Integrated circuit test parameter optimization method based on XGBoost algorithm

Abstract

This invention provides a method for optimizing integrated circuit test parameters based on the XGBoost algorithm, belonging to the field of integrated circuit testing technology. The invention first calculates the correlation coefficients between various test parameters, extracting parameter pairs capable of one-to-one mutual prediction. Then, it trains the XGBoost algorithm to predict these relationships, establishing a prediction model and selecting a set of input parameters suitable for predicting each target parameter. Next, it uses a first threshold to select initial removable parameters and verifies their input dependencies, followed by joint prediction. Finally, it dynamically backtracks and adjusts parameters exceeding a second threshold to determine the final set of retained parameters. This scheme combines the prediction results of the final deleted parameters with the measured results of the retained parameters, generating complete test parameter results while simplifying the measured parameters, thereby effectively shortening the Fourier Transform (FT) test time and reducing mass production testing costs.

Description

Technical Field

[0001] This invention relates to the field of integrated circuit testing technology, and more specifically to an integrated circuit testing parameter optimization method based on the XGBoost algorithm. Background Technology

[0002] As integrated circuit process nodes continue to evolve towards deeper submicron and nanometer scales, chip integration density and functional complexity are increasing exponentially, leading to a significant increase in the number of test parameters that need to be covered in the Final Test (FT) stage. These test parameters cover multiple core dimensions such as RF performance, power consumption, signal integrity, and timing characteristics. Traditional testing solutions require the verification of all preset test parameters one by one to ensure chip yield. However, the surge in test parameters directly leads to a significant increase in the testing time of automated testing equipment, a continuous rise in the unit chip testing cost during mass production, and the lengthy testing process may also increase the risk of chip damage, affecting mass production efficiency. In scenarios where chip mass production reaches hundreds of thousands or even millions of units, the efficiency shortcomings and cost pressures of the traditional "full-parameter traversal testing" mode become increasingly prominent, urgently requiring parameter optimization to simplify the testing process. In reality, the various test parameters of integrated circuits are not completely independent, but rather exhibit significant inherent correlations based on the physical connections and operating logic of the chip's underlying circuitry. For example, power consumption parameters and signal output parameters of the same functional module, as well as timing parameters and integrity parameters of different links, often show strong correlations due to shared circuit resources or causal relationships. This correlation means that the test results of some parameters can be derived from the measured data of other parameters. Repeatedly testing such redundant parameters not only fails to improve the fault detection rate but also wastes test resources. Summary of the Invention

[0003] (a) Technical problems to be solved To address the shortcomings of existing technologies, this invention provides a method for optimizing integrated circuit test parameters based on the XGBoost algorithm, which solves the technical problem of how to avoid repeated testing of redundant parameters.

[0004] (II) Technical Solution To achieve the above objectives, the present invention provides the following technical solution: A method for optimizing integrated circuit test parameters based on the XGBoost algorithm, comprising: Obtain the correlation coefficients between various test parameters; Test parameters whose correlation coefficients reach the correlation coefficient threshold are grouped in pairs to obtain several parameter pairs containing input parameters and target parameters, and the XGBoost algorithm is used to analyze the first prediction error of each target parameter. Compare the relationship between the first prediction error and the first threshold for each target parameter, and select the set of input parameters that can be used to predict the target parameter. Traverse the sequence of candidate deletion parameters consisting of target parameters whose first prediction error does not exceed the first threshold, check whether the input parameter set of the current candidate deletion parameter contains candidate deletion parameters that have not been traversed. If so, add them to the deletion set; otherwise, add them to the retention set. Extract all input parameters of each deleted parameter in the retention set to form the corresponding joint prediction input, and use the XGBoost algorithm to analyze the second prediction error of each deleted parameter in order to filter out the deleted parameters whose second prediction error exceeds the second threshold. The selected deletion parameters are moved from the deletion set to the retention set one by one to form a new joint prediction input. The XGBoost algorithm is then used to analyze the third prediction error of each deletion parameter in the current deletion set. Delete parameters whose third prediction error exceeds the second threshold are moved to the current retention set to obtain the final retention set.

[0005] Preferably, obtaining the correlation coefficients between the various test parameters includes: Convert the STDF file in the FT test to a CSV file, convert the dataset to a numerical type, and summarize the test parameters that need to be analyzed; Calculate the Pearson correlation coefficients between the various test parameters.

