Machine learning-guided broadband phonon blockade via Anderson localization in engineered Si/Ge nanowires

The localization of thermal phonons is a well-represented phenomenon in disordered atomic systems. Traditional disorder entities, such as random defects and aperiodicity, impede mid- and high-frequency phonons more than low-frequency modes, thus increasing the relative contributions of low-frequency phonon to thermal conductivity. Recognizing the constraints that this drawback imposes on the design of thermal functional materials, our research proposes a disordered structure for Si/Ge nanowires utilizing a checkerboard pattern. By combining molecular dynamics simulations with machine learning techniques, we identified an optimal configuration that minimizes thermal conductivity. This structure incorporates a distinctive design that achieves both strong phonon Anderson localization and enhanced interface scattering. This dual effect significantly hinders phonon transport across the broadband frequency spectrum, reducing the contribution of low-frequency phonons to thermal conductivity to a mere 9%. Our study deepens insights into phonon behavior in disordered structures and introduces a practical method for precise thermal conductivity control.