Graded screening method and system based on DRAM (Dynamic Random Access Memory) chip

By applying energy pulse sequences and analyzing dynamic responses to DRAM chips, combined with Fourier transform and similarity matrix calculation, multidimensional feature analysis and adaptive classification of DRAM chips are achieved. This solves the problem of single testing dimensions in existing technologies and improves the accuracy and applicability of chip screening.

CN121456702APending Publication Date: 2026-02-03HUIJU ELECTRONICS (DONGGUAN) IND CO LTD
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
CN202511533885.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-10-25
Publication Date
2026-02-03

AI Technical Summary

Technical Problem

Existing DRAM chip screening methods rely on a single testing dimension, leading to high-performance chips being mistakenly identified as unqualified or low-performance chips entering the market, resulting in resource waste and system instability.

Method used

A DRAM chip-based hierarchical screening method is adopted. By applying an energy pulse sequence to excite the deep response characteristics of the memory cell, energy absorption fluctuations are monitored, dynamic response curves are recorded and Fourier transforms are performed to derive a quantitative dataset of energy feature spectra. The similarity and Euclidean distance of feature vectors are calculated to generate a similarity matrix. Variable load tests are performed to simulate the actual working environment, and finally, hierarchical outputs with multiple performance levels are generated.

Benefits of technology

It expands the testing dimensions from single parameter detection to multi-dimensional feature analysis, enables adaptive classification, improves the accuracy and applicability of DRAM chip quality assessment, and forms a fully closed-loop hierarchical screening system, which significantly improves the accuracy and consistency of chip screening.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121456702A_ABST
    Figure CN121456702A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of dynamic random access memories, and particularly provides a hierarchical screening method and system based on a DRAM chip, and the method comprises the steps: monitoring the energy absorption fluctuation of each storage unit, and recording a dynamic response curve; after the dynamic response curve is subjected to Fourier transform processing, deriving a quantitative data set of the energy characteristic spectrum; performing similarity calculation on energy distribution modes in the energy characteristic spectrum, and generating a similarity matrix by calculating Euclidean distance and angle deviation between characteristic vectors; outputting a feature clustering group; a variable load condition is applied to each feature clustering group, and an actual working environment is simulated; monitoring a stability index of the storage unit; after a verification result is subjected to weighted scoring, final graded output is generated; the memory cells are divided into a plurality of performance levels. The system comprises a data set derivation module, a clustering group output module and a weighted scoring module. According to the method, the accuracy and the applicability of DRAM chip quality evaluation are remarkably improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of dynamic random access memory technology, and in particular to a hierarchical screening method and system based on DRAM chips. Background Technology

[0002] With the rapid development of information technology, DRAM (Dynamic Random Access Memory) chips, as core storage components in computers and electronic devices, directly affect the stability of the overall system due to their performance and reliability. However, during the manufacturing process, DRAM chips can exhibit significant performance differences due to issues such as process fluctuations, material defects, or environmental factors. Traditional DRAM chip screening methods typically employ a one-size-fits-all testing standard, subjecting all chips to the same testing process. While simple, this method has significant limitations: on the one hand, high-performance chips may be mistakenly deemed unqualified due to overly stringent testing standards, resulting in wasted resources; on the other hand, low-performance chips may enter the market due to insufficient testing, leading to instability in subsequent system operation.

[0003] Prior art 1, Chinese Patent Application No. 202110040675.9, discloses a chip testing method and system, including a first testing machine performing a first functional test on an unpackaged first chip and simultaneously generating a first quality grading code for the first chip; screening the first chips according to the first quality grading code and packaging the retained first chips to generate a second chip; and a second testing machine performing a second functional test on the second chip and simultaneously generating a second quality grading code for the second chip. Although the first quality grading of the chip is performed based on the test results before packaging, and a first quality grading code is generated, preserving the memory of the pre-packaging test, and then the chip is packaged and tested according to the corresponding level based on the chip category read from the first quality grading code, the memory of the test can be used to avoid performing higher-level tests on lower-level chips, thereby avoiding invalid testing processes and effectively improving testing efficiency; however, relying solely on functional test results for grading lacks in-depth analysis of the chip's physical characteristics, resulting in a single testing dimension. Prior art two, Chinese patent application number 202411326011.9, discloses a method and system for packaging memory chips. The method includes: first, obtaining the chip to be packaged, performing appearance inspection to screen out chips that pass appearance inspection, then performing electrical performance testing on it, classifying it according to storage type to obtain classified chip groups, searching for packaging materials, constructing a basic packaging structure using a substrate, installing the chip, bonding and fixing it, and sealing it to obtain a preliminary package, monitoring the packaging status after drying, detecting potential defects and marking them to obtain marked packages, repairing them and testing performance indicators, performing graded processing to obtain graded packages, querying the graded process and generating a packaging report for the chip to be packaged, realizing a complete process from chip detection to packaging, repair and report generation. Although this can improve the packaging level of memory chips, the use of traditional electrical performance testing and appearance inspection cannot identify potential stability defects.

[0004] Prior art three, Chinese patent application number 201810503034.0, discloses an online testing and classification method and system for flash memory chips with different packages. The method includes connecting the test chip to the testing system, selecting the blocks to be tested on a block-by-block basis, selecting appropriate test patterns based on the different packages of the flash memory chips to test the physical quantity information of the flash memory chips, and saving the test data in real time; using a lifetime prediction algorithm to predict the remaining lifetime of the flash memory chips based on the physical quantity information; processing the physical quantity information of the flash memory chips to obtain advanced data; and judging and screening flash memory chips individually or in combination based on the physical quantity information, remaining lifetime, and advanced data to classify the flash memory chips. Although using corresponding test patterns for different types of flash memory chips enables testing of flash memory chips with different packages and models; and classifying flash memory chips by grading the storage blocks to obtain a more accurate and realistic classification method, static testing based on physical quantity information cannot reflect performance under dynamic working conditions.

