An Efficient Variability-Aware Control Variate-Assisted Neural Network Model for Advanced Nanoscale Transistors

In this paper, a novel artificial neural network (ANN) has been developed for the efficient variation-aware modeling of current-voltage (I-V) characteristics of general nanoscale devices. The key innovation of this work lies in the development of a new control variate strategy to significantly shrink the number of technology computer-aided design (TCAD) device simulations required to train the ANN model. Consequently, the proposed ANN model can emulate the drain current of the target device as analytic functions of the device geometry, material, and bias voltages at much smaller computational costs than conventional ANN models. A validation example of a 14nm fin field effect transistor (FinFET) is provided in this paper.