A test system, cloud server and test machine of a power semiconductor device

Abstract

A testing system for power semiconductor devices, a cloud server, and a testing machine are disclosed. The system includes: a cloud server (100) and at least one testing machine (200); the cloud server (100) is used to pre-train an artificial intelligence model using historical data from multiple power semiconductor devices; the testing machine (200) is used to obtain data of the power semiconductor device to be tested; the cloud server (100) or the testing machine (200) is used to input the data of the power semiconductor device to be tested into the artificial intelligence model, and the output of the artificial intelligence model is the diagnostic result of the power semiconductor device to be tested. The powerful computing and storage capabilities of the cloud server (100) allow for the pre-training of an accurate artificial intelligence model. The cloud server (100) or the testing machine (200) then uses the trained artificial intelligence model to diagnose the power semiconductor device to be tested, obtaining accurate diagnostic results and eliminating abnormal power semiconductor devices to prevent malfunctions that could lead to circuit failures during actual product use.

Description

Technical Field

[0001] This application relates to the field of power semiconductor technology, and in particular to a test system, cloud server and test machine for power semiconductor devices. Background Technology

[0002] Power semiconductor devices generally refer to semiconductor devices that perform circuit switching functions, such as those used in power conversion circuits in electric vehicles, photovoltaic systems, or data centers. For example, in electric vehicles, power semiconductor devices can be used in the powertrain control circuit. In practical applications, to ensure the safety of the circuits in which power semiconductor devices are used, performance testing is required after production to reduce the risk of device failure.

[0003] Currently, the quality testing of power semiconductor devices mainly involves setting upper and lower limits for the performance parameters of power semiconductor devices using a testing machine, and eliminating power semiconductor devices whose performance parameters exceed the upper and lower limits.

[0004] However, the current method of testing using upper and lower limit values ​​requires confirmation through a large number of historical products, and the screening method based on upper and lower limit values ​​is highly dependent on experience. It cannot establish a complete correlation with the failure principle of the device and can only identify a portion of the problematic power semiconductor devices. Summary of the Invention

[0005] This application provides a testing system, cloud server, and testing machine for power semiconductor devices, which can more comprehensively screen out defective power semiconductor devices.

[0006] The power semiconductor device testing system provided in this application includes a cloud server and at least one testing machine. The cloud server is used to pre-train an artificial intelligence (AI) model using historical data from multiple power semiconductor devices. The testing machine is used to obtain data of the power semiconductor device to be tested. Both the cloud server and the testing machine can use the trained AI model to diagnose the power semiconductor device to be tested. That is, the cloud server or the testing machine inputs the data of the power semiconductor device to be tested into the AI ​​model, and the output of the AI ​​model is the diagnostic result of the power semiconductor device to be tested.

[0007] The solution provided in this application utilizes the powerful computing and storage capabilities of cloud servers to train a relatively accurate artificial intelligence model. Then, the pre-trained artificial intelligence model can be used to more accurately diagnose high-risk power semiconductor devices among the power semiconductor devices under test, thereby overcoming the limitations of traditional testing systems that only use upper limit data to screen high-risk power semiconductor devices. In other words, it can more accurately and comprehensively diagnose whether there is a risk in power semiconductor devices.

[0008] In one possible implementation, in this embodiment of the application, the cloud server is used to input the data of the power semiconductor device under test into an artificial intelligence model. The output of the artificial intelligence model is the diagnostic result of the power semiconductor device under test, and the diagnostic result is sent to the testing machine. It should be understood that in this embodiment of the application, the cloud server inputs data into the artificial intelligence model, obtains the diagnostic result, and sends the diagnostic result to the testing machine. Therefore, the testing machine can directly obtain the test result, saving the computing resources of the testing machine.

[0009] In one possible implementation, in this embodiment, the cloud server sends a pre-trained artificial intelligence model to the testing machine. The testing machine inputs data from the power semiconductor device under test into the artificial intelligence model, and the output of the artificial intelligence model is the diagnostic result of the power semiconductor device under test. It should be understood that the cloud server sends the trained artificial intelligence model to the testing machine, which then inputs the data and obtains the diagnostic result. Therefore, the testing machine completes the testing of the power semiconductor device under test independently, without needing to obtain diagnostic results from the cloud server. This reduces the network dependency between the cloud server and the testing machine and improves the real-time performance of the power semiconductor device testing system in this embodiment.

[0010] In one possible implementation, the test machine in this application embodiment is further used to collect historical data of multiple power semiconductor devices and send the collected historical data of multiple power semiconductor devices to a cloud server; the historical data includes at least one of the following: chip test data of multiple power semiconductor devices, functional test data of packaged modules, or relevant data when there are abnormalities in actual applications; the cloud server is further used to use the historical data of multiple power semiconductor devices to train an artificial intelligence model using at least one of supervised learning model or unsupervised learning model.

[0011] In one possible implementation, the cloud server uses an unsupervised learning model to train an artificial intelligence model, specifically used to obtain the failure types of multiple power semiconductor devices. It extracts all test items corresponding to each failure type from the test items of the power semiconductor devices using expert knowledge. All test items corresponding to each failure type form a subset of test items. For each subset of test items, the unsupervised learning model distinguishes between anomaly and normal subsets, assigning a first score and a second score to the anomaly subset and normal subset, respectively. The sum of the first and second scores of all test item subsets for each power semiconductor device is obtained as the total anomaly level score. When the total anomaly level score is greater than or equal to a preset score threshold, the power semiconductor device is determined to be an abnormal device. It should be understood that the embodiments of this application use an unsupervised learning model to train the AI ​​model, which can fully utilize expert knowledge to jointly detect anomalies in subsets of test items containing multiple test items. It can utilize the relationships between devices within the same subset of test items, or the physical principles corresponding to the same failure type, to detect the risk of power semiconductor device failure from multiple dimensions, thereby more accurately assessing the risk of device failure.

[0012] In one possible implementation, the cloud server in this embodiment is specifically used to assign different first scores to abnormal devices in different subsets of test items using expert knowledge. It should be understood that different subsets of test items may have different weights for judging abnormal devices; assigning different first scores to abnormal devices in different subsets of test items can yield more accurate diagnostic results.

[0013] In one possible implementation, in this embodiment of the application, the cloud server uses a supervised learning model to train an artificial intelligence model, specifically used to obtain relevant data when a power semiconductor device is abnormal as a data label, use the supervised learning model to extract data features of the data labels, use the data features to obtain the total score of the abnormality level of each power semiconductor device, and when the total score of the abnormality level is greater than a preset score threshold, the power semiconductor device is determined to be an abnormal device.

[0014] In one possible implementation, the testing machine in this embodiment is further used to compare the data of the power semiconductor device under test with preset upper and lower limit values. When the data of the power semiconductor device under test exceeds the upper and lower limit values, the power semiconductor device under test is determined to be an abnormal device. It should be understood that the power semiconductor device testing system provided in this embodiment can combine traditional testing methods with artificial intelligence models. Based on traditional testing methods, the artificial intelligence model in this embodiment is used for secondary detection, improving the accuracy of detection and achieving complementarity between traditional methods and AI model testing methods. This allows for a more comprehensive diagnosis of high-risk power semiconductor devices.

[0015] This application embodiment does not limit the number of test machines included in the testing system; it may include one test machine, or two or more test machines. When the testing system includes multiple test machines, the cloud server can first train a global AI model, and then, based on the differences between the various test machines, obtain a local AI model applicable to each test machine based on the global AI model. That is, in one possible implementation, the testing system provided in this application embodiment includes at least the following two test machines: a first test machine and a second test machine; the cloud server is specifically used to train a global artificial intelligence model based on first historical data sent by the first test machine and second historical data sent by the second test machine, adjust the global artificial intelligence model using the first historical data to obtain a first artificial intelligence model, adjust the global artificial intelligence model using the second historical data to obtain a second artificial intelligence model; use the first artificial intelligence model to test the power semiconductor device under test corresponding to the first test machine to obtain a first diagnostic result, use the second artificial intelligence model to test the power semiconductor device under test corresponding to the second test machine to obtain a second diagnostic result, send the first diagnostic result to the first test machine, and send the second diagnostic result to the second test machine. This application embodiment uses first and second historical data to train a global artificial intelligence model, which can fully utilize all historical data and improve the testing accuracy of the global artificial intelligence model. Furthermore, considering the potential performance differences between different testing machines, this application embodiment also uses the first historical data to adjust the global artificial intelligence model to obtain a first artificial intelligence model, and uses the second historical data to adjust the global artificial intelligence model to obtain a second artificial intelligence model. This results in a local artificial intelligence model (first artificial intelligence model) tailored to the characteristics of the first testing machine and a local artificial intelligence model (second artificial intelligence model) tailored to the characteristics of the second testing machine, further improving the testing accuracy of the artificial intelligence model used in actual testing.