[0006] Preferably, the step of pairwise grouping of test parameters with correlation coefficients exceeding a correlation coefficient threshold to obtain several parameter pairs containing input parameters and target parameters, and using the XGBoost algorithm to analyze the first prediction error of each target parameter, includes: If the correlation coefficient between any two test parameters reaches the preset correlation threshold, construct the corresponding parameter pair including the input parameter and the target parameter; Divide all parameter pairs into training and validation sets; The first prediction model is established by training the test parameters one-to-one prediction relationship using the training set and the XGBoost algorithm. The validation set is used as input to the first prediction model to obtain the first prediction error for each target parameter.

[0007] Preferably, the candidate deletion parameter sequence is composed in the following ways: The target parameters whose first prediction error does not exceed the first threshold are statistically analyzed and sorted in ascending order of their first prediction error values ​​to form a candidate deletion parameter sequence.

[0008] Preferably, before moving the selected deletion parameters from the deletion set to the retention set one by one, the selected deletion parameters are sorted in descending order according to the value of the second prediction error.

[0009] Preferably, one or more of the first prediction error, the second prediction error, and the third prediction error include the maximum correlation error and the mean absolute percentage error; wherein the maximum correlation error is the maximum percentage error of a single sample in each parameter, and the mean absolute percentage error is the average of the absolute percentage errors between the predicted value and the actual value.

[0010] An integrated circuit test parameter optimization system based on the XGBoost algorithm includes: The acquisition module is used to obtain the correlation coefficients between various test parameters; The analysis module is used to group test parameters whose correlation coefficients reach the correlation coefficient threshold in pairs, obtain several parameter pairs containing input parameters and target parameters, and use the XGBoost algorithm to analyze the first prediction error of each target parameter. The filtering module is used to compare the relationship between the first prediction error and the first threshold of each target parameter, and filter out the set of input parameters that can be used to predict the target parameter. The checking module is used to traverse the candidate deletion parameter sequence consisting of target parameters whose first prediction error does not exceed the first threshold, and check whether the input parameter set of the current candidate deletion parameter contains candidate deletion parameters that have not been traversed. If so, they are included in the deletion set; otherwise, they are included in the retention set. The first joint prediction module is used to extract all input parameters of each deletion parameter in the retention set, form the corresponding joint prediction input, and use the XGBoost algorithm to analyze the second prediction error of each deletion parameter in order to filter out the deletion parameters whose second prediction error exceeds the second threshold. The second joint prediction module is used to move the selected deletion parameters from the deletion set to the retention set one by one to form a new joint prediction input. It also uses the XGBoost algorithm to analyze the third prediction error of each deletion parameter in the current deletion set and moves the deletion parameters whose third prediction error exceeds the second threshold to the current retention set to obtain the final retention set.

[0011] A storage medium storing a computer program for optimizing integrated circuit test parameters based on the XGBoost algorithm, wherein the computer program causes a computer to execute the integrated circuit test parameter optimization method as described above.

[0012] An electronic device, comprising: One or more processors; a memory; and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the programs including methods for performing integrated circuit test parameter optimization as described above.

[0013] (III) Beneficial Effects This invention provides a method for optimizing integrated circuit test parameters based on the XGBoost algorithm. Compared with existing technologies, it has the following advantages: This invention first calculates the correlation coefficients between various test parameters to extract parameter pairs capable of one-to-one mutual prediction. Then, it uses the XGBoost algorithm to train the prediction relationships, establishing a prediction model and selecting a set of input parameters suitable for predicting each target parameter. Next, it uses a first threshold to filter out initial removable parameters and verifies their input dependencies, followed by joint prediction. Finally, it dynamically backtracks and adjusts parameters exceeding a second threshold to determine the final set of retained parameters. This scheme combines the prediction results of the final deleted parameters with the measured results of the retained parameters, generating complete test parameter results while simplifying the measured parameters, thereby effectively shortening the FT testing time and reducing mass production testing costs. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 A flowchart illustrating an integrated circuit test parameter optimization method based on the XGBoost algorithm, provided for an embodiment of the present invention; Figure 2 A schematic diagram of a prediction task matrix provided in an embodiment of the present invention; Figure 3 This invention provides a predictive directed graph as an embodiment of the present invention.

