Gate driver circuit and method for driving transistors

By introducing a deep neural network into the gate driver circuit, the transistor switching is dynamically tuned, which solves the problems of power loss and electromagnetic interference, extends the operating life of the transistor, and adapts to system-specific constraints and temperature changes.

CN123074468APending Publication Date: 2026-11-10INFINEON TECHNOLOGIES AG
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202610535712.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Priority Date
2025-05-09
Filing Date
2026-04-22
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

Existing gate driver circuits suffer from power loss and electromagnetic interference when switching transistors, affecting the transistor's operating life. Furthermore, their fixed configuration makes them unsuitable for different switching conditions.

Method used

A gate driver circuit including a computing module is employed, and the switching is dynamically tuned using a pre-trained function of a deep neural network (DNN). The output control signal is generated based on the input vector to minimize power loss and electromagnetic interference and optimize transistor operating life.

Benefits of technology

By dynamically adjusting and switching, unnecessary losses and electromagnetic interference are reduced, the operating life of transistors is extended, and the system-specific constraints and temperature changes are adapted.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123074468A_ABST
    Figure CN123074468A_ABST
Patent Text Reader

Abstract

This disclosure relates to a gate driver circuit and method for driving a transistor. The gate driver circuit and the method performed by the gate driver circuit include: a computation module configured to infer a pre-trained function to output a control signal to the transistor based on an output vector. The pre-trained function is configured to generate an output vector based on an input vector and an input control signal. The input vector includes at least one input value associated with transistor operation. A method for training a function to generate a pre-trained function for the gate driver circuit includes: receiving a training input vector associated with training transistor operation via a computer interface; applying the training input vector to an untrained function including an untrained neural network; and causing the neural network to generate a training output vector based on the training input vector while minimizing power dissipation of the training transistor. A training system includes: a training module configured to perform such methods.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to a gate driver circuit and a method for driving transistors, including a machine learning model. Background Technology

[0002] Typically, power loss or electromagnetic interference occurs when gate driver circuits switch transistors, which can affect the transistor's operational lifespan. Each switching event must generally comply with transistor-specific limitations (e.g., maximum voltage, specific current, or voltage gradient). However, prior art gate driver circuits often have fixed configurations, resulting in the aforementioned power loss or electromagnetic interference. Summary of the Invention

[0003] A gate driver circuit as defined in claim 1, a method performed by a gate driver circuit including a computing module as defined in claim 11, and a training system as defined in claim 22 are provided. The dependent claims define further embodiments.

[0004] According to one embodiment, a gate driver circuit is provided, including a computation module configured to infer a pre-trained function to output a control signal to a transistor based on an output vector. The pre-trained function is configured to generate an output vector based on an input control signal indicating a desired state of the transistor and an input vector. The input vector includes at least one input value associated with transistor operation.

[0005] According to another embodiment, a method is provided performed by a gate driver circuit including a computing module. The method includes inferring a pre-trained function and receiving an input vector and an input control signal indicating a desired state of a transistor. The input vector includes at least one input value associated with operation of the transistor. The method further includes applying the input vector to the pre-trained function to generate an output vector based on the input control signal and the input vector, and providing a control signal to the transistor according to the output vector.

[0006] According to another embodiment, a method is provided for training a function to generate a pre-trained function for a gate driver circuit. The method includes the steps of: receiving a training input vector associated with the operation of a training transistor via a computer interface. The method further includes applying the training input vector to an untrained function (which includes an untrained neural network), and causing the neural network to generate a training output vector based on the training input vector while minimizing power dissipation by the training transistor.

[0007] According to another embodiment, a training system is provided, including a training module configured to perform steps of a method for training functions to generate pre-trained functions for gate driver circuitry.

[0008] The above overview is intended only to briefly illustrate certain aspects of some embodiments and should not be construed as limiting in any way. Attached Figure Description

[0009] Figure 1 The illustration shows a gate driver circuit including a computing module according to an embodiment.

[0010] Figure 2 The diagram illustrates the different collector current transients depending on the gate resistance.

[0011] Figure 3 The illustration shows a deep neural network (DNN) for generating output vectors according to an embodiment.

[0012] Figure 4 The illustration shows a DNN according to an embodiment.

[0013] Figure 5 The illustration shows a gate driver circuit of a model including a computing module and an electrical layout according to an embodiment.

