Test method and system for chip simulation model

Abstract

The invention relates to the technical field of chip testing, in particular to a testing method and system for a chip simulation model. The method comprises the following steps: collecting a target chip simulation model and performing time sequence compensation to generate jitter compensation data, then performing eye pattern reconstruction to obtain waveform eye pattern data, performing amplitude normalization and time sequence phase correction on the waveform data to obtain purified alignment data, and performing phase correction on the purified alignment data. The method comprises the following steps: generating main modal data through orthogonal mapping processing and modal projection analysis, constructing nuclear space data, carrying out manifold cognitive aggregation to generate test coverage features, carrying out boundary analysis and density distribution analysis on the features to obtain clustering boundary data, finally adjusting and optimizing boundary precision, executing an adaptive test sequence, and generating a test case result. According to the invention, a more efficient chip simulation model test method is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of chip testing, and particularly to a method and system for testing a chip simulation model. Background Art

[0002] With the rapid development of semiconductor technology, the complexity and integration of chip design are constantly increasing. Simulation models play an increasingly important role in the chip development process. However, existing chip simulation technologies face problems such as timing jitter and signal interference, which affect the accuracy of simulation results. These problems not only increase the design cycle but also cause performance instability in actual chip applications, affecting the market competitiveness of products. During the chip testing process, the processing and analysis of waveform data are often key links. Existing technologies still have certain limitations in aspects such as waveform amplitude normalization and timing phase correction. Especially when dealing with high-frequency signals and complex data, traditional methods are difficult to effectively extract important features, resulting in insufficient test coverage and an inability to comprehensively evaluate the performance and reliability of the chip, further increasing the risks in the development process and affecting the quality and stability of the final product. Summary of the Invention

[0003] Based on this, it is necessary to provide a method and system for testing a chip simulation model to solve at least one of the above technical problems.

[0004] To achieve the above object, a method for testing a chip simulation model includes the following steps:

[0005] Step S1: Collect a target chip simulation model; perform complete timing compensation on the target chip simulation model to generate jitter compensation data; perform eye diagram reconstruction on the jitter compensation data to obtain waveform eye diagram data;

[0006] Step S2: Perform waveform amplitude normalization on the waveform eye diagram data to generate normalized data; perform timing phase correction on the normalized data to generate purified alignment data;

[0007] Step S3: Perform orthogonal mapping processing on the purified alignment data to generate orthogonal basis data; perform modal projection analysis based on the orthogonal basis data to generate main modal data;

[0008] Step S4: Perform vector mapping drive construction on the main modal data to generate kernel space data; perform manifold cognitive intelligent aggregation on the kernel space data to generate test coverage features;

[0009] Step S5: Perform boundary drive analysis on the test coverage features to generate density distribution data; perform stable feature precise calibration based on the density distribution data to obtain clustering boundary data;

[0010] Step S6: Optimize the boundary accuracy of the clustering boundary data to generate compensated result data; perform an adaptive test sequence execution on the target chip simulation model based on the compensated result data to generate test case results.

[0011] The present invention ensures the stability of signals through the timing complete compensation of the target chip simulation model. The generation of jitter compensation data improves the accuracy of the waveform. The acquisition of waveform eye diagram data provides an intuitive basis for signal quality assessment. The implementation of waveform amplitude normalization enhances the consistency of data. The generation of purified alignment data ensures the phase accuracy of signals. The implementation of orthogonal mapping processing provides an effective mathematical basis for subsequent analysis. The generation of main mode data can effectively extract the main features of signals. The construction of kernel space data provides support for in-depth analysis of signal features. The development of manifold cognitive intelligent aggregation enhances the comprehensiveness of test coverage. The generation of density distribution data provides a quantitative basis for the stability analysis of features. The formation of clustering boundary data can effectively identify the feature distribution of signals. The implementation of boundary accuracy optimization ensures the reliability of test results. The generation of compensated result data provides a basis for the execution of the adaptive test sequence. The formation of test case results can provide comprehensive support for the performance verification of the target chip, improving the scientificity and efficiency of chip simulation model testing as a whole, and providing effective technical means and solutions for chip development and optimization.

[0012] Preferably, step S1 includes the following steps:

[0013] Step S11: Collect the target chip simulation model; perform waveform injection scanning on the target chip simulation model to obtain original waveform data;

[0014] Step S12: Reconstruct the timing structure of the original waveform data to obtain timing deconstructed data; perform edge jitter detection on the timing deconstructed data to obtain edge jitter data;

[0015] Step S13: Perform jitter compensation on the timing deconstructed data according to the edge jitter data to generate jitter compensation data;

[0016] Step S14: Perform multi-domain sampling remapping on the jitter compensation data to obtain remapped data; perform clock recovery calibration on the remapped data to obtain calibrated data;

[0017] Step S15: Perform crosstalk elimination processing on the calibrated data to generate waveform noise reduction data; perform eye diagram reconstruction on the waveform noise reduction data to obtain waveform eye diagram data.

[0018] The present invention provides basic data for signal analysis through waveform injection scanning of the target chip simulation model. The acquisition of the original waveform data ensures the comprehensiveness of signal characteristics. The restructuring of the timing structure enhances the analyzability of the data. The generation of the deconstructed timing data provides an effective basis for subsequent processing. The edge jitter detection can accurately identify the jitter problems in the signal. The generation of the edge jitter data provides important parameters for jitter compensation. The formation of the jitter compensation data improves the stability and accuracy of the signal. The implementation of multi-domain sampling remapping provides support for the multi-dimensional analysis of the signal. The acquisition of the remapped data ensures the integrity of the signal. The clock recovery calibration enhances the time accuracy of the signal. The generation of the calibration data provides a guarantee for subsequent processing. The crosstalk elimination processing effectively eliminates noise interference. The formation of the waveform noise reduction data improves the clarity of the signal. The successful implementation of the eye diagram reconstruction provides a basis for the intuitive evaluation of the signal quality. Overall, it improves the scientificity and effectiveness of the chip simulation model test, and provides reliable support for chip performance verification and optimization.

[0019] Preferably, step S2 includes the following steps:

[0020] Step S21: Perform distortion compensation processing on the waveform eye diagram data to obtain compensated eye diagram data;

[0021] Step S22: Perform amplitude normalization on the compensated eye diagram data to generate normalized data;

[0022] Step S23: Perform phase alignment calibration on the normalized data to obtain phase-aligned eye diagram data;

[0023] Step S24: Perform inter-symbol de-scrambling processing on the phase-aligned eye diagram data to generate purified alignment data.

[0024] The present invention improves the accuracy and readability of the signal through the distortion compensation processing of the waveform eye diagram data. The generation of the compensated eye diagram data provides a basis for subsequent analysis. The implementation of amplitude normalization ensures the consistency of the data. The acquisition of the normalized data enhances the effectiveness of the analysis. The execution of phase alignment calibration improves the time accuracy of the signal. The formation of the phase-aligned eye diagram data can better reflect the signal characteristics. The inter-symbol de-scrambling processing effectively eliminates signal interference. The generation of the purified alignment data ensures the clarity and usability of the signal. Overall, it improves the scientificity and precision of the chip simulation model test, and provides reliable support for chip performance evaluation and optimization.

[0025] Preferably, step S3 includes the following steps:

[0026] Step S31: Perform modal decomposition processing on the purified alignment data to obtain modal feature data;

[0027] Step S32: Extract eigenvectors from the modal feature data to obtain vector group data; perform orthogonalization processing on the vector group data to generate orthogonal basis data;

[0028] Step S33: Based on the orthogonal basis data, perform feature projection simulation to generate a simulation projection matrix;

[0029] Step S34: Conduct modal correlation analysis on the orthogonal basis data according to the simulation projection matrix to obtain modal correlation data; based on the modal correlation data, perform modal screening to generate main modal data.

[0030] The present invention reveals the internal characteristics of the signal through the modal decomposition process of purifying the alignment data. The generation of the modal feature data provides a basis for subsequent analysis. The extraction of eigenvectors ensures the comprehensiveness of the signal features. The formation of the vector group data provides support for subsequent processing. The orthogonalization processing enhances the independence and effectiveness of the data. The generation of the orthogonal basis data provides a reliable basis for the feature projection simulation. The generation of the simulation projection matrix can effectively reflect the distribution of the signal features. The implementation of the modal correlation analysis reveals the correlation between the signal features. The formation of the modal correlation data provides a basis for modal screening. The generation of the main modal data can effectively extract the main features of the signal, overall improving the accuracy and scientificity of the chip simulation model test, and providing effective technical means and support for chip performance evaluation and optimization.

[0031] Preferably, step S4 includes the following steps:

[0032] Step S41: Map the main modal data to test vectors to obtain test sequence data;

[0033] Step S42: Perform kernel function transformation on the test sequence data to generate kernel space data;

[0034] Step S43: Perform manifold learning reconstruction on the kernel space data to generate manifold structure data;

[0035] Step S44: Evaluate the coverage depth of the manifold structure data to generate depth index data;

[0036] Step S45: Perform feature fusion processing on the depth index data to generate test coverage features.

