A Knowledge-Guided Neural Network-Based Large-Signal Model for InP HBTs With Self-Heating Effect Consideration

An efficient and accurate knowledge-guided neural network (KGNN) modeling method for microwave active devices is proposed. The KGNN-based modeling method specifically constructs a hybrid combined objective function (COF) to constrain and guide neural network modeling, greatly improving modeling accuracy and consistency. In particular, the COF consists of the training error term, the prior knowledge term, and the regularization term. The proposed method can be directly used for large-signal modeling. To improve the modeling accuracy, the proposed method is used to model both the bias-and temperature-dependent isothermal nonlinear current and thermal factor, which is introduced to characterize the influence of the self-heating effect. The depletion capacitance and delay time parameters are also modeled with the proposed method. The base–collector and base–emitter charge formulations are obtained by integrating the corresponding depletion capacitances and delay times. An improved small-signal extraction method for III–V heterojunction bipolar transistors (HBTs) is used to obtain the extrinsic parameters. The developed large-signal model (LSM) is implemented in the advanced design system and fully verified by the small-and large-signal measurements. Good agreement is obtained between measured and simulated results.