Product detection method and apparatus, storage medium, and electronic device

Abstract

The application provides a product detection method and device, a storage medium and an electronic device. The product detection method comprises: obtaining detection data of one or more test parameters of a target product; for each test parameter, obtaining a change trend of the test parameter by linear fitting based on the detection data of the test parameter, and generating an evaluation index for quantifying the degree of linearity of the change trend; selecting, from the plurality of test parameters, a test parameter with an evaluation index greater than or equal to a first preset value as a suspicious parameter; and identifying whether the suspicious parameter is an abnormal parameter of the product according to the change trend of the suspicious parameter and a preset fluctuation range. Thus, the accuracy of product detection is improved.

Description

Product testing methods, apparatus, storage media and electronic equipment Technical Field

[0001] This invention relates to the field of product testing technology, and in particular to a product testing method, apparatus, storage medium, and electronic device. Background Technology

[0002] Two key issues exist in product testing: first, oscillation issues, characterized by drastic fluctuations in test parameter data within a short period; and second, edge-drift issues, where some test parameter data gradually deviates from the normal range over time or across production batches, exhibiting a slow drift (e.g., gradually increasing or decreasing). Edge-drift issues are particularly insidious, impacting not only product reliability but also accumulating into potential quality risks. Traditional SPC (Statistical Process Control) methods struggle to effectively capture edge drift, especially when dealing with a large number of test parameters, where both the efficiency and accuracy of abnormal parameter detection face bottlenecks.

[0003] Currently, the industry commonly uses moving average control charts, traditional time series models, or traditional machine learning algorithms to monitor detection data in order to address edge cases. However, these methods have significant limitations in multi-parameter testing scenarios, making it difficult to filter out abnormal test parameters from multiple parameters. This often leads to problems such as detection lag and high false alarm rates when dealing with massive amounts of test data. Summary of the Invention

[0004] In view of the above, embodiments of this application provide a product testing method, apparatus, storage medium, and electronic device to solve at least one problem existing in the background art.

[0005] In a first aspect, one embodiment of this application provides a product testing method, the method comprising:

[0006] To obtain test data for one or more test parameters of the target product;

[0007] For each of the test parameters, based on the detection data of the test parameters, the changing trend of the test parameters is obtained through linear fitting, and an evaluation index is generated to quantify the linearity of the changing trend.

[0008] From the multiple test parameters, test parameters whose evaluation index is greater than or equal to a first preset value are selected as suspicious parameters;

[0009] Based on the changing trend of the suspicious parameters and the preset fluctuation range, identify whether the suspicious parameters are abnormal parameters of the product.

[0010] In conjunction with the first aspect of this application, in an optional embodiment, identifying whether the suspicious parameter is an abnormal parameter of the product based on the changing trend of the suspicious parameter and a preset fluctuation range includes:

[0011] Based on the changing trend of the suspected parameter, obtain the deviation of the change magnitude of the suspected parameter within a specified time window from the preset fluctuation range;

[0012] Based on whether the deviation of the suspected parameter exceeds a second preset value, the suspected parameter is identified as an abnormal parameter of the product.

[0013] In conjunction with the first aspect of this application, in an optional embodiment, obtaining the deviation of the change magnitude of the suspicious parameter within a specified time window relative to a preset fluctuation range based on the changing trend of the suspicious parameter includes:

[0014] Based on the changing trend of the suspicious parameter, obtain the first predicted value of the suspicious parameter at the start time and the second predicted value at the end time of the specified time window;

[0015] Based on the difference between the first predicted value and the second predicted value, the magnitude of change of the suspicious parameter is obtained;

[0016] Based on the change amplitude, the upper limit and lower limit of the preset fluctuation range, the deviation of the change amplitude relative to the preset fluctuation range is obtained.

[0017] In conjunction with the first aspect of this application, in an optional embodiment, obtaining the changing trend of the test parameter based on the detection data of the test parameter through linear fitting for each of the test parameters includes:

[0018] For each of the test parameters, a time-series-based recurrent neural network is used to extract features from the detection data of the test parameters to obtain the time-series features of the detection data.

[0019] Based on the temporal characteristics of the detection data, the changing trend of the test parameters is obtained through linear fitting.

[0020] In conjunction with the first aspect of this application, in an optional embodiment, when abnormal parameters of the product are identified, the method further includes:

[0021] Clustering is performed on the detection data of the abnormal parameters under different product dimensions, and the product dimensions with abnormal data are identified based on the clustering results corresponding to each product dimension.

[0022] In conjunction with the first aspect of this application, in an optional embodiment, the clustering result corresponding to each product dimension includes at least two data clusters; the step of identifying product dimensions with data anomalies based on the clustering result corresponding to each product dimension includes:

[0023] For each product dimension, the similarity between different data clusters in the clustering results is calculated.

[0024] The clustering result corresponding to the minimum similarity is determined as the target clustering result, and the product dimension corresponding to the target clustering result is determined as the product dimension with data anomalies.

[0025] In conjunction with the first aspect of this application, in an optional embodiment, the clustering process is a clustering process using an unsupervised Gaussian mixture model; wherein each data cluster in the clustering result is modeled as having a normal distribution; the calculation of the similarity between different data clusters in the clustering result includes:

[0026] Based on the probability distribution characteristics of each data cluster in the clustering results, the similarity between different data clusters in the clustering results is calculated.

[0027] In conjunction with the first aspect of this application, in an optional embodiment, the method further includes:

[0028] Generate and visualize the cumulative distribution function curve corresponding to each product dimension to help users identify the product dimensions with abnormal data.

[0029] In conjunction with the first aspect of this application, in an optional embodiment, the product dimension is used to perform attribution analysis on the detection data of the test parameters, and different product dimensions correspond to different physical units or different process units in the manufacturing process of the product.

[0030] Secondly, embodiments of this application provide a product testing device, comprising:

[0031] The acquisition module is used to acquire detection data of one or more test parameters of the target product;

[0032] The linear fitting module is used to obtain the changing trend of each test parameter based on the detection data of the test parameter through linear fitting, and generate an evaluation index for quantifying the linearity of the changing trend.

[0033] The first identification module is used to filter out test parameters from the plurality of test parameters whose evaluation index is greater than or equal to a first preset value, as suspicious parameters;

[0034] The second identification module is used to identify whether the suspicious parameter is an abnormal parameter of the product based on the changing trend of the suspicious parameter and the preset fluctuation range.

[0035] Thirdly, embodiments of this application provide a computer-readable storage medium storing instructions that, when executed by a processor of an electronic device, enable the electronic device to perform the product testing method described in any one of the first aspects.

[0036] Fourthly, embodiments of this application provide an electronic device, the electronic device comprising:

[0037] processor;

[0038] Memory used to store executable instructions for a computer;

[0039] The processor is configured to execute the computer-executable instructions to perform the product testing method according to any one of the first aspects.

