Model training methods, online junction temperature monitoring methods and devices for semiconductor devices

By constructing a Gaussian process regression model that combines on-state voltage drop and reverse voltage drop, the problem of SiC MOSFET junction temperature monitoring accuracy being affected by load current was solved, achieving higher accuracy and more stable junction temperature monitoring while reducing costs.

CN117572187BActive Publication Date: 2026-11-17TSINGHUA UNIVERSITY
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202311561391.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2023-11-22
Publication Date
2026-11-17
Estimated Expiration
2043-11-22

AI Technical Summary

Technical Problem

In existing technologies, the junction temperature monitoring accuracy of SiC MOSFETs is greatly affected by the load current, resulting in large monitoring errors and high costs.

Method used

By acquiring data from the training set, including junction temperature, on-state voltage drop, reverse voltage drop, and load current, an initial model is trained using Gaussian process regression to construct the first target model, eliminating the influence of load current on junction temperature monitoring, and using a combination of reverse voltage drop and on-state voltage drop for junction temperature monitoring.

Benefits of technology

It improves the accuracy and stability of junction temperature monitoring, reduces the impact of changes in operating conditions on monitoring results, and lowers monitoring costs.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117572187B_ABST
    Figure CN117572187B_ABST
Patent Text Reader

Abstract

This disclosure provides a model training method, an online junction temperature monitoring method for semiconductor devices, an apparatus, a device, and a storage medium, which can be applied to the field of power electronic device condition monitoring and thermal management technology. The method includes: acquiring a training set corresponding to a target semiconductor device, wherein the training set includes n sets of data, where n is a positive integer, and each set of data includes the junction temperature, the on-state voltage drop, the reverse voltage drop, and the load current corresponding to the junction temperature; training an initial model based on the training set to obtain a first target model used to eliminate the influence of the load current on junction temperature monitoring, wherein the initial model represents a functional model of the relationship between junction temperature and on-state voltage drop, reverse voltage drop, and load current, and the first target model represents a functional model of the relationship between junction temperature and on-state voltage drop and reverse voltage drop.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This disclosure relates to the field of power electronic device condition monitoring and thermal management, and in particular to a model training method, a method, apparatus, equipment, medium, and program product for online monitoring of junction temperature of semiconductor devices. Background Technology

[0002] Since their inception, SiC (Silicon Carbide) devices have occupied an important position in the converter market due to their excellent characteristics such as high power density, high switching frequency, and high switching speed. However, these characteristics also pose more challenges to their reliability and thermal management. Among them, high power density leads to greater thermal stress, and high switching frequency leads to greater switching losses.

[0003] As a representative of SiC devices, the reliability of SiC MOSFETs (Metal-Oxide-Semiconductor Field-Effect Transistors) has become a major concern. Statistics show that power device failure rates and repair costs are among the highest in converter systems. The main factors leading to SiC MOSFET failure are overtemperature and junction temperature fluctuations. Overtemperature can cause instantaneous failures such as thermal breakdown of the insulation, while junction temperature fluctuations accelerate device aging and shorten its lifespan.

[0004] As can be seen from the above, junction temperature monitoring of SiC devices is fundamental to their reliability. Therefore, improving the accuracy of junction temperature monitoring of semiconductor devices is a technical problem that needs to be solved in related technologies. Summary of the Invention

[0005] In view of the above problems, this disclosure provides a model training method, a method, apparatus, device, medium and program product for online monitoring of junction temperature of semiconductor devices.

[0006] According to a first aspect of this disclosure, a model training method is provided, comprising:

[0007] Obtain a training set corresponding to the target semiconductor device, wherein the training set includes n sets of data, where n is a positive integer, and each set of data includes the junction temperature, the on-state voltage drop, the reverse voltage drop, and the load current corresponding to the junction temperature; and

[0008] Based on the training set mentioned above, the initial model is trained to obtain the first target model used to eliminate the influence of load current on junction temperature monitoring. The initial model represents the functional model of the relationship between junction temperature and on-state voltage drop, reverse voltage drop and load current, and the first target model represents the functional model of the relationship between junction temperature and on-state voltage drop and reverse voltage drop.

[0009] According to embodiments of this disclosure, obtaining the training set corresponding to the target semiconductor device includes:

[0010] During each of the n double pulses, a set of data from the training set is measured, wherein the n double pulses are sent to the target semiconductor device.

[0011] According to embodiments of this disclosure, the aforementioned double pulse includes a first pulse and a second pulse, and during each of the n double pulses, a set of data in the aforementioned training set is measured, including:

[0012] For each of the above-mentioned double pulses, during the period of the first pulse, the load inductor on the heating plate is charged with DC current so that after the end of the first pulse, the load current of the load inductor is measured, wherein the load current of the load inductor characterizes the load current of the target semiconductor device.

[0013] Between the first pulse and the second pulse, the reverse voltage drop of the target semiconductor device is measured based on the freewheeling current of the load current in the semiconductor device located in the upper bridge arm, wherein the target semiconductor device is located in the lower bridge arm, and both the semiconductor device and the target semiconductor device are located on the heating plate.

[0014] During the second pulse, the on-state voltage drop of the target semiconductor device is measured based on the commutation of the load current from the semiconductor device to the target semiconductor device.

[0015] After the second pulse ends, the junction temperature of the target semiconductor device is measured, wherein the junction temperature of the target semiconductor device characterizes the temperature of the heating plate.

[0016] According to embodiments of this disclosure, the magnitude of the load current of the target semiconductor device is controlled by adjusting the width of the first pulse in the double pulse to obtain the on-state voltage drop, reverse voltage drop, and junction temperature under different load currents.

[0017] According to embodiments of this disclosure, the initial model is trained based on the training set to obtain a first target model for eliminating the influence of load current on junction temperature monitoring, including:

[0018] The above initial model is constructed based on the implicit functions corresponding to the on-state voltage drop, reverse voltage drop, and load current, wherein the above implicit functions follow a Gaussian distribution, and the above initial model follows a Gaussian distribution.

[0019] Determine the mean function and covariance matrix of the initial model described above;

[0020] Based on the Gaussian process corresponding to the training set and the Gaussian process corresponding to the prediction set, the joint distribution corresponding to the training set and the prediction set is obtained. The Gaussian process is determined by the mean function and the covariance matrix. The prediction set includes m sets of data, where m is a positive integer. In each set of data, the on-state voltage drop, reverse voltage drop, and load current are known, while the junction temperature is unknown.

[0021] Based on the joint distribution and prior distribution described above, the conditional distribution corresponding to the prediction set is obtained, wherein the prior distribution characterizes the Gaussian process corresponding to the training set; and

[0022] Based on the above conditional distribution, the first target model is determined. Using the first target model, the junction temperature corresponding to each set of data in the above prediction set is obtained based on the on-state pressure drop and reverse pressure drop in the above prediction set.

[0023] According to embodiments of this disclosure, determining the mean function and covariance matrix of the initial model includes:

[0024] Using a kernel function, an element in the covariance matrix is ​​determined for any two sets of data in the training set, where the mean function is set to zero, and the covariance matrix represents an n*n matrix.

[0025] According to embodiments of this disclosure, the first target model is determined based on the aforementioned conditional distribution. Using this first target model, and based on the on-state voltage drop and reverse voltage drop in the aforementioned prediction set, the junction temperature corresponding to each set of data in the aforementioned prediction set is obtained, including:

[0026] Based on the above conditional distribution, the first initial target model is determined;

[0027] Parameter estimation is performed on the unknown hyperparameters in the first initial target model to obtain the hyperparameter values ​​corresponding to each unknown hyperparameter; and

[0028] Based on the first initial target model and each hyperparameter value, the first target model is obtained.

[0029] According to embodiments of this disclosure, the above-described model training method further includes:

[0030] Using the training set described above, the first target model is tested to obtain test results, which characterize the performance of the first target model.

[0031] According to embodiments of this disclosure, the above-mentioned testing of the first target model using the training set to obtain test results includes:

[0032] The n sets of data in the training set are divided into K datasets, where K is an integer greater than or equal to 2.

[0033] Using each of the aforementioned K datasets sequentially as the test set, and the remaining K-1 datasets as the training set, the first target model is tested to obtain K test sub-results; and

[0034] Based on the above K test sub-results, the above test results are obtained.

[0035] According to embodiments of this disclosure, the first target model is tested by sequentially using each of the K datasets as a test set and the remaining K-1 datasets as a training set, resulting in K test sub-results, including:

[0036] Using each of the aforementioned K datasets as the test set, the remaining K-1 datasets are used as the training set to train the first target model, yielding training parameters; and

[0037] The second target model was tested using the aforementioned test set, and the aforementioned test results were obtained. The aforementioned second target model was determined based on the aforementioned first target model and the aforementioned training parameters.

[0038] A second aspect of this disclosure provides a method for online monitoring of the junction temperature of a semiconductor device, comprising:

[0039] The on-state voltage drop and reverse voltage drop of the semiconductor device are input into the first target model, and the junction temperature of the semiconductor device is output to adjust the semiconductor device according to the junction temperature.

