AI Accelerates OLED Material Discovery, Improving Design and Efficiency

Artificial intelligence is accelerating the discovery of organic light-emitting diode (OLED) materials, addressing limitations in conventional, empirical development methods. Recent advances, mirroring successes in drug discovery which have achieved up to 60% reductions in preclinical timelines, leverage machine learning and high-throughput virtual screening to navigate the vast chemical space – estimated between 1023 and 1060 possible compounds. Industrial platforms, such as Kyulux’s KyumaticTM system, are now exploring libraries exceeding 106 candidate molecules, integrating quantum chemistry calculations and machine learning to predict molecular properties and facilitate targeted design. This research is supported by funding from the Beijing and National Natural Science Foundations of China.

Advancements in OLED Technology

Over the past three decades, organic luminescent materials – the core components of OLEDs – have evolved from conventional fluorescent emitters to phosphorescent materials, thermally activated delayed fluorescence (TADF) systems, and, most recently, multiple resonance-type TADF (MR-TADF) materials. These advances have propelled theoretical internal quantum efficiency (IQE) towards 100%, with external quantum efficiencies (EQE) exceeding 20% across red, green, and blue devices. However, simultaneously achieving high efficiency, long operational lifetimes, and precise colour purity remains a significant challenge, limiting the full potential of OLED technology.

The conventional paradigm of OLED material development is predominantly empirical, relying on expert intuition and established molecular design principles. This approach is time-consuming, costly, and struggles to explore the vast chemical space – estimated to encompass between 1023 and 1060 theoretically possible compounds – required for modern display technologies. Current methodologies typically involve modifying known molecular scaffolds or manipulating core structures to tailor specific photophysical properties, though these incremental strategies are insufficient to meet the increasingly exacting requirements of ultra-high-definition displays.

Data-driven methods, particularly those powered by artificial intelligence (AI), represent a potential solution. Considered the fourth paradigm of scientific research, AI leverages vast datasets and advanced computational tools to address complex problems, demonstrated by recent breakthroughs in neural networks and AI-driven protein structure prediction, recognised with the 2024 Nobel Prizes in Physics and Chemistry. AI-driven strategies offer an efficient solution by employing algorithms to navigate complex data and uncover structure-property relationships.

Advanced machine learning (ML) frameworks, including predictive algorithms and deep learning (DL)-based generative models, have revolutionized molecular design by enabling data-driven forward and inverse design processes. These AI-guided strategies have demonstrated transformative impact in areas such as battery materials and advanced catalysts, and have impacted organic luminescent materials discovery, yielding promising results in molecular design and property prediction. High-throughput virtual screening (HTVS) integrated with ML has significantly reduced the number of experimental candidates in TADF emitter discovery. Industrial platforms, such as Kyulux’s KyumaticTM system, leverage ML and HTVS to explore libraries exceeding 106 candidate molecules, shortening discovery timelines and increasing hit rates. Analogous strategies in pharmaceutical research have achieved up to 60% reductions in preclinical timelines by systematically narrowing experimental scopes via predictive modelling, suggesting similar potential for OLED materials like blue TADF emitters. Large-scale AI-driven screening strategies have also systematically mapped synthetically accessible inverted singlet-triplet gap molecules, identifying thousands of candidates and establishing molecular design rules.

A comprehensive framework for AI-driven OLED material design has been presented, integrating quantum chemistry calculations, ML strategies, and generative models to address challenges throughout the OLED materials discovery pipeline, from molecular representation and property labelling to material screening and inverse design. This framework integrates three essential components: quantum chemistry calculations tailored for organic luminescent systems, comprehensive ML strategies for accurate molecular property prediction, and advanced DL-based generative models coupled with high-throughput virtual screening and inverse design for targeted molecular design and discovery.

