Setting of SOI MOSFET Model Parameters based on BP Neural Networks Algorithm

Abstract

With development of SOI technology, SOI MOSFET technology is developed as well. Most traditional MOSFET parameter models adopt semi-empirical and semi-physical model, and simplification assumption is introduced during modeling. However, as SOI MOSFET devices keep downsizing, the short-channel effect and quantum effect are more obvious so that it is more complicate to calculate and extract characteristic parameters of SOI MOSFET. This paper proposes a kind of SOI MOSFET characteristic parameter modeling method based on BP neutral networks algorithm. Compared to other semi-empirical models, this method needs not to calculate characteristic parameters of devices. In stead, it calculates current and voltage output characteristics and transfer characteristics of devices through BP neutral network models according to test data. Through verification, the trained and predicted output relative error is within 5%. The model has short operation time, high calculation precision and good stability. The models established may be applied extensively to other types of transistor, and feasible for practical engineering application.