�鶹��Ƶ

Although HOIS perovskites have been the subject of significant recent research our ability to accurately predict the HOIS material structure based on the choice of building blocks used is extremely limited. Experience and intuition currently guide the development of new perovskites. The mission of this project is to harness the power of machine learning, in combination with advanced computer modeling, state-of-the-art synthesis, and characterization to codify and understand the physical properties that determine structural arrangement and produce reliable and effective predictive models that will guide the synthesis of new HOIS perovskites based on building block selection.

Perovskite is a calcium titanium oxide (CaTiO₃) mineral, first discovered in the Ural Mountains of Russia (in 1839) and named after the Russian mineralogist Lev Perovski. The arrangement of atoms in this mineral were first observed in 1926 by Victor Goldschmidt. In the original mineral a combination of metal ions (positively charged atoms) are surrounded by oxygen ions (negatively charged atoms) in a three-dimensional framework. This arrangement creates a repeating pattern that looks like a cube, with the larger metal ions in the middle of the cube, and the smaller oxygen ions at the corners. In modern research the oxygen atoms have been replaced with halides, such as bromine or iodine, hence the name metal-halide perovskites.

Hybrid organic-inorganic structures (HOIS) of metal-halide perovskites are a special class of these materials. The organic component is made up from carbon-based molecules, that are relatively flexible and through structural modification can provide different physical properties. The inorganic component is made up from metal and non-metal ions and imparts the perovskite structure. By combining these two systems into a hybrid material we can benefit from the performance of the perovskite and tune the properties of the material through the modularity of the organic component.

The size, shape, atomic arrangement, and charge character of the organic molecules in HOIS have been shown to directly control the structure of the inorganic perovskite component. Perovskites can form 1D structures (think of the cubes lined up in single file along an x-axis), 2D structures (the cubes extend out into a layer, along both the x- and y-axis), and 3D structure (the 2D layers of cubes are stacked one on top of the other, going up the z-axis), all of which have different properties and performance. Despite the existing extensive research in this area, and thousands of HOIS perovskites already synthesized, our ability to accurately predict the HOIS from the choice of organic cation and metal halide building blocks is extremely limited.

The goal of this DMREF project is to take the process of HOIS perovskite formation from where it is today, one of intuition-led trial and error, to one based on predictive models and structural understanding that will enable accurate and fast prediction of HOIS perovskite materials based on the building blocks chosen. This goal is only possible by adopting a collaborative approach that assembles the right expertise. The team will use a combination of computational molecular simulation and machine learning (led by Hendrik Heinz at CU Boulder) that will first be taught through an existing database of HOIS perovskites and further refined by experimental structures. Crucially, the accuracy and reliability of this combined computational approach will be validated, verified, and upgraded through the iterative synthesis (Seth Marder at CU Boulder and David Mitzi at Duke) and characterization (Raúl Casteñeda at NMHU) of proposed structures, a process that will augment the machine learning process and enable the development of accurate predictive models. The tunability of the organic component of these materials will form the basis for these studies. The size, shape, flexibility, polarizability, and way in which neighboring organic molecules interact with each other can all be designed and distilled down to a series of descriptors that can be experimentally modulated, and computationally simulated.

Hendrik Heinz, the project lead, is excited about this team approach: “This project is a great opportunity to get know the team members, listen to, and develop scientific ideas through the course of this project, and do great science together. We have a fantastic combination of expertise, the research topic is currently receiving high attention in the materials research community, and we are well positioned to make pioneering contributions.” David Mitzi, who has been studying HOIS perovskites since the 1990s, remarks that “it has been a long-standing dream to be able to dial in desired structures and properties by independently selecting organic and inorganic HOIS components, and I am excited about the prospects of working together to address this grand challenge.”

Parametrization of how the structural features of the building blocks influence the macro structure of the material will enable the development of predictive models through a machine learning process that is verified by experimental findings and refined. Ultimately this project will develop a set of tools and workflow that is expected to significantly accelerate and expand the field of HOIS perovskites. Understanding the factors that influence perovskite materials structure will enable advances in the performance and stability of these exciting materials.

Heinz expects there to be a wide range of benefits that emerge from this program “The integration of expertise, multimodal information, uncertainties, and validation is essential to make progress up to 10 times faster. Another critical benefit of working in this team is workforce training. The integration of different views and approaches in team meetings, co-advising of students, and exchanges of students from NMHU (a Hispanic-Serving Institution and Native American-Serving Nontribal Institution) with partner institutions will prepare the trainees for working in multidisciplinary teams, appreciating different opinions, writing more impactful papers, and gaining competitiveness for exciting careers after graduation or completion of postdoctoral appointments.” Marder, who has worked with collaborators at NMHU for 20 years, says “It is exciting to have NSF support to build upon our previous collaborations (which included Dr. Casteñeda when he was a graduate student at NMHU), so we can partner to create opportunities for a next generation of students at NMHU as well as at CU-Boulder and Duke.” Working together as a team to train researchers is a key part of this program, Casteñeda adds “From our previous experience it is very motivating for NMHU students participating in cutting edge research, and this typically results in under-represented students applying to Ph.D. programs or working in a national laboratory such as Los Alamos.”