Reliable silicon devices depend on accurate predictions of fracture initiation and growth, which remain sensitive to crystal orientation and surface effects at small scales. We present a data‑driven calibration of phase‑field fracture models using molecular statics and dynamics (MS/MD) simulations that explicitly resolve atomic‑scale processes. Silicon samples are loaded under controlled boundary conditions to trigger crack initiation and propagation; we quantify orientation‑dependent responses by extracting energy release rates, crack paths, and traction–separation behavior. These atomistic observables are mapped to continuum parameters, like critical energy release rate $\mathcal{G}_c$, internal length scale $l_0$, and the degradation function choice, within a phase-field fracture framework suitable for brittle fracture in single crystals. The calibrated phase‑field fracture model reproduces key fracture metrics across selected crystal orientations, including path selection and energy dissipation, and achieves good agreement with MS/MD‑derived quantities. Sensitivity analyses show how $l_0$, $\mathcal{G}_c$, the degradation and surface density function choice govern crack regularization and dissipation, while orientation effects are captured through the calibrated constitutive response. Overall, the approach systematically bridges atomic‑scale insight with continuum‑scale prediction, providing a transferable foundation for fracture simulations in silicon‑based materials.