Non-Linear Modelling of HBTs Using Neural Network
Muhmmad Shah Alam and George Alastair Armstrong
This paper describes an artificial neural network (ANN) based non-linear modelling of HBTs. The neural network based model is concise when compared to the conventional modelling approach based on empirical equations and can demonstrate better accuracy. The framework for this ANN based model is a common-emitter large-signal equivalent circuit model. Intrinsic HBT model parameters voltage dependency of base-collector and base-emitter capacitances and resistances as well as base and collector currents is characterized by a 3 layered neural network, whose inputs are the base bias voltage Vb and collector bias voltage Vc. All extrinsic components of the HBT model are treated as bias independent. The corresponding “well-trained” neutral network has been found to give an excellent agreement for the base and collector currents and intrinsic parameters with the measured data and exhibits good extrapolation characteristics. The model has been implemented in Agilent ADS simulation environment and excellent agreement is obtained when compared with the results of bias dependent DC and S-parameters measurements. More significantly the neural network model predicts well the behaviour of higher order harmonics in a two tone test.