Artificial intelligence failure analysis system and method
By automatically integrating and analyzing test data through an artificial intelligence failure analysis system, and generating failure analysis models, the system solves the problem of low test data processing efficiency in existing technologies, and achieves efficient failure analysis and test process optimization.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- UNIVERSAL GLOBAL TECH KUNSHAN
- Filing Date
- 2026-01-09
- Publication Date
- 2026-04-24
AI Technical Summary
Existing failure analysis techniques require extensive manual recording and organization of test data, making it impossible to obtain test results in real time and to efficiently calculate the mean time between failures.
The AI-powered failure analysis system uses storage modules and processors to generate failure analysis models, automatically integrates and analyzes test data, calculates the mean time between failures (MTBF), and updates the model based on the application domain and worst-case scenario list.
It enables automated failure analysis, reduces testing time and manpower costs, allows for real-time acquisition of test results and dynamic adjustment of the testing process, and improves testing efficiency and accuracy.
Smart Images
Figure CN121920218A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of failure analysis technology, and more particularly to an artificial intelligence failure analysis system and method thereof. Background Technology
[0002] During the New Product Introduction phase, electronic components have corresponding operating range specifications defined based on their circuit characteristics and features. These operating range specifications are primarily determined by performing multi-parameter boundary tests to obtain the failure characteristics and related feature parameters at the limit points, followed by further life testing of the product. Life testing can be used to obtain the mean time between failures (MTBF) and failure rate (FIT) of electronic components.
[0003] Existing failure analysis requires users to obtain and record test data and parameters on-site, and input them into additional analysis software to calculate the mean time between failures. This requires a lot of time and manpower to monitor the test status, and it is impossible to know the status of the test object in real time. After the test, even more time is required to sort out the complex and scattered test data.
[0004] Therefore, it can be seen that there is currently a lack of an artificial intelligence failure analysis system and method in this field that can integrate statistical test data and parameters and calculate the mean failure interval time. As a result, relevant researchers are seeking solutions. Summary of the Invention
[0005] Therefore, the purpose of this application is to provide an artificial intelligence failure analysis system and method, which uses a first, second, and third procedure to find the analysis result data corresponding to the optimal test mode, thereby training a failure analysis model. In this way, the artificial intelligence failure analysis system and method of this application can update the failure analysis model based on the analysis result data, application field, worst-case list, and optimal mode, to achieve automatic acquisition, calculation, and judgment using artificial intelligence.
[0006] According to one embodiment of the system embodiments of this application, an artificial intelligence failure analysis system is provided, including a storage module and a processor. The storage module includes a historical test dataset and multiple test modes. The historical test dataset includes multiple test data sets. The processor is signal-connected to the storage module and configured to perform operations including the following steps: receiving historical test datasets and test modes; executing a first program, wherein the first program includes acquiring at least one of the test data, determining that at least one of the test data is test graphical data or test behavioral data, and testing at least one of the test data through the test modes to generate multiple analysis result data; executing a second program, wherein the second program includes analyzing the historical test dataset to calculate the mean time between failures (MTBF) of the historical test dataset, generating a worst-case list based on the MTBF, and determining the best mode among the test modes; executing a third program, wherein the third program includes identifying the application domain corresponding to the historical test dataset, and analyzing the historical test dataset based on the application domain to generate analysis result data; training the historical test dataset using artificial intelligence methods based on the analysis result data, application domain, worst-case list, and best mode to generate a failure analysis model; and testing the current test data using the failure analysis model to generate test result data corresponding to the current test data for failure analysis of the component.
[0007] An embodiment of the method according to this application provides an artificial intelligence failure analysis method, comprising: a driver processor receiving a historical test dataset and multiple test modes from a storage module, wherein the historical test dataset includes multiple test data; the driver processor executing a first program, wherein the first program includes acquiring at least one of the test data, determining that at least one of the test data is test graphic data or test behavior data, and testing at least one of the test data through the test modes to generate multiple analysis result data; the driver processor executing a second program, wherein the second program includes analyzing the historical test dataset to calculate the average failure interval time of the historical test dataset, generating a worst-case list based on the average failure interval time, and determining the best mode among the test modes; the driver processor executing a third program, wherein the third program includes identifying the application domain corresponding to the historical test dataset, and analyzing the historical test dataset according to the application domain to generate the analysis result data; and the driver processor training the historical test dataset using artificial intelligence methods based on the analysis result data, application domain, worst-case list, and best mode to generate a failure analysis model; and the driver processor testing the current test data through the failure analysis model to generate test result data corresponding to the current test data, for failure analysis of components. Attached Figure Description
[0008] Figure 1 A block diagram of an artificial intelligence failure analysis system according to a first embodiment of this application is shown; Figure 2 A flowchart of the artificial intelligence failure analysis method according to the second embodiment of this application is shown; Figure 3 It shows that according to Figure 2 The flowchart of the first procedure of the artificial intelligence failure analysis method; Figure 4 It shows that according to Figure 2 The flowchart of the second procedure of the artificial intelligence failure analysis method; Figure 5 It shows that according to Figure 2 The flowchart of the third procedure of the artificial intelligence failure analysis method; Figure 6 A flowchart of the artificial intelligence failure analysis method according to the third embodiment of this application is shown.
