A semiconductor device testing method, device and storage medium

Abstract

The present invention relates to the field of semiconductor device testing technology, and discloses a semiconductor device testing method, system, equipment and storage medium. The method comprises: obtaining historical device test data and an initial test plan; constructing and training a model based on the historical device test data to obtain a performance prediction model and device parameter characteristics; processing the device parameter characteristics using an adaptive large neighborhood search model to obtain a test plan; predicting the performance of the test plan based on the performance prediction model to obtain prediction data, and determining an optimal test plan based on the prediction data, so as to perform device testing according to the optimal test plan. The method effectively reduces test time and improves test efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of semiconductor device testing, and in particular to a semiconductor device testing method, device and storage medium. Background Art

[0002] In recent years, my country's integrated circuit industry has developed rapidly. As a key link to ensure product quality and performance, integrated circuit testing is becoming increasingly important. How to improve testing efficiency and reduce costs has gradually become the key to integrated circuit testing.

[0003] In traditional testing schemes for integrated circuits, all devices are subject to uniform test items, and the test scheme is usually static and adjusted only through offline analysis when necessary. However, with the advancement of manufacturing processes, this approach has exposed several problems. First, the front test items can often detect most faulty devices, while the back items take more time and detect relatively few faults. Second, there is a lot of redundancy between different test types, and the same fault may be detected repeatedly, which increases the test time but does not effectively improve the defect detection rate. Finally, due to dynamic changes in the manufacturing process, fixed test items are difficult to adapt to evolving defect modes.

[0004] During the testing process of the above technology, there is a lot of redundancy between different test types. The same fault may be detected repeatedly, which increases the test time, but does not effectively improve the defect detection rate, resulting in low test efficiency. Summary of the invention

[0005] The present invention provides a semiconductor device testing method, apparatus and storage medium to solve the problem that there is a large amount of redundancy between different test types during the testing process, the same fault may be detected repeatedly, which increases the testing time but does not effectively improve the defect detection rate, resulting in low testing efficiency.

[0006] In a first aspect, in order to solve the above technical problems, the present invention provides a semiconductor device testing method, comprising:

[0007] Obtain historical device test data and initial test plans;

[0008] Based on the historical device test data, a performance prediction model is constructed and trained to obtain device parameter characteristics; wherein the device parameter characteristics are obtained according to the weights of the decision operators in the process of training the model;

[0009] According to the device parameter characteristics, an adaptive large neighborhood search model is used to process and obtain a test plan;

[0010] The performance of the test scheme is predicted based on the performance prediction model to obtain prediction data, and an optimal test scheme is determined according to the prediction data, so as to perform device testing according to the optimal test scheme.

[0011] Preferably, the historical device test data includes test set data and training set data;

[0012] The training set data is used for training the performance prediction model, and the test set data is used for evaluating the prediction effect of the performance prediction model.

[0013] Preferably, the constructing and training a model based on the historical device test data to obtain a performance prediction model and device parameter characteristics includes:

[0014] Based on the historical device test data, a random forest is used for training modeling to obtain a performance prediction model;

[0015] The test set data is input into the performance prediction model to obtain the test results of various parameters, and the absolute values ​​of the differences between all the test results and the preset actual values ​​are calculated;

[0016] When the absolute values ​​of all the differences are less than the preset deviation threshold, the training is determined to be completed and the performance prediction model is obtained.

[0017] Preferably, after acquiring historical device test data and an initial test plan, the method further comprises:

[0018] The training set data is preprocessed, and the abnormal data or missing data of the faulty device is filled in according to the average value of the test value of the fault-free device obtained in advance.

[0019] Preferably, the process of using an adaptive large neighborhood search model to obtain a test solution based on the device parameter characteristics includes:

[0020] Initializing parameters of an adaptive large neighborhood search model, and setting the initial test scheme as a first test scheme; wherein the parameters include an initial search range, a preset number of iterations, an operator set, and an initial operator weight;

[0021] Performing an iterative search according to the device parameter characteristics to obtain a second test solution, and recording the number of iterations;

[0022] According to the preset evaluation indicators, determine whether the evaluation indicators of the second test scheme are better than the evaluation indicators of the first test scheme, and if so, update the second test scheme to the first test scheme; if not, abandon the second test scheme, wherein the preset evaluation indicators include test time and defect detection rate;

[0023] Continue to the next iteration, and when the number of iterations is greater than or equal to the preset number of iterations, output the current first test plan as the optimal test plan.

[0024] Preferably, the method further comprises:

[0025] During the iteration process, the weight of the operator is updated using the following formula:

[0026]

[0027] in, represents the weight of the operator, is the attenuation coefficient, Score the effectiveness of the operator, is the number of times the operator is used.

