CN-121978500-A - Mechanical stress wave monitoring method and system for junction temperature of SiC MOSFET

CN121978500ACN 121978500 ACN121978500 ACN 121978500ACN-121978500-A

Abstract

The invention discloses a mechanical stress wave monitoring method and a mechanical stress wave monitoring system for junction temperature of a SiC MOSFET, and relates to the technical field of semiconductor device detection. The method comprises the steps of enabling a device to reach a plurality of preset junction temperatures through a heating device, obtaining a plurality of off-current values through an adjusting circuit at each preset junction temperature, collecting mechanical stress wave signals excited by switching transients each time, synchronously collecting corresponding off-current values, extracting logarithmic root mean square, peak value and kurtosis from the mechanical stress wave signals to form acoustic feature vectors, combining the acoustic feature vectors with the off-current values to form input feature vectors, establishing a junction temperature prediction model by taking the input feature vectors as input and the preset junction temperature as output, and realizing real-time prediction of the junction temperature of the SiC MOSFET device by using the junction temperature prediction model. The invention realizes non-invasive and high-precision on-line monitoring of junction temperature of the SiC MOSFET device.

Inventors

He Binze
LI QIYING
PING YANG
DENG BAOYUAN
ZHANG JIE
YANG WENXUE
YUAN MAN
HE HONGYING

Assignees

湖南大学

Dates

Publication Date: 20260505
Application Date: 20260408

Claims (10)

1. A method for monitoring mechanical stress wave of junction temperature of a SiC MOSFET, the method comprising: The SiC MOSFET device is enabled to reach a plurality of different preset junction temperatures through the heating device; acquiring mechanical stress wave signals excited by each switching transient through an acoustic sensor and synchronously acquiring corresponding turn-off currents of the switching transient through a current acquisition unit to obtain the mechanical stress wave signals under different preset junctions and different turn-off currents; Extracting acoustic feature vectors at least comprising logarithmic root mean square, peak-to-peak value and kurtosis from mechanical stress wave signals under different preset junctions and different off-states, and combining the acoustic feature vectors with corresponding off-currents to form input feature vectors; Taking the input feature vector as input, taking the corresponding preset junction temperature as output, and establishing a mapping relation between the input feature and the junction temperature to obtain a junction temperature prediction model; and collecting the current mechanical stress wave signal and the current turn-off current of the SiC MOSFET device to be tested in real time, extracting the acoustic feature vector of the current mechanical stress wave signal, combining the acoustic feature vector with the current turn-off current, inputting the acoustic feature vector into the junction temperature prediction model, and outputting the current junction temperature predicted value of the SiC MOSFET device to be tested by the junction temperature prediction model.
2. The method for monitoring mechanical stress wave of junction temperature of SiC MOSFET according to claim 1, characterized in that the acoustic sensor is indirectly abutted to the SiC MOSFET device through a heat transfer barrier, specifically comprising: The method comprises the steps of abutting a first surface of a heat transfer barrier to a first surface of a SiC MOSFET device in a dry contact mode, abutting the acoustic sensor to a second surface of the heat transfer barrier, and coating an acoustic coupling agent between the acoustic sensor and the second surface of the heat transfer barrier, wherein the heating device is arranged on the side of a second surface, opposite to the first surface, of the SiC MOSFET device.
3. The method for monitoring mechanical stress waves of junction temperature of SiC MOSFET according to claim 2, wherein the heat transfer barrier element is a waveguide rod made of polyimide, and the acoustic couplant is a silicon-based couplant.
4. The method for monitoring mechanical stress waves of the junction temperature of the SiC MOSFET according to claim 1, wherein the method further comprises filtering the mechanical stress wave signals before extracting the acoustic feature vectors, and the frequency range of the filtering is set according to the noise distribution range determined by the lead breaking test so as to filter low-frequency mechanical background noise and retain effective signals.
5. The method for monitoring mechanical stress wave of junction temperature of SiC MOSFET according to claim 1, characterized in that the logarithmic root mean square is calculated according to the following formula: ; Wherein, F 1 represents the logarithmic root mean square, RMS represents the root mean square, N represents the total number of sampling points; a mechanical stress wave signal representing an nth sampling point; The peak-to-peak value is calculated according to the following formula: ; Wherein F 2 represents a peak-to-peak value; the kurtosis is calculated according to the following formula: ; Wherein F 3 represents kurtosis; Representing the mean value of the mechanical stress wave signal; representing the standard deviation of the mechanical stress wave signal.
6. The method of monitoring mechanical stress waves of junction temperature of a SiC MOSFET according to claim 1, wherein the acoustic feature vector further comprises a frequency domain feature or a time-frequency domain feature, the frequency domain feature comprising a frequency domain peak or a specific frequency band signal energy.
7. The method for monitoring mechanical stress waves of junction temperature of SiC MOSFET according to claim 1, wherein a machine learning algorithm is adopted to establish a mapping relation between input characteristics and junction temperature.
8. The method for monitoring mechanical stress waves of junction temperature of SiC MOSFET according to claim 7, wherein the machine learning algorithm is a gaussian process regression algorithm, and the mapping relationship between the input features and the junction temperature is established by using the gaussian process regression algorithm, comprising: Defining a kernel function, wherein the kernel function adopts an additive composite kernel function and is used for measuring covariance among input feature vectors; Initializing hyper-parameters of the kernel function, wherein the hyper-parameters comprise signal variance, length scale and noise variance; constructing a covariance matrix between input feature vectors based on the defined kernel function and the input feature vectors; And calculating the log marginal likelihood based on the covariance matrix, and solving the optimal super-parameters of the kernel function by an optimization algorithm with the aim of maximizing the log marginal likelihood to obtain the junction temperature prediction model.
9. The method for monitoring the junction temperature of the SiC MOSFET by using the mechanical stress wave according to claim 1, wherein the input eigenvector is composed of the acoustic eigenvector only and does not contain off current, and the junction temperature prediction model is a pure acoustic prediction model and is used for monitoring the junction temperature under the condition of no electric measurement assistance.
10. A mechanical stress wave monitoring system for SiC MOSFET junction temperature, the system comprising: The heating device is arranged on the surface of the SiC MOSFET device and is used for enabling the SiC MOSFET device to reach a plurality of different preset junction temperatures; The regulating circuit is used for enabling the SiC MOSFET device to obtain a plurality of different turn-off current values in a plurality of switching transients through regulating parameters at each preset junction temperature, and each switching transient corresponds to one turn-off current value; the acoustic sensor is used for collecting mechanical stress wave signals excited by each switching transient; the current acquisition unit is used for synchronously acquiring the turn-off current corresponding to each switching transient state; The characteristic extraction unit is used for extracting acoustic characteristic vectors at least comprising logarithmic root mean square, peak-to-peak value and kurtosis from mechanical stress wave signals under different preset junctions and different off-states, and combining the acoustic characteristic vectors with corresponding off-currents to form input characteristic vectors; the model construction unit is used for taking the input feature vector as input and corresponding preset junction temperature as output, and establishing a mapping relation between the input feature and the junction temperature to obtain a junction temperature prediction model; And the real-time monitoring unit is used for collecting the current mechanical stress wave signal and the current turn-off current of the SiC MOSFET device to be tested through the acoustic sensor and the current collecting unit in the real-time monitoring stage, calling the feature extracting unit to extract the current acoustic feature vector, combining the current acoustic feature vector with the current turn-off current, inputting the combination of the combination with the current turn-off current into the junction temperature prediction model, and outputting the current junction temperature predicted value of the SiC MOSFET device to be tested through the junction temperature prediction model.

