Material properties of low pressure chemical vapor deposited silicon nitride for modeling and calibrating the simulation of advanced isolation structures

The increasing cost and complexity of semiconductor process development has lead to the widespread use of multidimensional semiconductor process simulators. The success of a program like SUPREM−IV is primarily due to the fact that it is based on physical models, rather than empirical equations. This is in contrast to the first generation of process simulators, which calculated impurity profiles and oxide thickness in one dimension based on semiempirical approaches. SUPREM−IV incorporates two-dimensional coupled stress-dependent oxidation and impurity diffusion, which allows the accurate simulations of state-of-the-art integrated processes, provided that accurate model parameter sets are available. In this article we present an improved calibration methodology for simulation of advanced isolation technologies using SUPREM−IV, based on the experimental determination of the material properties of silicon nitride. The proposed strategy is applicable not only to SUPREM−IV but to any numerical simulator that uses the stress-dependent oxidation models to calculate oxide growth. In order to simulate experimental isolation boundary shapes, the oxidation models in SUPREM−IV must be calibrated. This requires a set of five fitting parameters, i.e., the material viscosities and activation volumes for stress-dependent diffusion, reaction rate, and critical stress. These parameters form a quintuplet but are not unique. Multiplying the viscosity values and dividing the activation volumes by a constant will yield exactly the same isolation structure boundary shape. The calculated stresses in the substrate however do not remain constant when different quintuplets are used. This has serious implications since isolation structures require the stress levels in the silicon substrate to remain well below the yield stress of silicon. If a nonoptimal parameter set is used, incorrect designs will result. Based on the experimental extraction of the silicon nitride viscosity by measuring the relative film thickness shrinkage rate as a function of temperature, a scaled parameter set has been derived. SUPREM−IV simulations predict identical boundary shapes but the simulated stress levels are different from the values obtained using the default parameter set. In order to validate the stress simulations, a comparison with experimental stress values measured by micro-Raman spectroscopy has been carried out. It has been found that the quintuplet based on the experimental nitride viscosity values yields the best agreement with the experimental stress data, supporting the accuracy of the new model parameter set.