[0037] The present invention effectively generates a targeted test sequence through the test vector mapping of the main modal data. The formation of the test sequence data provides a specific basis for signal evaluation. The implementation of the kernel function transformation improves the representation ability of data in the high-dimensional space. The generation of the kernel space data lays a foundation for the in-depth analysis of signal features. The execution of the manifold learning reconstruction can effectively reveal the internal structure of the signal. The generation of the manifold structure data provides an intuitive view for subsequent analysis. The implementation of the coverage depth evaluation quantifies the comprehensiveness of the test. The formation of the depth index data provides data support for the identification of test features. The successful implementation of the feature fusion processing enhances the expression ability of the test coverage features. Overall, it improves the scientificity and effectiveness of the chip simulation model test, and provides a reliable technical means and solution for chip performance verification and optimization.

[0038] Preferably, step S5 includes the following steps:

[0039] Step S51: Infer the limit conditions of the test coverage features to obtain the limit coverage parameters;

[0040] Step S52: Perform density estimation clustering on the limit coverage parameters to generate density distribution data;

[0041] Step S53: Locate the inflection points based on the density distribution data to obtain the inflection point feature data; Locate the boundaries according to the inflection point feature data to generate the boundary feature data;

[0042] Step S54: Perform stability evaluation based on the boundary feature data to generate stability index parameters; Cluster and optimize the stability index parameters to obtain the clustering boundary data.

[0043] The present invention provides the boundary parameters for signal testing through the inference of the limit conditions of the test coverage features. The generation of the limit coverage parameters provides a basis for the analysis of the limit performance of the signal. The implementation of the density estimation clustering effectively reveals the distribution characteristics of the signal features. The formation of the density distribution data provides support for subsequent analysis. The execution of the inflection point location can identify the key change points of the signal features. The generation of the inflection point feature data provides a foundation for boundary analysis. The formation of the boundary feature data ensures the accurate location of the signal features. The implementation of the stability evaluation quantifies the reliability of the signal. The generation of the stability index parameters provides data support for the analysis of signal characteristics. The successful implementation of the clustering optimization improves the accuracy and effectiveness of the boundary data. Overall, it improves the scientificity and precision of the chip simulation model test, and provides an effective technical means and solution for chip performance verification and optimization.

[0044] Preferably, step S53 includes the following steps:

[0045] Perform gradient transformation on the density distribution data to obtain gradient mapping data; perform local extreme value detection on the gradient mapping data to generate extreme point set data;

[0046] Calculate the curvature of the extreme point set data to obtain curvature characteristic parameters; perform threshold screening on the curvature characteristic parameters based on a preset inflection point curvature threshold to generate candidate inflection point data;

[0047] Perform local window segmentation on the candidate inflection point data to obtain window characteristic data; perform morphological processing on the candidate inflection point data based on the window characteristic data to generate inflection point morphological characteristics;

[0048] Locate the inflection point center according to the inflection point morphological characteristics to generate inflection point characteristic data;

[0049] Perform neighborhood expansion simulation on the inflection point characteristic data to obtain a simulated expansion area; perform directional analysis on the simulated expansion area to generate direction characteristic data;

[0050] Perform continuity identification on the direction characteristic data to obtain continuity index data; perform boundary tracking based on the continuity index data to generate initial boundary data;

[0051] Evaluate the smoothness of the initial boundary data to obtain a boundary smoothness parameter; perform precise repositioning on the initial boundary data based on the boundary smoothness parameter to generate boundary characteristic data.

[0052] The present invention improves the visualization effect of signal features through gradient transformation processing of density distribution data. The generation of gradient mapping data provides a basis for subsequent analysis. Local extreme value detection effectively identifies key change points in the signal. The formation of extreme point set data provides important information for further analysis. The implementation of curvature calculation reveals the shape changes of signal features. The generation of curvature feature parameters provides a basis for inflection point identification. The application of a preset inflection point curvature threshold ensures the accuracy of candidate inflection point data. The generation of window feature data enhances the fineness of inflection point analysis. Morphological processing improves the feature expression ability of candidate inflection point data. The successful implementation of inflection point center positioning ensures the accuracy of inflection point feature data. Neighborhood expansion simulation provides a broader perspective for subsequent analysis. Directional analysis reveals the change trend of signal features. The generation of direction feature data enhances the depth of signal analysis. The implementation of continuity identification quantifies the stability of signal features. The formation of continuity index data provides support for boundary tracing. The generation of initial boundary data lays a foundation for the comprehensive analysis of signal features. The implementation of smoothness evaluation ensures the refinement of boundary features. The generation of boundary smoothness parameters improves the accuracy and reliability of initial boundary data. The successful implementation of precise repositioning ensures the high quality of final boundary feature data, overall improving the scientificity and effectiveness of chip simulation model testing and providing strong support for chip performance verification and optimization.

[0053] Preferably, step S6 includes the following steps:

[0054] Step S61: Calculate the similarity of the clustering boundary data to obtain boundary similarity data;

[0055] Step S62: Compare the consistency of the boundary similarity data to generate comparison result parameters;

[0056] Step S63: Perform deviation analysis and correction based on the comparison result parameters to generate compensation result data; perform candidate test allocation based on the compensation result data to generate supplementary test data;

[0057] Step S64: Execute an adaptive test sequence on the target chip simulation model according to the supplementary test data and the compensation result data to generate test case results.

[0058] The present invention effectively reveals the correlation between boundary features through the similarity calculation of clustered boundary data. The generation of boundary similarity data provides a basis for subsequent consistency comparison. The implementation of consistency comparison ensures the accuracy of data analysis. The generation of comparison result parameters provides a basis for deviation analysis. The execution of deviation analysis correction improves the reliability of data. The formation of compensation result data can effectively guide candidate test allocation. The generation of supplementary test data ensures the comprehensiveness of testing. The execution of the adaptive test sequence provides the target chip simulation model with the ability of dynamic adjustment. The generation of test case results can effectively reflect signal characteristics and performance. Overall, it improves the scientificity and effectiveness of chip simulation model testing, and provides strong technical support for chip performance verification and optimization.

[0059] Preferably, step S63 includes the following steps:

[0060] Perform differential mapping processing on the comparison result parameters to obtain differential distribution data; perform region segmentation processing based on the differential distribution data to obtain region feature data;

[0061] Perform priority sorting on the region feature data based on the differential distribution data to generate deviation level data;

[0062] Perform threshold classification on the deviation level data according to a preset deviation level threshold to generate classification marker data; perform compensation rule matching on the classification marker data to generate matching compensation rules;

[0063] Perform compensation calculation on the classification marker data based on the matching compensation rules to generate compensation result data;

[0064] Perform residual calculation on the compensation result data to obtain compensation residual features; perform convergence analysis on the compensation residual features to generate convergence index data;

[0065] Perform coverage degree analysis on the convergence index data to generate compensation coverage features; perform test scenario mapping based on the compensation coverage features to generate a mapped scenario matrix;

[0066] Decompose test requirements according to the mapped scenario matrix to obtain requirement item data; supplement test conditions for the requirement item data to generate a test condition set;

[0067] Construct test cases based on the test condition set to generate a test task sequence; perform dependency analysis on the test task sequence to obtain task dependency relationships;

[0068] Arrange and allocate the test task sequence according to the task dependency relationships to generate supplementary test data.

[0069] The present invention reveals significant changes in data through differential mapping processing of result parameters. The generation of differential distribution data supports subsequent regional analysis. Region segmentation processing ensures clear identification of features. The formation of region feature data improves the accuracy of data analysis. The implementation of priority ranking effectively identifies important deviations. The generation of deviation level data provides a basis for threshold classification. The application of preset deviation level thresholds ensures the effectiveness of classification. The generation of classification marker data enhances the pertinence of compensation rules. The successful implementation of compensation rule matching improves the rationality of compensation effects. The execution of compensation calculation ensures the scientific nature of results. Residual calculation reveals errors in the compensation process. The generation of compensation residual features provides a basis for convergence analysis. The analysis of convergence index data quantifies the effectiveness of compensation. Coverage degree analysis supports the comprehensiveness of compensation results. The generation of compensation coverage features improves the applicability of test scenarios. The formation of mapping scenario matrix lays a foundation for the refinement of test requirements. The generation of requirement item data ensures comprehensive coverage of tests. The supplementation of test condition sets enhances the pertinence of tests. The implementation of test case construction improves the standardization of tests. Task dependency analysis effectively optimizes the test process. The arrangement and allocation of test task sequences ensure the effective utilization of resources. Overall, it improves the scientific nature and effectiveness of chip simulation model tests, providing strong technical support for chip performance verification and optimization.

