![\"\"](\"https:\/\/i0.wp.com\/semiengineering.com\/wp-content\/uploads\/2025\/08\/Synopsys_Materials-Modeling-Superconducting-Qubits-fig1.webp?ssl=1\")
<\/a><\/p>\nFig. 1: Modeling qubits in a realistic way involves large-scale atomistic models with possibly amorphous materials, disorder, interdiffusion, localized defects, and so on, in order to capture all effects that may contribute to noise, such as thermal effects and interface scattering.<\/p>\n
The most common atomistic simulation method on the quantum-mechanical level, density functional theory (DFT), has been used very successfully for decades to predict electronic properties, chemical reactions, material strength, phase transitions, thermal transport, and many other key material properties. DFT alone is, however, not enough for the study of materials for quantum computing, but must be supplemented both with algorithms that provide higher accuracy and more approximate methods to allow for large-scale simulations with sufficiently short turnaround time.<\/p>\n
An effective materials exploration platform for qubits needs to provide flexibility to transition seamlessly between all these methods and maximize the synergy between them, to allow materials engineers to:<\/p>\n
run large-scale molecular dynamics and structure prediction based on machine-learned potentials (MLPs);
\nuse DFT for electronic structure and defect level analysis;
\nswitch to tight-binding models, calibrated against DFT, to handle much larger structures;
\nemploy non-equilibrium Green\u2019s functions (NEGF) for quantum transport calculations;
\nvalidate DFT results against higher-accuracy models such as GW; and
\nimplement custom algorithms to study specific quantum-mechanical properties such as many-body spectra.<\/p>\n
All these simulation methods should be combined under a unified interface, with common data structures, eliminating the friction of moving data between disparate codes and enabling truly synergetic, multiscale workflows.<\/p>\n
For high-performance simulations, the calculations should furthermore be able to take advantage of hardware acceleration on graphics processing units (GPUs), which have been proven to provide 20X faster training and 5\u201315X faster execution of MLPs, with routine simulations of 100,000+ atoms at near-DFT accuracy. GPUs can also provide 5\u201310X speedups for DFT and tight-binding calculations, especially for larger models, and up to 100X gains for NEGF transport simulations.<\/p>\n
A working solution using atomistic simulation<\/p>\n
Synopsys QuantumATK<\/a> atomistic modeling and simulation software meets all these requirements, bridging classical and quantum frontiers in materials science. It is a unified ecosystem integrating multiple modeling approaches within a single framework. For more than a decade, QuantumATK has been in wide use by engineers at leading companies who rely on its robustness for critical tasks and a steady pace of performance improvements.<\/p>\n Integrating machine learning (ML) into atomistic simulation is more than a speedup play\u2014it reshapes how new materials are discovered. The QuantumATK framework provides built-in moment tensor potential (MTP) and neural-network models trained on DFT data, and also allows external MLP calculators to plug directly into the framework. These models can achieve near-quantum accuracy while running 1000\u2013100,000X faster than untrained methods. The platform also offers users a flexible application programming interface (API) to develop custom ML models in Python within a unified environment, with direct access to all the aforementioned atomistic methods and carefully designed data classes.<\/p>\n In addition to modeling qubits, QuantumATK can address problems of interest to quantum algorithms. While the timeline to general-purpose quantum computing remains debated\u2014with expert surveys pointing to 2035\u20132040 for practical materials application\u2014the Fermi-Hubbard model, a key testbed for early quantum hardware, can be studied classically in large lattices today using tight-binding and DFT parameterization. Recent quantum experiments have probed small Hubbard systems (16\u201320 qubits), observing metal-insulator transitions and magnetic ordering. QuantumATK can simulate significantly larger systems, providing correctness checks for quantum results and a performance baseline against which to measure quantum speedups. Moreover, classical improvements via GPU and MLFF may outpace early quantum hardware gains, ensuring that classical tools remain competitive as the quantum ecosystem matures.<\/p>\n Real-world results<\/p>\n