A semiconductor device detection method and system based on collected data

Abstract

The present invention discloses a semiconductor device detection method and system based on collected data, relating to the semiconductor field. The method includes collecting real-time data; obtaining the classified real-time data and a list of risk variables; obtaining a comprehensive risk index, and using an intelligent algorithm to output parameter indicators and resource allocations during the semiconductor device detection process; determining whether the comprehensive risk index exceeds an alarm threshold, and if it exceeds the threshold, sending an alarm signal, and performing a personalized detection process and environment setting according to the generated parameter indicators and resource allocations during the semiconductor device detection process; detecting the semiconductor device according to the personalized detection process and environment setting; and the system includes a real-time data collection module, a risk variable acquisition module, a recommendation module, a setting module, and a detection module. The present invention ensures the safety of the detection process through real-time risk assessment, while the personalized detection process recommendation improves the adaptability and flexibility of the detection process.

Description

Technical Field

[0001] The present invention relates to the field of semiconductors, and more particularly, to a method and system for detecting semiconductor devices based on collected data. Background Art

[0002] The semiconductor industry provides basic electronic components for electronic devices. With the development of technology, the demand for semiconductor devices is increasing day by day, covering multiple fields such as smartphones, computers, automobiles, industrial equipment, and medical equipment. Semiconductor devices include various types of transistors, integrated circuits (ICs), diodes, microprocessors, etc. These products need to go through multiple steps during the manufacturing process, including material preparation, lithography, etching, doping, chemical vapor deposition, physical vapor deposition, metallization, packaging, etc. Each step may introduce defects or affect the device performance, so strict detection is required.

[0003] Semiconductor device detection refers to a series of detection work carried out on semiconductor devices during or after the production process to ensure the performance, reliability, and compliance with design standards of the devices. These detections usually include electrical performance testing, physical defect detection, chemical composition analysis, etc. Semiconductor device detection is crucial for ensuring product quality and reliability because it can help identify and eliminate defective products, reduce rework and return rates, and protect the brand reputation.

[0004] Existing semiconductor device detection technologies, for example, Chinese Patent No. 201611217408.X discloses a device and method for detecting semiconductor device failures. It compares the converted low voltage with a preset first comparison threshold and a second comparison threshold, and judges whether the semiconductor device fails according to the comparison result and preset conditions in the judgment module. However, the above method still has the following deficiencies: relying solely on voltage data cannot comprehensively reflect the health status and potential risks of the device. If the response to risks is not rapid enough, it may not be possible to take timely measures before the failure causes system damage. At the same time, the preference information of the detection engineer is not considered, so a personalized detection process recommendation cannot be provided.

[0005] Regarding the problems in the related art, no effective solution has been proposed yet. Summary of the Invention

[0006] In view of the problems in the related art, the present invention proposes a method and system for detecting semiconductor devices based on collected data to overcome the above-mentioned technical problems existing in the existing related technologies.

[0007] To this end, the specific technical solutions adopted by the present invention are as follows:

[0008] According to one aspect of the present invention, there is provided a method for detecting semiconductor devices based on collected data. The method for detecting semiconductor devices based on collected data includes the following steps:

[0009] S1. Collect physical and electrical characteristic data of the semiconductor device to be tested, detection data of the same type of semiconductor device in history, and preference information of the detection engineer, and combine them to obtain real-time data.

[0010] S2. Classify the real-time data into risk variables within the normal operation range and those within the abnormal range, and obtain the classified real-time data and the list of risk variables.

[0011] S3. Calculate the potential failure probability, failure severity score, and emergency response index of the risk variables, and obtain the comprehensive risk index, and use an intelligent algorithm to output the parameter indicators and resource allocation in the process of detecting semiconductor devices.

[0012] S4. Determine whether the comprehensive risk index exceeds the alarm threshold. If it exceeds the threshold, send an alarm signal, and perform personalized detection process and environment setting according to the generated parameter indicators and resource allocation in the process of detecting semiconductor devices.

[0013] S5. Detect the semiconductor device according to the personalized detection process and environment setting, and generate a detection report.

[0014] Classifying the real-time data into risk variables within the normal operation range and those within the abnormal range, and obtaining the classified real-time data and the list of risk variables includes:

[0015] S21. Determine the normal operation range of each physical and electrical parameter of the semiconductor device;

[0016] S22. Set an abnormal threshold for identifying risk variables that exceed the normal operation range;

[0017] S23. Perform trend analysis on the data of risk variables that exceed the normal operation range.