[0028] Preferably, the performing performance prediction on the test scheme based on the performance prediction model to obtain prediction data, and determining the optimal test scheme according to the prediction data includes:

[0029] Judging the predicted data according to a preset error threshold;

[0030] When the error of the predicted data is greater than a preset error threshold, the test scheme is updated using the adaptive large domain search model;

[0031] When the error of the predicted data is less than the preset error threshold, the test scheme is determined to be the optimal test scheme.

[0032] Preferably, after performing performance prediction on the test scheme based on the performance prediction model to obtain prediction data, and determining the optimal test scheme according to the prediction data, the method further includes:

[0033] Calculating a test escape rate based on the predicted data;

[0034] When the test escape rate is higher than a preset escape threshold, the test plan is updated using the adaptive large domain search model to obtain an updated optimal test plan;

[0035] When the test escape rate is lower than a preset escape threshold, the test plan is recorded;

[0036] The test escape rate is the ratio of the number of faulty devices that the model incorrectly predicts to be normal devices to the total number of tested devices.

[0037] In a second aspect, the present invention provides a semiconductor device testing device, comprising:

[0038] An information acquisition unit, used to acquire historical device test data and an initial test plan;

[0039] A model building unit, used to build a performance prediction model based on the historical device test data and train it to obtain device parameter characteristics; wherein the device parameter characteristics are obtained according to the weight of the decision operator in the process of training the model;

[0040] A scheme generating unit, used for obtaining a test scheme by processing according to device parameter characteristics using an adaptive large neighborhood search model;

[0041] The performance prediction unit is used to perform performance prediction according to the test plan using the performance prediction model to obtain prediction data, and determine the optimal test plan according to the prediction data.

[0042] In a third aspect, the present invention further provides a computer-readable storage medium, which includes a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute any one of the semiconductor device testing methods described above.

[0043] Compared with the prior art, the present invention has the following beneficial effects:

[0044] The present invention provides a semiconductor device testing method, including: obtaining historical device test data and an initial test plan; constructing and training a model based on the historical device test data to obtain a performance prediction model and device parameter characteristics; processing the device parameter characteristics using an adaptive large neighborhood search model to obtain a test plan; predicting the performance of the test plan based on the performance prediction model to obtain prediction data, and determining an optimal test plan based on the prediction data, so as to perform device testing according to the optimal test plan. During the test process, a performance prediction model is established, and a test plan is obtained through processing with an adaptive large neighborhood search model, and the test plan is predicted using the performance prediction model to determine the optimal test plan, thereby reducing the test time and improving the test efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 is a schematic diagram of a semiconductor device testing method provided by a first embodiment of the present invention;

[0046] Figure 2 It is a structural schematic diagram of a semiconductor device testing device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0047] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0048] Reference Figure 1 The first embodiment of the present invention provides a semiconductor device test, comprising the following steps:

[0049] S1, obtain historical device test data and initial test plan;

[0050] S2, based on the historical device test data, constructing a performance prediction model and training the model to obtain device parameter characteristics;

[0051] S3, according to the device parameter characteristics, using an adaptive large neighborhood search model to process and obtain a test plan;

[0052] S4, performing performance prediction on the test scheme based on the performance prediction model to obtain prediction data, and determining an optimal test scheme according to the prediction data, so as to perform device testing according to the optimal test scheme.

[0053] It should be noted that in recent years, my country's integrated circuit industry has developed rapidly. Against this background, integrated circuit testing, as a key link throughout the entire cycle of product design, manufacturing, packaging and final application, has become increasingly strategic. It plays an irreplaceable role in ensuring product quality, optimizing product performance, accelerating product iteration and promoting the upgrading and development of the entire industry chain. With the continuous improvement of the complexity of integrated circuits and the increasing diversification of functions, higher requirements are placed on testing technology. The level of test efficiency directly affects the time to market, cost control and market competitiveness of products. Therefore, how to significantly improve test efficiency and reduce test costs while ensuring the comprehensiveness and accuracy of the test has become a core issue that needs to be urgently solved in the current field of integrated circuit testing.

[0054] As an indispensable component of electronic equipment, the quality and reliability of integrated circuits are directly related to the stability and performance of the entire system. Therefore, the integrated circuit testing process plays a vital role in the device production process. Traditionally, the integrated circuit test scheme uses a unified test item to evaluate all devices. Although this "one-size-fits-all" test scheme guarantees the standardization and consistency of the test to a certain extent, with the continuous advancement of manufacturing processes and the increase in device complexity, this method has gradually shown its limitations. First, the test items in the test scheme are usually executed in a fixed order. A major problem with this method is that the front test items can often detect most of the faulty devices, while those test items at the back have a relatively small number of faulty devices detected after consuming a lot of test time. This not only causes a waste of resources, but also reduces the efficiency of testing. Secondly, there is a lot of redundancy between different test types. In the actual test process, the same fault may be repeatedly detected by multiple test items. Although this phenomenon increases the probability of fault detection to a certain extent, it greatly increases the time cost of testing, but does not correspondingly increase the defect detection rate. This not only affects the economic efficiency of testing, but also makes the testing process cumbersome and inefficient. Finally, due to various dynamic changes in the integrated circuit manufacturing process, such as process fluctuations, changes in material properties, etc., these changes will cause the defect mode of the device to evolve continuously. Traditional fixed test items are difficult to adapt to such rapidly changing defect modes, which reduces the accuracy and reliability of the test results.