[0016] In this embodiment, each test machine corresponds to a global AI model, and the diagnostic results are obtained by a cloud server. Specifically, in one possible implementation, the testing system provided in this application embodiment includes at least two test machines: a first test machine and a second test machine; and a cloud server, specifically used to train a global artificial intelligence model based on first historical data sent by the first test machine and second historical data sent by the second test machine. The global artificial intelligence model is then used to test the power semiconductor device under test corresponding to the first test machine to obtain a first diagnostic result, and the same model is used to test the power semiconductor device under test corresponding to the second test machine to obtain a second diagnostic result. The first diagnostic result is sent to the first test machine, and the second diagnostic result is sent to the second test machine. The cloud server provided in this application embodiment utilizes first and second historical data to train the global artificial intelligence model, which can fully utilize all historical data, thereby improving the generalization ability of the global artificial intelligence model. Using a more accurate global artificial intelligence model to diagnose the power semiconductor device under test can improve the universality of the artificial intelligence model and increase the accuracy of diagnosing new devices.

[0017] In this embodiment, the cloud server obtains the local AI models corresponding to each test machine and sends them to each test machine, which then performs the diagnosis. Specifically, in one possible implementation, the power semiconductor device testing system in this embodiment includes at least two test machines: a first test machine and a second test machine. The cloud server is specifically used to train a global artificial intelligence model based on first historical data sent by the first test machine and second historical data sent by the second test machine. It then adjusts the global artificial intelligence model using the first historical data to obtain a first artificial intelligence model, which is sent to the first test machine. Finally, it adjusts the global artificial intelligence model using the second historical data to obtain a second artificial intelligence model, which is sent to the second test machine. The first test machine is specifically used to diagnose the corresponding power semiconductor device under test using the first artificial intelligence model. The second test machine is specifically used to diagnose the corresponding power semiconductor device under test using the second artificial intelligence model. The test machines in this embodiment can directly utilize the artificial intelligence model to complete the task of detecting the power semiconductor device under test independently, without needing to communicate with the cloud server. Therefore, it avoids the situation where the network between the board-side test machine and the cloud server is interrupted, preventing the power semiconductor device testing system in this embodiment from failing to obtain the diagnostic results for the power semiconductor device under test. Thus, the power semiconductor device testing system provided in this application embodiment can reduce the network dependence between the cloud server and the testing machine, and improve the real-time performance of the power semiconductor device testing system in this application embodiment in obtaining diagnostic results.

[0018] In this embodiment, the cloud server sends the global AI model to each test machine, which then uses the global AI model to complete the diagnosis. Specifically, in one possible implementation, the power semiconductor device testing system in this embodiment includes at least two test machines: a first test machine and a second test machine; a cloud server, specifically used to train a global artificial intelligence model based on first historical data sent by the first test machine and second historical data sent by the second test machine, and then sends the global artificial intelligence model to the first and second test machines; the first test machine is used to diagnose the corresponding power semiconductor device under test using the global artificial intelligence model; and the second test machine is used to diagnose the corresponding power semiconductor device under test using the global artificial intelligence model. In this embodiment, the test machines can complete the testing of the power semiconductor device under test independently based on the trained global artificial intelligence model. When the test machine obtains the diagnostic result, the cloud server does not need to send the diagnostic result back to the test machine; therefore, the test machine's acquisition of the diagnostic result is not affected by network failures, and its dependence on the network is low.

[0019] In one possible implementation, the test machine in this embodiment is further used to send updated data to the cloud server; the cloud server is further used to update the artificial intelligence model based on the updated data. Additionally, the test machine is also used to fine-tune the pre-trained AI model using data from its own side, and to use the fine-tuned model to diagnose the power semiconductor device under test, outputting diagnostic results. It should be understood that in this embodiment, the test machine can fine-tune the obtained AI model based on its own data or the characteristics of the data, thereby further improving the accuracy of the AI ​​model in this embodiment. The test machine can send the fine-tuned AI model to the cloud server for unified management.

[0020] Based on the power semiconductor device testing system provided in the above embodiments, this application also provides a cloud server. The advantages of the various embodiments of the above testing system are also applicable to the following server, and will not be repeated here. The server includes: a first transceiver device and a first controller; the first transceiver device is used to receive test data of the power semiconductor device under test sent by the test machine; the first controller is used to obtain an artificial intelligence model by pre-training using historical data of multiple power semiconductor devices; it is also used to input the data of the power semiconductor device under test into the artificial intelligence model, and the output of the artificial intelligence model is the diagnostic result of the power semiconductor device under test; the first transceiver device is also used to send the diagnostic result to the test machine; or, the first controller is used to send the artificial intelligence model to the test machine so that the test machine can use the artificial intelligence model to diagnose the power semiconductor device under test.

[0021] In one possible approach, the first controller in this embodiment is specifically used to obtain the failure types of multiple power semiconductor devices when training an artificial intelligence model using an unsupervised learning model. It then uses expert knowledge to extract all test items corresponding to each failure type from the test items of the power semiconductor devices. All test items corresponding to each failure type form a subset of test items. Anomaly level detection is performed on each subset of test items using an unsupervised learning model, resulting in a first score and a second score for abnormal devices and normal devices, respectively. The sum of the first score and the second score of all test item subsets for each power semiconductor device is obtained as the total anomaly level score. When the total anomaly level score is greater than a preset score threshold, the power semiconductor device is determined to be an abnormal device.

[0022] In one possible embodiment, the cloud server in this application corresponds to at least two test machines: a first test machine and a second test machine; a first controller, specifically configured to train a global artificial intelligence model based on first historical data sent by the first test machine and second historical data sent by the second test machine, adjust the global artificial intelligence model using the first historical data to obtain a first artificial intelligence model, adjust the global artificial intelligence model using the second historical data to obtain a second artificial intelligence model; test the power semiconductor device under test corresponding to the first test machine using the first artificial intelligence model to obtain a first diagnostic result, and test the power semiconductor device under test corresponding to the second test machine using the second artificial intelligence model to obtain a second diagnostic result; and a first transceiver device, specifically configured to send the first diagnostic result to the first test machine and also to send the second diagnostic result to the second test machine.

[0023] In one possible embodiment, the cloud server in this application corresponds to at least two test machines: a first test machine and a second test machine; a first controller, specifically used to train a global artificial intelligence model based on first historical data sent by the first test machine and second historical data sent by the second test machine, use the global artificial intelligence model to test the power semiconductor device under test corresponding to the first test machine to obtain a first diagnostic result, and use the global artificial intelligence model to test the power semiconductor device under test corresponding to the second test machine to obtain a second diagnostic result; and a first transceiver device, specifically used to send the first diagnostic result to the first test machine and also used to send the second diagnostic result to the second test machine.

[0024] In one possible embodiment, the first transceiver device in this application is further configured to receive updated data sent by the test machine; the first controller is further configured to update the artificial intelligence model based on the updated data.

[0025] Based on the power semiconductor device testing system and server provided in the above embodiments, this application also provides a testing machine. The advantages of the various embodiments of the above testing system are also applicable to the following testing machine, and will not be repeated here. The testing machine includes: a second transceiver device and a second controller; the second transceiver device is used to receive an artificial intelligence model sent by a cloud server, the artificial intelligence model being pre-trained by the cloud server using historical data from multiple power semiconductor devices; the second controller is used to obtain data of the power semiconductor device to be tested; the data of the power semiconductor device to be tested is input into the artificial intelligence model, and the output of the artificial intelligence model is the diagnostic result of the power semiconductor device to be tested.

[0026] In one possible approach, the second controller in this application embodiment is further configured to collect historical data of multiple power semiconductor devices and send the collected historical data of multiple power semiconductor devices to a cloud server; the historical data includes at least one of chip test data or packaged module functional test data of multiple power semiconductor devices.

[0027] In one possible embodiment, the second controller is further configured to compare the data of the power semiconductor device under test with preset upper and lower limit values. When the data of the power semiconductor device under test exceeds the upper and lower limit values, the power semiconductor device under test is determined to be an abnormal device.

