Nanotechnology Meets Machine Learning

This research uses deep learning to discover new materials and devices – particularly in renewable energy and increasingly for medical applications. Models aim to learn chemical and physical laws to 

1.  generate new materials and
2. simulate their behavior efficiently.

There are currently three areas of research:

Comparative Convolutional Neural Networks (CNNs) for Perovskite Solar Cells

CNNs can predict the performance of perovskite solar cells from images. Unlike classical methods, they analyze images of the same device in different states, such as before and after encapsulation, or compared to a reference image. This allows precise detection of relative changes in efficiency. Especially in scenarios with small datasets, this method achieves high accuracy and enables fast, scalable characterization of solar cells without time-consuming measurements.

AUGUR: Optimization of Adsorption Sites

The Aware of Uncertainty Graph Unit Regression (AUGUR) algorithm identifies optimal adsorption sites in molecules and clusters. By combining graph neural networks with Bayesian optimization, it creates a model that automatically quantifies uncertainties and works independently of symmetries, translations, or rotations. AUGUR is universally applicable, requires less computing time than classical methods, and can efficiently analyze molecules of any size. It allows large, complex systems to be optimized in a data- and resource-efficient manner.

ML for Lead-Free Perovskites: ML models enable the development of lead-free perovskites for photovoltaics. They can predict band gaps, stability, and energy, identifying promising material combinations without costly simulations. Using a database of 344 computed perovskite materials, the chemical space can be quickly explored to identify efficient, sustainable materials for the next generation of solar cells.