[0014] Figure 6 The diagram illustrates the voltage ramp generated based on the output vector.

[0015] Figure 7 This is a flowchart of a method performed by a gate driver circuit according to an embodiment.

[0016] Figure 8 This is a flowchart of a method for training functions to generate pre-trained functions for gate driver circuits, according to an embodiment.

[0017] Figure 9 The diagram illustrates ten voltage ramp curves for different temperatures over time. Detailed Implementation

[0018] In the following, embodiments will be described in detail with reference to the accompanying drawings. It should be understood that the following description of the embodiments should not be construed as limiting. The scope of this application is not intended to be limited to the embodiments or drawings described below, which are for illustrative purposes only. Those skilled in the art will recognize that various alternative and / or equivalent implementations can be chosen instead of the specific embodiments shown and described herein without departing from the scope of the invention as depicted. It is intended that this application cover all modifications or variations to the specific embodiments discussed herein. Therefore, it is intended that the invention be limited only to the claims and their equivalents.

[0019] The accompanying drawings should be considered schematic representations, and the elements illustrated in the drawings are not necessarily shown to scale. Rather, the various elements are represented in a manner that makes their general function obvious to those skilled in the art. Any connections or couplings between functional blocks, devices, components, or other physical or functional units shown in the drawings or described herein may also be achieved through indirect connections or couplings. Functional blocks may be implemented in hardware, firmware, software, or a combination thereof. In addition to the features explicitly shown and described, additional features may be provided in other embodiments, such as those used in conventional switching devices employing phase change materials.

[0020] In all the accompanying drawings and descriptions, the same reference numerals are used to describe the same features or parts.

[0021] Figure 1 A gate driver circuit according to an embodiment is illustrated. The gate driver circuit includes a computation module 10. The computation module 10 is configured to infer a pre-training function 200 to output a control signal 400 to a transistor 50 based on an output vector 250. The pre-training function 200 is configured to generate the output vector 250 based on an input control signal 100 and an input vector. The input control signal 100 indicates a desired state of the transistor 50, such as on or off. The input vector includes at least one input value 220 associated with the operation of the transistor 50.

[0022] The output vector 250 can be converted into a control signal 400 via the driver circuit 300.

[0023] Transistor 50 can be integrated into different systems, potentially creating system-specific constraints. Each switching event of the transistor may require compliance with these system-specific constraints, thus the control signal 400 may need to be adjusted accordingly. Such constraints may, for example, relate to power losses, electromagnetic interference tolerances, and / or transistor operating lifetime. For instance, if the current increases, conduction losses in the transistor may also increase. Higher currents may also result in steeper current transitions (di / dt) during switching, potentially triggering stronger electromagnetic interference (EMI). Furthermore, sustained high current levels can lead to thermal stress, which may shorten transistor lifetime over time. This can also affect switching losses if the voltage increases. Additionally, high voltage transients (dv / dt) can cause capacitive coupling effects and increased EMI, which may negatively impact the long-term stability and reliability of the transistor. As another example, increased temperature can lead to higher conduction and / or switching losses. This may be due to changes in on-resistance and / or carrier mobility during switching transitions, which may also affect the transistor's threshold voltage. Furthermore, variations in environmental and / or operating conditions can cause temperature fluctuations, resulting in system-specific thermal characteristics. Typically, temperature can affect aging mechanisms within a transistor, such as material fatigue, bond degradation, or gate oxide wear, which can shorten its operational life.

[0024] In a comparative example, the on-collector current is adjusted by implementing a gate resistor, thereby adjusting the switching operation. To illustrate the effect, Figure 2 The diagram illustrates the possible transient on-collector current of a transistor over time with varying gate resistances. However, while operating parameters may vary over time or throughout operation, such implementations are typically fixed for all switching conditions. Using fixed settings can lead to unnecessary losses or increased EMI during switching. Therefore, the calculation module 10 can be implemented to dynamically tune switching, thereby reducing such losses or EMI and potentially improving the operational life of the transistor device.

[0025] In some embodiments, the pre-trained function 200 includes a deep neural network (DNN) comprising at least two layers of interconnected neurons. Gate driver circuits including such DNNs are as follows: Figure 3 As shown in the diagram.

