Collaborative research: accelerated discovery of lead-free metal halide perovskitoids using automation and artificial intelligence

The National Science Foundation Division of Materials Research awarded Fordham University $393,021 under the Solid State and Materials Chemistry Program on February 15, 2026, to accelerate discovery of lead-free metal halide perovskitoids using automation and artificial intelligence.

The project develops machine learning models and generative AI methods to guide automated laboratory synthesis and characterization of lead-free perovskites and perovskitoids—materials with applications in photovoltaic solar cells, lighting, lasers, photodetectors, X-ray detectors, and catalysis. Work includes extraction and analysis of published experimental data to identify gaps in chemical knowledge, construction of machine learning models to optimize synthesis efforts, development of robotic-compatible synthetic routes, and creation of automated AI-based assistants for multistep research planning. The team shares general-purpose AI tools developed through this work with the broader materials chemistry community. The project also supports workforce development through interdisciplinary training in data science and chemical experimentation for undergraduate, graduate, and postdoctoral researchers, creates undergraduate-level teaching resources for distribution to other institutions, and develops age-appropriate educational games teaching materials design principles.

Performance occurs in Bronx, New York, with a period of performance from February 15, 2026, through January 31, 2029. The assistance type is a Project Grant under the Mathematical and Physical Sciences directorate (CFDA 47.049).

Abstract

Non-technical summary: halide perovskites and perovskitoids are a broad family of materials that have many applications including photovoltaic solar cells, lighting, lasers, photodetectors, x-RAY detectors, and catalysis. However, many of the applications use materials containing toxic metals, such as lead. In this project, with support from the solid state and materials chemistry program in the mathematical and physical sciences directorate, automated laboratory experiments are guided by artificial intelligence to discover and characterize new lead-free perovskites and perovskitoids with novel properties. The development of novel methods for using generative AI to improve understanding is accomplished by the extraction of data from published experiments to articulate the current limits of chemical knowledge and to identify gaps in experimental efforts. These data are used to construct machine learning models that help optimally guide exploratory synthesis efforts. The development of novel robotic compatible synthetic routes enables iterative testing and refinement of the machine learning models. To aid in this, automated AI-based assistants are developed to assist in complex multistep research planning. The value of this approach for lead-free halide perovskites demonstrates their wider importance, general-purpose AI tools are shared with the broader materials chemistry community. Undergraduate, graduate, and postdoctoral researchers participating in this project develop interdisciplinary skills combining data science and chemical experimentation, addressing a scientific workforce GAP. Additionally, undergraduate-level teaching resources are created, all of which are shared with other institutions. To help further improve the pipeline of scientists, age-appropriate games that teach principles of materials design to kindergarten-level students are also developed, created and tested. Technical summary: with support from the solid state and materials chemistry program in the mathematical and physical sciences directorate, this project develops automated high-throughput synthesis methods based on antisolvent vapor diffusion, inverse temperature crystallization, and ionic-liquid based methods for lead-free halide perovskites and perovskitoids, with an initial focus on copper halides. Copper halides exhibit high stability and luminescence (serving many of the same optoelectronic applications as lead halides) and also have the potential of displaying novel physical properties including chirality-induced spin selectivity, magnonic response, enantioselective electrocatalysis, and quantum information processing applications. Active learning is used for efficient experimental parameter exploration, combined with large-language model (LLM)-based literature extraction to identify past precedents and identify gaps in scientific understanding. Agentic LLM workflow strategies are studied, developed, and evaluated for their ability to assist in exploratory chemistry research. Broader impacts include workforce development of undergraduate researchers that have both data science and experimental skills, and simplifications of the automated experiments that can be performed on frugal twin automation systems in an undergraduate teaching laboratory setting. Additionally, additive manufacturing is used to prototype and play-test toys that enable pre-k and kindergarten aged students to see, touch, and explore concepts of materials design based on the 2D-ising model. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the foundation's intellectual merit and broader impacts review criteria.- subawards are not planned for this award.