[0016] Figure 4 This is a schematic diagram illustrating the prediction effect of a removable test parameter under a second threshold, provided by an embodiment of the present invention. Figure 5 This is a comparison chart showing the optimization effect under different second threshold schemes provided in an embodiment of the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention are described clearly and completely. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] The technical solution in this application is to solve the above-mentioned technical problems, and the general idea is as follows: Machine learning algorithms have demonstrated significant advantages in high-dimensional data mining and prediction. This invention constructs a predictive model to uncover complex relationships between parameters, filters and removes redundant parameters that can be accurately predicted, retaining only the core, essential parameters for testing, thereby optimizing the testing process. This prediction-based parameter optimization method can significantly shorten testing time and reduce testing costs while ensuring testing accuracy and fault coverage. Simultaneously, it provides complete data support for subsequent chip fault analysis and classification, which is of great significance to the economy and reliability of integrated circuit mass production testing.

[0019] A separate explanation of several technical terms used in the embodiments of this invention: 1. The XGBoost algorithm is an ensemble learning algorithm, short for eXtreme Gradient Boosting, belonging to the boosting method within ensemble learning. It is an improved and optimized version based on Gradient Boosting Decision Tree (GBDT), which builds a powerful predictive model by integrating multiple weak learners (usually decision trees).

[0020] 2. STDF (Standard Test Data Format) is a data format widely used in semiconductor testing, typically for storing test data generated by test equipment.

[0021] 3. CSV (Comma-Separated Values) is a simple and practical file format used to store and represent various types of data, including text and numbers.

[0022] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0023] Example 1: like Figure 1 As shown, this embodiment of the invention provides a method for optimizing integrated circuit test parameters based on the XGBoost algorithm, including: S1. Obtain the correlation coefficients between various test parameters; S2. Group the test parameters whose correlation coefficients reach the correlation coefficient threshold in pairs to obtain several parameter pairs containing input parameters and target parameters, and use the XGBoost algorithm to analyze the first prediction error of each target parameter. S3. Compare the relationship between the first prediction error and the first threshold for each target parameter, and select the set of input parameters that can be used to predict the target parameter. S4. Traverse the sequence of candidate deletion parameters consisting of target parameters whose first prediction error does not exceed the first threshold. Check whether the input parameter set of the current candidate deletion parameters contains candidate deletion parameters that have not been traversed. If so, include them in the deletion set; otherwise, include them in the retention set. S5. Extract all input parameters of each deletion parameter in the retention set, form the corresponding joint prediction input, and use the XGBoost algorithm to analyze the second prediction error of each deletion parameter in order to filter out the deletion parameters whose second prediction error exceeds the second threshold. S6. Move the selected deletion parameters from the deletion set to the retention set one by one to form a new joint prediction input. Use the XGBoost algorithm to analyze the third prediction error of each deletion parameter in the current deletion set. Move the deletion parameters whose third prediction error exceeds the second threshold to the current retention set to obtain the final retention set.

[0024] This invention simplifies test parameters while generating complete results for all removed test parameters through a predictive model. This ensures both test accuracy and fault coverage, and provides comprehensive data support for subsequent chip fault analysis and classification.

[0025] The following will detail each step of the above solution: In step S1, the correlation coefficients between the various test parameters are obtained.

[0026] S11. Convert the STDF file in the FT test to a CSV file, convert the dataset to a numerical type, and summarize the test parameters that need to be analyzed.

[0027] For example, the embodiments of the present invention use a total of 22 test parameters, and the 22 test parameters are numbered 00-21 respectively.

[0028] S12. Calculate the Pearson correlation coefficient between each test parameter. Specifically: Samples with missing values ​​are excluded; mean values ​​for all data points are calculated for each test parameter; all test parameters are grouped into pairs; for each pair, the sum of the products of the two deviations is calculated, as well as the sum of the squares of the two deviations is calculated separately; the sum of the products of the deviations is divided by the product of the square roots of the sums of the squares of the two deviations; finally, the Pearson correlation coefficient of the parameter pair (i.e., between each pair of test parameters) is obtained.

[0029] In step S2, the test parameters whose correlation coefficients reach the correlation coefficient threshold are grouped in pairs to obtain several parameter pairs containing input parameters and target parameters, and the XGBoost algorithm is used to analyze the first prediction error of each target parameter.

