Session: 16-01-01: Government Agency Student Poster Competition

Paper Number: 150925

150925 - Phonon Local Non-Equilibrium at Al/si Interface, Demonstrated by Machine Learning Molecular Dynamics 

All micro and nanoelectronics are equipped with metal/semiconductor junctions, resulting in resistance to thermal transport. Over the years, significant effort has gone into understanding and enhancing thermal transport across interfaces using theoretical and experimental techniques. The nanoscale phononic complexities like phonon local non-equilibrium and inelastic scattering add to computational and experimental characterization difficulty. Al/Si is a quintessential metal-semiconductor pair due to their prominence in electronic devices. After many years of efforts, experimental measurements have achieved consensus in the last decade. However, the simulation results and experimental measurements still show discrepancies with a lack of reliable theoretical explanation. We use a neural network potential (NNP) trained by ab initio data, demonstrating near-first-principles precision more accurate than classical potentials used in molecular dynamics (MD) simulations to predict interfacial thermal transport with Al/Si as a prototype system.

Our NNP-MD simulations show an unprecedented agreement of interfacial thermal conductance with previous experimental consensus while considering all the essential physics - interfacial bonding nature, phonon local non-equilibrium, and inelastic scattering. Previous Landauer-approach-based studies often do not entirely capture these effects, while MD simulations with empirical potentials are inaccurate and inconclusive. In addition to the interfacial interactions, our NNP provides a unique novelty through its ability to predict the atomic interactions in two distinct atomic environments - face-centered cubic (FCC) lattice of metallic Al and diamond cubic lattice of non-metallic Si. Next, we decompose the atomic trajectories from the NNP-MD simulations to obtain spectral insights into the interfacial behavior of phonons and demonstrate the presence of a very thin interfacial region. We observe the optical modes of Si interfacial region in the frequency range of 12<ω<16 THz pass their energy to lower frequencies to allow acoustic coupling of Al and Si modes. Our work using the NNP-NEMD thoroughly validates the previous findings of Feng et al. using classical MD through the NNP's ab initio accuracy of interfacial interactions.

Furthermore, we decompose the atomic trajectories using the phonon spectral temperature method to observe the phonon local non-equilibrium. The bulk longitudinal acoustic (LA) modes of Al are at a lower temperature in the bulk region. These phonon modes gain energy and show higher temperatures near the interface. The same trend is not observed in Al's transverse acoustic phonon modes. This indicates that the LA modes are receivers of phonon energy from the Si interfacial region. Our modal analysis shows the utility of the LA phonon modes, which act as a hypothetical bridge to cross the interfacial barrier for thermal transport from Si to Al at the atomic scale. Our work uses a novel machine learning method to employ high-fidelity MD to predict and extract accurate insights about interfacial thermal transport. Our accurate results advance the fundamental understanding and create a pathway toward further optimization of thermal transport at the interface.

Presenting Author: Krutarth Khot Purdue University

Presenting Author Biography: Krutarth is a fourth-year PhD student working with Prof. Xiulin Ruan in Mechanical Engineering at Purdue University. His research focuses on using machine learning techniques to perform atomistic simulations relevant to thermal transport. Krutarth's field of interest includes thermal management research in semiconductors and battery technologies.

Authors:

Krutarth Khot Purdue University
Boyuan Xiao Purdue University
Zherui Han Purdue University
Ziqi Guo Purdue University
Zixin Xiong Purdue University
Xiulin Ruan Purdue University