[0026] exist Figure 3In this example, the DNN comprises two layers, Layer 1 and Layer 2; however, this is merely an example, and the DNN may also comprise three, five, or more layers. Each layer comprises a set of computational elements Z11 to Z24, called neurons. Each neuron Z11-Z24 receives one or more input values, applies mathematical operations, and generates an output value. The first layer (Layer 1) may receive its input from input vector 150. In some embodiments, input vector 150 may include at least one of the following values ​​associated with the operation of transistor 50: the drain current or collector current of the transistor, the drain voltage or collector voltage of the transistor, and the transistor temperature. For example, these values ​​may be input value 220 received from transistor 50. These values ​​associated with the operation of the transistor can provide information about the power consumption and / or efficiency of the transistor at that moment of operation. Mathematical operations may include linear combinations, such as a weighted sum of input values. Mathematical operations may also include activation functions, such as the Rectified Linear Unit (ReLU) function. In some embodiments, mathematical operations may also include a normalization step, which can stabilize the training. These mathematical operations are merely examples, and other mathematical operations may be applied to replace or supplement them to generate output values. These output values ​​can then be passed to subsequent layers, enabling the network to learn patterns and hierarchical features from the input data, thus allowing it to perform tasks such as classification or prediction. For example, a DNN may include 50-1000 neurons (preferably 100-200 neurons) such that the transistor switching speed is in the range of 10 ns to 1 ms (preferably in the range of 100 ns to 1 us). In some embodiments, the computing module 10 adapts the switching speed by including a multi-core implementation comprising at least one processing core having a processing frequency in the range of 1 kHz to 50 kHz.

[0027] Figure 4 An example implementation of a DNN according to an embodiment is illustrated. In some embodiments, the DNN can be created in a physically inspired manner. For example, the DNN can be a Physically Inspired Neural Network (PiNN). A Physically Inspired Neural Network integrates physical equations into a data-driven model to ensure consistency with known physical properties of the system, enabling the network to extrapolate beyond the available data. In this example, the input vector 150 includes the drain voltage Vd and the gate voltage Vg. These values ​​are provided to multiple neurons Z. i These neurons can perform mathematical operations, such as linear combinations, which might involve calculating a weighted sum of input values. For example, a neuron can assign weights w and biases b to input values. In this example, neuron Z... iThis includes different activation functions. For example, at least some neurons use the sigmoid activation function sig, which maps the input to a range between 0 and 1. Other neurons may employ the hyperbolic tangent activation function tanh, which maps the input to a range between -1 and +1. The sigmoid activation function can be used for classification tasks, such as binary classification. The hyperbolic tangent activation function can be used to represent information in a symmetric, zero-centered range. These neurons can also support backpropagation, particularly during training, enabling the adjustment of weights and biases (e.g., based on computed gradients) to minimize prediction error. The output vector 250 may include a drain current Id, which can then be converted into a control signal 400 for transistor 50.

[0028] exist Figure 4 In the embodiment, the network structure implements two main rules. The first rule states that when Vd = 0, the network output (Id in this case) is forced to an exact zero because all biases b of the Vd subnet 201 are set to zero and tanh(0) = 0. The second rule states that changes in Vg are not allowed to alter the sign of the output Id. This can be achieved by implementing a sigmoid activation function for the Vg subnet 202 according to this rule. These rules are derived from known system physics characteristics, as they reflect the operating principles of transistors.

[0029] For example, pre-trained function 200 is pre-trained to achieve at least one of the following: minimizing transistor power loss during transistor switching operations, minimizing transistor electromagnetic interference (EMI) during transistor switching operations, maximizing transistor slave operation lifetime, and tolerating or compensating for transistor manufacturing variations. Such manufacturing variations may occur during manufacturing and may include, for example, variations in doping, lithographic misalignment, oxide thickness variations, or deviations in channel geometry. In some embodiments, pre-trained function 200 is pre-trained to handle such manufacturing tolerances or variations throughout the system, ensuring that these variations substantially do not affect the performance of the gate driver circuitry. In some embodiments, pre-trained function 200 is pre-trained for constraints specific to a particular system. Such constraints may stem from a variety of applications and requirements. These constraints may include specific requirements related to high-frequency operation, limitations on permissible parasitic inductance or capacitance for minimizing EMI, and constraints on critical components such as motor windings. Additionally, there may also be requirements related to operation lifetime, such as the duration of avalanche operation.