[0030] S21. If the correlation coefficient between any two test parameters reaches the preset correlation threshold, construct the corresponding parameter pair including the input parameter and the target parameter.

[0031] For example, in this embodiment of the invention, the correlation threshold is set to 0.9. If the correlation between the two test parameters of a parameter pair reaches the set threshold, it can be preliminarily determined that the two sets of test parameters can perform one-to-one mutual prediction. In this embodiment of the invention, a total of 39 sets of 1-to-1 mutual prediction parameters meet the requirement, namely: 00-01, 00-02, 00-03, 00-04, 01-02, 01-03, 01-04, 02-04, 02-05, 03-04, 03- 06, 03-07, 03-08, 04-08, 05-06, 06-07, 06-08, 07-08, 09-10, 11-12, 11-13, 11-14, 11-15, 11-16, 11-17, 12-14, 12-15, 12-16, 13-14, 13-15, 14-15, 14-17, 14-18, 14-19, 15-19, 16-17, 17-18, 17-19, 20-21.

[0032] S22. Divide all parameter pairs into training set and validation set in a 7:3 ratio. The training set is used to train the model, and the validation set is used to determine the training effect and prediction effect of the model.

[0033] S23. Using the training set and the XGBoost algorithm, train the one-to-one prediction relationship of the test parameters to establish the first prediction model. Specifically: Each test parameter is used as the target variable, and the corresponding mutually predictable test parameters are used as feature inputs. XGBoost is then used to build a prediction model.

[0034] To make the predictions more accurate, the hyperparameters used in the embodiments of the present invention are as follows: the tree type is a classification and regression tree, the number of decision trees is 200, the maximum depth of each tree is 6, the learning rate is 0.05, the feature sampling ratio of each tree is 0.7, the number of leaf nodes is 15, and the random number seed is 42.

[0035] S24. Use the validation set as input to the first prediction model to obtain the first prediction error for each target parameter. Specifically: The validation set is used to test the model training effect, and the first prediction error of each target parameter is calculated, including the maximum correlation error and the mean absolute percentage error between the true value and the predicted value in the validation set. The calculation methods are as follows: the maximum correlation error is the maximum percentage error of a single sample in each test parameter, and the mean absolute percentage error is the average of the absolute percentage errors between the predicted value and the actual value.

[0036] In step S3, the relationship between the first prediction error and the first threshold of each target parameter is compared to select the set of input parameters that can be used to predict the target parameter.

[0037] Continuing with the above example, in this embodiment of the invention, the first threshold is set as follows: the maximum correlation error does not exceed 5%, the average absolute percentage error does not exceed 1.5%, and if the error of the test parameter as the target parameter in the validation set reaches the specified first threshold, then it is defined that the input parameter can predict the target parameter.

[0038] For each target parameter, all relevant input parameters that can be used to predict that target parameter are selected, forming a correspondence unit of "input parameter group - target parameter". All correspondence units corresponding to the target parameters are then integrated to finally construct a system as follows: Figure 2 The matrix shown contains the prediction task matrix with all test parameters.

[0039] In step S4, the sequence of candidate deletion parameters consisting of target parameters whose first prediction error does not exceed the first threshold is traversed. The input parameter set of the current candidate deletion parameter is checked to see if it contains candidate deletion parameters that have not been traversed. If so, they are included in the deletion set; otherwise, they are included in the retention set.

[0040] S41, form a candidate deletion parameter sequence.

[0041] For example, this embodiment of the invention provides a feasible method for composing a candidate deletion parameter sequence: Target parameters with a first prediction error not exceeding a first threshold are statistically analyzed and sorted according to the first prediction error value from smallest to largest to form a candidate deletion parameter sequence. Specifically: First, based on the prediction task matrix obtained in step S3, the input parameters required when each test parameter is used as the prediction target are identified to form a basic dependency relationship; then, combined with the first prediction error obtained in step S2, the parameters are sorted from smallest to largest according to the value of the first prediction error to form a candidate deletion parameter sequence.

[0042] S42. Traverse the candidate deletion parameter sequence and check each candidate deletion parameter one by one. Only if the input parameter set of the current candidate deletion parameter contains a candidate deletion parameter that has not been traversed will it be included in the deletion set. Otherwise, it will be included in the retention set as the prediction input for the remaining test parameters.