[0070] The present invention also provides a test system for a chip simulation model, which is used to execute the test method of the chip simulation model as described above. The test system for the chip simulation model includes:

[0071] An eye diagram reconstruction module, configured to collect a target chip simulation model; perform timing complete compensation on the target chip simulation model to generate jitter compensation data; perform eye diagram reconstruction on the jitter compensation data to obtain waveform eye diagram data);

[0072] A phase correction module, configured to perform waveform amplitude normalization on the waveform eye diagram data to generate normalized data; perform timing phase correction on the normalized data to generate purified alignment data;

[0073] A modal analysis module, configured to perform orthogonal mapping processing on the purified alignment data to generate orthogonal basis data; perform modal projection analysis based on the orthogonal basis data to generate main modal data;

[0074] A feature mapping module, configured to perform vector mapping drive construction on the main modal data to generate kernel space data; perform manifold cognitive intelligent aggregation on the kernel space data to generate test coverage features;

[0075] A boundary analysis module, configured to perform boundary drive parsing on the test coverage features to generate density distribution data; perform stable feature precise calibration based on the density distribution data to obtain clustering boundary data;

[0076] A test optimization module is used to optimize the boundary accuracy of clustering boundary data to generate compensated result data; and perform an adaptive test sequence execution on the target chip simulation model based on the compensated result data to generate test case results.

[0077] The implementation of the present invention through the eye diagram reconstruction module provides a basis for the real-time monitoring of the target chip signal. The generation of jitter compensation data improves the stability and reliability of the signal. The acquisition of waveform eye diagram data can intuitively display the signal quality and characteristics. The function of the phase correction module ensures the consistency of the signal amplitude. The generation of normalized data enhances the accuracy of subsequent analysis. The acquisition of purified alignment data provides a guarantee for the correction of the signal phase. The orthogonal mapping process of the modal analysis module provides an effective method for the extraction of signal features. The generation of main modal data can effectively identify the main features and trends of the signal. The construction of kernel space data by the feature mapping module provides support for the in-depth analysis of signal features. The implementation of manifold cognitive intelligent aggregation improves the comprehensiveness of test coverage. The density distribution data of the boundary analysis module provides a quantitative basis for the stability analysis of features. The generation of clustering boundary data effectively identifies the distribution and changes of signal features. The boundary accuracy optimization ensures the reliability of test results. The generation of compensated result data provides a basis for the execution of the adaptive test sequence. The formation of test case results provides comprehensive support for the performance verification of the target chip. Overall, it improves the scientificity and efficiency of the chip simulation model test, and provides effective technical means and solutions for chip development and optimization. Description of the Drawings

[0078] Figure 1 It is a schematic diagram of the step flow of a test method for a chip simulation model;

[0079] Figure 2 It is a schematic diagram of the detailed implementation step flow of step S2;

[0080] Figure 3 It is a schematic diagram of the detailed implementation step flow of step S3.

[0081] The realization, functional characteristics and advantages of the object of the present invention will be further described in conjunction with the embodiments with reference to the drawings. Detailed Embodiment

[0082] The technical method of the present invention will be clearly and completely described below with reference to the drawings. Obviously, the described embodiments are a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts belong to the scope of protection of the present invention.

[0083] In addition, the accompanying drawings are only schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and thus repeated descriptions thereof will be omitted. Some of the block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities may be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor methods and / or microcontroller methods.

[0084] It should be understood that although the terms "first", "second", etc. may be used herein to describe various units, these units should not be limited by these terms. These terms are only used to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, the first unit may be referred to as the second unit, and similarly the second unit may be referred to as the first unit. The term "and / or" used herein includes any and all combinations of one or more of the listed associated items.

[0085] To achieve the above object, please refer to Figures 1 to 3 , a test method for a chip simulation model, comprising the following steps:

[0086] Step S1: Acquire a target chip simulation model; perform timing complete compensation on the target chip simulation model to generate jitter compensation data; perform eye diagram reconstruction on the jitter compensation data to obtain waveform eye diagram data;

[0087] Step S2: Perform waveform amplitude normalization on the waveform eye diagram data to generate normalized data; perform timing phase correction on the normalized data to generate purified alignment data;

[0088] Step S3: Perform orthogonal mapping processing on the purified alignment data to generate orthogonal basis data; perform modal projection analysis based on the orthogonal basis data to generate main modal data;

[0089] Step S4: Perform vector mapping driving construction on the main modal data to generate kernel space data; perform manifold cognitive intelligent aggregation on the kernel space data to generate test coverage features;

[0090] Step S5: Perform boundary driving analysis on the test coverage features to generate density distribution data; perform stable feature precise calibration based on the density distribution data to obtain clustering boundary data;

[0091] Step S6: Perform boundary accuracy tuning on the clustering boundary data to generate compensation result data; perform adaptive test sequence execution on the target chip simulation model based on the compensation result data to generate test case results.

[0092] The present invention ensures the stability of signals through the timing complete compensation of the target chip simulation model. The generation of jitter compensation data improves the accuracy of waveforms. The acquisition of waveform eye diagram data provides an intuitive basis for signal quality assessment. The implementation of waveform amplitude normalization enhances the consistency of data. The generation of purified alignment data ensures the phase accuracy of signals. The implementation of orthogonal mapping processing provides an effective mathematical basis for subsequent analysis. The generation of main mode data can effectively extract the main features of signals. The construction of kernel space data provides support for in-depth analysis of signal features. The development of manifold cognitive intelligent aggregation enhances the comprehensiveness of test coverage. The generation of density distribution data provides a quantitative basis for the stability analysis of features. The formation of clustering boundary data can effectively identify the feature distribution of signals. The implementation of boundary accuracy tuning ensures the reliability of test results. The generation of compensation result data provides a basis for the execution of adaptive test sequences. The formation of test case results can provide comprehensive support for the performance verification of the target chip. Overall, it improves the scientificity and efficiency of chip simulation model testing, and provides effective technical means and solutions for chip development and optimization.

[0093] In an embodiment of the present invention, the method for testing the chip simulation model includes the following steps:

[0094] Step S1: Collect the target chip simulation model; perform timing complete compensation on the target chip simulation model to generate jitter compensation data; perform eye diagram reconstruction on the jitter compensation data to obtain waveform eye diagram data;

[0095] In this embodiment, during the process of collecting the target chip simulation model, a high-precision data acquisition device is used to comprehensively sample the physical parameters of the target chip. The sampled signals include clock signals, data signals, and power supply signals. The storage of sampled data uses a lossless compression algorithm to ensure data integrity. After sampling, an adaptive timing compensation algorithm is used to perform timing complete compensation on the sampled data. During the compensation process, a time-domain cross-correction method is adopted. By inserting dynamically adjusted compensation nodes in the original data, the signal jitter amplitude is kept at the nanosecond level. After compensation, eye diagram reconstruction is performed on the data with corrected timing. The eye diagram reconstruction uses the sliding window averaging method to segment the sampled data, extracts characteristic parameters such as peak value, level, and overshoot in each time window, and constructs an eye diagram matrix based on statistical characteristics to generate complete waveform eye diagram data.

[0096] Step S2: Perform waveform amplitude normalization on the waveform eye diagram data to generate normalized data; perform timing phase correction on the normalized data to generate purified alignment data;

[0097] In this embodiment, during the process of waveform amplitude normalization of the waveform eye diagram data, first, the maximum amplitude and the minimum amplitude of the signal are calculated based on the amplitude distribution histogram. The amplitude stretching transformation method is used to adjust the amplitude values to the standardized range. During the normalization process, the interval scaling algorithm is adopted to linearly map the amplitude data to the [0, 1] interval. After the normalization is completed, the timing phase correction is performed. The correction method adopts the least squares fitting method. By calculating the phase deviation between the eye diagram data and the reference clock, the phase offset of the data is adjusted to make the zero crossing point of the signal consistent with the ideal clock signal, thereby generating the purified and aligned data.

[0098] Step S3: Perform orthogonal mapping processing on the purified and aligned data to generate orthogonal basis data; perform modal projection analysis based on the orthogonal basis data to generate main modal data;

[0099] In this embodiment, during the process of orthogonal mapping processing of the purified and aligned data, first, the singular value decomposition (SVD) method is used to decompose the data matrix, and the main eigenvectors are extracted as the orthogonal basis vectors. The principal component analysis (PCA) method is used to reduce the dimension of the orthogonal basis vectors, and more than 95% of the variance information is retained. After the mapping is completed, the modal projection analysis method is used to perform modal projection on the orthogonal basis data. The modal projection adopts the weighted projection algorithm to allocate weights to different modal features, and finally the main modal data is obtained.

[0100] Step S4: Perform vector mapping drive construction on the main modal data to generate kernel space data; perform manifold cognitive intelligent aggregation on the kernel space data to generate test coverage features;

[0101] In this embodiment, during the process of vector mapping drive construction of the main modal data, the multilayer perceptron (MLP) is used to map the modal features. During the mapping process, the weight sharing mechanism is used to optimize the computing resources. After the mapping is completed, the kernel function method is used to construct the kernel space data. The kernel function adopts the radial basis kernel (RBF) to calculate the mapping relationship of each modal vector in the high-dimensional space. After the kernel space data is generated, the intelligent aggregation method based on manifold learning is used to perform intelligent aggregation on the data. During the aggregation process, the manifold dimensionality reduction algorithm is used to construct a nonlinear low-dimensional space to obtain the test coverage features.