[0055] In order to facilitate the understanding of the present invention, some preferred embodiments of the present invention are further described below.

[0056] In step S1, historical device test data and an initial test plan are obtained.

[0057] The initial test plan refers to a set of test plans and projects that are preset before testing semiconductor devices. This initial test plan is usually formulated based on past test experience, industry standards, and specific requirements of device design. It clarifies the test objectives, required test equipment, specific test steps and sequence, and how to evaluate the performance and quality of the device through these tests.

[0058] It should be noted that the historical device test data includes test set data and training set data;

[0059] The training set data is used to train the performance prediction model, and the test set data is used to evaluate the prediction effect of the performance prediction model;

[0060] Exemplarily, the test set data is a set of device test data that has been tested and has known results, and these data are used to evaluate and verify the prediction effect of the performance prediction model. It usually comes from the actual device testing process, and is obtained by collection and organization to ensure that it contains various factors that may affect the performance of the device. In the model development process, the test set data is used for model evaluation in the final stage, and the accuracy and reliability of the model are evaluated by comparing the predicted results of the model with the actual results in the test set; the training set data is the original data set used to build and train the performance prediction model, which contains a large number of data points for learning device performance characteristics. It is usually collected through historical test records, including the response and performance indicators of the device under different test conditions. The training set data is preprocessed to ensure the quality of the data and the needs of model training. It is used in the model building stage, and by inputting these data into the training model, the model learns how to predict the performance of the device based on the input features. During the training process, the model automatically adjusts the parameters of the model based on these data to reduce the prediction error, and finally forms a mature model that can accurately predict the performance of the device.

[0061] It should be noted that after obtaining the historical device test data and the initial test plan, the method further includes:

[0062] The training set data is preprocessed, and the abnormal data or missing data of the faulty device is filled in according to the average value of the test value of the fault-free device obtained in advance.

[0063] Specifically, the test values ​​of fault-free devices are first summarized and statistically analyzed to calculate the average values ​​of various test indicators. These average values ​​reflect the expected performance of fault-free devices under normal working conditions. Subsequently, for the faulty devices in the training set, if their test data is abnormal or missing, that is, there is a significant deviation from the expected performance under normal working conditions, or the key indicators are missing, making it impossible to conduct effective analysis, the average values ​​calculated above are used to fill in the data, and the statistical characteristics of normal devices are used to estimate and replace the missing or abnormal data points in the faulty devices, thereby ensuring the integrity and consistency of the training data set and avoiding model training deviations caused by incomplete data.

[0064] In step S2, based on the historical device test data, a performance prediction model is constructed and trained to obtain device parameter characteristics.

[0065] It should be noted that, based on the historical device test data, random forest is used for training and modeling to obtain a performance prediction model.

[0066] Among them, the test set data is input into the performance prediction model to obtain the test results of various parameters, and the absolute values ​​of the differences between all the test results and the preset actual values ​​are calculated; when all the absolute values ​​of the differences are less than the preset deviation threshold, the training is determined to be completed and the performance prediction model is obtained.

[0067] Among them, the preset deviation threshold is used to determine whether the model has achieved sufficient training effect. The setting of this threshold is based on the tester's control of the model prediction error to ensure that the model does not overfit the training data during the training process. Specifically, when the absolute value of the difference between the model's prediction results and the actual values ​​for all samples is less than the preset deviation threshold, it is considered that the model has learned enough data features and the training effect meets the established standards. At this time, the training process will terminate automatically, thereby preventing the model from over-learning the noise and outliers in the training data and ensuring the generalization ability of the model. When the prediction error of all samples is less than this threshold, the training stops to avoid overfitting.

[0068] Random forest is an ensemble learning technique that improves the accuracy and robustness of the model by building multiple decision trees and voting or averaging predictions. In building the device test prediction model, the accuracy and robustness of the prediction is improved by creating a forest composed of multiple decision trees. Specifically, the model first applies bootstrap sampling to the original data set to generate multiple sub-data sets with replacement, ensuring that the data trained on each tree is slightly different. Then, at each node of each tree, the model randomly selects a portion of the features to search for the best split point instead of using all the features, which increases the randomness of the model and reduces the risk of overfitting. Each tree grows independently to its maximum extent without pruning until it cannot be further subdivided. Finally, the prediction results of all trees are integrated by voting or averaging to form the final prediction output. This method can balance the prediction bias of each tree while using their diversity to improve the generalization ability of the overall model, thereby achieving more accurate defect detection and performance evaluation in device testing.

