Towards an appropriate acceleration model for beol tddb

Abstract

TDDB lifetime projections at operating voltages for backend of line (BEOL) dielectrics have been based on accelerated testing at high fields and extrapolation to operating conditions based on electric field dependent dielectric wear-out models. In this paper, we examine the veracity of common TDDB models by using a large set of failure data spanning 3 pitches. Key components of the approach include (i) investigating the universality of failure time data at 3 pitches and different test structure areas using area scaling, (ii) using the universal curve of failure time data to test the acceleration model for 3 pitches and (iii) examining the ability of the models to extrapolate to low field data using high field data for fitting and vice versa. While almost all models seem appropriate when entire set of data (low and high field) is used to fit, fitting only high or low field data and testing the ability to predict low and high fields respectively, shows that power-law and impact damage models extrapolate much better than the root-E model that shows systematic deviation during extrapolation.