[0040] The product testing method, apparatus, storage medium, and electronic device provided in this application generate evaluation indicators that quantify the linearity of the changing trend of each test parameter. This transforms slow drifts, which are difficult to judge intuitively, into intuitive, interpretable, and comparable values. As a result, suspicious parameters with significant changing trends can be quickly and automatically screened from multiple test parameters. Then, the changing trend of the suspicious parameters is further judged by combining a preset fluctuation range to identify whether they are indeed abnormal parameters. Thus, through a dual judgment mechanism, abnormal parameters are automatically, efficiently, and accurately screened from multiple test parameters, reducing reliance on engineers' experience and achieving early and accurate identification of potential product quality problems. At the same time, it can reduce the cost of delayed problem discovery and provide reliable data support for engineers' analysis, judgment, and decision-making.

[0041] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0042] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0043] Figure 1 is a flowchart of the product testing method provided in an embodiment of this application;

[0044] Figure 2 is a schematic diagram of linear fitting;

[0045] Figure 3 is a schematic diagram of the Long Short-Term Memory network model;

[0046] Figure 4 is a schematic diagram of the data cluster distribution;

[0047] Figure 5 is a schematic diagram of the detection data and temporal features in a specific example of this application;

[0048] Figure 6 is a schematic diagram of the time series features and the fitted trend line in a specific example of this application;

[0049] Figure 7 is a schematic diagram of the cumulative distribution function curve of the batch dimension in a specific example of this application;

[0050] Figure 8 is a schematic diagram of the cumulative distribution function curve of the wafer dimension in a specific example of this application;

[0051] Figure 9 is a schematic diagram of the cumulative distribution function curve of the detection point dimension in a specific example of this application;

[0052] Figure 10 is a schematic diagram of the product testing device provided in an embodiment of this application;

[0053] Figure 11 is a schematic diagram of the structure of the electronic device provided in the embodiment of this application. Detailed Implementation

[0054] To make the technical solutions and beneficial effects of the present invention more apparent and understandable, the technical solutions in the embodiments of this application are clearly and completely described below by listing specific examples. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0055] To make the technical solution and beneficial effects of the present invention more apparent and understandable, a detailed description is provided below by listing specific embodiments. The accompanying drawings are not necessarily drawn to scale, and local features may be enlarged or reduced to more clearly show the details of the local features; unless otherwise defined, the technical and scientific terms used herein have the same meanings as those in the technical field to which this application pertains.

[0056] The embodiments in this application are not exhaustive, but merely illustrative of some embodiments, and are not intended to limit the scope of protection of this disclosure. Unless otherwise specified, each step in a particular embodiment can be implemented as an independent embodiment, and the steps can be arbitrarily combined. For example, a solution after removing some steps in a particular embodiment can also be implemented as an independent embodiment, and the order of the steps in a particular embodiment can be arbitrarily interchanged. Furthermore, the optional implementation methods in a particular embodiment can be arbitrarily combined; moreover, the embodiments can be arbitrarily combined, for example, some or all steps of different embodiments can be arbitrarily combined, and a particular embodiment can be arbitrarily combined with the optional implementation methods of other embodiments.

[0057] In each embodiment of this application, unless otherwise specified or in case of logical conflict, the terminology and / or descriptions of the embodiments are consistent and can be referenced by each other. Technical features in different embodiments can be combined to form new embodiments based on their inherent logical relationships.

[0058] The terminology used in the embodiments of this application is for the purpose of describing particular embodiments only and is not intended to limit the scope of this disclosure.

[0059] In the embodiments of this application, unless otherwise stated, elements expressed in the singular, such as "a", "an", "the", "the", "the", "the", "the", "this", etc., can mean "one and only one", or "one or more", "at least one", etc.

[0060] In the embodiments of this application, "multiple" refers to two or more.

[0061] In some embodiments, the terms “at least one (at least one, at least one item, at least one),” “one or more,” “multiple”, etc., may be used interchangeably.

[0062] In some embodiments, the notation "A or B" may include the following technical solutions, depending on the situation: in some embodiments, A (execution of A regardless of B); in some embodiments, B (execution of B regardless of A); in some embodiments, selective execution from A and B (A and B are selectively executed). The same applies when there are more branches such as A, B, and C.

[0063] The prefixes "first," "second," etc., used in the embodiments of this application are merely for distinguishing different descriptive objects and do not impose restrictions on the position, order, priority, value, or content of the descriptive objects. For descriptions of the descriptive objects, please refer to the contextual descriptions in the embodiments. The use of prefixes should not constitute unnecessary restrictions. For example, if the descriptive object is "level," then the ordinal numbers preceding "level" in "first level" and "second level" do not restrict the priority between "levels."

[0064] In some embodiments, “including A,” “containing A,” “for indicating A,” and “carrying A” can be interpreted as directly carrying A or indirectly indicating A.

[0065] In some embodiments, terms such as “in the case of,” “when,” “when,” “if,” “if,” etc., can be used interchangeably.

[0066] In some embodiments, the terms “greater than,” “greater than or equal to,” “not less than,” “more than,” “more than or equal to,” “not less than,” “higher than,” “higher than or equal to,” “not lower than,” and “above” can be used interchangeably, as can the terms “less than,” “less than or equal to,” “not greater than,” “less than,” “less than or equal to,” “not more than,” “lower than,” “lower than or equal to,” “not higher than,” and “below”.

[0067] In some embodiments, devices, etc., can be interpreted as physical or virtual, and their names are not limited to those described in the embodiments. Terms such as “device”, “equipment”, “circuit”, “network element”, “node”, “function”, “unit”, “component”, “system”, “network”, “chip”, “chip system”, “entity”, and “subject” can be used interchangeably.

[0068] In related technologies, edge cases of test parameters are mainly addressed through methods such as moving average control charts, traditional time series models, and traditional machine learning algorithms. Specifically, the Exponentially Weighted Moving Average (EWMA) method smooths historical data and assigns higher weights to recent data to detect minor drifts; the Autoregressive Integrated Moving Average (ARIMA) model captures linear trends, seasonality, and random fluctuations in time series through a combination of autoregression, differencing, and moving averages; and traditional machine learning algorithms predict future trends by assuming a linear relationship between data drift and time or machine status, using algorithmic fitting coefficients. However, these methods struggle to accurately filter out abnormal parameters from a variety of test parameters, resulting in problems such as detection lag and high false alarm rates.

