Atomistic simulations with QuantumATK are used in semiconductor, battery, and other materials R&D across a broad range of high-tech industries to predict material physical and chemical properties, such as electronic, structural, optical, thermal, etc. Modeling with QuantumATK reduces development time and cost by down-selection of promising material candidates with desired properties prior to going to the lab and dealing with expensive and possibly hazardous chemicals with unknown properties and behavior.
Density Functional Theory (DFT) atomistic approach is widely applied in predicting these properties accurately for any element in the periodic table in any kind of atomic arrangement, without the need for experimental input parameters. However, DFT simulations are computationally rather expensive and thus scanning of thousands of materials for target properties could take a long time. Machine Learning algorithms, such as Neural Networks, could highly reduce the computational cost of such a large-scale screening.
The goal of this project is to train Machine Learning algorithm(s) with calculated DFT data using QuantumATK for a selected dataset of materials (primarily, molecules). An essential part of this project will beto demonstrate that the trained Machine Learning algorithm can reliably and efficiently predict selected physical and chemical properties of a large pool of new materials that share some common composition and structural features with materials present in the selected dataset.
Benjamin Fogstrup Grundahl
R&D Engineer,
Synopsys QuantumATK, Copenhagen
MSc from DTU Chemistry, 2020
“I studied my master at DTU Chemistry where I did my thesis in collaboration with Synopsys QuantumATK implementing a solvation model called COSMO. It was a great experience and opportunity to challenge myself in a ‘real-life’ project while knowing that both DTU and Synopsys QuantumATK supported me if I needed any help. I also had the chance to discuss the project with the people working at Synopsys QuantumATK which was a fantastic learning.”
See the poster presentation here