[0040] A third aspect of this disclosure provides a model training apparatus, comprising: an acquisition module and a training module. The acquisition module is used to acquire a training set corresponding to a target semiconductor, wherein the training set includes n sets of data, where n is a positive integer, and each set of data includes the junction temperature, the on-state voltage drop, the reverse voltage drop, and the load current corresponding to the junction temperature. The training module is used to train an initial model based on the training set to obtain a first target model for eliminating the influence of load current on junction temperature monitoring, wherein the initial model represents a functional model of the relationship between junction temperature and on-state voltage drop, reverse voltage drop, and load current, and the first target model represents a functional model of the relationship between junction temperature and on-state voltage drop and reverse voltage drop.

[0041] A fourth aspect of this disclosure provides an online junction temperature monitoring device for a semiconductor device, comprising: an output module. The output module is configured to input the on-state voltage drop and reverse voltage drop of the semiconductor device into a first target model, and output the junction temperature of the semiconductor device to adjust the semiconductor device according to the junction temperature.

[0042] A fifth aspect of this disclosure provides an electronic device comprising: one or more processors; and a memory for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors perform the methods described above.

[0043] A sixth aspect of this disclosure also provides a computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, cause the processor to perform the methods described above.

[0044] The seventh aspect of this disclosure also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0045] According to the model training method, semiconductor device junction temperature online monitoring method, device, equipment, medium and program products provided in this disclosure, an initial model characterizing the relationship between junction temperature and on-state voltage drop, reverse voltage drop and load current is trained using the acquired training set corresponding to the target semiconductor device. A first target model characterizing the relationship between junction temperature and on-state voltage drop and reverse voltage drop is obtained. The first target model is used to monitor the junction temperature of the target semiconductor. Since the reverse voltage drop has high sensitivity over the entire load current range, and the on-state voltage drop has extremely high sensitivity when the current is large, the reverse voltage drop can compensate for the sensitivity of the on-state voltage drop in the small current range. Furthermore, the effect of the load current can be eliminated, so that the junction temperature monitoring is not affected by the operating conditions. At the same time, the accuracy of junction temperature monitoring can be improved over the entire current range. Attached Figure Description

[0046] The foregoing contents, as well as other objects, features, and advantages of this disclosure, will become clearer from the following description of embodiments with reference to the accompanying drawings, in which:

[0047] Figure 1 The illustrations schematically depict application scenarios of model training methods, online junction temperature monitoring methods for semiconductor devices, apparatus, devices, media, and program products according to embodiments of the present disclosure.

[0048] Figure 2 A typical on-state voltage drop measurement circuit in a prior art solution according to an embodiment of this disclosure is illustrated schematically;

[0049] Figure 3 This illustration schematically depicts a junction temperature monitoring strategy based on an embodiment of the present disclosure.

[0050] Figure 4 A flowchart illustrating a model training method according to an embodiment of the present disclosure is shown schematically.

[0051] Figure 5A flowchart illustrating the process of obtaining a set of data in a training set according to an embodiment of the present disclosure is shown schematically.

[0052] Figure 6 A schematic diagram of a dual-pulse test platform according to an embodiment of the present disclosure is shown.

[0053] Figure 7 The flowchart illustrating the process of obtaining a first target model according to an embodiment of the present disclosure is shown in the illustration.

[0054] Figure 8 A flowchart illustrating an online junction temperature monitoring method for a semiconductor device according to an embodiment of the present disclosure is shown schematically.

[0055] Figure 9 A schematic diagram of a circuit for measuring on-state voltage drop and reverse voltage drop according to an embodiment of the present disclosure is shown.

[0056] Figure 10 A schematic diagram illustrating a junction temperature monitoring strategy according to an embodiment of the present disclosure is shown.

[0057] Figure 11A The on-state voltage drop characteristics in the training set according to an embodiment of the present disclosure are illustrated schematically;

[0058] Figure 11B The reverse voltage drop characteristics in the training set according to an embodiment of the present disclosure are illustrated schematically;

[0059] Figure 12A The illustration shows the verification results of the GPR model according to an embodiment of the present disclosure;

[0060] Figure 12B The illustration schematically shows the verification results of the GRP model incorporating load current characteristics according to an embodiment of the present disclosure;

[0061] Figure 13 The diagram illustrates online measurement results of the on-state voltage drop and reverse voltage drop according to embodiments of the present disclosure;

[0062] Figure 14 This illustration schematically shows the error in online junction temperature monitoring according to an embodiment of the present disclosure;

[0063] Figure 15 A schematic block diagram of a model training apparatus according to an embodiment of the present disclosure is shown.

[0064] Figure 16 A schematic diagram illustrates a structural block diagram of an online junction temperature monitoring device for a semiconductor device according to an embodiment of the present disclosure; and

[0065] Figure 17A block diagram schematically illustrates an electronic device suitable for implementing a model training method and an online junction temperature monitoring method for semiconductor devices according to embodiments of the present disclosure. Detailed Implementation

[0066] The embodiments of the present disclosure will now be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the disclosure. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of the present disclosure for ease of explanation. However, it will be apparent that one or more embodiments may be practiced without these specific details. Furthermore, descriptions of well-known structures and techniques are omitted in the following description to avoid unnecessarily obscuring the concepts of the present disclosure.

[0067] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit this disclosure. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0068] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0069] When using expressions such as "at least one of A, B, and C", they should generally be interpreted in accordance with the meaning that is commonly understood by a person skilled in the art (e.g., "a system having at least one of A, B, and C" should include, but is not limited to, a system having A alone, a system having B alone, a system having C alone, a system having A and B, a system having A and C, a system having B and C, and / or a system having A, B, and C, etc.).

[0070] In the technical solution of this invention, the user information (including but not limited to user personal information, user image information, user device information, such as location information) and data (including but not limited to data used for analysis, stored data, and displayed data) involved are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of related data all comply with the relevant laws, regulations, and standards of the relevant countries and regions, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entry points for users to choose to authorize or refuse.

[0071] In implementing this disclosure, it was discovered that junction temperature typically refers to the operating temperature of a semiconductor chip. Due to chip packaging limitations, temperature measurement devices cannot directly contact the chip body. Therefore, thermistor electrical parameter methods are often used to indirectly measure junction temperature. The basic principle is to characterize the junction temperature using electrical parameters related to it. This method of measuring junction temperature offers fast response, low invasiveness, high accuracy, and is suitable for online applications.

[0072] Based on the state of the device during measurement, thermistor parameters can be divided into steady-state parameters and transient parameters. The most typical steady-state parameter is the on-state voltage drop, which is the voltage difference between the drain and source when the SiC MOSFET is in the on-state. Transient parameters are diverse and can include switching time, gate threshold voltage, gate on-current, etc. Although transient thermistor parameters have higher theoretical accuracy, they require extremely high bandwidth and sampling rate from the measurement equipment, resulting in significant additional costs in practical applications. Therefore, on-state voltage drop remains the most commonly used thermistor parameter.

[0073] However, junction temperature monitoring methods based on on-state voltage drop have two major problems: the on-state voltage drop is not only related to the junction temperature but also proportional to the load current. Therefore, the junction temperature monitoring results are greatly affected by the operating conditions. When the load current is small, the sensitivity of the on-state voltage drop to the junction temperature change is low, which directly limits the accuracy of junction temperature monitoring.

[0074] In practice, almost all thermistor parameter methods are affected by operating conditions, especially load current and bus voltage. However, it is impossible to cover all operating conditions when determining the relationship between thermistor parameters and junction temperature, leading to monitoring errors. Furthermore, existing thermistor parameter methods require specialized measuring equipment to monitor operating conditions, increasing the cost of the monitoring scheme. Therefore, improving the accuracy of junction temperature monitoring for semiconductor devices is a technical problem that needs to be solved in related technologies.

[0075] To this end, embodiments of this disclosure provide a model training method, comprising: acquiring a training set corresponding to a target semiconductor device, wherein the training set includes n sets of data, where n is a positive integer, and each set of data includes junction temperature, on-state voltage drop, reverse voltage drop, and load current corresponding to the junction temperature; training an initial model based on the training set to obtain a first target model for eliminating the influence of load current on junction temperature monitoring, wherein the initial model represents a functional model of the relationship between junction temperature and on-state voltage drop, reverse voltage drop, and load current, and the first target model represents a functional model of the relationship between junction temperature and on-state voltage drop and reverse voltage drop.

[0076] Figure 1 The illustration shows an application scenario of the model training method and the online junction temperature monitoring method for semiconductor devices according to embodiments of the present disclosure.

[0077] like Figure 1As shown, application scenario 100 according to this embodiment may include a first terminal device 101, a second terminal device 102, a third terminal device 103, a network 104, and a server 105. The network 104 serves as a medium for providing a communication link between the first terminal device 101, the second terminal device 102, the third terminal device 103, and the server 105. The network 104 may include various connection types, such as wired or wireless communication links, or fiber optic cables, etc.

