Waveform feature visualization method and system based on high and low temperature test

Abstract

The invention provides a waveform characteristic visualization method and system based on high and low temperature test, current time sequence data of a chip under high and low temperature test working conditions are collected through a Hall effect sensor, and after sliding window filtering and resampling preprocessing, a continuous and smooth current data fitting waveform is generated through local polynomial fitting; dividing the fitting waveform according to a time window, extracting wave crest and wave trough seed points, and obtaining a pulse width and period characteristic data set; an inherent time scale decomposition algorithm is called to strip waveform PR components, and a PR component density degree data set is obtained through discretization processing; fusing the time domain and frequency domain features to calculate a disorder coefficient, a trend coefficient and a warpage index, and constructing a multi-dimensional feature data set; through feature mapping, normalization processing, drawing of a time domain distribution curve and a thermodynamic diagram visualization chart, and generation of a waveform feature visualization data set, extraction and coupling of current waveform features are realized, and visual and reliable data are provided for chip high and low temperature test working condition analysis.

Description

Technical Field

[0001] This invention belongs to the field of chip measurement technology, and more specifically, relates to a waveform feature visualization method and system based on high and low temperature testing. Background Technology

[0002] With the rapid development of the semiconductor industry, the application scenarios of chips are becoming increasingly diversified, from consumer electronics to industrial control, aerospace and other fields, which put forward stringent requirements on the stability and reliability of chips in extreme high and low temperature environments.

[0003] High and low temperature testing, as a verification step before chip shipment, simulates the chip's operating state under different temperature conditions to determine whether its performance meets standards. Current mainstream chip high and low temperature testing methods mostly rely on a combination of a three-temperature sorter and a thermal hood. This method has certain advantages in testing large batches of chips with few varieties, but it has limitations when facing the testing needs of small batches and multiple varieties. Traditional methods process current timing data in a relatively coarse manner; the sawtooth effect caused by discrete sampling can mask the true characteristics of the waveform, affecting the accuracy of subsequent analysis and making it difficult to reflect the chip's state. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides a waveform feature visualization method and system based on high and low temperature testing. This addresses the issues in the prior art where traditional measurement methods handle current timing data in a coarse manner, and the sawtooth effect caused by discrete sampling obscures the true characteristics of the waveform, affecting the accuracy of subsequent analysis and making it difficult to reflect the chip status.

[0005] The purpose and effectiveness of the waveform feature visualization method and system based on high and low temperature testing of the present invention are achieved by the following specific technical means: A waveform feature visualization method based on high and low temperature testing includes the following steps: S1: Obtain chip current timing data sequence based on Hall effect sensor, preprocess and fit the chip current timing data sequence to obtain current data fitting waveform; S2: Based on the current data, the waveform is fitted and divided into time windows, and the peak and trough seed points of the waveform in each window are recorded to obtain the basic dataset of pulse width of each window and the periodic feature dataset of each window. S3: Based on the basic dataset of pulse width in the sub-window, perform waveform decomposition and discretization to obtain the dataset of density of PR components in the sub-window; S4: Analyze the disorder based on the basic dataset of pulse width in the sub-window and the dataset of waveform periodic features in the sub-window, enhance the sub-window PR component density dataset, and obtain the sub-window waveform pulse width sequence warp index dataset based on the analysis results and enhancement results. S5: Based on the segmented window waveform pulse width sequence warp index dataset, the segmented window PR component density dataset, and the segmented window waveform pulse width sequence warp index dataset, feature fusion is performed to obtain a waveform feature visualization dataset.

[0006] According to a preferred embodiment, the chip current timing data sequence acquired based on the Hall effect sensor includes: The chip is placed in a high and low temperature test equipment, a Hall effect sensor is deployed, a preset sampling interval is set, and the chip's operating current signal under high and low temperature test conditions is continuously collected to obtain the initial chip current timing data sequence. Based on the initial chip current timing data sequence, a sliding window filtering method is used to correct the spike outliers in the current signal. The current timing data sequence is obtained by unifying the time resolution of the signal through data resampling.