[0102] Step S5: Perform boundary drive analysis on the test coverage features to generate density distribution data; perform stable feature precise calibration based on the density distribution data to obtain clustering boundary data;

[0103] In this embodiment, during the process of boundary-driven analysis of test coverage features, the density peak clustering algorithm (DPC) is first used to cluster the test features, calculate the local density of each feature point, and mark the positions of boundary points. During the analysis process, the density distribution function is used to fit the feature boundaries, and finally density distribution data is generated. Based on the density distribution data, the minimum entropy optimization algorithm is used to accurately calibrate stable features. During the calibration process, the Kalman Filter is used to smooth the feature data to obtain the clustering boundary data.

[0104] Step S6: Optimize the boundary accuracy of the clustering boundary data to generate compensated result data; perform an adaptive test sequence execution on the target chip simulation model based on the compensated result data to generate test case results.

[0105] In this embodiment, during the process of optimizing the boundary accuracy of the clustering boundary data, the Iterative Closest Point (ICP) algorithm is used to match the boundary data with the ideal model. By iteratively optimizing the boundary error, the boundary error converges to a set threshold. After the accuracy optimization is completed, an adaptive test sequence execution is performed on the target chip simulation model based on the compensated result data. The test sequence uses a dynamic random sequence generation algorithm to dynamically adjust the input sequence according to the actual operating environment of the chip. During the execution process, a high-precision measuring instrument is used to collect the test results, and finally test case results are generated.

[0106] Preferably, step S1 includes the following steps:

[0107] Step S11: Collect the target chip simulation model; perform waveform injection scanning on the target chip simulation model to obtain the original waveform data;

[0108] Step S12: Reconstruct the timing structure of the original waveform data to obtain the timing deconstructed data; perform edge jitter detection on the timing deconstructed data to obtain the edge jitter data;

[0109] Step S13: Perform jitter compensation on the timing deconstructed data according to the edge jitter data to generate the jitter compensation data;

[0110] Step S14: Perform multi-domain sampling remapping on the jitter compensation data to obtain the remapped data; perform clock recovery calibration on the remapped data to obtain the calibrated data;

[0111] Step S15: Perform crosstalk elimination processing on the calibrated data to generate waveform noise reduction data; perform eye diagram reconstruction on the waveform noise reduction data to obtain the waveform eye diagram data.

[0112] In this embodiment, the register-transfer level (RTL) model of the target chip is loaded through the simulation tool Cadence Virtuoso. The working clock frequency is configured to be 500 MHz in the simulation environment, the voltage is set to 1.2 V, the signal source generator Keysight M8195A is connected to the chip input port, the waveform injection is performed using the PRBS31 pseudo-random code sequence, and the oscilloscope Tektronix DPO73304D is used to synchronously collect the signal data at the output end. The acquisition depth is set to 100 M points, and the time-domain resolution is set to 10 ps. After the data acquisition is completed, the original waveform data is exported as...CSV format files are used to perform preliminary visual analysis on the collected data using the signal analysis tool MATLAB, verify data integrity and ensure that the sampling range covers the entire test time window. The timing analysis tool Synopsys PrimeTime is used to perform timing parsing on the original waveform data. The collected waveform is divided into multiple clock cycle windows according to the cycle of the simulation clock signal. The zero-crossing detection algorithm is used to identify the rising and falling edges of the waveform. After reconstructing the complete timing structure, the Gaussian weighted moving average filtering method is applied to extract the signal edge features. The jitter amplitude of each signal edge is calculated through the timing edge detection module. The detection threshold is set to ±5 ps. After removing the abnormal points, the edge jitter data is generated. The timing deconstruction data is input into the timing correction module. Combining with the edge jitter data, the phase correction of the signal waveform is performed by the Lagrange interpolation method. The window smoothing method is used to adjust the jitter amplitude of each cycle. The correction step size is set to 2 ps. Nonlinear compensation is performed according to the distribution characteristics of the edge jitter data. After the compensation is completed, the standard reference clock signal is used for comparison and verification to ensure that the data after jitter correction meets the timing constraints of the target chip. Finally, the compensated signal waveform data is output. The jitter compensation data is loaded into the frequency domain conversion module. The fast Fourier transform (FFT) is used to perform frequency domain analysis on the time domain signal. The sampling rate is set to 10 GS / s. Multiple frequency bands are divided and remapping operations are performed. Band-pass filters are used to extract the baseband and harmonic components respectively. After completing the frequency domain reconstruction, clock recovery calibration is achieved through a phase-locked loop (PLL). The phase deviation is adjusted to within ±1 ps. Finally, the calibrated data with accurate clock recovery is generated. The calibrated data is input into the signal integrity analysis tool Ansys SIwave. The crosstalk interference sources between signals are identified based on the spatial proximity analysis method. The adaptive noise suppression algorithm is used to perform crosstalk elimination processing. The crosstalk suppression threshold is set to -30 dB. After completing the crosstalk elimination, a sample-and-hold circuit is used to stabilize the waveform. It is loaded into the eye diagram analysis module. The data is aligned according to the UI (Unit Interval). 1000 repeated samplings are used to generate a clear waveform eye diagram. Key parameters such as eye height, eye width, and jitter margin are calculated.

[0113] Preferably, step S2 includes the following steps:

[0114] Step S21: Perform distortion compensation processing on the waveform eye diagram data to obtain compensated eye diagram data;

[0115] Step S22: Normalize the amplitude of the compensated eye diagram data to generate normalized data;

[0116] Step S23: Perform phase alignment calibration on the normalized data to obtain phase-aligned eye diagram data;

[0117] Step S24: Perform inter-symbol de-scrambling on the phase-aligned eye diagram data to generate purified alignment data.

[0118] In this embodiment, the waveform eye diagram data is input into the signal distortion analysis module. The frequency domain characteristics of the waveform are analyzed using the Band-Limited Model. Based on distortion types such as amplitude distortion, phase distortion, and non-linear distortion, an optimization algorithm based on the least squares method is used to perform reverse compensation on the waveform data. The step size of the distortion compensation is set to 0.5 ns, and the compensation coefficient is adjusted in combination with actual test data. Finally, the compensated eye diagram data is obtained. The eye diagrams before and after compensation are compared through an eye diagram display tool. The compensated eye diagram data is input into the amplitude normalization module. The maximum and minimum values of the waveform are standardized using the normalization algorithm. The normalization range is set from 0 to 1, and the amplitude ratio is adjusted through dynamic normalization. In the specific operation process, first, the maximum and minimum amplitudes of the waveform data are calculated, and then the ratio scaling is performed according to the normalization formula. Finally, the normalized data is obtained. The time-domain data is converted into frequency-domain data through Fourier transform to analyze the phase deviation of the signal. Then, a phase adjustment algorithm is used, and the phase alignment target is set to an error range of 0.1 ps. During the signal phase correction process, the Bisection method is used to refine the adjustment step size and gradually reduce the correction error. After the adjustment is completed, the data is converted back to the time domain and phase alignment is performed to obtain accurately aligned eye diagram data. The eye diagram data before and after calibration is compared using an oscilloscope. The data is divided into time-domain windows, and the sliding window method is used to perform local processing on the data. The noise with a frequency lower than the signal bandwidth is filtered out through a high-pass filter. The cut-off frequency of the filter is set to 5 GHz. The Kalman Filtering algorithm is used to further eliminate the interference between symbols, and purified signal data is generated after correction. In data processing, the Recursive Least Squares (RLS) algorithm is used to estimate and correct the noise in real time. Finally, the processed purified data is verified, and the final purified eye diagram data is generated through an eye diagram tool.

[0119] Preferably, step S3 includes the following steps:

[0120] Step S31: Perform modal decomposition on the purified alignment data to obtain modal feature data;

[0121] Step S32: Extract eigenvectors from the modal feature data to obtain vector group data; perform orthogonalization processing on the vector group data to generate orthogonal basis data;

[0122] Step S33: Perform feature projection simulation based on the orthogonal basis data to generate a simulated projection matrix;

[0123] Step S34: Perform modal correlation analysis on the orthogonal basis data according to the simulated projection matrix to obtain modal correlation data; perform modal screening based on the modal correlation data to generate principal modal data.