[0069] Based on this, this application provides a product testing method, as shown in Figure 1, the product testing method includes:

[0070] Step S1: Obtain test data for one or more test parameters of the target product;

[0071] Step S2: For each test parameter, based on the detection data of the test parameter, obtain the changing trend of the test parameter through linear fitting, and generate an evaluation index for quantifying the linearity of the changing trend;

[0072] Step S3: Select test parameters whose evaluation index is greater than or equal to the first preset value from multiple test parameters as suspicious parameters;

[0073] Step S4: Based on the changing trend of the suspicious parameters and the preset fluctuation range, identify whether the suspicious parameters are abnormal parameters of the product.

[0074] This application generates an evaluation index for each test parameter to quantify the linearity of its changing trend. This transforms the slow drift, which is difficult to judge intuitively, into an intuitive, interpretable, and comparable value. This allows for the rapid and automatic screening of suspicious parameters with significant changing trends from a variety of test parameters. Furthermore, a second judgment is made on the changing trend of the suspicious parameters in conjunction with a preset fluctuation range to identify whether they are indeed abnormal parameters. Thus, through a dual judgment mechanism, abnormal parameters are automatically, efficiently, and accurately screened from a variety of test parameters, reducing reliance on engineers' experience and enabling early and accurate identification of potential product quality issues. At the same time, it can reduce the cost of delayed problem discovery and provide reliable data support for engineers' analysis, judgment, and decision-making.

[0075] First, perform step S1 to obtain test data for one or more test parameters of the target product.

[0076] Understandably, test parameters refer to quantifiable indicators used to measure the performance, specifications, or quality characteristics of a product during the production process or finished product testing stage. Typically, multiple parameters need to be tested for a given product. For example, in the semiconductor field, test parameters may include, but are not limited to, at least two of the following: wafer thickness, warpage, thin film refractive index, and electrical parameters such as chip threshold voltage, saturation current, and leakage current; as another example, in the battery field, test parameters may include, but are not limited to, at least two of the following: cell voltage, internal resistance, capacity, and capacity retention rate in cycle life testing; and as yet another example, in the mechanical manufacturing field, test parameters may include, but are not limited to, at least two of the following: dimensional tolerances, surface roughness, and hardness.

[0077] The detection data includes time-series data related to the test parameters, where each data point is associated with a corresponding detection time.

[0078] In some embodiments, after performing step S1, the product testing method may further include: preprocessing the testing data, wherein the preprocessing includes, but is not limited to, data cleaning (e.g., handling missing values ​​and outliers), normalization, etc., to facilitate more accurate subsequent processing of the testing data.

[0079] Next, step S2 is executed. For each test parameter, based on the detection data of the test parameter, the changing trend of the test parameter is obtained through linear fitting, and an evaluation index for quantifying the linearity of the changing trend is generated.

[0080] Therefore, an evaluation index is generated for each test parameter to quantify the linearity of its changing trend, transforming the slow drift that is difficult to judge intuitively into an intuitive, interpretable and comparable value. This allows for the rapid and automatic screening of suspicious parameters with significant changing trends from a variety of test parameters.

[0081] The trend of the test parameter includes the drift trend of the test parameter.

[0082] In some embodiments, the evaluation metric is used to quantify the linearity of the trend of change. For example, the evaluation metric may include the coefficient of determination (R-squared, R²) or a linear correlation coefficient related to the goodness of fit of the model, such as the Pearson correlation coefficient. The coefficient of determination can be used to assess the goodness of fit of a linear regression, representing the proportion of the variation in the dependent variable (Y) that can be explained by the independent variable (X) through a linear relationship. Its value ranges from 0 to 1. For example, when R² = 0.8, it means that 80% of the variation in the dependent variable can be explained by the independent variable in the model through a linear relationship, while the remaining 20% ​​may be caused by random noise or other factors not included in the model. Generally speaking, the larger the coefficient of determination R², the higher the degree of linear correlation, and the better it indicates that the test parameter has a linear trend that drifts with the detection time.

[0083] However, when the true drift pattern of test parameters (such as semiconductor test parameters) is nonlinear, an excessively high coefficient of determination (R²) is caused by overfitting of the model, making it difficult to effectively capture the true nonlinear drift pattern in the test parameters.

[0084] The Pearson correlation coefficient measures the strength and direction of the correlation between two variables. In linear regression, R² equals the square of the Pearson correlation coefficient (r) between the independent and dependent variables. Therefore, a very high |r| value usually corresponds to a very high R². Including an excessively large absolute value of the Pearson correlation coefficient in a linear regression model can also lead to an excessively large R², increasing the risk of overfitting. In some embodiments, obtaining the trend of change of the test parameters through linear fitting and generating an evaluation index to quantify the linearity of the trend may include:

[0085] The linear regression algorithm is used to fit the detection data of each test parameter to obtain the fitting trend line of each test parameter; the determination coefficient of each fitting trend line is calculated, and the determination coefficient is used as an evaluation index to quantify the linearity of the change trend.

[0086] For example, the formula for the linear regression algorithm is shown in formula (1):

[0087] (1)

[0088] Where y represents the detection data at time t, t is the detection time, β1 is the regression coefficient, and β0 is the intercept. This is the error term.

[0089] For example, as shown by the data points in Figure 2, each test parameter has a set of detection data (t). i y i For example, β1 and β0 can be calculated using the mean of t and the mean of y according to the least squares method, thus obtaining the fitted trend line (the sloping straight line in Figure 2). The formula for the fitted trend line is given in formula (2):

[0090] (2)

[0091] Among them, V P(t) This represents the predicted value at time t. A trend line is fitted to reflect the changing trend of the data points.

[0092] Understandably, Figure 2 schematically shows the fitting trend line corresponding to one test parameter. For multiple test parameters, fitting trend lines corresponding to each test parameter can be obtained.

[0093] Next, the coefficient of determination R is calculated according to formulas (1) and (2). 2 It should be noted that the coefficient of determination R... 2 The calculation method is well known to those skilled in the art, and will not be elaborated upon here.

[0094] Semiconductor parameter drift often exhibits nonlinear and multimodal characteristics (such as slow early drift, mid-term acceleration, and late-term saturation), which traditional methods struggle to capture comprehensively. In this embodiment, detection data is used as the dependent variable, and detection time is used as the independent variable, utilizing the coefficient of determination R... 2 It can assess the linear correlation between test data and test time, with a coefficient of determination R. 2A higher coefficient of determination (R²) indicates a stronger linear correlation, suggesting a more linear trend in the drift of the test parameter over time. However, an excessively high R² may be due to overfitting, making it difficult to effectively capture the true nonlinear drift in semiconductor parameters. To reduce such overfitting errors, this application filters out test parameters whose evaluation index is greater than or equal to a first preset value from multiple test parameters as suspicious parameters. This allows for the selection of suspicious parameters based on the R². 2 Whether it is greater than or equal to a first preset value, thus preliminarily screening out suspected abnormal test parameters. In some embodiments, for each test parameter, based on the detection data of the test parameter, the changing trend of the test parameter is obtained through linear fitting, which may include:

[0095] For each test parameter, a time-series-based recurrent neural network is used to extract features from the detection data of the test parameter to obtain the time-series features of the detection data; based on the time-series features of the detection data, the changing trend of the test parameter is obtained through linear fitting.