[0018] Furthermore, calculating the potential failure probability, failure severity score, and emergency response index of the risk variables, and obtaining the comprehensive risk index, and using an intelligent algorithm to output the parameter indicators and resource allocation in the process of detecting semiconductor devices includes the following steps:

[0019] S31. Collect real-time data of each risk variable, obtain the change trend of the real-time data, and predict the future state of the risk variable by establishing a statistical model to obtain the potential failure probability of each risk variable occurring within a preset future time period;

[0020] S32. Define different fault levels according to the impact of risk variables on the performance of semiconductor devices, and assign a scoring value to each fault level to represent the degree of impact of the fault on overall production and performance;

[0021] S33. Evaluate the severity of the fault based on the current state and predicted future state of the risk variable, and give the corresponding fault severity score;

[0022] S34. Based on the impact on the operation of the production line after the fault occurs, evaluate the duration required to take measures to solve the problem, and assign a time-related metric value to each risk variable according to the urgency of fault handling;

[0023] S35. Combine the potential fault probability, fault severity score, and time-related metric value of the risk variable to calculate the emergency response index to guide the priority of response measures;

[0024] S36. Assign weights to the potential fault probability, fault severity score, and emergency response index to reflect their relative importance in risk assessment;

[0025] S37. Add the weighted values of the potential fault probability, fault severity score, and emergency response index to obtain the comprehensive risk index of each risk variable, and sort the comprehensive risk indices of all risk variables to identify the risk variables of concern;

[0026] S38. Based on the comprehensive risk index of the risk variable and the preference information of the detection engineer, use an intelligent algorithm to generate parameter indicators and resource allocations in the semiconductor device detection process.

[0027] Furthermore, when calculating the emergency response index by combining the potential fault probability, fault severity score, and time-related metric value of the risk variable, the calculation formula is:

[0028] ;

[0029] In the formula, URI represents the emergency response index;

[0030] P represents the potential fault probability, S represents the fault severity score, T represents the time-related metric value, P threshold represents the threshold of the potential fault probability;

[0031] a , β , γ represent the non-linear weight exponents of the potential fault probability, fault severity score, and time-related metric value respectively.

[0032] Further, based on the comprehensive risk index of risk variables and the preference information of inspection engineers, using an intelligent algorithm to generate parameter indicators and resource allocation in the semiconductor device inspection process includes the following steps:

[0033] S381. Standardize the data in the comprehensive risk index and the preference information of inspection engineers, and select features related to the parameter indicators and resource allocation in the semiconductor device inspection process to obtain a dataset to be processed;

[0034] S382. Apply vector orthogonalization processing to convert the dataset to be processed into several independent principal components;

[0035] S383. Determine the optimal number of nearest neighbors, and randomly divide the dataset to be processed into a training set and a test set, and ensure that both the training set and the test set contain the comprehensive risk index and the preference information of inspection engineers, as well as the corresponding parameter indicators and resource allocation parameters;

[0036] S384. Use the features after vector orthogonalization and the optimal number of nearest neighbors, apply the K-nearest neighbor algorithm for training, and at each prediction, add new data to the training set to obtain a trained recommendation model;

[0037] S385. For the new comprehensive risk index and the preference information of inspection engineers, use the trained recommendation model for prediction;

[0038] S386. Generate targeted parameter indicators and resource allocation in the semiconductor device inspection process according to the prediction results of the recommendation model.

[0039] Further, after generating the targeted parameter indicators and resource allocation in the semiconductor device inspection process, the following steps are included:

[0040] Use the test set to evaluate the recommendation effect of the recommendation model;

[0041] Adjust the number of nearest neighbors or reselect features according to the evaluation results to optimize the recommendation model;

[0042] Deploy the optimized recommendation model to the inspection engineering environment.

[0043] Further, determining the optimal number of nearest neighbors includes the following steps:

[0044] S3831. Randomly select several data objects in the dataset to be processed as initial clustering centers to form m agents, and initialize the speed and position of the agents;

[0045] S3832. Calculate the distance from all data objects in each agent to the clustering center, and classify the data according to the nearest neighbor distance principle;

[0046] S3833. Calculate the fitness value of each agent, compare the fitness values, and update the optimal position of the agent.

[0047] S3834. Compare the fitness values of all agents, update the global extreme value, and update the speed and position of the agents.

[0048] S3835. Repeat S3832 - S3834 until a sufficiently good position is reached or the maximum number of iterations is stopped.

[0049] S3836. Output the optimal clustering center and determine the optimal number of nearest neighbors.

[0050] Furthermore, outputting the optimal clustering center and determining the optimal number of nearest neighbors includes the following steps:

[0051] S38361. Take the clustering center corresponding to the global optimal solution as the optimal clustering center.

[0052] S38362. Determine the optimal number of nearest neighbors through several runs. In each run, use a different number of nearest neighbors, compare the results of each run, and select the optimal number of nearest neighbors by evaluating the clustering quality.

[0053] Furthermore, personalized detection process and environment settings according to the generated parameter indicators and resource allocation in the semiconductor device detection process include the following steps:

[0054] S41. Allocate personnel and detection resources according to the generated parameter indicators and resource allocation in the semiconductor device detection process.

[0055] S42. Set the detection process and environment.

[0056] S43. Convert the data of the detection process and environment settings into a visual format data.

[0057] S44. Select a visualization tool and technology platform, bind the visual format data with visual elements, and ensure that users interact with the visual format data through the interface.

