Abstract: Silicon dioxide ( \textSi\textO_2 ) thin films have found extensive applications in semiconductor devices and superconducting electronics owing to their superior electrical insulation, optical transparency, and chemical stability. During the electron beam evaporation (EBE) of SiO₂, the pronounced "pit-digging effect" caused by extremely low thermal conductivity leads to a continuous deterioration of film thickness uniformity as deposition time increases. This study experimentally demonstrates that the conventional "small cone model" incurs a prediction error exceeding 8.80% for the edge-to-center thickness ratio during long-term deposition (> 15 min), primarily because it neglects the dynamic evolution of pit morphology. To address this limitation, a dynamic evolution model for frustum-shaped pits is developed based on thermal diffusion mechanisms and actual geometric characteristics under linear scanning processes. A quantitative functional relationship between deposition time and pit wall inclination is established, and a correction factor is introduced to compensate for the dynamic expansion of the effective evaporation area. A "segmented fitting and independent prediction" logic is employed: experimental data from the first three deposition durations (7, 10, and 17 min) are used to determine model constants, followed by an independent blind test on the fourth data point (22 min). Results show that the prediction error of the frustum model for the fourth point is less than 0.20%, representing an approximately 90% enhancement in precision compared to the conventional model. This research reveals that thermal diffusion is the core physical mechanism driving pit evolution, providing a robust theoretical foundation and quantitative tool for process optimization, uniformity prediction, and process window design in EBE applications.