The as-grown-generation (AG) model: A reliable model for reliability prediction under real use conditions

Abstract

Modeling the negative bias temperature instability (NBTI) can optimize circuit design. Several models have been proposed and all of them can fit test data well. These models are extracted typically by fitting short accelerated stress data. Their capability to predict NBTI aging outside the test range has not been fully demonstrated. This predictive capability for long term aging under low operation bias is what needed by circuit designers. In this work, we test the predictive capability of the well-known reaction-diffusion (RD) based framework for samples fabricated by a variety of processes. Results show that the RD model cannot make an acceptable generic prediction. The recently proposed As-grown-Generation (AG) model is then introduced. By dividing defects into two groups, as-grown and generated defects, and measuring the as-grown defects experimentally, we demonstrate that it can make reliable prediction for the same set of data where the RD model failed.