[0028] As can be seen from the above technical solutions, the embodiments of this application have the following advantages:

[0029] The testing system provided in this application includes a cloud server and a testing machine. The cloud server has higher computing and storage capabilities than the testing machine, thus leveraging its powerful computing and storage capabilities to pre-train an artificial intelligence model. Since the cloud server pre-trains the AI ​​model using a large amount of historical data from power semiconductor devices, it can accurately obtain the model's parameters. Therefore, the cloud server or testing machine can accurately diagnose the power semiconductor devices under test using the trained AI model, thereby obtaining accurate diagnostic results and eliminating abnormal power semiconductor devices to prevent malfunctions that could lead to circuit failures during actual product use. This testing system overcomes the shortcomings of traditional methods, which only use upper and lower limit values ​​to screen high-risk power semiconductor devices, sometimes failing to detect any. The testing system provided in this application can further screen out high-risk power semiconductor devices, providing a more comprehensive diagnosis of whether power semiconductor devices pose a risk. This solution compensates for the inaccuracy of testing machines simply using upper and lower limit values ​​to screen power semiconductor devices for defects. Attached Figure Description

[0030] Figure 1An architecture diagram of a test system for a power semiconductor device provided in an embodiment of this application;

[0031] Figure 2 A schematic diagram illustrating the data source of a power semiconductor device provided for implementation of this application;

[0032] Figure 3A A flowchart illustrating a testing method for a power semiconductor device provided in an embodiment of this application;

[0033] Figure 3B A flowchart illustrating another power semiconductor device testing method provided in this application embodiment;

[0034] Figure 4 A schematic diagram of a test system for another power semiconductor device provided in an embodiment of this application;

[0035] Figure 5 A schematic diagram of a test system for another power semiconductor device provided in an embodiment of this application;

[0036] Figure 6 A schematic diagram of a test system for another power semiconductor device provided in an embodiment of this application;

[0037] Figure 7 A schematic diagram of a test system for another power semiconductor device provided in an embodiment of this application;

[0038] Figure 8 A schematic diagram of a test system for a power semiconductor device including multiple testers, provided for an embodiment of this application;

[0039] Figure 9 A schematic diagram of another test system for a power semiconductor device including multiple testers, provided for an embodiment of this application;

[0040] Figure 10 A schematic diagram of another test system for a power semiconductor device including multiple testers, provided as an embodiment of this application;

[0041] Figure 11 A schematic diagram of another test system for a power semiconductor device including multiple testers, provided as an embodiment of this application;

[0042] Figure 12 This is a schematic diagram of the structure of a cloud server provided in an embodiment of this application;

[0043] Figure 13 This is a schematic diagram of the structure of a testing machine provided in an embodiment of this application. Detailed Implementation

[0044] The terms "first," "second," etc., used in the following description are for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature specified with "first," "second," etc., may explicitly or implicitly include one or more of that feature. In the description of this application, unless otherwise stated, "multiple" means two or more.

[0045] Furthermore, in this application, directional terms such as "upper" and "lower" may be defined relative to the orientation in which the components are schematically placed in the accompanying drawings. It should be understood that these directional terms can be relative concepts, used for relative description and clarification, and can change accordingly depending on the orientation in which the components are placed in the accompanying drawings.

[0046] In this application, unless otherwise expressly specified and limited, the term "connection" should be interpreted broadly. For example, "connection" can be a fixed connection, a detachable connection, or an integral part; it can be a direct connection or an indirect connection through an intermediate medium. Furthermore, the term "coupled" can refer to a method of electrical connection for signal transmission. "Coupled" can be a direct electrical connection or an indirect electrical connection through an intermediate medium.

[0047] This application relates to a testing system for power semiconductor devices, comprising a cloud server and at least one testing machine. The cloud server possesses powerful computing capabilities and storage space, enabling it to train an artificial intelligence model using extensive historical data from power semiconductor devices. Since the AI ​​module is trained on a large amount of historical data, the cloud server or testing machine can accurately diagnose the power semiconductor devices under test using the trained AI model, eliminating faulty devices to prevent them from malfunctioning and causing circuit failures during real-world use. This solution overcomes the inaccuracies of traditional methods that rely solely on pre-set upper and lower limits on the testing machine for defective screening of power semiconductor devices.

[0048] To enable those skilled in the art to better understand the technical solutions provided in the embodiments of this application, the testing system provided in the embodiments of this application is described in detail below with reference to the accompanying drawings. The testing system includes a cloud server and at least one testing machine.

[0049] This application does not specifically limit the number of test machines in its embodiments. The number can be set according to the actual number of product lines. One test machine can correspond to one product line, and one product line can correspond to multiple test machines. The types of power semiconductor devices corresponding to multiple product lines can be the same or different. The following is in conjunction with the appendix... Figure 1The example given is three test machines. It should be understood that there could also be one test machine, two test machines, or more test machines.

[0050] System Implementation Examples

[0051] See Figure 1 This figure is a test system architecture diagram of a power semiconductor device provided in an embodiment of this application.

[0052] like Figure 1 As shown in the embodiment of this application, the power semiconductor device testing system includes a cloud server 100, a first tester 201, a second tester 202, and a third tester 203. The cloud server 100 is located on a cloud platform, which may include multiple cloud servers; the figure only illustrates one cloud server as an example. The first tester 201, second tester 202, and third tester 203 on the board can communicate with the cloud server 100 in the cloud to exchange or transmit data.

[0053] Among them, cloud server 100 is used to pre-train AI models using historical data from multiple power semiconductor devices.

[0054] It should be understood that the embodiments of this application do not specifically limit the specific model used by the cloud server 100 for AI model training. For example, it can use a supervised learning model or an unsupervised learning model. The historical data used by the cloud server 100 to train the AI ​​model can be data uploaded by the first test machine 201, the second test machine 202, and the third test machine 203, or data uploaded by more test machines. No specific limitation is made in the embodiments of this application. It should be understood that the more historical data the cloud server 100 uses when training the AI ​​model, the more accurate the parameters of the trained AI model will be.

[0055] The first tester 201, the second tester 202, and the third tester 203 are used to obtain data of their respective power semiconductor devices under test.

[0056] In specific testing, the test can be conducted by the cloud server 100, or the cloud server 100 can send the trained AI model to the test machine for testing.

[0057] The first method involves testing cloud server 100.

[0058] Cloud server 100 is used to input data from the power semiconductor device under test into a pre-trained artificial intelligence model. The output of the artificial intelligence model is the diagnostic result of the power semiconductor device under test. The diagnostic result is then sent to the testing machine.

[0059] It should be understood that the term "power semiconductor device under test" here refers to any power semiconductor device that needs to be tested; and the term "testing machine" here refers to any testing machine.

[0060] The second method: testing with a testing machine.

[0061] The first tester 201, the second tester 202, and the third tester 203 are respectively used to input the data of their respective power semiconductor devices under test into pre-trained artificial intelligence models, and respectively obtain the output of the pre-trained models as the diagnostic results of their respective power semiconductor devices under test.

[0062] This application does not limit whether the first tester 201, the second tester 202, and the third tester 203 simultaneously test their respective corresponding power semiconductor devices under test, or test them sequentially, because each tester can complete its test action independently without interfering with each other.

[0063] It is understandable that this application embodiment utilizes an artificial intelligence model to test the performance of power semiconductor devices in order to improve the accuracy of intercepting defective power semiconductor devices. However, due to the limited computing and storage capabilities of the on-board testing machine, training the artificial intelligence model using the on-board testing machine would be time-consuming and consume excessive computing resources, affecting other functions of the testing machine. Therefore, the power semiconductor device testing system provided in this application embodiment utilizes the powerful computing and storage capabilities of a cloud server to train an artificial intelligence model using historical data from multiple power semiconductor devices. This artificial intelligence model is then used to test the power semiconductor device under test. The AI ​​model trained on the cloud server can be distributed to the testing machine, which uses the pre-trained AI model to test the power semiconductor device under test and ultimately outputs diagnostic results.

[0064] Therefore, the power semiconductor device testing system provided in this application can utilize an artificial intelligence model to diagnose the performance of power semiconductors and screen out defective power semiconductor devices. Furthermore, considering the limited computing and storage capabilities of the on-board testing machine, this application utilizes a cloud server with powerful computing and storage capabilities to train the artificial intelligence model, thereby obtaining an artificial intelligence model with high diagnostic accuracy.

[0065] The above content mainly introduces the architecture of the power semiconductor device testing system provided in the embodiments of this application. The following describes the specific process of AI model training on a cloud server with reference to the accompanying drawings.

[0066] See Figure 2 The figure is a schematic diagram of the data source of a power semiconductor device provided in this application.

[0067] In this embodiment, the cloud server utilizes a large amount of historical data to train and obtain the AI ​​model. Historical data may include production line data and application-side data.

[0068] The production line data D1 can originate from the following sources: chip (CP, ChipProbe) testing, functional testing (FT, Functional Test), single-board testing, and complete system testing. CP testing can be for individual chips, FT testing can be for packaged modules with specific circuit functions, single-board testing can be for board-level testing, and complete system testing can be for system-level testing. It can be understood that chip test data and power test data can be used separately for AI model training, or they can be used together for AI model training.

[0069] It should be noted that CP testing primarily targets individual chips. A single chip can be a standalone Insulated Gate Bipolar Transistor (IGBT), a standalone Metal-Oxide-Semiconductor Field-Effect Transistor (MOSFET), a diode, a bipolar junction transistor (BJT), a thyristor, an integrated gate-commutated thyristor (IGCT), or other power semiconductor devices. It can also be a chip formed by multiple IGBTs, or a chip formed by IGBTs and diodes. FT testing primarily targets a packaged module.