[0069] In an optional implementation, the process of obtaining a prediction model using random forest includes:

[0070] S21, obtain the training set data and initialize the parameters of the random forest.

[0071] The parameters mainly include the number of decision trees (M), the maximum depth of each tree (D), and the number of device parameter features (N).

[0072] When determining the number of decision trees, the cross-validation method is used to evaluate the impact of different numbers of decision trees on the model performance. By comparing the prediction performance of the model under different numbers of decision trees or observing the changing trend of the model performance as the number of decision trees increases, the number of decision trees that optimizes the model performance is found.

[0073] S22. According to the training set data, samples are drawn using the bootstrap sampling method to obtain multiple sub-datasets. Among them, each sub-dataset can construct a decision tree.

[0074] Among them, the bootstrap sampling method, also known as the bootstrap method, is a statistical technique for drawing samples from the original dataset and is used to construct each sub-dataset in the random forest. This method allows each sample to have the chance to be selected in each draw, and the sample is put back after each draw, so that the same sample may be repeatedly drawn in the same sub-dataset multiple times. In this way, the bootstrap sampling can generate multiple sub-datasets of the same size as the original dataset, but the sample combinations of each sub-dataset are different. In the random forest, the application of the bootstrap sampling method makes the training data of each decision tree have a certain degree of randomness, which helps to increase the diversity of the model, reduce the dependence on specific samples, and thus reduce the risk of overfitting. Specifically, the following steps are repeatedly executed: randomly draw a sample from the original dataset, record the draw result, then put the sample back, and continue the next draw until the required number of sub-datasets is obtained. The training of each tree is carried out independently, using different sub-datasets, so that each tree can learn the characteristics of the data from different perspectives, and finally improve the accuracy and robustness of the overall model by integrating the prediction results of these trees.

[0075] S23. In the sub-dataset, randomly select n features (where n < N) from the N device parameter features as the feature subset F for node splitting.

[0076] S24. Calculate the optimal splitting point of the features in the feature subset at the current node, and perform splitting according to the optimal splitting point to obtain two child nodes;

[0077] Among them, during the splitting process, calculate the Gini index of the feature subset to obtain the optimal splitting point. The Gini index of this feature subset can be calculated by the following formula:

[0078]

[0079] Where G is the Gini index, N is the total number of samples in the node, is the number of samples belonging to the k-th feature, and K is the total number of features of the samples in the node.

[0080] The Gini index is an indicator used to measure the purity of a data set, and its value ranges from 0 to 1. For a data set, if all samples belong to the same category, the Gini index of the data set is 0, indicating the highest purity; if the samples are evenly distributed in multiple categories, the Gini index is close to 1, indicating the lowest purity of the data set.

[0081] At each node of the decision tree, the model considers all possible features and all possible values ​​of these features. For each possible value of each feature, the model calculates the Gini index if the data set is split according to this value of this feature. After traversing all possible values ​​of the feature and calculating the Gini index for each possible split, the feature and value that reduces the Gini index the most are selected as the optimal split point.

[0082] The two child nodes will contain samples in the original node divided according to the split point value. After the split, the data set of each child node will only contain samples corresponding to the child node in the original node. In this way, each child node will become a candidate node for the next split.

[0083] S25, repeat step S24 to the maximum depth to construct a decision tree.

[0084] For each child node, a feature subset is randomly selected from all available features to obtain the feature subset for further splitting of the child node.

[0085] S26, repeat the above steps S23 to S25 until the Decision trees are constructed to form a random forest, and the prediction model is obtained.

[0086] Random forests are able to assess the importance of each device parameter feature to the prediction result. This is achieved by calculating the reduction in the Gini index for each feature when building a decision tree. The importance of device parameter features can be expressed by the following formula:

[0087]

[0088] in, is the characteristic of the device parameter, is the reduction in the Gini index.

[0089] In step S3, an adaptive large neighborhood search model is used to process the device parameter characteristics to obtain a test solution.

[0090] It should be noted that the test scheme is obtained by processing the device parameter characteristics using an adaptive large neighborhood search model, including:

[0091] Initializing parameters of an adaptive large neighborhood search model, and setting the initial test scheme as a first test scheme; wherein the parameters include an initial search range, a preset number of iterations, an operator set, and an initial operator weight;

[0092] Performing an iterative search according to the device parameter characteristics to obtain a second test solution, and recording the number of iterations;

[0093] According to the preset evaluation indicators, determine whether the evaluation indicators of the second test scheme are better than the evaluation indicators of the first test scheme, and if so, update the second test scheme to the first test scheme; if not, abandon the second test scheme, wherein the preset evaluation indicators include test time and defect detection rate;

[0094] Continue to the next iteration, and when the number of iterations is greater than or equal to the preset number of iterations, output the current first test plan as the optimal test plan.