[0005] Current technologies 1, 2, and 3 suffer from problems such as limited testing dimensions, rigid classification standards, and poor environmental adaptability. Therefore, this invention provides a hierarchical screening method and system based on DRAM chips. Summary of the Invention

[0006] To achieve the above objectives, the present invention adopts the following technical solution: One aspect of the present invention provides a hierarchical screening method based on DRAM chips, comprising the following steps: An energy pulse sequence generated by a waveform generator is applied to the memory cell set. The energy pulse sequence is designed to excite the deep response characteristics of the memory cells. The energy absorption fluctuations of each memory cell are monitored, and the dynamic response curve is recorded. After the dynamic response curve is processed by Fourier transform, a quantitative dataset of energy characteristic spectrum is derived. The similarity calculation is performed on the energy distribution patterns in the energy feature spectrum. By calculating the Euclidean distance and angular deviation between feature vectors, a similarity matrix is ​​generated. Based on the pattern density in the similarity matrix, the number and boundaries of clusters are automatically determined, and feature cluster groups are output. These are a set of labeled storage units, with each group representing a type of energy feature. Variable load conditions are applied to each feature cluster to simulate the actual working environment; the variable load conditions are generated by a stress generator, and the load parameters are adjusted according to the energy characteristic type; the stability index of the storage unit is monitored; the verification results are weighted and scored to generate the final graded output; the storage unit is divided into multiple performance levels.

[0007] In one alternative implementation, the quantization dataset contains the energy distribution pattern of each storage cell, including peak frequency, harmonic components, and attenuation coefficient characteristics.

[0008] In one alternative implementation, the process of deriving a quantized dataset of energy feature spectra includes the following steps: Each dynamic response curve is divided into multiple overlapping time windows, and the length of each time window is adaptively determined based on the main oscillation period of the pulse sequence. A local frequency domain reconstruction operation is applied to each independent time window to convert each complete dynamic response curve into a set of multiple local frequency domain segments, which is called a multi-scale response component. The information entropy contained in each local frequency domain segment of the multi-scale response component is analyzed, and a weighting coefficient is calculated. The segment with a higher weight indicates that it contains more significant response information generated by impulse excitation rather than random noise. Based on the weighting coefficient, the frequency domain information of multiple segments is weighted and synthesized into a single, highly representative synthetic spectrum. After feature mining, the synthesized spectrum yields three core quantization features: dominant oscillation mode, harmonic consistency coefficient, and energy decay trajectory. These three features together form a multidimensional vector, representing the energy feature spectrum of a memory cell. The set of vectors from all memory cells in the chip forms the quantization dataset.

[0009] In one alternative implementation, the process of calculating a weighting coefficient includes the following steps: An analysis called spectral distribution dispersion is performed on each local frequency domain segment in the multi-scale response components to obtain the distribution pattern of energy values ​​of all frequency points within a local frequency domain segment, thus obtaining an initial significance evaluation value; an initial response significance index is generated for each local frequency domain segment. The calibration of global consistency constraints is performed to obtain the global statistical characteristics of the set of significance indices of responses of all local frequency domain segments. After calibration, a set of provisional weights is obtained. An energy conservation constraint is applied to the provisional weights to ensure that the sum of all weight coefficients conforms to a preset specification, and a set of normalized weight coefficients for weighted synthesis is output.

[0010] In one alternative implementation, the process of outputting feature clusters includes the following steps: For any two feature vectors in storage units, two sets of independent calculations are performed in parallel. The first set of calculations calculates the absolute difference in the values ​​of each dimension of the vectors, generating an amplitude difference factor. The second set of calculations calculates the directional difference of the vectors in the multidimensional space, generating a direction deviation factor. The two factors are input into an adaptive fusion program, which outputs the inter-unit difference degree. All pairs of units are traversed, and the inter-unit difference degrees form a difference degree matrix. A global coupling transformation is performed on the dissimilarity matrix to transform the dissimilarity relationship between each cell and all other cells into a new index representing the density of its local data environment, which is the cell kernel density; the ratio of the kernel density of each storage cell to the kernel density of all its nearest dissimilarity neighbor storage cells is calculated to obtain an autonomy index. Each non-seed point unit is assigned to a potential energy field based on its difference relationship with various sub-points, forming a temporary affiliation. The boundary regions of the temporary affiliations are examined to find saddle points where the difference changes, thus defining the final cluster boundary and outputting stable and labeled feature clusters.

[0011] In one optional implementation, the process of outputting the difference between units includes the following steps: Based on the distribution width of the values ​​of all storage units in the quantized dataset across each dimension, a feature stability weight is calculated for each dimension; the absolute difference between two feature vectors across each dimension is obtained, multiplied by the dimension weight, and then weighted and summed to output a standardized scalar value, called the magnitude difference factor; Normalize the two eigenvectors by dividing them by the L2 norm to obtain two unit vectors. Obtain the cosine value of the angle difference between the vectors and a reference hyperplane defined by the principal direction feature of the entire quantized dataset. The principal direction feature is obtained through statistical analysis of the entire quantized dataset. Map the cosine value to amplify the difference in small angle deviations. The output is the direction deviation factor. The amplitude difference factor and the direction deviation factor are simultaneously input into an adaptive fusion program, which includes a dynamic balance parameter that adjusts the contribution ratio of the amplitude difference factor and the direction deviation factor, and outputs a scalar value of the inter-unit difference degree.