[0043] S43, Finally establish as follows Figure 3 The predicted directed graph shown illustrates the prediction path from retained parameters to deleted parameters, ensuring that each deleted parameter is supported by a complete input from the retained parameters. The arrows start at the test parameters as features and end at the test parameters as targets.

[0044] In step S5, all input parameters of each deletion parameter in the retention set are extracted to form the corresponding joint prediction input, and the second prediction error of each deletion parameter is analyzed using the XGBoost algorithm to filter out deletion parameters whose second prediction error exceeds the second threshold.

[0045] S51. Based on the predicted directed graph established in step S4, the initial retain set and the initial delete set were obtained by statistical analysis.

[0046] Continuing with the above example, the test parameters in the initial retained set in this embodiment of the invention are: 00, 03, 04, 10, 11, 15, 16, 17, 21; and the test parameters in the initial deleted set are: 01, 02, 05, 06, 07, 08, 09, 12, 13, 14, 18, 19, 20.

[0047] S52. For each deletion parameter in the initial deletion set, based on... Figure 3 Extract all input parameters from the initial retained parameter set to form the corresponding joint prediction input. Meanwhile, take each deleted parameter in the initial deleted set as the target variable. Referring to step S2, train the model using the training set and the XGBoost algorithm. Perform joint prediction using the model and calculate the second prediction error of each deleted parameter using the validation set.

[0048] Continuing with the above example, the second threshold set in this embodiment of the invention is: the maximum correlation error does not exceed 1.9% and the average absolute percentage error does not exceed 0.8%.

[0049] S53. Filter out the deletion parameters whose second prediction error exceeds the second threshold from the initial deletion set.

[0050] Continuing with the above example, in this embodiment of the invention, by comparing the prediction error with the second threshold, the deletion parameters in the initial deletion parameter set can be divided into two categories: one category is the parameter set that does not exceed the second threshold, namely: 01, 02, 07, 08, 12, 14, 19; the other category is the parameter set that exceeds the second threshold, namely: 05, 06, 09, 13, 18, 20.

[0051] It should be noted that, similar to the first prediction error, the second prediction error mentioned above can also include the maximum correlation error and the mean absolute percentage error, which will not be elaborated here.

[0052] In step S6, the selected deletion parameters are moved from the deletion set to the retention set one by one to form a new joint prediction input. The XGBoost algorithm is used to analyze the third prediction error of each deletion parameter in the current deletion set. Delete parameters whose third prediction error exceeds the second threshold are moved to the current retention set to obtain the final retention set.

[0053] S61. Sort the filtered deletion parameters in descending order of the second prediction error value.

[0054] S62. After sorting, each of the filtered deletion parameters is placed into the initial retain set and removed from the initial deletion set.

[0055] S63, Based on the current deletion parameters Figure 2 The target parameters corresponding to the prediction task matrix shown are used as input parameters. Using the updated retained set and referring to step S2, joint prediction is performed on all parameters still in the initial deletion set.

[0056] S64. After re-prediction, if a deletion parameter is found to exceed the second threshold, it is transferred from the current retention set.

[0057] S65. Continue to repeat steps S62 to S64 until all the deleted parameters have been filtered and the final set is output.

[0058] It should be noted that, similar to the first prediction error, the third prediction error mentioned above can also include the maximum correlation error and the mean absolute percentage error, which will not be discussed further here.

[0059] Furthermore, it is understood that the final retained set is output, while the remaining test parameters are saved to the final deleted set. Following this process, and continuing with the example above, the third prediction error of the deletable test parameters on the validation set can be obtained as follows: Figure 4 As shown, below the second threshold, nine test parameters can be deleted, accounting for 40.91% of the total test parameters. The average absolute percentage error of all deleted parameters is only 0.35%, and the average maximum correlation error is only 1.05%, which strongly demonstrates the effectiveness of the embodiments of the present invention.

[0060] To further demonstrate the effectiveness of the embodiments of the present invention, experiments were conducted at different second thresholds. The mean absolute percentage errors were 1.1%, 0.95%, 0.8%, 0.65%, 0.5%, and 0.35%, respectively, and the maximum correlation errors were 2.5%, 2.2%, 1.9%, 1.6%, 1.3%, and 1.0%, respectively. Finally, the average value of the mean absolute percentage error of the removable parameters, the average value of the maximum correlation error, and the number of removable parameters were obtained under different thresholds. The results are as follows: Figure 5As shown, it can be seen that as the threshold requirements become increasingly stringent, the number of parameters that can be deleted gradually decreases, and the average error also gradually decreases. Below the threshold with a maximum relative error of 2.5%, the number of parameters that can be deleted reaches 12, accounting for 55% of all test parameters. By changing the threshold, the prediction error and the number of deletions of the test parameters also change accordingly, and the results further demonstrate the effectiveness of the embodiments of the present invention.