[0007] According to a preferred embodiment, the step of fitting the current data fitting waveform based on the chip current timing data sequence includes: Based on the current time series data sequence, local neighborhood window division and kernel function weight configuration are performed to generate a local fitting dataset with weight constraints. Obtain the low-order polynomial basis functions and weighted least squares solution rules in the preset polynomial fitting model; calculate the weight coefficients and basis function mapping values ​​of each sampling point in the local fitting dataset, generate the polynomial coefficient matrix, and perform point-by-point fitting on the discrete sampling points according to the polynomial coefficient matrix to obtain the local continuous fitting segment. The local continuous fitting segments are smoothed and interpolated to generate a global continuous waveform curve. The global continuous waveform curve is then input into the noise filtering layer of the model to eliminate the sawtooth effect caused by discrete sampling and obtain the current data fitting waveform.

[0008] According to a preferred embodiment, the process of dividing the waveform based on current data fitting into time windows and recording the peak and trough seed points of the waveform within each window to obtain the basic dataset of pulse width for each window and the dataset of periodic features of the waveform for each window includes: Based on the current data fitting waveform, the window length and sliding step size are set, and the current data fitting waveform is evenly divided into several continuous and non-overlapping time windows. The correspondence between each window and the waveform segment is established to obtain the sub-window current waveform dataset. Based on the segmented window current waveform dataset, the extreme value detection algorithm is used to identify the peak and trough seed points of the waveform in each window, generate peak and trough sequences in time order, calculate the time span between adjacent peaks, determine the pulse width parameters of each cycle, and obtain the basic dataset of segmented window pulse width. Based on the segmented current waveform dataset, the rising edge slope and falling edge slope of the waveform in each window are calculated, and the waveform amplitude change rate within the period is statistically analyzed to obtain the segmented waveform periodic feature dataset.

[0009] According to a preferred embodiment, the step of performing waveform decomposition and discretization based on a windowed pulse width dataset to obtain a windowed PR component density dataset includes: Based on the current data fitting waveform, the inherent time scale decomposition algorithm is called to decompose the current fitting waveform within the window into PR components of a preset number of layers, realize the layered stripping of waveform frequency domain features, and obtain a multi-scale PR component dataset. Based on the multi-scale PR component dataset, the continuous PR components are discretized using the equal-interval value method, so that the amount of discretized data is consistent with the amount of current time series data in the corresponding window. The absolute deviation of each PR component data from the component mean is calculated to obtain the PR component density dataset by window.

[0010] According to a preferred embodiment, the step of analyzing disorder based on a windowed pulse width base dataset and a windowed waveform periodicity feature dataset, enhancing based on a windowed PR component density dataset, and obtaining a windowed waveform pulse width sequence warp index dataset based on the analysis and enhancement results includes: Based on the basic dataset of pulse width in the window and the feature dataset of waveform period in the window, the mean and variance of all period pulse widths in each window are statistically analyzed. A preset adjustment parameter is introduced to avoid the denominator being zero. The absolute value of the relative deviation between each period pulse width and the mean is calculated. The mean of all absolute values ​​of deviation is obtained to obtain the disorder coefficient dataset of waveform pulse width in the window. Based on the density dataset of PR components in each window, the density difference of PR components at different levels in each window is calculated, the divergence matrix between components is constructed, and the divergence feature dataset of PR components in each window is obtained. The sequence warp index dataset is obtained by fusing the windowed waveform pulse width disorder coefficient dataset and the windowed PR component divergence feature dataset.