[0012] In one optional implementation, the process of simultaneously inputting the amplitude difference factor and the direction deviation factor into an adaptive fusion program includes the following steps: The sum of the L2 norms of the two eigenvectors is obtained to get a composite amplitude intensity; the composite amplitude intensity is then subjected to nonlinear mapping processing; and finally, the composite amplitude intensity is converted into a fusion modulation coefficient between zero and one. Multiplying the amplitude difference factor by the fusion modulation coefficient yields a modulated amplitude contribution; simultaneously, multiplying the directional deviation factor by the result of subtracting the fusion modulation coefficient yields a modulated directional contribution. The modulated amplitude contribution and modulated directional contribution are algebraically added, and then processed by a dynamic range compressor. Based on the global distribution range of the modulated amplitude contribution and modulated directional contribution calculated for all units in the current batch, the result of this addition is scaled to fall into a standardized, unitless numerical range, and the inter-unit difference is output.

[0013] In one optional implementation, the process of algebraically adding the modulated amplitude contribution and the modulated directional contribution includes the following steps: The modulated amplitude contribution and the modulated directional contribution are preliminarily algebraically added to obtain a transient fusion value; Based on a global statistical calibration mechanism, in the current screening batch, the set of all transient fusion values ​​obtained by adding the amplitude contribution and directional contribution of all storage units to the generated infinite number of modulations in real time is calculated; the median and interquartile range of all transient fusion values ​​are extracted to form a global distribution statistic. The transient fusion values ​​are mapped to a standardized, unitless numerical scale, and the inter-unit variability is output, which represents the combined variability after weighting and calibration in two different dimensions: amplitude and direction.

[0014] In one alternative implementation, the process of extracting the median and interquartile range of all transient fusion values ​​includes the following steps: The entire set of transient fused values ​​is processed into an ordered serialization. Through a logic circuit structure, multiple sets of comparison and exchange operations are performed synchronously to locate the key quantiles in the set of transient fused values. A set of flag signals is output to identify the values ​​at several specific positions. After processing, the key value located at the 50th percentile position is obtained, which is the median value of the sequence. The position at which the data in the set of values ​​is located 50 percentile below the median of the sequence is called the lower quartile; at the same time, the position at which the data is located 50 percentile above the median of the sequence is called the upper quartile; the arithmetic difference between the upper quartile and the lower quartile is called the half-range dispersion. The median and half-range dispersion are combined to form a binary statistic, which is the final global distribution statistic. The median provides a central reference for the data distribution, and the half-range dispersion provides the scale of variation of the data around this central reference.

[0015] In another aspect, the present invention provides a hierarchical screening system based on DRAM chips, wherein implementing the hierarchical screening method based on DRAM chips includes: The dataset derivation module is responsible for applying an energy pulse sequence generated by a waveform generator to the set of storage cells. The energy pulse sequence is designed to excite the deep response characteristics of the storage cells. It monitors the energy absorption fluctuations of each storage cell and records the dynamic response curve. After the dynamic response curve is processed by Fourier transform, a quantitative dataset of energy feature spectrum is derived. The clustering group output module is responsible for calculating the similarity of energy distribution patterns in the energy feature spectrum. It generates a similarity matrix by calculating the Euclidean distance and angular deviation between feature vectors. Based on the pattern density in the similarity matrix, it automatically determines the number and boundaries of clusters and outputs feature clusters, which are a set of labeled storage units, each group representing a type of energy feature. The weighted scoring module is responsible for applying variable load conditions to each feature cluster group to simulate the actual working environment. The variable load conditions are generated by a stress generator, and the load parameters are adjusted according to the energy characteristic type. The stability index of the storage unit is monitored. After the verification results are weighted and scored, the final graded output is generated. The storage unit is divided into multiple performance levels, such as high stability, medium stability and low stability.

[0016] This invention breaks through the limitations of traditional threshold screening, realizing the expansion of testing dimensions from single parameter detection to multi-dimensional feature analysis; the upgrade of intelligent decision-making from fixed standards to adaptive classification; the improvement of performance prediction capabilities from laboratory conditions to working scenarios; and finally forming a hierarchical screening system with a complete closed loop of feature perception, pattern recognition, and environmental verification, which significantly improves the accuracy and applicability of DRAM chip quality assessment. Attached Figure Description

[0017] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1This is a flowchart of the hierarchical screening method based on DRAM chips provided in Embodiment 1 of the present invention; Figure 2 This is a schematic diagram of the hierarchical screening method based on DRAM chips provided in Embodiment 1 of the present invention; Figure 3 This is a process diagram of generating a quantized dataset of energy feature spectra provided in Embodiment 2 of the present invention; Figure 4 This is a process diagram of the output feature clustering group provided in Embodiment 4 of the present invention; Figure 5 This is a process diagram of generating the final hierarchical output provided in Embodiment 9 of the present invention; Figure 6 This is a block diagram of the DRAM chip-based hierarchical screening system provided in Embodiment 10 of the present invention.

[0018] Figure 7 A block diagram of the electronic device provided by the present invention; Figure 8 A block diagram of a computer-readable storage medium provided for this invention. Detailed Implementation

[0019] The technical solutions of the present invention will now be described with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0020] Hereinafter, the terms "first," "second," etc., are used for descriptive convenience only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined with "first," "second," etc., may explicitly or implicitly include one or more of that feature. In the description of this invention, unless otherwise stated, "a plurality of" means two or more.