[0078] Users can interact with server 105 via network 104 using at least one of the first terminal device 101, second terminal device 102, and third terminal device 103 to receive or send messages, etc. Various communication client applications can be installed on the first terminal device 101, second terminal device 102, and third terminal device 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social media platform software, etc. (for example only).

[0079] The first terminal device 101, the second terminal device 102, and the third terminal device 103 can be various electronic devices with displays and support web browsing, including but not limited to smartphones, tablets, laptops, and desktop computers.

[0080] Server 105 can be a server that provides various services, such as a backend management server that supports websites browsed by users using the first terminal device 101, the second terminal device 102, and the third terminal device 103 (this is just an example). The backend management server can analyze and process data such as received user requests, and feed back the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal devices.

[0081] For example, a training set corresponding to the target semiconductor device can be obtained through server 105. The training set includes n sets of data, where n is a positive integer. Each set of data includes the junction temperature and the corresponding on-state voltage drop, reverse voltage drop, and load current. Based on the training set, the initial model is trained to obtain a first target model for eliminating the influence of load current on junction temperature monitoring. The initial model represents the functional model of the relationship between junction temperature and on-state voltage drop, reverse voltage drop, and load current, while the first target model represents the functional model of the relationship between junction temperature and on-state voltage drop and reverse voltage drop.

[0082] It should be noted that the model training method and the online junction temperature monitoring method for semiconductor devices provided in this disclosure can generally be executed by server 105. Correspondingly, the model training device and the online junction temperature monitoring device for semiconductor devices provided in this disclosure can generally be located in server 105. The model training method and the online junction temperature monitoring method for semiconductor devices provided in this disclosure can also be executed by a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105. Correspondingly, the model training device and the online junction temperature monitoring device for semiconductor devices provided in this disclosure can also be located in a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105.

[0083] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0084] The following will be based on Figure 1 The described scene, through Figures 2 to 14 The model training method and the online junction temperature monitoring method of semiconductor devices according to the disclosed embodiments are described in detail.

[0085] Figure 2 The illustration schematically shows a typical on-state voltage drop measurement circuit in a prior art solution according to an embodiment of the present disclosure.

[0086] According to embodiments of this disclosure, existing methods for online monitoring of SiC MOSFET junction temperature based on steady-state characteristics are quite similar, typically using on-state voltage drop as the junction temperature monitoring indicator. Since the on-state voltage drop amplitude is small and measurement is difficult, a specially designed circuit for measurement is required. The differences between existing methods mostly lie in the measurement circuit.

[0087] like Figure 2 As shown, a typical on-state voltage drop measurement circuit is demonstrated. By controlling the conduction and cutoff of the diode, the bus voltage can be shielded, thereby realizing the online measurement of the on-state voltage drop.

[0088] Figure 3 The illustration schematically depicts a junction temperature monitoring strategy of a prior art solution according to an embodiment of this disclosure.

[0089] like Figure 3 As shown, V ON It can represent the on-state voltage drop, with the unit being V (volts); i DS It can represent the load current, with the unit being A (ampere); θjunction It can represent the estimated junction temperature, in °C.

[0090] According to embodiments of this disclosure, the basic strategy of the SiC MOSFET junction temperature online monitoring method based on steady-state characteristics is consistent, such as... Figure 3 As shown. First, offline calibration is required to obtain the relationship between on-state voltage drop and junction temperature under various load currents, and save it as follows: Figure 3 The form shown is used. When the semiconductor device is running, its on-state voltage drop and load current are measured online, and the junction temperature can then be calculated based on the calibration results.

[0091] Based on the above, the accuracy and results of junction temperature monitoring are closely related to the load current. However, the load current itself does not contain any information related to junction temperature; therefore, measuring the load current wastes computational resources and increases the cost of measurement equipment and monitoring. Furthermore, calibration cannot cover all operating conditions of the device. When the device is in an uncalibrated condition, significant monitoring errors will occur. Moreover, the sensitivity of the on-state voltage drop to junction temperature determines the theoretical accuracy of junction temperature monitoring. Clearly, as... Figure 3 As shown, when the load current is small, the on-state voltage drop is less sensitive to changes in junction temperature, which can lead to a large monitoring error.

[0092] Figure 4 A flowchart illustrating a model training method according to an embodiment of the present disclosure is shown schematically.

[0093] like Figure 4 As shown, the method 400 includes operations S410 and S420.

[0094] In operation S410, a training set corresponding to the target semiconductor device is obtained.

[0095] According to embodiments of this disclosure, the training set may include n sets of data, where n is a positive integer, and each set of data may include junction temperature, on-state voltage drop, reverse voltage drop and load current corresponding to junction temperature.

[0096] According to embodiments of this disclosure, the target semiconductor device may include a SiC MOSFET. The on-state voltage drop characterizes the voltage drop between the drain and source of the SiC MOSFET in the on-state; the reverse voltage drop characterizes the voltage drop generated when the body diode of the SiC MOSFET is reverse-biased.

[0097] According to embodiments of this disclosure, a target semiconductor can be calibrated offline using a double-pulse test to obtain the relationship between the junction temperature and the on-state voltage drop, reverse voltage drop, and load current of the target semiconductor. The calibration characterizes the relationship between the selected thermistor parameters determined through offline experiments and the junction temperature; the thermistor parameters may include the on-state voltage drop and the reverse voltage drop.

[0098] According to embodiments of this disclosure, each set of data can be denoted as [V] ds V sd i L ,T j ], where V ds It can characterize the on-state pressure drop, V sd It can characterize the reverse voltage drop, i L It can characterize the load current, T j It can characterize the junction temperature.

[0099] The training set requires at least 600 sets of data, i.e., n is greater than or equal to 600.

[0100] In operation of S420, the initial model is trained based on the training set to obtain the first target model for eliminating the influence of load current on junction temperature monitoring.

[0101] According to embodiments of this disclosure, the initial model can be a functional model characterizing the relationship between junction temperature and on-state voltage drop, reverse voltage drop and load current, and the first target model can be a functional model characterizing the relationship between junction temperature and on-state voltage drop and reverse voltage drop.

[0102] According to embodiments of this disclosure, an initial model can be trained based on the above-mentioned training set using Gaussian process regression to obtain a first target model for eliminating the influence of load current on junction temperature monitoring.

[0103] According to embodiments of this disclosure, a new steady-state parameter, namely the reverse voltage drop of the body diode of the target semiconductor, is introduced to form a new combination of junction temperature monitoring characteristics. Both the reverse voltage drop and the on-state voltage drop are only related to the load current and the junction temperature. Therefore, by using both parameters simultaneously to estimate the junction temperature, the effect of the load current can be eliminated, making the junction temperature monitoring unaffected by operating conditions.

[0104] According to embodiments of this disclosure, an initial model characterizing the relationship between junction temperature and on-state voltage drop, reverse voltage drop, and load current is trained using the acquired training set corresponding to the target semiconductor device. This yields a first target model characterizing the relationship between junction temperature and on-state voltage drop and reverse voltage drop. The first target model is then used to monitor the junction temperature of the target semiconductor. Since the reverse voltage drop is characterized by high sensitivity across the entire load current range, and the on-state voltage drop is characterized by extremely high sensitivity at high currents, the reverse voltage drop can compensate for the sensitivity of the on-state voltage drop within a small current range. Furthermore, the effect of the load current can be eliminated, thereby making junction temperature monitoring unaffected by operating conditions. Simultaneously, the accuracy of junction temperature monitoring can be improved across the entire current range.

[0105] According to embodiments of this disclosure, obtaining a training set corresponding to a target semiconductor device includes: measuring a set of data in the training set during each of n double pulses, wherein the n double pulses are sent to the target semiconductor device.

[0106] According to embodiments of this disclosure, each set of data in the training set can be obtained through double-pulse testing, wherein one set of data in the training set can be measured during each double-pulse period. That is, to obtain a training set including n sets of data, n double pulses need to be applied sequentially to the target semiconductor device.

[0107] According to embodiments of this disclosure, applying n double pulses to a target semiconductor device can obtain a training set including n sets of data, which can then be used to train a model.

[0108] Figure 5 A flowchart illustrating the process of obtaining a set of data in a training set according to an embodiment of the present disclosure is shown.

[0109] like Figure 5 As shown, the method 500 includes operations S510 to S540.

[0110] In operation of S510, for each double pulse, during the first pulse, the load inductor on the heating plate is charged with DC current so that the load current of the load inductor can be measured after the first pulse ends.

[0111] According to embodiments of this disclosure, the load current of the load inductor can characterize the load current of the target semiconductor device. The dual pulse may include a first pulse and a second pulse.

[0112] Figure 6 A schematic diagram of a dual-pulse test platform according to an embodiment of the present disclosure is shown.

