal factors such as readout noise, random telegraph noise (RTN), and local weak pages can easily cause drastic fluctuations in statistical values. This instability in statistical values ​​makes it difficult to effectively filter out a few outliers, which may even dominate the overall statistical results, leading to dose misjudgment or calibration curve drift, ultimately resulting in poor accuracy of radiation dose measurement results. Summary of the Invention

[0005] This invention provides a radiation dose measurement method and system based on dual decision levels and vertical fusion, which solves the technical problem that existing radiation dose measurement methods ultimately lead to poor accuracy of radiation dose measurement results.

[0006] The first aspect of this invention provides a radiation dose measurement method based on dual-decision level and vertical fusion, comprising:

[0007] A dedicated monitoring area is configured in the 3D NAND flash memory chip, and the idle physical blocks in the dedicated monitoring area are integrated to obtain an independent monitoring area;

[0008] Programming and multi-reference voltage scanning are performed on multiple main monitoring blocks within the independent monitoring area, and dual decision levels are output.

[0009] Based on the physical page mapping relationship of each main monitoring block in the independent monitoring area, multiple monitoring sampling groups are formed according to the vertical position, and the total number of bits of each monitoring sampling group is counted.

[0010] Based on the dual decision levels and the total number of bits in each monitoring sampling group, dual reference voltage readings are performed on each monitoring sampling group, and the cumulative probability after clipping of each monitoring sampling group is output.

[0011] The inverse function of the cumulative distribution function of Gaussian distribution is used to filter each monitoring sampling group based on the dual decision level and the cumulative probability after pruning of each monitoring sampling group, outputting multiple effective monitoring sampling groups, and extracting the mean threshold distribution of each effective monitoring sampling group;

[0012] Baseline alignment and dose inversion are performed on the mean threshold distribution of each effective monitoring sampling group to output radiation dose estimates.

[0013] Optionally, the step of performing programming write and multi-reference voltage scanning on multiple master monitoring blocks within the independent monitoring area, and outputting dual decision levels, includes:

[0014] The physical pages of each main monitoring block are written into a fixed pattern according to the preset target programming state to obtain multiple programmed main monitoring blocks;

[0015] Read the threshold distribution data of the physical pages of each programmed master monitoring block, and calculate the mean and standard deviation of the threshold distribution under the baseline state based on the threshold distribution data of the physical pages of each programmed master monitoring block.

[0016] The dual-decision level is calculated based on the mean and standard deviation of the threshold distribution under the baseline state.

[0017] Optionally, the step of forming multiple monitoring sampling groups according to vertical position based on the physical page mapping relationship of each main monitoring block within the independent monitoring area, and calculating the total number of bits of each monitoring sampling group, includes:

[0018] Based on the physical page mapping relationship of each main monitoring block, the vertical position coordinates of multiple physical pages of the corresponding main monitoring block are extracted and divided into multiple continuous vertical intervals.

[0019] In each of the aforementioned continuous vertical intervals, a preset number of physical pages are selected to form corresponding monitoring sampling groups;

[0020] The number of bits per page in each monitoring sampling group is obtained by querying the physical page mapping relationship and summing them up to get the total number of bits in each monitoring sampling group.

[0021] Optionally, the cumulative probability after clipping includes the cumulative probability after clipping the first reference voltage and the cumulative probability after clipping the second reference voltage; the step of performing dual reference voltage readings on each monitoring sampling group based on the dual decision level and the total number of bits in each monitoring sampling group, and outputting the cumulative probability after clipping of each monitoring sampling group, includes:

[0022] A first reference voltage and a second reference voltage are obtained, and the first reference voltage and the second reference voltage are applied to the storage unit of each physical page in each monitoring sampling group, respectively, and the first binary decision data and the second binary decision data of each physical page in each monitoring sampling group are output.

[0023] Within each monitoring sampling group, the number of on-cell units or non-on-cell units is counted in the first binary decision data and the second binary decision data of each physical page to obtain the first reference voltage cumulative bit count and the second reference voltage cumulative bit count for each monitoring sampling group.

[0024] Based on the total number of bits, the cumulative number of bits of the first reference voltage, and the cumulative number of bits of the second reference voltage for each monitoring sampling group, the cumulative probability after clipping the first reference voltage and the cumulative probability after clipping the second reference voltage for each monitoring sampling group are calculated.

[0025] Optionally, the step of using the inverse function of the Gaussian cumulative distribution function to filter each monitoring sampling group based on the dual decision level and the cumulative probability after pruning of each monitoring sampling group, outputting multiple valid monitoring sampling groups, and extracting the mean threshold distribution of each valid monitoring sampling group, includes:

[0026] The cumulative probability after clipping the first reference voltage and the cumulative probability after clipping the second reference voltage of each monitoring sampling group are transformed by the inverse function of the cumulative distribution function of the Gaussian distribution, respectively, to obtain the first normal domain value and the second normal domain value of each monitoring sampling group.

[0027] A monitoring sampling group in which the difference between any value in the first normal domain and any value in the second normal domain is greater than or equal to a preset difference threshold is considered a valid monitoring sampling group.

[0028] The threshold distribution mean of each effective monitoring sampling group is calculated based on the dual decision levels, the first normal domain value, and the second normal domain value of each effective monitoring sampling group.

[0029] Optionally, the step of baseline alignment and dose inversion of the mean threshold distribution of each of the effective monitoring sampling groups to output a radiation dose estimate includes:

[0030] Extract the mean of the baseline threshold distribution corresponding to each of the effective monitoring sampling groups;

[0031] Subtract the mean of the baseline threshold distribution and the mean of the threshold distribution corresponding to each effective monitoring sampling group to output the increment of the mean of the threshold distribution for each effective monitoring sampling group;

[0032] The mean increment of the threshold distribution of each effective monitoring sampling group is fused and calculated to obtain a unified dose signal;

[0033] The number of groups corresponding to the effective monitoring sampling groups and the inter-group dispersion of the mean increment are statistically analyzed.

[0034] If the number of groups corresponding to the effective monitoring sampling group is greater than or equal to a preset group number threshold and the inter-group dispersion of the mean increment corresponding to the effective monitoring sampling group is less than or equal to a preset dispersion threshold, then the unified dose signal is taken as a reliable dose signal.

[0035] The radiation dose estimate is calculated using the pre-stored calibration coefficients and the reliable dose signal.

[0036] A second aspect of the present invention provides a radiation dose measurement system based on dual-decision level and vertical fusion, comprising:

[0037] An integration module is used to configure a dedicated monitoring area in a 3D NAND flash memory chip and integrate the idle physical blocks in the dedicated monitoring area to obtain an independent monitoring area;

[0038] The write and scan module is used to perform programming write and multi-reference voltage scan on multiple main monitoring blocks within the independent monitoring area, and output dual decision levels;

[0039] The module for building and counting is used to build multiple monitoring sampling groups according to their vertical positions based on the physical page mapping relationship of each main monitoring block in the independent monitoring area, and to count the total number of bits of each monitoring sampling group.

[0040] The reading module is used to perform dual reference voltage reading on each of the monitoring sampling groups according to the dual decision level and the total number of bits of each monitoring sampling group, and output the cumulative probability of each monitoring sampling group after clipping;

[0041] The filtering and extraction module is used to filter each monitoring sampling group based on the dual decision level and the cumulative probability after pruning of each monitoring sampling group using the inverse function of the cumulative distribution function of Gaussian distribution, output multiple effective monitoring sampling groups, and extract the threshold distribution mean of each effective monitoring sampling group;

[0042] The output module is used to perform baseline alignment and dose inversion on the mean threshold distribution of each of the effective monitoring sampling groups, and output the radiation dose estimate.

[0043] A third aspect of the present invention provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of the radiation dose measurement method based on dual decision levels and vertical fusion as described above.

[0044] The fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed, it implements the radiation dose measurement method based on dual-decision level and vertical fusion as described above.

[0045] The fifth aspect of the present invention provides a computer program product comprising a computer program stored on a non-transitory computer-readable storage medium, the computer program comprising program instructions, wherein, when the program instructions are executed by a computer, the computer performs the steps of the radiation dose measurement method based on dual-decision level and vertical fusion as described above.

[0046] As can be seen from the above technical solutions, the present invention has the following advantages:

[0047] The above-mentioned technical solution of the present invention provides a radiation dose measurement method based on dual-decision level and vertical fusion. A dedicated monitoring area is configured in a 3D NAND flash memory chip, and the idle physical blocks in the dedicated monitoring area are integrated to obtain an independent monitoring area. Programming and multi-reference voltage scanning are performed on multiple main monitoring blocks within the independent monitoring area to output dual-decision levels. Based on the physical page mapping relationship of each main monitoring block within the independent monitoring area, multiple monitoring sampling groups are formed according to vertical position, and the total number of bits in each monitoring sampling group is calculated. According to the dual-decision level and the total number of bits in each monitoring sampling group, dual-reference voltage readings are performed on each monitoring sampling group, and the pruned cumulative probability of each monitoring sampling group is output. The inverse function of the Gaussian cumulative distribution function is used to filter each monitoring sampling group according to the dual-decision level and the pruned cumulative probability of each monitoring sampling group, outputting multiple effective monitoring sampling groups, and extracting the threshold distribution mean of each effective monitoring sampling group. Baseline alignment and dose inversion are performed on the threshold distribution mean of each effective monitoring sampling group to output a radiation dose estimate. Based on the above method... This invention, through the configuration and integration of independent monitoring areas, ensures the specificity and stability of monitoring samples, laying the foundation for improving measurement accuracy. The dual-decision-level output provides a unified and standardized basis for subsequent sampling and effective group screening, avoiding measurement deviations caused by inconsistent judgment criteria. The design of multiple monitoring sampling groups based on vertical positions enables multi-point vertical sampling, effectively expanding the effective sample size for online monitoring. Simultaneously, it covers physical pages at different vertical positions, specifically suppressing the impact of vertical position differences on the consistency of measurement output. Combining dual-decision-level analysis with the total number of bits in each sampling group to perform dual-reference voltage readings and calculate the cumulative probability after clipping suppresses statistical fluctuations caused by non-ideal factors. Screening of effective monitoring sampling groups directly eliminates outliers, preventing them from dominating the overall statistical results. Baseline alignment eliminates the interference of inherent sample bias on measurement results, ultimately significantly improving the accuracy of radiation dose measurement. Attached Figure Description

[0048] To more clearly illustrate the technical solutions in the embodiments of the present invention 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 of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0049] Figure 1 The flowchart illustrates the steps of a radiation dose measurement method based on dual-decision level and vertical fusion, as provided in Embodiment 1 of the present invention.

[0050] Figure 2 This is a schematic diagram of multi-point sampling and fusion provided in Embodiment 1 of the present invention;

[0051] Figure 3 This is a schematic diagram of the inversion principle provided in Embodiment 1 of the present invention;

[0052] Figure 4 This is a schematic diagram of the fusion signal fitting effect provided in Embodiment 1 of the present invention;

[0053] Figure 5 This is a schematic diagram illustrating the consistency of dose inversion provided in Embodiment 1 of the present invention;

[0054] Figure 6 This is an overall framework diagram of a radiation dose measurement method based on dual-decision level and vertical fusion, provided in Embodiment 1 of the present invention.

