Extending Moore's Law Computing with Quantum Monte Carlo

PI Anouar Benali, Argonne National Laboratory
HfO2 semiconductor with oxygen vacancies
Project Summary

There is extreme urgency in identifying the possible paths forward to extending Moore’s law beyond Si-CMOS based computing technologies. A fundamental materials problem that must be addressed is leakage current in semiconductors. Large intrinsic carrier density is in general responsible of the leakage. By increasing the bandgap energy of a material, we can reduce current leakage; as a result, devices can be reliably operated at higher frequencies, beyond Si-CMOS limits. Current computational studies are limited to density functional theory, which does not have the necessary fidelity to study band gaps in correlated materials. This project aims to use QMCPACK to study the influence of impurities on band gaps near transition metal oxides interfaces using mult-ireference quantum Monte Carlo methods. Because of the size of the systems, such calculations can only be enabled by the peak performance offered by Aurora.

Project Description

There is extreme urgency in identifying the possible paths forward to extending Moore’s law in silicon (Si) complementary metal-oxide-semiconductor (CMOS) based computing technologies. A fundamental materials problem that must be addressed is leakage current in semiconductors, particularly through the HfO2 gate dielectric. By increasing the bangap energy of a material, we can reduce current leakage; as a result, devices can be reliably operated at higher frequencies, beyond Si-CMOS limits. Current computational studies are limited to density functional theory (DFT), which does not have the necessary fidelity to study band gaps in correlated materials. This project aim to use QMCPACK to study the influence of impurities on band gaps near transition metal oxides interfaces using multi-reference quantum Monte Carlo (MR-QMC) methods. Because of the number of electrons required in these simulations, they necessitate and will be enabled by the large aggregate memory and peak performance offered by Aurora.

The research team’s software development efforts will focus on porting and optimizing the multireference evaluation of the trial wavefunction on Aurora’s GPUs. This work will also enhance the usability of Quantum Package, the multireference code which generates the initial trial wave functions for QMCPACK. Efforts to achieve performance portability of diffusion Monte Carlo (DMC), wavefunction optimizers and B-Splines evaluations are outside the scope of Aurora ESP and are funded through an Exascale Computing Project (ECP).

The team believes this Aurora ESP, along with complementary ECP efforts, will be an excellent testbed for the theoretical and algorithmic developments that are outcomes of a currently funded DOE Center for Material Science focused on functional materials.

The proposed research aims to advance knowledge of the semiconductor interface necessary to extend current Si-CMOS technology; it also aims to advance fundamental science necessary to advance computing beyond Moore’s Law and current Si-CMOS by focusing on central problems that gate potential applications of topological insulators.

Project Type