[0113] like Figure 6 As shown, the dual-pulse test platform 600 includes an auxiliary power supply 601, a DC power supply 602, a control board 603, a heating plate 604, a load inductor 605, a sensor 606, and an oscilloscope 607.

[0114] According to embodiments of this disclosure, the dual-pulse circuit can be provided by Rohm. The dual-pulse circuit can be disposed on the heating plate 604, and the target semiconductor device can characterize M on the heating plate. L Device.

[0115] According to the embodiments of this disclosure, the premise for performing a double-pulse test on the target semiconductor device is to form a circuit with electrical functions. That is, a single lower transistor cannot generate a double pulse, so another semiconductor device needs to be set as the upper transistor of the bridge arm, and the target semiconductor device is used as the lower transistor of the bridge arm.

[0116] Among them, the heating plate includes M L Devices and M U Device, M L The device is the lower tube of the bridge arm, M. U The device is the upper tube of the bridge arm; that is, the device whose junction temperature needs to be monitored is placed on the lower tube of the bridge arm.

[0117] According to embodiments of this disclosure, M can be secured by screws. L Devices and M U The device is fixed on the heating plate 604, with thermally conductive silicone and an insulating sheet between the device and the heating plate 604. The control board 603 has a digital signal processor (DSP), which can send a set of double pulse (DP) signals to the target semiconductor device on the lower transistor, while the semiconductor device on the upper transistor is always in the off state. The auxiliary power supply 601 can provide power to the control board 603.

[0118] According to an embodiment of this disclosure, for each double pulse, during the first pulse, the DC power supply 602 can charge the load inductor 605, causing the load current flowing through the load inductor 605 to increase linearly. After the first pulse ends, the load current of the load inductor can be measured by the sensor 606 and displayed on the oscilloscope 607, thereby obtaining the load current of the target semiconductor device.

[0119] During operation S520, between the first and second pulses, the reverse voltage drop of the target semiconductor device is measured based on the freewheeling current of the load current in the semiconductor device located on the upper bridge arm.

[0120] According to an embodiment of this disclosure, the target semiconductor device is located on the lower bridge arm, and both the semiconductor device and the target semiconductor device are located on the heating plate.

[0121] For example, such as Figure 6 As shown, the target semiconductor device can characterize M on the heating plate 604. L Devices, semiconductor devices can characterize the M on the heating plate 604 U Device.

[0122] According to embodiments of this disclosure, between the first pulse and the second pulse, the load current in the upper transistor (semiconductor device M) U The body diode of the device can be used for freewheeling. At this time, the reverse voltage drop of the target semiconductor device can be measured by the voltage probe of sensor 606 and read in oscilloscope 607.

[0123] During operation S530, during the second pulse, the on-state voltage drop of the target semiconductor device is measured based on the commutation of the load current from the semiconductor device to the target semiconductor device.

[0124] According to embodiments of this disclosure, during the second pulse, the load current flows from the upper transistor (semiconductor device M)... U The voltage is commutated from the body diode of the lower transistor (target semiconductor device). At this time, the on-state voltage drop of the target semiconductor device of the lower transistor can be measured by the voltage probe of the sensor 606 and read in the oscilloscope 607.

[0125] During operation of S540, the junction temperature of the target semiconductor device is measured after the second pulse ends.

[0126] According to embodiments of this disclosure, the junction temperature of the target semiconductor device can characterize the temperature of the heating plate 604; that is, the junction temperature of the target semiconductor device can be regarded as the temperature of the heating plate 604. Therefore, the temperature of the heating plate 604 can be measured after the second pulse ends to obtain the junction temperature of the target semiconductor device.

[0127] According to embodiments of this disclosure, the relationship between the junction temperature and on-state voltage drop, reverse voltage drop and load current of the target semiconductor device can be obtained through double-pulse testing. Thus, a training set for training an initial model can be constructed based on the junction temperature, reverse voltage drop, on-state voltage drop and load current obtained under each double pulse.

[0128] According to embodiments of this disclosure, the magnitude of the load current of the target semiconductor device is controlled by adjusting the width of the first pulse in the double pulse to obtain the on-state voltage drop, reverse voltage drop, and junction temperature under different load currents.

[0129] For example, under a given load current, five double pulses can be applied to the target semiconductor device to obtain five sets of data under that load current. Then, by adjusting the width of the first pulse in the double pulses to obtain a different load current, five more double pulses can be applied to the target semiconductor device to obtain five sets of data under that load current.

[0130] According to embodiments of this disclosure, the magnitude of the load current can be controlled by adjusting the width of the first pulse in the double pulse, thereby obtaining the on-state voltage drop, reverse voltage drop, and junction temperature under different load currents. This enriches the data in the training set and improves the accuracy of the first target model trained using the training set for junction temperature monitoring.

[0131] Figure 7 The flowchart illustrating the process of obtaining a first target model according to an embodiment of the present disclosure is shown in the illustration.

[0132] like Figure 7 As shown, the method 700 includes operations S710 to S750.

[0133] When operating S710, an initial model is constructed based on implicit functions corresponding to on-state voltage drop, reverse voltage drop, and load current.

[0134] According to embodiments of this disclosure, the implicit function follows a Gaussian distribution, and the initial model follows a Gaussian distribution.

[0135] According to embodiments of this disclosure, Gaussian process regression can be used to train an initial model based on the aforementioned training set. A Gaussian process is essentially the joint distribution of an infinite number of random variables in a continuous domain. The joint distribution of any number of random variables that follow a certain Gaussian process still follows a Gaussian distribution.

[0136] According to embodiments of this disclosure, the junction temperature, as a continuous variable, obviously also follows a Gaussian process. The training set obtained by offline calibration using a double-pulse test can be expressed as follows (1).

[0137]

[0138] Each vector x can be represented as [V ds V sd i L ], each can represent the junction temperature corresponding to vector x, and n can represent that the training set includes n sets of data, where n is greater than or equal to 600.

[0139] According to embodiments of this disclosure, since the junction temperature of the target semiconductor device is related to the on-state voltage drop, reverse voltage drop, and load current, it is desirable to obtain a form through training. The model.

[0140] According to the embodiments of this disclosure, the relationship between junction temperature and on-state voltage drop, reverse voltage drop and load current can be assumed. The initial model characterizing the relationship between junction temperature and on-state voltage drop, reverse voltage drop and load current can be expressed as the following formula (2).

[0141] y i =f(x) i )+ε i (2)

[0142] Where is an integer greater than or equal to 1 and less than or equal to n, y i x can represent the junction temperature in the i-th group of data in the training set. i ε can represent the on-state voltage drop, reverse voltage drop, and load current in the i-th data set of the training set. i This can represent noise, which follows a Gaussian distribution N(0, 2 ), f(x) i ) can characterize implicit functions.

[0143] According to embodiments of this disclosure, since the process of obtaining the training set through testing is usually noisy, it is necessary to introduce noise ε. i This is to ensure the accuracy of model construction.

[0144] According to embodiments of this disclosure, since the junction temperature follows a Gaussian distribution and the noise follows a Gaussian distribution, the resulting initial model also follows a Gaussian distribution.

[0145] In operation S720, the mean function and covariance matrix of the initial model are determined.

[0146] According to embodiments of this disclosure, the mean function of the Gaussian process is typically set to 0; the covariance matrix K can be determined by the kernel function k(x, x).

[0147] In operation S730, the joint distribution corresponding to the training set and the prediction set is obtained based on the Gaussian process corresponding to the training set and the Gaussian process corresponding to the prediction set.

[0148] According to embodiments of this disclosure, the Gaussian process can be determined by the mean function and the covariance matrix. The prediction set can include m sets of data, where m is a positive integer. In each set of data, the on-state voltage drop, reverse voltage drop, and load current are known, while the junction temperature is unknown.

[0149] According to an embodiment of this disclosure, based on the mean function and covariance matrix of the obtained initial model, a Gaussian process corresponding to the training set is determined: f(X)~GP(0,K(X,X).

[0150] According to embodiments of this disclosure, for a prediction set with an unknown junction temperature, the on-state voltage drop, reverse voltage drop, and load current in each data set can be expressed as X. * That is, the corresponding model for estimating the junction temperature can be expressed as: For the prediction set, the above operation S720 can be performed to obtain the mean function and covariance matrix corresponding to the prediction set, thereby determining the Gaussian process corresponding to the prediction set: f(X) * )~GP(0,K(X * ,X * )).

[0151] According to embodiments of this disclosure, it can be assumed that the prediction set includes m sets of data. Since f(X) follows a Gaussian distribution, for n variables of a Gaussian process, adding m new variables will still result in an m+n dimensional Gaussian distribution.

[0152] According to an embodiment of this disclosure, the m+n dimensional Gaussian distribution can be expressed as the following formula (3).

[0153]

[0154] According to an embodiment of this disclosure, the joint distribution corresponding to the training set and the prediction set obtained by simplifying the above formula (3) can be expressed as the following formula (4).