[0055] Figure 7 This is a structural block diagram of a radiation dose measurement system based on dual decision levels and vertical fusion, provided in Embodiment 2 of the present invention. Detailed Implementation

[0056] This invention provides a radiation dose measurement method and system based on dual-decision level and vertical fusion, which solves the technical problem that existing radiation dose measurement methods ultimately lead to poor accuracy of radiation dose measurement results.

[0057] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. It should be noted that in the optional embodiments of the present invention, the object information and other related data involved require the permission or consent of the object when the embodiments of the present invention are applied to specific products or technologies, and the collection, use, and processing of related data must comply with the relevant laws, regulations, and standards of the relevant countries and regions. That is to say, if the embodiments of the present invention involve data related to the object, it needs to be obtained with the authorization and consent of the object, the authorization and consent of the relevant departments, and in compliance with the relevant laws, regulations, and standards of the country and region. If personal information is involved in the embodiments, the acquisition of all personal information requires the consent of the individual. If sensitive information is involved, the separate consent of the information subject is required, and the embodiments also need to be implemented with the authorization and consent of the object.

[0058] Terminology Explanation:

[0059] 3D NAND Flash: 3D NAND Flash is a non-volatile memory formed by stacking memory cells vertically, achieving higher density through vertical stacking. Compared to planar NAND, it exhibits more significant process gradients and structural differences in the vertical direction (e.g., structural variations due to etching aspect ratio, stress distribution differences, etc.), resulting in potentially different threshold voltage distributions, drift rates, and noise levels at different vertical locations. This invention uses the threshold voltage distribution drift of 3D NAND as an "observable measure" of radiative accumulation effects and reduces the impact of vertical location differences on dose output consistency by sampling and fusing at multiple vertical locations.

[0060] Deck (Stack / Sub-Stack): In some 3D NAND structures, the entire vertically stacked wafer can be divided into multiple sub-stacks (commonly called decks) according to the manufacturing or connection method. The initial threshold distribution, readout noise, and radiation-induced drift may differ between different decks. This invention can use the deck as one of the sampling dimensions: either a single deck can be selected to establish a dosimeter, or monitoring pages from multiple decks can be selected simultaneously and fused to improve cross-structure consistency and stability.

[0061] Wordline / Physical Page: A wordline is a wire connecting the control gates of memory cells in the same row; in engineering implementations, a single read / write operation is typically measured in units of "pages." A physical page refers to the smallest independently readable and writable data page unit in the NAND physical address space (independent of logical mapping). This invention uses physical pages as the basic sampling object. By selecting several physical pages and performing grouped statistical analysis, a stable cumulative probability estimate is obtained, thereby enabling dose signal calculation without introducing additional analog measurements.

[0062] Monitor Block / Monitor Region: A monitor block / monitor region is a set of one or more physical blocks or pages reserved for dose monitoring, typically not carrying user data. Its functions include: fixing the monitored objects, ensuring that the read statistics correspond to the same batch of controllable units, and avoiding confounding factors introduced by changes in business data; fixing the programming state, allowing preset patterns to be written into the monitoring page or units to be written to the same target state as much as possible to form a repeatable threshold distribution; facilitating maintenance, allowing operations such as rewriting / refreshing / rebuilding the baseline to be performed within the maintenance window, and supporting redundant monitor blocks to avoid the impact of bad blocks.

[0063] Threshold voltage V th With threshold distribution: threshold voltage V this the equivalent gate voltage parameter required to make the memory cell transistor reach the conduction condition. Due to factors such as random doping, interface states, and size fluctuations, even in the same programming state, the V of a large number of cells th also exhibits a statistical distribution (threshold distribution). Radiation (especially the total ionizing dose effect) can introduce or change the oxide trap charges and interface states, resulting in an overall drift, broadening, or morphological change of the threshold distribution. The present invention focuses on the change of the "location parameter" of the threshold distribution and estimates it with a finite number of readout points.

[0064] Decision Level / Reference Voltage: The decision level (reference voltage) is the reference point for comparison during NAND reading, which is equivalent to making a threshold decision on the threshold distribution at a certain voltage. Changing the reference voltage is equivalent to "sampling" the cumulative probability points of the threshold distribution at different voltage positions. In engineering, multi-reference voltage reading can be achieved by configuring the read reference voltage register or the Read-Retry / Shift Read mechanism. The present invention utilizes this ability to complete statistics at two fixed decision levels, thereby avoiding a full-range voltage scan.

[0065] Cumulative probability p(V) (CDF point): The cumulative probability is defined as: p(V)=P(V th <V). It represents the proportion of cells with a threshold lower than the reference voltage V. In engineering implementation, after the monitoring area is written into a preset state, reading with the reference voltage V can obtain the binary result of each cell under this decision. By counting the number of bits "judged to be on the low-threshold side" and dividing by the total number of bits, p(V) can be estimated. The present invention obtains p1 = p(V1) and p2 = p(V2) at V1 and V2 respectively, as the input for subsequent inversion. For the convenience of system implementation, the cumulative count and the total count (Count / Total) can also be output simultaneously instead of directly outputting the probability value.

[0066] (·), (·) (standard normal distribution function and quantile function): (·) is the cumulative distribution function of the standard normal distribution,<00​​​​​​​​​​​​Characterizes the broadening (dispersion) of the radiation distribution. Within the dose range where radiation primarily causes overall drift, It is often closer to a calibrable signal that is directly related to the dose; It can be used as an auxiliary quantity for quality monitoring and anomaly detection. This invention uses... (or Δ after baseline alignment) () serves as the main output signal of the dosimeter.

[0068] Inversion: Inversion refers to the process of deducing distribution parameters from measurable readout statistics. This invention utilizes two cumulative probability points (p1, p2) to solve for the distribution parameters under the Gaussian approximation. , This falls under the category of "parameter inversion." To ensure the stability of the inversion, this invention incorporates probability pruning and validity determination (e.g., z2-z1≠0, ...). Mechanisms such as >0 are used to avoid numerical divergence or abnormal results affecting the final dose output.

[0069] Probability clipping (Clipping Epsilon): When p is close to 0 or 1, z = (p) will increase sharply, leading to numerical instability. Probability clipping refers to limiting the probability to an interval [...]. ,1- ].in A small positive number (e.g., 10^-6 to 10^-3) can be selected and fixed as a system parameter. Pruning can also be seen as a protection against extreme statistical fluctuations (e.g., too few samples or abnormal readouts).

[0070] Monitoring Sampling Group / Vertical Sampling Group: The monitoring sampling group is a key organizational method in this invention used to replace "single-page monitoring": within the monitoring area, several physical pages are selected according to their physical page numbers (which can approximately correspond to vertical positions) to form a set P. k (k is the sampling group index), each set is called a sampling group. Sampling groups serve the following purposes: 1. Suppressing random noise: A single sampling group contains multiple pages; averaging within the group can reduce the impact of single-page readout noise and occasional anomalies. 2. Covering vertical differences: Different sampling groups correspond to different vertical position bands, used to characterize the impact of vertical position differences on the dose signal. 3. Supporting robust fusion: The outputs of multiple sampling groups can be fused subsequently to obtain a more stable and consistent output. Sampling groups can be constructed as follows: within each deck, available physical pages are sorted by page number, and then grouped according to a fixed group size N. g Segmentation can be performed; alternatively, a predefined "key page number set" can be used for non-equidistant sampling.

[0071] Within-group averaging: Within-group averaging refers to the averaging of the sampled group P. k The counts / probabilities obtained from each physical page under the same reference voltage are averaged or robustly statistically analyzed (e.g., median, clipped mean) to obtain a more stable p1. k p2 k Or after page-by-page inversion The within-group average was obtained Intra-group statistics can be combined with "abnormal page removal": for example, pages that significantly deviate from the intra-group distribution can be removed or downgraded to enhance long-term robustness.

[0072] Multi-point sampling fusion: Multi-point sampling fusion refers to combining the outputs of multiple sampling groups. } or {Δ The signal is synthesized into a single dose signal through a fusion operator. Common fusion methods include: 1. Mean fusion: simple to implement and highly efficient when noise is approximately symmetrical; 2. Median fusion: more robust to outlier sampling groups; 3. Pruned mean fusion: calculates the mean after removing a certain percentage from both ends, balancing accuracy and robustness; 4. Weighted fusion (optional): sets weights based on noise levels, linearity, historical stability, etc., for each group. The purpose of fusion is to suppress the offset caused by vertical position differences and local anomalies, making the output easier to uniformly calibrate and stabilize the inverted dose.

[0073] Baseline Alignment: Baseline alignment refers to eliminating threshold voltage deviations caused by differences in the initial state or inherent structure of the chip. It is usually achieved by measuring the initial threshold distribution mean (i.e., baseline) of each physical page or sampling group of the chip before irradiation (or at the factory), and then subtracting the baseline value from the currently measured threshold distribution mean during subsequent online monitoring to extract the threshold distribution mean increment (Δμ) purely caused by irradiation.

[0074] Calibration and Dose Estimation: Calibration is the process of establishing a reversible mapping between the fusion output and the dose. Based on " Based on the empirical rule that "approximates one dose within a certain dosage range," this invention can be calibrated using a linear function: (D)≈aD+b. This was obtained during online monitoring. Then, it can be accessed via: D hat =( -b) / a. Inversion yields the estimated dose D hat a and b can be obtained during the manufacturing or maintenance phase and stored in the controller's non-volatile media.

[0075] Quality Metrics and Validity Check: Quality metrics are used to determine the reliability of the current readout and inversion results. Typical metrics include: 1. Whether p1 and p2 are close to saturation (whether they have reached the clipping boundary); 2. Whether z2-z1 is too small, causing inversion instability; 3. The inversion results... Is it positive and within a reasonable range?

[0076] TID (Total Ionizing Dose): Total ionizing dose (TID) describes the intensity of the cumulative ionizing effect in a radiation environment. Commonly used units include rad (Si) or krad (Si). The target quantity output by this invention is TID (or a dose index monotonically corresponding to TID), and it supports dose inversion through calibration.

[0077] Please see Figure 1 , Figure 1 This is a flowchart illustrating the steps of a radiation dose measurement method based on dual-decision level and vertical fusion, as provided in Embodiment 1 of the present invention.

[0078] This invention provides a radiation dose measurement method based on dual-decision level and vertical fusion, comprising:

[0079] Step 101: Configure a dedicated monitoring area in the 3D NAND flash memory chip and integrate the idle physical blocks in the dedicated monitoring area to obtain an independent monitoring area.

[0080] 3D NAND flash memory chips are non-volatile memory devices with a three-dimensional stacked structure. By vertically stacking memory cells, storage density is increased, and they are the core carrier for radiation dose monitoring in this invention.

