Session: K9-10: NANOSCALE THERMAL TRANSPORT MODELING AND MACHINE LEARNING II

Paper Number: 138632

138632 - Jax-Bte: A Differentiable Hybrid Neural Solver for Deep Learning Accelerated Thermal Modeling of Nanoelectronics 

Abstract: 

Numerical simulation of nanoscale thermal transport plays a crucial role in understanding thermal management in integrated circuits. Up to this date, Fourier’s law with effective thermal conductivity (TC) calibrated against phonon Boltzmann transport equation (BTE) simulations of specific transistor geometries, represents the state-of-the-art methodologies for addressing the circuit-level thermal transport. However, the temperature distribution within the transistor remains unknown, limited by the computational challenges of modeling non-diffusive transport, i.e. ballistic transport, across multiple spatial scales from transistor to circuit. To address this issue, we introduce JAX-BTE: a fully differentiable GPU-accelerated Python package for efficiently solving non-gray phonon BTE. JAX-BTE also allows end-to-end, sequence-to-sequence deep learning (DL) accelerated simulation of complex multiscale thermal transport, leveraging DL and numerical methods. Our proposed approach is essentially a differentiable numerical solver based on discrete ordinates method (DOM), which can simulate both steady-state and transient conditions. The entire solver is constructed using differentiable programming in JAX, resulting in a significant speedup compared to conventional numerical BTE solvers, thanks to GPU acceleration and the potential coupling with deep neural networks through automatic differentiation. The proposed method allows the integration of known BTE physics with DL techniques, making it more data efficient and generalizable compared to purely blackbox, data-driven DL models. 

Presenting Author: Bo Zhang University of Notre Dame

Presenting Author Biography: Bo Zhang is a postdoctoral research associate at the Department of Aerospace and Mechanical Engineering at the University of Notre Dame. He obtained his Ph.D. degree at Baylor University and afterwards, he studied as a postdoctoral fellow at Georgia Institute of Technology. His research focuses on liquid fuel atomization, droplet dynamics, shock waves, aerodynamics, reacting flows, advanced multiphase forming for paper-making, and machine learning methods for fluid mechanics.

Authors:

Bo Zhang University of Notre Dame
Wenjie Shang University of Notre Dame
Jyoti Panda University of Notre Dame
Jiahang Zhou University of Notre Dame
Jianxun Wang University of Notre Dame
Tengfei Luo University of Notre Dame