[0058] Furthermore, setting the detection process and environment includes the following steps:

[0059] S421. Set the detection environment of the semiconductor device.

[0060] S422. Set the parameters of the detection equipment.

[0061] S423. Determine the test steps, sequence, and time arrangement.

[0062] Furthermore, the semiconductor device is detected according to the personalized detection process and environmental settings, and a detection report is generated, including:

[0063] S51. Detect the same type of semiconductor devices according to the personalized detection process and environmental settings;

[0064] S52. Collect the detection data of the semiconductor devices, and identify faults and anomalies in the detection data of the semiconductor devices;

[0065] S53. Generate a detection report based on the results of fault and anomaly identification.

[0066] According to another aspect of the present invention, a semiconductor device detection system based on collected data is provided. The semiconductor device detection system based on collected data includes a real-time data collection module, a risk variable acquisition module, a recommendation module, a setting module, and a detection module.

[0067] Among them, the real-time data collection module is used to collect the physical and electrical characteristic data of the semiconductor device to be tested, the detection data of the same type of semiconductor devices in history, and the preference information of the detection engineer, and combine them to obtain real-time data.

[0068] The risk variable acquisition module is used to classify the real-time data into risk variables within the normal operation range and the abnormal range, and obtain the classified real-time data and the list of risk variables.

[0069] The recommendation module is used to calculate the potential failure probability, failure severity score, and emergency response index of the risk variables, obtain the comprehensive risk index, and output the parameter indicators and resource allocation in the semiconductor device detection process using an intelligent algorithm.

[0070] The setting module is used to determine whether the comprehensive risk index exceeds the alarm threshold. If it exceeds the threshold, an alarm signal is sent, and the personalized detection process and environmental settings are performed according to the parameter indicators and resource allocation in the semiconductor device detection process generated.

[0071] The detection module is used to detect the semiconductor device according to the personalized detection process and environmental settings, and generate a detection report.

[0072] The beneficial effects of the present invention are:

[0073] (1) A semiconductor device detection method and system provided by the present invention ensure the safety of the detection process through real-time risk assessment, while the personalized detection process recommendation improves the adaptability and flexibility of the detection process.

[0074] (2) By dividing the real-time data into risk variables within normal and abnormal ranges, the risk variables that require special attention and further analysis are determined. Calculate the potential failure probability, failure severity, and emergency response indicators of the risk variables, generate a comprehensive risk index, and provide it as input to the intelligent algorithm to optimize the detection parameters and resource allocation. At the same time, conduct actual semiconductor device detection according to the personalized detection process and environmental settings, and generate a detection report to provide a basis for quality control and subsequent decision-making.

[0075] (3) When using the intelligent algorithm to generate parameter indicators and resource allocation in the semiconductor device detection process based on the comprehensive risk index of risk variables and the preference information of detection engineers, this invention standardizes and performs principal component analysis on the data to reduce dimensions and improve the model performance. Through optimization techniques, determine the optimal K value, split the dataset, train the recommendation model, and make predictions, which improves the prediction accuracy and applicability of the recommendation model.

[0076] (4) By allocating personnel and resources, setting the detection process and environment, converting the setting data into a visual format, and presenting the detection process using a visualization tool, this invention enhances the user's interaction experience and the transparency of the process. Description of the Drawings

[0077] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0078] Figure 1 is a flowchart of a semiconductor device detection method based on collected data according to an embodiment of the present invention;

[0079] Figure 2 is a schematic block diagram of a semiconductor device detection system based on collected data according to an embodiment of the present invention.

[0080] In the figure:

[0081] 1. Real-time data collection module; 2. Risk variable acquisition module; 3. Recommendation module; 4. Setting module; 5. Detection module. Detailed Embodiments

[0082] To further illustrate the embodiments, the present invention provides accompanying drawings, which are part of the disclosure of the present invention. These drawings are mainly used to illustrate the embodiments and can be combined with the relevant descriptions in the specification to explain the operating principles of the embodiments. With reference to these contents, those of ordinary skill in the art should be able to understand other possible implementation manners and the advantages of the present invention. The components in the drawings are not drawn to scale, and similar component symbols are generally used to represent similar components.

[0083] According to an embodiment of the present invention, a method and system for detecting semiconductor devices based on collected data are provided.

[0084] Now, the present invention will be further described in conjunction with the accompanying drawings and specific implementation manners. As Figure 1 shown, according to an embodiment of the present invention, a method for detecting semiconductor devices based on collected data is provided. The method for detecting semiconductor devices based on collected data includes the following steps:

[0085] S1. Collect the physical and electrical characteristic data of the semiconductor device to be tested, the detection data of the same type of semiconductor devices in history, and the preference information of the detection engineer, and combine them to obtain real-time data.

[0086] Among them, measure and record the physical parameters of the semiconductor device, such as size, weight, material composition, etc. Measure the electrical characteristics of the semiconductor device through testing equipment, such as resistance, capacitance, leakage current, threshold voltage, etc.