[0081] The dedicated monitoring area is a specific area specially designated within the 3D NAND flash memory chip for radiation dose monitoring operations. It is distinct from the chip's normal data storage area to ensure the exclusivity of the monitoring operations.

[0082] Idle physical blocks are physical blocks in 3D NAND flash memory chips that have not been allocated for normal data storage and are in an idle state, and have the potential to be integrated for monitoring related operations.

[0083] The independent monitoring area is an independent space formed by screening and integrating the idle physical blocks in the dedicated monitoring area. It serves as the basic carrier for the subsequent monitoring-related component settings and operations, and has the characteristics of being independent of the normal storage area of ​​the chip.

[0084] It should be noted that a dedicated monitoring area is configured within the 3D NAND flash memory chip, and the idle physical blocks within this dedicated monitoring area are integrated to obtain an independent monitoring area. This configuration and integration process delineates a dedicated monitoring space and filters and integrates available idle physical blocks, providing a stable and dedicated foundation for subsequent radiation dose monitoring operations and ensuring that the monitoring process does not interfere with the chip's normal storage function.

[0085] It is worth mentioning that, in order to decouple the dose output from the operational data, this invention configures a dedicated monitoring area inside the NAND flash memory to store monitoring pages with fixed patterns and in a preset programming state. This monitoring area can consist of one or more physical blocks and is marked as a reserved area in the firmware to avoid entering the user data mapping and garbage collection process, thereby ensuring the consistency of the monitored object during long-term operation; at the same time, redundant monitoring blocks can be configured to deal with bad blocks or local degradation.

[0086] Step 102: Perform programming and multi-reference voltage scanning on multiple master monitoring blocks within the independent monitoring area, and output dual decision levels.

[0087] The main monitoring block is a physical block selected within an independent monitoring area to perform core monitoring operations such as programming and multi-reference voltage scanning. It is the core unit for data acquisition and processing during radiation dose monitoring.

[0088] The dual decision level refers to two reference level values ​​(V1 and V2) calculated based on the data obtained from the multi-reference voltage scan. These values ​​serve as the standard basis for determining the conduction status of the memory cell during subsequent monitoring and sampling group reading operations.

[0089] It should be noted that, based on the independent monitoring area formed by the aforementioned integration, multiple main monitoring blocks within the area are selected, and the two operations of programming and multi-reference voltage scanning are completed in sequence. By processing the data obtained from the scanning, dual decision levels are finally output, providing a unified judgment benchmark for the subsequent reading and filtering operations of each monitoring sampling group.

[0090] Furthermore, step 102 may include the following sub-steps:

[0091] S21. Write the physical pages of each main monitoring block into a fixed pattern according to the preset target programming state to obtain multiple programmed main monitoring blocks;

[0092] S22. Read the threshold distribution data of the physical pages of each programmed main monitoring block, and calculate the mean and standard deviation of the threshold distribution under the baseline state based on the threshold distribution data of the physical pages of each programmed main monitoring block.

[0093] S23. Calculate the dual decision level based on the mean and standard deviation of the threshold distribution under the baseline state.

[0094] The target programming state is a pre-defined programming state used to unify the initial storage state of the physical pages of the main monitoring block, and it is the basic state setting for monitoring data acquisition.

[0095] The fixed pattern is a pre-determined, unified data format used to write to the physical page of the main monitoring block, ensuring the consistency of the initial state of each monitoring block.

[0096] The baseline state is the initial standard state after the main monitoring block has been programmed and written, and has not been affected by factors such as radiation and aging. It is the benchmark state for subsequent parameter comparison and calculation.

[0097] In 3D NAND flash memory chips, a deck is a structural unit composed of multiple vertically stacked storage cell layers, which is the basic unit for dividing vertical intervals and selecting boundary pages.

[0098] Boundary pages are the physical pages at the top and bottom of each deck, used to estimate more conservative dual-decision level candidate values ​​to improve the stability of the worst-case page.

[0099] It should be noted that the preferred method for writing the monitoring page is to use a fixed pattern and lock the target programming state, so that the monitoring page forms a repeatable threshold distribution in the target state. For example, in a TLC (Triple-Level Cell), a high-level state that is sensitive to radiation and has a relatively stable distribution pattern is selected as the target state (such as the P7 state). Within the maintenance window, the long-term drift caused by data retention and cyclic aging can be reduced by refreshing or rebuilding the baseline to reduce the mixed effects of long-term drift on dose output.

[0100] like Figure 2 As shown, considering the process gradients and structural differences in the vertical direction of 3D NAND, the response curves to radiation at different vertical locations (different word line regions or physical page number ranges) may differ significantly. Therefore, this invention selects monitoring sampling groups covering multiple vertical locations within the same monitoring area to improve representativeness. Specifically, within each deck (stack / sub-stack), adjacent pages can be grouped into a sampling group P after being sorted by physical page number. k Each group contains N g Each physical page is used to reduce the impact of random noise and local weak pages by averaging across pages within the group; the sampling group should cover multiple height ranges such as low, middle and high in the vertical direction, and can be selected simultaneously across decks to further improve robustness.

[0101] Furthermore, the present invention only requires two fixed decision levels V1 and V2 (satisfying V1 < V2) in each measurement cycle to complete statistics and parameter inversion, thereby achieving a balance between observability and read overhead. To ensure inversion stability and dynamic range, the selection of V1 and V2 should avoid long-term entry into the probability saturation region within the target dose range and enable the distance between the two points in the normal domain to have sufficient numerical resolution; an easy-to-implement engineering selection method is to estimate the mean of the target state threshold distribution under the reference condition of dose≈0 and standard deviation , and let V1 be taken at the left tail of the distribution to enhance the sensitivity to drift (for example , m can take 1 - 3), and at the same time let V2 be taken in the middle or slightly to the left of the middle of the distribution (for example V2≈ or ) and ensure V2 > V1; to take into account the stability of the most unfavorable page, a small number of boundary pages can also be selected at the top and bottom of each deck to estimate its( ), and thus obtain a more conservative candidate value for V1 and select V2 near the mean value of the full-page set, thereby reducing the sensitivity of the reference voltage configuration to cross-vertical differences.

[0102] In this embodiment, the physical pages of each main monitoring block are written with a fixed pattern according to a preset target programming state, obtaining multiple programmed main monitoring blocks, and then the threshold distribution data of the physical pages of each programmed main monitoring block are read. Based on the read threshold distribution data, the sum of the threshold voltages of all monitoring samples is divided by the total number of valid samples to obtain the mean value of the threshold distribution in the baseline state , and then the difference between each sample threshold voltage and this threshold distribution mean is squared and summed, and the obtained result is divided by the total number of valid samples and then square-rooted to obtain the standard deviation of the threshold distribution in the baseline state , and then based on the mean value of the threshold distribution in this baseline state and the standard deviation of the threshold distribution to determine the dual decision levels V1 and V2 (satisfying V1 < V2), where V1 can be selected at the left tail of the distribution to enhance the sensitivity to drift, for example, let , m can take 1 - 3, and at the same time let V2 be taken in the middle or slightly to the left of the middle of the distribution (for example V2≈ or ) and ensure V2 > V1; to take into account the stability of the most unfavorable page, a small number of boundary pages can also be selected at the top and bottom of each deck to estimate its( , This yields a more conservative candidate value for V1, and V2 is selected near the mean of the entire set, thereby reducing the sensitivity of cross-vertical differences to the reference voltage configuration. The dual decision level is determined through this standardized selection method, ensuring both inversion stability and dynamic range while avoiding prolonged probability saturation within the target dose range. This ensures sufficient numerical resolution of the distance between the two points in the normal domain, and is easy to implement in engineering. It can estimate the mean of the target state threshold distribution under the baseline condition of dose≈0. with standard deviation This provides a precise and unified judgment standard for the subsequent reading and screening of sampling groups, reduces the interference of benchmark deviation on radiation dose measurement results, and improves the accuracy of radiation dose measurement.

[0103] Step 103: Based on the physical page mapping relationship of each main monitoring block in the independent monitoring area, multiple monitoring sampling groups are formed according to the vertical position, and the total number of bits of each monitoring sampling group is calculated.

[0104] It should be noted that, based on the physical page mapping relationship of each main monitoring block in the aforementioned independent monitoring area, the vertical position information corresponding to each physical page is extracted, continuous intervals are divided according to the vertical position, physical pages are selected in each interval to form a monitoring sampling group, and then the number of single-page bits of all physical pages in each sampling group is accumulated and counted to obtain the total number of bits of each monitoring sampling group, which provides basic data support for subsequent dual reference voltage reading and cumulative probability calculation.

[0105] Furthermore, step 103 may include the following sub-steps:

[0106] S31. Based on the physical page mapping relationship of each main monitoring block, extract the vertical position coordinates of multiple physical pages of the corresponding main monitoring block and divide them into multiple continuous vertical intervals.

[0107] S32. Select a preset number of physical pages in each continuous vertical interval to form a corresponding monitoring sampling group;

[0108] S33. Query the number of bits per page of all physical pages in each monitoring sampling group through the physical page mapping relationship, and sum them up to obtain the total number of bits of each monitoring sampling group.

[0109] Vertical position coordinates are a quantitative identifier used to characterize the specific hierarchical position of a physical page in the vertical stacking structure of 3D NAND flash memory chips, and are the core basis for dividing vertical intervals.

[0110] A continuous vertical interval is a segment of intervals that are continuous and non-overlapping in the vertical direction, divided according to the vertical position coordinate range of the physical page, and is used to define the vertical coverage range of the monitoring sampling group.

[0111] The preset number of physical pages is the number of physical pages selected in each vertical interval to form a monitoring sampling group, ensuring the consistency of sample size in each sampling group.

[0112] The number of bits per page is the number of storage bits contained in a single physical page, and it is the basic unit data for calculating the total number of bits in a monitoring sampling group.

[0113] It should be noted that, based on the physical page mapping relationship of each main monitoring block, the vertical position coordinates of each physical page within the corresponding main monitoring block are extracted from the mapping relationship. Multiple continuous and non-overlapping vertical intervals are obtained according to the coordinate range. Then, within each vertical interval, physical pages are selected according to a preset sampling number. The physical pages selected within the same interval are combined into corresponding monitoring sampling groups. Afterwards, the number of bits per page of all physical pages in each monitoring sampling group is queried through the physical page mapping relationship. The number of bits per page of each physical page is then accumulated sequentially to obtain the total number of bits corresponding to each monitoring sampling group. By uniformly dividing the intervals according to the vertical position and forming sampling groups, vertical multi-point sampling is achieved, effectively expanding the effective sample size. Simultaneously, covering physical pages at different vertical positions can suppress the impact of vertical position differences on the consistency of measurement output, providing reliable basic data for subsequent dual-reference voltage readings and cumulative probability calculations, thereby improving the accuracy of radiation dose measurement. The aforementioned design of "dividing continuous intervals across vertical locations and establishing multiple monitoring sampling groups" is a key technical approach for addressing the vertical process gradient (Layer-to-layer variation) in 3D NAND. By selecting physical pages in different vertical intervals to form sampling groups, multi-point sampling can be achieved, providing sufficient sampling data for subsequent signal fusion. This reduces inter-group dispersion, decreases the need for location-specific calibration, and makes it easier for the entire system to complete unified calibration and dose inversion using a linear function, thereby improving the accuracy and stability of radiation dose estimation.

