Switching Characteristics Testing And Modeling Of Medium And High Voltage IGBT Power Module

Insulated-gate bipolar transistors (IGBTs) are widely used in medium and high power converter. Switching characteristics determine the switching losses, device voltage and current stresses, converter maximum switching frequency, power density and electromagnetic compatibility as well as thermal design, therefore the switching characteristics directly determine the performance and reliability of the converter. IGBT switching characteristics under actual operating conditions are closely related to their working environments. The switching characteristics of IGBTs are critical for power converter design, performance, efficiency and reliability.Because of the short switching time, high dv/dt and di/dt as well as serious electromagnetic interference, the measurement of high-voltage IGBT switching characteristics meets great challenges. The design criteria for the switching characteristics test system of IGBT voltage ratings from1200V to6500V have been proposed. With the criteria, a universal off-line switching characteristics test bench for medium and high voltage IGBTs has been developed. The test bench covers voltage ratings from1200V to6500V and wide current range of40A to1500A, applicable to a variety of topologies and package structures. The test bench realizes:1.adjusting working environmental parameters automatically according to user’s need;2.recording IGBT switching transients accurately;3.completing oscilloscope settings, repeated measurement and data preservation automatically;4.extracting comprehensive information of switching characteristics far more than the vendor-supplied datasheet;5.acquiring IGBT switching characteristics trends with environmental parameters for converter design, optimization and performance evaluation as well as device loss calculation;6.recording device fault transients and providing evidence for failure analysis.IGBT modules of different topologies, technologies and capacities have been tested on the test bench. Distribution trends of switching characteristics parameters with environmental parameters have been acquired. Based on experimental data:1.The influence of loop parasitic inductance on IGBT switching characteristics has been explored.2.An extraction method for parasitic inductance within the package has been proposed.3.The mechanism for IGBT dynamic failure during test has been analyzed.4.A comparison of IGBTs with the same capacity but different technologies has been presented.5.A new type of ANPC power module realized by reverse blocking IGBT has been tested and analyzed, and the device losses ANPC of and NPC3-level topologies have been calculated and compared. The result shows that the conduction losses of ANPC is lower than NPC while its switching losses is higher. This indicates that it has device losses advantage in a certain frequency range.The prediction and simulation of switching characteristics and device losses are extremely important for device manufacturers and users. Physical modeling is extremely complex while its accuracy is difficult to guarantee and device structure and process parameters can hardly be obtained, therefore it is not easy for device users. An error back-propagation multi-layer feed-forward neural network model has been established in this paper based on the experimental data from the test bench, realizing reliable prediction of IGBT switching characteristics as device stresses and switching losses under actual working conditions. Compared to physical modeling, the specified BP neural network prediction model for purchased IGBTs is convenient to use and easy to build. Environmental influences and IGBT physical mechanisms can be uniformly considered, avoiding explicit description of physical mechanisms.In this paper, an optimized switching characteristics artificial neural network prediction model has been further proposed. BP neural network algorithm has certain limitations. The number of hidden neurons, training goal, initial network weights and biases have great impacts on neural network performance, and designing network structure by using trial and error method may make the impacts uncontrollable. The optimizing process of BP algorithm is vulnerable to local minima, and when falling into local minima, the BP algorithm cannot escape from the pole. Global optimization algorithms are introduced:genetic algorithm has been used for hidden neuron number and training goal optimization; genetic algorithm, simulated annealing algorithm and particle swarm optimization algorithm have been applied for network initial weights and biases optimization. The global optimization algorithms have improved the performance of the BP neural network prediction model greatly.After training, the improved BP neural network prediction model has realized accurate predictions of IGBT hard switching performance such as switching time, switching losses, voltage overshoot and current spike under different environmental parameters:parasitic inductance, junction temperature, driving voltage and resistance, collector voltage and current. By high precision forecast, quasi-online analysis for actual working conditions as device losses and system efficiency calculation, device voltage and current stress evaluation, circuit parasitic parameter optimization, and reasonable dead-time design can be performed. It is convenient and has great guiding significance for engineers to design heat sink, circuit, structure and assess device and system performance.