Data Driven Composition and Octahedral Engineering of Halide Perovskites

Halide perovskites are highly desired for optoelectronic applications, because of the excellent tunability of their properties from composition engineering and octahedral distortion and rotation. However, the infinite possibilities of compositions via cation and/or anion-site mixing and octahedral manipulation raise considerable challenges when exploring the stability and properties of halide perovskites, both experimentally and computationally. In this work, we address this issue using high-throughput density functional theory (DFT) computations and discover trends and design rules to aid prediction and screening across large chemical spaces.<br/><br/>We generate a large DFT dataset of ABX3 halide perovskite alloys with several elemental and molecular choices at A, B and X sites, and arbitrary mixing allowed on each site.¹ This dataset covers ~ 500 compounds across a 14-dimensional ionic space and is currently restricted to alloying with pseudo-cubic 2x2x2 supercells. To further explore property tuning, we create hundreds of random distorted and rotated structures for Cs/MAPbX3 perovskites (MA = methylammonium, X = I/Br/Cl). Many critical properties are calculated using standard GGA-PBE and hybrid HSE06 functionals, including decomposition and mixing energies, electronic band gap, and Spectroscopic Limited Maximum Efficiency (SLME) based on computed optical absorption spectra.²<br/><br/>A screening process is applied across the dataset of ~ 500 pure and mixed composition perovskites, in terms of the stability, bandgap, deviation from cubicity (to account of any excessive distortion in lattice parameters), degree of octahedral distortion, and finally, some established perovskite stability metrics such as the tolerance and octahedral factors .^3-5 This leads us to 49 stable compositions with attractive properties which are ranked according to SLME and compared with the literature as well as recommended for synthesis and characterization. Using Pearson correlation coefficients and unsupervised learning techniques such as PCA and T-SNE, we unravel qualitative relationships between elemental/molecular combinations, octahedral distortion, and computed electronic and optical properties. We further develop a unique perovskite descriptor with compositional, physical, and octahedral information, which are used to train regression-based machine learning (ML) models and predict properties of thousands of novel compounds.<br/><br/>In order to make the dataset openly available to researchers, we developed NanoHub^6-7 tools that help visualize our entire DFT database, create new structures with arbitrary octahedral distortion and rotation, and make ML predictions for any compositions of interest. The high-throughput DFT framework and tools from this work are promising for accelerating the design of halide perovskites for various applications such as solar cells, photodiodes, electronics, and infrared sensors, and are currently being coupled with active experiments for validation and discovery.<br/><br/><br/>1. Mannodi-Kanakkithodi, Energy & Environmental Science 2022, 15, 1930-1949.<br/>2. Yang, J.; A High-Throughput Computational Dataset of Halide Perovskite Alloys. In Preparation.<br/>3. Bartel Christopher, J. Science Advances, 5, eaav0693.<br/>4. Sampson, M. D.; Journal of Materials Chemistry A 2017, 5, 3578-3588.<br/>5. V. M. Goldschmidt, Nach Untersuchungen Gemeinsam Mitt. Barth, G. Lunde, W. Zachariasen.<br/>6. Madhavan, K.; Nanotechnology Reviews 2013, 2, 107-117.<br/>7. Klimeck, G.; Computing in Science & Engineering 2008, 10, 17-23.