Cs-Pb-Br halide perovskite phases: photoluminescence, in-situ studies, and structure prediction with machine learning

The materials comprising combinations of Cs, Pb and Br feature unique optoelectronic properties, and both their fundamental behavior and the synthesis processes need to be fully understood in order to achieve meaningful applications. In this thesis we cover fundamental physical properties of some of these materials, follow their dynamics as a function of temperature and investigate their synthesis parameter space using machine learning. Specifically, we examined the luminescence properties of zero-dimensional (0D) Cs4PbBr6, exploring the steady-state and time-resolved photoluminescence (TRPL) behavior of millimeter-scale Cs4PbBr6 crystals across a temperature range of 80 to 360 K. This analysis included observations of exciton binding energy, phonon energy, and changes in PL luminescence peak with temperature. High-resolution transmission electron microscopy (HRTEM) characterization and TRPL results confirmed that nanoscale CsPbBr3 crystals are embedded within the Cs4PbBr6 bulk matrix, as hypothesized in the literature. Furthermore, we fabricated white light-emitting diodes (WLEDs) device based on Cs4PbBr6/CsPbBr3, achieving a luminous efficiency of 64.56 lm/W and color coordinates of (0.34, 0.31), closely matching standard white coordinates. The device’s color gamut was measured at 128.66% of the National Television System Committee (NTSC) standard. We employed in-situ characterization techniques, including X-ray diffraction (XRD) and TEM, to analyze 0D Cs4PbBr6 and their dynamics as a function of temperature. In-situ XRD heating revealed that pure 0D Cs4PbBr6 nanocrystals (NCs) start converting to 3D CsPbBr3 at approximately 150 °C, increasing with temperature up to 400 °C. We also monitored this transformation through in-situ heating high-resolution TEM (HRTEM). We observed that not all 0D Cs4PbBr6 would transform into 3D CsPbBr3 under heating. However, as TEM imaging can only provide information about a limited number of crystals, we couldn’t draw conclusions at ensemble level. This prompted us to investigate in more detail the interplay between three different Cs-Pb-Br related materials, normally referred to as 0D, 2D and 3D perovskites. To understand how their nucleation, growth and evolution lead to different products, we applied machine learning (ML) to summarize and predict the crystal phase outcomes based on experimental parameters. Using data from 220 documented synthesis runs in the literature as a training dataset for eight different ML models, we achieved a prediction accuracy of 84.1%. This trained model was subsequently validated by predicting the structure of 10 experimental runs based on random parameters, achieving an experimental verification accuracy of 80%. Our findings highlight the critical influence of the Cs/Pb molar ratio, reaction time, and concentration of organic compounds (ligands) on the crystal structure of Cs-Pb-Br.