Switching circuit control method based on wide bandgap semiconductors
By combining multi-source synchronous sampling, data fusion, and a lightweight gated recursive unit network, the problem of multi-source synchronous acquisition and data fusion in the control of existing wide-bandgap semiconductor switching circuits is solved. This enables accurate identification and timely adjustment of the switching state, ensuring a high degree of matching and long-term stability between the control logic and the dynamic characteristics of the device switching.
Patent Information
- Application Number
- CN202610955674.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-06-30
- Publication Date
- 2026-11-10
AI Technical Summary
Existing wide-bandgap semiconductor switching circuit control technology cannot achieve multi-source synchronous acquisition and data fusion, lacks the ability to predict the timing of switching states, and causes the control logic to rely on fixed parameters, which is prone to adjustment lag or malfunction. Furthermore, it does not set a coordinated constraint between the turn-off threshold voltage and the gate drive current, resulting in poor adaptability to operating conditions and poor long-term operational stability.
By employing multi-source synchronous sampling, data fusion, trend prediction, and dynamic collaborative adjustment, an extended feature set is constructed, and a lightweight gated recursive unit network is used to determine the switching state. A hierarchical response control logic is established to achieve collaborative regulation of the turn-off threshold voltage and the gate drive current. Furthermore, the network weights are optimized through a closed-loop feedback calibration mechanism.
It achieves accurate identification and timely adjustment of the switching circuit state, avoids meaningless adjustments, ensures a high degree of matching between the control logic and the dynamic characteristics of the device switching, guarantees the accuracy and timeliness of the control action, and ensures long-term stable operation through collaborative constraint conditions and closed-loop calibration mechanism.
Smart Images

Figure CN123074466A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of electronic device control technology, specifically to a switching circuit control method based on wide bandgap semiconductors. Background Technology
[0002] Wide bandgap semiconductor materials, with their unique physical properties, have become core components of next-generation power electronic equipment and are widely used in key industrial fields such as new energy power generation, rail transportation, industrial frequency conversion, and aerospace. As power electronics technology continues to iterate and upgrade towards higher frequencies, higher efficiency, and smaller sizes, switching circuits, as the core functional components of various power electronic devices, directly determine the overall working state of the equipment through their control logic and operating performance. Currently, the industry has set higher standards for the adaptability of wide bandgap semiconductor switching circuits to different operating conditions and their level of intelligent control. Precise control of the switching characteristics of devices under complex operating environments has become a problem that the industry needs to overcome.
[0003] However, existing wide-bandgap semiconductor switching circuit control technologies have some limitations. Most existing solutions use single-parameter detection methods, which cannot achieve multi-source synchronous acquisition and data fusion processing of key characteristics of the switching process. They are difficult to fully characterize the true switching state of the device. The control logic relies on fixed parameter settings and lacks the ability to predict the timing of switching state deviations. It is impossible to predict the device's operating trend in advance. The adjustment mechanism adopts a direct trigger mode without graded response and parameter pre-calculation process, which is prone to adjustment lag or malfunction. At the same time, the lack of coordinated constraints between the turn-off threshold voltage and the gate drive current can easily lead to parameter adjustment conflicts. In addition, existing technologies have not established a closed-loop calibration mechanism based on actual operating results, resulting in poor adaptability to operating conditions and poor long-term operational stability. Summary of the Invention
[0004] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a switching circuit control method based on wide-bandgap semiconductors. This invention acquires key feature data of the switching process through multi-source synchronous sampling, extracts and constructs a complete extended feature set through data fusion, outputs the switch state offset coefficient based on a lightweight gated recursive unit network and determines the offset state, and establishes a hierarchical response control logic. This technology achieves accurate state identification based on the full-dimensional features of the device switch. Through a hierarchical mechanism of pre-calculating moderate offsets and executing instructions for significant offsets, the adjustment parameter calculation is completed in advance to avoid the response delay of real-time calculation. At the same time, the control action is accurately triggered according to the offset state to avoid meaningless adjustments, so that the control logic is highly matched with the switching dynamic characteristics of the wide-bandgap semiconductor device, ensuring the accuracy and timeliness of the control action.