[0061] Example 2: This invention provides an integrated circuit test parameter optimization system based on the XGBoost algorithm, comprising: The acquisition module is used to obtain the correlation coefficients between various test parameters; The analysis module is used to group test parameters whose correlation coefficients reach the correlation coefficient threshold in pairs, obtain several parameter pairs containing input parameters and target parameters, and use the XGBoost algorithm to analyze the first prediction error of each target parameter. The filtering module is used to compare the relationship between the first prediction error and the first threshold of each target parameter, and filter out the set of input parameters that can be used to predict the target parameter. The checking module is used to traverse the candidate deletion parameter sequence consisting of target parameters whose first prediction error does not exceed the first threshold, and check whether the input parameter set of the current candidate deletion parameter contains candidate deletion parameters that have not been traversed. If so, they are included in the deletion set; otherwise, they are included in the retention set. The first joint prediction module is used to extract all input parameters of each deletion parameter in the retention set, form the corresponding joint prediction input, and use the XGBoost algorithm to analyze the second prediction error of each deletion parameter in order to filter out the deletion parameters whose second prediction error exceeds the second threshold. The second joint prediction module is used to move the selected deletion parameters from the deletion set to the retention set one by one to form a new joint prediction input. It also uses the XGBoost algorithm to analyze the third prediction error of each deletion parameter in the current deletion set and moves the deletion parameters whose third prediction error exceeds the second threshold to the current retention set to obtain the final retention set.

[0062] Example 3: This invention provides a storage medium storing a computer program for optimizing integrated circuit test parameters based on the XGBoost algorithm, wherein the computer program causes a computer to execute the integrated circuit test parameter optimization method as described in Embodiment 1.

[0063] Example 4: This invention provides an electronic device, comprising: One or more processors; a memory; and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the programs including methods for performing integrated circuit test parameter optimization as described in Example 1.

[0064] It is understood that the integrated circuit test parameter optimization system, storage medium and electronic device based on the XGBoost algorithm provided in the embodiments of the present invention correspond to the integrated circuit test parameter optimization method based on the XGBoost algorithm provided in the embodiments of the present invention. The explanation, examples and beneficial effects of the relevant contents can be referred to the corresponding parts of the method, and will not be repeated here.

[0065] In summary, compared with existing technologies, it has the following beneficial effects: 1. The embodiments of the present invention use the XGBoost machine learning algorithm to construct a one-to-one joint prediction model for test parameters. By calculating the maximum relative error and the average absolute percentage error, the predictability and correlation strength of each test parameter can be accurately quantified.

[0066] 2. The embodiments of the present invention construct a prediction directed graph based on the prediction task matrix, and combine error sorting and dynamic backtracking adjustment mechanism to screen out the minimum number of retained parameters that meet the error threshold requirements, determine the parameters that can be deleted, and use the retained parameters to jointly predict them, which can effectively reduce the number of test parameters in the FT test, thereby shortening the test time.

[0067] 3. In this embodiment of the invention, while simplifying test parameters, the complete results of all deleted test parameters are generated through a predictive model, which not only ensures test accuracy and fault coverage, but also provides comprehensive data support for subsequent chip fault analysis and classification.

[0068] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0069] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An integrated circuit test parameter optimization method based on an XGBoost algorithm, characterized by, include: Obtain the correlation coefficients between various test parameters; Test parameters whose correlation coefficients reach the correlation coefficient threshold are grouped in pairs to obtain several parameter pairs containing input parameters and target parameters, and the XGBoost algorithm is used to analyze the first prediction error of each target parameter. Compare the relationship between the first prediction error and the first threshold for each target parameter, and select the set of input parameters that can be used to predict the target parameter. Traverse the sequence of candidate deletion parameters consisting of target parameters whose first prediction error does not exceed the first threshold, check whether the input parameter set of the current candidate deletion parameter contains candidate deletion parameters that have not been traversed. If so, add them to the deletion set; otherwise, add them to the retention set. Extract all input parameters of each deleted parameter in the retention set to form the corresponding joint prediction input, and use the XGBoost algorithm to analyze the second prediction error of each deleted parameter in order to filter out the deleted parameters whose second prediction error exceeds the second threshold. The selected deletion parameters are moved from the deletion set to the retention set one by one to form a new joint prediction input. The XGBoost algorithm is then used to analyze the third prediction error of each deletion parameter in the current deletion set. Delete parameters whose third prediction error exceeds the second threshold are moved to the current retention set to obtain the final retention set.