[0087] Extract the historical detection data of the same type of devices from the database or record file. Communicate with the detection engineer to understand their preferences and requirements for aspects such as the detection process and parameter settings.

[0088] For example, when testing a batch of newly produced microprocessors, collect their physical and electrical data such as current and voltage, and refer to the test data of similar microprocessors in the past. At the same time, the detection engineer provides information about test preferences, such as they are more concerned about current stability.

[0089] S2. Classify the real-time data into risk variables within the normal operation range and within the abnormal range to obtain the classified real-time data and the list of risk variables. Specifically, it includes the following steps:

[0090] Determine the normal operation range of each physical and electrical parameter of the semiconductor device. The range is based on design specifications, manufacturer data, historical performance data, or industry standards.

[0091] Set an abnormal threshold for identifying risk variables that exceed the normal operation range. Screen the real-time data, and label each data point as normal or abnormal according to whether it falls within the predefined normal operation range.

[0092] Identify key risk variables from the abnormal data, where these risk variables indicate early signs of failures or performance issues. And perform trend analysis on the data within the abnormal range to identify whether there are persistent or deteriorating problems.

[0093] In a further embodiment, classify the real-time data into risk variables within the normal operating range and within the abnormal range, and the classified real-time data and the list of risk variables include:

[0094] S21. Determine the normal operating range of each physical and electrical parameter of the semiconductor device;

[0095] S22. Set an abnormal threshold for identifying risk variables outside the normal operating range;

[0096] S23. Perform trend analysis on the data of risk variables outside the normal operating range.

[0097] S3. Calculate the potential failure probability, failure severity score, and emergency response index of the risk variables, and obtain a comprehensive risk index, and use an intelligent algorithm to output the parameter indicators and resource allocation in the semiconductor device detection process.

[0098] In a further embodiment, calculating the potential failure probability, failure severity score, and emergency response index of the risk variables, and obtaining a comprehensive risk index, and using an intelligent algorithm to output the parameter indicators and resource allocation in the semiconductor device detection process includes the following steps:

[0099] S31. Collect the real-time data of each risk variable, obtain the trend of the real-time data change, and predict the future state of the risk variable by establishing a statistical model, and obtain the potential failure probability of each risk variable occurring within a preset time period in the future.

[0100] Among them, the statistical model is a mathematical model that uses historical data to analyze the relationships between variables and predicts future events or behaviors through these relationships. In semiconductor device detection, the statistical model is used to predict whether specific risk variables (such as current, voltage, temperature, etc.) may become abnormal in the future.

[0101] Select a suitable statistical model to analyze the risk variables. Commonly included are linear regression, logistic regression, time series analysis, survival analysis, etc. And train and validate the statistical model. For each risk variable, the statistical model will give a predicted value.

[0102] For example, to predict whether a semiconductor device will fail within a certain period in the future, the statistical model calculates the probability of failure based on the current values of the risk variables. If this probability exceeds the threshold, the risk of the device becoming abnormal within a certain period in the future is relatively high.

[0103] S32. Define different fault levels (such as minor, medium, severe) according to the impact of risk variables on the performance of semiconductor devices, and assign a scoring value to each fault level to indicate the degree of impact of the fault on overall production and performance;

[0104] Among them, when defining different fault levels according to the impact of risk variables on the performance of semiconductor devices, determine the performance standards and key performance indicators of semiconductor devices, including power loss, switching frequency, maximum current-carrying capacity, thermal stability, etc. Evaluate the degree of impact of risk variables (such as voltage fluctuations, temperature increases, current abnormalities, etc.) on the above performance standards. Conduct a failure mode and effects analysis to determine the impact of different failure modes on device performance. Develop a grading standard based on the severity of the impact of risk variables on device performance. For example, a minor fault may only cause a slight decline in performance, while a severe fault may cause the device to completely fail. Define fault levels, such as minor (the device performance slightly decreases, but does not affect the main functions), medium (the performance significantly decreases, and may affect some functions), severe (the device fails or there is a safety risk).

[0105] S33. Evaluate the severity of the fault based on the current state and predicted future state of the risk variable, and give the corresponding fault severity score;

[0106] Among them, determine the corresponding fault severity level according to the current and predicted states of the risk variable. For example, if the risk variable shows a continuous deteriorating trend, it may be rated as a medium or severe fault. Assign a scoring value to each fault severity level. This score should reflect the potential impact of the fault. For example, a minor fault may be assigned a low score, while a severe fault is assigned a high score.

[0107] S34. Based on the impact on the operation of the production line after the fault occurs, evaluate the duration of measures needed to solve the problem, and assign a time-related metric value to each risk variable according to the urgency of fault handling;

[0108] Among them, estimate the actual time required to solve the problem (such as repairing or replacing components) according to the nature and severity of the fault. Consider the necessary diagnosis, obtaining replacement components, actual repair, and test time. Assign a time-related metric value to each risk variable, which reflects the time required from discovering the problem to completely solving the problem. For example, labels such as "urgent", "high priority", "medium priority", and "low priority" can be used to indicate.

