Machine-Learning-Guided Discovery of Boron-Free Narrowband Blue Thermally Activated Delayed Fluorescence Emitters

Thermally activated delayed fluorescence (TADF) emitters enable color-pure, energy-efficient organic light-emitting diodes (OLEDs), yet many ultra-narrowband designs rely on synthetically demanding boron incorporation. Here we discover boron-free emitters that reconcile a robust nitrogen-only polycyclic scaffold, Rec. 2020-grade pure-blue emission, and small singlet-triplet gaps (ΔE_ST) conducive to TADF. We enumerated 19 518 13-ring carbazole-containing polycyclic aromatic hydrocarbons (Cz-PAHs) and obtained quantum-chemical reference values for ∼1000 molecules: emission energies and ΔE_ST at S₁-optimized geometries. A graph neural network enabled library-wide prediction and revealed a pronounced negative correlation between emission energy and ΔE_ST (R = -0.75), implicating configuration mixing in S₁ and indicating that small-gap candidates are enriched in the pure-blue regime. Two selected Cz-PAH emitters were synthesized and display deep-blue, exceptionally narrow photoluminescence (full widths at half maximum 17-19 nm) with small experimental ΔE_ST (0.13-0.19 eV). OLEDs incorporating these emitters deliver narrowband blue electroluminescence (λ_EL 456 and 481 nm); device A reaches chromaticity coordinates of (0.139, 0.069), close to the Rec. 2020 blue primary, whereas the sensitized device D achieves a maximum external quantum efficiency of 35.2%. Collectively, this work expands the boron-free TADF design space and establishes an experimentally validated route to discovering high-color-purity organic emitters.