A fast TCAD-based methodology for Variation analysis of emerging nano-devices

Variability analysis of nanoscale transistors and circuits is emerging as a necessity at advanced technology nodes. Technology Computer Aided Design (TCAD) tools are powerful ways to get an accurate insight of Process Variations (PV). However, obtaining both fast and accurate device simulations is impractical with current TCAD solvers. In this paper, we propose an automated output prediction method suited for fast PV analysis. Coupled with TCAD simulations, our methodology can substantially reduce the time complexity and cost of variation analysis for emerging technologies. We overcome the simulation obstacles and preserve accuracy, using a neural network based regression to predict the output of individual process simulations. Experiments indicate that, after the training process, the proposed methodology effectively accelerate TCAD-based PV simulations close to compact-model-based simulations. Therefore, the methodology can be an excellent opportunity in enabling extensive statistical simulations such as Monte-Carlo for emerging nano-devices.