[0109] S35. Combine the potential failure probability, fault severity score, and time-related metric value of the risk variable to calculate an emergency response indicator to guide the priority of response measures.

[0110] When calculating the emergency response index by combining the potential failure probability of risk variables, the failure severity score, and the time-related index value, the calculation formula is:

[0111] ;

[0112] In the formula, URI represents the emergency response index;

[0113] P represents the potential failure probability, S represents the failure severity score, T represents the time-related index value, P threshold represents the threshold value of the potential failure probability;

[0114] a , β , γ respectively represent the non-linear weight exponents of the potential failure probability, the failure severity score, and the time-related index value.

[0115] S36. Assign weights to the potential failure probability, the failure severity score, and the emergency response index to reflect their relative importance in risk assessment.

[0116] S37. Add the weighted values of the potential failure probability, the failure severity score, and the emergency response index to obtain the comprehensive risk index of each risk variable, and sort the comprehensive risk indices of all risk variables to identify the risk variables that require special attention.

[0117] S38. Based on the comprehensive risk index of risk variables and the preference information of the inspection engineer, use an intelligent algorithm to generate the parameter indicators and resource allocation in the semiconductor device inspection process.

[0118] In a further embodiment, based on the comprehensive risk index of risk variables and the preference information of the inspection engineer, using an intelligent algorithm to generate the parameter indicators and resource allocation in the semiconductor device inspection process includes the following steps:

[0119] S381. Standardize the data in the comprehensive risk index and the preference information of the inspection engineer, and select the features related to the parameter indicators and resource allocation in the semiconductor device inspection process to obtain the dataset to be processed. For example, select the equipment failure rate, the historical accident frequency, the engineer's sensitivity to cost, etc. as features.

[0120] S382. Apply vector orthogonalization processing to convert the dataset to be processed into several independent principal components; principal component analysis can identify the main variation directions in the data and project the original data onto these directions to form a new set of linearly independent features. The new features are called principal components and are used to replace the original features for model training and prediction, which usually can improve the efficiency and performance of the algorithm.

[0121] S383. Determine the optimal number of nearest neighbors (K value), and randomly divide the dataset to be processed into a training set and a test set, and ensure that both the training set and the test set contain the comprehensive risk index and the preferences of the inspection engineer, as well as the corresponding parameter indicators and resource allocation parameters.

[0122] S384. Use the features after vector orthogonalization and the optimal number of nearest neighbors, and apply the K-nearest neighbor algorithm for training. And in each prediction, add the new data to the training set to obtain the trained recommendation model. The K-nearest neighbor algorithm is instance-based learning, which predicts the classification or regression value of the new sample according to the similarity between the new sample and the samples in the training set.

[0123] S385. For the new comprehensive risk index and the preferences of the inspection engineer, use the trained recommendation model for prediction; utilize the historical cases that are most similar to the new input predicted, and recommend parameter indicators and resource allocation accordingly.

[0124] S386. Generate the parameter indicators and resource allocation in the semiconductor device detection process according to the prediction results of the recommendation model. Considering the comprehensive risk index and the preferences of the inspection engineer, optimize the detection process to improve efficiency and effectiveness.

[0125] By calculating the potential failure probability, potential problems that may occur in semiconductor devices can be identified in advance, so as to take preventive measures before the actual occurrence of the failure, avoid potential risks from evolving into actual failures, and reduce downtime and economic losses.

[0126] The failure severity score can help quantify the impact degree of semiconductor device failures on the production process, making the detection process more targeted. This helps to prioritize those problems that have the greatest impact on production and ensure that resources are reasonably allocated to the most critical tasks.

[0127] The emergency response index can guide the priority of response measures according to the urgency of semiconductor device failure handling. This means that problems that need to be resolved immediately can be responded to faster, reducing the impact on the operation of the production line.

[0128] In a further embodiment, after generating the parameter indicators and resource allocation in the semiconductor device detection process, the following steps are included:

[0129] Evaluate the recommendation effect of the recommendation model using the test set; use appropriate evaluation metrics, such as accuracy, recall, F1 score, ROC curve, etc.

[0130] Adjust the number of nearest neighbors or reselect features according to the evaluation results to optimize the recommendation model; feature selection is based on the understanding of the problem and the importance of features, and use feature selection algorithms or feature importance scores to assist in the selection.

[0131] Deploy the optimized recommendation model to the detection engineering environment, for example, integrate it into the working platform of detection engineers.

[0132] In a further embodiment, determining the optimal number of nearest neighbors includes the following steps:

[0133] S3831: Randomly select several data objects in the dataset to be processed as initial clustering centers, form m agents, and initialize the speed and position of the agents; the position of the agent represents a possible solution (the position of the clustering center).

