Thermal oxidation of silicon is a basic process in integrated circuit fabrication. Its application in advanced technologies, such as shallow trench oxidation, active corner rounding, ultrathin gate oxide growth, and some others, provides device features especially critical to product quality. 1 Another example is local oxidation of silicon (LOCOS), which has been used as the major isolation scheme for three decades, exhibiting a unique combination of process robustness, simplicity, and low cost. LOCOS isolation is still applied in ultralarge scale integration (ULSI) device manufacturing but requires careful tuning. 2-4 Its optimization can be efficiently carried out only with the assistance of process simulation means. [5][6][7] Existing two-and three-dimensional process simulators employ elaborate stress-dependent viscous and viscoelastic oxidation models. [8][9][10][11][12][13] In these models, viscosities of silicon oxide and nitride, oxidant diffusivity, and oxidation reaction rate are functions of stress. The utilization of commercially available process simulators using default values of oxide and nitride viscosities often results in a significant error in the simulated isolation structure geometry and stress. Therefore, it is recommended to calibrate the parameters characterizing viscous properties of these materials for each relevant technology. Fitting the shapes of isolation structures obtained by simulation to the microscopy images of actual devices is usually employed for this calibration. 6,14 A layer of low-pressure chemical vapor deposited (LPCVD) silicon nitride is used in LOCOS as a mask for protecting active areas from oxidation. Depending on deposition conditions, the structure and therefore mechanical properties of silicon nitride may differ significantly. For example, Griffin and Rafferty reported a higher viscosity for silicon-rich nitride films. 10 While most mechanical parameters of Si 3 N 4 films can be determined quite accurately using wellestablished methods, there is no standard approach to obtaining the value of nitride viscosity. Several researchers have derived it through a comparison of simulated and processed LOCOS structures. 6,9,15 Rational measurements of nitride viscosity were recently performed by Smeys et al. 16 They measured nitride film stress and decrease in film thickness after high-temperature annealing. Nitride viscosity was calculated under the assumption that film shrinkage occurred only due to viscous flow of nitride. However, since the nature of this shrinkage might be different, especially at the beginning of an anneal, the reported nitride viscosity values can be underestimated.In this work, we have conducted a series of experiments with nitride films involving two different high-temperature anneal schemes. The changes in film thickness and intrinsic stress were measured. The values for nitride viscosity at anneal temperatures were calculated based on the viscoelastic approximation. The contribution of the initial stage of annealing to film shrinkage was excluded...