[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a switching circuit control method based on wide bandgap semiconductors, the specific steps of which are as follows:
[0006] Multi-source synchronous sampling: Through the charge integration circuit and low-voltage auxiliary detection circuit integrated into the gate drive circuit, four types of related data are collected in real time and transmitted to the controller.
[0007] Data fusion and extraction: The controller performs sliding window filtering and timing alignment on the four types of received associated data, extracts four-dimensional switch state feature vectors, calculates the residual net gate charge and Miller plateau time ratio, forms an extended feature set, and then transmits it to the lightweight gated recursive unit network.
[0008] Trend prediction and judgment: The lightweight gated recursive unit network receives the extended feature set, outputs the normalized switch state offset coefficient, determines the corresponding offset state based on the switch state offset coefficient, and transmits the offset state to the controller.
[0009] Dynamic Cooperative Adjustment: After receiving the offset status, the controller sends an adjustment command to the drive circuit when the offset status is significant. The drive circuit adjusts the turn-off threshold voltage through the adjustable reference voltage source, adjusts the gate drive current by switching the number of parallel branches in the drive current source array, and applies cooperative constraint conditions at the same time.
[0010] Closed-loop feedback calibration: The controller counts the switching success rate within each time window. After completing a preset number of switching cycles, the network weights of the lightweight gated recursive unit network are adjusted based on the statistical results of the switching success rate.
[0011] Furthermore, in the multi-source synchronous sampling, the charge integration circuit uses a high-precision operational amplifier and a capacitor to form an integration loop, respectively acquiring the amount of gate charge accumulated from the gate rising from zero potential to the end of the Miller plateau in each switching cycle and the total amount of charge released during the gate turn-off process; the low-voltage auxiliary detection circuit detects the start and end times of the Miller plateau of the gate-source voltage through a high-speed comparator, and acquires the time interval from the gate-source voltage entering the Miller plateau to exiting the Miller plateau; and obtains the instantaneous peak value of the drain current through a non-inductive resistor connected in series in the main circuit and a high-speed sample-and-hold circuit; the Miller plateau is a voltage range in which the gate-source voltage is approximately flat due to the charging and discharging effect of the Miller capacitance between the gate and drain during the switching process of a wide-bandgap semiconductor power device.
[0012] Furthermore, in the data fusion and extraction process, the controller uses a sliding window of length 3-7 to perform mean filtering on the four types of associated data. The four types of associated data are synchronized to the same switching cycle reference through timestamp alignment. The four-dimensional switch state feature vector is then extracted using the four-dimensional switch state feature vector formula, which is: ,in, This is a four-dimensional switch state feature vector. The amount of gate charge in the current cycle. This represents the total charge released during the gate turn-off process in the current cycle. This represents the time interval from when the gate-source voltage enters the Miller plateau to when it exits the Miller plateau in the current cycle. This represents the instantaneous peak value of the drain current in the current cycle.
[0013] Furthermore, in the data fusion extraction, the residual net gate charge is obtained by subtracting the total charge released during the gate turn-off process of the current cycle from the current cycle's gate charging charge. The Miller plateau time ratio is obtained by comparing the time interval between entering and exiting the Miller plateau from the gate-source voltage in the current cycle with the time interval between entering and exiting the Miller plateau from the gate-source voltage in the previous cycle. The extended feature set is composed of the four-dimensional switch state feature vector, the residual net gate charge, and the Miller plateau time ratio, specifically: ,in, To expand the feature set, This is the four-dimensional switch state feature vector for the current period. The remaining net gate charge for the current cycle. This represents the time interval from when the gate-source voltage enters the Miller plateau to when it exits the Miller plateau in the current cycle. This is the time interval from when the gate-source voltage enters the Miller plateau to when it exits the Miller plateau in the previous cycle.