[0096] The drift of product test parameters often exhibits complex nonlinear and multimodal characteristics, such as slow drift in the early stage, acceleration in the middle stage, and saturation in the later stage. Traditional time series models, represented by moving averages and ARIMA, as well as traditional machine learning algorithms, are based on linear assumptions and excel at capturing linear trends and seasonal variations with fixed periods, but they cannot effectively model the nonlinear dynamic characteristics commonly found in parameter drift. Traditional time series models have stringent requirements for data integrity and stationarity, and are extremely sensitive to missing values ​​or isolated outliers. To ensure model operation, complex preprocessing (such as interpolation and filtering) is usually required. This process is not only cumbersome but may also introduce artificial noise or cause loss of original information, thus masking true abnormal signals. Traditional machine learning algorithms, when faced with high-dimensional data or complex models, easily capture noise, leading to decreased generalization ability and the risk of overfitting. Therefore, the product testing methods used in these technologies struggle to accurately analyze the drift trend of test parameters, resulting in insufficient monitoring capabilities for potential product quality issues.

[0097] This application utilizes a recurrent neural network to learn and extract time-series features that characterize the complex changes in parameters (including nonlinear drift) from the original time-series data, overcoming the linear limitations of traditional time-series models; it can also automatically filter out random noise, reduce reliance on complex preprocessing, and reduce trend misjudgments caused by data noise interference.

[0098] Furthermore, recurrent neural networks can include Long Short-Term Memory (LSTM) networks. LSTM, through its unique gating mechanism (input gate, forget gate, output gate), can selectively memorize and transmit information, thereby learning long-term dependencies. The resulting temporal features are refined core information that better represents the long-term variation patterns of parameters. Linear fitting based on this yields a fitting trend line that better reflects the true, potential drift direction, making the final linearity evaluation index and anomaly detection based on a preset fluctuation range more accurate and reliable.

[0099] Suppose the input detection data sequence X = {x1, x2, ..., xt}, where xt is the detection data at the current time step.

[0100] Referring to Figure 3, the LSTM calculation process for each time step can be represented as follows:

[0101] (3)

[0102] (4)

[0103] (5)

[0104] (6)

[0105] (7)

[0106] (8)

[0107] Among them, i t W serves as the input gate for the current time step, determining how much of the detection data from the current time step can be written into the cell state. xi It is the first input gate weight matrix and W hi It is the weight matrix of the second input gate, x t This is the detection data at the current time step, h t-1 b is the hidden state of the previous time step. i This is the input gate bias, σ is the sigmoid function used to compress the input value to between 0 and 1; f t W is the forget gate for the current time step, used to determine how much of the detection data at that time step should be discarded from the cell state. xf It is the first forget gate weight matrix and W hf This is the first forget gate weight matrix, b f It is a forgetting gate bias; For the candidate cell states at the current time step, the tanh function compresses their values ​​to between -1 and 1.xc It is the first cell state weight matrix and W hc It is the second cell state rematrix, b c It is cell state bias, c t c represents the cell state at the current time step. t-1 This refers to the cell state at the previous time step; t W is the output gate for the current time step, used to determine which part of the cell state will be output as the hidden state. xo It is the first output gate weight matrix and W ho It is the first output gate weight matrix, b o It is the output gate bias; h t The hidden state at the current time step.

[0108] The detection data for each detection parameter includes time-series data related to the test parameter, which serves as the input to the recurrent neural network. The detection data x at the current time step... t After input, combine the hidden state h from the previous time step. t-1 The input gate i for the current time step is determined according to formulas (3), (4), (5), and (6). t The forget gate at the current time step f t Output gate of the current time step t and the candidate cell state at the current time step Calculate; then, according to formula (7), calculate the forget gate f for the current time step. t The cell state c at the previous time step t-1 Perform vector multiplication, and then combine it with the input gate i at the current time step. t and the candidate cell state at the current time step The vector product is calculated, the vector sum is used to update the cell state, and the cell state c at the current time step is obtained. t Finally, combine the output gate o of the current time step t and the cell state c at the current time step t According to formula (8), update the hidden state to obtain the hidden state h at the current time step. t This is then output as a temporal feature. For each time step t in the sequence, the above steps are repeated until all time steps have been processed, resulting in the temporal feature sequence of the test parameter.

[0109] In this embodiment, to accurately identify potential drift in test parameters, a time-series-based recurrent neural network is used to extract features from the time-series data of the detection data for each parameter, obtaining the time-series features of the detection data. This effectively captures the complex nonlinear time-series dependencies in the parameter data. Based on the extracted time-series features, a linear regression algorithm is used for fitting, quantifying its trend over time. The fitting process establishes a mapping relationship between the time-series features and the time series, and outputs an evaluation index (e.g., coefficient of determination) to assess the linearity of the trend. This combines the powerful nonlinear feature extraction capability of the recurrent neural network with the highly interpretable advantage of the linear model, improving the accuracy and reliability of parameter drift identification, thereby helping to improve the accuracy of subsequent suspicious parameter screening.

[0110] Next, step S3 is executed to filter out test parameters whose evaluation index is greater than or equal to the first preset value from multiple test parameters, and these parameters are designated as suspicious parameters.

[0111] The evaluation index reflects the correlation between the drift of the test parameter and the detection time. A high correlation indicates that the test parameter does indeed have a drift trend over the detection time. Therefore, by comparing the evaluation index with a first preset value, if the evaluation index is greater than or equal to the first preset value, the test parameter is judged to have a drift risk. In this way, suspicious parameters that may have a drift risk are initially screened from multiple test parameters, reducing the computational load in the subsequent abnormal parameter identification stage and improving efficiency.

[0112] As an example, the evaluation metric is the coefficient of determination R. 2 The first preset value can be set according to expert experience. Optionally, the first preset value is set to 0.45, but this application is not limited to this.

[0113] Next, step S4 is executed to identify whether the suspicious parameter is an abnormal parameter of the product based on the changing trend of the suspicious parameter and the preset fluctuation range.