[0114] It is worth mentioning that this invention addresses two key aspects: in the monitoring area configuration stage, "configuring a single, fixed dedicated monitoring area in the 3D NAND flash memory chip, and integrating the idle physical blocks within that area to form an independent monitoring area containing N main monitoring blocks," and in the sampling group formation stage, "selecting a single target programming state and forming a monitoring sampling group limited to that area." Specifically, this invention employs a sampling implementation method combining a single monitoring object and a fixed sampling range. Specifically, the storage controller configures a single, fixed dedicated monitoring area in the 3D NAND flash memory chip, integrates the idle physical blocks within that area to form an independent monitoring area containing N main monitoring blocks, selects a single preset target programming state S, reads the physical page mapping relationship of all main monitoring blocks under that programming state, extracts the vertical position coordinates to divide the interval, and forms a monitoring sampling group. The sampling range is limited to the vertical page set within this single monitoring area. Besides this implementation, alternative methods can be used: multi-state / multi-block / multi-chip expansion, i.e., simultaneously selecting multiple programming states (e.g., high-level state and medium-level state) to calculate the average threshold voltage distribution. Feature fusion is then performed to enhance adaptability to different dose ranges or individual devices; or the monitoring range can be expanded to multiple monitoring blocks, multiple semiconductor device dies, multiple channels or even multiple chips to form a larger spatial sampling set and perform hierarchical fusion in the spatial dimension, thereby further reducing the impact of local defects, bad blocks and occasional weak pages on the output. Moreover, this alternative still follows the overall idea of ​​"multi-point sampling - inversion / feature extraction - fusion - calibration".

[0115] Step 104: Based on the dual decision levels and the total number of bits in each monitoring sampling group, perform dual reference voltage readings on each monitoring sampling group and output the cumulative probability after clipping of each monitoring sampling group.

[0116] The cumulative probability after clipping includes the cumulative probability after clipping the first reference voltage and the cumulative probability after clipping the second reference voltage.

[0117] It should be noted that, based on the aforementioned dual decision levels as a unified judgment standard, and combined with the total number of bits counted for each monitoring sampling group, two reference voltage reading operations corresponding to the dual decision levels are performed on each monitoring sampling group. The decision data generated by the readings is combined with the total number of bits to complete the statistical calculation, and finally the cumulative probability after pruning of each monitoring sampling group is output, providing core data for the subsequent screening of monitoring sampling groups.

[0118] Furthermore, step 104 may include the following sub-steps:

[0119] S41. Obtain the first reference voltage and the second reference voltage, and apply the first reference voltage and the second reference voltage to the storage unit of each physical page in each monitoring sampling group respectively, and output the first binary decision data and the second binary decision data of each physical page in each monitoring sampling group.

[0120] S42. Count the number of conducting units or the number of non-conducting units in the first binary decision data and the second binary decision data of each physical page in each monitoring sampling group to obtain the first reference voltage cumulative bit number and the second reference voltage cumulative bit number of each monitoring sampling group.

[0121] S43. Based on the total number of bits in each monitoring sampling group, the cumulative number of bits of the first reference voltage, and the cumulative number of bits of the second reference voltage, calculate the cumulative probability of the first reference voltage after clipping and the cumulative probability of the second reference voltage after clipping for each monitoring sampling group.

[0122] It should be noted that in the circuit and register implementation, the decision level (Read Reference Voltage) is usually configured through the NAND internal read reference voltage register or the read-retry mechanism. The internal DAC (Digital-to-Analog Converter) generates the corresponding analog reference voltage based on the register value and applies it to the word line where the target memory cell is located. Subsequently, the sense amplifier evaluates the conduction status of the memory cell under the current reference voltage by detecting the discharge current or voltage change of the bit line, outputs a binary decision result, latches it into the page buffer, and finally transmits it to the controller's buffer via the NAND interface (Toggle / ONFI).

[0123] Furthermore, within a single measurement cycle, the controller monitors each sampling group P. k Perform two reads sequentially: First, configure the reference voltage to V1, read each page in the sampling group, and count the number of bits that meet the threshold decision condition (Count1). k Combined with the pre-stored total number of bits in this sampling group k (Total number of bits) k (Statistics have been completed during the initialization or baseline phase, and do not need to be repeated during subsequent online monitoring phases) to obtain the cumulative probability p1 after the first reference voltage cutoff. k =Count1 k / Total k Subsequently, the reference voltage is configured as V2, and the number of bits that meet the threshold decision condition (Count2) is repeatedly read and counted. k Combined with the same pre-stored total number of bits (Total) k The cumulative probability p2 after the second reference voltage is cut is obtained. k =Count2 k / Total k To avoid the need to calculate z= later. (p) When p is close to 0 or 1, numerical divergence or noise amplification occurs. This invention addresses p1. k With p2 k Perform probabilistic pruning, restricting p to... ,in For the clipping parameters (e.g., 10) -6 ~10 -3 The clipped probabilities are used as parameters for inversion input.

[0124] It is worth noting that for features of the decision level and reference voltage type, they can be interchangeably expressed as decision level, reference voltage, read threshold / decision threshold, and read cutoff; for features of the dual-decision level type (i.e., dual decision level), they can be interchangeably expressed as dual-decision level, dual threshold, dual reference voltage, and dual-threshold / two-Vref / two-point; for features of the cumulative probability p type (i.e., cumulative probability after clipping), they can be interchangeably expressed as cumulative probability p, cumulative count ratio, cumulative distribution function point (CDFPoint), and the proportion of thresholds below the threshold.

[0125] Meanwhile, for the cumulative probability calculation step of "performing bit-level statistics on page cache data to obtain the cumulative number of bits and the total number of bits, and calculating the cumulative probabilities p1 and p2 after pruning", this invention adopts a direct bit-level statistical cumulative probability calculation method. Specifically, after the storage controller sends V1 and V2 read instructions to the 3D NAND flash memory and receives the binary decision data of each monitoring sampling group, it directly performs bit-level statistics on the page cache data to obtain the cumulative number of reference voltages V1, V2, and the total number of bits for each monitoring sampling group, and calculates the cumulative probabilities p1 and p2 after pruning V1 and V2 based on this, which are used as inputs for subsequent inversion steps. In addition to this implementation, alternative approaches include firmware statistics / ECC multiplexing / hardware acceleration. This involves multiplexing statistical results provided by an ECC (Error Correcting Code) engine or read path counter, or implementing a dedicated bit-count acceleration unit in the controller to reduce power consumption and latency. Alternatively, the "cumulative probability" can be replaced with equivalent statistics (e.g., deviation counts based on fixed pattern comparisons, or partition counts based on read threshold decisions) and p1, p2, or directly mapped to position parameters through equivalent conversion. This alternative approach does not change the physical source and calculation target of the dose signal, only altering the statistical implementation path, making it easier to implement under different controller architectures and resource constraints.

[0126] Secondly, regarding the aforementioned operation of "counting the cumulative number of bits and calculating the cumulative probability after clipping": the core of obtaining the cumulative probability p is "counting the read results to obtain the cumulative count and the total count and then calculating the ratio," where "cumulative count" is the number of bits that meet the corresponding reference voltage decision condition (such as the total number of true values ​​(which are 1) when V1 is read), and "total count" is the total number of bits in the monitoring sampling group. In engineering implementation, the counting operation can be implemented by firmware bit statistics, or it can be implemented by reusing ECC (Error Correcting Code) statistics, read path counters, or hardware popcount (bit count) units; at the same time, "Count(V)" can be uniformly expressed as "the number of units / bits that meet the threshold decision condition" to cover the equivalent counting paths under different NAND read return formats and different controller architectures.

[0127] Furthermore, this invention addresses two key aspects of the decision level initialization process: "determining fixed V1 and V2 during initialization or factory testing and maintaining them unchanged during online monitoring," and "using fixed V1 and V2 to perform read operations" during online reading. Specifically, this invention employs a fixed decision level combined with probability pruning for the following two stages: First, during initialization or factory testing, the storage controller determines fixed dual decision levels V1 and V2 based on the mean and standard deviation of the threshold distribution under baseline conditions, maintaining these levels unchanged during online monitoring. Second, during online monitoring, the determined fixed V1 and V2 are consistently used to send read commands to the 3D NAND flash memory, performing targeted read operations. Probability pruning and validity determination are used to address probability saturation and abnormal situations. Alternatively, an adaptive decision level (adaptive V1 / V2) can be used, where, during online operation, the decision level is adjusted based on the currently measured p1 and p2 or the estimated p2. The decision level is adaptively fine-tuned to keep p1 and p2 within a preset range for a long time (e.g., avoiding saturation near 0 / 1 and maintaining sufficient...). The spacing expands the effective dynamic range and reduces the probability of failure due to long-term drift. The alternative still uses the "two-point inversion + fusion + calibration" framework as the main framework, but allows V1 and V2 as updatable parameters, and provides update triggering conditions, stepping strategies and rollback mechanisms to ensure online stability.

[0128] In this embodiment, the storage controller sends a reference voltage configuration command to the 3D NAND flash memory via a standard NAND interface. Upon receiving the command, the 3D NAND flash memory configures the digital parameters corresponding to the dual decision levels using an internal reference voltage register or a read-retry mechanism. A digital-to-analog converter (DAC) then converts these digital parameters into the corresponding first and second reference voltages, which are allocated to the sensing amplifiers corresponding to each physical page within each monitoring sampling group. Subsequently, the controller performs two read operations for each monitoring sampling group: first, it configures the reference voltage as the first reference voltage, applies the first reference voltage to the storage cells of each physical page within each monitoring sampling group, checks the conduction status of each storage cell, outputs a binary decision result, and latches it into the page cache. After reading the decision data from the page cache, the controller counts the number of conducting or non-conducting cells to obtain the cumulative number of bits for the first reference voltage. This is then combined with the pre-stored total number of bits for the sampling group. k (The total number of bits has been counted during the initialization or baseline phase and does not need to be counted again during the subsequent online monitoring phase.) Calculate the first original cumulative probability; then configure the reference voltage as the second reference voltage, apply the second reference voltage to the storage cells of each physical page in each monitoring sampling group, check whether each storage cell is conducting or not, repeat the above latching and counting process, and count the number of conducting cells or the number of non-conducting cells to obtain the cumulative number of bits of the second reference voltage, combined with the total number of bits stored in the same pre-stored Total. k The second original cumulative probability is calculated; to avoid numerical divergence or noise amplification in the region where p is close to 0 or 1 during subsequent inversion calculations, a probability pruning operation is performed on the two original cumulative probabilities, limiting the probability values ​​to [ ,1- Within the effective range of [ ], the cumulative probability after clipping the first reference voltage and the cumulative probability after clipping the second reference voltage for each monitoring sampling group are finally obtained. By combining the standardized reading, statistics and clipping process implemented by hardware circuit, the accuracy and consistency of the decision data are guaranteed, and the statistical fluctuations and numerical divergence caused by non-ideal factors are effectively suppressed, thereby improving the stability of subsequent parameter inversion and improving the accuracy of radiation dose measurement results.

