Machine Learning Force Fields Simulate Large Atomistic Systems, Improving Fidelity for Quantum Qubits and Sensors

Simulating the behaviour of materials at the atomic level is crucial for advancing superconducting qubit and sensor technology, but conventional methods struggle with both accuracy and scale. Søren Smidstrup, Shela Aboud, and Ricardo Borges, alongside colleagues at Synopsys, Inc, address these limitations by integrating machine learning into atomistic modelling. Their work introduces a new approach that combines the strengths of traditional density-functional theory with machine-learned force fields, enabling the simulation of larger, more complex systems with improved fidelity. This advancement allows researchers to accurately model key phenomena like superconductivity and surface states, and crucially, to simulate the thermal properties essential for practical quantum computing applications, ultimately paving the way for more robust and reliable quantum devices.

Density-functional theory (DFT) struggles to fully capture the complex physics governing superconductivity and surface states, particularly when simulating larger systems crucial for quantum computing applications. The QuantumATK platform addresses these challenges by combining DFT, based on localized atomic orbitals, with non-equilibrium Green’s functions to characterize interfaces between superconductors and insulators, and to analyze the surface states of topological insulators. Additionally, the software utilizes machine-learned forcefields to simulate thermal properties and to generate.

Multi-Scale Simulations for Qubit Materials Engineering

Scientists employ a multi-model simulation approach to overcome the limitations of materials engineering for qubits, integrating density-functional theory (DFT), semi-empirical tight-binding (TB) models, and machine-learned forcefields (MLFFs) within the QuantumATK platform. This platform enables seamless transitions between different levels of computational complexity, allowing researchers to leverage the strengths of each technique for comprehensive analysis. The study pioneers the use of DFT, based on localized orbital basis sets, to perform accurate large-scale electronic structure calculations on systems containing thousands of atoms, even with limited computational resources. To broaden the scope of simulations, scientists harness semi-empirical tight-binding models for faster calculations, parameterizing them with DFT data when necessary, and utilize empirical potentials for efficient molecular dynamics simulations of systems with millions of atoms.

Recognizing the difficulty of developing accurate forcefields for novel materials, the research team leverages recent advances in machine-learning algorithms to construct machine-learned forcefields, enabling simulations of complex materials with interfaces and multiple chemical elements. This approach allows for the inclusion of temperature-dependent effects in DFT calculations through electron-phonon interaction or by generating thermodynamic ensembles of atomistic structures. Furthermore, the study introduces a unique Green’s function model, coupled with DFT and tight-binding, utilizing open boundary conditions to study electron transport on the quantum-mechanical atomistic level. This technique overcomes the limitations of conventional DFT in accurately describing the surface band structure of topological insulators. By integrating these diverse computational methods within the QuantumATK platform, scientists achieve unprecedented capability in simulating complex quantum systems, facilitating the development of advanced qubit technologies and materials.

Atomic Scale Modeling Predicts Superconductor Properties

Scientists developed a comprehensive modeling platform, QuantumATK, to guide the development of qubits and sensors by accurately simulating material properties at the atomic scale. The work combines density-functional theory with non-equilibrium Green’s functions to characterize interfaces between superconductors and insulators, as well as the surface states of topological insulators, enabling detailed analysis of these complex systems. Researchers successfully modeled the critical temperature of several diboride materials, achieving excellent agreement with experimental results and demonstrating the tool’s ability to screen candidate materials combining high critical temperatures with desired topological properties. Furthermore, simulations revealed that applying strain can enhance the critical temperature, providing new avenues for realizing high-temperature superconductivity.

The team also created an atomistic model of a two-level system, representing material defects that impact qubit coherence, utilizing a discrete quantum-mechanical description to compute energy level splitting and other properties. Simulations of a double quantum dot system, containing over 1 million atoms, were performed using a semi-empirical tight-binding model, significantly faster than full diagonalization methods. These calculations allowed researchers to extract quantities such as coherent oscillations and Coulomb blockade diagrams, which can be used to simulate qubit operation and decoherence. By varying the detuning potential between quantum dots, scientists observed transitions between singlet and triplet states, defining the operations for qubit read-out and manipulation.

Measurements of many-body energy levels for interacting electrons in a GaAs coupled double quantum dot system confirmed the ability to accurately model complex quantum phenomena. The simulations distinguished the influence of effects difficult to observe experimentally, such as the impact of impurities versus intrinsic breaking of time-reversal symmetry. Researchers demonstrated the tool’s ability to model superconducting tunnel junctions, critical for quantum computing and high-sensitivity sensors, and optimize these devices for improved performance and reliability. The platform enables precise modeling of atomic-scale variability in junctions, contributing to the development of more efficient, scalable, and reliable quantum devices.

Atomistic Modeling Advances Quantum Material Simulation

This work demonstrates the power of atomistic modeling to guide the development of qubits and quantum sensors, addressing limitations inherent in traditional simulation techniques. Researchers successfully combined diverse computational methods, including forcefields, tight-binding models, and advanced first-principles calculations with Green’s functions, to comprehensively model material properties relevant to quantum technologies. This integrated approach allows for detailed investigation of structural, thermal, and electronic characteristics, providing insights into complex phenomena such as superconductivity and topological states. The team’s simulations accurately describe the behavior of superconducting tunnel junctions, critical components in quantum computing and high-sensitivity sensors, and enable precise modeling of atomic-scale variability within these devices.

Furthermore, they demonstrated the ability to model double quantum dot systems, accurately capturing the energy levels and transitions essential for qubit operation and manipulation. This capability allows researchers to distinguish between subtle influences on material behavior, such as the effects of impurities versus intrinsic properties, which are often difficult to isolate experimentally. The authors acknowledge that computational cost remains a limitation, particularly when simulating large-scale systems or extended timescales. Future work will focus on extending these methods to model more complex qubit architectures and to incorporate environmental effects that contribute to decoherence. By providing a powerful platform for in silico experimentation, this research accelerates materials discovery and optimization, ultimately contributing to the development of more efficient and reliable quantum devices.