[0030] For example, meeting these constraints can be achieved by optimizing the switching duration. Switching too slowly can lead to increased power loss, while switching too quickly can lead to increased EMI. By balancing switching speed to minimize both losses and EMI, transistor operating life can be improved because mechanisms such as material fatigue, bond degradation, and gate oxide wear can be reduced.

[0031] For example, the switching speed can be adjusted by the gate voltage. In some embodiments, the output vector 250 includes a single voltage value that is clocked and applied to transistor 50 at discrete steps, for example, via driver circuitry 300. Different clock cycles can be used to adjust the voltage ramp over time. In some embodiments, the output vector 250 includes multiple gate voltage values ​​or gate current values. These values ​​can be arranged in a time sequence such that the multiple gate voltage values ​​form a time-dependent voltage ramp for switching transistor 50. This is as follows: Figure 5 As illustrated. Output vector 250 includes multiple gate voltage values ​​V. g0 To V gn These values ​​are assigned to the time series t0 to tn via driver circuit 300, thereby converting output vector 250 into control signal 400. Figure 5 In the example, output vector 250 includes 10 gate voltage values, but this is just an example, and output vector 250 may also include 10 or fewer, or even more, such as 20, 50 or more gate voltage values.

[0032] Figure 6 The diagram illustrates multiple gate voltage values ​​V. g0 To V gn How to arrange the time sequence to form a voltage ramp (which corresponds, for example, to control signal 400).

[0033] In some embodiments, the pre-trained function 200 can be pre-trained based on a model of the electrical layout 30, which includes at least one electrical connection formed between the gate driver circuit and the transistor 50, such as... Figure 4 As shown. Electrical layout 30 may correspond to at least a portion to which transistor 50 in a particular system or network may be integrated. For example, electrical layout 30 may include components such as: voltage and current sources; switching components or loads; components or subsystems that consume power and can affect EMI and / or thermal conditions; power paths and / or signal traces between power supplies, switches, and loads; or heat sinks.

[0034] For example, the pre-trained function 200 can also be pre-trained to handle the manufacturing tolerances of the electrical layout 30 and / or its components.

[0035] In some embodiments, the pre-trained function 200 can also be pre-trained based on a parasitic capacitance model simulated according to the electrical layout 30. This pre-trained function can be configured to generate an output vector 250 based on the input vector 150 and the parasitic capacitance model. By taking into account the electrical layout 30 and / or other environmental components associated with the target system, the control of the transistor 50 can be further optimized to operate the specific system. Furthermore, by providing a model for the target implementation, the network can be trained for that specific model.

[0036] For example, the pre-trained function 200 can be generated through a training process executed by the computation module 10, or the training system can provide the pre-trained function 200 to the gate driver circuit for execution after the training process is completed. In this way, the pre-trained function 200 includes operational information before its first use in a practical application, providing a solid foundation for subsequent optimization.

[0037] In some embodiments, the input vector 150 may further include at least one historical gate voltage value used to operate the transistor 50 before generating the output vector 250. For example, the pre-trained function 200 may be retrained to generate the output vector 250 based on the at least one historical gate voltage value. For example, the at least one historical gate voltage value indicates degradation of the transistor 50. The retrained function may be configured to generate the output vector 250, which minimizes power losses that occur during operation of the gate driver circuitry due to transistor degradation over time.

[0038] Generally, considering historical gate voltage values ​​allows for the analysis of transistor operating performance over time. For example, this might enable DNNs to analyze how the gate voltage changes under different conditions, thereby optimizing transistor operation and / or predicting transistor lifetime. Furthermore, this can help identify deviations or trends that can facilitate fault diagnosis.

[0039] Figure 7 The illustration depicts a method according to an embodiment, which is executed by a gate driver circuit including a computing module. To avoid repetition, reference will be made to... Figures 1 to 3 Gate driver circuit description Figure 7 The method can be implemented using a gate driver circuit including the computing module 10, but it can also be implemented in other circuits. The features detailed for the gate driver circuit also apply to the method described herein.

[0040] In step 1001, the method includes inferring a pre-training function 200. In step 1002, an input control signal 100 and an input vector 150 are received. The input control signal 100 indicates a desired state of the transistor (e.g., on or off). The input vector 150 includes at least one input value 220 associated with the operation of the transistor. In step 1003, the input vector 150 is applied to the pre-training function 200 to generate an output vector 250 based on the input control signal 100 and the input vector 150. In step 1004, a control signal 400 is provided to the transistor 50 according to the output vector 250.