[0011] According to a preferred embodiment, the step of fusing the sequence warp index based on the segmented-window waveform pulse width disorder coefficient dataset and the segmented-window PR component divergence feature dataset to obtain the segmented-window waveform pulse width sequence warp index dataset includes: Based on the disorder coefficient dataset of waveform pulse width and the divergence feature dataset of PR component in the window, the disorder coefficients of each window are constructed into a time series in chronological order. The series is evenly divided into several sub-windows. The ratio of the range of disorder coefficients in each sub-window to the corresponding time interval is calculated. The mean of the ratios of all sub-windows is obtained to obtain the trend coefficient dataset of waveform pulse width sequence in the window. Based on the trend coefficient dataset and the density dataset of PR component in the segmented window waveform pulse width sequence, the trend coefficient and the density of PR component are multiplied window by window to achieve coupling of time domain and frequency domain features and obtain the warp index dataset of the segmented window waveform pulse width sequence.

[0012] According to a preferred embodiment, the step of fusing features based on the windowed waveform pulse width sequence warp index dataset, the windowed PR component density dataset, and the windowed waveform pulse width sequence warp index dataset to obtain a waveform feature visualization dataset includes: Based on the windowed waveform pulse width sequence warp index dataset, the windowed PR component density dataset, and the windowed waveform pulse width sequence warp index dataset, establish the mapping relationship between time windows and various feature parameters, construct the feature parameter topological correlation matrix, and obtain the original dataset of waveform features; The original waveform feature dataset is normalized using minimum and maximum normalization methods to map the numerical range of various feature parameters to the [0-1] interval, eliminating the dimensional differences between different feature dimensions, and obtaining the waveform feature dataset. Based on the waveform feature dataset, a visualization chart is drawn as the waveform feature visualization dataset.

[0013] A waveform feature visualization system based on high and low temperature testing, comprising: The current data acquisition module includes a Hall effect sensor for acquiring chip current timing data sequences; The waveform time-domain feature extraction module is connected to the current data acquisition module. The waveform time-domain feature extraction module is used to fit the waveform based on the current data, divide it into time windows, and record the peak and trough seed points of the waveform in each window to obtain the basic dataset of pulse width of each window and the periodic feature dataset of each window. The waveform frequency domain feature extraction module is connected to the current data acquisition module and is used to decompose and discretize the waveform based on the segmented pulse width basic dataset to obtain the segmented PR component density dataset. The feature quantization fusion module is connected to the waveform time-domain feature extraction module and the waveform frequency-domain feature extraction module, respectively. It is used to analyze the disorder based on the segmented window pulse width basic dataset and the segmented window waveform periodic feature dataset, enhance the density based on the segmented window PR component density dataset, and then obtain the segmented window waveform pulse width sequence warp index dataset based on the analysis results and enhancement results.

[0014] Compared with the prior art, the present invention has the following beneficial effects: 1. By correcting the spike outliers in the original current signal through the sliding window filtering method, and combining data resampling to unify the signal time resolution, the original data is preprocessed. At the same time, based on local neighborhood window partitioning, kernel function weight configuration, and weighted least squares solution rules, polynomial fitting is completed to map the discrete current time series data into a continuous and smooth waveform curve, eliminating the sawtooth effect caused by discrete sampling, restoring the true characteristics of the current waveform, and solving the problems of coarse current time series data processing and discrete sampling sawtooth effect masking the true characteristics of the waveform in traditional measurement methods.

[0015] 2. Based on time window segmentation, peak and trough seed points of the waveform are extracted. The PR component of the waveform is removed and its density is calculated through the inherent time scale decomposition algorithm to mine the frequency domain features of the waveform. On this basis, through disorder coefficient calculation, divergence feature enhancement and warp index fusion, the deep coupling of time domain and frequency domain features is achieved, realizing the comprehensive extraction of multi-dimensional features of current waveform. This breaks through the limitation of single feature extraction dimension of traditional methods, and allows the extracted feature parameters to correspond to the changes in the working state of the chip under high and low temperature environments in all aspects. Attached Figure Description

[0016] Figure 1 This is a flowchart of the steps of a waveform feature visualization method based on high and low temperature testing according to the present invention.

[0017] Figure 2 This is a block diagram illustrating the principle of a waveform feature visualization system based on high and low temperature testing according to the present invention. Detailed Implementation

[0018] The embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are used to illustrate the technical solutions of the present invention, but should not be used to limit the scope of protection of the present invention.

