Deep learning approach to genome of two-dimensional materials with flat electronic bands

Many-body electron-electron correlations play central role in condensed matter physics; they govern a wide range of phenomena, stretching from superconductivity to magnetism, and are behind numerous technological applications. Two-dimensional (2D) materials with flat electronic bands provide a natural playground to explore this rich interaction-driven physics, thanks to their highly localized electrons. The search for novel 2D flat band materials has since attracted intensive efforts, especially now with the development of open science databases encompassing thousands of materials with computed electronic bands. In this work, we automate the otherwise daunting task of materials search and classification by combining supervised and unsupervised machine learning algorithms. To this end, a convolutional neural network was employed to identify 2D flat band materials, which were then subjected to symmetry-based analysis using a bilayer unsupervised learning algorithm. Such hybrid approach of exploring materials databases allowed us to reveal completely new material classes outside the known flat band paradigms, with high efficiency and accuracy. Our results present a genome of 2D materials hosting flat electronic bands that can be further explored to enrich the physics of electron-electron interactions.