[0041] In some embodiments, the method further includes generating a pre-trained function 200 by pre-training the function to achieve at least one of the following: minimizing the power loss of transistor 50 during switching operations of transistor 50, minimizing the electromagnetic interference of transistor 50 during switching operations of transistor 50, and maximizing the operating lifetime of transistor 50.

[0042] In some embodiments, the method further includes receiving at least one historical gate voltage value used to operate the transistor 50 before generating the output vector 250. Additionally, the method may include retraining a pre-trained function 200 to generate the output vector 250 based on the at least one historical gate voltage value. For example, the at least one historical gate voltage value may indicate degradation of the transistor 50, wherein the retrained function is configured to generate the output vector 250 to minimize power losses that occur during gate driver circuitry operation due to time-dependent degradation of the transistor 50.

[0043] Figure 8 The illustration depicts a method for training a function to generate a pre-trained function for a gate driver circuit, according to an embodiment. In step 2001, a training input vector associated with the operation of a training transistor is received via a computer interface. In some embodiments, the method further includes receiving a training input control signal indicating a desired state of the transistor (e.g., on or off). In step 2002, the training input vector is applied to an untrained function (which includes an untrained neural network). In step 2003, a training output vector is generated. This training output vector is generated by the neural network and, based on the training input vector, minimizes power dissipation by the training transistor and / or minimizes electromagnetic interference of the training transistor, and / or maximizes the operating lifetime of the training transistor.

[0044] The value of the optimization objective, such as minimizing power loss dissipated by the training transistor, minimizing electromagnetic interference (EMI) of the training transistor, and maximizing the operating lifetime of the training transistor, can be obtained through measurement and / or simulation. For example, voltage transients (e.g., drain-source voltage or collector-emitter voltage) across the training transistor and / or current transients (e.g., drain current or collector current) through the training transistor can be measured. Based on these transients, conduction and switching losses can be calculated to determine power loss. Furthermore, fast transients and high dv / dt or di / dt values ​​may indicate EMI sources. Operating lifetime can be estimated using temperature and electrical stress data over time. Additionally or alternatively, simulation techniques such as SPICE or behavioral modeling can be used to predict power loss and transient behavior under various operating conditions. Electromagnetic simulation or circuit models that consider parasitics help estimate EMI and identify system and / or layout-related constraints that may affect integrity and / or reliability.

[0045] Then, this pre-trained function can be used in the computation module that operates the real transistors.

[0046] The training input vector may include at least one of a training drain current value, a training drain voltage value, and / or a training transistor temperature value. Such training values ​​may, for example, resemble typical values ​​for a particular system or target implementation.

[0047] In some embodiments, the training output vector may include a plurality of gate voltage values ​​or gate current values ​​arranged in a time sequence, such that the plurality of gate voltage values ​​form a time-dependent voltage ramp for switching the training transistor.

[0048] In some embodiments, the method further includes simulating a training input vector based on a training electrical layout. The training electrical layout may include at least one electrical connection formed between the gate driver circuitry and the training transistor. For example, such a training electrical layout may include components such as: voltage and current sources; switching components or loads; components or subsystems that consume power and can affect EMI and / or thermal conditions; power paths and / or signal traces between power supplies, switches, and loads; or heat sinks.

[0049] The method may also include receiving at least one measurement value from a measuring device, the measuring device including gate driver circuitry and a physical implementation of a training electrical layout coupled to a training transistor. The at least one measurement value may form a training input vector.

[0050] Figure 9 The diagram illustrates ten voltage ramps based on time for different temperatures.

[0051] In this example, the training input vector consists of training transistor temperature values, and the training output vector consists of 25 output gate voltage values ​​assigned to a 10 ns time step. The DNN consists of two layers, with 64 neurons in the first layer and 32 neurons in the second layer. The ReLU function is used as the activation function. To train this function, seven out of ten temperatures were used, with the gate voltage values ​​at 40°C, 120°C, and 180°C predicted by the DNN. The symbols denote the training data.