[0019] Example: As attached Figure 1 As shown: This invention provides a waveform feature visualization method based on high and low temperature testing, comprising the following steps: S1: Obtain chip current timing data sequence based on Hall effect sensor, preprocess and fit the chip current timing data sequence to obtain current data fitting waveform; In this embodiment, the chip is placed in the designated test station of the high and low temperature test equipment, a Hall effect sensor is deployed, and the sensor's detection end is placed close to the current path of the chip. A preset sampling interval is set, and the high and low temperature test equipment and current acquisition program are started to continuously acquire the chip's operating current signal at different temperature nodes under high and low temperature test conditions, thereby obtaining an initial chip current time series data sequence. Based on the initial chip current time series data sequence, a sliding window filtering method is used, with the window length and sliding step size parameters set, to traverse the entire current time series data sequence, correcting the spike anomalies in the current signal. Then, through data resampling, the time resolution of the signal is unified, eliminating the problem of inconsistent sampling intervals that may occur in different acquisition periods, and obtaining a standardized current time series data sequence.

[0020] Furthermore, based on the current time-series data sequence, a local neighborhood window is divided according to a preset neighborhood range parameter. A kernel function weight is configured for each sampling point within the neighborhood, with the highest weight for the sampling point at the center of the neighborhood and the weight decreasing sequentially for sampling points farther from the center. This generates a local fitting dataset with weight constraints. The low-order polynomial basis functions and weighted least squares solution rules in the preset polynomial fitting model are obtained. The sampling point timestamps and corresponding current values ​​in the local fitting dataset are substituted into the data. The weight coefficients and basis function mapping values ​​of each sampling point in the local fitting dataset are calculated. A system of linear equations is constructed and solved to generate a polynomial coefficient matrix. The discrete sampling points are fitted point by point according to the polynomial coefficient matrix to obtain a local continuous fitting segment corresponding to each local neighborhood. The local continuous fitting segments are smoothed and spliced ​​to eliminate the gaps between segments. Then, an interpolation point supplementation operation is performed to supplement a sufficient number of interpolation points between adjacent sampling points to generate a global continuous waveform curve. The global continuous waveform curve is input to the noise filtering layer of the model to filter out the sawtooth effect caused during discrete sampling and retain the core features of the current waveform, such as the true period and pulse width, to obtain the current data fitting waveform.

[0021] S2: Based on the current data, the waveform is fitted and divided into time windows. The peak and trough seed points of the waveform in each window are recorded to obtain the basic dataset of pulse width of each window and the periodic feature dataset of each window waveform.

[0022] In this embodiment, based on the current data fitting waveform, and combined with the typical periodic characteristics of the chip current waveform, a matching window length and sliding step size are set. The current data fitting waveform is uniformly divided into several continuous and non-overlapping time windows. Each time window is assigned a unique time interval identifier, and a one-to-one correspondence between each window and the corresponding waveform segment is established to obtain a sub-window current waveform dataset. This dataset provides an independent and complete analysis unit for subsequent window-by-window feature extraction. Based on the sub-window current waveform dataset, an extreme value detection algorithm is used to traverse the current fitting waveform data in each window, compare the current value of each data point with the current value of its adjacent data points, mark the positions where the current value is higher than the adjacent data points on both sides as peak seed points, and mark the positions where the current value is lower than the adjacent data points on both sides as trough seed points. The peak seed points and trough seed points in each window are sorted in chronological order to generate peak sequences and trough sequences corresponding to each window. The timestamps corresponding to two adjacent peak seed points in the peak sequence are extracted, and the difference between the two timestamps is calculated to obtain the time span between adjacent peaks. This time span is determined as the pulse width parameter of each period. The pulse width parameters of all windows are integrated to obtain the sub-window pulse width basic dataset.

