High electric field stress model of n-channel VDMOSFET based on artificial neural network

Abstract

In VDMOSFETs (Vertical Double-Diffused Metal-Oxide Semiconductor Field-Effect Transistor), in the cases when the voltage at the gate contact comes close to the breakdown voltage, a high electric field (HEF) is formed in the gate oxide. This leads to the generation of fixed and mobile charged defects in the gate oxide and at the silicon/oxide interface. The mechanisms of defects formation are very complex, so it is impossible to create a unique physical model that could describe the behavior of components, i.e., its electrical characteristics, depending on the stress voltage and the stress time. In cases like this, the application of artificial neural network (ANN) has proved to be one of the acceptable solution. In fact, this approach can give the very accurate models for the define range of changes of the input parameters, without going into very complex physical and chemical processes that occur in the structure of the components. All that is necessary to generate the ANN model is the sufficiently large number of the measured data and correct selection of the neural network structure and appropriate training algorithm. In this paper, we have applied the multilayer feedforward neural networks to the model of high electric field stress effects (HEFS) in the n-channel VDMOS power transistor. Experimentally measured transfer characteristics for different times of stress were used to train the neural networks of different configurations. In addition, the influence of several different training algorithms on the accuracy of the model was studied. It is shown that a properly selected neural network structure and training algorithm provide the ANN model of HEFS of the n-channel VDMOS power transistor that very precisely gives its electrical characteristics (transfer characteristic and threshold voltage) for a given stress voltage and temperature in the whole range of the time stress changes from 0 to 150 min.