[0155]

[0156] In operation S740, the conditional distribution corresponding to the prediction set is obtained based on the joint distribution and the prior distribution.

[0157] According to embodiments of this disclosure, the prior distribution can characterize the Gaussian process corresponding to the training set. Training a model based on existing calibration data (training set) to evaluate the junction temperature corresponding to new data is equivalent to finding the conditional probability distribution under the existing prior data. According to Bayes' theorem, the conditional probability distribution is equal to the joint probability distribution divided by the marginal probability distribution (prior distribution). Since both the joint distribution and the prior distribution follow a Gaussian distribution, the conditional distribution corresponding to the prediction set also follows a Gaussian distribution.

[0158] According to embodiments of this disclosure, based on the above content, the conditional probability distribution corresponding to the prediction set can be expressed as the following formula (5).

[0159] p(Y * |Y)=N(μ * ,∑ * (5)

[0160] Where, μ * This can be represented by the mean function of the conditional probability distribution, ∑ * It can represent the covariance matrix of the conditional probability distribution.

[0161] According to embodiments of this disclosure, the mean function μ of the conditional probability distribution is obtained based on the above formula (4) and the Gaussian process corresponding to the training set: f(X)~GP(0,K(X,X)). * It can be expressed as the following formula (6).

[0162] μ * =k * (k+σ 2 I) -1 Y (6)

[0163] According to embodiments of this disclosure, the covariance matrix ∑ of the conditional probability distribution is obtained based on the above formula (4) and the Gaussian process corresponding to the training set: f(X)~GP(0,K(X,X)). * It can be expressed as the following formula (7).

[0164]

[0165] When operating the S750, a first target model is determined based on the conditional distribution. Using the first target model, the junction temperature corresponding to each set of data in the prediction set is obtained based on the on-state pressure drop and reverse pressure drop in the prediction set.

[0166] According to embodiments of this disclosure, the junction temperature is typically estimated using a Gaussian distribution mean function, meaning the mean obtained through the mean function can be used to represent the junction temperature.

[0167] According to an embodiment of this disclosure, the first target model determined according to the above formula (6) can be expressed as the following formula (8).

[0168] Y * =K * (K+σ 2 I) -1 Y (8)

[0169] According to an embodiment of this disclosure, based on the above formula (8), the corresponding junction temperature can be estimated from the on-state voltage drop and reverse voltage drop in the prediction set.

[0170] According to embodiments of this disclosure, based on the above formula (8), the form can be obtained as follows: The first target model.

[0171] According to embodiments of this disclosure, a Gaussian process regression is introduced to train an initial model to obtain a first target model using on-state voltage drop and reverse voltage drop as feature variables. The effect of load current is eliminated, and at the same time, in the small current range, the reverse voltage drop can compensate for the sensitivity of the on-state voltage drop, thereby improving the junction temperature monitoring accuracy across the entire current range.

[0172] According to embodiments of this disclosure, determining the mean function and covariance matrix of an initial model includes: using a kernel function to determine an element in the covariance matrix based on any two sets of data in the training set.

[0173] According to embodiments of this disclosure, the mean function is typically set to zero, and the covariance matrix can represent an n*n matrix.

[0174] According to embodiments of this disclosure, the kernel function used to determine the covariance matrix can characterize a quadratic rational function, which can be expressed as the following formula (9).

[0175]

[0176] Where, x i and x j They can represent the i-th and j-th data sets in the training set, respectively, and α and l can both represent unknown hyperparameters.

[0177] According to the embodiments of this disclosure, the on-state voltage drop, reverse voltage drop, and load current of any two sets of data in the training set can be substituted into the above formula (9) to obtain an element of the covariance matrix corresponding to the implicit function. Since there are n sets of data in the training set, the covariance value of any two sets of data can be obtained through the above formula (9), that is, the obtained covariance matrix is ​​an n*n matrix.

[0178] According to an embodiment of this disclosure, the covariance matrix is ​​determined using the kernel function shown in formula (9) above to ensure the accuracy of the first target model obtained through subsequent training.

[0179] According to an embodiment of this disclosure, a first target model is determined based on a conditional distribution. Using the first target model, the junction temperature corresponding to each set of data in the prediction set is obtained based on the on-state voltage drop and reverse voltage drop in the prediction set. This includes: determining a first initial target model based on the conditional distribution; estimating the unknown hyperparameters in the first initial target model to obtain hyperparameter values ​​corresponding to each unknown hyperparameter; and obtaining the first target model based on the first initial target model and each hyperparameter value.

[0180] According to embodiments of this disclosure, in the above formula (8), there are three unknown hyperparameters α, α, and α. The first initial target model can characterize the function model with unknown hyperparameters in the above formula (8). These three unknown parameters can be denoted as α. Maximum likelihood estimation is performed on α to obtain the trained first target model. The first target model can characterize the function model with known hyperparameters in the above formula (8).

[0181] According to an embodiment of this disclosure, the three unknown hyperparameters in the above formula (8) are solved by maximum likelihood estimation to obtain a first target model for characterizing the relationship between junction temperature and on-state voltage drop and reverse voltage drop.

[0182] According to embodiments of this disclosure, the above-described model training method further includes: using a training set to test the first target model and obtain test results.

[0183] According to embodiments of this disclosure, the test results can be used to characterize the performance of the first target model. The accuracy of the trained first target model can be verified using K-fold cross-validation.

[0184] According to embodiments of this disclosure, a first target model is tested using a training set to evaluate the performance of the first target model for junction temperature monitoring based on the obtained test results, thereby making the evaluation of the performance of the first target model more accurate.

[0185] According to an embodiment of this disclosure, the first target model is tested using a training set to obtain test results, including: dividing the n sets of data in the training set into K datasets; using each of the K datasets as a test set and the remaining K-1 datasets as a training set to test the first target model and obtain K test sub-results; and obtaining the test results based on the K test sub-results.

[0186] According to embodiments of this disclosure, K is an integer greater than or equal to 2. The accuracy of the first target model trained can be verified using K-fold cross-validation. The test sub-results can characterize the accuracy of the model trained using the remaining datasets as training sets, with one dataset as the test set, for junction temperature estimation. The test result can characterize the average of the K test sub-results.

[0187] According to an embodiment of this disclosure, when K is 10, the n sets of data in the training set can be divided into 10 datasets. One of these 10 datasets is used as the test set, and the remaining 9 datasets are used as the training set to test the first target model. This yields 10 test sub-results, which are equivalent to 10 accuracy rates.

[0188] According to embodiments of this disclosure, the average of the 10 obtained accuracies can be used to obtain test results, which can be used to evaluate the performance of the first target model for junction temperature monitoring based on the average accuracy.

[0189] According to embodiments of this disclosure, if the accuracy of the test results representing the first target model is lower than a set threshold, a more accurate first target model can be obtained by optimizing the hyperparameters in the first target model.

[0190] According to an embodiment of this disclosure, if the accuracy of the test results characterizing the first target model is higher than a set threshold, the accuracy of the first target model has reached the standard, and the first target model can be used for junction temperature monitoring of the target semiconductor device.

[0191] According to embodiments of this disclosure, the performance of a first target model in monitoring the junction temperature of a target semiconductor device can be evaluated by testing the first target model.

[0192] According to an embodiment of this disclosure, each of the K datasets is used sequentially as a test set, and the remaining K-1 datasets are used as a training set to test the first target model and obtain K test sub-results. This includes: using each of the K datasets as a test set, training the first target model using the remaining K-1 datasets as a training set to obtain training parameters; and using the test set to test the second target model to obtain test sub-results.

[0193] According to embodiments of this disclosure, the second target model is determined based on the first target model and training parameters.

[0194] According to an embodiment of this disclosure, when a certain dataset is used as the test set and the remaining K-1 datasets are used as the training set, the training parameters can be obtained by training the first target model using the remaining K-1 datasets as the training set.

[0195] According to embodiments of this disclosure, a second target model can be obtained based on training parameters and a first target model, and the second target model can be tested using a test set to obtain test sub-results.

[0196] The test sub-results may include the accuracy of the second target model for junction temperature monitoring when the dataset is used as the test set and the remaining K-1 datasets are used as the training set.

[0197] According to embodiments of this disclosure, training parameters can be obtained by using the remaining K-1 datasets as a training set, and a second target model can be obtained based on the training parameters. This allows the obtained second target model to be tested using a test set to evaluate the performance of the second target model for junction temperature monitoring.

[0198] Figure 8 A flowchart illustrating an online junction temperature monitoring method for a semiconductor device according to an embodiment of the present disclosure is shown.

[0199] like Figure 8 As shown, the method 800 includes operation S810.

[0200] In operation S810, the on-state voltage drop and reverse voltage drop of the semiconductor device are input into the first target model, and the junction temperature of the semiconductor device is output to adjust the semiconductor device according to the junction temperature.

