Machine learning enabled characterization of charge transport in organic semiconducting devices

Abstract

While organic semiconducting materials have found immense potential in device applications such as field-effect transistors and photovoltaics in the past few decades, the mechanism of charge transport in these devices has not been fully understood. A scrutiny of the existing models suggests that there is no unified approach to characterize transport in these materials. With machine learning algorithms that learn input-output relationships, device data can aid in the determination of transport parameters, offering physical insights about transport and material design. Furthermore, these techniques can aid in improving device figure-of-merit by screening for novel materials. In this thesis, we examine the extent of applicability of conventional theories on transport in semiconducting devices. We demonstrate different data-driven and physics-inspired approaches, independent of conventional theories, to characterise transport using fewer parameters. We also study structure at the molecular level by employing machine learning methods to predict thermoelectric properties of single molecular junctions.