[0095] An optional implementation method of obtaining a test solution by processing using an adaptive large neighborhood search model includes:

[0096] S31, initializing the parameters of the adaptive large neighborhood search model, and setting the initial test scheme as the first test scheme, generating an initial solution according to the first test scheme , let the current solution ;

[0097] S32, start iterative search, select the Destroy operator and the Repair operator according to the weight of each operator in the operator set, and update the number of times the operator is used.

[0098] Among them, the Destroy operator is a mechanism used to perturb the current solution in the adaptive large neighborhood search model. Its core purpose is to break the local optimal trap that may be encountered during the search process to explore a wider solution space. This operator generates one or more new candidate solutions by randomly or strategically changing part of the current solution, such as exchanging, moving or deleting certain features or parameters. If the test solution obtained after the repair operator fails to provide improvement, the operator will adjust its destruction strategy based on feedback to increase the possibility of finding a better solution.

[0099] The Repair operator is a key component in the adaptive large neighborhood search model. Its role is to adjust and optimize the candidate solutions generated by the Destroy operator that may not fully meet the problem constraints. The introduction of the Repair operator is based on the understanding that the solution generated by the Destroy operator may violate certain constraints of the problem during the destruction process, and its feasibility needs to be restored in some way. These operators usually use local search techniques, such as neighborhood search or simulated annealing, to fine-tune the destroyed solution to ensure that it meets all given constraints while maintaining or improving the quality of the solution as much as possible to generate new test solutions. After the Destroy operator generates a new solution, the Repair operator will be triggered to perform necessary repairs on these solutions; then, the repaired solution will re-enter the evaluation phase of the model and compete with other solutions to determine whether it can become the new current solution. In this way, the Repair operator not only ensures the feasibility of the solution, but also helps the model gradually approach the optimal solution to the problem while maintaining diversity.

[0100] The selection of Destroy and Repair operators is usually based on a roulette mechanism. The probability of each operator being selected is proportional to its weight, which is determined by the following formula:

[0101]

[0102] in, is the probability of each operator being selected, is the weight of each operator.

[0103] S33, the Destroy operator and the Repair operator process the first test scheme in turn to obtain the second test scheme, and generate a test solution according to the second test scheme. Compare to current solution to update.

[0104] like , then accept the test solution As the current solution , determine that the evaluation index of the second test scheme is better than the evaluation index of the first test scheme, and update the second test scheme to the first test scheme; if , it is determined that the evaluation index of the second test solution is not better than the evaluation index of the first test solution, and the test solution The probability of being accepted as the current solution is determined by the following formula:

[0105]

[0106] in, Test solution The probability of being accepted, is the test index function value of the current solution, is the test index function value of the test solution, is the search scope.

[0107] As the test index function value of the test solution, it represents the performance or quality of the solution in a specific optimization problem. This value comes from the performance evaluation of the test solution based on preset evaluation indicators, such as test time and defect detection rate. In the adaptive large neighborhood search model, The value of the test indicator function used to compare with the current solution Compare to decide whether to accept the new solution. Less than or equal to , then the test solution is at least as good as the current solution and can be directly accepted. Greater than , then it is necessary to use the probability acceptance criterion to decide whether to accept this possibly poor solution to promote a wider exploration of the solution space.

[0108] As the test index function value of the test solution, it is the performance evaluation of the test solution obtained according to the preset evaluation index during the model iteration process. As the model progresses, when a better solution is found, will be updated to the test indicator function value of the better solution. In the probability acceptance criterion, Serves as a benchmark for evaluating whether the newly generated test solutions are acceptable. The purpose of using is to ensure that the model can jump out of the local optimum while pursuing the local optimum and search towards the global optimal solution.

[0109] In the probability acceptance formula, The cost difference between the current solution and the test solution is calculated. If Less than , that is, the test metric of the test solution is better (which means better in the minimization problem), then is positive, which will make A value greater than 0 and close to 1 indicates that the test solution is likely to be accepted. On the contrary, if Greater than , that is, the test index of the current solution is better, is negative, which will make A value of less than 1 and close to 0 indicates that the test solution is unlikely to be accepted.

[0110] Furthermore, the search scope Determines the probability that the model accepts a poor solution. In the early stages of the model, Higher, the model is more inclined to explore, even if Higher than , the test solution may also be accepted. As the model progresses, As it decreases, the model is more inclined to exploit, that is, it only accepts solutions with better indicators, thus balancing exploration and exploitation in the search process.

[0111] In each iteration, Usually, the adjustment is made according to the geometric progression reduction scheme. The specific formula is as follows:

[0112]

[0113] in The search range to be used for the next iteration, is the search range used in the current iteration, a is the decreasing proportional coefficient, the value range is 0 to 1, and is set by the developer according to the preset number of iterations.

[0114] S34, update the weights of the Destroy operator and the Repair operator using the following formula:

[0115]

[0116] in, represents the weight of the operator, is the attenuation coefficient, Score the effectiveness of the operator, is the number of times the operator is used.