[0021] In this invention, unless otherwise explicitly specified and limited, the term "connection" should be interpreted broadly. For example, "connection" can be a fixed mechanical connection, a detachable mechanical connection, or an integral part; or, "connection" can be a direct connection or an indirect connection through an intermediate medium. Furthermore, unless otherwise explicitly specified and limited, the term "coupling" should be interpreted broadly. For example, "coupling" can be a direct electrical connection, such as physical contact and electrical conduction between two components; it can also be understood as an electrical connection between different components in a circuit structure through physical lines capable of transmitting electrical signals, such as copper foil or wires on a printed circuit board (PCB), to transmit electrical signals; or, "coupling" can be an indirect electrical connection between two components through an intermediate medium; or, "coupling" can be an electrical connection between two components in a non-contact manner, such as an electrical connection between two components using capacitive coupling to transmit electrical signals.

[0022] In this embodiment of the invention, directional terms such as "up," "down," "left," and "right" may be defined relative to the orientation of the components shown in the accompanying drawings. It should be understood that these directional terms can be relative concepts, used for relative description and clarification, and can change accordingly depending on the orientation of the components in the accompanying drawings.

[0023] Example 1: like Figure 1 As shown, this embodiment of the invention provides a hierarchical screening method based on DRAM chips, comprising the following steps: Step S100: Apply an energy pulse sequence generated by a waveform generator to the set of memory cells. The energy pulse sequence is designed to excite the deep response characteristics of the memory cells. Monitor the energy absorption fluctuations of each memory cell and record the dynamic response curve. After the dynamic response curve is processed by Fourier transform, a quantitative dataset of energy characteristic spectrum is derived. The quantization dataset contains the energy distribution pattern of each storage cell, including features such as peak frequency, harmonic components, and attenuation coefficient; Step S200: Perform similarity calculation on the energy distribution patterns in the energy feature spectrum. Generate a similarity matrix by calculating the Euclidean distance and angular deviation between feature vectors. Based on the pattern density in the similarity matrix, automatically determine the number and boundaries of clusters and output feature cluster groups, which are a set of labeled storage units, each group representing a specific energy feature type. Step S300: Apply variable load conditions to each feature cluster group to simulate the actual working environment; the variable load conditions are generated by a stress generator, and the load parameters are adjusted according to the energy characteristic type; monitor the stability index of the storage unit; after the verification results are weighted and scored, the final graded output is generated; the storage unit is divided into multiple performance levels, such as high stability, medium stability and low stability.

[0024] In the above embodiments, the present invention breaks through the limitations of traditional threshold screening, realizes the expansion of testing dimensions from single parameter detection to multi-dimensional feature analysis; upgrades intelligent decision-making from fixed standards to adaptive classification; improves performance prediction capabilities from laboratory conditions to working scenarios; and ultimately forms a hierarchical screening system with a complete closed loop of feature perception, pattern recognition, and environmental verification, which significantly improves the accuracy and applicability of DRAM chip quality assessment.

[0025] Example 2: like Figure 2 As shown, based on Example 1, the process of deriving the quantized dataset of energy feature spectra in step S100 of this embodiment of the invention includes the following steps: Step S101: Divide each dynamic response curve into multiple overlapping time windows. The length of each time window is adaptively determined based on the main oscillation period of the pulse sequence. Apply a local frequency domain reconstruction operation to each independent time window to convert each complete dynamic response curve into a set consisting of multiple local frequency domain segments, called multi-scale response components. Step S102: Analyze the information entropy contained in each local frequency domain segment of the multi-scale response component and calculate a weighting coefficient; the segment with a higher weight indicates that it contains more significant response information generated by impulse excitation rather than random noise; based on the weighting coefficient, the frequency domain information of multiple segments is weighted and synthesized into a single, highly representative synthetic spectrum. Step S103: The synthesized spectrum undergoes feature mining to extract three core quantization features: dominant oscillation mode, harmonic consistency coefficient, and energy decay trajectory. These three features together form a multi-dimensional vector, representing the energy feature spectrum of a memory cell. The set of vectors from all memory cells in the chip forms a quantization dataset.

[0026] In the above embodiments, this embodiment integrates time-frequency analysis, entropy-weighted noise reduction, and multi-dimensional feature mining at the system level. The final output quantized dataset can not only retain the essential characteristics of the original dynamic response, but also eliminate the interference of measurement noise and individual differences, thus realizing high-precision digital representation of the energy characteristics of the storage unit.

[0027] Example 3: Based on Example 2, the process of calculating a weighting coefficient in step S102 of this embodiment of the invention includes the following steps: Step S1021: Perform an analysis called spectral distribution dispersion on each local frequency domain segment in the multi-scale response components to obtain the distribution pattern of energy values ​​of all frequency points within a local frequency domain segment, and obtain an initial significance evaluation value; generate an initial response significance index for each local frequency domain segment. Step S1022: Perform calibration of global consistency constraints, obtain global statistical characteristics of the set of significance indices of responses of all local frequency domain segments, and obtain a set of provisional weights after calibration; Step S1023: Apply an energy conservation constraint to the provisional weights, that is, ensure that the sum of all weight coefficients conforms to the preset specification, and output a set of normalized weight coefficients for weighted synthesis.

[0028] In the above embodiments, this embodiment uses a progressive optimization of local evaluation, global correction, and energy constraint to ensure that the final weight coefficients can accurately reflect the information contribution of each frequency band and conform to the physical conservation laws of the system, thereby providing a highly reliable quantitative basis for subsequent weighted synthesis.