[0014] Furthermore, in the trend prediction judgment, the lightweight gated recursive unit network adopts a single hidden layer temporal prediction architecture, consisting of an input layer, a hidden layer, and a fully connected output layer connected in series. The number of neurons in the input layer is consistent with the dimension of the extended feature set. The hidden layer contains 8-32 gated recursive unit neurons, each of which consists of two independent sigmoid gate structures (update gate and reset gate) and a tanh candidate state generation structure. The output of the hidden layer is connected to a ReLU activation function. The fully connected output layer contains one neuron and uses a sigmoid activation function. The lightweight... The gated recurrent unit network is trained on a wide-bandgap semiconductor device switching characteristic dataset, which includes device switching process feature data under different ambient temperatures, bus voltages, load currents, and aging levels. The training process employs a mean squared error loss function and an Adam optimizer for iterative optimization. After training, the network weights of the lightweight gated recurrent unit network are quantized and stored in the controller's storage unit. The input to the lightweight gated recurrent unit network is an extended feature set, and the output is the switching state offset coefficient. .
[0015] Furthermore, in the trend prediction and judgment, a first threshold of 0.3 and a second threshold of 0.7 are preset for the switch state offset coefficient. The first and second thresholds are determined based on the safe operating area data and the fluctuation range of the switching characteristics of the wide bandgap semiconductor device. Then, the corresponding offset state is determined based on the switch state offset coefficient. When the switch state offset coefficient... When the switch state offset coefficient is at that time, it is determined to be a slight offset. When the switch state offset coefficient is at that time, it is determined to be a moderate offset. When the value is significantly offset, it is considered a significant deviation.
[0016] Furthermore, in the dynamic coordinated adjustment, when the offset state is slight, the controller does not perform any operation; when the offset state is moderate, the controller pre-calculates the target turn-off threshold voltage and the target gate drive current according to the current switch state offset coefficient, and temporarily stores the pre-calculation results in the controller's cache; when the offset state is significant, the controller packages the pre-calculation results in the cache into an adjustment command and sends it to the drive circuit.
[0017] Furthermore, in the aforementioned dynamic coordinated adjustment, the adjustable reference voltage source is implemented using a digital-to-analog converter, with an output voltage range of -5V to 0V, and the turn-off threshold voltage is adjusted using a turn-off threshold voltage adjustment formula. The formula for adjusting the turn-off threshold voltage is: ,in, The initial turn-off threshold voltage is determined according to the datasheet of the wide-bandgap semiconductor device used. The first coordination coefficient is determined based on the gate threshold voltage characteristic parameters in the datasheet of the wide-bandgap semiconductor device used. This is the switch state offset coefficient;
[0018] The drive current source array consists of 4-16 parallel constant current source branches, each with an output current of 20-100mA. The gate drive current is adjusted using a gate drive current regulation formula. The formula for adjusting the gate drive current is: ,in, The initial drive current is determined based on the design values of the drive circuit. The second coordination coefficient is determined based on the gate charge characteristic parameters in the datasheet of the wide-bandgap semiconductor device used. This is the switch state offset coefficient;
[0019] The cooperative constraint is that when the turn-off threshold voltage is increased by more than 0.5V, the reduction in gate drive current does not exceed 30% of the initial value.
[0020] Furthermore, in the closed-loop feedback calibration, the switching success rate... The success rate of the switch is calculated using the following formula: ,in, To meet the requirement of no shoot-through, no gate overvoltage or undervoltage, and an absolute value of the rate of change of turn-off time less than a preset threshold, the preset threshold is determined based on the allowable fluctuation range of the switching characteristics. This represents the total number of switches within the given time window.
[0021] The time window length is set to 500-2000 switching cycles. When the switching success rate is higher than the high threshold for 2-4 consecutive time windows, the network weight of the lightweight gated recursive unit network is adjusted to reduce the response to small fluctuations in the switching state and reduce erroneous adjustments. When the switching success rate is lower than the low threshold in any time window, the weight of the lightweight gated recursive unit network is adjusted to improve the response speed to changes in the switching state. The high and low thresholds are determined based on the safe operating area of the wide bandgap semiconductor device and the allowable fluctuation range of the switching characteristics.
