On the afternoon of November 15, a team of scientists from the University of Science and Technology of China (USTC) virtually attended the 2022 Supercomputing Conference (SC22) held in Dallas, Texas, reporting their achievement in first realizing first-principles (ab initio) computing simulation of the electronic structures of complex metallic heterostructures with 2.5 million atoms. This is USTC’s second time nominated for the Gordon Bell Prize as the first contributor. The related article was published in Science Bulletin.
Advanced materials form an important foundation for the transformation and upgrading of manufacturing industry. While experiments for verifying material properties are complex in design and cost, ab initio methods allow a rather precise calculation of the fundamental structures and physiochemical properties of given materials based on quantum mechanics. In addition, the contrivance of high-performance computing makes the ab initio modeling viable in investigating the electronic structures of ultra-large-scale metallic systems containing tens of thousands of atoms.
Galerkin density functional theory (DGDFT), as an ab initio application, combines the strengths of both small localized and large uniform basis sets. It could reduce the number of basis functions similar to numerical atomic basis sets, while maintaining the high precision comparable to that of plane-wave basis set.
In particular, the DGDFT method jointly developed by USTC and CAS adopts a combination of a two-level parallelization strategy and the most advanced pole expansion and selected inversion (PEXSI) technique, conducting key innovations in numerical algorithm, parallel sparse-matrix solution and data communication schemes.
Based on the above innovations, the team ran the DGDFT on the new Sunway supercomputer and for the first time realized ab initio computing simulation of the electronic structures of complex metallic heterostructures with 2.5 million atoms, which is expected to be applied in the construction of 2D-materials-based transistors (2DFETs). The size of the material simulation is hundreds of times larger than the previous software with the same plane wave precision.
In the future, with the supercomputer computational performance reaching 10 exaflops, the high scalability of DGDFT software could allow further extension of first principles material simulations to macroscopic scale (>1000nm), realizing the simulation of real materials and devices. Ultimately, it could pave the way for the industrial application of integrated software and hardware for first principles material simulations.
Journal
Science Bulletin