Automatic Identification of Traps in Molecular Charge Transport Networks of Organic Semiconductors

TL;DR

This paper presents a novel spectral clustering-based method to identify traps in molecular charge transport networks of organic semiconductors, enabling trap detection without explicit charge dynamics simulation.

Contribution

The proposed method uniquely combines eigenvalue analysis and spectral clustering to identify various trap types in molecular networks, improving over existing techniques.

Findings

Successfully identifies single and multiple traps in simulated networks

Applicable to diverse trap types without explicit charge dynamics

Works on systems with or without detailed balance

Abstract

This paper introduces a method to identify traps in molecular charge transport networks as obtained by multiscale modeling of organic semiconductors. Depending on the materials, traps can be defect-like single molecules or clusters of several neighboring ones, and can have a significant impact on the dynamics of charge carriers. Our proposed method builds on the random walk model of charge dynamics on a directed, weighted graph, the molecular transport network. It comprises an effective heuristic to determine the number of traps or trap clusters based on the eigenvalues and eigenvectors of the random walk Laplacian matrix and uses subsequent spectral clustering techniques to identify these traps. In contrast to currently available methods, ours enables identification of trap molecules in organic semiconductors without having to explicitly simulate the charge dynamics and is applicable to a variety of energy- or topology-based traps in homomolecular or mixed systems with or without detailed-balance. As a prototypical system we simulate an amorphous morphology of bathocuproine, a material with known high energetic disorder and charge trapping. Based on a first-principle multiscale model, we first obtain a reference charge transport network and then purposefully modify its properties to represent different trap characteristics. In contrast to currently available methods, our approach successfully identifies both single trap, multiple distributed traps, and a combination of a single-molecule trap and trap regions on an equal footing.