[0070] Application-side data D2 mainly includes relevant data corresponding to failures that occur during actual use of power semiconductor devices after production. In other words, it includes actual operating condition data of power semiconductor devices reported by the customer. This data can include test data sources or verification data sources.

[0071] As an example, the test data of the power semiconductor device under test in this application embodiment may include, but are not limited to, the following breakdown voltage, leakage current, on-state voltage drop, parasitic capacitance, gate charge, turn-on loss, and turn-off loss.

[0072] The following describes the principle of the power semiconductor device test system provided in this application, which uses test data to obtain diagnostic results.

[0073] This application provides AI models trained on cloud servers, which can be trained using unsupervised ensemble learning models, supervised learning models, or a combination of supervised and unsupervised learning models. No specific limitations are made in this application.

[0074] The following describes how the AI ​​model provided in this application is obtained by training an unsupervised ensemble learning model.

[0075] As an example, the unsupervised learning model in this application embodiment can be an unsupervised ensemble learning model. It is understood that the unsupervised model in this application embodiment combines unsupervised learning algorithms with expert knowledge to construct an unsupervised ensemble learning model, which can fully utilize expert experience and reduce the impact of too few failed samples on the accuracy of the artificial intelligence model.

[0076] In this embodiment of the application, when the cloud server uses an unsupervised learning model to train an artificial intelligence model, as one possible implementation method, the cloud server is specifically used to obtain the failure types of multiple power semiconductor devices. It uses expert knowledge to extract all test items corresponding to each failure type from the test items of the power semiconductor devices. All test items corresponding to each failure type form a test item subset. For each test item subset, an unsupervised learning model is used to distinguish between abnormal subsets and normal subsets. The abnormal subset and the normal subset are respectively recorded as a first score and a second score. The sum of the first score and the second score of all test item subsets of each power semiconductor device is obtained as the total abnormality level score. When the total abnormality level score is greater than or equal to a preset score threshold, the power semiconductor device is judged to be an abnormal device.

[0077] Specifically, the failure types can be various types of failures that cannot be intercepted during individual testing of power semiconductor devices. For example, an early failure may occur in a power semiconductor device during actual use, but it may not have been detected during controllable testing.

[0078] Furthermore, a single failure type may correspond to problems in many test items. For example, gate leakage might indicate a problem at the physical layer, affecting multiple test items. Therefore, all test items associated with a particular failure type are extracted to form a subset of test items corresponding to that failure type.

[0079] As an example, the following describes in detail, with reference to Tables 1 and 2, the scheme for obtaining an artificial intelligence model by training an unsupervised ensemble learning model provided in the embodiments of this application.

[0080] Table 1

[0081]

[0082] Table 2

[0083] Subset1 Subset2 Subset3 …… Total score of abnormal level Device A 0 0 0 0 Device B 0 1 0 1 Device C 0 0 0 0 …… ……

[0084] Among them, Gate failure, PN junction failure, or withstand ring failure are different failure types; IGES (GE leakage current), VTH (threshold voltage), VCESAT (conduction voltage drop between collector and emitter), ICES (CE leakage current), and Eon (turn-on loss) are different test items. As can be seen from Table 1, different failure types include different test items, and each failure type includes multiple test items. Therefore, the data used in the AI ​​model training of this application embodiment are all multi-dimensional data, not traditional one-dimensional data. Therefore, the AI ​​model obtained by training is more accurate and more comprehensive, and the diagnostic results output when using the trained AI model for diagnosis are also more accurate.

[0085] Devices A, B, and C are different power semiconductor devices. Subset1, Subset2, and Subset3 are different test item subsets; that is, each device includes at least three test item subsets. The content and number of test items included in each test item subset differ. In the table, the abnormal subset is assigned a second score of 1, and the normal subset is assigned a first score of 0. The sum of the first and second scores of all test item subsets for each power semiconductor device is used as the total abnormality score; that is, the total abnormality score for device A is 0, the total abnormality score for device B is 1, and the total abnormality score for device C is 0. If the total abnormality score is greater than or equal to a preset score threshold, the power semiconductor device is judged to be an abnormal device. For example, if the preset score threshold is 1, then device B is an abnormal device.

[0086] See Figure 3A The figure is a flowchart of a testing method for a power semiconductor device provided in an embodiment of this application.

[0087] This application provides a testing method for power semiconductor devices, including:

[0088] S301: Obtain the failure types of multiple power semiconductor devices.

[0089] As an example, failure types such as gate failure, PN junction failure, or breakdown ring failure can be obtained. Understandably, the more historical data available for the same power semiconductor device, the more diverse and comprehensive the failure types collected will be.

[0090] S302: Utilize expert knowledge to extract all test items corresponding to each failure type from the test items of power semiconductor devices, and form a subset of test items corresponding to each failure type.

[0091] In this example, gate failure corresponds to multiple test items such as IGES, VTH, VCESAT, and Eon, forming a test item subset 1. PN junction failure corresponds to test items such as VCESAT and ICES, forming a test item subset 2. Withstand ring failure corresponds to test items such as ICES, forming a test item subset 3.

[0092] S303: For each subset of test items, use an unsupervised learning model to distinguish between abnormal subsets and normal subsets, and record the first score and the second score for the abnormal subset and the normal subset respectively.

[0093] For example, an unsupervised learning model is used to detect outliers on the subset of test items corresponding to gate failures. Normal devices A, B, and C are assigned a second score. Similarly, an unsupervised learning model is used to detect outliers on the subset of test items corresponding to PN junction failures. Normal devices A and C are assigned a second score, while outlier B is assigned a first score. In the example in Table 2, the first score is 1, and the second score is 0.

[0094] In practical applications, to more accurately assess the risk of failure for each device, as a possible implementation, the cloud server also utilizes expert knowledge to assign different initial scores to abnormal devices in different test item subsets. That is, a weight corresponding to the test item subset is added to the initial score and the second score. For example, there are three failure types, corresponding to test item subsets a, b, and c, respectively. The weight of test subset a is 0.2, the weight of test subset b is 0.3, and the weight of test subset c is 0.5. Accordingly, if devices A and B in test subset b are normal, then devices A and B can each receive a second score of 0. If device C malfunctions, device C can receive an initial score of 1 * the weight of test subset b (0.3) = a first score of 0.3.

[0095] S304: Obtain the sum of the first and second scores of all test item subsets for each power semiconductor device as the total score for the anomaly level.

[0096] In the examples corresponding to Tables 1 and 2, the total anomaly level score for device A is the sum of the second scores of test item subset 1 (0), test item subset 2 (0), and test item subset 3 (0), which is 0. The total anomaly level score for device B is the sum of the second scores of test item subset 1 (0), test item subset 2 (1), and test item subset 3 (0), which is 1. The total anomaly level score for device C is the sum of the second scores of test item subset 1 (0), test item subset 2 (0), and test item subset 3 (0), which is 0.

[0097] S305: When the total score of the abnormal level is greater than or equal to the preset score threshold, the power semiconductor device is determined to be an abnormal device.

[0098] In this example, the preset score threshold can be 1. Of course, other values ​​can be obtained according to the actual situation, and this application does not limit this. When the preset score threshold is 1, since the total abnormality score of device B is 1, which is greater than or equal to the preset score threshold, device B is determined to be an abnormal device.

[0099] Therefore, it can be seen that by using multiple test items corresponding to each failure type and performing anomaly detection on a subset of test items containing multiple test items, the risk of power semiconductor device failure can be detected from multiple dimensions by utilizing the relationship between devices in the same subset of test items or the physical principle corresponding to the same failure type, thereby enabling a more accurate assessment of the risk of device failure.

[0100] The above embodiments describe a cloud server training an artificial intelligence model using an unsupervised learning model. As another possible implementation, we will now describe a cloud server training an artificial intelligence model using a supervised learning model.

[0101] See Figure 3B The figure is a flowchart of another power semiconductor device testing method provided in an embodiment of this application.

[0102] The testing method for power semiconductor devices provided in this application includes:

[0103] S311: Obtain relevant data when power semiconductor devices malfunction, and use it as data tags. For example, a very small number of failure samples can be selected as data tags.

[0104] S312: Use supervised learning models to extract data features from data labels.

[0105] S313: Utilize data features to obtain the total score of the anomaly level for each power semiconductor device.

[0106] S314: When the total score of the abnormal level is greater than the preset score threshold, the power semiconductor device is determined to be an abnormal device. Therefore, the solution provided in this application embodiment can obtain an artificial intelligence model for diagnosing power semiconductor devices through supervised or unsupervised learning models.

