<p>Single-crystal silicon, a widely used substrate material for Micro-Electro-Mechanical Systems (MEMS), has been extensively studied for its quasi-static mechanical properties. However, data concerning its mechanical properties at high strain rates are scarce, and corresponding research is insufficient, thereby severely limiting the extended applications of silicon-based MEMS. This study first conducted three sets of high strain rate tests on single-crystal silicon using a Split Hopkinson pressure bar (SHPB) system to successfully acquire its dynamic stress–strain curves. Subsequently, the experimental results were utilized to calibrate and validate the finite element simulations performed in ABAQUS, ensuring the accuracy of the numerical modeling. Finally, a machine learning model based on a Backpropagation Neural Network (BPNN) was developed. This model utilizes the incident wave parameters from SHPB experiments as input features to predict the mechanical properties of single-crystal silicon at high strain rates. The results indicate that the developed BPNN-based machine learning prediction model exhibits excellent applicability. It can effectively circumvent the extensive and time-consuming processes of finite element simulation modeling, analysis, and post-processing. Furthermore, it accurately predicts the mechanical properties of single-crystal silicon at high strain rates and demonstrates the potential to extrapolate stress–strain curves to even higher strain rates. This research provides a valuable reference for subsequent work on establishing strain rate-dependent constitutive models for single-crystal silicon, optimizing the design of silicon-based MEMS structures, and expanding the applications of silicon-based MEMS.</p>