Parameter extraction in thin film transistors using artificial neural networks

This work presents a method based on supervised learning for the extraction of parameters in Indium Gallium Zinc Oxide Thin-Film Transistors with aluminium contacts, as an alternative regarding analytical and optimisation methods. The method consists of generating a set of I-V curves of the device of interest using Spice software. These curves are the input samples of the Artificial Neural Networks, from which it is intended to predict the different parameters such as threshold voltage, transconductance and contact resistance, from each sample curve. By generating the training set itself, it is possible to label each sample curve, which allows the type of learning to be supervised. The results show that ANNs provide parameters with which it is possible to model physical measurements with error rates of less than 5% when extracting the first two parameters, and errors of between 0.06% and 4.62%, when extracting the three parameters. In addition, a comparison was made between the results of the ANNs and the analytical extraction of parameters.