[0052] In some embodiments, the method may further include receiving at least one historical gate voltage value used to operate the training transistor before generating the training output vector, and retraining a pre-trained function to generate the training output vector based on the at least one historical gate voltage value. The at least one historical gate voltage value may indicate degradation of the training transistor. The retrained function may generate the training output vector to minimize power loss due to time-dependent degradation of the training transistor during operation of the gate driver circuitry.

[0053] This method can be executed by a training system that includes a training module.

[0054] Some embodiments are defined by the following examples:

[0055] Example 1: A gate driver circuit, comprising: The computation module is configured to infer a pre-trained function to output control signals to the transistors based on the output vector. The pre-training function is configured to generate an output vector based on an input control signal indicating the desired state of the indicator transistor. The input vector includes at least one input value associated with the operation of the transistor.

[0056] Example 2: Based on the gate driver circuit of Example 1, The pre-trained function is pre-trained to achieve at least one of the following: Minimize transistor power loss during transistor switching operations. Minimize the electromagnetic interference of transistors during transistor switching operations, and Maximize the operating life of transistors.

[0057] Example 3: Gate driver circuit according to any of the preceding examples, The output vector includes multiple gate voltage values ​​or gate current values ​​arranged in a time sequence, such that the multiple gate voltage values ​​form a time-dependent voltage ramp for switching transistors.

[0058] Example 4: Gate driver circuit according to any of the preceding examples, The input vector includes at least one of the following values: The drain current or collector current of a transistor. The drain voltage or collector voltage of a transistor, and Transistor temperature.

[0059] Example 5: Gate driver circuit according to any of the preceding examples, The pre-trained function is pre-trained based on the following terms: The electrical layout model includes at least one electrical connection formed between the gate driver circuitry and the transistor, and Based on the parasitic capacitance model of this electrical layout simulation, The pre-training function is configured to generate an output vector based on the input vector and the parasitic capacitance model.

[0060] Example 6: Gate driver circuit according to any of the preceding examples, The pre-trained function includes a deep neural network (DNN), which comprises at least two layers of interconnected neurons.

[0061] Example 7: Based on the gate driver circuit of Example 6, The DNN comprises 50-1000 neurons, preferably 100-200 neurons, such that the switching speed of the transistor is in the range of 10 ns to 1 ms, preferably in the range of 100 ns to 1 us.

[0062] Example 8: Based on the gate driver circuit of Example 7, The computing module adapts to the switching speed through a multi-core implementation, which includes at least one processing core with a processing frequency in the range of 1 kHz to 50 kHz.

[0063] Example 9: Gate driver circuit according to any of the preceding examples, The pre-trained function is generated through a training process, which is executed by the computation module or by the training system providing the pre-trained function to the gate driver circuit for execution after the training process is completed.

[0064] Example 10: Gate driver circuit according to any of the preceding examples, The input vector also includes: At least one historical gate voltage value used to operate the transistor before generating the output vector, and The pre-trained function is retrained to generate an output vector based on the at least one historical gate voltage value. Wherein at least one historical gate voltage value indicates transistor degradation, and The retrained function is configured to generate an output vector to minimize power loss due to transistor degradation over time during gate driver circuit operation.

[0065] Example 11: A method performed by a gate driver circuit, the gate driver circuit including a computing module, the method comprising: Infer the pre-trained function. Receives an input vector and an input control signal indicating the desired state of the transistor. The input vector includes at least one input value associated with the operation of the transistor. The input vector is applied to a pre-trained function to generate an output vector based on the input control signal and the input vector. The control signal is provided to the transistor based on the output vector.

[0066] Example 12: Based on the method in Example 11, it also includes: Generate pre-trained functions using pre-trained functions to achieve at least one of the following: Minimize transistor power loss during transistor switching operations. Minimize electromagnetic interference of transistors during transistor switching operations, and Maximize the operating life of transistors.

[0067] Example 13: Based on the method of any one of Examples 11-12, The output vector includes multiple gate voltage values ​​or gate current values ​​arranged in a time sequence, such that the multiple gate voltage values ​​form a time-dependent voltage ramp for switching transistors.

[0068] Example 14: Based on any one of the methods in Examples 11-13, it also includes: Receive an input vector that includes at least one of the following values: The drain current or collector current of a transistor. The drain voltage or collector voltage of a transistor, and Transistor temperature.