[0129] Step 105: Using the inverse function of the Gaussian cumulative distribution function, each monitoring sampling group is screened based on the dual decision level and the cumulative probability after clipping of each monitoring sampling group, multiple valid monitoring sampling groups are output, and the mean threshold distribution of each valid monitoring sampling group is extracted.

[0130] The inverse function of the cumulative distribution function of the Gaussian distribution is the Probit function.

[0131] It should be noted that, based on the aforementioned dual decision levels as the judgment benchmark, and combined with the cumulative probability after clipping corresponding to each monitoring sampling group, each monitoring sampling group is screened through preset validity judgment rules. Sampling groups with cumulative probabilities deviating from the reasonable range or exhibiting abnormal fluctuations are eliminated, and multiple valid monitoring sampling groups are output. Then, based on the cumulative probability after clipping of the valid monitoring sampling groups, and combined with the reference voltage corresponding to the dual decision levels, the threshold distribution mean of each valid monitoring sampling group is calculated by inversion, providing reliable core parameters for subsequent baseline alignment and dose inversion.

[0132] Furthermore, step 105 may include the following sub-steps:

[0133] S51. The cumulative probability after clipping the first reference voltage and the cumulative probability after clipping the second reference voltage of each monitoring sampling group are transformed by the inverse function of the cumulative distribution function of Gaussian distribution, respectively, to obtain the first normal domain value and the second normal domain value of each monitoring sampling group.

[0134] S52. A monitoring sampling group in which the difference between any value in the first normal domain and the value in the second normal domain is greater than or equal to a preset difference threshold shall be regarded as a valid monitoring sampling group.

[0135] S53. Calculate the mean threshold distribution of each effective monitoring sampling group based on the dual decision level, the first normal domain value, and the second normal domain value of each effective monitoring sampling group.

[0136] It should be noted that, as Figure 3 As shown, under the target programming state, this invention uses the Gaussian approximation to model the threshold voltage distribution, that is, it is assumed that the threshold voltage at the same dose point D satisfies Vth ~ N( (D), (D)), where (D) Characterizes the location of the distribution center and serves as the main signal for the dosimeter. (D) can be used as an optional output for quality monitoring or confidence assessment. At the decision level V, the cumulative probability satisfies:

[0137] ;

[0138] in, (·) represents the cumulative distribution function of the standard normal distribution. For the same sample group, p1 is obtained after clipping under V1 and V2. k With p2 k Then, define:

[0139] z1 k = (p1 k );

[0140] z2 k = (p2 k );

[0141] Wherein, D is the radiation dose value, which is the radiation dose estimate that needs to be output through inversion in this invention. At the same dose point, the threshold voltage distribution characteristics of the storage cell will change accordingly. Vth is the threshold voltage of the storage cell in the 3D NAND flash memory chip, which is the core physical quantity constituting the threshold distribution data. Its distribution characteristics will shift or change in dispersion with the change of radiation dose D. (D) represents the mean value of the threshold voltage distribution in the storage cell when the radiation dose is D (i.e., the mean value of the threshold distribution). As the main signal of the dosimeter, it characterizes the center position of the threshold voltage distribution under this radiation dose. (D) represents the standard deviation of the threshold voltage distribution of the storage cell when the radiation dose is D. It is used to characterize the dispersion of the threshold voltage distribution under this radiation dose and can be output for quality monitoring or confidence assessment. V is any decision level (read reference voltage / read threshold), which is the reference voltage value for determining the conduction state of the storage cell. The dual decision levels V1 and V2 are two specific values ​​of this variable. p(V,D) is the probability that the threshold voltage Vth of the storage cell is less than the decision level V under the condition that the radiation dose is D and the decision level is V, i.e., the cumulative probability value. z1 k To calculate the cumulative probability p1 after clipping the first reference voltage for the k-th monitoring sampling group. k The standard normal distribution quantile, z2, is calculated using the inverse function of the cumulative distribution function of the standard normal distribution. k To calculate the cumulative probability p2 after clipping the second reference voltage for the k-th monitoring sampling group. k The standard normal distribution quantiles are calculated using the inverse function of the cumulative distribution function of the standard normal distribution.

[0142] It can be solved by the two-point equation. and further obtained ;when Too small or calculated When the value is ≤0, the page or the sampling group should be marked as invalid to avoid misjudgments caused by division by zero and abnormal amplification. Specifically, at the firmware implementation level... (·) This can be achieved through a lookup table (LUT) combined with interpolation or piecewise approximation polynomials, or, if the system architecture allows, by the host / management software, which can then transmit the calculation results back; within the sampling group , Robustness can be improved by inverting page by page and then averaging, or in scenarios with stronger cost constraints, the group output can be obtained by averaging the probabilities within the group first and then inverting.

[0143] Among them, symbols Specifically, it refers to the mean of the threshold voltage distribution (i.e., the mean of the threshold distribution) / location parameter, which is used to characterize the center position of the threshold voltage distribution of the memory cell. At the same time, this invention also covers equivalent location parameters that are substantially equivalent to this parameter, including but not limited to the threshold voltage distribution center index obtained by non-Gaussian distribution modeling, offline mapping fitting, etc. All of the above parameters belong to the distribution location-related features defined by this invention.

[0144] It is worth mentioning that, for the core logic of the parameter inversion stage based on the mean and standard deviation of the inversion threshold distribution of dual decision levels and corresponding cumulative probabilities, this invention adopts a Gaussian approximation combined with analytical inversion. That is, the storage controller assumes that the target state threshold distribution is approximately Gaussian and calls the standard normal distribution quantile function. (p) The two cropped cumulative probabilities of each monitoring sampling group are converted into corresponding normal domain values, and then the mean and standard deviation of the threshold distribution are analytically solved using a two-point closed-form inversion formula; where the standard normal quantile function (p) is only one specific calculation approach for two-point inversion and not the only implementation method. In engineering practice, it can be implemented through equivalent schemes such as look-up table (LUT) + interpolation, piecewise approximate polynomial, CORDIC algorithm, or host-side calculation and backhaul. At the same time, this invention is also compatible with alternative implementation schemes for non-Gaussian distributions. The threshold distribution can be modeled as a distribution form that is more in line with the actual form, such as skewed distribution, logistic distribution, exponentially corrected Gaussian distribution, Gaussian mixture model, etc., or the distribution form is not explicitly assumed. A look-up table and regression mapping model from probability points to equivalent position parameters are established through offline calibration. The dosimeter signal can be obtained directly from the statistical probability points online. This scheme can still maintain observation consistency when the target device distribution deviates from Gaussian and meets the engineering constraint of relying only on standard reading operations.

[0145] Secondly, regarding the operations of "calculating the cumulative probability after pruning" and "marking invalid groups and retaining valid groups" mentioned above, when the cumulative probability p approaches 0 or 1, the value of the standard normal quantile function will diverge, thereby amplifying noise and causing the threshold distribution mean to be affected. Standard deviation The estimation results are unstable. Therefore, pruning is performed on the cumulative probability p. (That is, obtaining the cumulative probability after pruning) is necessary, and at the same time, validity criteria need to be set (such as the difference between the two normal domain values). Meets preset threshold and standard deviation Pages with a weight greater than 0 or less will be removed or downgraded from invalid pages or invalid monitoring sampling groups. The details of the above probability pruning and invalidation determination can be further improved by clarifying engineering implementation details such as the "pruning threshold range, invalidation determination threshold, and invalidation handling strategy," thereby enhancing the robustness of the solution.

[0146] In addition, in the dual decision level setting, read statistics, and parameter inversion processes of the present invention, a parameter estimation implementation method combining dual decision levels with two-point analytical inversion is adopted. That is, the storage controller determines the dual decision levels V1 and V2 that satisfy V1 < V2 based on the mean and standard deviation of the threshold distribution in the baseline state, which are benchmark distribution statistics. Here, V1 can be set near the tail of the threshold distribution, and V2 can be set near the middle of the distribution. These decision levels can be determined and fixed during the initialization calculation of the baseline distribution parameters or can be adaptively fine-tuned during the online monitoring stage to maintain the corresponding cumulative probability within a preset interval, covering both fixed threshold and adaptive threshold engineering implementation methods. Subsequently, read operations are performed based on V1 and V2 to obtain the trimmed cumulative probabilities p1 and p2 corresponding to each monitoring sampling group. Then, by combining the dual decision levels and the normal domain values, the mean and optional standard deviation of the threshold distribution of each effective monitoring sampling group are inverted through a two-point closed-form formula. After multi-point fusion and linear calibration, the radiation dose is output. In addition, the present invention also supports an alternative scheme for expanding to multiple decision levels, that is, adding one or more decision levels based on V1 and V2 to form a decision level set {V i} with N ≥ 3. After reading the corresponding cumulative probabilities {p i} for each sampling group, distribution parameters are fitted by methods such as least squares and maximum likelihood estimation. The mean or equivalent position parameter obtained by fitting is used as the dosimeter signal, and the baseline alignment, fusion, and calibration processes are continued. This scheme sacrifices a moderate increase in the number of reads to improve the robustness and noise resistance of parameter estimation and can achieve calibrated dose output without full-range scanning, further enhancing the flexibility and applicability of the scheme.

[0147] In this embodiment, the inverse function of the Gaussian distribution cumulative distribution function (which can be implemented by methods such as look-up table + interpolation, piecewise approximate polynomial, or CORDIC algorithm) is used to convert the trimmed cumulative probability of the first reference voltage and the trimmed cumulative probability of the second reference voltage for each monitoring sampling group respectively, obtaining the first normal domain value and the second normal domain value corresponding to the monitoring sampling group. Then, the difference between the first normal domain value and the second normal domain value of the same monitoring sampling group is calculated. The monitoring sampling groups with a difference greater than or equal to the preset difference threshold are screened as effective monitoring sampling groups to eliminate abnormal sampling groups caused by noise, saturation, or local weak pages. Finally, based on the dual decision levels V1 and V2, combined with the first normal domain value z

[0148] Step 106: Baseline alignment and dose inversion are performed on the mean threshold distribution of each effective monitoring sampling group to output the radiation dose estimate.

[0149] It should be noted that the mean threshold distribution of each effective monitoring sampling group is calibrated and matched with the mean threshold distribution under the baseline state to complete the baseline alignment and eliminate the interference of non-radiation factors. Then, the aligned mean threshold distribution is substituted into the preset dose inversion model to complete the calculation and finally output the radiation dose estimate, providing the final result for accurate monitoring of radiation dose.