[0023] Based on the segmented current waveform dataset, for each period of the waveform within each window, the rising edge is extracted from the trough seed point to the corresponding peak seed point, and the falling edge is extracted from the peak seed point to the next trough seed point. The rising edge slope is obtained by calculating the ratio of the change in current amplitude to the change in time at the rising edge, and the falling edge slope is obtained by calculating the ratio of the change in current amplitude to the change in time at the falling edge. At the same time, the waveform amplitude change rate within the period is obtained by calculating the ratio of the maximum change in current amplitude to the period duration. The rising edge slope, falling edge slope, and amplitude change rate data of all windows are integrated to obtain the segmented waveform periodic feature dataset.

[0024] S3: Based on the basic dataset of pulse width in the sub-window, perform waveform decomposition and discretization to obtain the dataset of density of PR components in the sub-window; In this embodiment, the inherent time-scale decomposition algorithm is invoked based on the current data fitting waveform. A multi-window current waveform dataset is input, and a preset number of PR component decomposition layers is set. The algorithm is then activated to decompose the current fitting waveform within each time window layer by layer, separating the waveform signal into PR components containing different frequency characteristics. Different PR component layers correspond to the high-frequency detail features and low-frequency trend features of the waveform, respectively, achieving layered separation of waveform frequency domain features and obtaining a multi-scale PR component dataset. Based on the multi-scale PR component dataset, the continuous PR components are discretized using an equal-interval sampling method, according to the sampling points of the corresponding window current time series data. The value interval is determined, and data points with equal intervals are selected on the time axis of each continuous PR component to ensure that the amount of discretized data is consistent with the amount of current time series data in the corresponding window, thus eliminating the impact of inconsistent dimensions of continuous data on subsequent calculations. Then, for each discretized PR component, the mean of all data points of that component is calculated. Each data point within the component is traversed, and the absolute deviation between that data point and the component mean is calculated. The average of all absolute deviations is calculated and used as the density index of the corresponding PR component. The density indices of PR components in all time windows are integrated to obtain the density dataset of PR components in each window.

[0025] S4: Analyze the disorder based on the basic dataset of pulse width in the sub-window and the dataset of waveform periodic features in the sub-window, enhance the sub-window PR component density dataset, and obtain the sub-window waveform pulse width sequence warp index dataset based on the analysis results and enhancement results. In this embodiment, based on the segmented window pulse width basic dataset and the segmented window waveform period feature dataset, the pulse width parameter and period feature parameter corresponding to each time window are traversed. The mean and variance of all period pulse widths within each window are calculated. A preset adjustment parameter greater than zero is introduced to avoid the case of zero denominator in subsequent calculations. The difference between each period pulse width and the mean pulse width within that window is calculated one by one. Then, the ratio of the absolute value of the difference to the sum of the variance and the adjustment parameter is calculated. The mean of this ratio for all periods within each window is obtained. This mean is used as the waveform pulse width disorder coefficient for the corresponding window. The disorder coefficient data of all time windows are integrated to obtain the segmented window waveform pulse width disorder coefficient dataset. This dataset can reflect the disorder coefficient of each time window. The stability of the pulse width of the current waveform within each window; based on the density dataset of PR components in each window, for the density values ​​of different levels of PR components within each time window, the absolute value of the difference between the density of any two levels of PR components is calculated, and the divergence matrix between the components is constructed according to the hierarchical order of the PR components. The element values ​​in the matrix correspond to the characteristic difference degree of different levels of PR components. The divergence matrix data of all time windows are integrated to obtain the divergence feature dataset of PR components in each window. This dataset can realize the enhancement processing of frequency domain features; based on the disorder coefficient dataset of waveform pulse width in each window and the divergence feature dataset of PR components in each window, the sequence warp index is fused to obtain the sequence warp index dataset of waveform pulse width in each window.

