Machine-learning-assisted performance improvements for multi-resonance thermally activated delayed fluorescence...

Area of Science:

• Organic optoelectronics
• Materials science
• Computational chemistry

Background:

• Multi-resonance thermally activated delayed fluorescence (MR-TADF) molecules are crucial for high-definition displays due to their color purity.
• Optimizing MR-TADF molecules is challenging due to limited chemical space and reliance on traditional expertise.

Purpose of the Study:

• To optimize the performance of the classic MR-TADF molecule, DABNA-1, using machine learning (ML).
• To rapidly explore the chemical space adjacent to DABNA-1 for novel high-performance molecules.

Main Methods:

• Generated adjacent chemical space of DABNA-1 via molecular morphing.
• Trained an ML model on a limited dataset to predict molecular properties.
• Evaluated electronic structures, reorganization energy, singlet-triplet energy gaps, and emission spectra of predicted molecules.

Main Results:

• ML identified top-performing molecules with excellent electronic structures, including small reorganization energy and singlet-triplet energy gaps.
• Analysis via molecular orbital (MO) theory elucidated improvements in electronic structures.
• Top molecules exhibited reduced vibronic peaks, leading to superior color purity compared to DABNA-1.
• Molecule M2 showed a high Reverse Intersystem Crossing (RISC) rate, indicating high efficiency potential.

Conclusions:

• The ML-assisted approach enables rapid optimization of existing MR-TADF molecules.
• This strategy addresses the need for efficient molecular design in organic optoelectronics.
• The findings pave the way for developing next-generation high-efficiency MR-TADF materials for advanced displays.