[0022] Compared with existing technologies, this switching circuit control method based on wide bandgap semiconductors has the following advantages:
[0023] I. This invention acquires key feature data of the switching process through multi-source synchronous sampling, extracts and constructs a complete extended feature set through data fusion, outputs the switch state offset coefficient based on a lightweight gated recursive unit network and determines the offset state, and establishes a hierarchical response control logic. This technology achieves accurate state identification based on the full-dimensional features of the device switch. Through a hierarchical mechanism of pre-calculating moderate offsets and executing instructions for significant offsets, the adjustment parameter calculation is completed in advance to avoid the response delay of real-time calculation. At the same time, the control action is accurately triggered according to the offset state to avoid meaningless adjustment. This makes the control logic highly matched with the switching dynamic characteristics of wide bandgap semiconductor devices, ensuring the accuracy and timeliness of the control action.
[0024] Second, this invention achieves synchronous regulation of the turn-off threshold voltage and gate drive current through dynamic collaborative adjustment, and applies collaborative constraints to avoid parameter adjustment conflicts. At the same time, a closed-loop feedback calibration mechanism dynamically adjusts the weight of the lightweight gated recursive unit network based on the switching success rate. The collaborative control of the turn-off threshold voltage and gate drive current is aligned with the physical logic of the device drive, and the constraints ensure the safety of the adjustment process. The closed-loop calibration mechanism continuously optimizes the lightweight gated recursive unit network based on actual operating data, which can adapt to different operating conditions and device state changes, forming a self-optimizing control closed loop to ensure the long-term stable operation of the switching circuit.
[0025] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description
[0026] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.
[0027] Figure 1 The flowchart shows a switching circuit control method based on wide bandgap semiconductors.
[0028] Figure 2 This is a framework diagram for data fusion and extraction in a switching circuit control method based on wide bandgap semiconductors;
[0029] Figure 3 This is a framework diagram of a switching circuit control method based on wide bandgap semiconductors. Detailed Implementation
[0030] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.
[0031] Example:
[0032] In high-frequency industrial power electronic conversion scenarios, wide-bandgap semiconductor switching circuits undertake the core function of efficient power conversion and are key components to ensure the stable operation of power electronic devices. The hardware architecture adopted in this embodiment includes wide-bandgap semiconductor power devices, gate drive circuits with integrated sampling circuits, controllers, adjustable reference voltage sources, and drive current source arrays.
[0033] Multi-source synchronous sampling: Through a charge integration circuit and a low-voltage auxiliary detection circuit integrated into the gate drive circuit, four types of related data on the switching state are acquired synchronously. The charge integration circuit consists of a high-precision operational amplifier and a non-inductive capacitor forming an integration loop. Within a single switching cycle, it acquires the amount of charge accumulated from the gate rising from zero potential to the end of the Miller plateau, and the total charge released during gate turn-off. These two types of charge data directly reflect the charge balance state of the gate drive. The low-voltage auxiliary detection circuit uses a high-speed comparator to monitor the gate-source voltage waveform, accurately identifying the start and end times of the Miller plateau, and thus obtaining the duration of the Miller plateau. The Miller plateau is an inherent physical characteristic of wide-bandgap semiconductor power devices, caused by the charging and discharging effect of the Miller capacitance between the gate and drain. It manifests as a voltage range where the gate-source voltage is approximately flat, and is a core characteristic characterizing the device's switching process. The instantaneous peak value of the drain current is obtained through a non-inductive sampling resistor connected in series in the main circuit, combined with a high-speed sample-and-hold circuit, to reflect the current stress state at the moment the device is turned on. The four types of related data are acquired synchronously under a unified clock reference, with no timing deviation. After acquisition, the data is directly transmitted to the controller. Figure 1 As shown.
