Data-Driven Design Rules for TADF Emitters from a High-Throughput Screening of 747 Molecules

TL;DR

This work tackles the design of TADF emitters by performing a validated high-throughput virtual screening of 747 molecules to uncover universal, quantitative design rules that unify thermodynamic and kinetic requirements. It reveals a clear architectural hierarchy favoring D–A–D and MR-TADF systems, identifies a donor–acceptor torsional window of $50^{\circ}$ to $90^{\circ}$ as optimal, and demonstrates the fundamental trade-off between reducing the singlet–triplet gap $ΔE_{ST}$ and maintaining oscillator strength through HOMO–LUMO overlap. Data-driven clustering delineates distinct molecular families, with MR-TADF forming a blue-emission paradigm, and the study provides an actionable shortlist of 127 candidates predicted to have $ΔE_{ST} < 0.1$ eV and $f > 0.1$. The authors also present a framework for leveraging the HTS dataset in machine learning to accelerate discovery, offering concrete guidelines for synthetic targets and paving the way for rapid, predictive TADF emitter design. These insights bridge thermodynamic and kinetic design principles, enabling accelerated, rational discovery of next-generation TADF materials.

Abstract

The rational design of thermally activated delayed fluorescence (TADF) emitters is hindered by a complex interplay of thermodynamic and kinetic factors. To unravel these relationships, we performed a comprehensive computational analysis of \num{747} experimentally known TADF molecules to establish large-scale, quantitative design principles. Our validated semi-empirical protocol systematically reveals how molecular architecture, conformational geometry, and electronic structure govern photophysical properties. We establish a clear performance hierarchy, with Donor-Acceptor-Donor (D-A-D) architectures being statistically superior for minimizing the singlet-triplet energy gap ($ΔE_{\text{ST}}$). Crucially, we identify an optimal D-A torsional angle window of \qtyrange{50}{90}{\degree} that resolves the key trade-off between a small $ΔE_{\text{ST}}$ and the non-zero spin-orbit coupling (SOC) required for efficient reverse intersystem crossing (RISC). Data-driven clustering further identifies a distinct family of high-performance candidates and confirms Multi-Resonance (MR) emitters as a unique paradigm for high-efficiency blue emission. These findings culminate in a set of actionable design rules and the identification of \num{127} high-priority candidates predicted to have $ΔE_{\text{ST}}< \qty{0.1}{\electronvolt}$ and oscillator strength $f \num{> 0.1}$. This work provides a data-driven framework that unifies thermodynamic and kinetic principles to accelerate the discovery of next-generation TADF emitters.

Figures (5)

• Figure 1: Principal component analysis score plots for the 747.0 TADF emitters in the gas phase. The tight clustering of data along the first two principal components, which collectively capture 71.9 of the variance, illustrates the low intrinsic dimensionality of the TADF property space. This finding suggests that the complex design challenge can be rationalized by optimizing a few key orthogonal properties.
• Figure 2: TADF performance stratified by molecular architecture. The box plots show the distribution of the singlet-triplet energy gap ($\Delta E_{\text{ST}}$) for each class. D-A-D architectures demonstrate statistically superior characteristics (lower median and narrower distribution of $\Delta E_{\text{ST}}$) compared to other systems, confirming the efficacy of this design principle for consistently achieving small energy gaps.
• Figure 3: K-means clustering (k=4) of the 747.0 TADF emitters based on their standardized photophysical properties in both the gas phase (a) and toluene solvent (b). The analysis partitions the dataset into distinct families, most notably identifying a cluster of high-efficiency candidates (Cluster 1) with desirable properties for TADF applications.
• Figure 4: Validation of the optimal torsional angle design rule. The bar chart shows the percentage of molecules that are efficient TADF emitters ($\Delta E_{\text{ST}}$$<$0.3) within the optimal (5090°) versus suboptimal torsional angle ranges. Molecules in the optimal range exhibit a statistically significant ($p < 0.001$) increase in TADF efficiency.
• Figure 5: Correlation analysis between HOMO-LUMO overlap and key photophysical properties. The strong negative correlation with $\Delta E_{\text{ST}}$ (left) validates the spatial separation model for TADF design, while the strong positive correlation with oscillator strength (right) highlights the fundamental design trade-off that molecular engineers must navigate.