[0117] It should be noted that the attenuation coefficient It is a key parameter in the adaptive large neighborhood search model, which is used to adjust the operator weights to reflect that as the search progresses, some operators may gradually lose their initial effectiveness. It is usually a value between 0 and 1, which determines the proportion of the operator weight reduction after each iteration. This parameter allows the model to explore a wide range of possible search paths in the early stage, and gradually tend to those operators that show better performance as the iteration progresses, thereby achieving more effective exploration of the search space.

[0118] Specifically, operator effect scoring It is an indicator based on the quality of the solution generated by the operator during the iteration process. It is determined by comparing the performance difference between the test solution generated after the operator is applied and the current solution. If the test solution generated by an operator is better than the current solution in terms of the evaluation index, then the operator will receive a higher score, indicating that it is effective and should be considered more in future searches.

[0119] S35, repeat steps S32 to S34, and record the number of iterations. When the number of iterations is greater than or equal to the preset number of iterations, the current solution is The test plan output is taken as the optimal test plan.

[0120] The preset number of iterations defines the maximum number of iterations that the model will perform before termination. This parameter is set based on a comprehensive consideration of model performance, computing resources, and problem complexity. Reasonable selection of the preset number of iterations helps to obtain high-quality solutions under limited computing resources. Developers will select an expected number of iterations that can achieve a satisfactory solution based on the model's performance in previous similar tasks and their understanding of the scale and characteristics of the current problem.

[0121] In step S4, the performance of the test scheme is predicted based on the performance prediction model to obtain prediction data, and an optimal test scheme is determined according to the prediction data, so as to perform device testing according to the optimal test scheme.

[0122] It should be noted that the performing performance prediction on the test scheme based on the performance prediction model to obtain prediction data, and determining the optimal test scheme according to the prediction data, includes:

[0123] Judging the predicted data according to a preset error threshold;

[0124] When the error of the predicted data is greater than a preset error threshold, the test scheme is updated using the adaptive large domain search model;

[0125] When the error of the predicted data is less than the preset error threshold, the test scheme is determined to be the optimal test scheme.

[0126] Among them, the preset error threshold is a predetermined standard used in the adaptive large neighborhood search model to evaluate the quality of the output and compare it with the expected target. It is an important parameter in the model decision-making process, reflecting the acceptable deviation range of the model output, and is used to determine whether the accuracy of the prediction model or test scheme meets specific requirements. It is usually set by developers based on domain knowledge and historical data requirements, and can also be determined through data analysis and statistical methods based on the performance of previous similar systems or problems. In addition, the threshold can also be optimized through experiments and adjustments to balance the performance and efficiency of the model.

[0127] Prediction data are performance indicators obtained when testing the plan, including defect detection rate and test time.

[0128] In addition, it should be noted that after the performance prediction of the test scheme is performed based on the performance prediction model to obtain prediction data, and the optimal test scheme is determined according to the prediction data, the method further includes:

[0129] Calculating the test escape rate according to the predicted data; when the test escape rate is higher than the preset escape threshold, updating the test plan using the adaptive large domain search model to obtain the updated optimal test plan; when the test escape rate is lower than the preset escape threshold, recording the test plan;

[0130] Among them, the test escape rate is a key indicator to measure the accuracy of the semiconductor device test plan. It is defined as the ratio of the number of faulty devices that are incorrectly predicted to be normal devices during the test process to the total number of tested devices. This indicator directly reflects the risk that the test plan may miss faulty devices. The specific calculation formula is as follows:

[0131]

[0132] in, represents the test escape rate, The number of failed devices that were mistakenly judged as qualified devices during the test process. is the total number of devices.

[0133] This ratio is obtained by dividing the number of faulty devices that were misidentified by the total number of devices tested. It is usually counted and analyzed after the test plan is completed to evaluate the effectiveness of the test plan. By reviewing the prediction results generated by the test plan, it can be determined how many faulty devices were not correctly identified. This ratio is obtained by dividing the number of faulty devices that were misidentified by the total number of devices tested. It is usually counted and analyzed after the test plan is completed to evaluate the effectiveness of the test plan.

[0134] The preset escape threshold is a predetermined standard used to define the acceptable escape rate of the test plan in semiconductor device testing. It is a key parameter for test quality control and is defined as the maximum test escape rate allowed without affecting the test quality. Testers can determine a reasonable escape threshold by analyzing the escape instance data in the historical test data. If the actual test escape rate is lower than or equal to this threshold, the test plan can be considered effective and can be recorded. If the test escape rate is higher than the preset escape threshold, it indicates that there may be problems with the test plan and further adjustments and optimizations are needed to improve the test model, enhance the test coverage, or increase the strictness of the test parameters to ensure that the device quality meets the standards.