Quantum chemistry calculations constitute a fundamental component of the AI framework, providing a robust and scalable means of generating molecular descriptors often inaccessible experimentally. In OLED research, where acquiring large, high-quality experimental datasets is hindered by the complexity, cost, and time-consuming nature of molecular synthesis and device fabrication, first-principles methods provide an alternative. Machine learning (ML) has achieved remarkable success across a wide range of scientific and engineering disciplines. The process of constructing ML-driven property prediction models begins with acquiring high-quality datasets sourced from experimental measurements, theoretical calculations, or existing databases. Subsequent feature processing transforms molecular structures into a format suitable for AI algorithms.

The predictive models established form the computational foundation for AI-guided screening workflows. By enabling rapid property estimation across extensive chemical spaces, these models allow high-throughput virtual screening and generative design to move beyond purely empirical exploration, systematically prioritising candidate molecules based on predicted properties and facilitating data-driven material discovery.

The Rise of AI-Driven Material Discovery

A distinctive feature of this review lies in its cross-disciplinary perspective, drawing upon established methodologies from more mature AI-driven domains such as drug discovery and materials informatics. This approach facilitates the application of proven techniques to the specific challenges of OLED material discovery, potentially accelerating innovation in the field.

The authors declare no conflict of interest, ensuring the objectivity and reliability of the presented research. This work was supported by the Beijing Natural Science Foundation and the National Natural Science Foundation of China, acknowledging the funding sources that enabled the investigation. The authors also thank Dr. Wei Xu from the TCL AI Lab for valuable advice on molecular generation models, recognising the contributions of external collaborators. Yiming Shi conceived the review framework, conducted the literature survey, and wrote the initial manuscript, detailing the primary author’s contributions. Ming Sun and Haochen Shi provided suggestions on machine learning model development and algorithmic design, acknowledging their expertise in computational methods. Zhiqin Liang and Bo Qiao contributed to the work, completing the team responsible for the review’s creation.

Integrating Quantum Chemistry and Machine Learning This cross-disciplinary perspective draws upon established methodologies from more mature AI-driven domains such as drug discovery and materials informatics, facilitating the application of proven techniques to the specific challenges of OLED material discovery. The authors declare no conflict of interest, ensuring the objectivity and reliability of the presented research. This work was supported by the Beijing Natural Science Foundation and the National Natural Science Foundation of China, acknowledging the funding sources that enabled the investigation. The authors also thank Dr. Wei Xu from the TCL AI Lab for valuable advice on molecular generation models, recognising the contributions of external collaborators. Yiming Shi conceived the review framework, conducted the literature survey, and wrote the initial manuscript, detailing the primary author’s contributions. Ming Sun and Haochen Shi provided suggestions on machine learning model development and algorithmic design, acknowledging their expertise in computational methods. Zhiqin Liang and Bo Qiao contributed to the work, completing the team responsible for the review’s creation.

A Cross-Disciplinary Approach to OLED Design

The integration of quantum chemistry calculations with machine learning (ML) strategies represents a crucial advancement in OLED material discovery. Quantum chemistry calculations provide a robust and scalable means of generating molecular descriptors often inaccessible through experimental means, particularly valuable in OLED research where acquiring large, high-quality experimental datasets is hampered by the complexity, cost, and time-consuming nature of molecular synthesis and device fabrication. First-principles methods offer an alternative approach to data acquisition in these circumstances.

The construction of ML-driven property prediction models begins with acquiring high-quality datasets sourced from experimental measurements, theoretical calculations, or existing databases. Subsequent feature processing transforms molecular structures into a format suitable for AI algorithms, creating the computational foundation for AI-guided screening workflows. These predictive models enable rapid property estimation across extensive chemical spaces, allowing high-throughput virtual screening and generative design to move beyond purely empirical exploration.

By systematically prioritising candidate molecules based on predicted properties, these models facilitate data-driven material discovery. This approach allows for a more targeted exploration of the vast chemical space – estimated to encompass between 1023 and 1060 theoretically possible compounds – than traditional methods, addressing a key limitation of conventional OLED material development.