[0107] It should be understood that the method for screening defective power semiconductor devices using an artificial intelligence model in this application embodiment is an improvement upon the traditional method of setting upper and lower limit values ​​for diagnosis. Specifically, the power semiconductor device testing method in this application embodiment involves a testing machine first diagnosing the power semiconductor device under test using pre-set upper and lower limit values. The testing machine also compares the data of the power semiconductor device under test with the preset upper and lower limit values. If the data of the power semiconductor device under test exceeds the upper and lower limit values, the device is determined to be defective. Then, the testing machine or cloud server uses an AI model to diagnose the power semiconductor device under test again, and the diagnostic results from the two methods are logically ORed. Therefore, the testing system provided in this application embodiment can complement the two methods, thereby providing a more comprehensive diagnosis of the power semiconductor device under test.

[0108] It is understood that the power semiconductor device testing system provided in this application embodiment, while utilizing an artificial intelligence model to diagnose power semiconductor devices, can also use traditional methods to compare the data of the power semiconductor device under test with preset upper and lower limit values, thereby diagnosing the power semiconductor device. As a possible implementation, when at least one of the artificial intelligence model's approach and the upper and lower limit value approach diagnoses a power semiconductor device failure, the power semiconductor device is determined to be faulty; that is, the two diagnostic results are logically ORed. Thus, the power semiconductor device testing system provided in this application embodiment can combine traditional testing methods with an artificial intelligence model, further improving the accuracy of detection.

[0109] For example, if the data of power semiconductor device X does not exceed the preset upper and lower limits, but the AI ​​model outputs a failure result after inputting the data of power semiconductor device X, then power semiconductor device X is determined to be faulty. If the data of power semiconductor device Y exceeds the preset upper and lower limits, but the AI ​​model outputs a normal failure result after inputting the data of power semiconductor device Y, then power semiconductor device Y is determined to be faulty. If the data of power semiconductor device Z exceeds the preset upper and lower limits, and the AI ​​model outputs a failure result after inputting the data of power semiconductor device Z, then power semiconductor device Z is determined to be faulty.

[0110] In the power semiconductor device testing system provided in this application embodiment, in order to make the diagnosis of the artificial intelligence model more accurate, as a possible implementation, the testing machine is also used to send updated data to the cloud server; the cloud server is also used to update the artificial intelligence model according to the updated data.

[0111] It should be noted that the step of inputting the data of the power semiconductor device to be tested into a pre-trained artificial intelligence model and obtaining diagnostic results in this embodiment can be completed by a cloud server in the cloud or by a test machine on the board. This embodiment does not limit the scope of the application. The following will describe two different implementation methods in detail through embodiments.

[0112] See Figure 4 The figure is a schematic diagram of a test system for another power semiconductor device provided in an embodiment of this application.

[0113] The test machine 200 is also used to collect historical data from multiple power semiconductor devices and send the collected historical data from multiple power semiconductor devices to the cloud server 100. The cloud server 100 can train an artificial intelligence model using the historical data from multiple power semiconductor devices.

[0114] The cloud server 100 is also used to train an artificial intelligence model using at least one of a supervised learning model or an unsupervised learning model, based on historical data from multiple power semiconductor devices.

[0115] In this embodiment, historical data may include at least one of the following: chip test data of multiple power semiconductor devices, functional test data of packaged modules, or relevant data when anomalies occur in actual applications. Specifically, as a possible implementation, historical data may include... Figure 2 One or more items from the training data.

[0116] It should be noted that the artificial intelligence model trained in this embodiment is stored in a cloud server. When it is necessary to use the artificial intelligence model to detect the power semiconductor device under test, the cloud server uses the artificial intelligence model to detect the power semiconductor device under test and obtains the detection result.

[0117] The following section, with reference to the accompanying diagram, details the specific scheme by which the cloud server uses an artificial intelligence model to detect the power semiconductor device under test in this example.

[0118] See Figure 5 The figure is a schematic diagram of a test system for another power semiconductor device provided in an embodiment of this application.

[0119] The cloud server 100 is used to input data from the power semiconductor device under test obtained from the test machine into a pre-trained artificial intelligence model. The output of the pre-trained artificial intelligence model is the diagnostic result of the power semiconductor device under test. Then, the cloud server 100 sends the diagnostic result to the test machine 200, which finally outputs the diagnostic result.

[0120] It is understood that in the solution provided in this application embodiment, the cloud server 100 stores its trained artificial intelligence model. When it is necessary to use the artificial intelligence model to test the power semiconductor device under test, the cloud server 100 can obtain the data of the power semiconductor device under test from the test machine 200, and input this data into the pre-trained artificial intelligence model to obtain diagnostic results. Then, the cloud server 100 sends the obtained diagnostic results to the test machine 200, so that the test machine 200 on the end board outputs the diagnostic results. That is, the test machine 200 only receives the diagnostic results and does not test the power semiconductor device itself.

[0121] As another possible implementation, the step of inputting the data of the power semiconductor to be tested into a pre-trained artificial intelligence model and obtaining diagnostic results in this embodiment can also be completed by the testing machine at the board end. The training process of the AI ​​model will be introduced below.

[0122] See Figure 6 The figure is a schematic diagram of a test system for another power semiconductor device provided in an embodiment of this application.

[0123] The test machine 200 is also used to collect historical data from multiple power semiconductor devices and send the collected historical data to the cloud server 100, thereby enabling the cloud server 100 to train an artificial intelligence model using the historical data from the multiple power semiconductor devices. The historical data includes at least one of the chip test data or packaged module functional test data of the multiple power semiconductor devices in the above embodiments. It should be noted that after the artificial intelligence model is trained in this embodiment, the cloud server 100 sends the obtained artificial intelligence model to the board-side test machine 200, and the test machine 200 stores the artificial intelligence model.

[0124] The following section, with reference to the accompanying drawings, details the specific scheme by which the test machine uses an artificial intelligence model to detect the power semiconductor device under test in this example.

[0125] See Figure 7 The figure is a schematic diagram of a test system for another power semiconductor device provided in an embodiment of this application.

[0126] The cloud server 100 is used to send the pre-trained artificial intelligence model to the test machine 200. The test machine 200 is used to input the data of the power semiconductor device under test into the pre-trained artificial intelligence model, and the output of the pre-trained artificial intelligence model is the diagnostic result of the power semiconductor device under test.

[0127] It is understood that in the solution provided in this application embodiment, the test machine 200 will receive the artificial intelligence model sent by the cloud server 100 and store the artificial intelligence model. When it is necessary to use the artificial intelligence model to detect the power semiconductor device under test, the test machine 200 will input the data of the power semiconductor device under test into the artificial intelligence model, obtain and output the diagnostic results.

[0128] It should be noted that, in this embodiment, since the artificial intelligence model is stored in the testing machine, the testing machine can directly utilize the artificial intelligence model to complete the task of testing the power semiconductor device under test independently, without needing to communicate with the cloud server. Therefore, it avoids the situation where the network between the testing machine on the board and the cloud server is interrupted, preventing the power semiconductor device testing system in this embodiment from failing to obtain diagnostic results for the power semiconductor device under test. Thus, the power semiconductor device testing system provided in this embodiment reduces the network dependency between the cloud server and the testing machine, improving the real-time performance of obtaining diagnostic results.

[0129] In summary, the testing system provided in this application can either input the data of the power semiconductor under test into a pre-trained artificial intelligence model via a cloud server in the cloud and obtain diagnostic results, or input the data of the power semiconductor under test into a pre-trained artificial intelligence model via a test machine on the board and obtain diagnostic results. When obtaining diagnostic results from the cloud server, the cloud server can directly store the trained artificial intelligence model without needing to distribute it to the test machine on the board. When obtaining diagnostic results from the test machine on the board, the cloud server needs to distribute the pre-trained artificial intelligence model to the test machine, which can then complete the testing of the power semiconductor device under test independently. When obtaining diagnostic results from the test machine, the cloud server does not need to distribute the diagnostic results to the test machine; therefore, the test machine's acquisition of diagnostic results is not affected by network failures and has low network dependence.

[0130] The testing system provided in this application embodiment may include one testing machine or multiple testing machines. The working principle of the testing system with multiple testing machines provided in this application embodiment is described below with reference to the accompanying drawings.

[0131] See Figure 8 The figure is a schematic diagram of a test system for a power semiconductor device that includes multiple testers, provided in an embodiment of this application.

[0132] The power semiconductor device testing system provided in this application embodiment includes at least the following two testing machines: a first testing machine 201 and a second testing machine 202;

[0133] The cloud server 100 is specifically used to train a global artificial intelligence model based on the first historical data sent by the first test machine 201 and the second historical data sent by the second test machine 202, to test the power semiconductor device under test corresponding to the first test machine 201 using the global artificial intelligence model to obtain a first diagnostic result, to test the power semiconductor device under test corresponding to the second test machine 202 using the global artificial intelligence model to obtain a second diagnostic result, to send the first diagnostic result to the first test machine 201, and to send the second diagnostic result to the second test machine 202.