[0134] S3832: Calculate the distance from all data objects in each agent to the clustering center, and classify the data according to the nearest neighbor distance principle, that is, each data point is assigned to the class represented by its nearest clustering center.

[0135] S3833: Calculate the fitness value of each agent, and compare the fitness values to update the optimal position of the agent; the fitness value is based on a pre-defined evaluation criterion (such as minimizing the distance within the cluster), and update the optimal position of each agent according to the fitness value.

[0136] S3834: Compare the fitness values of all agents, update the global extreme value, and update the speed and position of the agents to explore new possible solutions in the next iteration.

[0137] S3835: Repeat S3832 - S3834 until a sufficiently good position is reached or the maximum number of iterations stops.

[0138] S3836: Output the optimal clustering center and determine the optimal number of nearest neighbors.

[0139] In a further embodiment, outputting the optimal clustering center and determining the optimal number of nearest neighbors includes the following steps:

[0140] S38361. Use the cluster center corresponding to the global optimal solution as the optimal cluster center; S38362. Determine the optimal number of nearest neighbors through multiple runs, using a different number of nearest neighbors each time, comparing the results of each run, and selecting the optimal number of nearest neighbors by evaluating the clustering quality, such as using the Silhouette Coefficient or other clustering validity metrics.

[0141] S4. Determine whether the comprehensive risk index exceeds the alarm threshold. If it exceeds the threshold, send an alarm signal, and perform personalized detection process and environment settings based on the generated parameter indicators and resource allocations during the semiconductor device detection process.

[0142] In a further embodiment, performing personalized detection process and environment settings based on the generated parameter indicators and resource allocations during the semiconductor device detection process includes the following steps:

[0143] S41. Allocate personnel and detection resources according to the generated parameter indicators and resource allocations during the semiconductor device detection process; comprehensively consider the complexity of the detection tasks, the availability of detection resources (such as equipment, instruments, and tools), and the skills and experience of personnel to ensure that each detection task has appropriate personnel and resource support.

[0144] S42. Set the detection process and environment.

[0145] S43. Convert the data of the detection process and environment settings into a visual format data, such as JSON, CSV, or XML, etc., to ensure that the data can be effectively processed by visualization tools and technology platforms.

[0146] S44. Select visualization tools and technology platforms, such as Tableau, PowerBI, D3.js, etc., to bind the visual format data with visual elements and ensure that users can interact with the visual format data through the interface.

[0147] In a further embodiment, setting the detection process and environment includes the following steps:

[0148] S421. Set the detection environment of the semiconductor device, including temperature, humidity, clean room conditions, etc.

[0149] S422. Set the parameters of the detection equipment, such as voltage, current, test frequency, etc.

[0150] S423. Determine the test steps, sequence, and time arrangement.

[0151] S5. Detect the semiconductor device according to the personalized detection process and environment settings and generate a detection report.

[0152] Among them, analyze the test results. If there is a fault, determine the cause of the fault. The test report is reviewed by the person in charge to ensure the accuracy and integrity of the information. Submit the test report to the relevant engineers, project management team or customers to ensure that all stakeholders are aware of the test results.

[0153] In a further embodiment, the semiconductor device is tested according to a personalized test process and environmental settings, and a test report is generated, including:

[0154] S51. Test semiconductor devices of the same type according to the personalized test process and environmental settings.

[0155] S52. Collect semiconductor device test data and identify faults and anomalies in the semiconductor device test data;

[0156] Specifically, the electrical characteristic data of the semiconductor device collected after the personalized test process and environmental settings. Retrieve and record the corresponding test data of the same type of semiconductor device previously to facilitate comparison with the current data.

[0157] Determine the normal operating range, and define the normal operating range of each electrical parameter based on design specifications, manufacturer data, historical performance data or industry standards.

[0158] Screen the collected real-time data, and label each data point as normal or abnormal according to whether it is within the predefined normal operating range.

[0159] S53. Generate a test report based on the results of fault and anomaly identification.

[0160] As Figure 2 shown, according to another embodiment of the present invention, a semiconductor device detection system based on collected data is provided. The semiconductor device detection system based on collected data includes a real-time data collection module 1, a risk variable acquisition module 2, a recommendation module 3, a setting module 4 and a detection module 5;

[0161] Among them, the real-time data collection module 1 is used to collect the physical and electrical characteristic data of the semiconductor device to be tested, the test data of the same type of semiconductor device in history, and the preference information of the test engineer, and combine them to obtain real-time data.

[0162] The risk variable acquisition module 2 is used to classify the real-time data into risk variables within the normal operating range and outside the normal operating range, and obtain the classified real-time data and the list of risk variables.

[0163] A recommendation module 3 is configured to calculate the potential failure probability, failure severity score, and emergency response metrics of risk variables, obtain a comprehensive risk index, and output parameter metrics and resource allocation in the semiconductor device detection process using an intelligent algorithm.