[0114] Understandably, the goal of the first judgment phase is to identify parameters from a variety of test parameters that exhibit a clear and continuous direction of change (i.e., high linearity). However, a trend alone does not equate to an anomaly. A test parameter may show a very obvious linear drift trend, but the overall magnitude of its drift may still be within the safe range allowed by product specifications or processes; for example, the value of a parameter may decrease slowly over time, but within the monitoring period, all values ​​may still be far above the threshold that would cause product malfunction. Furthermore, linear fitting itself aims to find a straight line that minimizes the overall distance from all data points. When there are individual extreme outliers or local nonlinear fluctuations in the data, the overall trend line fitted to fit these points may be skewed, producing a trend that appears significant but does not actually represent the long-term, true direction of parameter change. This overfitting or false trend caused by data noise or local fluctuations could very well generate a high evaluation index, thus misjudging the parameter as a suspicious parameter.

[0115] This embodiment of the application, after screening out suspicious parameters, performs a secondary judgment on the test parameters based on the changing trends and preset fluctuation ranges of the suspicious parameters, reducing the possibility of misjudgment. Thus, through this dual judgment mechanism, abnormal parameters are automatically, efficiently, and accurately screened from multiple test parameters, reducing reliance on engineer experience, achieving early and accurate identification of potential product quality issues, reducing costs incurred due to delayed problem discovery, and providing reliable data support for engineers' analysis, judgment, and decision-making.

[0116] In some embodiments, identifying whether a suspicious parameter is an abnormal parameter of the product based on the changing trend of the suspicious parameter and a preset fluctuation range may include:

[0117] Based on the changing trend of the suspicious parameter, obtain the deviation of the change of the suspicious parameter within a specified time window relative to the preset fluctuation range; based on whether the deviation of the suspicious parameter exceeds the second preset value, identify whether the suspicious parameter is an abnormal parameter of the product.

[0118] The specified time window can be the production cycle of a product batch, or it can be a specified continuous monitoring duration, such as 24 hours, 7 days, or 30 days. The length and start time of the specified time window can be adjusted and set according to factors such as the product's process characteristics and parameter stability requirements, and this application does not impose specific limitations on this. In this embodiment, by obtaining the deviation of the change amplitude of the suspicious parameter within the specified time window from the preset fluctuation range, the relationship between the change trend of the suspicious parameter and the preset fluctuation range is quantified, directly reflecting the degree of deviation of the suspicious parameter from the preset fluctuation range.

[0119] The second preset value can be a fixed threshold set based on expert experience. Alternatively, the second preset value can be set based on a preset fluctuation range, such as multiplying the width of the preset fluctuation range by a proportional coefficient (e.g., 50%). This application does not specifically limit this.

[0120] In this embodiment of the application, by setting a second preset value, a clear, unified, and objective judgment threshold is established for anomaly judgment, eliminating ambiguity and subjectivity in the judgment process.

[0121] In some embodiments, obtaining the deviation of the change magnitude of the suspicious parameter within a specified time window relative to a preset fluctuation range based on the changing trend of the suspicious parameter may include:

[0122] Based on the changing trend of the suspicious parameter, obtain the first predicted value of the suspicious parameter at the start time and the second predicted value at the end time of the specified time window; based on the difference between the first predicted value and the second predicted value, obtain the change amplitude of the suspicious parameter; based on the change amplitude, the upper limit and lower limit of the preset fluctuation range, obtain the deviation of the change amplitude from the preset fluctuation range.

[0123] The first prediction value is the fitted or predicted value of the change trend of the suspected parameter at the beginning of the specified time window, and the second prediction value is the fitted or predicted value of the change trend of the suspected parameter at the end of the specified time window.

[0124] Therefore, the difference between the first and second predicted values ​​is used to calculate the magnitude of change, instead of using the actual detection data. This ensures that the magnitude of change is derived entirely from the inherent trend revealed by the linear fitting, rather than from any random fluctuations or outliers that may exist in the actual detection data.

[0125] In this embodiment, because the calculation process of the variation amplitude of the suspicious parameter effectively filters out random noise interference in the original detection data, the calculation result of the variation amplitude can more accurately reflect the true drift of the test parameter. Based on this, the accuracy of the obtained variation amplitude relative to the deviation of the preset fluctuation range is higher, thereby effectively improving the accuracy and reliability of subsequent abnormal parameter identification.

[0126] The deviation of the change amplitude from the preset fluctuation range can be achieved through an amplitude detection algorithm, which can be expressed as:

[0127] (9)

[0128] Where ACR (Amplitude Change Ratio) represents the deviation of the change amplitude from the preset fluctuation range, Abs is calculated as an absolute value, and V P0V represents the first predicted value. Pn The second predicted value is represented by UCL, which represents the upper control line of the preset fluctuation range, and LCL represents the lower control line of the preset fluctuation range.

[0129] The second preset value can be set based on expert experience. As an example, the second preset value is 0.35, but this application is not limited to this.

[0130] Once abnormal parameters are identified, an early warning can be issued based on the judgment result. Users can then manually verify the abnormal parameters according to the warning.

[0131] Therefore, by combining the LSTM model for extracting time-series features, the Linear Regression (LR) algorithm for fitting the changing trends of parameters, and the ACR algorithm for detecting the magnitude of parameter deviations, the LLA streaming framework is formed. Through hybrid modeling and streaming processing, the determination of abnormal parameters is completed automatically and accurately.

[0132] When abnormal parameters of a product are identified, the method may also include: classifying and organizing the detection parameters of the abnormal parameters according to different product dimensions.

[0133] Product dimensions are used for attribution analysis of test parameter detection data. Different product dimensions correspond to different physical units or different process units in the product manufacturing process.

[0134] In some embodiments, product dimensions may include multi-level product dimensions, such as first-level product dimensions, second-level product dimensions, and third-level product dimensions. The first-level product dimension refers to the dimension of different batches of products produced by the same production equipment, the second-level product dimension refers to the dimension of different products within the same production batch, and the third-level product dimension refers to the dimension of different detection points on the same product.

[0135] In the semiconductor field, the second-level product dimension refers to the dimension of different wafers within the same production batch, while the third-level product dimension refers to the dimension of different inspection points on the same wafer. When the wafer specifically refers to a chip wafer, the third-level product dimension refers to the dimension of different chips on the same wafer. Product dimensions include batch (Lot) dimension, wafer (Wafer) dimension, and inspection point (Site) or chip (Die) dimension. Wafers and chips can be understood as different physical units in the chip manufacturing process.

[0136] Therefore, when classifying and organizing, the numerous detection data in the abnormal parameters can be classified according to product dimensions.

[0137] When abnormal parameters of a product are identified, the method may further include: performing clustering processing on the detection data of abnormal parameters under different product dimensions, and identifying the product dimension with abnormal data based on the clustering results corresponding to each product dimension.

[0138] The detection data of abnormal parameters are clustered under different product dimensions. Figure 4 schematically shows the clustering results under a certain product dimension, which has three data clusters. Based on the distribution of each data cluster, it can be determined whether there is abnormal data in this product dimension, thereby quickly locating the investigation scope from the entire production process to a specific link or equipment. This realizes the attribution analysis of the problem dimension in the abnormal parameters, which greatly accelerates the speed of problem diagnosis and resolution.

