Tailoring optical properties of perovskite nanocrystals by machine learning

Abstract

Edisonian trial-and-error and intuition methods have been a method that has led the design of chemical synthesis development for decades. These methods are inefficient and their predictability is limited as increasingly complex studies progress. Machine learning (ML) technology has rapidly advanced to make faster and more accurate predictions beyond the limits of intuition. In the field of perovskite nanocrystals (PNCs), ML is mainly used to discover new compositions or materials structures. It is difficult to predict the optical properties of PNCs because particle are formed in a multi-phase structure during synthesis. Herein, we introduce an innovative prediction method for the optical properties of PNCs with ML algorithms. We built a database with more than 250 experimental results with a pipetting robot arm. Looking at the database, PNCs were synthesized in various structures and sizes. PNCs produced by multi-phase were classified into peak groups according to the emission wavelength range and intensity groups by luminance. Classified peak positions and intensities were predicted by decision trees, random forests, and gradient boosting. The optical properties were predicted by converting the predicted peak position and intensity values into a Gaussian graph.