[0029] Example 4: like Figure 4 As shown, based on Example 1, the process of outputting feature cluster groups in step S200 of this embodiment of the invention includes the following steps: Step S201: For any two feature vectors of storage units, perform two sets of independent calculations in parallel; the first set of calculations is the absolute difference of the values ​​of each dimension of the vector, generating an amplitude difference factor; the second set of calculations is the directional difference of the vector in the multidimensional space, generating a direction deviation factor; input the two factors into an adaptive fusion program, and output the inter-unit difference degree; traverse all unit pairs, and the inter-unit difference degree constitutes a difference degree matrix; Step S202: Perform a global coupling transformation on the dissimilarity matrix to transform the dissimilarity relationship between each cell and all other cells into a new index representing the density of its local data environment, which is the cell kernel density; calculate the ratio of the kernel density of each storage cell to the kernel density of all its nearest neighbor storage cells to obtain an autonomy index. Step S203: Each non-seed point unit is assigned to the potential energy field according to its difference relationship with various sub-points, forming a temporary affiliation; the boundary region of the temporary affiliation is checked, the saddle point where the difference changes is found, the final cluster boundary is delineated, and a stable and labeled feature cluster group is output.

[0030] In the above embodiments, this embodiment organically combines difference measurement, density field analysis and potential energy boundary optimization. The final output feature cluster group not only retains the topological structure of the original data, but also achieves adaptive determination of the number of clusters through autonomy index and saddle point detection, thus meeting the unsupervised classification requirements of complex energy feature spectra.

[0031] Example 5: Based on Example 4, the process of outputting the inter-unit difference degree in step S201 provided in this embodiment of the invention includes the following steps: Step S2011: Based on the distribution width of the values ​​of all storage units in the quantized dataset in each dimension, calculate a feature stability weight for each dimension; obtain the absolute difference between two feature vectors in each dimension, multiply it with the dimension weight, perform weighted summation, and output the standardized scalar value, called the amplitude difference factor; Step S2012: Normalize the two feature vectors by dividing them by the L2 norm to obtain two unit vectors. Obtain the cosine value of the angle difference between the vectors and a reference hyperplane defined by the principal direction feature of the entire quantization dataset. The principal direction feature is obtained through statistical analysis of the entire quantization dataset. Map the cosine value to amplify the difference in small angle deviations. The output is the direction deviation factor. Step S2013: Input the amplitude difference factor and the direction deviation factor into an adaptive fusion program, which includes a dynamic balance parameter that adjusts the contribution ratio of the amplitude difference factor and the direction deviation factor, and outputs a scalar value of the inter-unit difference degree.

[0032] In the above embodiments, this embodiment uses a cascaded process of stability weighting, principal direction constraint and dynamic fusion to make the final output inter-unit difference both numerically robust and directionally resolvable, providing a difference measurement benchmark with both global consistency and local sensitivity for subsequent cluster analysis.

[0033] Example 6: Based on Example 5, the process of simultaneously inputting the amplitude difference factor and the direction deviation factor into an adaptive fusion program in step S2013 of this embodiment of the invention includes the following steps: Step S20131: Obtain the sum of the L2 norms of the two feature vectors to get a comprehensive amplitude intensity; perform nonlinear mapping processing on the comprehensive amplitude intensity; convert the comprehensive amplitude intensity into a fusion modulation coefficient between zero and one; Step S20132: Multiply the amplitude difference factor by the fusion modulation coefficient to obtain a modulated amplitude contribution; at the same time, multiply the direction deviation factor by the result of subtracting the fusion modulation coefficient to obtain a modulated direction contribution. Step S20133: The modulated amplitude contribution and the modulated directional contribution are algebraically added, and then processed by a dynamic range compressor. Based on the global distribution range of the modulated amplitude contribution and the modulated directional contribution calculated for all units in the current batch, the result of this addition is scaled to fall into a standardized, unitless numerical range, and the inter-unit difference is output.

[0034] In the above embodiments, this embodiment achieves optimal fusion of amplitude and direction differences through dynamic modulation and global normalization: First, an adaptive fusion coefficient is generated based on the comprehensive amplitude intensity of the feature vector to achieve a nonlinear dynamic balance between amplitude contribution and direction contribution. Finally, through batch-aware global distribution scaling, it is ensured that the output inter-unit difference retains the structural information of the original feature space and has stable comparability across data batches, forming a standardized difference metric that matches the requirements of subsequent clustering algorithms.

[0035] Example 7: Based on Example 6, the process of algebraically adding the modulated amplitude contribution and the modulated direction contribution in step S20133 of this embodiment of the invention includes the following steps: Step S201331: Perform a preliminary algebraic summation of the modulated amplitude contribution and the modulated directional contribution to obtain a transient fusion value; Step S201332: Based on the calibration mechanism of global statistics, in the current screening batch, calculate in real time the set of all transient fusion values ​​obtained by adding the amplitude contribution and directional contribution of all storage units to the generated infinite number of modulations respectively; extract the median and interquartile range of all transient fusion values ​​to form a global distribution statistic; Step S201333: Map the transient fusion value to a standardized, unitless numerical scale, and output the inter-unit difference, which represents the combined difference after weighting and calibration in two different dimensions: amplitude and direction.

[0036] In the above embodiments, this embodiment achieves cross-sample consistency of difference measurement through batch adaptive normalization: transient fusion values ​​capture local feature differences, dynamic calibration based on global median and interquartile range eliminates distribution offset between batches, and finally the original fusion values ​​are linearly projected to the standardized scale through robust statistics, so that the output inter-unit difference degree simultaneously satisfies local sensitivity (preserving subtle differences in feature space) and global comparability (results from different batches can be directly compared).

