Quantifying the impact of interatomic potentials on silicon fracture simulations using unified mode-I test platforms

Atomistic simulations provide valuable insight into fracture and deformation in single-crystal silicon, but their accuracy strongly depends on the choice of interatomic potentials. To evaluate these effects, we systematically assess twelve silicon potentials within a unified mode-I fracture test framework, including SW, modified SW (SW2) with enhanced brittleness, Tersoff (T3), T3 with an extended cutoff (T3C), MEAM, EDIP, ReaxFF, two MOD variants (MOD1 and MOD2), ADP, and the machine-learning-based L-SNAP and Q-SNAP. Among these, MOD, ADP, and L-/Q-SNAP are recently developed models that have not been thoroughly assessed for their ability to predict silicon fracture properties. Simulations performed for three crystallographic orientations reveal that failure behaviour varies considerably depending on the potential. SW, SW2, T3C, MEAM, ReaxFF, ADP, and the SNAP models predict well-defined fracture modes, whereas T3, MOD1, MOD2, and EDIP exhibit enhanced crack resistance. Lattice and elastic constants, surface energy, stress–strain responses during fracture, and fracture toughness are compared. Furthermore, tensile and shear strengths are examined to elucidate the underlying mechanisms, and a novel stress ratio parameter is introduced to characterise and predict fracture behaviour. The accuracy of force and energy calculations of T3 and T3C is also evaluated in comparison with the machine-learning-based L-SNAP and Q-SNAP potentials.