[0150] Furthermore, step 106 may include the following sub-steps:

[0151] S61. Extract the mean value of the baseline threshold distribution corresponding to each valid monitoring sampling group;

[0152] S62. Subtract the mean of the baseline threshold distribution and the mean of the threshold distribution corresponding to each effective monitoring sampling group, and output the increment of the mean of the threshold distribution for each effective monitoring sampling group.

[0153] S63. The mean increment of the threshold distribution of each effective monitoring sampling group is fused and calculated to obtain a unified dose signal;

[0154] S64. Statistically measure the number of groups corresponding to the effective monitoring sampling groups and the inter-group dispersion of the mean increment;

[0155] S65. If the number of groups corresponding to the effective monitoring sampling group is greater than or equal to the preset group number threshold and the inter-group dispersion of the mean increment corresponding to the effective monitoring sampling group is less than or equal to the preset dispersion threshold, then the unified dose signal will be used as the reliable dose signal.

[0156] S66. Using pre-stored calibration coefficients and reliable dose signals, calculate the radiation dose estimate.

[0157] It should be noted that, as Figures 4-5 As shown, due to the inherent bias at different vertical positions, direct fusion... Significant inter-group differences may occur even at zero dose; therefore, this invention can record the data of each sampling group at the baseline time of dose=0. (0), and convert subsequent measurements into increments. This eliminates fixed bias, thereby improving the stability of unified calibration. During the fusion phase, the system performs fusion on multiple sampling groups. or Δ A uniform output is obtained by using robust statistical methods such as mean, median, or clipped mean. or Δ The system also outputs quality metrics for reliability assessment during fusion, such as the number of effective sample groups (Keff), inter-group dispersion (standard deviation and range), and the range band of p1 / p2. The statistical characteristics; when the number of valid groups is too low, p1 / p2 is close to the pruning boundary, or the dispersion is abnormally increased, the system should output an insufficient confidence or saturation alarm and trigger maintenance strategies, such as reconfiguring V1 and V2, switching monitoring blocks, refreshing the monitoring page, or rebuilding the baseline.

[0158] During the calibration and dose inversion phase, the system obtains [dosage values] at several known dose points. (D) or Δ (D) Fits a linear function relationship within the target dose range. (D)≈a·D+b or Δ (D)≈a·D+b, where b represents the residual bias of the system when using the mean increment for fitting, and ideally approaches zero. The obtained calibration coefficients a and b are then compared with the applicable conditions (target state, V1 / V2, ...). The sampling group configuration and version timestamp are saved together to the parameter storage unit; during online monitoring, they are saved as follows: or Perform dose inversion and output dose estimates and alarm status via registers, logs, or host-side API.

[0159] It is worth mentioning that, in the signal fusion stage, this invention adopts a signal processing approach that combines basic spatial fusion with simple quality judgment for the step of "using spatial fusion strategies such as mean, median, or truncated mean to obtain a unified dose signal for the threshold distribution mean increment of all valid groups". Specifically, the storage controller performs fusion calculation on the threshold distribution mean increment of all valid monitoring sampling groups using spatial fusion strategies such as mean, median, or truncated mean to obtain a unified dose signal; the reliability of the dose signal is determined only by statistically analyzing the number of valid monitoring sampling groups and the inter-group dispersion of the mean increment, without introducing stronger robustness processing and time-domain noise suppression. In addition to this implementation, alternative approaches include robust fusion and time-domain filtering, such as using more robust spatial fusion methods like weighted median, Huber robust estimation, M-estimator, and RANSAC (Random Sample Consensus), or introducing explicit outlier detection and weighting rules (e.g., thresholds based on historical distribution or MAD (Median Absolute Deviation)) before fusion; simultaneously, adjustments can be made in the time dimension. Δ Or D hatTemporal filtering methods such as moving average, exponential smoothing, or Kalman filtering are introduced to suppress short-term noise and improve long-term stability. This alternative does not change the main path of "obtaining the dose signal from standard readout statistics," but only replaces the fusion and robustness implementation.

[0160] Furthermore, the operations described above, namely "extracting the mean of the baseline threshold distribution and calculating the mean increment" and "fusing the mean increment to obtain a reliable dose signal," can be defined in engineering as "baseline alignment / offset compensation." Their core function is to offset the inherent static bias (i.e., baseline deviation) of each monitoring sampling group, making the fused dose signal closer to "only reflecting the threshold voltage drift caused by radiation," thus improving the accuracy of dose estimation. In addition, the fusion operator described above, "performing fusion calculations according to a preset fusion strategy," can employ mean, median, clipped mean, weighted fusion, and other robust statistical methods. It is not limited to a specific implementation and can be flexibly selected according to actual engineering needs, further enhancing the applicability and robustness of the solution.

[0161] In this embodiment, the mean value of the baseline threshold distribution for each effective monitoring sampling group under the baseline condition of dose≈0 is first extracted from the pre-stored baseline calibration data. The mean value of the current threshold distribution of each effective monitoring sampling group is subtracted from the mean value of the corresponding baseline threshold distribution to obtain the increment of the mean value of the threshold distribution for each effective monitoring sampling group. This increment represents the degree of shift of the threshold distribution center after being affected by radiation. Then, the increments of the mean values ​​of the threshold distribution of each effective monitoring sampling group are fused and calculated, for example, by using a weighted average or median statistical method to remove the influence of extreme values ​​and obtain a uniform dose signal. Subsequently, the number of groups corresponding to the effective monitoring sampling groups is counted, and the inter-group values ​​of each mean increment are calculated. Dispersion (such as standard deviation or coefficient of variation); if the number of effective monitoring sampling groups is greater than or equal to the preset group number threshold, and the inter-group dispersion of the mean increment is less than or equal to the preset dispersion threshold, then the current unified dose signal is determined to be a reliable dose signal; finally, the pre-stored calibration coefficients (the mapping relationship between dose and mean increment obtained from calibration experiments) and the reliable dose signal are used to perform linear or nonlinear operations to calculate the radiation dose estimate. Through multi-dimensional validity verification and fusion calculation, the accuracy and confidence of the radiation dose estimate are effectively improved, the interference of abnormal sampling and non-ideal factors on the measurement results is reduced, and the problem of poor accuracy of existing radiation dose measurement methods is improved.

[0162] For comparison of technical effects, existing technologies can be referenced. In extreme scenarios such as aerospace, high-altitude environments, and the nuclear industry, electronic devices are exposed to cumulative radiation fields for extended periods, facing severe challenges in total ionizing dose (TID). For 3D NAND Flash devices, radiation energy causes numerous charge traps and interface states to form in the tunneling layer and interface region. These defects not only trap transmitted charges but also weaken the ability to control stored charges, leading to unintended shifts in the threshold voltage (Vth) of the memory cells. As the dose accumulates, this shift will exceed the limits of error correction capabilities, causing severe data readout errors and memory failures.

[0163] Based on the above, existing technology 1 proposes a threshold voltage scanning / histogram reading scheme, which involves scanning multiple reference voltages (e.g., a set of Vrefs). i By repeatedly reading the same monitoring page, a set of cumulative probability points p(Vref) is obtained. i The method uses an equivalent counting curve to reconstruct the approximate shape of the threshold distribution and further extracts parameters such as the distribution center location and broadening as radiation characteristics. This type of method provides sufficient information and can directly estimate the drift and broadening of the threshold distribution. It has good accuracy and interpretability, but it requires multiple read operations (even relying on test mode / long scan), which results in long monitoring time, high online overhead, and can easily affect storage service performance and increase firmware implementation and verification costs.

[0164] Existing technology 2 proposes a single-decision-level cumulative count / bit error statistical monitoring (single-point CDF) scheme. This scheme uses a fixed reference voltage Vref to periodically read and count the cumulative quantities such as "counts on the side below the threshold", bit error count, or logic count for the monitored page. The cumulative quantity is directly used as the output signal of the dosimeter or for fitting the dose mapping. This scheme is simple to implement, has low read overhead, and is easy to run online. However, it is essentially a single-point observation of CDF. As the threshold distribution drifts, it is easy to enter the saturation region and generate obvious nonlinearity. It is also highly sensitive to the position of Vref. At the same time, in 3D NAND, it is easily affected by vertical position differences and local weak pages, often requiring positional calibration or resulting in large inversion errors.

[0165] Existing technology three proposes a dedicated reference unit / analog circuit dosimeter solution. This solution involves adding a dedicated reference unit, sensing structure, or analog measurement circuit within the chip or system. By measuring current, voltage, or differential signals, radiation-related parameters are output, thereby obtaining dose estimates. Under specific design conditions, this solution may achieve high sensitivity and good observability, but it usually requires hardware modifications, has high design and verification costs, and is difficult to directly reuse in commercial standard memories. In addition, analog quantities are easily affected by temperature, aging, and process drift, and system-level calibration and long-term maintenance are highly complex.

[0166] Furthermore, existing technologies one, two, and three each have their own drawbacks in radiation dose monitoring applications, specifically as follows: Technology one (threshold voltage scanning / histogram reading) suffers from the disadvantage that while it can obtain relatively rich distribution information, it often requires repeated readings of the same monitoring page under multiple reference voltages, accompanied by statistical and fitting processes. This results in long monitoring cycles, high read bandwidth consumption, and can easily affect the normal business throughput of the storage system during online operation. Simultaneously, multiple reads and data processing make firmware implementation more complex, verification costs higher, and may increase energy consumption and read interference risks, thus failing to meet the engineering requirements of "low overhead and long-term online deployment." Technology two (cumulative counting / error statistics monitoring under a single decision level) suffers from the disadvantage that while this scheme has low read overhead, its output is essentially a single-point observation of the threshold distribution CDF. With radiation-induced distribution drift, it easily enters the saturation region, exhibiting significant nonlinearity, leading to sensitivity decay, limited effective dynamic range, and high sensitivity to reference voltage position. Furthermore, single-point statistics struggle to distinguish between different effects such as "overall drift" and "broadening / morphological changes," resulting in poor calibration stability, especially in 3D... In NAND flash memory, vertical position differences and factors such as local weak / bad pages can further amplify the effects of position dependence and outliers, often requiring location-specific calibration or generating significant inversion errors. The disadvantage of the third technology (dedicated reference cell / analog circuit dosimeter) is that this type of solution usually relies on dedicated device structures or analog measurement circuits, requiring chip-level hardware modifications or additional peripheral circuit support, which increases area, power consumption, and cost, and significantly raises the design and verification threshold, making it difficult to directly reuse on commercial standard device (COTS) platforms. At the same time, analog signals are sensitive to temperature, aging, and process drift, and often require more complex system-level compensation and recalibration mechanisms during long-term operation, resulting in higher maintenance complexity.

[0167] Therefore, this invention mainly addresses the following specific problems existing in current 3D NAND flash memory radiation resistance monitoring technology:

[0168] 1) Addressing the issues of nonlinearity and saturation in single-decision-level cumulative counting signals: Existing techniques, based on a single reference voltage, are equivalent to single-point sampling of the threshold distribution (CDF). When radiation causes a distribution shift, the sampled value rapidly approaches the saturation region (probability 0 or 1), resulting in limited dynamic range and nonlinear sensitivity decay. Furthermore, this method is highly sensitive to the reference voltage position, readout noise, and environmental temperature drift, leading to poor output curve stability and difficulty in supporting long-term reliable dose calibration and inversion.