[0034] Data fusion and extraction: The controller receives four types of related data, such as Figure 2 As shown, a sliding window of length 3-7 is first used to perform mean filtering on the four types of correlated data to remove high-frequency noise caused by electromagnetic interference at the site, while retaining the true trend of switch characteristics. Then, a timestamp alignment operation is used to synchronize data from different acquisition channels to the same switch cycle reference, solving the problem of time sequence misalignment of multi-source data. Finally, a four-dimensional switch state feature vector is extracted using the four-dimensional switch state feature vector formula, which is: ,in, This is a four-dimensional switch state feature vector. The amount of gate charge in the current cycle. This represents the total charge released during the gate turn-off process in the current cycle. This represents the time interval from when the gate-source voltage enters the Miller plateau to when it exits the Miller plateau in the current cycle. The instantaneous peak value of the drain current in the current cycle is given. Based on this, two sets of derived features are calculated: the residual net gate charge is obtained by subtracting the total charge released during turn-off from the current cycle's gate charge, used to determine the balance state of the gate charge; the Miller plateau time ratio is obtained by dividing the current cycle's Miller plateau time interval by the previous cycle, used to reflect the periodic change trend of the device's switching characteristics. Finally, the four-dimensional switching state feature vector, the residual net gate charge, and the Miller plateau time ratio are combined into an extended feature set, which serves as the input data for the lightweight gated recursive unit network, specifically: ,in, To expand the feature set, This is the four-dimensional switch state feature vector for the current period. The remaining net gate charge for the current cycle. This represents the time interval from when the gate-source voltage enters the Miller plateau to when it exits the Miller plateau in the current cycle. This is the time interval from when the gate-source voltage enters the Miller plateau to when it exits the Miller plateau in the previous cycle; this extended feature set takes into account both static parameters and dynamic trends, and can fully characterize the switching state of the device.
[0035] Trend prediction and judgment: A lightweight gated recurrent unit network (GRU) is used as the core to complete the quantitative prediction and hierarchical judgment of switch state shifts. This lightweight GRU adopts a single hidden layer temporal prediction architecture, consisting of an input layer, a hidden layer, and a fully connected output layer connected in series. The number of neurons in the input layer is consistent with the dimension of the extended feature set to ensure complete feature data input. The hidden layer contains 8-32 gated recurrent unit neurons, each consisting of two sigmoid gate structures (update gate and reset gate) and a tanh candidate state generation structure, enabling selective memorization of historical switch features and effective fusion of new features. The output of the hidden layer uses the ReLU activation function to improve feature extraction efficiency. The fully connected output... One neuron is set at the output layer, paired with a sigmoid activation function, to output a normalized switch state offset coefficient in the 0-1 interval. A lightweight gated recurrent unit network is pre-trained offline, with the training dataset covering switching process feature data under different ambient temperatures, bus voltages, load currents, and device aging levels. Before training, the data is normalized, and iterative optimization is performed using the mean squared error loss function and the Adam optimizer. After training, the network weights of the lightweight gated recurrent unit network are quantized and stored in the controller's storage unit. After the lightweight gated recurrent unit network outputs the switch state offset coefficient, the offset state is determined based on a preset first threshold and a second threshold: when the switch state offset coefficient... When the switch state offset coefficient is at that time, it is determined to be a slight offset. When the switch state offset coefficient is at that time, it is determined to be a moderate offset. When this occurs, it is determined to be a significant offset; the first threshold is 0.3, and the second threshold is 0.7. The first and second thresholds are determined based on the safe operating area data and switching characteristic fluctuation range of the wide-bandgap semiconductor device; after the determination is completed, the offset status is transmitted to the controller, such as... Figure 3 As shown.
