Synergic Exploitation of TCAD and Deep Neural Networks for Nonlinear FinFET Modeling

Accurate large-signal (LS) modeling of Fin Field-Effect Transistors (FinFETs) plays an important role in designing microwave circuits for the next generation of communication systems and quantum sensing. In this work we propose preliminary results on a novel approach to LS FinFET modeling, that translates the X-parameters from physical TCAD analysis into deep neural networks (DNNs). The proposed method includes two phases. First, the X-parameters of nonlinear active device are extracted trough accurate TCAD physical simulations. Then, a long short-term memory (LSTM)-based DNN is employed for ANN modelling, to reproduce the scattered waves for any given incident waves up to the 5th harmonic. Similarly to X-parameters, the proposed DNN model simulates the transistor behavior around the large-signal operating point. Unlike the original X-parameter method, though, the DNN approach can incorporate the dependency on bias or other technological and physical parameters in a seamless and numerically efficient way. Hence, once implemented into circuit simulators, it allows for faster and more accurate circuit design.