Accelerated computational discovery of high-performance materials for organic photovoltaics by means of cheminformatics

Abstract

In this perspective we explore the use of strategies from drug discovery, pattern recognition, and machine learning in the context of computational materials science. We focus our discussion on the development of donor materials for organic photovoltaics by means of a cheminformatics approach. These methods enable the development of models based on molecular descriptors that can be correlated to the important characteristics of the materials. Particularly, we formulate empirical models, parametrized using a training set of donor polymers with available experimental data, for the important current-voltage and efficiency characteristics of candidate molecules. The descriptors are readily computed which allows us to rapidly assess key quantities related to the performance of organic photovoltaics for many candidate molecules. As part of the Harvard Clean Energy Project, we use this approach to quickly obtain an initial ranking of its molecular library with 2.6 million candidate compounds. Our method reveals molecular motifs of particular interest, such as the benzothiadiazole and thienopyrrole moieties, which are present in the most promising set of molecules.

Lead candidates for high-performance organic photovoltaics from high-throughput quantum chemistry – the Harvard Clean Energy Project

“…It is designed and implemented to identify lead compounds, in particular organic semiconductors for photovoltaic applications. The CEP framework utilizes rst-principles electronic structure theory (augmented by techniques from cheminformatics/materials informatics and machine learning 5,6 ) to characterize millions of molecular motifs and assess their potential. The massive amount of computing time required for this research is provided by distributed volunteer computing by means of IBM's World Community Grid (WCG).…”

The Harvard Clean Energy Project: Large-Scale Computational Screening and Design of Organic Photovoltaics on the World Community Grid

“…In Ref. [97] we give an introduction to this approach with a detailed discussion of the systematic construction and optimization of descriptor models.…”

“…The early stages of this work [97] utilized descriptors from the Marvin code by ChemAxon [82] and for the modeling we employed the R statistics package [98]. Recently, we started using the more comprehensive descriptor set from Dragon [99] and the specialized modeling code StarDrop [100].…”

From the computer to the laboratory: materials discovery and design using first-principles calculations

“…This review already presented many examples of materials discovered through high-throughput computing (e.g., LiZnSb in thermoelectrics, BiPt in catalysis, a few new cathode materials in Li-ion batteries and a specific mixture of LiNH 2 -MgH 2 in hydrogen storage). Highthroughput computing studies have already been performed in numerous fields such as catalysis [170], hydrogen storage [180], Li-ion batteries [181,182], organic photovoltaics [183], thermoelectrics [114,184], scintillators [185,186], or photocatalysts [187]. In parallel, this large amount of computed properties is getting compiled in databases publicly available through web interfaces as the Materials Project and others [188][189][190].…”