Machine Learning Speeds up Predicting the Mechanical Properties of SiC Nanophononic Heterostructures

K Ren and ZL Zhuang and K Wang and Y Wei and JP Li and HB Shu, JOURNAL OF PHYSICAL CHEMISTRY C, 129, 15462-15470 (2025).

DOI: 10.1021/acs.jpcc.5c04526

This work investigates the mechanical properties of a silicon carbide (SiC) nanophononic heterostructure (NPH) tuned by temperatures and pore geometries, using molecular dynamics (MD) simulations in conjunction with the machine learning method. The SiC NPH, constructed with pure SiC and phononic crystal pores, exhibits temperature-dependent fracture behavior, showing a decrease in mechanical strength decreasing as the temperature increases from 50 to 500 K. The fracture strength and strain are markedly influenced by pore size, with a particular emphasis on pore length, whereas pore width exerts a negligible influence. In addition, armchair interfaces can cause a higher mechanical strength compared with zigzag interfaces. The presence of larger phononic pores amplifies stress concentrations and thermal effects, consequently resulting in more significant reductions in mechanical performance. More importantly, a random forest model has been developed to accurately predict fracture characteristics of the SiC NPH, achieving a high degree of precision (R 2 = 0.99) and offering a 600-fold increase in computational speed compared to traditional MD methods. These findings provide valuable insights for designing robust SiC-based nanodevices with tunable mechanical properties.

Return to Publications page