Machine Learning-Aided Non-Functionalized Graphene Chemitransistors for Highly Selective Chemisensing in Liquid Media

The development of highly selective and miniaturized chemisensors for liquid media can enable newer discoveries and empower a wide range of disciplines from biology to human health, marine ecology to climate change, and industrial process monitoring to food safety. While the two-terminal chemiresistor architecture has been remarkably successful in this regard, they rely on appropriate surface functionalization to achieve selectivity for a specific target analyte. Furthermore, sensor drift and sensor-to-sensor variation in a multiplexed chemisensors array can severely impact their reliability and obscure the response of individual sensors to the target analytes. In this article, we overcome these challenges by introducing a new sensing paradigm that integrates an array of three-terminal chemitransistors based on monolayer graphene with an ensemble of machine learning (ML) algorithms. The extraordinary transport properties of graphene combined with its electrochemically inert basal plane allow a wide range of chemical species to electrostatically control the channel conductivity in a liquid medium through voltage gating. This in turn expands the parameter space for the sensing elements which enables the integration of various unsupervised and supervised ML algorithms to eliminate the need for surface functionalization, eradicate the need for sensor calibration, and minimize the impact of sensor-to-sensor variation. Our ML-aided, non-functionalized, and reusable graphene chemitransistors were found to be highly selective to de-ionized (DI) water, multiple chloride salts including NaCl, LiCl, CaCl2, and MgCl2 that are relevant to biosensing, several weak acids including acetic, propionic, and citric acid that are used in the food industry, as well as more complex solutions and processes such as decaying juices extracted from different fruits (grape, orange, watermelon, and pineapple).