[0009] Figure reference numerals: 100: Artificial intelligence failure analysis system; 110: Storage module; 111: Historical test dataset; 120: Processor; 200, 200a: Artificial intelligence failure analysis method; M1: Failure analysis model; P1: First program; P2: Second program; P3: Third program; S01, S02, S021, S022, S023, S024, S03, S031, S032, S033, S034, S04, S041, S042, S043, S044, S05, S06, S11, S12, S13, S14, S15, S16, S17: Steps; TM: Test mode. Detailed Implementation
[0010] Please see Figure 1 and Figure 2 , Figure 1 A block diagram of an artificial intelligence failure analysis system 100 according to a first embodiment of this application is shown. Figure 2 A flowchart of an artificial intelligence failure analysis method 200 according to a second embodiment of this application is shown. The artificial intelligence failure analysis system 100 includes a storage module 110 and a processor 120. The storage module 110 includes a historical test dataset 111 and multiple test modes TM. The historical test dataset 111 includes multiple test data. The processor 120 is signal-connected to the storage module 110 and configured to implement the artificial intelligence failure analysis method 200. The artificial intelligence failure analysis method 200 includes steps S01, S02, S03, S04, S05, and S06.
[0011] Specifically, the storage module 110 may be a memory or other storage device; the processor 120 may be a central processing unit (CPU) or other electronic computing device; the historical test dataset 111 may include the mean time between failures, failure rate or other component life test-related parameters corresponding to multiple electronic components under specific voltage, current, stress, impedance, temperature, acceleration, delay, and operating mode conditions; the test mode TM may include specific parameter conditions for specific electronic components, but this application is not limited thereto.
[0012] Step S01 includes receiving a historical test dataset 111 and a test mode TM. Step S02 includes executing a first procedure P1. The first procedure P1 includes acquiring at least one of these test data, determining that at least one of the test data is test graphical data or test behavioral data, and testing at least one of the test data through the test mode TM to generate multiple analysis result data. Step S03 includes executing a second procedure P2. The second procedure P2 includes analyzing the historical test dataset 111 to calculate the average failure interval time of the historical test dataset 111, generating a worst-case list based on the average failure interval time, and determining the optimal mode in the test mode TM. Step S04 includes executing a third procedure P3. The third procedure P3 includes identifying the application domain corresponding to the historical test dataset 111, and analyzing the historical test dataset 111 based on the application domain to generate these analysis result data. Steps S05 and S06 include training the historical test dataset 111 using the analysis result data, application domain, worst-case list, and optimal mode to generate a failure analysis model M1 for failure analysis of components. Thus, the artificial intelligence failure analysis method 200 of this application can automatically obtain the historical test dataset 111 in the storage module 110, and integrate the analysis results data, application fields, worst case list and best mode to update the failure analysis model M1, so as to save testing time and testing manpower costs.
[0013] Specifically, failure analysis may include failure analysis (FA); test graphical data may include at least one of voltage waveform data, current data, and first delay data; test behavior data may include at least one of strain data, impedance data, acceleration data, second delay data, and operating mode; the component may be an electronic component, but this application is not limited thereto.
[0014] Please refer to the following: Figures 1 to 3 , Figure 3 It shows that according to Figure 2The flowchart of the first procedure P1 of the artificial intelligence failure analysis method 200 is provided. The first procedure P1 includes steps S021, S022, S023, and S024. Step S021 determines whether the test data is test graphic data or test behavior data. Step S022 obtains multiple test modes TM from the storage module 110 based on whether the test data is test graphic data or test behavior data. Step S023 executes the multiple test modes TM on the test data respectively. Step S024 generates multiple analysis result data. The aforementioned multiple analysis result data correspond to multiple test modes TM. The analysis result data may include at least one of defect type data, size data, location data, junction temperature data, circuit data, and software application data, but this application is not limited thereto.
[0015] Please refer to the following: Figures 1 to 4 , Figure 4 It shows that according to Figure 2 The flowchart of the second procedure P2 of the artificial intelligence failure analysis method 200 is shown below. The second procedure P2 includes steps S031, S032, S033, and S034. Step S031 analyzes the historical test dataset 111 to calculate multiple mean failure interval times (MFRs) for the historical test dataset 111. These multiple MFRs correspond to multiple electronic components. Step S032 generates a worst-case list based on the MFRs. The worst-case list may include parameters with a high probability of electronic component failure. Step S033 tests the test data using test modes TM to generate multiple test results. Step S034 determines the best test mode TM corresponding to the best result among the test results as the optimal mode.