[0069] Example 15: Based on the method of any one of Examples 11-14, The pre-trained function is pre-trained based on the following: A model of an electrical layout, comprising at least one electrical connection formed between a gate driver circuit and a transistor, and Based on the parasitic capacitance model of this electrical layout simulation, The method also includes generating an output vector based on the input vector and the parasitic capacitance model.

[0070] Example 16: Based on any one of the methods in Examples 11-15, it also includes: Receive at least one historical gate voltage value used to operate the transistor before generating the output vector, and The pre-trained function is retrained to generate an output vector based on at least one historical gate voltage value. The at least one historical gate voltage value indicates transistor degradation, and the retrained function is configured to generate an output vector to minimize power loss due to transistor degradation over time during gate driver circuit operation.

[0071] Example 17: A method for training functions to generate pre-trained functions for gate driver circuits, the method comprising the following steps: The training input vector associated with the operation of the training transistor is received via a computer interface. Applying the training input vector to an untrained function that includes an untrained neural network, and This enables the neural network to generate a training output vector based on the training input vector, while minimizing the power loss dissipated by the training transistors.

[0072] Example 18: Following the method in Example 17, The training input vector includes at least one of the following: Training drain current value, Training drain voltage value, and / or Training transistor temperature values.

[0073] Example 19: Following the method of Example 17 or 18, The training output vector includes multiple gate voltage values ​​or gate current values ​​arranged in a time sequence, such that the multiple gate voltage values ​​form a time-dependent voltage ramp for switching the training transistor.

[0074] Example 20: The method according to any one of Examples 17-19 also includes: The training electrical layout is based on the training input vector of the training electrical layout simulation, which includes at least one electrical connection formed between the gate driver circuit and the training transistor, and / or At least one measurement value is received from a measuring device, which includes a gate driver circuit and a physical implementation of a training electrical layout coupled to a training transistor, wherein the at least one measurement value forms a training input vector.

[0075] Example 21: Based on the method of Example 20, it also includes: Receive at least one historical gate voltage value used to operate the training transistor before generating the training output vector, and The pre-trained function is retrained to generate a training output vector based on at least one historical gate voltage value. Wherein the at least one historical gate voltage value indicates the degradation of the training transistor, and The retrained function generates a training output vector to minimize power loss during operation of the gate driver circuit due to the degradation of the training transistor over time.

[0076] Example 22: A training system including a training module configured to perform the method steps of any of Examples 17-21.

Claims

1. A gate driver circuit, comprising: The computation module is configured to infer a pre-trained function to output a control signal (400) to the transistor (50) based on the output vector (250). The pre-training function (200) is configured to generate the output vector based on an input control signal (100) indicating the desired state of the transistor (50) and based on an input vector (150). The input vector (150) includes at least one input value (220) associated with the operation of the transistor (50).

2. The gate driver circuit according to claim 1, in, The pre-trained function (200) is pre-trained to achieve at least one of the following: To minimize the power loss of the transistor (50) during its switching operation, To minimize electromagnetic interference of the transistor (50) during switching operations. To maximize the operating lifetime of the transistor (50), and Production deviations of the transistor (50) are permitted.

3. The gate driver circuit according to any one of the preceding claims, in, The output vector (250) includes a plurality of gate voltage values ​​or gate current values ​​arranged in a time sequence, such that the plurality of gate voltage values ​​form a time-dependent voltage ramp for switching the transistor (50).

4. The gate driver circuit according to any one of the preceding claims, in, The input vector (150) includes at least one of the following values: The drain current or collector current of the transistor (50), The drain voltage or collector voltage of the transistor (50), and Transistor temperature.

5. The gate driver circuit according to any one of the preceding claims, in, The pre-trained function (200) is pre-trained based on the following: The model of the electrical layout (30) includes at least one electrical connection formed between the gate driver circuit and the transistor (50), and Based on the parasitic capacitance model of the electrical layout (30) simulation, The pre-training function (200) is configured to generate the output vector (250) based on the input vector (150) and the parasitic capacitance model.

6. The gate driver circuit according to any one of the preceding claims, in, The pre-trained function (200) includes a deep neural network (DNN) comprising at least two layers of interconnected neurons.

7. The gate driver circuit according to any one of the preceding claims, in, The pre-trained function (200) is generated through a training process, which is executed by the computing module or by the training system providing the pre-trained function (200) to the gate driver circuit for execution after the training process is completed.