[0169] 2) Addressing the time and system overhead issues caused by multi-voltage scanning / multiple reads: To correct nonlinearity, existing technologies often employ multi-voltage scanning or read-retry to fit the distribution, which introduces significant read latency and computational overhead, consuming storage controller bandwidth and reducing service throughput. Simultaneously, high-frequency reads increase energy consumption and the risk of read interference. How to obtain highly linear, calibrable dose signals at extremely low read frequencies is a key challenge for practical engineering implementation.

[0170] 3) Addressing the issue of inconsistent dose response curves caused by vertical positional differences in 3D NAND: The inherent process gradients (such as aperture differences) in the vertical stacking structure of 3D NAND result in significant variations in the initial threshold distribution, noise level, and radiative drift rate of different layers. Monitoring at a single location lacks representativeness, leading to a strong positional dependence in the dose response curves, making it difficult to universalize calibration parameters and resulting in large inversion errors. It is necessary to address how to suppress the interference of physical differences in devices on output consistency.

[0171] 4) Addressing the challenge of achieving calibrated dose output without adding dedicated analog hardware: In monitoring scenarios based on Commercially Available Standard (COTS) devices, it is impossible to introduce dedicated analog sensing hardware. Relying solely on the limited statistics read from standard digital interfaces often results in biased and drifting outputs due to insufficient information dimensions and individual device variations. A continuous, reusable calibration model independent of individual variations needs to be constructed without altering the hardware architecture.

[0172] 5) Addressing the issue of output instability caused by small sample noise and outliers: Due to the limited available sample size for online monitoring, non-ideal factors such as readout noise, random telegraph noise (RTN), and local weak pages can easily cause drastic fluctuations in statistical values. A few outliers may dominate the overall statistical results, leading to dose misjudgment or calibration curve drift. Robust statistical methods need to be introduced under small sample conditions to suppress abnormal interference and improve the reliability and stability of monitoring signals.

[0173] To address the above problems, this invention provides a radiation dose measurement method based on dual-decision level and vertical fusion, which aims to:

[0174] (1) Obtain the cumulative probabilities p1 and p2 under the condition that only two fixed decision levels are required, and invert the mean of the threshold distribution based on the Gaussian approximation. This makes the dosimeter output closer to a linear function within the target dose range, thereby improving linearity and effective dynamic range.

[0175] (2) By constructing monitoring sampling groups at different vertical locations and performing multi-point sampling fusion, the influence of vertical position differences and local anomalies on output consistency is suppressed, resulting in a dose output signal that is easier to uniformly calibrate and can be stably inverted (e.g. or Δ ).

[0176] (3) Provide engineering-applicable parameter selection and robustness mechanisms, including V1 and V2 selection principles and probability pruning. The solution includes handling of invalid sampling groups / outliers, enabling long-term stable operation in the controller firmware and facilitating maintenance.

[0177] Specifically, such as Figure 6 As shown, this invention selects multiple monitoring sampling groups (composed of several physical pages) covering different vertical positions within a reserved monitoring area. The cumulative probabilities p1 and p2 are read and statistically analyzed under two fixed decision levels V1 and V2, and the mean threshold distribution is obtained through two-point inversion under the premise of Gaussian approximation. (Standard deviation can be obtained) This was used as a dosimeter signal. Subsequently, multiple sampling groups were... (or Δ after baseline alignment) Perform multi-point fusion to obtain a unified output. (or Δ) The calibration and dose inversion are completed through a linear function, and the dose estimate and confidence / alarm information are output.

[0178] Furthermore, this embodiment of the invention provides a radiation monitoring system for storage devices. This system mainly consists of a storage controller and a 3D NAND flash memory chip, and can be widely used in solid-state drives, embedded storage, or other storage devices containing NAND media. In terms of system architecture, the storage controller, as the core processing unit, establishes a communication connection with the flash memory chip through a standard NAND interface. It is responsible for both routine data read / write management and executing the radiation monitoring process of this invention, ultimately providing a dose query interface and status alarm information to the host or upper-level management software.

[0179] From a hardware composition and data interaction perspective, the 3D NAND flash memory chip is used to carry the preset monitoring area and supports responding to the controller's probe commands by adjusting the read reference voltage. The storage controller integrates a microprocessor, NAND interface control unit, and cache unit, possessing the ability to perform complex logical operations and data transfer. Furthermore, the system maintains a parameter storage unit in the non-volatile memory area to store critical configuration information, including reference voltage configuration, probability pruning parameters, calibration coefficients, and version timestamps, ensuring that the system maintains computational consistency and accuracy even after power failure and restart.

[0180] The core functional logic of the controller can be implemented either through firmware code or in a collaborative manner between firmware and dedicated hardware. In terms of functional partitioning, the controller mainly includes several key processing links such as decision level configuration, data counting and statistics, parameter inversion calculation, multi-point data fusion, and calibration inversion. Its working flow is roughly as follows: First, the controller controls the flash memory to read the monitoring page data at different decision levels, and then performs bit-level statistics on the read results to obtain the cumulative probability. Next, through the internal algorithm module, the statistical data is subjected to normal domain transformation and multi-point fusion processing to extract the characteristic values reflecting the threshold voltage drift. Finally, the current radiation dose estimate value is deduced by combining the pre-stored calibration coefficients, and the corresponding quality indicators and alarm status are output based on this value.

[0181] In summary, the core method of the present invention is double decision level and double point inversion , that is, at two fixed decision levels V1 < V2, the cumulative probabilities p1 = P(Vth < V1) and p2 = P(Vth < V2) are read for the same monitoring sampling group, and the probabilities are mapped to the normal domain z = (p) to establish a double point equation and solve for the mean value of the threshold distribution (and optionally the standard deviation ), and or its increment Δ is used as the main output signal of the radiation dosimeter, so as to obtain a dose characterization quantity closer to a linear function under the condition of low read overhead; the core strategies include probability clipping and inversion effectiveness determination. Perform clipping on p1 and p2 to suppress the numerical divergence of the quantile function when p → 0 / 1, and perform effectiveness determination and invalid marking on conditions such as [[ID=二十一]], > 0, whether p1 / p2 approaches the clipping boundary, etc., to ensure the availability of the inversion process when noise, saturation, and outlier pages exist, and provide a credibility basis for online output, as well as the construction of vertical multi-point sampling groups and intra-group statistics. Select multiple monitoring sampling groups P composed of several physical pages at different vertical positions and / or different decks in 3D NAND k , and stable p1 is obtained through intra-group page-by-page averaging or robust statistics k , p2 k and the inversely obtained therefrom, reducing the dominant effect of small sample noise and local weak pages on the results, and enhancing the representativeness and consistency across positions; the core algorithm is multi-point fusion and dose output with baseline alignment. The output of multiple sampling groups { } or {Δ } adopts fusion strategies such as mean / median / truncated mean / weighting to obtain a unified dosimeter signal or Δ And a baseline can be recorded by setting the dose=0. (0) Implementation Alignment is used to weaken the fixed bias. Based on this, the coefficients (a, b) are obtained by linear function calibration of the fused signal. Online calibration is then performed. or The system performs dose inversion and outputs quality indicators such as inter-group dispersion and number of effective groups, as well as alarm conditions such as insufficient saturation / reliability, to ensure the maintainability and traceability of long-term online monitoring.

[0182] Compared to existing technologies, the multi-decision level scanning / histogram reading scheme reconstructs the threshold distribution by repeatedly reading the same monitoring page under multiple reference voltages. This provides ample information and strong interpretability, and is generally considered a superior dose monitoring approach. However, it inevitably leads to multiple read-retry operations, a longer monitoring cycle, and significant read bandwidth consumption. Furthermore, the scanning and fitting process increases firmware implementation complexity and long-term verification costs, and may introduce additional energy consumption and read interference risks during online deployment. This invention addresses the engineering constraints of "low-overhead online monitoring" by compressing the observation points into two fixed decision levels, V1 and V2. Only two reads are needed in each sampling group to obtain p1 and p2, and the results are obtained using the Gaussian approximation through z= The two-point equation of (p) directly inverts the distribution location parameters. (Optional) This allows for the extraction of "distributed displacement information that requires multiple scans" with fewer reads, significantly reducing monitoring overhead and improving deployability from a causal chain perspective.

[0183] Furthermore, compared to the single-decision level cumulative counting / cumulative probability scheme, the output of this invention is no longer a single-point sampled value of the CDF, but rather a distribution mean obtained by the joint constraints of two points. Because the total ionizing dose effect often manifests as an overall shift in the threshold distribution, It more directly represents the change in the "distribution center position," thus making it easier to approximate a linear function relationship with the dose within the target dose range. This alleviates the nonlinearity and limited dynamic range problems caused by the sensitivity of the operating point selection and the ease with which the probability enters the saturation region in single-point schemes. Furthermore, by probabilistically pruning p1 and p2... And on , The present invention uses conditions such as >0 to determine validity. It can suppress the propagation of numerical instability when the probability is close to 0 / 1, noise is amplified, or readout is abnormal, so that the output has stronger robustness and availability.

[0184] Furthermore, this invention addresses the vertical process gradient problem in 3D NAND by constructing multiple monitoring and sampling groups covering different vertical positions and / or different decks and outputting the results. Combined with baseline alignment A unified signal is obtained by fusion with multiple points. or Δ This mechanism statistically weakens the "position-dependent fixed bias" and "occasional outlier / weak page perturbations," thereby significantly reducing the impact of cross-layer differences on dose output consistency. This makes unified calibration and dose inversion more stable and reduces maintenance costs. Since the above process is entirely based on standard NAND read operations and controller digital calculations, it can output dose estimates and quality indicators / alarm conditions without additional analog sensing hardware. Therefore, it has comprehensive advantages such as low cost, easy integration, upgradeability, and traceability.