[0016] The second procedure P2 may further include testing the test data using test mode TM to generate multiple test results, and determining the test mode TM corresponding to the best result among the test results as the optimal mode. In other words, steps S033 and S034 involve selecting one test mode TM as the current optimal test mode TM, and testing the current test data using other test modes TM. When the test results obtained by testing the test data using other test modes TM are better than the test results of the optimal test mode TM, the aforementioned other test modes TM are updated to new optimal test mode TM, until the optimal test mode TM corresponding to the best test result is found. The test results may be mean time between failures (MTBF) or failure rate, but this application is not limited to these.
[0017] Please refer to the following: Figures 1 to 5 , Figure 5 It shows that according to Figure 2The flowchart illustrates the third procedure P3 of the artificial intelligence failure analysis method 200. The third procedure P3 includes steps S041, S042, S043, and S044. Step S041 analyzes the historical test dataset 111. Step S042 determines the application domain corresponding to the historical test dataset 111. Step S043 executes the test mode TM for the corresponding application domain. Step S044 generates analysis result data. In other words, when acquiring data from the historical test dataset 111, the third procedure P3 determines the test mode TM based on the application domain of each data element and the current process. When the application domain of an element differs, its input / output parameters, the features to be extracted, and the analysis results desired by the failure analysis model M1 will differ. For example, the application domain may include at least one of the semiconductor manufacturing field, product field, component field, and packaging field, but this application is not limited thereto.
[0018] When one of the data applications is in the semiconductor manufacturing field or at the product stage, the input test graphical data may include voltage data, current data, and delay data, while the test behavior data may include voltage data, current data, strain data, impedance data, acceleration data, delay data, and operating mode data. The analysis results generated by failure analysis model M1 may be yield.
[0019] Thus, the artificial intelligence failure analysis method 200 of this application can calculate the yield rate after obtaining the analysis result data, and can be dynamically adjusted and updated, so that users can know the specific process test most relevant to the test result, and obtain the test result without having to perform all the tests, thereby reducing the test time and cost.
[0020] When one of the data applications is in the component or packaging field, the input test graphical data may include voltage data, current data, and delay data, while the test behavior data may include voltage data, current data, strain data, impedance data, acceleration data, delay data, and operating mode data. The analysis result generated by failure analysis model M1 may be the mean time between failures (MTBF).
[0021] Thus, the artificial intelligence failure analysis method 200 of this application can be dynamically adjusted and updated when applied to the field of components or packaging, allowing users to adjust the test time, test patterns or test behavior according to the mean time between failures, and determine whether to extend the test time and test cost based on the characteristics of a specific component and the reliability value of the test.
[0022] For example, certain electronic components need to be tested in a high-temperature and high-humidity environment for a long time to detect whether there is a problem of low-density impurity contamination in the production line. The artificial intelligence failure analysis method 200 of this application can appropriately adjust the testing time of electronic components through the analysis results generated by the failure analysis model M1, so as to save testing time and testing costs.
[0023] Please refer to the following: Figure 2 Step S05 involves training the historical test dataset 111 using artificial intelligence methods to generate a failure analysis model M1 based on the analysis results data, application area, worst-case list, and best-case scenario. Step S06 involves testing the current test data using the failure analysis model M1 to generate corresponding test result data for failure analysis of the component.
[0024] Specifically, after analyzing the analysis results data, application areas, worst-case scenario list, and best-case scenario through steps S02, S03, and S04, the historical test dataset 111 can be used to train and generate a failure analysis model M1. The failure analysis model M1 can be used to analyze and calculate the failure analysis parameters for electronic components.
[0025] Please see Figure 1 , Figure 2 and Figure 6 , Figure 6 A flowchart of the artificial intelligence failure analysis method 200a according to a third embodiment of this application is shown. The artificial intelligence failure analysis method 200a includes steps S11, S12, S13, S14, S15, S16, and S17. In the third embodiment, steps S11, S12, S13, S14, S16, and S17 are implemented in the same way as steps S01, S02, S03, S04, S05, and S06 of the first embodiment, and will not be described again. Specifically, the artificial intelligence failure analysis method 200a may further include step S15. Step S15 includes executing a fourth procedure. The fourth procedure includes stopping the test when the test data is tested through one of the test modes TM and an anomaly information is generated.
[0026] In other words, during failure testing, if the machine or hardware malfunctions, or if a situation not recorded in the historical test dataset 111 occurs, step S15 can send an anomaly message and alert the testers. Simultaneously, the test is stopped, and the historical test dataset 111 and the failure analysis model M1 are updated using the anomaly message. Thus, the artificial intelligence failure analysis method 200a of this application can automatically update the failure analysis model M1 in real time in response to abnormal situations during the testing process.