[0139] In some embodiments, the clustering results corresponding to each product dimension include at least two data clusters; based on the clustering results corresponding to each product dimension, identifying product dimensions with abnormal data may include: calculating the similarity between different data clusters in the clustering results for each product dimension; determining the clustering result corresponding to the minimum similarity as the target clustering result, and determining the product dimension corresponding to the target clustering result as the product dimension with abnormal data.

[0140] By calculating the quantifiable mathematical metric of similarity between different data clusters and following explicit rules to select the results with the lowest similarity (i.e., the greatest difference), the root cause localization process is automated, ensuring the consistency and repeatability of the analysis results. This avoids engineers having to browse through numerous data cluster distribution maps, thereby alleviating the inefficiency and low accuracy of manual attribution analysis.

[0141] In this embodiment, the clustering results include normal data clusters and abnormal data clusters.

[0142] Understandably, the goal of clustering is to divide a dataset into different clusters, making data points within the same cluster as similar as possible, while data points between different clusters are as different as possible. Therefore, if a certain product dimension is the root cause of anomalies, there should be a very significant difference between normal and abnormal data in the detection data for that product dimension. This difference would allow the clustering algorithm to quickly divide them into two separate clusters. From the data cluster distribution diagram, the normal data clusters and the abnormal data clusters appear to be separated.

[0143] By calculating similarity, the differences between data clusters can be quantified. By comparing the similarity of different product dimensions, if the similarity of a certain product dimension is the smallest, it means that the data cluster differences in the clustering results of this product dimension are more significant, and it can better distinguish abnormal data from normal data, which meets the expectations for the root cause dimension. This product dimension can be identified as the product dimension with abnormal data.

[0144] Optionally, the clustering process is a clustering process using an unsupervised Gaussian mixture model; wherein each data cluster in the clustering result is modeled as having a normal distribution; the similarity between different data clusters in the clustering result is calculated, including: calculating the similarity between different data clusters in the clustering result based on the probability distribution characteristics of each data cluster in the clustering result.

[0145] By modeling each data cluster as a normal distribution using a Gaussian Mixture Model (GMM), which can be fully described by the mean and variance, similarity calculation can be specifically performed by calculating the statistical distance between two probability distributions based on the mean and variance of each data cluster. This statistical distance is then used as a measure of similarity (SCR). Compared to simply comparing the Euclidean distance between the centers of different data clusters, this method can capture the complete statistical characteristics of the data distribution and identify more subtle and fundamental differences.

[0146] In the clustering process of the GMM model, initialization is performed first, and K parameters (including mean, covariance, and weights) of a Gaussian distribution are randomly set. Then, the E (Expectation) step and the M (Maximization) step are executed cyclically until the convergence stopping condition is met. In the E step, the probability value of each data point belonging to each component is calculated. In the M step, the parameters (mean, covariance, and weights) of each component are updated using the probability values ​​weighted. The stopping condition is that the parameter change is less than a threshold or the log-likelihood no longer increases.

[0147] In practical applications, the value of K can be set according to actual needs. For example, K=2.

[0148] Therefore, by combining the GMM model used for clustering and the SCR similarity algorithm used to reflect the similarity of data clusters, the GSR streaming framework is formed. This approach, through hybrid modeling and streaming processing, automatically locates product problem dimensions and improves the efficiency of engineers in discovering the root causes of product quality issues.

[0149] In some embodiments, the method may further include: generating a cumulative distribution function curve corresponding to each product dimension and visualizing it to help users identify product dimensions with abnormal data.

[0150] Clustering based on the GMM model models the data clusters as normally distributed. The cumulative distribution probability can be calculated using the mean and variance of the normal distribution. Then, a cumulative distribution function (CDF) curve is generated, with the numerical values ​​of the detected data as the x-axis and the cumulative distribution probability as the y-axis. The CDF curve visually displays the numerical distribution of each detected data point within the product dimension. Users can observe the degree of separation of each CDF curve to intuitively determine whether there are anomalies in the detected data, and thus verify the similarity calculation results.

[0151] Next, taking semiconductor products as an example, we will explain the streaming processing of the LLA framework and GSR framework in detail with reference to Figures 5 to 9.

[0152] For example, the test parameters are the EMV-SPACET-CD parameters of semiconductor products. Test data from 185 wafers are selected, with a preset upper limit of UCL=4.6, a lower limit of LCL=3.4, a first preset value of 0.45, and a second preset value of 0.35. It should be noted that in practice, multiple production batches can be selected, with the same number of wafers (185 wafers) selected for each batch. Multiple test points are selected on each wafer, and the test points are located at the same positions on each wafer. Test data is obtained by testing these test points.

[0153] The first step is to preprocess the detection data.

[0154] The second step, please refer to Figure 5, is to extract temporal features from the detection data using the LSTM model.

[0155] The third step, please refer to Figure 6, is to use a linear regression algorithm to perform linear fitting on each time series feature, obtain the fitting trend line, and calculate the coefficient of determination R. 2 =0.76.

[0156] The fourth step is to determine the coefficient of determination R. 2 If the value is 0.76, which is greater than the first preset value of 0.45, then the test parameter is determined to be a suspicious parameter.

[0157] The fifth step is to calculate the rate of change of the detection parameters (ACR) using an amplitude detection algorithm; specifically, please refer to Figure 6 to obtain the first predicted value V at the start of the specified time window. P(0) And the second predicted value V at the end time P(n) According to formula (8), ACR=0.67 is calculated; if ACR=0.67 is greater than the second preset value of 0.35, then the test parameter is determined to be an abnormal parameter.

[0158] The sixth step is to classify and organize the test data according to batch, wafer, and test point dimensions.

[0159] The seventh step involves using the GMM model to cluster the inspection data for each product dimension, resulting in two data clusters. Specifically, data clusters C1 and C2 are obtained for the batch dimension, data clusters C3 and C4 are obtained for the wafer dimension, and data clusters C5 and C6 are obtained for the inspection point dimension.

[0160] Step 8: Calculate the similarity between two data clusters within the same product dimension, and compare the similarity across different product dimensions. The similarity calculation results for different product dimensions are shown in Table 1.

[0161] Table 1

[0162]

[0163] Specifically, the similarity is compared using formula (10). Formula (10) is expressed as:

[0164] (10)

[0165] Where SCR1 represents the similarity between data clusters C1 and C2, SCR2 represents the similarity between data clusters C3 and C4, and SCR3 represents the similarity between data clusters C5 and C6. The calculated SCRs... min =SCR1, then the batch dimension is determined to be the product dimension that caused the data anomaly.

