Neural networks for large- and small-signal modeling of MESFET/HEMT transistors

The design of microwave and millimeter-wave circuits and the increasing integration of hybrid and monolithic circuits has reinforced the need of accurate large and small-signal device models to improve the performance of these circuits and to minimize the number of design and fabrication steps required. Therefore, it is very important for efficient CAD tools to have good modeling approaches able to predict the small and large-signal nonlinear dynamic behavior of GaAs devices, such as metal semiconductor field effect transistor (MESFET), or high electron mobility transistor (HEMT). Conventional nonlinear techniques applied to device modeling, such as closed-form equations [1,2], Volterra series [3], or the use of look-up tables [4], are difficult to implement in commercial simulators because of their high memory requirements or their computational burden. Recently, some attempts have been made to model the nonlinear behavior of active devices and circuits by using neural networks [5,6]. Neural networks have the capability of approximating any nonlinear function and the ability to learn from experimental data; therefore, they are good candidates to solve device-modeling problems. However, practically all these neural approaches only consider the use of the Multilayer Perceptron (MLP) and, in this case, the memory requirements to give a good approximation, and the computational requirements to carry out the training of the network are high. In order to avoid these problems we have proposed two models: the Generalized Radial Basis Functions (GRBF) network [7], for small-signal modeling, and the Smoothed Piecewise Linear (SPWL) model [8] for largesignal regimes. Both models require a low number of parameters and, at the same time, their computational requirements to train the models are lower than those of the above mentioned methods (including the MLP). The paper is organized as follows. In Section 2 the problem of modeling microwave transistors is stated. In Section 3 the GRBF and SPWL models are described. In Section 4 the main results obtained are presented, and in Section 5 a global model is proposed, which characterize the whole device behavior. Finally, in Section 6, the main conclusions are exposed.