[0124] In this embodiment, the purified alignment data is input into the modal decomposition module. The principal component analysis (PCA) method is used to perform time-domain decomposition on the signal. The number of principal components for modal decomposition is set to 5. The principal component coefficients of each signal are calculated using the PCA algorithm and sorted according to their contribution degrees. The principal components with a contribution degree higher than 85% are selected for analysis. These principal components are used to construct modal feature data, and finally, modal data containing significant features is obtained. The effectiveness of the modal data is further verified through frequency-domain analysis. The frequency-domain reconstruction is carried out using the toolbox of MATLAB. The modal feature data is input into the feature vector extraction module. The singular value decomposition (SVD) method is used to perform matrix decomposition on the modal feature data. The accuracy threshold for matrix decomposition is set to 0.01. During the decomposition process, several singular values and corresponding eigenvectors are generated. The first 5 eigenvectors in the eigenvector matrix are selected to generate vector group data. The “svd” function in R language is used to extract eigenvectors from the data. By comparing the contribution degrees of each eigenvector, it is ensured that the selected eigenvectors cover the features of most of the data. The extracted eigenvector data is input into the orthogonalization processing module. The Gram-Schmidt orthogonalization method is used to orthogonalize the vector group data. The error threshold for each vector during the orthogonalization process is set to 10^-4. First, the original vectors are normalized, and then the inner product values between the vectors are gradually adjusted to make the orthogonal error between the vectors reach the set threshold, obtaining orthogonal basis data. This operation is implemented using the NumPy library in Python. By calculating the angles between the vectors, it is ensured that the orthogonal basis data fully conforms to the orthogonal standard. The orthogonal basis data is input into the feature projection simulation module. The projection analysis algorithm is used to project each eigenvector into a new space. The dimension of the projection space is set to 3. The projection matrix is fitted using the least squares method to obtain the simulated projection matrix. During the projection process, the projection error is set to 0.001. By gradually adjusting the projection angle and offset, it is ensured that the projection matrix can accurately reflect the features of the orthogonal basis data. The “proj” function in Matlab is used for data projection, and the proximity of each projection point to the actual data point is checked. The simulated projection matrix and the orthogonal basis data are input into the modal correlation analysis module. The Pearson Correlation Coefficient method is used to calculate the correlation between each modal. The correlation threshold is set to 0.8. The correlation coefficients of each pair of modal data are calculated. By analyzing the correlation between each modal, the modal pairs with a correlation greater than 0.8 are selected as valid modes, obtaining modal correlation data. The “cor” function in R language is used to calculate the correlation of the modal data. The modal correlation data is input into the modal screening module. The threshold-based screening algorithm is used, and the correlation threshold is set to 0.85. Select the modalities with a correlation greater than this threshold, and finally generate the main modality data. During the screening process, set the correlation score of each modality to be between 0 and 1. Through regression analysis of the screened data, determine the selection of the main modality data. Use the "sklearn" library in Python to process the screening process, and finally obtain accurate main modality data.

[0125] Preferably, step S4 includes the following steps:

[0126] Step S41: Perform a test vector mapping on the main modality data to obtain test sequence data;

[0127] Step S42: Perform a kernel function transformation on the test sequence data to generate kernel space data;

[0128] Step S43: Perform a manifold learning reconstruction on the kernel space data to generate manifold structure data;

[0129] Step S44: Perform a coverage depth evaluation on the manifold structure data to generate depth index data;

[0130] Step S45: Perform a feature fusion process on the depth index data to generate test coverage features.

[0131] In this embodiment, the main modal data is input into the test vector mapping module, and the mapping is performed using the method of Support Vector Machine (SVM). The kernel function of the support vector machine is set as the Radial Basis Function (RBF). During the mapping process, the mapping dimension is selected as 6, and the mapping value of the main modal data in the high-dimensional space is calculated through the SVM algorithm. The "svm" toolbox of MATLAB is used for mapping. During the mapping process, the classification accuracy is controlled by setting the C value to 1 and the γ value to 0.5. By post-processing the result of the test vector mapping, the test sequence data is obtained. This data contains the test vectors of each group of main modal data after mapping and their corresponding output category information. The test sequence data is input into the kernel function transformation module, and the Gaussian Kernel is selected as the transformation function. The parameter σ of the kernel function is set to 0.8, and the "kernel_transform" function in the "sklearn" library of Python is used to perform the Gaussian kernel transformation on the test sequence data. By mapping the original data to the high-dimensional space, the kernel space data is obtained. During the kernel function transformation process, by calculating the similarity between the input data points and the mapping points, a new data representation is generated, making the data points more distinguishable in the high-dimensional space. The transformed data has higher classification characteristics. The kernel space data is input into the manifold learning reconstruction module, and the Laplacian Eigenmaps (LE) algorithm is used to reconstruct the data. The neighborhood size is set to 10, and the "laplacian_eigenmaps" function in the R language is used to reconstruct the data. First, the adjacency matrix between the data points is constructed, the weight function is set as the Gaussian function, and the eigen-decomposition is performed based on the Laplacian matrix to obtain the low-dimensional representation of the data. By visualizing the result of the manifold reconstruction, it is ensured that the obtained manifold structure data can effectively retain the local structure characteristics of the data. During the reconstruction process, by setting the reconstruction error threshold to 0.005, the manifold structure data is input into the coverage depth evaluation module, and the K-Nearest Neighbors (KNN) method is used to evaluate the coverage depth. The K value is set to 5, and the depth index data is obtained by calculating the local density and coverage depth of each data point in the manifold structure. The "KNN" function in Matlab is used to implement this operation. By analyzing the distance relationship between the data points, the coverage depth of each point is determined. During the evaluation process, the depth evaluation threshold is set to 0.02 to ensure the accuracy of the depth evaluation. By continuously adjusting the K value and the depth threshold, the final depth index data is generated. The depth index data is input into the feature fusion processing module, and the weighted fusion method is used to fuse different features. The weighting factors are set to 0.6 and 0.4, representing the weights of the depth metric data and the test sequence data respectively. By calculating the weighted sum of each feature, the fused test coverage feature is obtained. The feature fusion is performed using the weighted average function in the "numpy" library in Python. By analyzing the fusion result, it is ensured that the test coverage feature can reflect the comprehensive information of both the test sequence data and the depth metric data, and finally the test coverage feature data is generated.

[0132] Preferably, step S5 includes the following steps:

[0133] Step S51: Infer the extreme conditions of the test coverage feature to obtain the extreme coverage parameters;

[0134] Step S52: Perform density estimation clustering on the extreme coverage parameters to generate density distribution data;

[0135] Step S53: Locate the inflection points based on the density distribution data to obtain the inflection point feature data; perform boundary location according to the inflection point feature data to generate the boundary feature data;

[0136] Step S54: Perform stability evaluation based on the boundary feature data to generate the stability index parameters; perform clustering optimization on the stability index parameters to obtain the clustering boundary data.

[0137] In this embodiment, the test coverage feature data is input into the extreme condition inference module, and the extreme value theory (EVT) is used for data analysis. The threshold is set as the 95% quantile of the data. By performing maximum clustering on the test coverage feature data, the behavior of the data under extreme conditions is inferred. The "evm" toolbox in Matlab is used for extreme condition inference, and the tail characteristics of the data are modeled in combination with the extreme value distribution to determine the extreme coverage parameters. The inference result will include the response limit of the coverage feature under extreme conditions. By calculating the statistical characteristics of the extreme coverage parameters, the tolerance of the parameters is set to 0.1. The extreme coverage parameters are input into the density estimation clustering module, and the kernel density estimation (KDE) algorithm is used to estimate the density of the extreme coverage parameters. The kernel function is set as the Gaussian kernel, and the kernel width (bandwidth) is selected as 0.5. The "density" function in the R language is used for density estimation clustering. During the estimation process, the local density of each extreme coverage parameter point is calculated to form a probability density function (PDF). By integrating the density function, density distribution data is generated. The density distribution data will reflect the distribution characteristics of the extreme coverage parameters. The density distribution data is input into the inflection point location module, and the second derivative method is used to analyze the inflection points of the data. The "find_peaks" function in the "scipy" library of Python is used for inflection point location. The threshold for inflection point detection is set to 0.02. The second derivative of the density distribution curve is calculated to detect and calibrate the local maxima and minima of the curve. The inflection points will be extracted as key feature points to obtain the inflection point feature data. The inflection point data will include the position and the corresponding density value. The inflection point feature data is input into the boundary location module, and the boundary segmentation algorithm is used to process the inflection point data. The minimum distance for boundary recognition is set to 0.3. Connect the inflection points by a method based on Dynamic Programming (DP) to form the boundary interval of the data. Generate the boundary feature data through this method. The boundary positioning process determines the starting and ending points of the boundary by analyzing the relative positions and density changes between the inflection points. The generated boundary feature data includes the specific positions of the boundary and their corresponding density values. Input the boundary feature data into the stability evaluation module, and use the Lyapunov exponent method to evaluate the stability of the boundary. Set the evaluation window size to 5, and calculate the Lyapunov exponent using the "lyapunov" function in Python. Analyze the dynamic changes of the boundary feature data within different time windows to obtain the stability index parameters. During the stability evaluation process, the volatility and uncertainty of the data are considered. The calculated Lyapunov exponent reflects the stability of the system. If the exponent is negative, it indicates that the system is stable. The generated stability index parameters include the Lyapunov exponent value, standard deviation, mean, etc. Input the stability index parameters into the clustering optimization module, and use the K-Means Clustering algorithm to optimize the stability index data. Set the value of K to 3, and perform clustering processing using the "sklearn.cluster.KMeans" function in Python. During the optimization process, adjust the value of K to make the clustering effect optimal, and generate the clustering boundary data. The boundary data generated after clustering will be grouped according to the distribution of the stability index to obtain the boundary positions and stability characteristics of each group, and finally generate the clustering boundary data.

