A research paper titled "Inverse design workflow discovers hole-transport materials tailored for perovskite solar cells" was published in the journal Science by Christoph J. Brabec's team at the University of Erlangen-Nuremberg. Jianchang Wu is the first author, and Jianchang Wu, Luyao Wang, Sang Il Seok, Pascal Friederich, and Christoph J. Brabec are co-corresponding authors.

Key Highlights: This paper demonstrates a closed-loop workflow that combines high-throughput synthesis of organic semiconductors to create large datasets with Bayesian optimization to discover novel hole-transport materials with tailored properties for solar cell applications. Power conversion efficiencies as high as 26.2% (certified 25.9%) were achieved in perovskite solar cells.

The design of hole-transport materials (HTMs) for perovskite solar cells (PSCs) has primarily relied on experimentalists qualitatively identifying patterns in the HTM structure to improve device performance. However, this approach lacks mechanistic understanding of novel HTMs and requires pattern recognition on high-dimensional datasets. Machine learning (ML) has been used to discover meaningful patterns across various fields of science and technology, including organic synthesis, materials science, and process optimization. However, the discovery of new materials with optimized properties for semiconductor device functionality has not yet been applied to the emerging field of photovoltaics. Previous researchers have primarily focused on using ML to optimize manufacturing processes or predict device performance and stability based on manufacturing processes. Only recently have researchers combined ML with organic synthesis.

However, autonomous optimization algorithms require not only sufficient data volume but also data diversity, which necessitates the synthesis of structurally diverse molecules. To address these issues, we developed a joint knowledge- and data-based strategy implemented in a high-throughput (HT) organic synthesis platform. This strategy can synthesize and purify over 100 solution-processable small-molecule semiconductors with diverse structures and consistent quality within a few weeks and across multiple synthetic campaigns. This approach provides a sufficiently large and diverse dataset for training ML models that couple the structural features of HTMs to the properties of the corresponding p-i-n PSCs and for implementing data-based exploration strategies.

To address this issue, a team led by Jianchang Wu and Christoph J. Brabec of the University of Erlangen-Nuremberg, Pascal Friederich of the Karlsruhe Institute of Technology, Sang Il Seok of the Ulsan National Institute of Science and Technology in South Korea, and Luyao Wang of Xiamen University demonstrated a closed-loop workflow that combines high-throughput synthesis of organic semiconductors to create a large dataset with Bayesian optimization to discover novel hole-transporting materials with tailored properties for solar cell applications. A predictive model based on molecular descriptors enables the connection between the structure of these materials and their properties. From a minimal set of suggestions, a series of high-performance molecules were identified, achieving power conversion efficiencies of up to 26.2% (certified at 25.9%) in perovskite solar cells.

The study synthesized a library of conjugated organic molecules to create a large dataset and evaluated these molecules as hole transporters. A Bayesian model, trained on device performance, was used to generate new candidates based on the molecular descriptions. This inverse design approach has identified high-performance organic hole-transporting semiconductors for perovskite solar cells. This approach could be extended to other application areas and change the way we develop and optimize materials for cutting-edge technological applications.