Abstract

We construct a large-signal artificial neural network (ANN) model for SiGe HBTs, directly from time-domain large-signal measurements. It is known that HBTs are very sensitive to self-heating and therefore we explicitly study the effect on the model accuracy of the incorporation of the self-heating effect in the behavioural model description. Finally, we show that this type of models can be accurate at extreme operating conditions, where classical compact models start to fail.