Deep Neural Network RF Model for SiGe HBT in Breakdown Regime

A silicon germanium heterojunction bipolar transistor (SiGe HBT) model based on the deep learning technique in the breakdown region is first proposed. The avalanche model with a deep neural network (DNN) is developed using back propagation and adaptive moment estimation (Adam) optimization methods. The results from the deep learning technique, measurement, and theoretical calculation in the breakdown and active region can achieve very good agreement. Compared to the conventional shallow neural network, the mean square error (MSE) of this technique is lower. This DNN can predict accurately measured S-parameters even for the significant change of the imaginary part of the output reflection coefficient from capacitive to inductive characteristics in the avalanche region. This method for breakdown characteristics builds the multibias S-parameter behavioral model in the range of practical interests. The accurate predicted results comparable to measured data are used for radio frequency (RF) circuit application to reduce measurement data points for the SiGe HBT.