[0135] The present invention provides a semiconductor device testing method, including: obtaining historical device test data and an initial test plan; constructing and training a model based on the historical device test data to obtain a performance prediction model and device parameter characteristics; processing the device parameter characteristics using an adaptive large neighborhood search model to obtain a test plan; predicting the performance of the test plan based on the performance prediction model to obtain prediction data, and determining an optimal test plan based on the prediction data, so as to perform device testing according to the optimal test plan. During the test process, a performance prediction model is established, and a test plan is obtained through processing with an adaptive large neighborhood search model, and the test plan is predicted using the performance prediction model to determine the optimal test plan, thereby reducing the test time and improving the test efficiency.

[0136] In summary, the present invention provides a semiconductor device testing method, including: obtaining historical device test data and an initial test plan; based on the historical device test data, building and training a model to obtain a performance prediction model and device parameter characteristics; according to the device parameter characteristics, using an adaptive large neighborhood search model to process to obtain a test plan; based on the performance prediction model, the test plan is predicted to obtain prediction data, and the optimal test plan is determined according to the prediction data, so as to perform device testing according to the optimal test plan. During the test process, a performance prediction model is established, and a test plan is obtained through processing with an adaptive large neighborhood search model, and the test plan is predicted using the performance prediction model to determine the optimal test plan, thereby reducing the test time and improving the test efficiency.

[0137] Reference Figure 2 A second embodiment of the present invention provides a semiconductor device testing device, comprising:

[0138] An information acquisition unit, used to acquire historical device test data and an initial test plan;

[0139] A model building unit, used to build a performance prediction model based on the historical device test data and train it to obtain device parameter characteristics; wherein the device parameter characteristics are obtained according to the weight of the decision operator in the process of training the model;

[0140] A scheme generating unit, used for obtaining a test scheme by processing according to device parameter characteristics using an adaptive large neighborhood search model;

[0141] The performance prediction unit is used to perform performance prediction according to the test plan using the performance prediction model to obtain prediction data, and determine the optimal test plan according to the prediction data.

[0142] It should be noted that a semiconductor device testing apparatus provided in an embodiment of the present invention is used to execute all the steps of a semiconductor device testing method in the above embodiment, and the working principles and beneficial effects of the two correspond one to one, so they will not be described in detail.

[0143] An embodiment of the present invention further provides an electronic device. The electronic device includes: a processor, a memory, and a computer program stored in the memory and executable on the processor, such as a semiconductor device test program. When the processor executes the computer program, the steps in the above-mentioned semiconductor device test method embodiments are implemented, such as Figure 1 Alternatively, when the processor executes the computer program, the functions of each module / unit in the above-mentioned device embodiments are implemented, such as the model building module.

[0144] Exemplarily, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to implement the present invention. The one or more modules / units may be a series of computer program instruction segments capable of implementing specific functions, which are used to describe the execution process of the computer program in the electronic device.

[0145] The electronic device may be a computing device such as a desktop computer, a notebook, a PDA, and a smart tablet. The electronic device may include, but is not limited to, a processor and a memory. Those skilled in the art will appreciate that the above components are merely examples of electronic devices and do not constitute a limitation on the electronic device. The electronic device may include more or fewer components than the above components, or may combine certain components, or different components. For example, the electronic device may also include input and output devices, network access devices, buses, etc.

[0146] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the electronic device, and uses various interfaces and lines to connect various parts of the entire electronic device.

[0147] The memory can be used to store the computer program and / or module, and the processor realizes various functions of the electronic device by running or executing the computer program and / or module stored in the memory, and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, an application required for at least one function (such as a sound playback function, an image playback function, etc.), etc.; the data storage area can store data created according to the use of the mobile phone (such as audio data, a phone book, etc.), etc. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0148] Wherein, if the module / unit integrated in the electronic device is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the present invention implements all or part of the schemes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Wherein, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electric carrier signals and telecommunication signals.

[0149] It should be noted that the device embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. In addition, in the accompanying drawings of the device embodiments provided by the present invention, the connection relationship between the modules indicates that there is a communication connection between them, which may be specifically implemented as one or more communication buses or signal lines. A person of ordinary skill in the art may understand and implement it without paying any creative effort.

[0150] The specific embodiments described above further illustrate the purpose, technical solutions and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. It is particularly pointed out that for those skilled in the art, any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the scope of protection of the present invention.