[0201] Figure 9 A schematic diagram of a circuit for measuring on-state voltage drop and reverse voltage drop according to an embodiment of the present disclosure is shown.

[0202] like Figure 9 As shown, during the operation of the test converter, the on-state voltage drop and reverse voltage drop measurement circuits are connected in parallel with the device under test.

[0203] According to embodiments of this disclosure, the lower tube M L They can characterize the device under test (DUT). R1 ​​and R2 can characterize two 1000-ohm resistors, and D1 and D2 can characterize two diodes. M aS1 can represent an auxiliary MOSFET, and LOM can represent a small current source. LOM can represent a logic controller that outputs high or low levels based on the logic state of the bridge arm drive signals. This is true if and only if M... L When the LOM is conducting in either forward or reverse direction, it will output a low level to assist the M. a Turn off. At this time, S1 will turn on D1, so the measurement can be performed at point P. m The on-state and reverse voltage drops of the device can be directly measured. When the converter starts running, it is only necessary to measure the on-state and reverse voltage drops of the lower transistor before and after commutation, respectively, based on the existing model. Estimating junction temperature does not require load current measurement. On-state voltage drop and reverse voltage drop share a dedicated measurement circuit, without adding extra hardware cost.

[0204] Among them, G U M can represent the drive signal for the upper bridge arm device. U It can characterize the upper bridge arm device, G L It can represent the drive signal of the lower bridge arm device, D U It can characterize the upper bridge arm device M U The body diode, D L It can characterize the lower bridge arm device M L Body diode; D R It can represent the direction signal of the load current; flowing into the bridge arm can be represented as 1, and flowing out of the bridge arm can be represented as 0; G a It can characterize the auxiliary MOSFET M a The driving signal is also the output signal of the logic unit LOW.

[0205] According to embodiments of this disclosure, the control of the converter can be adjusted accordingly based on the obtained junction temperature monitoring results to improve the reliability of device operation and extend the service life of the device.

[0206] For example, adjustments made based on junction temperature monitoring results may include: if the junction temperature is too high, exceeding a certain threshold, the operating power can be reduced to achieve cooling; when the junction temperature fluctuates significantly, the lifespan is reduced, so the fluctuation of the junction temperature should be reduced, as the fluctuation of the junction temperature directly affects the lifespan of the device.

[0207] According to embodiments of this disclosure, a first target model obtained through training is used to obtain the corresponding junction temperature based on the on-state voltage drop and reverse voltage drop of the semiconductor device, so as to regulate the semiconductor device according to the junction temperature and increase the lifetime of the semiconductor device.

[0208] Figure 10 A schematic diagram of a junction temperature monitoring strategy according to an embodiment of the present disclosure is shown.

[0209] like Figure 10As shown, the junction temperature monitoring strategy 1000 includes a target semiconductor device 1001, a training set 1002, a first target model 1003, a controller 1004, and a measurement circuit 1005.

[0210] According to an embodiment of this disclosure, the above-described operation S410 is performed on the target semiconductor device 1001 using a double-pulse test, that is, the target semiconductor device 1001 is calibrated offline to obtain a training set 1002; based on the training set 1002, the initial model is trained using GPR (Gaussian Process Regression) to obtain the first target model 1003, and the above-described operation S420 can then be performed.

[0211] According to embodiments of this disclosure, the trained first target model 1003 can be stored in the controller 1004. The on-state voltage drop and reverse voltage drop of the target semiconductor device 1001 are measured using the measurement circuit 1005, and the measured on-state voltage drop and reverse voltage drop are input to the controller 1004 so as to perform junction temperature estimation using the first target model 1003 stored in the controller 1004.

[0212] The measurement circuit 1005 can be adopted as follows: Figure 9 The circuit shown is for measuring on-state voltage drop and reverse voltage drop.

[0213] According to embodiments of this disclosure, a junction temperature regulation target semiconductor device 1001 can be obtained based on a first target model 1004 stored in the controller 1004, so as to improve the service life of the target semiconductor 1001.

[0214] Figure 11A The schematic illustration shows the on-state voltage drop characteristics in the training set according to an embodiment of the present disclosure.

[0215] Figure 11B The diagram illustrates the reverse voltage drop characteristics in the training set according to an embodiment of the present disclosure.

[0216] According to embodiments of this disclosure, the on-state voltage drop feature and the reverse voltage drop feature can be combined to characterize the calibrated training set. For example, the training set contains 600 sets of data.

[0217] According to embodiments of this disclosure, by Figure 11A and Figure 11B It can be seen that the on-state voltage drop V ds Sensitivity varies with junction temperature T j and load current i L The pressure increases as it rises, while the reverse pressure drop V... sd The sensitivity of the two remains almost constant with junction temperature and load current. Therefore, they can support each other within different junction temperature and current ranges, thereby improving the overall junction temperature monitoring accuracy.

[0218] Figure 12A The illustration shows the verification results of the GPR model according to an embodiment of the present disclosure.

[0219] Figure 12B The illustration schematically shows the verification results of the GRP model with added load current characteristics according to an embodiment of the present disclosure.

[0220] According to embodiments of this disclosure, performing GPR regression on the above training set yields the hyperparameters θ = {0.9222, 0.0082, 0.4273}. K-fold cross-validation is then performed on this model, and the results are as follows: Figure 12A As shown, the characteristic selected at this time is the on-state voltage drop V. ds and reverse voltage drop V sd Add load current i L The validation result after feature post is as follows Figure 12B As shown. It can be seen that regardless of whether the load current i... L Characterized by this feature, the trained models all exhibit good accuracy. Therefore, the accuracy of this junction temperature monitoring method is independent of operating conditions such as load current. On the other hand, it can be clearly seen that as the load current increases (each set of data from left to right), the monitoring error does not change significantly.

[0221] Wherein, Response(Y) can represent the junction temperature T j Predict can represent the model's predicted data, True can represent the real data, Error can represent the training error, and Data number can represent the number of samples in the training set.

[0222] Figure 13 The diagram illustrates the online measurement results of the on-state voltage drop and reverse voltage drop according to an embodiment of the present disclosure.

[0223] According to embodiments of this disclosure, the following is employed: Figure 9 The measurement circuit shown can measure the reverse voltage drop and the on-state voltage drop before and after commutation, respectively. The measurement results of the reverse voltage drop and the on-state voltage drop are as follows: Figure 13 As shown.

[0224] Among them, V GL It can characterize the lower bridge arm drive voltage, V GU It can characterize the upper arm drive voltage, V DS It can characterize the drain-source voltage, V mea It can characterize the output voltage signal of the measuring circuit, and the unit is V (volt).

[0225] Figure 14 The illustration shows the error in online junction temperature monitoring according to an embodiment of the present disclosure.

[0226] Figure 14 The online monitoring error of this method is demonstrated, with the reference junction temperature obtained using an infrared thermal imager. Here, LR1 can represent a model obtained through linear regression characterized by load current and on-state voltage drop; LR2 can represent a model obtained through linear regression characterized by load current and reverse voltage drop; SVM (Support Vector Machine) can represent a model obtained through support vector machine regression characterized by on-state / reverse voltage drop; NN (Neural Network) can represent a model obtained through neural network regression characterized by on-state / reverse voltage drop; GPR can represent the method involved in this application. It is evident that the error of this junction temperature monitoring method is smaller than that of traditional methods.

[0227] Based on the above model training method, this disclosure also provides a model training apparatus. The following will combine... Figure 15 The device is described in detail.

[0228] Figure 15 A schematic block diagram of a model training apparatus according to an embodiment of the present disclosure is shown.

[0229] like Figure 15 As shown, the model training device 1500 of this embodiment includes an acquisition module 1510 and a training module 1520.

[0230] The acquisition module 1510 is used to acquire a training set corresponding to the target semiconductor. The training set includes n sets of data, where n is a positive integer. Each set of data includes the junction temperature, the on-state voltage drop, the reverse voltage drop, and the load current corresponding to the junction temperature. In one embodiment, the acquisition module 1510 can be used to perform the operation S410 described above, which will not be repeated here.

[0231] The training module 1520 is used to train the initial model based on the training set to obtain a first target model for eliminating the influence of load current on junction temperature monitoring. The initial model represents a functional model of the relationship between junction temperature and on-state voltage drop, reverse voltage drop, and load current, while the first target model represents a functional model of the relationship between junction temperature and on-state voltage drop and reverse voltage drop. In one embodiment, the training module 1520 can be used to perform the operation S420 described above, which will not be repeated here.

[0232] According to embodiments of this disclosure, the acquisition module 1510 includes a measurement unit.

[0233] A measurement unit is used to measure a set of data from a training set during each of n double pulses, wherein the n double pulses are sent to a target semiconductor device.

[0234] According to embodiments of this disclosure, the measurement unit includes a first measurement subunit, a second measurement subunit, a third measurement subunit, and a fourth measurement subunit.

