USC researchers are using ML to create materials for energy-efficient electronics.
New scholarly work from the USC Viterbi School of Engineering:
“Ferroelectric materials exhibit a rich spectrum of complex polar topologies, but their study under optical excitation far from equilibrium is largely unexplored due to the difficulty of quantum-mechanically modeling the multiple spatiotemporal scales involved. To study optical excitation at spatiotemporal scales at which these topologies arise, we performed quantum molecular dynamics simulations with neural networks in multiscale excited states, integrating the quantum mechanical description of electronic excitation and molecular dynamics machine learning of billions of atoms, to describe ultrafast polarization control in an archetype ferroelectric oxide, lead titanate. Far-from-equilibrium quantum simulations show a clear photoinduced change in the electronic energy landscape and a resulting transition from ferroelectric to octahedral tilting topological dynamics within picoseconds. The coupling and frustration of these dynamics in turn creates topological defects in the form of polar strings. The demonstrated linkage of multiscale quantum simulation and machine learning will not only advance the emerging field of ferroelectric topotronics, but also broader optoelectronic applications.”
Funding: Computational Materials Sciences Program funded by the US DOE, Office of Science. The simulations were performed at the Argonne Leadership Computing Facility as part of the DOE INCITE and Aurora Early Science programs, and at USC’s Center for Advanced Research Computing.
Visit Semiconductor Engineering’s library of technical documents here and discover many other scientific works of the chip industry.