[0138] Preferably, step S53 includes the following steps:

[0139] Perform gradient transformation processing on the density distribution data to obtain gradient mapping data; perform local extreme value detection on the gradient mapping data to generate extreme point set data;

[0140] Calculate the curvature of the extreme point set data to obtain curvature characteristic parameters; perform threshold screening on the curvature characteristic parameters based on a preset inflection point curvature threshold to generate candidate inflection point data;

[0141] Perform local window segmentation on the candidate inflection point data to obtain window feature data; perform morphological processing on the candidate inflection point data based on the window feature data to generate inflection point morphological characteristics;

[0142] Locate the inflection point center according to the inflection point morphological characteristics to generate inflection point feature data;

[0143] Perform neighborhood expansion simulation on the inflection point feature data to obtain a simulated expansion area; perform directional analysis on the simulated expansion area to generate direction feature data;

[0144] Perform continuity recognition on the direction feature data to obtain continuity index data; perform boundary tracing based on the continuity index data to generate initial boundary data;

[0145] Perform smoothness evaluation on the initial boundary data to obtain boundary smoothness parameters; perform precise repositioning on the initial boundary data based on the boundary smoothness parameters to generate boundary feature data.

[0146] In this embodiment, the density distribution data is input into the gradient transformation processing module. The gradient of each data point is calculated using the Central Difference Method. The window size is set to 3. By calculating the gradient of each data point and its neighborhood data, gradient mapping data is obtained. In the gradient transformation process, by calculating the local change rate of the density distribution, significant regions of data change are identified. The obtained data includes the gradient magnitude and direction of each point. The processing is implemented using the "gradient" function in the NumPy library of Python. The gradient mapping data is input into the local extreme value detection module. The second derivative method is used to detect local extreme values. The extreme value detection threshold is set to 0.1, and the "findpeaks" function in Matlab is used for detection. By analyzing the second derivative of each data point and its neighborhood points, local extreme value points are found, and extreme value point set data is generated. The extreme value point set data will include the position of each extreme value point and the corresponding gradient value. Extreme value points are key change points extracted from the gradient mapping data, representing significant turning points of the density distribution data. The extreme value point set data is input into the curvature calculation module. The curve fitting method is used to calculate the curvature. The fitting function is set as a quadratic polynomial, and the "scipy.optimize.curve_fit" function in Python is used to fit the data. According to the fitting result, the second derivative of the curve is calculated, and then the curvature characteristic parameters are obtained. During the curvature calculation process, the concavity and convexity of points are judged by the change speed of the curve. The generated curvature characteristic parameters include the curvature value of each point, indicating the degree of bending of the point on the curve. The curvature characteristic parameters are input into the inflection point screening module. The curvature threshold is set to 0.05, and the curvature characteristic parameters are screened by the threshold. Points with curvature values greater than the threshold are selected as candidate inflection point data. The "numpy.where" function in Python is used for threshold screening. The generated candidate inflection point data includes points with curvature greater than the set threshold and their corresponding curvature values. These points represent the turning positions of the curve. The candidate inflection point data is input into the local window segmentation module. The sliding window method is used to segment the data. The window size is set to 5. By moving the window to divide the candidate inflection points, it is ensured that the data within each window contains at least one inflection point. Within each window, the maximum and minimum values within the window are extracted to generate window feature data. The sliding window method analyzes the candidate inflection point data by dividing it into multiple local regions for subsequent processing. The generated window feature data includes the extreme value points within each window and their change trends. The window feature data is input into the morphological processing module. Erosion and Dilation operations are used to perform morphological processing on the data. The structuring element is set as a 3x3 matrix, and the "cv2.erode" and "cv2.Operate using the "dilate" function. By expanding and contracting the data, generate the morphological features of the inflection points. Morphological processing can remove noise and extract the main structure of the data. The finally generated morphological feature data of the inflection points contains the shape and spatial position of each inflection point. Input the morphological features of the inflection points into the center positioning module. Use the Centroid Method to calculate the center position of the inflection points. Set the positioning accuracy to 0.01. Calculate the centroid of the inflection point morphological data. Use the "regionprops" function in Matlab to calculate the centroid and obtain the position of the inflection point center, generating inflection point feature data. The inflection point center positioning process generates the inflection point feature data by determining the geometric center of the data points, which contains the exact position of the inflection points and their corresponding attributes. Input the inflection point feature data into the neighborhood expansion simulation module. Use the Voronoi Diagram algorithm for neighborhood expansion. Set the neighborhood expansion radius to 0.1. Use the "scipy.spatial.Voronoi" function in Python to calculate the neighborhood area of each inflection point. The expansion process generates the expansion area of each inflection point based on the position of the inflection point, obtaining the simulated expansion area. The result of the neighborhood expansion simulation contains the influence range of each inflection point, indicating the influence of the inflection point on the surrounding area. Input the simulated expansion area into the directional analysis module. Use the Gradient Direction Method to analyze the expansion area. Set the analysis accuracy to 0.02. Calculate the gradient direction of the data within each area to obtain the directional features of the expansion area. Use the "gradient" function in Matlab for gradient direction analysis. The generated direction feature data includes the main direction and gradient value of each area, representing the main flow direction of each expansion area. Input the direction feature data into the continuity recognition module. Use the time series analysis method to identify the continuity of the data. Set the time window to 10. Use the "statsmodels.tsa" library in Python for time series modeling and analyze the change trend of each data point to obtain the continuity index data. Continuity recognition identifies the stable change areas by analyzing the time sequence of the data. The generated continuity index data includes the change trend and stability of each data point. Input the continuity index data into the boundary tracking module. Use the Boundary Tracking Algorithm to track the data. Set the tracking threshold to 0.05. Use "skimage.measure." in Python.The "label" function marks the data in regions, identifies continuous regions of the boundary, generates initial boundary data. The boundary tracking process generates the initial boundary data containing the shape and position of the boundary by analyzing the change trend in the continuity index data. The initial boundary data is input into the smoothness evaluation module, and the least squares method is used to calculate the smoothness of the boundary data. The fitting accuracy is set to 0.01, and the boundary is fitted using the "scipy.optimize.curve_fit" function in Python to calculate the smoothness parameters of the boundary. The generated smoothness parameters will include the fitting curve and residuals of the boundary, representing the smoothness degree of the boundary. The boundary smoothness parameters are input into the precise repositioning module, and the minimum distance repositioning method is used to adjust the boundary. The repositioning accuracy is set to 0.01, and the boundary is repositioned by adjusting each point of the boundary points and using the "interpolate" function in Matlab for data interpolation, generating boundary feature data. The repositioned boundary feature data will include a more precise position and shape.

[0147] Preferably, step S6 includes the following steps:

[0148] Step S61: Calculate the similarity of the clustering boundary data to obtain boundary similarity data;

[0149] Step S62: Compare the consistency of the boundary similarity data to generate comparison result parameters;

[0150] Step S63: Perform deviation analysis and correction based on the comparison result parameters to generate compensation result data; perform candidate test allocation based on the compensation result data to generate supplementary test data;

[0151] Step S64: Execute an adaptive test sequence on the target chip simulation model according to the supplementary test data and the compensation result data to generate test case results.

[0152] In this embodiment, the Euclidean Distance is used as a metric to compare each data point of each clustering boundary one by one, calculate the similarity between them, use the standard calculation method to compare each pair of boundary data points pairwise, obtain the similarity values between each clustering boundary and other clustering boundaries, obtain the similarity of the boundary by calculating the shortest distance from each point to all other points, and finally obtain a similarity matrix, where each item represents the similarity between two clustering boundary data. After the calculation for each boundary is completed, the obtained similarity matrix will be used as the basic data for subsequent consistency comparison. The boundary similarity data obtained in the previous step is input into the consistency analysis algorithm. This algorithm uses the matching degree algorithm to compare each group of data, compares the similarity of two groups of boundaries according to a preset threshold. If the similarity exceeds the threshold, it is determined to be consistent and generates comparison result parameters. The entire comparison process scans the boundary data step by step, and gradually determines the consistency of the boundary data based on the sorting of the similarity values, and outputs a result including the matching degree of each group of data. Usually, cluster analysis or the K-means algorithm is used for the sorting and comparison of the matching degree. The generated comparison result parameters are represented by binary values, where 1 represents consistency and 0 represents inconsistency. Filter out the inconsistent data pairs from the matching results, and use these data pairs as the objects of deviation analysis. The deviation analysis adjusts them by comparing the deviation amounts of the two groups of boundary data, and uses the weighted average method to correct the inconsistent boundary data. During the correction process, the correction coefficients of each boundary data are calculated, and the compensated result data is generated after correction. This compensated result data will play an important role in subsequent test allocation. The compensated result data is mainly composed of the corrected boundary parameters, including the compensated position, direction, and morphological features. After the compensated result data is generated, it is allocated through the test allocation algorithm. First, according to the characteristics of the compensated result data, candidate test data is generated. The candidate test allocation is adjusted based on the parameters in the compensated result, and the genetic algorithm is used for optimal allocation. During the optimization process, the error range of the compensated result data is calculated to dynamically allocate test tasks and generate a group of candidate test data. These candidate data are effectively extended based on the compensated data to ensure the maximum matching degree between the test data and the target chip simulation model. The candidate test data includes the corrected values after compensation and the optimized test path parameters. Based on the supplementary test data and the compensated result data, an adaptive test sequence for the target chip simulation model is executed. First, according to the rules of the test sequence, for each test data point, the test order and execution path are adjusted based on the previous compensated result data, and each test case is simulated and executed using an automated test platform. The execution results of the test cases are generated through simulation software (such as Simulink).The execution of test cases is dynamically adjusted based on the previously supplemented test data. During the execution process, test data feedback is obtained in real time, and the test data is updated through the feedback results. Finally, the execution of the adaptive test sequence is completed, and the generated test case results will be used for subsequent chip performance evaluation.