[0026] Furthermore, based on the windowed waveform pulse width disorder coefficient dataset and the windowed PR component divergence feature dataset, the disorder coefficients of each window are arranged according to the order of the time windows to construct a waveform pulse width disorder coefficient time series. The number of sub-windows is set according to the total length of the sequence, and the time series is evenly divided into several sub-windows. The difference between the maximum and minimum values ​​of the disorder coefficient in each sub-window is calculated, i.e., the range. Then, the ratio of this range to the time interval corresponding to the sub-window is calculated, and the mean of this ratio for all sub-windows is obtained. This mean is used as the waveform pulse width sequence trend coefficient for the corresponding window. The trend coefficient data of all time windows are integrated to obtain the windowed waveform pulse width sequence trend coefficient dataset. This dataset can reflect the changing trend of the disorder coefficient in the time dimension. Based on the windowed waveform pulse width sequence trend coefficient dataset and the windowed PR component density dataset, the trend coefficient corresponding to each time window is multiplied with the PR component density value window by window to achieve coupling of the time domain features and frequency domain features of the current waveform. The calculation results of all time windows are integrated to obtain the windowed waveform pulse width sequence warp index dataset.

[0027] S5: Based on the segmented window waveform pulse width sequence warp index dataset, the segmented window PR component density dataset, and the segmented window waveform pulse width sequence warp index dataset, feature fusion is performed to obtain a waveform feature visualization dataset.

[0028] In this embodiment, based on the windowed waveform pulse width sequence warp index dataset, the windowed PR component density dataset, and the windowed waveform pulse width sequence warp index dataset, a one-to-one mapping relationship is established between each time window and its corresponding warp index parameter and PR component density parameter, using the time window as the index dimension. A feature parameter topology association matrix is ​​constructed according to the time window order and feature parameter type. The row dimension of the matrix is ​​the unique identifier of each time window, the column dimension is the name of each type of feature parameter, and the elements in the matrix are the specific values ​​of the feature parameters under the corresponding time window. All feature data are integrated through this matrix to obtain the original waveform feature dataset, which can centrally present the multi-dimensional feature parameter information under each time window. The original waveform feature dataset is subjected to minimum and maximum normalization methods, traversing each feature parameter dimension. For each feature parameter dimension, all values ​​are determined, including the maximum and minimum values. All values ​​are mapped to the [0-1] interval to eliminate numerical differences caused by different dimensions, ensuring all feature parameters are within the same comparable range and obtaining a standardized waveform feature dataset. Based on this dataset, visualization charts are created according to the type of feature parameter. A time-domain distribution curve is plotted for the current-fitted waveform, a spatial heatmap is plotted for the density of the PR component, and a bar chart is plotted for the waveform pulse width sequence warp index. Simultaneously, a multi-feature parameter linkage analysis chart is created, establishing a time window correlation filtering function between the charts. All completed visualization charts are integrated to obtain a waveform feature visualization dataset. This dataset can transform abstract feature parameters into intuitive visualization content.

[0029] Please see as follows Figure 2 As shown, the present invention also provides a waveform feature visualization system based on high and low temperature testing, comprising: The current data acquisition module includes a Hall effect sensor for acquiring chip current timing data sequences; The waveform time-domain feature extraction module is connected to the current data acquisition module. The waveform time-domain feature extraction module is used to fit the waveform based on the current data, divide it into time windows, and record the peak and trough seed points of the waveform in each window to obtain the basic dataset of pulse width of each window and the periodic feature dataset of each window. The waveform frequency domain feature extraction module is connected to the current data acquisition module and is used to decompose and discretize the waveform based on the segmented pulse width basic dataset to obtain the segmented PR component density dataset. The feature quantization and fusion module is connected to the waveform time-domain feature extraction module and the waveform frequency-domain feature extraction module, respectively. It is used to analyze disorder based on the segmented-window pulse width base dataset and the segmented-window waveform periodic feature dataset, enhance the waveform based on the segmented-window PR component density dataset, and then obtain the segmented-window waveform pulse width sequence warp index dataset based on the analysis and enhancement results. The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A waveform feature visualization method based on high and low temperature testing, characterized in that, It includes the following steps: S1: Obtain chip current timing data sequence based on Hall effect sensor, preprocess and fit the chip current timing data sequence to obtain current data fitting waveform; S2: Based on the current data, the waveform is fitted and divided into time windows, and the peak and trough seed points of the waveform in each window are recorded to obtain the basic dataset of pulse width of each window and the periodic feature dataset of each window. S3: Based on the basic dataset of pulse width in the sub-window, perform waveform decomposition and discretization to obtain the dataset of density of PR components in the sub-window; S4: Analyze the disorder based on the basic dataset of pulse width in the sub-window and the dataset of waveform periodic features in the sub-window, enhance the sub-window PR component density dataset, and obtain the sub-window waveform pulse width sequence warp index dataset based on the analysis results and enhancement results. S5: Based on the segmented window waveform pulse width sequence warp index dataset, the segmented window PR component density dataset, and the segmented window waveform pulse width sequence warp index dataset, feature fusion is performed to obtain a waveform feature visualization dataset.

