運用機器學習輔助熱激活延遲螢光材料開發

Abstract

本研究針對有機發光二極體（OLED）中的熱激活延遲螢光（TADF）材料進行深入探討以提升元件的外部量子效率（EQE），傳統的TADF材料開發方法主要依賴於合成實驗，耗時且成本高，因此本研究引入機器學習技術輔助材料開發，提供更高效且系統化的策略。本研究建構了涵蓋2000年至2023年期間TADF材料數據的資料庫，包括材料結構、光電性質（如激發態能階、速率常數等）及元件表現等數據，並基於此資料庫採用梯度提升回歸演算法分別建立三種模型，首先是性質特徵與元件表現的預測模型，基於光電性質數據建立了EQE的預測模型，並通過特徵重要性分析找出影響EQE的關鍵性質，研究結果顯示光致發光量子產率(ηPL)是影響 EQE 最顯著的性質，顯示出提升ηPL對於優化OLED效率的重大意義。接著是分子結構與性質特徵的預測模型，由於前述模型篩選出ηPL做為關鍵特徵，因此透過結構描述符將材料導入模型進行訓練，並進行關鍵片段的篩選，篩選出影響輻射衰變與非輻射衰變的關鍵結構片段，例如B_Aromatic、Spiro 與 Triazine 等正向片段，為設計高效能的放光材料提供了指導。最後是分子結構與元件表現的關係模型，為了建立更加直觀的模型，我們進一步將結構導入模型並建立 EQE 預測模型，最終不僅重點分析了正向片段如 B_Aromatic、 Spiro 與 Triazine 等對於分子間的正向影響，也探討如 Methyl_Group 與 tButyl_Group 對分子間激子猝滅現象的抑制作用，並且分析負向片段的劣勢及其修飾策略，為提升整體元件效率提供了解決方案。本研究系統性分析了TADF材料結構、光電性質與元件表現之間的關聯，並提供了準確的預測工具和高表現的材料設計策略，為實驗端提供快速且可靠的材料篩選和優化指南。

This study investigates thermally activated delayed fluorescence (TADF) materials in organic light-emitting diodes (OLEDs) to enhance external quantum efficiency (EQE). Traditional development methods often rely on costly and time-consuming synthesis experiments, limiting efficiency and scalability. To address this, the research integrates machine learning techniques to create a more systematic and effective approach to material development. Database was established, covering TADF material data from 2000 to 2023. This database includes structures, optoelectronic properties and device performance metrics like EQE. Using gradient boosting regression algorithms, predictive models were developed to analyze and optimize the relationship between these variables. The results highlight photoluminescence quantum yield (ηPL) as the most critical factor influencing EQE, underscoring its significance in improving OLED efficiency. Further analysis using structure descriptors identified key structural fragments, such as B_Aromatic, Spiro, and Triazine, that enhance radiative efficiency and suppress non-radiative decay, providing actionable guidance for designing high-performance emissive materials. By directly linking molecular structures to EQE performance, the study also revealed the positive impact of structural fragments like B_Aromatic and Triazine on device performance. In addition, it examined the inhibitory effects of fragments such as Methyl_Group and t-Butyl_Group on exciton quenching, proposing modification strategies to mitigate the influence of negative fragments. This study systematically analyzes the relationships between TADF material structure, optoelectronic properties, and device performance. It offers predictive tools and actionable insights, providing a rapid and reliable guide for material screening and optimization in experimental research.