Accurate atomistic and electronic-structure calculations based on density functional theory (DFT) and DFT-based ab initio molecular dynamics (AIMD) calculations underpin much of modern computational chemistry and materials science[1]. DFT calculations offer quantitative insights into chemical bonding, charge transport, phase stability, reaction pathways, etc[2]. However, this accuracy comes with a high computational cost. As the system size increases, the numerical solution of the Kohn−Sham equations becomes more challenging. The self-consistent-field (SCF) iterations required for DFT calculations are both time- and energy-consuming on conventional von Neumann digital hardware. Atomistic simulations involving thousands of atoms or long-time scales are often prohibitively costly, restricting the routine application of quantum-accurate methods in materials simulations and design. Alongside algorithmic advances, two complementary trends have emerged to tackle this bottleneck problem.

Link:Software−hardware co-design accelerates materials simulations - IOPscience