[0185] In this embodiment of the invention, a radiation dose measurement method based on dual-decision level and vertical fusion is provided. A dedicated monitoring region is configured in a 3D NAND flash memory chip, and idle physical blocks within the dedicated monitoring region are integrated to obtain an independent monitoring region. Programming and multi-reference voltage scanning are performed on multiple main monitoring blocks within the independent monitoring region, outputting dual-decision levels. Based on the physical page mapping relationship of each main monitoring block within the independent monitoring region, multiple monitoring sampling groups are formed according to vertical positions, and the total number of bits in each monitoring sampling group is calculated. According to the dual-decision levels and the total number of bits in each monitoring sampling group, dual-reference voltage readings are performed on each monitoring sampling group, outputting the pruned cumulative probability of each monitoring sampling group. The inverse function of the Gaussian cumulative distribution function is used to filter each monitoring sampling group based on the dual-decision levels and the pruned cumulative probability of each monitoring sampling group, outputting multiple effective monitoring sampling groups, and extracting the mean threshold distribution of each effective monitoring sampling group. Baseline alignment and dose inversion are performed on the mean threshold distribution of each effective monitoring sampling group, outputting a radiation dose estimate. Based on the above method… This invention, through the configuration and integration of independent monitoring areas, ensures the specificity and stability of monitoring samples, laying the foundation for improving measurement accuracy. The dual-decision-level output provides a unified and standardized basis for subsequent sampling and effective group screening, avoiding measurement deviations caused by inconsistent judgment criteria. The design of multiple monitoring sampling groups based on vertical positions enables multi-point vertical sampling, effectively expanding the effective sample size for online monitoring. Simultaneously, it covers physical pages at different vertical positions, specifically suppressing the impact of vertical position differences on the consistency of measurement output. Combining dual-decision-level analysis with the total number of bits in each sampling group to perform dual-reference voltage readings and calculate the cumulative probability after clipping suppresses statistical fluctuations caused by non-ideal factors. Screening of effective monitoring sampling groups directly eliminates outliers, preventing them from dominating the overall statistical results. Baseline alignment eliminates the interference of inherent sample bias on measurement results, ultimately significantly improving the accuracy of radiation dose measurement.

[0186] Please see Figure 7 , Figure 7 This is a structural block diagram of a radiation dose measurement system based on dual decision levels and vertical fusion, provided in Embodiment 2 of the present invention.

[0187] This invention provides a radiation dose measurement system based on dual-decision level and vertical fusion, comprising:

[0188] Integration module 701 is used to configure a dedicated monitoring area in a 3D NAND flash memory chip and integrate the free physical blocks in the dedicated monitoring area to obtain an independent monitoring area;

[0189] The write and scan module 702 is used to perform programming write and multi-reference voltage scan on multiple master monitoring blocks in an independent monitoring area, and output dual decision levels;

[0190] The assembly and statistics module 703 is used to assemble multiple monitoring sampling groups according to their vertical positions based on the physical page mapping relationship of each main monitoring block in an independent monitoring area, and to count the total number of bits of each monitoring sampling group.

[0191] The reading module 704 is used to perform dual reference voltage reading on each monitoring sampling group according to the dual decision level and the total number of bits of each monitoring sampling group, and output the cumulative probability after clipping of each monitoring sampling group;

[0192] The filtering and extraction module 705 is used to filter each monitoring sampling group based on the inverse function of the Gaussian cumulative distribution function according to the dual decision level and the cumulative probability after clipping of each monitoring sampling group, output multiple valid monitoring sampling groups, and extract the threshold distribution mean of each valid monitoring sampling group.

[0193] Output module 706 is used to perform baseline alignment and dose inversion on the mean threshold distribution of each effective monitoring sampling group and output radiation dose estimates.

[0194] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of the system and modules described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0195] This invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program; when the computer program is executed by the processor, the processor performs the steps of the radiation dose measurement method based on dual decision levels and vertical fusion as described in the above embodiments.

[0196] This invention also provides a computer-readable storage medium storing a computer program / instructions thereon, which, when executed by a processor, implements the steps of the radiation dose measurement method based on dual-decision level and vertical fusion as described in the above embodiments.

[0197] This invention also provides a computer program product, including a computer program stored on a non-transitory computer-readable storage medium, the computer program including program instructions, wherein when the program instructions are executed by a computer, the computer performs the steps of the radiation dose measurement method based on dual decision levels and vertical fusion as described in the above embodiments.

[0198] In the several embodiments provided in this application, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0199] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0200] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0201] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0202] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A radiation dose measurement method based on dual decision level and vertical fusion, characterized in that, include: A dedicated monitoring area is configured in the 3D NAND flash memory chip, and the idle physical blocks in the dedicated monitoring area are integrated to obtain an independent monitoring area; Programming and multi-reference voltage scanning are performed on multiple main monitoring blocks within the independent monitoring area, and dual decision levels are output. Based on the physical page mapping relationship of each main monitoring block in the independent monitoring area, multiple monitoring sampling groups are formed according to the vertical position, and the total number of bits of each monitoring sampling group is counted. Based on the dual decision levels and the total number of bits in each monitoring sampling group, dual reference voltage readings are performed on each monitoring sampling group, and the cumulative probability after clipping of each monitoring sampling group is output. The inverse function of the cumulative distribution function of Gaussian distribution is used to filter each monitoring sampling group based on the dual decision level and the cumulative probability after pruning of each monitoring sampling group, outputting multiple effective monitoring sampling groups, and extracting the mean threshold distribution of each effective monitoring sampling group; Baseline alignment and dose inversion are performed on the mean threshold distribution of each effective monitoring sampling group to output radiation dose estimates.

2. The radiation dose measurement method based on dual-decision level and vertical fusion according to claim 1, characterized in that, The process of programming and writing to multiple master monitoring blocks within the independent monitoring area and scanning multiple reference voltages to output dual decision levels includes: The physical pages of each main monitoring block are written into a fixed pattern according to the preset target programming state to obtain multiple programmed main monitoring blocks; Read the threshold distribution data of the physical pages of each programmed master monitoring block, and calculate the mean and standard deviation of the threshold distribution under the baseline state based on the threshold distribution data of the physical pages of each programmed master monitoring block. The dual-decision level is calculated based on the mean and standard deviation of the threshold distribution under the baseline state.

3. The radiation dose measurement method based on dual-decision level and vertical fusion according to claim 1, characterized in that, Based on the physical page mapping relationship of each main monitoring block within the independent monitoring area, multiple monitoring sampling groups are formed according to their vertical positions, and the total number of bits of each monitoring sampling group is calculated, including: Based on the physical page mapping relationship of each main monitoring block, the vertical position coordinates of multiple physical pages of the corresponding main monitoring block are extracted and divided into multiple continuous vertical intervals. In each of the aforementioned continuous vertical intervals, a preset number of physical pages are selected to form corresponding monitoring sampling groups; The number of bits per page in each monitoring sampling group is obtained by querying the physical page mapping relationship and summing them up to get the total number of bits in each monitoring sampling group.

4. The radiation dose measurement method based on dual-decision level and vertical fusion according to claim 1, characterized in that, The cumulative probability after clipping includes the cumulative probability after clipping the first reference voltage and the cumulative probability after clipping the second reference voltage; the step of performing dual reference voltage readings on each monitoring sampling group based on the dual decision level and the total number of bits in each monitoring sampling group, and outputting the cumulative probability after clipping of each monitoring sampling group, includes: A first reference voltage and a second reference voltage are obtained, and the first reference voltage and the second reference voltage are applied to the storage unit of each physical page in each monitoring sampling group, respectively, and the first binary decision data and the second binary decision data of each physical page in each monitoring sampling group are output. Within each monitoring sampling group, the number of on-cell units or non-on-cell units is counted in the first binary decision data and the second binary decision data of each physical page to obtain the first reference voltage cumulative bit count and the second reference voltage cumulative bit count for each monitoring sampling group. Based on the total number of bits, the cumulative number of bits of the first reference voltage, and the cumulative number of bits of the second reference voltage for each monitoring sampling group, the cumulative probability after clipping the first reference voltage and the cumulative probability after clipping the second reference voltage for each monitoring sampling group are calculated.

5. The radiation dose measurement method based on dual-decision level and vertical fusion according to claim 4, characterized in that, The method employs the inverse function of the Gaussian cumulative distribution function to filter each monitoring sampling group based on the dual decision levels and the cumulative probability after pruning, outputting multiple valid monitoring sampling groups, and extracting the mean threshold distribution of each valid monitoring sampling group, including: The cumulative probability after clipping the first reference voltage and the cumulative probability after clipping the second reference voltage of each monitoring sampling group are transformed by the inverse function of the cumulative distribution function of the Gaussian distribution, respectively, to obtain the first normal domain value and the second normal domain value of each monitoring sampling group. A monitoring sampling group in which the difference between any value in the first normal domain and any value in the second normal domain is greater than or equal to a preset difference threshold is considered a valid monitoring sampling group. The threshold distribution mean of each effective monitoring sampling group is calculated based on the dual decision levels, the first normal domain value, and the second normal domain value of each effective monitoring sampling group.

6. The radiation dose measurement method based on dual-decision level and vertical fusion according to claim 1, characterized in that, The process of baseline alignment and dose inversion of the mean threshold distribution of each of the effective monitoring sampling groups, and outputting a radiation dose estimate, includes: Extract the mean of the baseline threshold distribution corresponding to each of the effective monitoring sampling groups; Subtract the mean of the baseline threshold distribution and the mean of the threshold distribution corresponding to each effective monitoring sampling group to output the increment of the mean of the threshold distribution for each effective monitoring sampling group; The mean increment of the threshold distribution of each effective monitoring sampling group is fused and calculated to obtain a unified dose signal; The number of groups corresponding to the effective monitoring sampling groups and the inter-group dispersion of the mean increment are statistically analyzed. If the number of groups corresponding to the effective monitoring sampling group is greater than or equal to a preset group number threshold and the inter-group dispersion of the mean increment corresponding to the effective monitoring sampling group is less than or equal to a preset dispersion threshold, then the unified dose signal is taken as a reliable dose signal. The radiation dose estimate is calculated using the pre-stored calibration coefficients and the reliable dose signal.

7. A radiation dose measurement system based on dual-decision level and vertical fusion, characterized in that, include: An integration module is used to configure a dedicated monitoring area in a 3D NAND flash memory chip and integrate the idle physical blocks in the dedicated monitoring area to obtain an independent monitoring area; The write and scan module is used to perform programming write and multi-reference voltage scan on multiple main monitoring blocks within the independent monitoring area, and output dual decision levels; The module for building and counting is used to build multiple monitoring sampling groups according to their vertical positions based on the physical page mapping relationship of each main monitoring block in the independent monitoring area, and to count the total number of bits of each monitoring sampling group. The reading module is used to perform dual reference voltage reading on each of the monitoring sampling groups according to the dual decision level and the total number of bits of each monitoring sampling group, and output the cumulative probability of each monitoring sampling group after clipping; The filtering and extraction module is used to filter each monitoring sampling group based on the dual decision level and the cumulative probability after pruning of each monitoring sampling group using the inverse function of the cumulative distribution function of Gaussian distribution, output multiple effective monitoring sampling groups, and extract the threshold distribution mean of each effective monitoring sampling group; The output module is used to perform baseline alignment and dose inversion on the mean threshold distribution of each of the effective monitoring sampling groups, and output the radiation dose estimate.

8. An electronic device, characterized in that, The device includes a memory and a processor, wherein the memory stores a computer program that, when executed by the processor, causes the processor to perform the steps of the radiation dose measurement method based on dual-decision level and vertical fusion as described in any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed, it implements the radiation dose measurement method based on dual-decision level and vertical fusion as described in any one of claims 1-6.

10. A computer program product, characterized in that, The computer program product includes a computer program stored on a non-transitory computer-readable storage medium, the computer program including program instructions, wherein when the program instructions are executed by a computer, the computer performs the steps of the radiation dose measurement method based on dual-decision level and vertical fusion as described in any one of claims 1-6.

Images