[0134] It is understood that the cloud server provided in this application embodiment uses the first historical data and the second historical data to train the global artificial intelligence model. It can make full use of all historical data, thereby improving the generalization ability of the global artificial intelligence model. Using the more accurate global artificial intelligence model to diagnose the power semiconductor device under test can improve the universality of the artificial intelligence model and increase the accuracy of diagnosing new devices.

[0135] The cloud server provided in this application embodiment can either send diagnostic results directly to the test machine on the board, or send a trained global artificial intelligence model to the test machine for each test machine to complete the test. This application embodiment does not limit this.

[0136] See Figure 9 The figure is a schematic diagram of another test system for power semiconductor devices that includes multiple testers, provided in an embodiment of this application.

[0137] The power semiconductor device testing system provided in this application includes at least the following two testing machines: a first testing machine 201 and a second testing machine 202.

[0138] Specifically, the cloud server 100 is used to train a global artificial intelligence model based on the first historical data sent by the first test machine 201 and the second historical data sent by the second test machine 202, and then send the global artificial intelligence model to the first test machine 201 and the second test machine 202.

[0139] The first tester 201 is used to test the corresponding power semiconductor device under test using a global artificial intelligence model; the second tester 202 is used to test the corresponding power semiconductor device under test using a global artificial intelligence model.

[0140] The testing system provided in this application embodiment allows the testing machine to directly utilize the stored global AI model, enabling it to perform the task of testing the power semiconductor device under test independently, without needing to communicate with the cloud server. This avoids the situation where the testing system cannot obtain diagnostic results for the power semiconductor device under test when the network between the testing machine on the board and the cloud server is interrupted. Thus, the testing system for power semiconductor devices provided in this application embodiment reduces the dependence on network stability between the cloud server and the testing machine, improving the real-time performance of obtaining diagnostic results.

[0141] As can be seen from the above, the power semiconductor device testing system provided in this application embodiment can obtain a global artificial intelligence model through all historical data, and then use this global artificial intelligence model to obtain the test results. However, considering that the testing machine may have certain systematic errors, the systematic errors in the data measured by different testing machines are different, and this error may be related to the testing machine itself. Therefore, this application embodiment also provides a power semiconductor device testing system that, after obtaining the global artificial intelligence model, fine-tunes the global artificial intelligence model based on the historical data obtained by different testing machines to obtain local artificial intelligence models corresponding to different testing machines.

[0142] See Figure 10 The figure is a schematic diagram of another test system for power semiconductor devices including multiple testers provided in an embodiment of this application.

[0143] like Figure 10 As shown, the power semiconductor device test system provided in this application includes at least the following two test machines: a first test machine 201 and a second test machine 202.

[0144] Specifically, the cloud server 100 is used to train a global artificial intelligence model based on the first historical data sent by the first test machine 201 and the second historical data sent by the second test machine 202; to adjust the global artificial intelligence model using the first historical data to obtain a first artificial intelligence model; to adjust the global artificial intelligence model using the second historical data to obtain a second artificial intelligence model; to test the power semiconductor device under test corresponding to the first test machine 201 using the first artificial intelligence model to obtain a first diagnostic result; to test the power semiconductor device under test corresponding to the second test machine 202 using the second artificial intelligence model to obtain a second diagnostic result; to send the first diagnostic result to the first test machine 201; and to send the second diagnostic result to the second test machine 202.

[0145] It should be noted that, in this embodiment, the first historical data sent by the first test machine 201 is the historical data corresponding to the first test machine 201. That is, the first historical data includes data directly or indirectly measured by the first test machine 201. The second historical data sent by the first test machine 202 is the historical data corresponding to the second test machine 202.

[0146] This application embodiment uses first and second historical data to train a global artificial intelligence model, which can fully utilize all historical data and improve the testing accuracy of the global artificial intelligence model. Furthermore, considering the potential performance differences between different testing machines, this application embodiment also uses the first historical data to adjust the global artificial intelligence model to obtain a first artificial intelligence model, and uses the second historical data to adjust the global artificial intelligence model to obtain a second artificial intelligence model. This results in a local artificial intelligence model (first artificial intelligence model) tailored to the characteristics of the first testing machine and a local artificial intelligence model (second artificial intelligence model) tailored to the characteristics of the second testing machine, further improving the testing accuracy of the artificial intelligence model used in actual testing.

[0147] For example, the power semiconductor device testing system provided in this application includes a test machine Q and a test machine W. Test machine Q sends its historical data to a cloud server, and test machine W sends its historical data to the cloud server. The cloud server uses the historical data from test machine Q and test machine W together to train a global artificial intelligence model. Then, the cloud server uses the historical data from test machine Q to adjust the global artificial intelligence model, obtaining a local artificial intelligence model corresponding to test machine Q, and uses this local artificial intelligence model to obtain the power semiconductor device to be tested corresponding to test machine Q for testing. The cloud server uses the historical data from test machine W to adjust the global artificial intelligence model, obtaining a local artificial intelligence model corresponding to test machine W, and uses this local artificial intelligence model to obtain the power semiconductor device to be tested corresponding to test machine W for testing.

[0148] The cloud server provided in this application embodiment can either send diagnostic results directly to the test machine on the board, or send a trained artificial intelligence model to the test machine on the board. The test machine on the board can then use the trained artificial intelligence model to test the power semiconductor device and obtain diagnostic results. The following is a detailed description in conjunction with the accompanying drawings.

[0149] See Figure 11 The figure is a schematic diagram of another power semiconductor device testing system including multiple testers provided in the embodiments of this application.

[0150] The power semiconductor device test system provided in this application includes at least the following two test machines: a first test machine 201 and a second test machine 202;

[0151] The cloud server 100 is specifically used to train a global artificial intelligence model based on the first historical data sent by the first test machine 201 and the second historical data sent by the second test machine 202, adjust the global artificial intelligence model using the first historical data to obtain a first artificial intelligence model and send it to the first test machine 201, and adjust the global artificial intelligence model using the second historical data to obtain a second artificial intelligence model and send it to the second test machine 202.

[0152] The first tester 201 is specifically used to test the corresponding power semiconductor device under test using the first artificial intelligence model; the second tester 202 is specifically used to test the corresponding power semiconductor device under test using the second artificial intelligence model.

[0153] In this embodiment, since the cloud server directly distributes the first and second artificial intelligence models to the first and second test machines respectively, the test machines can directly utilize the corresponding artificial intelligence models to complete the task of testing the power semiconductor device under test independently, without needing to communicate with the cloud server. This avoids the situation where the power semiconductor device testing system in this embodiment cannot obtain the diagnostic results of the power semiconductor device under test when the network between the test machine on the board and the cloud server is interrupted. Thus, the power semiconductor device testing system provided in this embodiment reduces the dependence on network stability between the cloud server and the test machine, improving the real-time performance of obtaining diagnostic results.

[0154] As can be seen from the above, the power semiconductor device testing system provided in this application embodiment can either directly use all historical data to obtain a global artificial intelligence model and use it directly, or after obtaining the global artificial intelligence model, use the historical data corresponding to different test machines to adjust the global artificial intelligence model to obtain a local artificial intelligence model, and use the local artificial intelligence model to test the power semiconductor device of the corresponding test machine.

[0155] To obtain a more accurate artificial intelligence model, as one possible implementation method, this application embodiment can further test the power semiconductor devices diagnosed as faulty to verify whether the power semiconductor devices are indeed faulty, and thus verify the accuracy of the AI ​​model. If the power semiconductor devices are found not to be faulty, it indicates that the AI ​​model is inaccurate and needs to be adjusted. For power semiconductor devices determined not to be faulty, their operational data on the application side is continuously collected, and then the data from both types of power semiconductor devices is uploaded to a cloud server. The cloud server uses this data to adjust the AI ​​model.

[0156] To improve the accuracy of the artificial intelligence model in this application embodiment, as a possible implementation, the testing machine in this application embodiment is further used to fine-tune the pre-trained model by combining data from the testing machine side, and to use the fine-tuned model to diagnose the power semiconductor device under test and output diagnostic results. It should be noted that fine-tuning in this application embodiment may include parameter optimization of the pre-trained model. As an example, transfer learning methods can be used to fine-tune the pre-trained AI model. For example, the fine-tuned AI model can be used to test whether a wafer is faulty.