[0164] A setting module 4 is configured to determine whether the comprehensive risk index exceeds an alarm threshold. If it exceeds the threshold, an alarm signal is sent, and a personalized detection process and environment are set according to the parameter metrics and resource allocation in the generated semiconductor device detection process.

[0165] A detection module 5 is configured to detect the semiconductor device according to the personalized detection process and environment settings and generate a detection report.

[0166] In summary, a semiconductor device detection method and system provided by the present invention ensure the safety of the detection process through real-time risk assessment, while the personalized detection process recommendation improves the adaptability and flexibility of the detection process. By dividing real-time data into risk variables within normal and abnormal ranges, risk variables that require special attention and further analysis are determined. The potential failure probability, failure severity, and emergency response metrics of the risk variables are calculated to generate a comprehensive risk index, which provides input for an intelligent algorithm to optimize detection parameters and resource allocation. At the same time, the actual semiconductor device detection is performed according to the personalized detection process and environment settings, and a detection report is generated to provide a basis for quality control and subsequent decision-making. Based on the comprehensive risk index of risk variables and the preference information of detection engineers, when using an intelligent algorithm to generate parameter metrics and resource allocation in the semiconductor device detection process, the data is standardized and principal component analysis is performed to reduce dimensions and improve the model performance. Through optimization techniques, the optimal K value is determined, the data set is segmented, a recommendation model is trained, and predictions are made to improve the prediction accuracy and applicability of the recommendation model. The present invention enhances the user's interaction experience and the transparency of the process by allocating personnel and resources, setting the detection process and environment, converting the setting data into a visual format, and presenting the detection process using a visualization tool.

[0167] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A semiconductor device detection method based on collected data, characterized in that: include: S1. Collect physical and electrical characteristic data of the semiconductor device to be tested, historical test data of the same type of semiconductor devices, and preference information of test engineers, and combine them to obtain real-time data; S2. Classify the real-time data into risk variables within the normal operating range and the abnormal range, and obtain a classified list of real-time data and risk variables, including: S21, determining the normal operating range of various physical and electrical parameters of the semiconductor device; S22. Setting abnormal thresholds to identify risk variables that exceed normal operating ranges; S23. Conduct trend analysis on data of risk variables that are beyond the normal operating range; S3. Calculate the potential failure probability, failure severity score and emergency response index of the risk variable, and obtain a comprehensive risk index, and use an intelligent algorithm to output parameter indicators and resource allocation in the semiconductor device detection process, where: Combining the potential failure probability, failure severity score and time-related index values ​​of the risk variables, the calculation formula for the emergency response index is: ； In the formula, URI represents an emergency response indicator; P represents the potential failure probability, S represents the fault severity score, T Indicates the time-related indicator value, P threshold A threshold value representing the probability of potential failure; a , β , γ The nonlinear weight indexes representing the potential failure probability, failure severity score and time-related index value respectively; S4. Determine whether the comprehensive risk index exceeds the alarm threshold. If it exceeds the threshold, an alarm signal is issued, and a personalized detection process and environment setting are performed according to the parameter indicators and resource configuration generated in the semiconductor device detection process; S5. Test the semiconductor device according to the personalized test process and environment settings, and generate a test report.

2. A semiconductor device detection method based on collected data according to claim 1, characterized in that: The method of calculating the potential failure probability, failure severity score and emergency response index of the risk variables and obtaining a comprehensive risk index, and outputting parameter indicators and resource allocation in the semiconductor device detection process using an intelligent algorithm also includes the following steps: S31. Collect the real-time data of each risk variable, obtain the change trend of the real-time data, and predict the future state of the risk variable by establishing a statistical model to obtain the potential failure probability of each risk variable in a pre-set time period in the future; S32. Define different fault levels based on the impact of risk variables on semiconductor device performance, and assign a score value to each fault level to indicate the degree of impact of the fault on overall production and performance; S33. Evaluate the severity of the fault based on the current state and predicted future state of the risk variable, and give a corresponding fault severity score; S34. Based on the impact of the fault on the production line operation, evaluate the time required to take measures to solve the problem, and assign time-related indicator values ​​to each risk variable based on the urgency of fault handling; S35. Combine the potential failure probability, failure severity score and time-related index values ​​of the risk variables to calculate the emergency response index to guide the priority of the response measures; S36. Assign weights to potential failure probabilities, failure severity scores, and emergency response indicators to reflect their relative importance in the risk assessment; S37, adding the weighted values ​​of the potential failure probability, the failure severity score and the emergency response index to obtain a comprehensive risk index for each risk variable, and sorting the comprehensive risk indexes of all risk variables to identify the risk variables of concern; S38. Based on the comprehensive risk index of risk variables and the preference information of inspection engineers, an intelligent algorithm is used to generate parameter indicators and resource allocation in the semiconductor device inspection process.