2. The waveform feature visualization method based on high and low temperature testing according to claim 1, characterized in that, The chip current timing data sequence obtained based on the Hall effect sensor includes: The chip is placed in a high and low temperature test equipment, a Hall effect sensor is deployed, a preset sampling interval is set, and the chip's operating current signal under high and low temperature test conditions is continuously collected to obtain the initial chip current timing data sequence. Based on the initial chip current timing data sequence, a sliding window filtering method is used to correct the spike outliers in the current signal. The current timing data sequence is obtained by unifying the time resolution of the signal through data resampling.

3. The waveform feature visualization method based on high and low temperature testing according to claim 2, characterized in that, The process of obtaining the current data fitting waveform based on the chip current timing data sequence includes: Based on the current time series data sequence, local neighborhood window division and kernel function weight configuration are performed to generate a local fitting dataset with weight constraints. Obtain the low-order polynomial basis functions and weighted least squares solution rules in the preset polynomial fitting model; calculate the weight coefficients and basis function mapping values ​​of each sampling point in the local fitting dataset, generate the polynomial coefficient matrix, and perform point-by-point fitting on the discrete sampling points according to the polynomial coefficient matrix to obtain the local continuous fitting segment. The local continuous fitting segments are smoothed and interpolated to generate a global continuous waveform curve. The global continuous waveform curve is then input into the noise filtering layer of the model to eliminate the sawtooth effect caused by discrete sampling and obtain the current data fitting waveform.

4. The waveform feature visualization method based on high and low temperature testing according to claim 1, characterized in that, The waveform fitting based on current data is divided into time windows, and the peak and trough seed points of the waveform within each window are recorded to obtain the basic dataset of pulse width for each window and the dataset of periodic features of the waveform for each window, including: Based on the current data fitting waveform, the window length and sliding step size are set, and the current data fitting waveform is evenly divided into several continuous and non-overlapping time windows. The correspondence between each window and the waveform segment is established to obtain the sub-window current waveform dataset. Based on the segmented window current waveform dataset, the extreme value detection algorithm is used to identify the peak and trough seed points of the waveform in each window, generate peak and trough sequences in time order, calculate the time span between adjacent peaks, determine the pulse width parameters of each cycle, and obtain the basic dataset of segmented window pulse width. Based on the segmented current waveform dataset, the rising edge slope and falling edge slope of the waveform in each window are calculated, and the waveform amplitude change rate within the period is statistically analyzed to obtain the segmented waveform periodic feature dataset.

5. The waveform feature visualization method based on high and low temperature testing according to claim 1, characterized in that, The waveform decomposition and discretization based on the segmented pulse width dataset to obtain the segmented PR component density dataset includes: Based on the current data fitting waveform, the inherent time scale decomposition algorithm is called to decompose the current fitting waveform within the window into PR components of a preset number of layers, realize the layered stripping of waveform frequency domain features, and obtain a multi-scale PR component dataset. Based on the multi-scale PR component dataset, the continuous PR components are discretized using the equal-interval value method, so that the amount of discretized data is consistent with the amount of current time series data in the corresponding window. The absolute deviation of each PR component data from the component mean is calculated to obtain the PR component density dataset by window.

