Feature extractions from transfer characteristics of hybrid GO FETs for selective detection of volatile organic compounds

Abstract

The current work concerns a new strategy for selective classifications of multiple volatile organic compounds (VOCs) by using an array of field effect transistor (FET) type sensors and extraction of device-specific features from transfer characteristics of the FET-sensors to use in classification algorithm(s). A few layered graphene oxide (GO) was used as the base channel materials which was then functionalized very precisely with different metal oxides (WO3, TiO2) and metals (Au, Pd) nanoforms to construct an array of five FET structure sensors on SiO2/Si platform. The deviation in Id-Vgs characteristics in different VOCs were accounted with a distinctive features. The features were organized based on their importance score and used to construct different feature matrices. A minimum four sensors and six features were required to successful classification of seven VOCs by linear discriminant analysis (LDA). The classification accuracy was tested with supportive vector machine (SVM) classifier and it was 92.86%.