[0166] Step 9: Refer to Figures 7 through 9 to generate and visualize the cumulative distribution function curves for each product dimension. Comparing Figures 7, 8, and 9, it is easy to see that the cumulative distribution function curves shown in Figure 7 are more dispersed. This confirms that the batch dimension has large differences in the detection data, thus proving that the batch dimension is the product dimension that causes the data anomalies.

[0167] Therefore, by using the LLA and GSR frameworks, not only are abnormal parameters automatically and accurately identified, but product problem dimensions are also automatically located. The identification results can be visualized to assist users in reviewing the data, greatly improving the accuracy and efficiency of product testing.

[0168] This application embodiment also provides a product testing device. Please refer to Figure 10. The product testing device includes:

[0169] The acquisition module is used to acquire detection data of one or more test parameters of the target product;

[0170] The linear fitting module is used to obtain the changing trend of each test parameter based on the detection data of the test parameter through linear fitting, and generate an evaluation index to quantify the linearity of the changing trend.

[0171] The first identification module is used to filter out test parameters whose evaluation index is greater than or equal to a first preset value from multiple test parameters, and to identify them as suspicious parameters.

[0172] The second identification module is used to identify whether the suspicious parameters are abnormal parameters of the product based on the changing trend of the suspicious parameters and the preset fluctuation range.

[0173] This application generates an evaluation index for each test parameter to quantify the linearity of its changing trend. This transforms the slow drift, which is difficult to judge intuitively, into an intuitive, interpretable, and comparable value. This allows for the rapid and automatic screening of suspicious parameters with significant changing trends from multiple test parameters. Furthermore, a preset fluctuation range is used to perform a secondary judgment on the changing trend of suspicious parameters to identify whether they are indeed abnormal parameters. Thus, through a dual judgment mechanism, abnormal parameters are automatically, efficiently, and accurately screened from multiple test parameters, reducing reliance on engineers' experience and enabling early and accurate identification of potential product quality issues. At the same time, it can reduce the cost of delayed problem discovery and provide reliable data support for engineers' analysis, judgment, and decision-making.

[0174] Optionally, the product testing device further includes a first preprocessing module for preprocessing the test data of the test parameters.

[0175] In some embodiments, the second identification module is specifically used to obtain the deviation of the change range of the suspicious parameter within a specified time window relative to a preset fluctuation range based on the changing trend of the suspicious parameter; and to identify whether the suspicious parameter is an abnormal parameter of the product based on whether the deviation of the suspicious parameter exceeds a second preset value.

[0176] Optionally, the second identification module is specifically used to: obtain a first predicted value of the suspicious parameter at the start time and a second predicted value at the end time of a specified time window based on the changing trend of the suspicious parameter; obtain the change amplitude of the suspicious parameter based on the difference between the first predicted value and the second predicted value; and obtain the deviation of the change amplitude from the preset fluctuation range based on the change amplitude, the upper limit value and the lower limit value of the preset fluctuation range.

[0177] In some embodiments, the linear fitting module is specifically used to: for each test parameter, use a time-based recurrent neural network to extract features from the detection data of the test parameter to obtain the time-series features of the detection data; and based on the time-series features of the detection data, obtain the changing trend of the test parameter through linear fitting.

[0178] Optionally, the product testing device may also include an early warning module for providing early warnings of abnormal parameters.

[0179] Optionally, the product testing device further includes a second preprocessing module, used to classify and organize the detection parameters of abnormal parameters according to different product dimensions.

[0180] Optionally, the product testing device further includes a third identification module, which is used to perform clustering processing on the test data of abnormal parameters under different product dimensions, and identify the product dimension with abnormal data based on the clustering results corresponding to each product dimension.

[0181] In some embodiments, the clustering result corresponding to each product dimension includes at least two data clusters; the third identification module is specifically used to: calculate the similarity between different data clusters in the clustering result for each product dimension; determine the clustering result corresponding to the minimum similarity as the target clustering result, and determine the product dimension corresponding to the target clustering result as the product dimension with data anomalies.

[0182] Optionally, the clustering process is an unsupervised Gaussian mixture model clustering process; wherein each data cluster in the clustering result is modeled as having a normal distribution; the third identification module is specifically used to: calculate the similarity between different data clusters in the clustering result based on the probability distribution characteristics of each data cluster in the clustering result.

[0183] Optionally, the product testing device also includes a visualization module for generating and visualizing the cumulative distribution function curve corresponding to each product dimension, in order to help users identify product dimensions with abnormal data.

[0184] This application also provides a computer-readable storage medium. The computer-readable storage medium stores instructions that, when executed by a processor of an electronic device, enable the electronic device to perform the product testing method described above.

[0185] Embodiments of this application may be systems, methods, and / or computer program products. A computer program product may include a computer-readable storage medium on which computer-readable program instructions are loaded to cause a processor to implement various aspects of this application. The computer program product may be written in any combination of one or more programming languages ​​to perform operations of embodiments of this application. Programming languages ​​include object-oriented programming languages ​​such as Java, C++, etc., and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code may execute entirely on a user's computing device, partially on a user's device, as a standalone software package, partially on a user's computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computers, the remote computer may be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuits, such as programmable logic circuits, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), are personalized by utilizing state information of computer-readable program instructions. These electronic circuits can execute computer-readable program instructions to implement various aspects of this application.

[0186] Computer-readable storage media can take the form of any combination of one or more readable media. A readable medium can be a readable signal medium or a readable storage medium. A computer-readable storage medium is a tangible device capable of holding and storing instructions for use by an instruction execution device. A readable storage medium can be, for example, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples (a non-exhaustive list) of readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, mechanical encoding devices, such as punch cards or recessed protrusions storing instructions thereon, and any suitable combinations thereof. The computer-readable storage medium as used herein is not to be construed as a transient signal itself, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses through fiber optic cables), or electrical signals transmitted through wires.

[0187] The computer-readable program instructions described herein can be downloaded from computer-readable storage media to various computing / processing devices, or downloaded via a network, such as the Internet, local area network, wide area network, and / or wireless network, to an external computer or external storage device. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to the computer-readable storage media in the respective computing / processing device.

[0188] Various aspects of this application are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer-readable program instructions.

[0189] These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that, when executed by the processor of the computer or other programmable data processing apparatus, they create means for implementing the functions / actions specified in one or more blocks of the flowchart and / or block diagram. These computer-readable program instructions can also be stored in a computer-readable storage medium that causes a computer, programmable data processing apparatus, and / or other device to operate in a particular manner; thus, the computer-readable medium storing the instructions comprises an article of manufacture that includes instructions for implementing aspects of the functions / actions specified in one or more blocks of the flowchart and / or block diagram.

[0190] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions executed on the computer, other programmable data processing apparatus, or other device to perform the functions / actions specified in one or more boxes of a flowchart and / or block diagram.