6. The waveform feature visualization method based on high and low temperature testing according to claim 1, characterized in that, The analysis of disorder based on the segmented window pulse width base dataset and the segmented window waveform period feature dataset, the enhancement based on the segmented window PR component density dataset, and the acquisition of the segmented window waveform pulse width sequence warp index dataset based on the analysis and enhancement results include: Based on the basic dataset of pulse width in the window and the feature dataset of waveform period in the window, the mean and variance of all period pulse widths in each window are statistically analyzed. A preset adjustment parameter is introduced to avoid the denominator being zero. The absolute value of the relative deviation between each period pulse width and the mean is calculated. The mean of all absolute values ​​of deviation is obtained to obtain the disorder coefficient dataset of waveform pulse width in the window. Based on the density dataset of PR components in each window, the density difference of PR components at different levels in each window is calculated, the divergence matrix between components is constructed, and the divergence feature dataset of PR components in each window is obtained. The sequence warp index dataset is obtained by fusing the windowed waveform pulse width disorder coefficient dataset and the windowed PR component divergence feature dataset.

7. The waveform feature visualization method based on high and low temperature testing according to claim 6, characterized in that, The step of fusing the sequence warp index based on the segmented-window waveform pulse width disorder coefficient dataset and the segmented-window PR component divergence feature dataset to obtain the segmented-window waveform pulse width sequence warp index dataset includes: Based on the disorder coefficient dataset of waveform pulse width and the divergence feature dataset of PR component in the window, the disorder coefficients of each window are constructed into a time series in chronological order. The series is evenly divided into several sub-windows. The ratio of the range of disorder coefficients in each sub-window to the corresponding time interval is calculated. The mean of the ratios of all sub-windows is obtained to obtain the trend coefficient dataset of waveform pulse width sequence in the window. Based on the trend coefficient dataset and the density dataset of PR component in the segmented window waveform pulse width sequence, the trend coefficient and the density of PR component are multiplied window by window to achieve coupling of time domain and frequency domain features and obtain the warp index dataset of the segmented window waveform pulse width sequence.

8. The waveform feature visualization method based on high and low temperature testing according to claim 1, characterized in that, The waveform feature visualization dataset obtained by fusing features from the windowed waveform pulse width sequence warp index dataset, the windowed PR component density dataset, and the windowed waveform pulse width sequence warp index dataset includes: Based on the windowed waveform pulse width sequence warp index dataset, the windowed PR component density dataset, and the windowed waveform pulse width sequence warp index dataset, establish the mapping relationship between time windows and various feature parameters, construct the feature parameter topological correlation matrix, and obtain the original dataset of waveform features; The original waveform feature dataset is normalized using minimum and maximum normalization methods to map the numerical range of various feature parameters to the [0-1] interval, eliminating the dimensional differences between different feature dimensions, and obtaining the waveform feature dataset. Based on the waveform feature dataset, a visualization chart is drawn as the waveform feature visualization dataset.

9. A waveform feature visualization system based on high and low temperature testing, characterized in that: Including, The current data acquisition module includes a Hall effect sensor for acquiring chip current timing data sequences; The waveform time-domain feature extraction module is connected to the current data acquisition module. The waveform time-domain feature extraction module is used to fit the waveform based on the current data, divide it into time windows, and record the peak and trough seed points of the waveform in each window to obtain the basic dataset of pulse width of each window and the periodic feature dataset of each window. The waveform frequency domain feature extraction module is connected to the current data acquisition module and is used to decompose and discretize the waveform based on the segmented pulse width basic dataset to obtain the segmented PR component density dataset. The feature quantization fusion module is connected to the waveform time-domain feature extraction module and the waveform frequency-domain feature extraction module, respectively. It is used to analyze the disorder based on the segmented window pulse width basic dataset and the segmented window waveform periodic feature dataset, enhance the density based on the segmented window PR component density dataset, and then obtain the segmented window waveform pulse width sequence warp index dataset based on the analysis results and enhancement results.

Images