In Silico Study of Degradation in Host Molecular Materials for OLEDs

We investigate degradation in three OLED host materials: TBSO (4-(m-tolylsulfonyl)-1,1-biphenyl), DPPCarba (9-(3,5-diphenylphenyl)carbazole), and DPPTriphen (2-(3,5-diphenylphenyl)triphenylene), using static electronic-structure calculations, non-adiabatic molecular dynamics (NAMD), and machine-learning-accelerated NAMD (ML-NAMD). Static analysis identifies DPPTriphen as having the most favorable overall stability profile, whereas TBSO shows the greatest degradation susceptibility, with S–aryl cleavage emerging as an important pathway. All three hosts are also vulnerable under high-energy exciton–exciton annihilation conditions. Direct NAMD shows that TBSO can undergo bond dissociation on a sub-picosecond timescale. A fully connected neural-network potential reproduces the reference data with 𝑅$^2$ =0.999 for energies and 𝑅$^2$ =0.998 for forces, while providing an approximately 4.2 ×10$^3$ speedup over direct NAMD. This enables 3519 trajectories to be propagated to 1 ps and yields 29 degradation events from the first excited state. These results show that static descriptors alone do not fully capture degradation propensity and that explicit excited-state dynamics provides complementary mechanistic insight for designing more stable OLED host materials.