[0157] Considering that the test machine in this embodiment stores a small amount of historical data it has collected, but this data has not been uploaded to the cloud server, as a possible implementation, the data on the test machine side in this embodiment may also include data that has not been uploaded to the cloud server. It is understood that the test machine in this embodiment typically uploads its collected data to the cloud server periodically. The test machine in this embodiment typically stores a certain amount of historical data that has not been uploaded. Therefore, the test machine can use this unuploaded data to fine-tune the AI ​​model, and then test power semiconductor devices based on the fine-tuned AI model. After fine-tuning the AI ​​model, the test machine in this embodiment can also upload the fine-tuned AI model to the cloud server so that the cloud server can manage the AI ​​model uniformly.

[0158] In summary, the power semiconductor device testing system provided in this application, considering the limited computing and storage capabilities of the testing machine, utilizes a cloud server with greater computing and storage capabilities to train an artificial intelligence model. This allows for the direct acquisition of diagnostic results using the artificial intelligence model on the cloud server, or the sending of the artificial intelligence model to the on-board testing machine for diagnostic results. Alternatively, the system can send an incompletely trained artificial intelligence model to the on-board testing machine, which then trains the model based on its stored data and uses the trained model to obtain diagnostic results.

[0159] Server Example

[0160] Based on the power semiconductor device testing system provided in the above embodiments, this application also provides a cloud server.

[0161] See Figure 12 The figure is a schematic diagram of the structure of a cloud server provided in an embodiment of this application.

[0162] like Figure 12 As shown in the embodiment of this application, the cloud server provided includes: a first transceiver device 1201 and a first controller 1202.

[0163] The first transceiver 1201 is used to receive test data of the power semiconductor device under test sent by the tester.

[0164] The first controller 1202 is used to train an artificial intelligence model in advance using historical data from multiple power semiconductor devices; it is also used to input data from the power semiconductor device under test into the artificial intelligence model, and the output of the pre-trained AI model is the diagnostic result of the power semiconductor device under test; the first transceiver device is also used to send the diagnostic result to the test machine.

[0165] Alternatively, the first controller 1202 is used to send a pre-trained artificial intelligence model to the test machine so that the test machine can use the pre-trained artificial intelligence model to diagnose the power semiconductor device under test.

[0166] For a detailed description of the cloud server provided in this application embodiment, please refer to the above test system embodiment for an introduction to the cloud server. Here, we will only provide a brief overview.

[0167] In one possible implementation of this application, the first controller is specifically used to obtain the failure types of multiple power semiconductor devices when training an artificial intelligence model using an unsupervised learning model. It uses expert knowledge to extract all test items corresponding to each failure type from the test items of the power semiconductor devices. All test items corresponding to each failure type form a subset of test items. Anomaly level detection is performed on each subset of test items using an unsupervised learning model, resulting in a first score and a second score for abnormal devices and normal devices, respectively. The sum of the first score and the second score of all test item subsets for each power semiconductor device is obtained as the total anomaly level score. When the total anomaly level score is greater than a preset score threshold, the power semiconductor device is determined to be an abnormal device.

[0168] In one possible implementation of this application, the cloud server corresponds to at least two test machines: a first test machine and a second test machine. Specifically, the first controller is used to train a global artificial intelligence model based on first historical data sent by the first test machine and second historical data sent by the second test machine; adjust the global artificial intelligence model using the first historical data to obtain a first artificial intelligence model; adjust the global artificial intelligence model using the second historical data to obtain a second artificial intelligence model; test the power semiconductor device under test corresponding to the first test machine using the first artificial intelligence model to obtain a first diagnostic result; and test the power semiconductor device under test corresponding to the second test machine using the second artificial intelligence model to obtain a second diagnostic result. The first transceiver device is specifically used to send the first diagnostic result to the first test machine and also to send the second diagnostic result to the second test machine.

[0169] In one possible implementation of this application, the cloud server corresponds to at least two test machines: a first test machine and a second test machine. Specifically, the first controller is used to train a global artificial intelligence model based on first historical data sent by the first test machine and second historical data sent by the second test machine. The global artificial intelligence model is then used to test the power semiconductor device under test corresponding to the first test machine to obtain a first diagnostic result, and the global artificial intelligence model is used to test the power semiconductor device under test corresponding to the second test machine to obtain a second diagnostic result. The first transceiver is specifically used to send the first diagnostic result to the first test machine and also to send the second diagnostic result to the second test machine.

[0170] In one possible implementation of this application, the first transceiver device is further configured to receive update data sent by the test machine; the first controller is further configured to update the artificial intelligence model based on the update data.

[0171] The cloud server provided in this application embodiment, in addition to including a first transceiver device and a first controller, may also include a first storage device (not shown in the figure). The first storage device can be used to store historical data and updated data uploaded by the test machine, facilitating subsequent processing by the first controller. Before the first controller uses historical data to train the AI ​​model, in order to make the parameters of the trained AI model more accurate, some interfering data can be removed; specifically, data cleaning can be performed.

[0172] In summary, the cloud server in this embodiment can utilize its powerful computing and storage capabilities to train the artificial intelligence model, thereby obtaining an AI model with high diagnostic accuracy. After obtaining the AI ​​model, diagnostic results can be obtained directly from the AI ​​model on the cloud server, or the AI ​​model can be provided only to the test machine without providing direct diagnostic results.

[0173] Test machine example

[0174] Based on the power semiconductor device testing system and cloud server provided in the above embodiments, this application also provides a testing machine, which will be described in detail below with reference to the accompanying drawings.

[0175] See Figure 13 The figure is a schematic diagram of the structure of a testing machine provided in an embodiment of this application.

[0176] like Figure 13 As shown, the test machine in this embodiment includes: a second transceiver device 1301 and a second controller 1302;

[0177] The second transceiver device 1301 is used to receive the artificial intelligence model sent by the cloud server. The artificial intelligence model is obtained by the cloud server in advance by training with historical data from multiple power semiconductor devices.

[0178] The second controller 1302 is used to obtain data from the power semiconductor device under test; input the data from the power semiconductor device under test into a pre-trained artificial intelligence model, and output the pre-trained AI model as the diagnostic result of the power semiconductor device under test.

[0179] The test machine provided in this application embodiment refers to any test machine. One cloud server can correspond to multiple test machines provided in this application embodiment. For the specific working method of the test machine, please refer to the introduction of the test machine in the above test system embodiment. Only a brief description is given here.

[0180] In one possible implementation of this application, the second controller is further configured to collect historical data of multiple power semiconductor devices and send the collected historical data of multiple power semiconductor devices to a cloud server; the historical data includes at least one of chip test data or packaged module functional test data of multiple power semiconductor devices.

[0181] In one possible implementation of this application, the second controller is further configured to compare the data of the power semiconductor device under test with preset upper and lower limit values. If the data of the power semiconductor device under test exceeds the upper and lower limit values, the power semiconductor device under test is determined to be an abnormal device. Furthermore, in this application, the testing machine can directly obtain diagnostic results sent from the cloud server, or it can obtain an artificial intelligence model sent from the cloud server and obtain diagnostic results within the testing machine using the artificial intelligence model.

[0182] In summary, the testing machine provided in this application embodiment has limited computing and storage capabilities. Therefore, it sends data to a cloud server and uses the powerful computing and storage capabilities of the cloud server to train the artificial intelligence model, thereby obtaining an artificial intelligence model with high diagnostic accuracy.

[0183] It should be understood that in this application, "at least one (item)" means one or more, and "more than" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0184] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A testing system for power semiconductor devices, characterized in that, include: A cloud server and at least one test machine; The cloud server is used to pre-train an artificial intelligence model using historical data from multiple power semiconductor devices. The testing machine is used to obtain data from the power semiconductor device under test; The cloud server or the testing machine is used to input the data of the power semiconductor device under test into the artificial intelligence model, and the output of the artificial intelligence model is the diagnostic result of the power semiconductor device under test; The cloud server is used to input the data of the power semiconductor device under test into the artificial intelligence model, the output of the artificial intelligence model is the diagnostic result of the power semiconductor device under test, and the diagnostic result is sent to the testing machine. The cloud server trains the artificial intelligence model using an unsupervised learning model. Specifically, it obtains the failure types of the multiple power semiconductor devices, extracts all test items corresponding to each failure type from the test items of the power semiconductor devices using expert knowledge, forms a test item subset with all test items corresponding to each failure type, distinguishes between abnormal and normal subsets for each test item subset using the unsupervised learning model, and assigns a first score and a second score to the abnormal subset and the normal subset respectively. The sum of the first score and the second score of all test item subsets of each power semiconductor device is obtained as the total abnormality level score. When the total abnormality level score is greater than or equal to a preset score threshold, the power semiconductor device is determined to be an abnormal device.

2. The system according to claim 1, characterized in that, The cloud server is used to send pre-trained artificial intelligence models to the test machine; The testing machine is used to input the data of the power semiconductor device under test into the artificial intelligence model, and the output of the artificial intelligence model is the diagnostic result of the power semiconductor device under test.

