Exciton diffusion in amorphous organic semiconductors: Reducing simulation overheads with machine learning.

Area of Science:

• Computational materials science
• Organic electronics
• Machine learning applications

Background:

• Simulations of exciton and charge hopping in amorphous organic materials require numerous physical parameters.
• Calculating these parameters via ab initio methods presents a significant computational overhead, hindering large-scale material studies.
• Existing machine learning approaches for parameter prediction often involve lengthy training times, adding to the overall simulation burden.

Purpose of the Study:

• To develop a new machine learning architecture for predicting intermolecular exciton coupling parameters.
• To reduce the training time for predictive models compared to traditional methods like Gaussian process regression or kernel ridge regression.
• To demonstrate the efficacy of this architecture in accelerating exciton diffusion simulations in amorphous organic materials.

Main Methods:

• Development of a novel machine learning architecture designed for faster training.
• Building a predictive model using the new architecture to estimate intermolecular exciton coupling parameters.
• Utilizing the predicted parameters in an exciton hopping simulation for amorphous pentacene.
• Comparing simulation results with those obtained using parameters computed via density functional theory.

Main Results:

• The developed machine learning architecture significantly reduces model training time.
• Exciton hopping simulations using predicted coupling parameters achieve excellent accuracy for exciton diffusion tensor elements and other properties.
• The simulation results closely match those obtained using parameters derived from ab initio calculations (density functional theory).

Conclusions:

• The novel machine learning architecture effectively reduces computational overhead in simulating exciton and charge diffusion in amorphous organic materials.
• This approach offers a faster and computationally efficient alternative to traditional methods for parameter calculation.
• Machine learning, particularly with optimized architectures, holds significant potential for advancing the study of charge and energy transport in organic materials.