Session: 09-03: Computational Methods in Micro/ Nanoscale Transport

Paper Number: 131217

131217 - Mode-Resolved Phonon Transmittance Across Ga2o3/sic Interface Using Lattice Dynamics With Machine Learning Potentials 

Abstract: 

Ga2O3, an emerging semiconductor material with an ultra-wide bandgap, has demonstrated considerable potential for enabling next-generation power electronics. However, the extremely low intrinsic thermal conductivity of Ga2O3 presents a formidable challenge to heat dissipation and thermal management in prospective Ga2O3-based devices. To mitigate this issue, the utilization of SiC substrate, with its superior thermal conductivity, has been proposed as a viable solution for facilitating effective heat removal in Ga2O3 devices. Gaining a fundamental understanding of the thermal boundary conductance (TBC) at the Ga2O3/SiC heterointerface thus becomes critical for the thermal engineering and performance optimization of these promising material systems.

However, accurately predicting TBC at a complex heterogeneous material interface is an extremely difficult undertaking. This difficulty stems from two primary factors: (1) the lack of precise experimental characterization and theoretical modeling of the atomic-level interactions occurring at the interface, and (2) the absence of computational methodologies that can efficiently and accurately calculate the interfacial thermal transport from first principles.

To overcome these challenges, this work develops an innovative computational approach that combines lattice dynamics simulations and machine learning interatomic potentials to enable the efficient and accurate prediction of TBC across the Ga2O3/SiC interface for the first time. Critically, calculating the mode-resolved phonon transmittance across the interface provides fundamental insights into the interfacial heat conduction mechanisms beyond what frequency-averaged descriptions can offer. However, widely used models like the acoustic mismatch model (AMM) and diffuse mismatch model (DMM) only provide frequency-resolved transmission, while atomistic methods like molecular dynamics and atomic Green's function incorporate interface structure but remain limited to frequency-level information.

Lattice dynamics (LD) simulations enable calculating the transmittance of individual phonon modes across an interface by explicitly tracking the coupled phonon transport and scattering processes. However, existing LD techniques lack proper enforcement of energy conservation, which restricts their applicability to real material interfaces. To overcome this limitation, this work develops an enhanced LD methodology that incorporates energy conservation constraints through linear algebra transformations and projection gradient descent iterative solving.

The improved LD technique is applied to study the mode-resolved phonon transmission across Ga2O3/SiC interfaces. To enable accurate interatomic force predictions, a machine learning (ML) potential for the Ga2O3/SiC system is developed using the state-of-the-art MACE framework, which employs a higher-order graph neural network architecture to learn complex many-body atomic interactions, achieving density functional theory (DFT) level accuracy at a fraction of the computational cost. To enable accurate modeling of both crystalline and disordered phases, the potential was trained on diverse reference data encompassing crystalline, interfacial, and amorphous configurations. Efficient sampling of the potential energy surface was enabled through random structure searching. In total, the training database contained approximately 300,000 local atomic environments and enabled the ML potential to reproduce DFT energies and forces with high fidelity, as evidenced by force errors below 50 meV/Å on test structures.

The simulation results reveal that the mode-resolved transmittance across Ga2O3/SiC interfaces is highly anisotropic, arising from the low symmetry of the monoclinic Ga2O3 crystal structure. The presence of interfacial stress significantly amplifies this anisotropy by altering phonon mode coupling across the interface. Introducing an amorphous interlayer is found to reduce the interfacial transmittance anisotropy by weakening the phonon coupling between the crystalline Ga2O3 and SiC.

Overall, this work establishes a robust computational framework for investigating mode-resolved phonon transmission at realistic material interfaces. The insights gained into phonon transport at the Ga2O3/SiC interface can guide phonon engineering across interfaces for more effective thermal management in electronics.

Presenting Author: HongAo Yang Tsinghua University

Presenting Author Biography: HongAo Yang is a third year graduate student at the School of Aerospace Engineering, Tsinghua University, advised by Prof. BingYang Cao. His research interests include interfacial heat transport, first principles, and lattice dynamics simulations. He is committed to developing new simulation tools that can advance heat management technologies. In his presentation, HongAo will discuss his work utilizing lattice dynamics and machine learning potentials to better understand heat conduction at realistic interfaces.

Authors:

HongAo Yang Tsinghua University
Yuanbin Liu University of Oxford
Bingyang Cao Tsinghua University