[0036] Dynamic Coordinated Adjustment: A graded response mechanism is adopted to balance control response speed and operational stability. When the offset state is slight, the device switching characteristic fluctuation is within a safe and permissible range, and the controller does not perform any operation, maintaining the original control parameters for stable operation. When the offset state is moderate, the controller pre-calculates the target turn-off threshold voltage and target gate drive current based on the current switching state offset coefficient, and temporarily stores the calculation results in the controller's cache unit without outputting adjustment commands, thereby eliminating the real-time calculation delay when significant offset is triggered and adapting to the fast response requirements of high-frequency switching scenarios. When the offset state is significant, the controller packages the pre-calculated parameters in the cache into adjustment commands and sends them to the drive circuit. After receiving the adjustment commands, the drive circuit performs dual-parameter coordinated adjustment: the adjustable reference voltage source uses a digital-to-analog converter to adjust the turn-off threshold voltage within the range of -5V to 0V, and uses the turn-off threshold voltage adjustment formula to adjust the turn-off threshold voltage. The formula for adjusting the turn-off threshold voltage is: ,in, The initial turn-off threshold voltage is determined according to the datasheet of the wide-bandgap semiconductor device used. The first coordination coefficient is determined based on the gate threshold voltage characteristic parameters in the datasheet of the wide-bandgap semiconductor device used. This represents the switching state offset coefficient; the drive current source array consists of 4-16 parallel constant current source branches. The gate drive current is adjusted by switching the number of parallel branches. The output current of a single branch is 20-100mA, and the gate drive current is adjusted using the gate drive current adjustment formula. The formula for adjusting the gate drive current is: ,in, The initial drive current is determined based on the design values of the drive circuit. The second coordination coefficient is determined based on the gate charge characteristic parameters in the datasheet of the wide-bandgap semiconductor device used. This is the switching state offset coefficient; during the adjustment process, a cooperative constraint condition is applied: when the turn-off threshold voltage is increased by more than 0.5V, the gate drive current is reduced by no more than 30% of the initial value, so as to avoid parameter adjustment conflicts that may cause device overvoltage, overcurrent or abnormal increase in switching losses.
[0037] Closed-loop feedback calibration: The controller uses 500-2000 switching cycles as a statistical time window, and calculates the switching success rate within each time window. The switching success rate calculation formula is as follows: ,in, To meet the requirement of no shoot-through, no gate overvoltage or undervoltage, and an absolute value of the rate of change of turn-off time less than a preset threshold, the preset threshold is determined based on the allowable fluctuation range of the switching characteristics. The total number of switches within the given time window is defined as follows: the switch success rate is the ratio of the number of successful switches to the total number of switches within the time window. A successful switch is defined as a switch action with no shoot-through, no gate overvoltage / undervoltage, and a turn-off time change rate within a preset threshold. After completing a preset number of switch cycle statistics, the network weights are dynamically adjusted based on the switch success rate: when the switch success rate is higher than the high threshold for 2-4 consecutive time windows, the weights of the lightweight gated recursive unit network are fine-tuned to reduce the network's response to minor fluctuations in the switch state and reduce false triggering and invalid pre-calculation; when the switch success rate is lower than the low threshold in any time window, the network weights are fine-tuned to improve the network's response speed to abnormal changes in the switch state and quickly identify device offset trends. Both the high and low thresholds are determined based on the safe operating area of the wide bandgap semiconductor device and the allowable fluctuation range of the switch characteristics. The weight adjustment process is completed online without interrupting the normal operation of the switching circuit, forming a continuously optimized control closed loop.
[0038] In summary, this embodiment constructs a complete wide-bandgap semiconductor switching circuit control system through multi-source synchronous sampling, data fusion extraction, time-series trend prediction, hierarchical collaborative adjustment, and online closed-loop calibration. This system can effectively adapt to the operational requirements of high-frequency industrial power electronic conversion scenarios and ensure the long-term stable operation of the switching circuit.
[0039] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.