[0037] Example 8: Based on Example 7, the process of extracting the median and interquartile range of all transient fusion values ​​in step S201332 of this embodiment of the invention includes the following steps: Step S2013321: Perform ordered serialization processing on the entire set of transient fusion values. Through the logic circuit structure, perform multiple sets of comparison and exchange operations synchronously to locate the key quantiles in the set of transient fusion values, and output a set of flag signals to identify the values ​​at several specific positions. After processing, the key value located at the 50th percentile position is obtained, which is the median value of the sequence. Step S2013322: Locate the 50th percentile of the data in the set that is lower than the median of the sequence, called the lower quartile; at the same time, locate the 50th percentile of the data that is higher than the median of the sequence, called the upper quartile; obtain the arithmetic difference between the upper quartile and the lower quartile, called the half-range dispersion. Step S2013323: Combine the median of the sequence with the half-range dispersion to form a binary tuple, which is the final global distribution statistic; the median of the sequence provides the central benchmark of the data distribution, and the half-range dispersion provides the scale of variation of the data around this central benchmark.

[0038] In the above embodiments, this embodiment achieves efficient and robust distribution modeling through a parallel quantile computing architecture: multiple sets of comparison and exchange operations quickly locate the median, and hierarchical recursive solution of quartiles obtains the dispersion against outliers. The final output median-half-range dispersion binary not only characterizes the central tendency of transient fusion values, with the median being more stable than the mean, but also eliminates extreme value interference through the robust dispersion measure of the interquartile range, providing a distribution reference benchmark with both computational efficiency and statistical reliability for subsequent normalization.

[0039] Example 9: like Figure 5 As shown, based on Embodiment 1, the process of dividing the storage unit into multiple performance levels in step S300 of this embodiment of the invention includes the following steps: Step S301: Identify the feature cluster group to which each storage unit belongs, and call a predefined specific weight configuration for the feature cluster group; multiply each stability index of each storage unit by the corresponding specific weight, and then perform weighted summation to generate a weighted stability score; Step S302: Obtain the statistical distribution of the weighted stability scores of all storage units in the current screening batch, and obtain its global performance benchmark and dispersion range; compare the original weighted stability score of each storage unit with this global performance benchmark to obtain the relative offset, and normalize the offset to a standard scale using the dispersion range; output the calibration performance index. Step S303: Analyze the distribution pattern of all calibrated performance indices in the current batch, find the boundary points between the multiple dense regions that are naturally formed, i.e., the troughs in the distribution curve, and determine the troughs as the dynamic thresholds for classifying performance levels such as high stability, medium stability and low stability; each storage unit is classified into the final performance level according to the comparison result between its own calibrated performance index and the dynamic threshold, and completes the entire graded screening process.

[0040] In the above embodiments, this embodiment achieves accurate performance evaluation of storage units through a dynamic adaptive grading strategy: based on the specific weighting of feature clusters, key stability indicators are highlighted; global benchmark normalization eliminates batch differences and generates comparable performance indices; finally, natural boundary points are automatically identified through distribution pattern analysis, so that the performance level division reflects the real difference pattern between storage units, avoids subjective threshold setting deviations, and can dynamically adapt to the performance distribution characteristics of different batches, such as automatically tightening the threshold in high-density areas, to achieve stable and reliable grading and screening.

[0041] Example 10: like Figure 6 As shown, based on Embodiments 1-9, the DRAM chip-based hierarchical screening system provided in this embodiment of the invention includes: Dataset Derivation Module 1 is responsible for applying an energy pulse sequence generated by a waveform generator to the set of storage cells. The energy pulse sequence is designed to excite the deep response characteristics of the storage cells. It monitors the energy absorption fluctuations of each storage cell and records the dynamic response curve. After the dynamic response curve is processed by Fourier transform, a quantitative dataset of energy feature spectrum is derived. The quantization dataset contains the energy distribution pattern of each storage cell, including features such as peak frequency, harmonic components, and attenuation coefficient; Clustering group output module 2 is responsible for calculating the similarity of energy distribution patterns in the energy feature spectrum. It generates a similarity matrix by calculating the Euclidean distance and angular deviation between feature vectors. Based on the pattern density in the similarity matrix, it automatically determines the number and boundaries of clusters and outputs feature clusters, which are a set of labeled storage units, each group representing a specific energy feature type. The weighted scoring module 3 is responsible for applying variable load conditions to each feature cluster group to simulate the actual working environment. The variable load conditions are generated by a stress generator, and the load parameters are adjusted according to the energy feature type. The stability index of the storage unit is monitored. After the verification results are weighted and scored, the final graded output is generated. The storage unit is divided into multiple performance levels, such as high stability, medium stability and low stability.

[0042] In the above embodiments, this embodiment obtains the deep feature spectrum of the memory cell through energy pulse excitation and dynamic response analysis, establishes energy feature type classification by using pattern similarity clustering, and evaluates stability performance under simulated load conditions, ultimately achieving accurate classification based on the correlation between energy features and stability. The system can automatically adapt to the characteristics of different chip batches, improve the accuracy and consistency of classification through energy feature clustering and dynamic threshold division, and ensure the correlation between the evaluation results and the actual working scenario by combining simulated load testing.

[0043] Figure 7 A block diagram of an exemplary electronic device suitable for implementing embodiments of the present invention is shown.

[0044] The electronic device may include a central processing unit / microprocessor / main control chip, etc. 4; and a storage medium 5, coupled to the central processing unit / microprocessor / main control chip, etc. 4, and storing computer-executable instructions therein for performing the steps of various methods of embodiments of the present invention when executed by the processor.

[0045] The central processing unit / microprocessor / main control chip, etc., can include, but are not limited to, one or more processors or microprocessors.

[0046] Storage medium 5 may include, but is not limited to, random access memory (RAM), read-only memory (ROM), flash memory, EPROM memory, EEPROM memory, registers, computer storage media (e.g., hard disk, floppy disk, solid-state drive, removable disk, CD-ROM, DVD-ROM, Blu-ray disc, etc.).