Description

Mechanical stress wave monitoring method and system for junction temperature of SiC MOSFET Technical Field The invention belongs to the technical field of semiconductor device detection, and particularly relates to a mechanical stress wave monitoring method and system for junction temperature of a SiC MOSFET. Background Silicon carbide (SiC) MOSFETs have become core devices of high-power-density power electronic systems such as electric vehicles and smart grids due to their excellent characteristics of high switching speed, high withstand voltage, low on-loss, and the like. However, under long-term high-frequency switching and high-voltage high-current stress, material fatigue and aging easily occur inside the device due to thermal stress concentration, resulting in an increased risk of failure. Therefore, accurate junction temperature on-line monitoring is a technical basis for realizing active thermal management and guaranteeing long-term reliable operation of the SiC MOSFET. At present, the junction temperature monitoring technology of the power device is mainly divided into three types: (1) And estimating junction temperature by establishing a thermal resistance-capacitance network of the device package. The method is low in cost and easy to implement, but the accuracy is seriously dependent on the accuracy of model parameters and the accuracy of real-time loss calculation. Because of the temperature dependence of the thermal conductivity of the packaging material, the thermal network parameters can fluctuate widely in a wide temperature range, resulting in significant estimation errors. (2) The temperature-sensitive electrical parameter method (TSEP) is to use the electrical parameters (such as conduction voltage drop, switch delay time, etc.) which are strongly related to the temperature to reversely push the junction temperature. The method has high sensitivity and high response speed, and is a mainstream technology at present. The method has the inherent defects that the measurement circuit and the main power circuit are electrically coupled, a complex electrical isolation design is needed, and meanwhile, the requirement on the bandwidth and the precision of the sampling circuit is high, so that the system cost and the implementation difficulty are increased. (3) In recent years, the research discovers that the instantaneous power loss generated by the switching transient state of the power device can form a huge temperature gradient inside the chip, and the mechanical stress wave is excited and propagates outwards based on the thermoelastic effect. Theoretically, junction temperature can be estimated indirectly by analyzing the stress wave signal, with the natural advantages of non-invasive and electrical isolation. However, the existing monitoring scheme based on mechanical stress waves has two major bottlenecks in practical application: First is the thermal degradation problem of the hardware acquisition side. According to the traditional scheme, the acoustic emission sensor is directly attached to the surface of a high-temperature device, the pyroelectric effect is induced after heat is conducted to the piezoelectric element, meanwhile, the interface acoustic couplant is invalid, so that the sensitivity of the sensor is irreversibly and non-linearly attenuated, and the stable mapping relation between the amplitude of stress wave and junction temperature is thoroughly destroyed. And secondly, the problem of single characteristic of the data processing end is solved. The mechanical stress wave is the dynamic modulation result of the coupling action of the electric-thermal-mechanical multi-physical field intensity, the working current determines the loss power, the thermal shock intensity is influenced, and the junction temperature influences the waveform form by changing the Young modulus and the damping coefficient of the material. However, the prior art has mostly used conventional acoustic emission detection concepts, where only a single energy feature (such as signal energy or root mean square value) is extracted and temperature estimation is performed in combination with a simple polynomial fit. The extreme dimension reduction process compresses rich waveform distortion information caused by transient thermal shock gradient and material high-temperature softening, so that a single-feature model cannot decouple nonlinear cooperative modulation action of current and temperature on mechanical stress waves, and under actual working conditions, junction temperature estimation errors are larger, so that engineering application requirements are difficult to meet. In summary, the prior art cannot solve the problems of thermal degradation of the sensor and the characteristic single decoupling current-temperature nonlinear relationship, and a method for on-line monitoring of junction temperature of the SiC MOSFET with high precision, high anti-interference performance and complete electrical isolation