[0153] Preferably, step S63 includes the following steps:

[0154] Perform differential mapping processing on the comparison result parameters to obtain differential distribution data; perform region segmentation processing based on the differential distribution data to obtain region feature data;

[0155] Perform priority sorting on the region feature data based on the differential distribution data to generate deviation level data;

[0156] Perform threshold classification on the deviation level data according to the preset deviation level threshold to generate classification marker data; perform compensation rule matching on the classification marker data to generate a matching compensation rule;

[0157] Perform compensation calculation on the classification marker data based on the matching compensation rule to generate compensation result data;

[0158] Perform residual calculation on the compensation result data to obtain compensation residual features; perform convergence analysis on the compensation residual features to generate convergence index data;

[0159] Perform coverage degree analysis on the convergence index data to generate compensation coverage features; perform test scenario mapping based on the compensation coverage features to generate a mapped scenario matrix;

[0160] Decompose the test requirements according to the mapped scenario matrix to obtain requirement item data; supplement the test conditions for the requirement item data to generate a test condition set;

[0161] Construct test cases based on the test condition set to generate a test task sequence; perform dependency analysis on the test task sequence to obtain task dependencies;

[0162] Arrange and allocate the test task sequence according to the task dependencies to generate supplementary test data.

[0163] In this embodiment, a comparison result parameter data set is obtained. The difference mapping algorithm is used to compare the comparison result parameters with the data of the target chip simulation model item by item. The difference values between the parameters are converted into difference mapping data. The difference values are quantified by the Euclidean Distance. The quantified value of each difference is mapped to a specified region to obtain the complete difference distribution data. The difference value of each region reflects the deviation of the chip model under different operating conditions during the test process. When performing region segmentation processing, based on the difference distribution data, the K-means clustering algorithm is used to segment the data. The regions with larger difference values are marked as key points. The average difference value of each region is calculated as the representative feature value of the region. The region feature data obtained by this method includes the main difference features and relative positions of each region. The segmented region data provides a basis for the subsequent generation of deviation levels. During the region segmentation process, the input difference distribution data and the number of clustering centers of the K-means algorithm are optimized through experimental data. When prioritizing the region feature data based on the difference distribution data, first calculate the features of each region, and sort their priorities according to the difference values of the region features. The calculation of the priority is weighted according to the absolute size of the difference value and the range of the difference mapping. The sorting of the region feature data is carried out according to the weight factor sorting rule. The obtained deviation level data includes the deviation level of each region. The regions with larger values have higher deviation levels. When classifying the deviation level data according to the preset deviation level threshold, first divide the deviation level data into different categories according to a predetermined standard, and set multiple thresholds, such as high, medium, and low levels. The high level indicates a large difference and needs to be corrected first, while the medium and low levels are corrected according to the degree of difference respectively. During classification, the data is divided into different categories according to the difference ranges of each level, and the generated classification mark data is stored in the form of category numbers. When matching the compensation rules, according to the classification mark data, the data of each category is matched with the rules. During the matching process, a predefined compensation rule library is used. The rule library includes data compensation strategies for different deviation levels. The corresponding compensation rules are applied to each classification mark data through an algorithm. The compensation rules are matched according to the characteristics of the deviation level data, and the matched compensation rules are applied to the relevant data to generate the matched compensation rules. When performing compensation calculation on the classification mark data based on the matched compensation rules, first use the matched compensation rules to perform compensation operations on the classification mark data. During the calculation process, according to the parameters in the compensation rules, the deviation values of each piece of data are corrected one by one through the compensation algorithm. The compensation calculation uses the weighted average method to correct each piece of data to ensure that each data point is appropriately corrected according to its deviation level to obtain the compensation result data. During the calculation of the compensation residual features, the compensation result data is compared with the original data, and the residual value of each data point is calculated.The residual value is the difference before and after compensation. In this way, the compensated residual feature is obtained. During the residual calculation process, the mean squared error (MSE) is used to quantify the compensated residual of each data point, and a data set of compensated residual features is obtained. When performing convergence analysis, based on the compensated residual feature, the data is analyzed. During the analysis process, a convergence detection algorithm, such as a common convergence determination criterion, is used to check whether the data residual tends to be stable. When the residual value is lower than the preset convergence threshold, it is determined to converge. The generated convergence index data is based on the change trend of the residual value and records the convergence process of the residual value. When performing compensated coverage feature analysis, based on the convergence index data, its coverage degree is analyzed in detail. The coverage degree analysis uses a graphical algorithm to calculate the coverage range of data points. During the analysis process, according to the distribution range of data points, the covered area is calculated, and covered feature data is generated. The covered feature data reflects the distribution of the compensated data points in the area. By graphically displaying the coverage area of each compensated point, the compensated coverage feature is generated. When mapping the compensated coverage feature to a specified test scenario, the compensated coverage feature is mapped to the specified test scenario. During mapping, the area covered by the compensation is corresponded to the parameters of the test scenario through a mapping algorithm to ensure that the compensated data can cover the area required for actual testing, and a mapped scenario matrix is generated. The mapped scenario matrix details the corresponding positions and test requirements of each compensated point in the test scenario. When decomposing the test requirements according to the mapped scenario matrix, first, the test scenarios in the mapped scenario matrix are disassembled, and the test conditions are allocated as needed. The generated requirement item data includes the specific test requirements of each test point. The decomposition process ensures that each test requirement is independently identified and listed in detail. When supplementing the test conditions for the requirement item data, first, according to the characteristics of the requirement item data, relevant test conditions are supplemented. The test conditions are extended according to the test requirements, and a test condition set is generated after supplementation. The test condition set includes all the detailed conditions of the requirement items. When constructing test cases based on the test condition set, detailed test cases are generated by combining the data in the test condition set. The test cases are constructed based on the previously collected compensation result data to ensure that the test cases can cover all test requirements and dynamically generate test task sequences according to different test conditions. The generation of the task sequence is optimized according to the dependency relationship between use cases to ensure that each test case can be executed in a specific order. When analyzing the dependency relationship of the test task sequence, the dependency relationships of all test cases are sorted out, and the sequence of each test task is analyzed to ensure that the test tasks can be allocated according to the dependency relationship, and the task dependency relationship is obtained. The dependency relationship analysis ensures that there are no conflicts or repeated executions during the execution of the test cases, and finally, a complete test task sequence is generated. When arranging and allocating the test task sequence according to the task dependency relationship, first, the test tasks are sorted according to the task dependency relationship,Ensure that the execution order of test tasks conforms to the dependency relationship, allocate test tasks through an algorithm, and finally generate supplementary test data. The supplementary test data is reasonably allocated according to the test task sequence to ensure the efficient execution of the test work.

[0164] The present invention also provides a test system for a chip simulation model, which is used to execute the test method for the chip simulation model as described above. The test system for the chip simulation model includes:

[0165] An eye diagram reconstruction module, which is used to collect a target chip simulation model; perform timing complete compensation on the target chip simulation model to generate jitter compensation data; perform eye diagram reconstruction on the jitter compensation data to obtain waveform eye diagram data);

[0166] A phase correction module, which is used to perform waveform amplitude normalization on the waveform eye diagram data to generate normalized data; perform timing phase correction on the normalized data to generate purified alignment data;

[0167] A modal analysis module, which is used to perform orthogonal mapping processing on the purified alignment data to generate orthogonal basis data; perform modal projection analysis based on the orthogonal basis data to generate main modal data;

[0168] A feature mapping module, which is used to perform vector mapping drive construction on the main modal data to generate kernel space data; perform manifold cognitive intelligent aggregation on the kernel space data to generate test coverage features;

[0169] A boundary analysis module, which is used to perform boundary-driven parsing on the test coverage features to generate density distribution data; perform stable feature precise calibration based on the density distribution data to obtain clustering boundary data;

[0170] A test optimization module, which is used to perform boundary accuracy tuning on the clustering boundary data to generate compensation result data; perform adaptive test sequence execution on the target chip simulation model based on the compensation result data to generate test case results.