[0235] The first measurement subunit is configured to charge the load inductor on the heating plate with DC current during the first pulse for each double pulse, so as to measure the load current of the load inductor after the end of the first pulse, wherein the load current of the load inductor characterizes the load current of the target semiconductor device.

[0236] The second measurement subunit is used to measure the reverse voltage drop of the target semiconductor device between the first pulse and the second pulse, based on the load current freewheeling in the semiconductor device located on the upper bridge arm, wherein the target semiconductor device is located on the lower bridge arm, and both the semiconductor device and the target semiconductor device are located on the heating plate.

[0237] The third measurement subunit is used to measure the on-state voltage drop of the target semiconductor device based on the commutation of the load current from the semiconductor device to the target semiconductor device during the second pulse.

[0238] The fourth measurement subunit is used to measure the junction temperature of the target semiconductor device after the second pulse ends, wherein the junction temperature of the target semiconductor device characterizes the temperature of the heating plate.

[0239] According to embodiments of this disclosure, the training module 1520 includes a construction unit, a first determination unit, a first acquisition unit, a second acquisition unit, and a second determination unit.

[0240] The building unit is used to construct the initial model based on the implicit functions corresponding to the on-state voltage drop, reverse voltage drop and load current, where the implicit functions follow a Gaussian distribution and the initial model follows a Gaussian distribution.

[0241] The first determining unit is used to determine the mean function and covariance matrix of the initial model.

[0242] The first acquisition unit is used to obtain the joint distribution corresponding to the training set and the prediction set based on the Gaussian process corresponding to the training set and the Gaussian process corresponding to the prediction set. The Gaussian process is determined by the mean function and the covariance matrix. The prediction set includes m sets of data, where m is a positive integer. In each set of data, the on-state voltage drop, reverse voltage drop and load current are known, and the junction temperature is unknown.

[0243] The second acquisition unit is used to obtain the conditional distribution corresponding to the prediction set based on the joint distribution and the prior distribution, wherein the prior distribution represents the Gaussian process corresponding to the training set.

[0244] The second determining unit is used to determine the first target model based on the conditional distribution, so as to use the first target model to obtain the junction temperature corresponding to each set of data in the prediction set based on the on-state pressure drop and reverse pressure drop in the prediction set.

[0245] According to embodiments of this disclosure, the first determining unit includes a first determining subunit.

[0246] The first determining subunit is used to determine an element in the covariance matrix based on any two sets of data in the training set using a kernel function. The mean function is set to zero, and the covariance matrix represents an n*n matrix.

[0247] According to embodiments of this disclosure, the second determining unit includes a second determining subunit, a first obtaining subunit, and a second obtaining subunit.

[0248] The second determining sub-unit is used to determine the first initial target model based on the conditional distribution.

[0249] The first obtaining sub-unit is used to estimate the unknown hyperparameters in the first initial target model and obtain the hyperparameter values ​​corresponding to each unknown hyperparameter.

[0250] The second obtaining sub-unit is used to obtain the first target model based on the first initial target model and each hyperparameter value.

[0251] According to embodiments of this disclosure, the model training apparatus 1500 further includes a testing module.

[0252] The testing module is used to test the first target model using the training set and obtain the test results, where the test results characterize the performance of the first target model.

[0253] According to embodiments of this disclosure, the testing module includes a partitioning unit, a testing unit, and a third acquisition unit.

[0254] The partitioning unit is used to divide the n groups of data in the training set into K datasets, where K is an integer greater than or equal to 2.

[0255] The test unit is used to test the first target model by taking each of the K datasets as the test set in turn and the remaining K-1 datasets as the training set, and obtaining K test sub-results.

[0256] The third obtaining unit is used to obtain the test result based on the K test sub-results.

[0257] According to embodiments of this disclosure, the test unit includes a training subunit and a test subunit.

[0258] The training subunit is used to train the first target model using the remaining K-1 datasets as the training set, with each of the K datasets serving as the test set, to obtain the training parameters.

[0259] The test subunit is used to test the second target model using the test set and obtain the test sub-result, wherein the second target model is determined based on the first target model and training parameters.

[0260] According to embodiments of this disclosure, any plurality of modules in the acquisition module 1510 and training module 1520 may be combined into one module, or any one of these modules may be split into multiple modules. Alternatively, at least a portion of the functionality of one or more of these modules may be combined with at least a portion of the functionality of other modules and implemented in one module. According to embodiments of this disclosure, at least one of the acquisition module 1510 and training module 1520 may be at least partially implemented as hardware circuitry, such as a field-programmable gate array (FPGA), a programmable logic array (PLA), a system-on-a-chip, a system-on-a-substrate, a system-on-package, an application-specific integrated circuit (ASIC), or any other reasonable means of integrating or packaging circuitry, or implemented in software, hardware, or firmware, or in any suitable combination of any of these three implementation methods. Alternatively, at least one of the acquisition module 1510 and training module 1520 may be at least partially implemented as a computer program module, which, when run, can perform corresponding functions.

[0261] Based on the above model training method, this disclosure also provides an online junction temperature monitoring device for semiconductor devices. The following will combine... Figure 16 The device is described in detail.

[0262] Figure 16 A schematic block diagram of a junction temperature online monitoring device for a semiconductor device according to an embodiment of the present disclosure is shown.

[0263] like Figure 16 As shown, the junction temperature online monitoring device 1600 for semiconductor devices in this embodiment includes an input module 1610.

[0264] The input module 1610 is used to input the on-state voltage drop and reverse voltage drop of the semiconductor device into the first target model, and output the junction temperature of the semiconductor device to adjust the semiconductor device according to the junction temperature. In one embodiment, the input module 1610 can be used to perform the operation S810 described above, which will not be repeated here.

[0265] According to embodiments of this disclosure, the input module 1610 can be at least partially implemented as hardware circuitry, such as a field-programmable gate array (FPGA), a programmable logic array (PLA), a system-on-a-chip, a system-on-a-substrate, a system-on-package, an application-specific integrated circuit (ASIC), or any other reasonable means of integrating or packaging the circuitry, or implemented in software, hardware, or firmware, or in any suitable combination of any of these three methods. Alternatively, the input module 1610 can be at least partially implemented as a computer program module, which, when run, can perform corresponding functions.

[0266] Figure 17 A block diagram schematically illustrates an electronic device suitable for implementing a model training method and an online junction temperature monitoring method for semiconductor devices according to embodiments of the present disclosure.

[0267] like Figure 17 As shown, an electronic device 1700 according to an embodiment of the present disclosure includes a processor 1701, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1702 or a program loaded from a storage portion 1708 into a random access memory (RAM) 1703. The processor 1701 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 1701 may also include onboard memory for caching purposes. The processor 1701 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present disclosure.

[0268] RAM 1703 stores various programs and data required for the operation of electronic device 1700. Processor 1701, ROM 1702, and RAM 1703 are interconnected via bus 1704. Processor 1701 performs various operations of the method flow according to embodiments of the present disclosure by executing programs in ROM 1702 and / or RAM 1703. It should be noted that the programs may also be stored in one or more memories other than ROM 1702 and RAM 1703. Processor 1701 may also perform various operations of the method flow according to embodiments of the present disclosure by executing programs stored in said one or more memories.

[0269] According to embodiments of this disclosure, the electronic device 1700 may further include an input / output (I / O) interface 1705, which is also connected to a bus 1704. The electronic device 1700 may also include one or more of the following components connected to the input / output (I / O) interface 1705: an input section 1706 including a keyboard, mouse, etc.; an output section 1707 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 1708 including a hard disk, etc.; and a communication section 1709 including a network interface card such as a LAN card, modem, etc. The communication section 1709 performs communication processing via a network such as the Internet. A drive 1710 is also connected to the input / output (I / O) interface 1705 as needed. A removable medium 1711, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 1710 as needed so that computer programs read from it can be installed into the storage section 1708 as needed.

[0270] This disclosure also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs that, when executed, implement the method according to the embodiments of this disclosure.

[0271] According to embodiments of this disclosure, the computer-readable storage medium may be a non-volatile computer-readable storage medium, such as including, but not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this disclosure, the computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, according to embodiments of this disclosure, the computer-readable storage medium may include ROM 1702 and / or RAM 1703 and / or one or more memories other than ROM 1702 and RAM 1703 described above.

[0272] Embodiments of this disclosure also include a computer program product comprising a computer program containing program code for performing the methods shown in the flowchart. When the computer program product is run on a computer system, the program code is used to cause the computer system to implement the methods provided in the embodiments of this disclosure.

[0273] When the computer program is executed by the processor 1701, it performs the functions defined in the system / apparatus of this disclosure embodiments. According to embodiments of this disclosure, the systems, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0274] In one embodiment, the computer program may rely on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may also be transmitted and distributed in the form of signals over a network medium, and may be downloaded and installed via the communication section 1709, and / or installed from a removable medium 1711. The program code contained in the computer program can be transmitted using any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination thereof.

