Deep learning modeling of oxygen redistribution and thermal transport in silicon on insulator and buried oxide layers

Silicon on insulator technology requires precise control of buried oxide layers and the associated heat transfer across interfaces. Current approaches struggle to predict oxygen distribution and layer thickness after implantation and annealing, and they face challenges in computing interfacial thermal resistance under complex conditions. Here we show a computational framework that integrates machine-learned interatomic potentials with molecular dynamics to resolve oxygen diffusion kinetics and the evolution of layer thickness. The framework predicts depth-dependent oxygen redistribution during annealing and reproduces the measured dimensions of fabricated samples with high accuracy. It also enables reliable determination of interfacial thermal resistance, surpassing conventional molecular dynamics method in agreement with experimental benchmarks. By linking atomic scale oxygen migration with macroscopic heat transport, this work establishes quantitative relationships between structure, process, and properties. The framework provides guidance for optimizing implantation energy, dose, and annealing protocols, and represents a step toward rational design of high-performance silicon on insulator technologies. Silicon on insulator technology demands precise control of buried oxide layers and heat transfer across interfaces, yet current methods fall short in predicting oxygen distribution and interfacial thermal resistance. Here, the authors present a computational framework combining machine-learned interatomic potentials with molecular dynamics, accurately predicting oxygen redistribution and thermal resistance, thereby advancing the design of high-performance silicon on insulator technologies.