2. The integrated circuit test parameter optimization method of claim 1, wherein, The process of obtaining the correlation coefficients between various test parameters includes: Convert the STDF file in the FT test to a CSV file, convert the dataset to a numerical type, and summarize the test parameters that need to be analyzed; Calculate the Pearson correlation coefficients between the various test parameters.

3. The integrated circuit test parameter optimization method of claim 1, wherein, The process involves pairwise grouping of test parameters whose correlation coefficients exceed a threshold to obtain several parameter pairs containing input and target parameters, and using the XGBoost algorithm to analyze the first prediction error of each target parameter, including: If the correlation coefficient between any two test parameters reaches the preset correlation threshold, construct the corresponding parameter pair including the input parameter and the target parameter; Divide all parameter pairs into training and validation sets; The first prediction model is established by training the test parameters one-to-one prediction relationship using the training set and the XGBoost algorithm. The validation set is used as input to the first prediction model to obtain the first prediction error for each target parameter.

4. The integrated circuit test parameter optimization method of claim 1, wherein, The candidate deletion parameter sequence is composed in the following ways: The target parameters whose first prediction error does not exceed the first threshold are statistically analyzed and sorted in ascending order of their first prediction error values ​​to form a candidate deletion parameter sequence.

5. The integrated circuit test parameter optimization method of claim 1, wherein, Before moving the selected deletion parameters from the deletion set to the retention set one by one, the selected deletion parameters are sorted from largest to smallest according to the second prediction error value.

6. The integrated circuit test parameter optimization method as described in claim 1, characterized in that, One or more of the first prediction error, the second prediction error, and the third prediction error include the maximum correlation error and the mean absolute percentage error; wherein the maximum correlation error is the maximum percentage error of a single sample in each parameter, and the mean absolute percentage error is the average of the absolute percentage errors between the predicted value and the actual value.

7. An integrated circuit test parameter optimization system based on the XGBoost algorithm, characterized in that, include: The acquisition module is used to obtain the correlation coefficients between various test parameters; The analysis module is used to group test parameters whose correlation coefficients reach the correlation coefficient threshold in pairs, obtain several parameter pairs containing input parameters and target parameters, and use the XGBoost algorithm to analyze the first prediction error of each target parameter. The filtering module is used to compare the magnitude of the first prediction error and the first threshold of each target parameter, and filter out the set of input parameters that can be used to predict the target parameter. The checking module is used to traverse the candidate deletion parameter sequence consisting of target parameters whose first prediction error does not exceed the first threshold, and check whether the input parameter set of the current candidate deletion parameter contains candidate deletion parameters that have not been traversed. If so, they are included in the deletion set; otherwise, they are included in the retention set. The first joint prediction module is used to extract all input parameters of each deletion parameter in the retention set, form the corresponding joint prediction input, and use the XGBoost algorithm to analyze the second prediction error of each deletion parameter in order to filter out the deletion parameters whose second prediction error exceeds the second threshold. The second joint prediction module is used to move the selected deletion parameters from the deletion set to the retention set one by one to form a new joint prediction input. It also uses the XGBoost algorithm to analyze the third prediction error of each deletion parameter in the current deletion set and moves the deletion parameters whose third prediction error exceeds the second threshold to the current retention set to obtain the final retention set.

8. A storage medium, characterized in that, It stores a computer program for optimizing integrated circuit test parameters based on the XGBoost algorithm, wherein the computer program causes the computer to execute the integrated circuit test parameter optimization method as described in any one of claims 1 to 6.

9. An electronic device, characterized in that, include: One or more processors; Memory; And one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the programs including methods for performing integrated circuit test parameter optimization as described in any one of claims 1 to 6.

Images