[0047] In addition, the electronic device may also include (but is not limited to) a data bus 6, an input / output bus / external bus / device bus 7, a display 8, and input / output devices 9 (e.g., keyboard, mouse, speaker, etc.).

[0048] The central processing unit / microprocessor / main control chip, etc. 4 can communicate with external devices (8, 9, etc.) via I / O bus 7 through wired or wireless network (not shown).

[0049] The storage medium 5 may also store at least one computer-executable instruction for performing the steps of various functions and / or methods in the embodiments described herein when the central processing unit / microprocessor / main control chip, etc., 4 is running.

[0050] In one embodiment, the at least one computer-executable instruction may also be compiled into or comprise a software product, wherein one or more computer-executable instructions are executed by a processor to perform the steps of the various functions and / or methods in the embodiments described herein.

[0051] Figure 8 A schematic diagram of a computer-readable storage medium according to an embodiment of the present invention is shown.

[0052] like Figure 8 As shown, the non-transitory computer-readable storage medium 11 stores instructions, such as computer-readable instructions 10. When the computer-readable instructions 10 are executed by a processor, the various methods described above can be performed. The non-transitory computer-readable storage medium includes, but is not limited to, volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and / or cache memory. Non-transitory non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc. For example, the non-transitory computer-readable storage medium 11 can be connected to a computing device such as a computer, and then, when the computing device executes the computer-readable instructions 10 stored on the computer-readable storage medium 11, the various methods described above can be performed.

[0053] In the several embodiments provided by this invention, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus 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 apparatuses or units may be electrical, mechanical, or other forms.

[0054] 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.

[0055] 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.

[0056] 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 this 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 for executing all or part of the steps of the methods of the various embodiments of this invention through a computer device (which may be a personal computer, server, or network device, etc.). The aforementioned storage medium includes: USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, optical disks, and other media capable of storing program code.

[0057] 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 hierarchical screening method based on DRAM chips, characterized in that, Includes the following steps: A sequence of energy pulses generated by a waveform generator is applied to the array of memory cells. The energy pulse sequence is designed to excite the deep response characteristics of the memory cells. The energy absorption fluctuations of each memory cell are monitored, and the dynamic response curve is recorded. After the dynamic response curve is processed by Fourier transform, a quantitative dataset of energy characteristic spectrum is derived. The similarity calculation is performed on the energy distribution patterns in the energy feature spectrum. By calculating the Euclidean distance and angular deviation between feature vectors, a similarity matrix is ​​generated. Based on the pattern density in the similarity matrix, the number and boundaries of clusters are automatically determined, and feature cluster groups are output. These are a set of labeled storage units, with each group representing a type of energy feature. Variable load conditions are applied to each feature cluster to simulate the actual working environment; the variable load conditions are generated by a stress generator, and the load parameters are adjusted according to the energy characteristic type; the stability index of the storage unit is monitored; the verification results are weighted and scored to generate the final graded output; the storage unit is divided into multiple performance levels.

2. The hierarchical screening method based on DRAM chips as described in claim 1, characterized in that, The quantization dataset contains the energy distribution pattern of each storage cell, including peak frequency, harmonic components, and attenuation coefficient characteristics.

3. The hierarchical screening method based on DRAM chips as described in claim 1, characterized in that, The process of deriving a quantitative dataset of energy feature spectra includes the following steps: Each dynamic response curve is divided into multiple overlapping time windows, and the length of each time window is adaptively determined based on the main oscillation period of the pulse sequence. A local frequency domain reconstruction operation is applied to each independent time window to convert each complete dynamic response curve into a set of multiple local frequency domain segments, which is called a multi-scale response component. The information entropy contained in each local frequency domain segment of the multi-scale response component is analyzed, and a weighting coefficient is calculated. The segment with a higher weight indicates that it contains more significant response information generated by impulse excitation rather than random noise. Based on the weighting coefficient, the frequency domain information of multiple segments is weighted and synthesized into a single, highly representative synthetic spectrum. After feature mining, three core quantitative features were extracted from the synthesized spectrum: dominant oscillation mode, harmonic consistency coefficient, and energy decay trajectory. The three features together form a multi-dimensional vector, representing the energy feature spectrum of a storage unit; The set of vectors from all storage units of the chip forms a quantized dataset.

4. The hierarchical screening method based on DRAM chips as described in claim 3, characterized in that, The process of calculating a weighting coefficient includes the following steps: An analysis called spectral distribution dispersion is performed on each local frequency domain segment in the multi-scale response components to obtain the distribution pattern of energy values ​​of all frequency points within a local frequency domain segment, thus obtaining an initial significance evaluation value; an initial response significance index is generated for each local frequency domain segment. The calibration of global consistency constraints is performed to obtain the global statistical characteristics of the set of significance indices of responses of all local frequency domain segments. After calibration, a set of provisional weights is obtained. An energy conservation constraint is applied to the provisional weights to ensure that the sum of all weight coefficients conforms to a preset specification, and a set of normalized weight coefficients for weighted synthesis is output.

5. The hierarchical screening method based on DRAM chips as described in claim 1, characterized in that, The process of outputting feature clusters includes the following steps: For any two feature vectors in storage units, two sets of independent calculations are performed in parallel. The first set of calculations calculates the absolute difference in the values ​​of each dimension of the vectors, generating an amplitude difference factor. The second set of calculations calculates the directional difference of the vectors in the multidimensional space, generating a direction deviation factor. The two factors are input into an adaptive fusion program, which outputs the inter-unit difference degree. All pairs of units are traversed, and the inter-unit difference degrees form a difference degree matrix. A global coupling transformation is performed on the difference matrix to transform the difference relationship between each unit and all other units into a new index representing the density of its local data environment, which is called the unit kernel density. The ratio of the kernel density of each memory cell to the kernel density of all its nearest-neighbor memory cells with different degrees of difference is used to obtain an autonomy index. Each non-seed point unit is assigned to a potential energy field based on its difference relationship with various sub-points, forming a temporary affiliation. The boundary regions of the temporary affiliations are examined to find saddle points where the difference changes, thus defining the final cluster boundary and outputting stable and labeled feature clusters.