Claims

1. A semiconductor device testing method, characterized in that: include: Obtain historical device test data and initial test plans; Based on the historical device test data, a performance prediction model is constructed and trained to obtain device parameter characteristics; wherein the device parameter characteristics are obtained according to the weights of the decision operators in the process of training the model; According to the device parameter characteristics, an adaptive large neighborhood search model is used to process and obtain a test plan; Performing performance prediction on the test scheme based on the performance prediction model to obtain prediction data, and determining an optimal test scheme according to the prediction data, so as to perform device testing according to the optimal test scheme; Wherein, the process of using an adaptive large neighborhood search model to process the device parameter characteristics to obtain a test solution includes: Initializing parameters of an adaptive large neighborhood search model, and setting the initial test scheme as a first test scheme; wherein the parameters include an initial search range, a preset number of iterations, an operator set, and an initial operator weight; Performing an iterative search according to the device parameter characteristics to obtain a second test solution, and recording the number of iterations; According to the preset evaluation indicators, determine whether the evaluation indicators of the second test scheme are better than the evaluation indicators of the first test scheme, and if so, update the second test scheme to the first test scheme; if not, abandon the second test scheme, wherein the preset evaluation indicators include test time and defect detection rate; Continue to the next iteration, and when the number of iterations is greater than or equal to the preset number of iterations, output the current first test plan as the optimal test plan; Wherein, the predicted data are performance indicators obtained when testing the scheme, including defect detection rate and test time; The step of performing performance prediction on the test scheme based on the performance prediction model to obtain prediction data, and determining the optimal test scheme according to the prediction data, includes: Judging the predicted data according to a preset error threshold; When the error of the predicted data is greater than a preset error threshold, the test scheme is updated using the adaptive large neighborhood search model; When the error of the predicted data is less than the preset error threshold, the test scheme is determined to be the optimal test scheme.

2. The semiconductor device testing method according to claim 1, characterized in that: The historical device test data includes test set data and training set data; The training set data is used for training the performance prediction model, and the test set data is used for evaluating the prediction effect of the performance prediction model.

3. The semiconductor device testing method according to claim 2, characterized in that: Based on the historical device test data, a performance prediction model is constructed and trained to obtain device parameter characteristics, including: Based on the historical device test data, a random forest is used for training modeling to obtain a performance prediction model; The test set data is input into the performance prediction model to obtain the test results of various parameters, and the absolute values ​​of the differences between all the test results and the preset actual values ​​are calculated; When the absolute values ​​of all the differences are less than the preset deviation threshold, the training is determined to be completed and the performance prediction model is obtained.

4. The semiconductor device testing method according to claim 2, characterized in that: After acquiring historical device test data and an initial test plan, the method further includes: The training set data is preprocessed, and the abnormal data or missing data of the faulty device is filled in according to the average value of the test value of the fault-free device obtained in advance.

5. The semiconductor device testing method according to claim 1, characterized in that: The method further comprises: During the iteration process, the weight of the operator is updated using the following formula: in, represents the weight of the operator, is the attenuation coefficient, Score the effectiveness of the operator, is the number of times the operator is used.

6. The semiconductor device testing method according to claim 1, characterized in that: After predicting the performance of the test scheme based on the performance prediction model to obtain prediction data, and determining the optimal test scheme according to the prediction data, the method further includes: Calculating a test escape rate based on the predicted data; When the test escape rate is higher than a preset escape threshold, the test plan is updated using the adaptive large neighborhood search model to obtain an updated optimal test plan; When the test escape rate is lower than a preset escape threshold, the test plan is recorded; The test escape rate is the ratio of the number of faulty devices that the model incorrectly predicts to be normal devices to the total number of tested devices.

7. A semiconductor device testing device, characterized in that: include: An information acquisition unit, used to acquire historical device test data and an initial test plan; A model building unit, used to build a performance prediction model based on the historical device test data and train it to obtain device parameter characteristics; wherein the device parameter characteristics are obtained according to the weight of the decision operator in the process of training the model; A scheme generating unit, used for obtaining a test scheme by processing according to device parameter characteristics using an adaptive large neighborhood search model; A performance prediction unit, configured to perform performance prediction using the performance prediction model according to the test scheme, obtain prediction data, and determine an optimal test scheme according to the prediction data; Wherein, the process of using an adaptive large neighborhood search model to process the device parameter characteristics to obtain a test solution includes: Initializing parameters of an adaptive large neighborhood search model, and setting the initial test scheme as a first test scheme; wherein the parameters include an initial search range, a preset number of iterations, an operator set, and an initial operator weight; Performing an iterative search according to the device parameter characteristics to obtain a second test solution, and recording the number of iterations; According to the preset evaluation indicators, determine whether the evaluation indicators of the second test scheme are better than the evaluation indicators of the first test scheme, and if so, update the second test scheme to the first test scheme; if not, abandon the second test scheme, wherein the preset evaluation indicators include test time and defect detection rate; Continue to the next iteration, and when the number of iterations is greater than or equal to the preset number of iterations, output the current first test plan as the optimal test plan; Wherein, the predicted data are performance indicators obtained when testing the scheme, including defect detection rate and test time; The step of performing performance prediction on the test scheme based on the performance prediction model to obtain prediction data, and determining the optimal test scheme according to the prediction data, includes: Judging the predicted data according to a preset error threshold; When the error of the predicted data is greater than a preset error threshold, the test scheme is updated using the adaptive large neighborhood search model; When the error of the predicted data is less than the preset error threshold, the test scheme is determined to be the optimal test scheme.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored computer program, wherein when the computer program is executed, the device where the computer-readable storage medium is located is controlled to execute the semiconductor device testing method according to any one of claims 1 to 6.

Images