AI Predicts Behaviour of Quantum Systems
Scientists from HSE University, in collaboration with researchers from the University of Southern California, have developed an algorithm that rapidly and accurately predicts the behaviour of quantum systems, from quantum computers to solar panels. This methodology enabled the simulation of processes in the MoS₂ semiconductor and revealed that the movement of charged particles is influenced not only by the number of defects but also by their location. These defects can either slow down or accelerate charge transport, leading to effects that were previously difficult to account for with standard methods. The study has been published in Proceedings of the National Academy of Sciences (PNAS).
Modern electronics rely on quantum effects. Devices such as semiconductors, LEDs, and solar panels all depend on the behaviour of electrons in materials. Accurately describing these processes is challenging, as their simulation demands immense computing power. Simulating the motion of electrons in a material with thousands of atoms requires supercomputers to perform millions of calculations.
Typically, quantum systems are modelled using the molecular dynamics method, which enables predictions of how atoms and electrons will move over time. However, when electron states change rapidly, standard modelling methods become excessively resource intensive.
Researchers at MIEM HSE solved this problem by leveraging machine learning. The new algorithm analyses small fragments of the material, learns from their local properties, and then predicts the behaviour of the entire system. The scientists studied the two-dimensional semiconductor molybdenum disulfide (MoS₂), a promising material for optoelectronics and photovoltaics. In particular, it can function as the active layer in solar cells. Ideally, molybdenum (Mo) and sulphur (S) atoms form an ordered lattice, but in real materials, the structure is rarely perfect, as defects may occur. Defects are disruptions in the arrangement of atoms. In MoS₂, defects can manifest as vacancies (the absence of sulphur or molybdenum atoms), excess atoms between layers, local displacements, or other deviations from the ideal lattice. Defects can alter the behaviour of electrons: in some cases, they may impair conductivity, while in others, they can impart new properties to the material, such as increasing its sensitivity to light or its ability to conduct charge.
Dongyu Liu
'To understand how defects impact electron movement, we focused on small fragments of the material. The algorithm first analysed the local properties of the system and then predicted the behaviour of the entire structure. It’s similar to learning a language: first, you memorise individual words, and then you begin to understand whole sentences,' says Dongyu Liu, Assistant Professor at MIEM HSE.
It turns out that not only does the number of defects matter, but also their location. Defects can either delay or accelerate the movement of charged particles, creating traps for charge carriers within the semiconductor's band gap. Standard methods struggle to calculate these effects accurately, as the calculations must account for interactions both between defects and with the atoms of the material, which is difficult when using small computational cells. Machine learning helps overcome these dimensional limitations and account for the synergistic effects of multiple defects in the material.
Andrey Vasenko
'Importantly, this method not only speeds up calculations but also facilitates the study of real quantum systems,' explains Andrey Vasenko, Professor at MIEM HSE. 'The results of our research will help bridge the gap between theoretical modelling and experimental studies of materials. We have developed a new approach to studying charge motion in complex systems by combining high-precision computing, molecular dynamics, and machine learning. This method will help investigate materials in which electrons carry energy and information. This is crucial for electronics and energy production.'