6. The hierarchical screening method based on DRAM chips as described in claim 5, characterized in that, The process of outputting the inter-unit differences includes the following steps: Based on the distribution width of the values ​​of all storage units in the quantized dataset across each dimension, a feature stability weight is calculated for each dimension. Obtain the absolute difference between two feature vectors in each dimension, multiply it by the dimension weights, perform a weighted summation, and output the standardized scalar value, called the magnitude difference factor. Normalize the two eigenvectors by dividing them by the L2 norm to obtain two unit vectors. Obtain the cosine value of the angle difference between the vectors and a reference hyperplane defined by the principal direction feature of the entire quantized dataset. The principal direction feature is obtained through statistical analysis of the entire quantized dataset. Map the cosine value to amplify the difference in small angle deviations. The output is the direction deviation factor. The amplitude difference factor and the direction deviation factor are simultaneously input into an adaptive fusion program, which includes a dynamic balance parameter that adjusts the contribution ratio of the amplitude difference factor and the direction deviation factor, and outputs a scalar value of the inter-unit difference degree.

7. The hierarchical screening method based on DRAM chips as described in claim 6, characterized in that, The process of simultaneously inputting the amplitude difference factor and the direction deviation factor into an adaptive fusion program includes the following steps: The sum of the L2 norms of the two eigenvectors is obtained to get a composite amplitude intensity; the composite amplitude intensity is then subjected to nonlinear mapping processing; and finally, the composite amplitude intensity is converted into a fusion modulation coefficient between zero and one. Multiplying the amplitude difference factor by the fusion modulation coefficient yields a modulated amplitude contribution; simultaneously, multiplying the directional deviation factor by the result of subtracting the fusion modulation coefficient yields a modulated directional contribution. The modulated amplitude contribution and modulated directional contribution are algebraically added, and then processed by a dynamic range compressor. Based on the global distribution range of the modulated amplitude contribution and modulated directional contribution calculated for all units in the current batch, the result of this addition is scaled to fall into a standardized, unitless numerical range, and the inter-unit difference is output.

8. The hierarchical screening method based on DRAM chips as described in claim 7, characterized in that, The process of algebraically adding the modulated amplitude contribution and the modulated directional contribution includes the following steps: The modulated amplitude contribution and the modulated directional contribution are preliminarily algebraically added to obtain a transient fusion value; Based on a global statistical calibration mechanism, in the current screening batch, the set of all transient fusion values ​​obtained by adding the amplitude contribution and directional contribution of all storage units to the generated infinite number of modulations in real time is calculated; the median and interquartile range of all transient fusion values ​​are extracted to form a global distribution statistic. The transient fusion values ​​are mapped to a standardized, unitless numerical scale, and the inter-unit variability is output, which represents the combined variability after weighting and calibration in two different dimensions: amplitude and direction.

9. The hierarchical screening method based on DRAM chips as described in claim 8, characterized in that, The process of extracting the median and interquartile range of all transient fusion values ​​includes the following steps: The entire set of transient fused values ​​is processed into an ordered serialization. Through a logic circuit structure, multiple sets of comparison and exchange operations are performed synchronously to locate the key quantiles in the set of transient fused values. A set of flag signals is output to identify the values ​​at several specific positions. After processing, the key value located at the 50th percentile position is obtained, which is the median value of the sequence. The position at which the data in the set of values ​​is located 50 percentile below the median of the sequence is called the lower quartile; at the same time, the position at which the data is located 50 percentile above the median of the sequence is called the upper quartile; the arithmetic difference between the upper quartile and the lower quartile is called the half-range dispersion. The midpoint of the sequence is combined with the half-range scatter to form a binary tuple, which is the final global distribution statistic. The midpoint of the sequence provides a central benchmark for the data distribution, while the half-range dispersion provides a scale of variation of the data around this central benchmark.

10. A hierarchical screening system based on DRAM chips, implementing the hierarchical screening method based on DRAM chips as described in any one of claims 1 to 9, characterized in that, include: The dataset derivation module is responsible for applying a sequence of energy pulses generated by a waveform generator to the set of storage cells. The energy pulse sequence is designed to excite the deep response characteristics of the storage cells. Monitor the energy absorption fluctuations of each storage unit and record the dynamic response curve; After the dynamic response curve is processed by Fourier transform, a quantitative dataset of energy characteristic spectrum is derived. The clustering group output module is responsible for calculating the similarity of energy distribution patterns in the energy feature spectrum. It generates a similarity matrix by calculating the Euclidean distance and angular deviation between feature vectors. Based on the pattern density in the similarity matrix, it automatically determines the number and boundaries of clusters and outputs feature clusters, which are a set of labeled storage units, each group representing a type of energy feature. The weighted scoring module is responsible for applying variable load conditions to each feature cluster group to simulate the actual working environment. The variable load conditions are generated by a stress generator, and the load parameters are adjusted according to the energy characteristic type. The stability index of the storage unit is monitored. After the verification results are weighted and scored, the final graded output is generated. The storage unit is divided into multiple performance levels, such as high stability, medium stability and low stability.

Citation Information

Patent Citations

  • A method and system for online testing and classification of flash memory chips with different packages

    CN108682442B

  • Chip testing method and system

    CN112881885A

  • A packaging method and system for implementing a memory chip

    CN118866773B