Research On The Behavior Model Of High Power Semiconductor Devices Based On Artificial Intelligence

RF/microwave circuits occupy an extremely important position in hardware circuits,and power semiconductor devices represented by GaN HEMTs are most widely used in microwave circuits.In the circuit simulation design,the model of each device in the circuit is needed,so establishing an accurate model of the power device is very important for the microwave circuit design.In recent years,with the development of big data,Artificial Intelligence has become a hot and promising research direction.As an effective method of artificial intelligence,Neural Network has become more and more concerned with the application to build device models.This paper mainly studies the behavioral model of the device using the Neural Network method in the field of Artificial Intelligence.When proceeding device model studies,they are often divided into small signal models and large signal models.The device used in this article is CREE's CGH40010.In terms of small-signal modeling,an equivalent circuit containing 15 components was selected for modeling based on device characteristics.This paper applies the substep method in the 15components equivalent circuit small-signal modeling.The parasitic component values and the intrinsic component values are obtained.The extracted component values are used to build the circuit in the ADS,and the tuning is added according to the actual parameters of the device.The model verification results are obtained.In the application of Neural Network to establish a small-signal behavior model of devices,a BP Neural Network is proposed to establish S-parameter behavior model.The established small-signal model can directly reflect the small-signal characteristics of GaN HEMT devices.Comparing it with the a small-signal equivalent circuit modeling,the BP Network eliminate the need to extract component values,which greatly simplifies the modeling process.And the results show that the obtained model also has higher precision.In the aspect of establishing large-signal model,a BP Neural Network is proposed to replace the traditional empirical formula modeling method.The BP Neural Network behavior modeling is carried out for the I_ds,C_gs and C_gd with the strongest nonlinear performance.Modeling using this method reduces the parameter extraction process and improves modeling efficiency.Because the model error of the BP Network is small,the modeling accuracy is improved.Finally,in the aspect of large-signal modeling,the paper also proposes to apply the Extreme Learning Machine to establish the X-parameter behavior model.Compared with the traditional BP Network,the Extreme Learning Machine has higher modeling efficiency which because the parameters in the network are obtained by calculation once,without iteration.This method is used to establish the X-parameter behavior model.The results show that the method is suitable for X-parameter behavior modeling and has higher modeling speed than BP Network.