8. The gate driver circuit according to any one of the preceding claims, in, The input vector (150) also includes: At least one historical gate voltage value used to operate the transistor (50) before generating the output vector (250), and The pre-trained function (200) is retrained to generate the output vector (250) based on the at least one historical gate voltage value. Wherein, the at least one historical gate voltage value indicates the degradation of the transistor (50), and The retrained function is configured to generate the output vector (250) to minimize the power loss that occurs over time due to the degradation of the transistor (50) during the operation of the gate driver circuit.

9. A method performed by a gate driver circuit, the gate driver circuit including a computing module, the method comprising: Infer the pre-trained function (200). It receives an input vector (150) and an input control signal (100) indicating the desired state of the transistor. The input vector (150) includes at least one input value (220) associated with the operation of the transistor (50). The input vector (150) is applied to the pre-training function (200) to generate an output vector (250) based on the input control signal (100) and the input vector (150). A control signal (400) is provided to the transistor (50) according to the output vector (250).

10. The method of claim 9, further comprising: At least one of the following is achieved by pre-training the function to generate the pre-trained function (200): To minimize the power loss of the transistor (50) during its switching operation. To minimize electromagnetic interference of the transistor (50) during switching operations. To maximize the operating lifetime of the transistor (50), and Production deviations of the transistor (50) are permitted.

11. The method according to any one of claims 9-10, in, The output vector (250) includes a plurality of gate voltage values ​​or gate current values ​​arranged in a time sequence, such that the plurality of gate voltage values ​​form a time-dependent voltage ramp for switching the transistor (50).

12. The method according to any one of claims 9-11, further comprising: The input vector (150) is received, and the input vector (150) includes at least one of the following values: The drain current or collector current of the transistor (50), The drain voltage or collector voltage of the transistor (50), and Transistor temperature.

13. The method according to any one of claims 9-12, in, The pre-trained function (200) is pre-trained based on the following: The model of the electrical layout (30) includes at least one electrical connection formed between the gate driver circuit and the transistor (50), and Based on the parasitic capacitance model of the electrical layout (30) simulation, The method further includes generating the output vector (250) based on the input vector (150) and the parasitic capacitance model.

14. The method according to any one of claims 9-13, further comprising: Receive at least one historical gate voltage value used to operate the transistor (50) before generating the output vector (250), and The pre-trained function (200) is retrained to generate the output vector (250) based on the at least one historical gate voltage value. The at least one historical gate voltage value indicates the degradation of the transistor (50), and the retrained function is configured to generate the output vector (250) to minimize the power loss that occurs over time due to the degradation of the transistor (50) during the operation of the gate driver circuit.

15. The method according to any one of claims 9-14, wherein, The pre-trained function (200) is pre-trained based on the following: The training input vector associated with the operation of the training transistor is received via a computer interface. The training input vector is applied to an untrained function, including an untrained neural network. The neural network generates a training output vector based on the training input vector while minimizing the power loss dissipated by the training transistor.

16. The method according to claim 15, in, The training input vector includes at least one of the following: Training drain current value, Training drain voltage value, and / or Training transistor temperature values.

17. The method according to claim 15 or 16, in, The training output vector includes a plurality of gate voltage values ​​or gate current values ​​arranged in a time sequence, such that the plurality of gate voltage values ​​form a time-dependent voltage ramp for switching the training transistor.

18. The method according to any one of claims 15-17, further comprising: The training input vector is simulated based on a training electrical layout (30), the training electrical layout including at least one electrical connection formed between the gate driver circuit and the training transistor, and / or At least one measurement value is received from a measuring device, the measuring device including the gate driver circuit and a physical implementation of the training electrical layout (30) coupled to the training transistor, wherein the at least one measurement value forms the training input vector.

19. The method of claim 18, further comprising: Receive at least one historical gate voltage value used to operate the training transistor before generating the training output vector, and The pre-trained function (200) is retrained to generate the training output vector based on the at least one historical gate voltage value. Wherein, the at least one historical gate voltage value indicates the degradation of the training transistor, and The retrained function generates the training output vector to minimize power loss due to the degradation of the training transistor over time during operation of the gate driver circuit.

20. A training system comprising a training module configured to perform the method steps of any one of claims 15-19.