Deep learning based charge mobility prediction for organic semiconductors

Abstract

Electronic coupling is an imperative indicator that aid the determination of the charge mobility of organic semiconductor molecules, and it is highly dependent on molecular packing motifs. The quantum mechanical simulations done to determine the charge mobility are usually time-consuming and unaffordable, but fortunately, with the advances in deep learning techniques nowadays, the task has been accelerated much further. This paper conducts the study by extracting the spatial coordinates of dimers (molecular pairs) and arranging them into an overlap matrix, which will then be used to calculate the electronic coupling using quantum chemical calculations. The overlap matrix and electronic coupling will be used to predict the charge mobility accurately and effectively for pentacene by developing a deep learning model based on artificial neural networks (ANNs). The results achieved have a very high determination coefficient 𝑅2 of 0.97 and Mean Absolute Error (MAE) of 3.96 meV, showing its reliability. However, the current model can still be further refined to study the charge mobility of other organic molecules given more datasets.