Novel Machine Learning Method for Stability and Energy Bandgap Prediction of Lead Free Perovskite Materials
Alessio Gagliardi a, Jared Stanley a
a Technische Universität München, Karlstraße 45, München, 80333, Germany
Proceedings of International Conference on Hybrid and Organic Photovoltaics (HOPV19)
Roma, Italy, 2019 May 12th - 15th
Organizers: Prashant Kamat, Filippo De Angelis and Aldo Di Carlo
Oral, Alessio Gagliardi, presentation 011
Publication date: 11th February 2019

The identification of suitable lead-free perovskites is crucial for their envisioned applications in photovoltaics. Homovalent substitution of lead with Sn- and Ge-based compounds are under intense investigation as potential alternatives, but suffer from stability issues, for example, due to the susceptibility of these ions toward the 4+ oxidation state. Mixed compositions, with two or more possible ions for each lattice position, have been proposed for overcoming these issues and enhancing performance [1, 2]. However, as it is computationally and experimentally prohibitive to measure the vast configuration space available to the mixed perovskites, statistical learning techniques are needed to find a more efficient mapping of mixing parameters to the properties of interest.

Efficient and accurate vetting of perovskites for a range of properties has recently been accomplished in high-throughput Density Functional Theory (DFT) studies of compounds by use of Kernel Ridge Regression (KRR) [3, 4]. Crucial to their success is the determination of adequate material fingerprints which uniquely define the materials and capture the property of interest. Here we demonstrate how one such important screening parameter, the fundamental bandgap, can be predicted for a family of inorganic mixed halide perovskites using novel globally valid material fingerprints based solely on the atomic configurations of arbitrary unit cells. The Partial Radial Distribution Function method [5] is expanded upon to include densities for a variety of elemental properties, enabling us to define a more robust material fingerprint while illuminating the underlying drivers of target properties in a chemically intuitive manner. The results are supplemented with thermodynamic and geometric data to identify the best compositions and the features responsible for them.

This work was supported by the excellence cluster Nanosystems Initiative Munich (NIM).

© Fundació Scito
We use our own and third party cookies for analysing and measuring usage of our website to improve our services. If you continue browsing, we consider accepting its use. You can check our Cookies Policy in which you will also find how to configure your web browser for the use of cookies. More info