3. The system according to claim 1 or 2, characterized in that, The testing machine is also used to collect historical data of the plurality of power semiconductor devices and send the collected historical data of the plurality of power semiconductor devices to the cloud server; the historical data includes at least one of the following: chip test data of the plurality of power semiconductor devices, functional test data of the packaged module, or relevant data when there are abnormalities in actual applications; The cloud server is also used to train the artificial intelligence model using at least one of a supervised learning model or an unsupervised learning model based on historical data from the plurality of power semiconductor devices.

4. The system according to claim 1, characterized in that, The cloud server is specifically used to assign different first scores to abnormal devices in different subsets of test items using expert knowledge.

5. The system according to claim 3, characterized in that, The cloud server uses the supervised learning model to train the artificial intelligence model, specifically to obtain relevant data when the power semiconductor device is abnormal as data labels, use the supervised learning model to extract data features of the data labels, use the data features to obtain the total abnormality level score of each power semiconductor device, and when the total abnormality level score is greater than a preset score threshold, the power semiconductor device is determined to be an abnormal device.

6. The system according to claim 1 or 2, characterized in that, The testing machine is also used to compare the data of the power semiconductor device under test with preset upper and lower limit values. When the data of the power semiconductor device under test exceeds the upper and lower limit values, the power semiconductor device under test is determined to be an abnormal device.

7. The system according to claim 1, characterized in that, The system includes at least two testing machines: a first testing machine and a second testing machine; The cloud server is specifically used to train a global artificial intelligence model based on first historical data sent by the first test machine and second historical data sent by the second test machine; to adjust the global artificial intelligence model using the first historical data to obtain a first artificial intelligence model; to adjust the global artificial intelligence model using the second historical data to obtain a second artificial intelligence model; to test the power semiconductor device under test corresponding to the first test machine using the first artificial intelligence model to obtain a first diagnostic result; to test the power semiconductor device under test corresponding to the second test machine using the second artificial intelligence model to obtain a second diagnostic result; and to send the first diagnostic result to the first test machine and the second diagnostic result to the second test machine.

8. The system according to claim 1, characterized in that, The system includes at least two testing machines: a first testing machine and a second testing machine; The cloud server is specifically used to train a global artificial intelligence model based on the first historical data sent by the first test machine and the second historical data sent by the second test machine, use the global artificial intelligence model to test the power semiconductor device under test corresponding to the first test machine to obtain a first diagnostic result, use the global artificial intelligence model to test the power semiconductor device under test corresponding to the second test machine to obtain a second diagnostic result, send the first diagnostic result to the first test machine, and send the second diagnostic result to the second test machine.

9. The system according to claim 2, characterized in that, The system includes at least two testing machines: a first testing machine and a second testing machine; The cloud server is specifically used to train a global artificial intelligence model based on the first historical data sent by the first test machine and the second historical data sent by the second test machine, adjust the global artificial intelligence model using the first historical data to obtain a first artificial intelligence model and send it to the first test machine, and adjust the global artificial intelligence model using the second historical data to obtain a second artificial intelligence model and send it to the second test machine. The first test machine is specifically used to diagnose the corresponding power semiconductor device under test using the first artificial intelligence model; The second testing machine is specifically used to diagnose the corresponding power semiconductor device under test using the second artificial intelligence model.

10. The system according to claim 2, characterized in that, The system includes at least two testing machines: a first testing machine and a second testing machine; The cloud server is specifically used to train a global artificial intelligence model based on the first historical data sent by the first test machine and the second historical data sent by the second test machine, and to send the global artificial intelligence model to the first test machine and the second test machine. The first test machine is used to diagnose the corresponding power semiconductor device under test using the global artificial intelligence model; The second test machine is used to diagnose the corresponding power semiconductor device under test using the global artificial intelligence model.

11. The system according to claim 1 or 2, characterized in that, The test machine is also used to send update data to the cloud server; The cloud server is also used to update the artificial intelligence model based on the updated data.

12. The system according to claim 2, characterized in that, The test machine is also used to fine-tune the pre-trained AI model by combining the data from the test machine side, and to use the fine-tuned model to diagnose the power semiconductor device under test and output the diagnostic results.

13. A cloud server, characterized in that, include: First transceiver and first controller; The first transceiver is used to receive test data of the power semiconductor device under test sent by the test machine; The first controller is configured to obtain an artificial intelligence model by pre-training using historical data from multiple power semiconductor devices; it is also configured to input data from the power semiconductor device under test into the artificial intelligence model, and the output of the artificial intelligence model is a diagnostic result of the power semiconductor device under test; the first transceiver is further configured to send the diagnostic result to the testing machine; or, The first controller is configured to send the artificial intelligence model to the test machine, so that the test machine can use the artificial intelligence model to diagnose the power semiconductor device under test; Specifically, the first controller, when training the artificial intelligence model using an unsupervised learning model, is used to obtain the failure types of the multiple power semiconductor devices, extract all test items corresponding to each failure type from the test items of the power semiconductor devices using expert knowledge, form a subset of test items corresponding to each failure type, perform anomaly level detection on each subset of test items using an unsupervised learning model, and obtain a first score and a second score for abnormal devices and normal devices respectively, and obtain the sum of the first score and the second score of all subsets of test items for each power semiconductor device as the total anomaly level score; when the total anomaly level score is greater than a preset score threshold, the power semiconductor device is determined to be an abnormal device.

14. The cloud server according to claim 13, characterized in that, The cloud server corresponds to at least two test machines: a first test machine and a second test machine; The first controller is specifically configured to train a global artificial intelligence model based on first historical data sent by the first test machine and second historical data sent by the second test machine; adjust the global artificial intelligence model using the first historical data to obtain a first artificial intelligence model; adjust the global artificial intelligence model using the second historical data to obtain a second artificial intelligence model; test the power semiconductor device under test corresponding to the first test machine using the first artificial intelligence model to obtain a first diagnostic result; and test the power semiconductor device under test corresponding to the second test machine using the second artificial intelligence model to obtain a second diagnostic result. The first transceiver is specifically used to send the first diagnostic result to the first test machine, and also to send the second diagnostic result to the second test machine.

15. The cloud server according to claim 13, characterized in that, The cloud server corresponds to at least two test machines: a first test machine and a second test machine; The first controller is specifically used to train a global artificial intelligence model based on the first historical data sent by the first test machine and the second historical data sent by the second test machine, and to use the global artificial intelligence model to test the power semiconductor device under test corresponding to the first test machine to obtain a first diagnostic result, and to use the global artificial intelligence model to test the power semiconductor device under test corresponding to the second test machine to obtain a second diagnostic result. The first transceiver is specifically used to send the first diagnostic result to the first test machine, and also to send the second diagnostic result to the second test machine.

16. The cloud server according to any one of claims 13-15, characterized in that, The first transceiver device is also used to receive update data sent by the test machine; The first controller is further configured to update the artificial intelligence model based on the updated data.

17. A testing machine, characterized in that, include: Second transceiver and second controller; The second transceiver device is used to receive an artificial intelligence model sent by a cloud server. The artificial intelligence model is obtained by the cloud server in advance by training with historical data from multiple power semiconductor devices. The second controller is used to obtain data from the power semiconductor device under test; input the data from the power semiconductor device under test into the artificial intelligence model, and the output of the artificial intelligence model is the diagnostic result of the power semiconductor device under test; Alternatively, the second controller is configured to receive the diagnostic results of the power semiconductor device under test sent by the cloud server, wherein the diagnostic results of the power semiconductor device under test are obtained by the cloud server inputting the data of the power semiconductor device under test into the artificial intelligence model; The artificial intelligence model is trained by the cloud server using an unsupervised learning model. The second controller is specifically used to obtain the failure types of the multiple power semiconductor devices, extract all test items corresponding to each failure type from the test items of the power semiconductor devices using expert knowledge, form a test item subset with all test items corresponding to each failure type, distinguish between abnormal subset and normal subset using an unsupervised learning model for each test item subset, and record a first score and a second score for the abnormal subset and the normal subset respectively, and obtain the sum of the first score and the second score of all test item subsets of each power semiconductor device as the total score of the abnormality level. When the total score of the abnormal level is greater than or equal to a preset score threshold, the power semiconductor device is determined to be an abnormal device.

18. The testing machine according to claim 17, characterized in that, The second controller is further configured to collect historical data of the plurality of power semiconductor devices and send the collected historical data of the plurality of power semiconductor devices to the cloud server; the historical data includes at least one of the chip test data or the packaged module functional test data of the plurality of power semiconductor devices.

19. The testing machine according to claim 17 or 18, characterized in that, The second controller is further configured to compare the data of the power semiconductor device under test with preset upper and lower limit values, and when the data of the power semiconductor device under test exceeds the upper and lower limit values, determine that the power semiconductor device under test is an abnormal device.

Images