Machine Learning-Assisted Development of Multi-Component Organic Photovoltaics via High-Throughput In-Situ Formulation

Abstract

Organic photovoltaics (OPVs) have witnessed in next generation energy source due to their outstanding potentials such as light-weight, flexibility, semi-transparency, color-tunability and roll-to-roll (R2R) processability. In addition, OPVs have recently achieved great progress in power conversion efficiency (PCE) of >18%. One key breakthrough is an emergence of non-fullerene acceptors (NFAs), which allows easy tuning of energy level compared fullerene counterparts, give opportunity to explore high open-circuit voltages. A development of ternary system is further contributed to high performance of NFA-based OPVs, which enhances short-circuit current density from complementary absorption of two different donors or acceptors absorption region. Despite of such great advantages and achievements, current manufacturing technology known as one variable at a time experimentation (Edonesian) still has remained far behind the expectations in terms of time consuming and human resource. Therefore, high-throughput experimentation approach is highly in demand. This thesis covers NFA-based ternary OPVs and their applications with a new experimental approach; Firstly, a ternary combination consists of PTB7-Th, IEICO-4F and two simple NFAs based on a bithiophene core with rhodanine end-groups (T2-ORH and T2-OEHRH) were explored and their photovoltaic properties were systematically investigated. PTB7-Th and IEICO-4F are generally known as narrow band gap donor and acceptor and two NFAs retain ultra-wide ban gap, hence, the ternary systems were further utilized to achieve controllable device coloration. We successfully demonstrated blend films with tunable colors including cyan → blue → purple → reddish purple colors, which were controlled by the ratios of IEICO-4F:T2-ORH or IEICO-4F:T2-OEHRH with PTB7-Th. Additionally, optical properties of blend films were studied via absorption and transmission measurements, while the range of colors achieved was quantified using CIE chromaticity and CIELAB color space then represented as RGB color models. Next, we introduced a new research approach to develop OPVs via industrial R2R slot die coating in conjunction with in-situ formulation technique and machine learning (ML) technology. Various PM6:Y6:IT-4F ternary blends, one of the highest performing ternary systems to date, are formulated in-situ and deposited on continuously moving substrates resulting in high-throughput fabrication of OPV with various compositions. The system is used to produce training data of ML technology. Composition/deposition parameters, referred as deposition densities, and efficiencies of 2218 devices are used to screen ML algorithms and to train an ML model based on Random Forest regression algorithm. Generated model is used to predict high-performance formulations and the prediction is experimentally validated resulting in 10.2% efficiency, the highest efficiency from R2R processed OPVs to date.