[0171] The implementation of the eye diagram reconstruction module in this invention provides a basis for the real-time monitoring of the target chip signal. The generation of jitter compensation data improves the stability and reliability of the signal. The acquisition of waveform eye diagram data can intuitively display the signal quality and characteristics. The function of the phase correction module ensures the consistency of the signal amplitude. The generation of normalized data enhances the accuracy of subsequent analysis. The acquisition of purified alignment data provides a guarantee for the correction of the signal phase. The orthogonal mapping process of the modal analysis module provides an effective method for the extraction of signal features. The generation of the main modal data can effectively identify the main features and trends of the signal. The construction of kernel space data by the feature mapping module provides support for the in-depth analysis of signal features. The implementation of manifold cognitive intelligent aggregation improves the comprehensiveness of test coverage. The density distribution data of the boundary analysis module provides a quantitative basis for the stability analysis of features. The generation of clustering boundary data effectively identifies the distribution and changes of signal features. The boundary accuracy tuning ensures the reliability of test results. The generation of compensation result data provides a basis for the execution of the adaptive test sequence. The formation of test case results provides comprehensive support for the performance verification of the target chip. Overall, it improves the scientificity and efficiency of the chip simulation model test, and provides effective technical means and solutions for chip development and optimization.

[0172] Therefore, from any perspective, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the above description. Therefore, all changes falling within the meaning and scope of the equivalent elements of the application document are intended to be encompassed within the present invention.

[0173] The above description is only a specific implementation manner of the present invention, enabling those skilled in the art to understand or implement the present invention. Various modifications to these embodiments will be obvious to those skilled in the art. The general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to these embodiments shown herein, but rather will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A chip simulation model testing method, characterized in that: The following steps are involved: Step S1: Acquire a target chip simulation model; perform timing complete compensation on the target chip simulation model to generate jitter compensation data; Performing eye diagram reconstruction on the jitter compensation data to obtain waveform eye diagram data; Step S2: normalizing the waveform eye diagram data by waveform amplitude to generate normalized data; performing timing phase correction on the normalized data to generate purified alignment data; Step S3: performing orthogonal mapping processing on the cleaned aligned data to generate orthogonal basis data; performing modal projection analysis based on the orthogonal basis data to generate primary modal data; Step S4: vector mapping driven construction is performed on the main modal data to generate kernel space data; manifold cognitive intelligent aggregation is performed on the kernel space data to generate test coverage features; Step S5: performing boundary-driven analysis on the test coverage features to generate density distribution data; performing precise calibration of stable features based on the density distribution data to obtain cluster boundary data; Step S6: Optimize the boundary accuracy of cluster boundary data to generate compensation result data; An adaptive test sequence is executed on the target chip simulation model based on the compensation result data to generate test case results.

2. The chip simulation model testing method according to claim 1, characterized in that: Step S1 includes the following steps: Step S11: Acquire a target chip simulation model; perform waveform injection scanning on the target chip simulation model to obtain original waveform data; Step S12: reorganizing the original waveform data into time series structure to obtain time series deconstructed data; performing edge jitter detection on the time series deconstructed data to obtain edge jitter data; Step S13: performing jitter compensation on the timing deconstructed data according to the edge jitter data to generate jitter compensation data; Step S14: performing multi-domain sampling remapping on the jitter compensation data to obtain remapped data; performing clock recovery calibration on the remapped data to obtain calibrated data; Step S15: performing crosstalk elimination processing on the calibration data to generate waveform noise reduction data; performing eye diagram reconstruction on the waveform noise reduction data to obtain waveform eye diagram data.

3. The chip simulation model testing method according to claim 1, characterized in that: Step S2 includes the following steps: Step S21: performing distortion compensation processing on the waveform eye diagram data to obtain compensated eye diagram data; Step S22: normalizing the amplitude of the compensated eye diagram data to generate normalized data; Step S23: performing phase alignment calibration on the normalized data to obtain phase-aligned eye diagram data; Step S24: performing inter-symbol de-interference processing on the phase-aligned eye diagram data to generate cleansed aligned data.

4. The chip simulation model testing method according to claim 1, characterized in that: Step S3 includes the following steps: Step S31: performing modal decomposition processing on the purified aligned data to obtain modal feature data; Step S32: extracting feature vectors from the modal feature data to obtain vector group data; orthogonalizing the vector group data to generate orthogonal basis data; Step S33: performing feature projection simulation based on orthogonal basis data to generate a simulation projection matrix; Step S34: performing modal correlation analysis on the orthogonal basis data according to the simulated projection matrix to obtain modal correlation data; performing modal screening based on the modal correlation data to generate main modal data.

5. The chip simulation model testing method according to claim 1, characterized in that: Step S4 includes the following steps: Step S41: performing test vector mapping on the main modal data to obtain test sequence data; Step S42: performing kernel function transformation on the test sequence data to generate kernel space data; Step S43: performing manifold learning reconstruction on the kernel space data to generate manifold structure data; Step S44: performing coverage depth evaluation on the manifold structure data to generate depth index data; Step S45: Perform feature fusion processing on the depth index data to generate test coverage features.

6. The chip simulation model testing method according to claim 1, characterized in that: Step S5 includes the following steps: Step S51: perform limit condition inference on the test coverage feature to obtain limit coverage parameters; Step S52: performing density estimation clustering on the extreme coverage parameters to generate density distribution data; Step S53: performing inflection point positioning based on the density distribution data to obtain inflection point feature data; performing boundary positioning based on the inflection point feature data to generate boundary feature data; Step S54: Perform stability assessment based on the boundary feature data to generate stability index parameters; perform cluster optimization on the stability index parameters to obtain cluster boundary data.

7. The chip simulation model testing method according to claim 6, characterized in that: Step S53 includes the following steps: Performing gradient transformation processing on density distribution data to obtain gradient mapping data; performing local extreme value detection on gradient mapping data to generate extreme value point set data; The curvature of the extreme point set data is calculated to obtain the curvature characteristic parameters; the curvature characteristic parameters are threshold screened based on the preset inflection point curvature threshold to generate candidate inflection point data; Performing local window segmentation on the candidate inflection point data to obtain window feature data; performing morphological processing on the candidate inflection point data based on the window feature data to generate inflection point morphological features; Locate the inflection point center according to the inflection point morphological characteristics and generate inflection point feature data; Perform neighborhood expansion simulation on the inflection point feature data to obtain a simulated expansion area; perform directional analysis on the simulated expansion area to generate directional feature data; Perform continuity recognition on the directional feature data to obtain continuity index data; perform boundary tracking based on the continuity index data to generate initial boundary data; The smoothness of the initial boundary data is evaluated to obtain boundary smoothness parameters; based on the boundary smoothness parameters, the initial boundary data is precisely repositioned to generate boundary feature data.

8. The chip simulation model testing method according to claim 1, characterized in that: Step S6 includes the following steps: Step S61: performing similarity calculation on cluster boundary data to obtain boundary similarity data; Step S62: performing consistency comparison on the boundary similarity data and generating comparison result parameters; Step S63: performing deviation analysis and correction based on the comparison result parameters to generate compensation result data; performing candidate test allocation based on the compensation result data to generate supplementary test data; Step S64: performing an adaptive test sequence on the target chip simulation model according to the supplementary test data and the compensation result data to generate a test case result.

9. The chip simulation model testing method according to claim 8, characterized in that: Step S63 includes the following steps: The result parameters are compared and processed by difference mapping to obtain difference distribution data; based on the difference distribution data, regional segmentation is performed to obtain regional feature data; Prioritize regional feature data based on differential distribution data to generate deviation level data; Performing threshold classification on the deviation level data according to a preset deviation level threshold to generate classification mark data; performing compensation rule matching on the classification mark data to generate matching compensation rules; Perform compensation calculation on the classified labeled data based on the matching compensation rule to generate compensation result data; Perform residual calculation on the compensation result data to obtain compensation residual characteristics; perform convergence analysis on the compensation residual characteristics to generate convergence index data; Perform coverage analysis on the convergence index data to generate compensation coverage features; perform test scenario mapping based on the compensation coverage features to generate a mapping scenario matrix; Decompose the test requirements according to the mapping scenario matrix to obtain the requirement item data; supplement the test conditions for the requirement item data to generate a test condition set; Construct test cases based on the test condition set and generate a test task sequence; perform dependency analysis on the test task sequence to obtain task dependencies; The test task sequences are arranged and allocated according to task dependencies to generate supplementary test data.

10. A chip simulation model testing system, characterized in that: A chip simulation model testing method according to claim 1, wherein the chip simulation model testing system comprises: The eye diagram reconstruction module is used to collect the target chip simulation model; perform timing complete compensation on the target chip simulation model to generate jitter compensation data; perform eye diagram reconstruction on the jitter compensation data to obtain waveform eye diagram data); A phase correction module is used to normalize the waveform amplitude of the waveform eye diagram data to generate normalized data; perform timing phase correction on the normalized data to generate purified alignment data; The modal analysis module is used to perform orthogonal mapping on the purified alignment data to generate orthogonal basis data; perform modal projection analysis based on the orthogonal basis data to generate main modal data; The feature mapping module is used to perform vector mapping-driven construction on the main modal data to generate kernel space data; perform manifold cognitive intelligent aggregation on the kernel space data to generate test coverage features; The boundary analysis module is used to perform boundary-driven analysis on test coverage features to generate density distribution data; based on the density distribution data, the stable features are accurately calibrated to obtain cluster boundary data; The test optimization module is used to perform boundary accuracy tuning on cluster boundary data and generate compensation result data; based on the compensation result data, an adaptive test sequence is executed on the target chip simulation model to generate test case results.

Images