[0027] Although the above embodiments have been disclosed in this application, they are not intended to limit this application. Those skilled in the art can make various modifications and refinements without departing from the spirit and scope of this application. Therefore, the scope of protection of this application shall be determined by the scope of protection of the claims.
Claims
1. An artificial intelligence failure analysis system, characterized in that, include: Storage module, comprising: Historical test dataset, which includes multiple test datasets; Multiple test modes; and A processor, signal-connected to the storage module, is configured to perform operations including the following steps: Receive the historical test dataset and the multiple test modes; Execute a first procedure, wherein the first procedure includes obtaining at least one of the plurality of test data, determining that the at least one of the plurality of test data is test graphic data or test behavior data, and testing the at least one of the plurality of test data through the plurality of test modes to generate a plurality of analysis result data; Execute a second procedure, wherein the second procedure includes analyzing the historical test dataset to calculate the average failure interval time of the historical test dataset, generating a worst-case list based on the average failure interval time, and determining the best mode among the plurality of test modes; Execute a third procedure, wherein the third procedure includes determining the application domain of the historical test dataset and analyzing the historical test dataset according to the application domain, thereby generating the multiple analysis result data; Based on the multiple analysis results, the application domain, the worst-case scenario list, and the best-case scenario, an artificial intelligence method is used to train the historical test dataset to generate a failure analysis model; and The failure analysis model is used to test the current test data to generate test result data corresponding to the current test data, which is used for failure analysis of the component.
2. The artificial intelligence failure analysis system according to claim 1, characterized in that, The processor is further configured to perform the following operations: Execute a fourth procedure, wherein the fourth procedure includes stopping the test when the plurality of test data are tested through one of the plurality of test modes and an anomaly is generated.
3. The artificial intelligence failure analysis system according to claim 1, characterized in that, The second procedure further includes: The multiple test data are tested using the multiple test modes to generate multiple test results, and one of the multiple test modes corresponding to the best result among the multiple test results is determined as the best mode.
4. The artificial intelligence failure analysis system according to claim 1, characterized in that, The test graphical data includes at least one of voltage waveform data, current data, and first delay data; the test behavior data includes at least one of strain data, impedance data, acceleration data, second delay data, and operating mode; the multiple analysis result data includes at least one of defect type data, size data, location data, junction temperature data, circuit data, and software application data.
5. The artificial intelligence failure analysis system according to claim 1, characterized in that, The application areas mentioned therein include at least one of the following: semiconductor manufacturing, product manufacturing, component manufacturing, and packaging.
6. An artificial intelligence failure analysis method, characterized in that, include: The driver processor self-storage module receives historical test datasets and multiple test modes, wherein the historical test dataset includes multiple test data; The processor is driven to execute a first program, wherein the first program includes obtaining at least one of the plurality of test data, determining that the at least one of the plurality of test data is test graphic data or test behavior data, and testing the at least one of the plurality of test data through the plurality of test modes to generate a plurality of analysis result data; The processor is driven to execute a second program, wherein the second program includes analyzing the historical test dataset to calculate the average failure interval time of the historical test dataset, generating a worst-case list based on the average failure interval time, and determining the best mode among the plurality of test modes; The processor is driven to execute a third program, wherein the third program includes determining the application domain corresponding to the historical test dataset, and analyzing the historical test dataset according to the application domain to generate the multiple analysis result data; The processor is driven to train the historical test dataset using artificial intelligence methods based on the multiple analysis result data, the application domain, the worst-case list, and the best-case pattern to generate a failure analysis model; as well as The processor is driven to perform tests on the current test data through the failure analysis model to generate test result data corresponding to the current test data, which is used to perform failure analysis on the component.
7. The artificial intelligence failure analysis method according to claim 6, characterized in that, Further includes: The processor is driven to execute a fourth program, wherein the fourth program includes stopping the test when the plurality of test data are tested through one of the plurality of test modes and an exception is generated.
8. The artificial intelligence failure analysis method according to claim 6, characterized in that, The second procedure further includes: The processor is driven to test the multiple test data through the multiple test modes to generate multiple test results, and one of the multiple test modes corresponding to the best result of the multiple test results is determined as the best mode.
9. The artificial intelligence failure analysis method according to claim 6, characterized in that, The test graphical data includes at least one of voltage waveform data, current data, and first delay data; the test behavior data includes at least one of strain data, impedance data, acceleration data, second delay data, and operating mode; the multiple analysis result data includes at least one of defect type data, size data, location data, junction temperature data, circuit data, and software application data.
10. The artificial intelligence failure analysis method according to claim 6, characterized in that, The application areas mentioned therein include at least one of the following: semiconductor manufacturing, product manufacturing, component manufacturing, and packaging.