Claims
1. A switching circuit control method based on wide bandgap semiconductors, characterized in that, The specific steps of this method are as follows: Multi-source synchronous sampling: Through the charge integration circuit and low-voltage auxiliary detection circuit integrated into the gate drive circuit, four types of related data are collected in real time and transmitted to the controller. Data fusion and extraction: The controller performs sliding window filtering and timing alignment on the four types of received associated data, extracts four-dimensional switch state feature vectors, calculates the residual net gate charge and Miller plateau time ratio, forms an extended feature set, and then transmits it to the lightweight gated recursive unit network. Trend prediction and judgment: The lightweight gated recursive unit network receives the extended feature set, outputs the normalized switch state offset coefficient, determines the corresponding offset state based on the switch state offset coefficient, and transmits the offset state to the controller. Dynamic Cooperative Adjustment: After receiving the offset status, the controller sends an adjustment command to the drive circuit when the offset status is significant. The drive circuit adjusts the turn-off threshold voltage through the adjustable reference voltage source, adjusts the gate drive current by switching the number of parallel branches in the drive current source array, and applies cooperative constraint conditions at the same time. Closed-loop feedback calibration: The controller counts the switching success rate within each time window. After completing a preset number of switching cycles, the network weights of the lightweight gated recursive unit network are adjusted based on the statistical results of the switching success rate.
2. The switching circuit control method based on wide bandgap semiconductors according to claim 1, characterized in that, In the multi-source synchronous sampling, the charge integration circuit uses a high-precision operational amplifier and a capacitor to form an integration loop, and respectively collects the amount of gate charge accumulated from the rise of the gate from zero potential to the end of the Miller plateau in each switching cycle and the total amount of charge released during the gate turn-off process; the low-voltage auxiliary detection circuit detects the start and end times of the Miller plateau of the gate-source voltage through a high-speed comparator, and collects the time interval from the entry of the gate-source voltage into the Miller plateau to the exit of the Miller plateau; and obtains the instantaneous peak value of the drain current through a non-inductive resistor connected in series in the main circuit and a high-speed sample-and-hold circuit.
3. The switching circuit control method based on wide bandgap semiconductors according to claim 1, characterized in that, In the data fusion and extraction process, the controller uses a sliding window of length 3-7 to perform mean filtering on the four types of associated data. The four types of associated data are synchronized to the same switching cycle reference through timestamp alignment. The four-dimensional switch state feature vector is then extracted using the four-dimensional switch state feature vector formula, which is: ,in, This is a four-dimensional switch state feature vector. The amount of gate charge in the current cycle. This represents the total charge released during the gate turn-off process in the current cycle. This is the time interval from when the gate-source voltage enters the Miller plateau to when it exits the Miller plateau in the current cycle. This represents the instantaneous peak value of the drain current in the current cycle.
4. The switching circuit control method based on wide bandgap semiconductors according to claim 1, characterized in that, In the data fusion extraction, the residual net gate charge is obtained by subtracting the total charge released during the gate turn-off process in the current cycle from the current cycle's gate charging charge. The Miller plateau time ratio is obtained by comparing the time interval between entering and exiting the Miller plateau from the gate-source voltage in the current cycle with the time interval between entering and exiting the Miller plateau from the gate-source voltage in the previous cycle. The extended feature set is composed of the four-dimensional switch state feature vector, the residual net gate charge, and the Miller plateau time ratio, specifically: ,in, To expand the feature set, This is the four-dimensional switch state feature vector for the current period. The remaining net gate charge for the current cycle. This is the time interval from when the gate-source voltage enters the Miller plateau to when it exits the Miller plateau in the current cycle. This is the time interval from when the gate-source voltage enters the Miller plateau to when it exits the Miller plateau in the previous cycle.
