Parameter Extraction from a Resistive Load Inverter Circuit Using Supervised Learning Methods

This paper presents a proposal for parameter extraction of a resistive load inverter circuit, with a Thin Film Transistor (TFT), using Artificial Neural Networks, Random Forest, Decision Trees and Support Vector Regression. Although analytical and optimization methods are usually used for this purpose, they have disadvantages such as the need for expertise or complex implementation. This work shows that these supervised learning methods are useful for this task because they can learn the parameters of the device transfer curves, obtaining a good fit between the measurements and the extracted parameters. The different methods were trained using a data set constructed from simulations performed with AIM-Spice software, where the parameters affecting different regions of the inverter characteristic curve were extracted. In the experimental stage, the Neural Networks obtained better results, with an average error rate of 6.04%. The method was also applied to real NMOS measurements and yielded minimum errors of up to 0.43%.