GaN Power Device Based On A Neural Network Optimization Of Thermal Performance Of Package

With the increasing demand for power,high frequency,high voltage and high integration of electronic devices,it is particularly important to provide a reliable power supply for electronic devices.Traditional power devices have shown initial weakness in high power performance requirements,which drives power devices towards the direction of higher power and higher reliability.As the third-generation semiconductor,GaN material has become a new material in the field of power devices because of its excellent material characteristics such as large forbidden band width,high mobility and high electron saturation.For the high-power type of GaN power device,its heat generation is large.If the packaging structure of the device does not have a proper heat dissipation way,the heat will be accumulated in the device for a long time,and the Schottky junction will deteriorate rapidly,resulting in device failure.Therefore,how to effectively enhance the heat dissipation capacity of devices has become a key problem in the field of GaN power devices.This thesis used the method of combining thermal analysis and intelligent algorithm to solve the cooling problem of GaN power device.First made the thermal simulation analysis of the device model,and got the simulation results affecting cooling factors as sample data.Then established the neural network prediction model based on MATLAB,and through the genetic algorithm for the best package parameter combination.Finally measuring temperature based on the optimal parameter combination verified the feasibility of the optimization method,so as to improve the cooling characteristics of GaN power device.The main research works are described as follows:1.For the thermal simulation analysis of GaN power device packaging structure,the GaN power device package structure model was established based on ANSYS finite element simulation software.Then,the thermal characteristic parameters of different packaging structures such as epoxy,plastic sealing materials and thermal sink materials were simulated and analyzed,the maximum junction temperature of the chip being taken as the evaluation standard,to calculate the thermal resistance and change trend of different components,and the feasibility of the thermal characteristic parameters was verified.Finally,the analysis determined four main factors affecting the heat dissipation of the device.And taking the maximum junction temperature of the chip as the evaluation standard,the sample database of 4 input parameters and 1 output parameters was established,which provides the sample data support for the establishment of the neural network model.2.The sample data was obtained through the established device thermal simulation analysis model.In the MATLAB environment,the BP neural network prediction model with the maximum junction temperature of the chip as the evaluation standard was established,and the maximum prediction error of the verified neural network prediction model was 1.32%.the error was within a reasonable range and the model could be used as a mathematical model for the maximum junction temperature prediction of the device chip.3.On the basis of establishing BP neural network model,which influencing device cooling package structure was optimized by the genetic algorithm and got the best package structure parameters,basing on the best package structure parameters for standard made corresponding physical and verified the feasibility of the method by measuring the device temperature that the maximum junction temperature error of the chip was 3.08%.And it improved the heat dissipation characteristics of GaN power devices.