Scientists have developed a new software called QuaTrEx to simulate very small parts in future computer chips. These devices are needed because technology keeps getting smaller, following Moore's law. This allows for more powerful computers, smartphones, and other gadgets. As transistors shrink to nanometers, the strange quantum behaviors of particles at tiny scales, where electrons interact in ways that classical physics cannot predict, affecting how devices work.
Until now, computer simulations could only handle structures with a few atoms because larger ones were too complex for even the biggest computers. The new software combines three methods to overcome this. Density functional theory, or DFT, is a way to calculate quantum properties of materials. The GW approximation fixes DFT's limits by better describing excited states, where electrons gain energy. Non-equilibrium Green's function, or NEGF, models how electrons flow when voltage is applied, using open boundary conditions that let energy or particles pass through the simulation edges like in real devices.
Key advances in QuaTrEx
The software speeds up calculations for boundary conditions and applies open boundaries to the GW part for the first time. It also uses a new algorithm to split tasks across many graphics processing units (GPUs) for handling complex math. Researchers ran QuaTrEx on two supercomputers: Alps in Switzerland and Frontiers in the USA. They simulated a nanoribbon - a thin strip of material used in transistors - with 42,240 atoms, similar to real industry designs. This work earned an honorable mention in a major computing award for achieving over one exaflop, or a quintillion calculations per second.
Looking ahead, improvements could include using mixed-precision math, where numbers are stored with fewer bits to speed things up, or machine learning to predict starting data instead of calculating it slowly. This could lead to simulating full transistors or logic gates, which are basic circuits for operations in computers. Overall, QuaTrEx helps design better nanoscale devices for AI, robotics, and more.
This research is published in the Proceedings of SC 25.