[0191] This application also provides an electronic device. Figure 11 shows a schematic diagram of the structure of an electronic device provided in an embodiment of this application. As shown in the figure, the electronic device 100 includes: one or more processors 101 and a memory 102; the memory 102 stores computer-executable instructions; the processor 101 is used to execute the computer-executable instructions to implement the above-mentioned product testing method.

[0192] The processor 101 may be a central processing unit (CPU) or other form of processing unit with data processing capabilities and / or instruction execution capabilities, and may control other components in the electronic device to perform desired functions.

[0193] The memory 102 may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and / or cache memory. Non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage medium, and the processor 1501 may execute the program instructions to implement the steps in the text recognition methods of the various embodiments of this application described above, and / or other desired functions.

[0194] In one example, the electronic device 100 may also include input devices and output devices, which are interconnected via a bus system and / or other forms of connection mechanism (not shown in the figure).

[0195] In addition, input devices may include, for example, a keyboard, a mouse, a microphone, etc. Output devices can output various information to the outside, and may include, for example, a monitor, speakers, a printer, and communication networks and their connected remote output devices, etc.

[0196] Of course, for simplicity, Figure 11 only shows a portion of the components of the electronic device 100 relevant to this application, omitting components such as buses, input devices / output interfaces, etc. In addition, the electronic device 100 may include any other suitable components depending on the specific application.

[0197] It should be noted that the product testing method embodiments, product testing device embodiments, computer-readable storage medium embodiments, and electronic device embodiments provided in this application belong to the same concept; the technical features in the technical solutions described in each embodiment can be arbitrarily combined without conflict.

[0198] It should be understood that the above embodiments are exemplary and not intended to encompass all possible implementations. Various modifications and changes can be made to the above embodiments without departing from the scope of this disclosure. Similarly, the various technical features of the above embodiments can be arbitrarily combined to form other embodiments of the present invention that may not be explicitly described. Therefore, the above embodiments only illustrate several implementations of the present invention and do not limit the scope of protection of this patent.

Claims

1. A product testing method, characterized in that, The method includes: acquiring detection data of one or more test parameters of a target product, wherein the target product is a semiconductor product; for each test parameter, based on the detection data of the test parameter, obtaining the changing trend of the test parameter through linear fitting, and generating an evaluation index for quantifying the linearity of the changing trend; selecting test parameters from multiple test parameters whose evaluation index is greater than or equal to a first preset value as suspicious parameters; identifying whether the suspicious parameter is an abnormal parameter of the target product based on the changing trend of the suspicious parameter and a preset fluctuation range; in the case of identifying an abnormal parameter of the target product, performing clustering processing on the detection data of the abnormal parameter under different product dimensions, and identifying the product dimension with abnormal data based on the clustering results corresponding to each product dimension; wherein the clustering results corresponding to each product dimension include at least two data clusters; the step of identifying the product dimension with abnormal data based on the clustering results corresponding to each product dimension includes: calculating the similarity between different data clusters in the clustering results for each product dimension; determining the clustering result corresponding to the minimum value of the similarity as the target clustering result, and determining the product dimension corresponding to the target clustering result as the product dimension with abnormal data.

2. The product testing method according to claim 1, characterized in that, The step of identifying whether the suspicious parameter is an abnormal parameter of the target product based on the changing trend of the suspicious parameter and the preset fluctuation range includes: obtaining the deviation of the change amplitude of the suspicious parameter within a specified time window relative to the preset fluctuation range based on the changing trend of the suspicious parameter; and identifying whether the suspicious parameter is an abnormal parameter of the target product based on whether the deviation of the suspicious parameter exceeds a second preset value.

3. The product testing method according to claim 2, characterized in that, The step of obtaining the deviation of the change amplitude of the suspicious parameter within a specified time window relative to a preset fluctuation range based on the change trend of the suspicious parameter includes: obtaining a first predicted value of the suspicious parameter at the start time and a second predicted value at the end time of the specified time window based on the change trend of the suspicious parameter; obtaining the change amplitude of the suspicious parameter based on the difference between the first predicted value and the second predicted value; and obtaining the deviation of the change amplitude relative to the preset fluctuation range based on the change amplitude, the upper limit value and the lower limit value of the preset fluctuation range.

4. The product testing method according to claim 1, characterized in that, The step of obtaining the changing trend of each test parameter by linear fitting based on the detection data of the test parameter includes: for each test parameter, using a time-series-based recurrent neural network to extract features from the detection data of the test parameter to obtain the time-series features of the detection data; and obtaining the changing trend of the test parameter by linear fitting based on the time-series features of the detection data.

5. The product testing method according to claim 1, characterized in that, The clustering process is an unsupervised Gaussian mixture model clustering process; wherein, each data cluster in the clustering result is modeled as having a normal distribution; the calculation of the similarity between different data clusters in the clustering result includes: calculating the similarity between different data clusters in the clustering result based on the probability distribution characteristics of each data cluster in the clustering result.

6. The product testing method according to claim 1, characterized in that, The method further includes generating a cumulative distribution function curve corresponding to each product dimension and visually displaying it to help users identify the product dimensions with abnormal data.

7. The product testing method according to claim 1, characterized in that, The product dimension is used to perform attribution analysis on the detection data of the test parameters. Different product dimensions correspond to different physical units or different process units in the manufacturing process of the target product.

8. A product testing device, characterized in that, include: The acquisition module is used to acquire detection data of one or more test parameters of a target product, wherein the target product is a semiconductor product; The linear fitting module is used to obtain the changing trend of each test parameter based on the detection data of the test parameter through linear fitting, and generate an evaluation index for quantifying the linearity of the changing trend. The first identification module is used to filter out test parameters from the plurality of test parameters whose evaluation index is greater than or equal to a first preset value, as suspicious parameters; The second identification module is used to identify whether the suspicious parameter is an abnormal parameter of the target product based on the changing trend of the suspicious parameter and the preset fluctuation range. The third identification module is used to cluster the detection data of abnormal parameters under different product dimensions, and identify the product dimensions with abnormal data based on the clustering results corresponding to each product dimension; wherein, the clustering results corresponding to each product dimension include at least two data clusters; the third identification module is specifically used to: calculate the similarity between different data clusters in the clustering results for each product dimension; determine the clustering result corresponding to the minimum value of the similarity as the target clustering result, and determine the product dimension corresponding to the target clustering result as the product dimension with abnormal data.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores instructions that, when executed by a processor of an electronic device, enable the electronic device to perform the product testing method according to any one of claims 1 to 7.

10. An electronic device, characterized in that, The electronic device includes: a processor; a memory for storing computer-executable instructions; the processor being configured to execute the computer-executable instructions to perform the product testing method according to any one of claims 1 to 7.

Images