5. The switching circuit control method based on wide bandgap semiconductors according to claim 1, characterized in that, In the trend prediction and judgment, the lightweight gated recursive unit network adopts a single hidden layer temporal prediction architecture, which consists of an input layer, a hidden layer, and a fully connected output layer connected in series. The number of neurons in the input layer is consistent with the dimension of the extended feature set. The hidden layer contains 8-32 gated recursive unit neurons. Each gated recursive unit consists of two independent sigmoid gate structures (update gate and reset gate) and a tanh candidate state generation structure. The output of the hidden layer is connected to the ReLU activation function. The fully connected output layer contains one neuron and uses the sigmoid activation function. The lightweight gated recursive unit network is trained on a wide-bandgap semiconductor device switching characteristic dataset. During the training process, the mean squared error loss function and the Adam optimizer are used for iterative optimization. After training, the network weights of the lightweight gated recursive unit network are quantized and stored in the controller's storage unit. The lightweight gated recurrent unit network takes an extended feature set as input and outputs switch state offset coefficients. .
6. The switching circuit control method based on wide bandgap semiconductor according to claim 1, characterized in that, In the trend prediction and judgment, a first threshold of 0.3 and a second threshold of 0.7 are preset for the switch state offset coefficient. The first and second thresholds are determined based on the safe operating area data and the fluctuation range of the switching characteristics of the wide bandgap semiconductor device. Then, the corresponding offset state is determined based on the switch state offset coefficient. When the switch state offset coefficient... When the switch state offset coefficient is at that time, it is determined to be a slight offset. When the switch state offset coefficient is at that time, it is determined to be a moderate offset. When the value is significantly offset, it is considered a significant deviation.
7. The switching circuit control method based on wide bandgap semiconductors according to claim 1, characterized in that, In the dynamic coordinated adjustment, when the offset state is slight, the controller does not perform any operation; when the offset state is moderate, the controller pre-calculates the target turn-off threshold voltage and the target gate drive current according to the current switching state offset coefficient, and temporarily stores the pre-calculation results in the controller's cache. When the offset state is significant, the controller packages the pre-calculated results in the cache into adjustment instructions and sends them to the drive circuit.
8. The switching circuit control method based on wide bandgap semiconductor according to claim 7, characterized in that, In the aforementioned dynamic coordinated adjustment, the adjustable reference voltage source is implemented using a digital-to-analog converter, with an output voltage range of -5V to 0V, and the turn-off threshold voltage is adjusted using a turn-off threshold voltage adjustment formula. The formula for adjusting the turn-off threshold voltage is: ,in, The initial turn-off threshold voltage is determined according to the datasheet of the wide-bandgap semiconductor device used. The first coordination factor is determined based on the gate threshold voltage characteristic parameters in the datasheet of the wide-bandgap semiconductor device used. This is the switch state offset coefficient; The drive current source array consists of 4-16 parallel constant current source branches, each with an output current of 20-100mA. The gate drive current is adjusted using a gate drive current regulation formula. The formula for adjusting the gate drive current is: ,in, The initial drive current is determined based on the design values of the drive circuit. The second coordination coefficient is determined based on the gate charge characteristic parameters in the datasheet of the wide-bandgap semiconductor device used. This is the switch state offset coefficient; The cooperative constraint is that when the turn-off threshold voltage is increased by more than 0.5V, the reduction in gate drive current does not exceed 30% of the initial value.
9. The switching circuit control method based on wide bandgap semiconductor according to claim 1, characterized in that, In the closed-loop feedback calibration, the switching success rate The success rate of the switch is calculated using the following formula: ,in, To meet the requirement of no shoot-through, no gate overvoltage or undervoltage, and an absolute value of the rate of change of turn-off time less than a preset threshold, the preset threshold is determined based on the allowable fluctuation range of the switching characteristics. This represents the total number of switches within the given time window. The time window length is set to 500-2000 switching cycles. When the switching success rate is higher than the high threshold for 2-4 consecutive time windows, the network weight of the lightweight gated recursive unit network is adjusted to reduce the response to small fluctuations in the switching state and reduce erroneous adjustments. When the switching success rate is lower than the low threshold in any time window, the weight of the lightweight gated recursive unit network is adjusted to improve the response speed to changes in the switching state. The high and low thresholds are determined based on the safe operating area of the wide bandgap semiconductor device and the allowable fluctuation range of the switching characteristics.