3. A semiconductor device detection method based on collected data according to claim 2, characterized in that: The method of using an intelligent algorithm to generate parameter indicators and resource allocation in a semiconductor device testing process based on a comprehensive risk index of risk variables and preference information of testing engineers includes the following steps: S381. Standardize the data in the comprehensive risk index and the preference information of the test engineer, and select features related to parameter indicators and resource allocation in the semiconductor device testing process to obtain a data set to be processed; S382, applying vector orthogonalization processing to convert the data set to be processed into a number of independent principal components; S383. Determine the optimal number of nearest neighbors, and randomly divide the data set to be processed into a training set and a test set, and ensure that both the training set and the test set contain the comprehensive risk index and the preference information of the inspection engineer, as well as the corresponding parameter indicators and resource configuration parameters; S384, using the features after vector orthogonalization and the optimal number of nearest neighbors, applying the K nearest neighbor algorithm for training, and adding new data to the training set at each prediction, and obtaining a recommended model after training; S385. For the new comprehensive risk index and the preference information of the testing engineer, use the trained recommendation model to make predictions; S386. Generate targeted parameter indicators and resource configurations in the semiconductor device testing process based on the prediction results of the recommended model.

4. A semiconductor device detection method based on collected data according to claim 3, characterized in that: The generation of parameter indicators and resource configuration in the targeted semiconductor device detection process includes the following steps: Use the test set to evaluate the recommendation effect of the recommendation model; Adjust the nearest neighbor value or reselect features based on the evaluation results to optimize the recommendation model; Deploy the optimized recommendation model into the detection engineering environment.

5. A semiconductor device detection method based on collected data according to claim 4, characterized in that: Determining the optimal nearest neighbor quantity value comprises the following steps: S3831, randomly selecting a number of data objects in the data set to be processed as initial cluster centers, forming m agents, and initializing the speed and position of the agents; S3832, calculating the distance from all data objects in each agent to the cluster center, and classifying the data according to the nearest neighbor distance principle; S3833, calculating the fitness value of each agent, and comparing the fitness values, and updating the optimal position of the agent; S3834, comparing the fitness values ​​of all agents, updating the global extreme value, and updating the speed and position of the agent; S3835, repeat S3832-S3834 until a good enough position is reached or the maximum number of iterations is reached; S3836. Output the optimal cluster center and determine the optimal number of nearest neighbors.

6. A semiconductor device detection method based on collected data according to claim 5, characterized in that: The outputting of the optimal cluster center and determining the optimal number of nearest neighbors includes the following steps: S38361. Taking the cluster center corresponding to the global optimal solution as the optimal cluster center; S38362. Determine the optimal number of nearest neighbors by running several times, using a different number of nearest neighbors each time, and compare the results of each run, and select the optimal number of nearest neighbors by evaluating the clustering quality.

7. A semiconductor device detection method based on collected data according to claim 6, characterized in that: The personalized testing process and environment setting according to the parameter indicators and resource configuration generated in the semiconductor device testing process includes the following steps: S41, allocating personnel and testing resources according to the generated parameter indicators and resource configuration in the semiconductor device testing process; S42, setting up the testing process and environment; S43, converting the data of the detection process and the environment setting into a data format to obtain visual format data; S44. Select visualization tools and technology platforms to achieve the binding of visualization format data with visual elements and ensure that users can interact with visualization format data through the interface; The setting of the detection process and environment includes the following steps: S421, setting a detection environment for semiconductor devices; S422, setting parameters of the detection equipment; S423. Determine the test steps, sequence and time schedule.

8. The semiconductor device detection method based on collected data according to claim 7, characterized in that: The testing of semiconductor devices according to the personalized testing process and environment settings and generating a testing report include: S51, testing semiconductor devices of the same type according to personalized testing procedures and environment settings; S52, collecting semiconductor device detection data, and identifying faults and abnormalities of the semiconductor device detection data; S53: Generate a test report based on the fault and abnormality identification results.

9. A semiconductor device detection system based on collected data, characterized in that: Used to implement the semiconductor device detection method based on collected data as described in any one of claims 1 to 8, the semiconductor device detection system based on collected data includes a real-time data collection module, a risk variable acquisition module, a recommendation module, a setting module and a detection module; The real-time data collection module is used to collect the physical and electrical characteristic data of the semiconductor device to be tested, the historical test data of the same type of semiconductor devices and the preference information of the test engineers, and combine them to obtain the real-time data; The risk variable acquisition module is used to classify the real-time data into risk variables within a normal operating range and an abnormal range, and obtain the classified real-time data and a risk variable list; The recommendation module is used to calculate the potential failure probability, failure severity score and emergency response index of the risk variable, and obtain a comprehensive risk index, and use an intelligent algorithm to output parameter indicators and resource configuration in the semiconductor device detection process; The setting module is used to determine whether the comprehensive risk index exceeds the alarm threshold. If it exceeds the threshold, an alarm signal is issued, and personalized detection process and environment settings are performed according to the parameter indicators and resource configuration generated in the semiconductor device detection process; The detection module is used to detect semiconductor devices according to personalized detection procedures and environmental settings, and generate a detection report.

Images