[0275] In such an embodiment, the computer program can be downloaded and installed from a network via the communication section 1709, and / or installed from the removable medium 1711. When the computer program is executed by the processor 1701, it performs the functions defined in the system of this disclosure embodiment. According to embodiments of this disclosure, the systems, devices, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0276] According to embodiments of this disclosure, program code for executing the computer programs provided in embodiments of this disclosure can be written in any combination of one or more programming languages. Specifically, these computational programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, languages ​​such as Java, C++, Python, "C", or similar programming languages. The program code can execute entirely on the user's computing device, partially on the user's device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0277] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0278] Those skilled in the art will understand that the features described in the various embodiments and / or claims of this disclosure can be combined or combined in various ways, even if such combinations or combinations are not explicitly described in this disclosure. In particular, the features described in the various embodiments and / or claims of this disclosure can be combined or combined in various ways without departing from the spirit and teachings of this disclosure. All such combinations and / or combinations fall within the scope of this disclosure.

[0279] The embodiments of this disclosure have been described above. However, these embodiments are for illustrative purposes only and are not intended to limit the scope of this disclosure. Although various embodiments have been described above, this does not mean that the measures in the various embodiments cannot be used advantageously in combination. The scope of this disclosure is defined by the appended claims and their equivalents. Various substitutions and modifications can be made by those skilled in the art without departing from the scope of this disclosure, and all such substitutions and modifications should fall within the scope of this disclosure.

Claims

1. A model training method, comprising: Obtain a training set corresponding to the target semiconductor device, wherein the training set includes n sets of data, where n is a positive integer, and each set of data includes the junction temperature, on-state voltage drop, reverse voltage drop, and load current corresponding to the junction temperature; and Based on the training set, the initial model is trained to obtain a first target model for eliminating the influence of load current on junction temperature monitoring. This includes: constructing the initial model based on implicit functions corresponding to on-state voltage drop, reverse voltage drop, and load current, wherein the implicit functions follow a Gaussian distribution, and the initial model also follows a Gaussian distribution; determining the mean function and covariance matrix of the initial model; and obtaining a joint distribution corresponding to the training set and the prediction set based on the Gaussian process corresponding to the training set and the Gaussian process corresponding to the prediction set, wherein the Gaussian process is determined by the mean function and the covariance matrix, and the prediction set includes m sets of data, where m is a positive integer. In the dataset, the on-state voltage drop, reverse voltage drop, and load current are known, while the junction temperature is unknown. Based on the joint distribution and prior distribution, a conditional distribution corresponding to the prediction set is obtained, where the prior distribution represents a Gaussian process corresponding to the training set. Based on the conditional distribution, a first target model is determined. Using the first target model, based on the on-state voltage drop and reverse voltage drop in the prediction set, the junction temperature corresponding to each set of data in the prediction set is obtained. The initial model represents a functional model of the relationship between junction temperature and on-state voltage drop, reverse voltage drop, and load current, while the first target model represents a functional model of the relationship between junction temperature and on-state voltage drop and reverse voltage drop.

2. The method according to claim 1, wherein, The acquisition of the training set corresponding to the target semiconductor device includes: During each of the n double pulses, a set of data in the training set is measured, wherein the n double pulses are sent to the target semiconductor device.

3. The method according to claim 2, wherein, The double pulse includes a first pulse and a second pulse, and during each of the n double pulses, a set of data in the training set is measured, including: For each of the two pulses, during the first pulse, a load inductor on the heating plate is charged with DC current so that after the first pulse ends, the load current of the load inductor is measured, wherein the load current of the load inductor characterizes the load current of the target semiconductor device. Between the first pulse and the second pulse, the reverse voltage drop of the target semiconductor device is measured based on the freewheeling of the load current on the semiconductor device located in the upper bridge arm, wherein the target semiconductor device is located in the lower bridge arm, and both the semiconductor device and the target semiconductor device are located on the heating plate; During the second pulse, the on-state voltage drop of the target semiconductor device is measured based on the load current commutated from the semiconductor device to the target semiconductor device; After the second pulse ends, the junction temperature of the target semiconductor device is measured, wherein the junction temperature of the target semiconductor device characterizes the temperature of the heating plate.

4. The method according to claim 3, wherein, The load current of the target semiconductor device is controlled by adjusting the width of the first pulse in the dual pulse to obtain the on-state voltage drop, reverse voltage drop, and junction temperature under different load currents.

5. The method according to claim 1, wherein, Determining the mean function and covariance matrix of the initial model includes: Using a kernel function, an element in the covariance matrix is ​​determined for any two sets of data in the training set, wherein the mean function is set to zero, and the covariance matrix represents an n*n matrix.

6. The method according to claim 1, wherein, The step of determining the first target model based on the conditional distribution, and then using the first target model to obtain the junction temperature corresponding to each set of data in the prediction set based on the on-state voltage drop and reverse voltage drop in the prediction set, includes: Based on the conditional distribution, a first initial target model is determined; Parameter estimation is performed on the unknown hyperparameters in the first initial target model to obtain the hyperparameter values ​​corresponding to each unknown hyperparameter; and The first target model is obtained based on the first initial target model and each hyperparameter value.

7. The method according to claim 1, further comprising: The first target model is tested using the training set to obtain test results, wherein the test results characterize the performance of the first target model.

8. The method according to claim 7, wherein, The step of testing the first target model using the training set to obtain test results includes: The n sets of data in the training set are divided into K datasets, where K is an integer greater than or equal to 2. Using each of the K datasets sequentially as the test set and the remaining K-1 datasets as the training set, the first target model is tested to obtain K test sub-results; and The test result is obtained based on the K test sub-results.

9. The method according to claim 8, wherein, The first target model is tested by sequentially using each of the K datasets as a test set and the remaining K-1 datasets as a training set, resulting in K test sub-results, including: Using each of the K datasets as a test set, the first target model is trained using the remaining K-1 datasets as a training set to obtain training parameters; and The second target model is tested using the test set to obtain the test sub-result, wherein the second target model is determined based on the first target model and the training parameters.

10. A method for online monitoring of the junction temperature of a semiconductor device, comprising: The on-state voltage drop and reverse voltage drop of the semiconductor device are input into a first target model, and the junction temperature of the semiconductor device is output to adjust the semiconductor device according to the junction temperature. The first target model is trained by the model training method according to any one of claims 1 to 9.

11. A model training device, comprising: The acquisition module is used to acquire a training set corresponding to the target semiconductor, wherein the training set includes n sets of data, where n is a positive integer, and each set of data includes the junction temperature, the on-state voltage drop, the reverse voltage drop, and the load current corresponding to the junction temperature; and The training module is used to train the initial model based on the training set to obtain a first target model for eliminating the influence of load current on junction temperature monitoring. The initial model represents a functional model of the relationship between junction temperature and on-state voltage drop, reverse voltage drop and load current, and the first target model represents a functional model of the relationship between junction temperature and on-state voltage drop and reverse voltage drop. The training module includes a construction unit, a first determination unit, a first acquisition unit, a second acquisition unit, and a second determination unit. A construction unit is used to construct the initial model based on the implicit functions corresponding to the on-state voltage drop, reverse voltage drop, and load current, wherein the implicit functions follow a Gaussian distribution, and the initial model follows a Gaussian distribution; The first determining unit is used to determine the mean function and covariance matrix of the initial model; The first obtaining unit is used to obtain a joint distribution corresponding to the training set and the prediction set based on the Gaussian process corresponding to the training set and the Gaussian process corresponding to the prediction set. The Gaussian process is determined by the mean function and the covariance matrix. The prediction set includes m sets of data, where m is a positive integer. In each set of data, the on-state voltage drop, reverse voltage drop, and load current are known, while the junction temperature is unknown. The second obtaining unit is used to obtain the conditional distribution corresponding to the prediction set based on the joint distribution and the prior distribution, wherein the prior distribution represents the Gaussian process corresponding to the training set; The second determining unit is used to determine the first target model based on the conditional distribution, so as to use the first target model to obtain the junction temperature corresponding to each set of data in the prediction set based on the on-state voltage drop and reverse voltage drop in the prediction set.

12. An online junction temperature monitoring device for a semiconductor device, comprising: An output module is used to input the on-state voltage drop and reverse voltage drop of a semiconductor device into a first target model and output the junction temperature of the semiconductor device to adjust the semiconductor device according to the junction temperature, wherein the first target model is trained by the model training method according to any one of claims 1 to 9.

13. An electronic device, comprising: One or more processors; Storage device for storing one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors perform the model training method according to any one of claims 1 to 9 or the online junction temperature monitoring method for semiconductor devices according to claim 10.

14. A computer-readable storage medium having stored executable instructions thereon, which, when executed by a processor, cause the processor to perform the model training method according to any one of claims 1 to 9 or the online junction temperature monitoring method for a semiconductor device according to claim 10.

15. A computer program product comprising a computer program that, when executed by a processor, implements the model training method according to any one of claims 1 to 9 or the online junction temperature monitoring method for a semiconductor device according to claim 10.