A regression based methodology to estimate SNM for improving yield of 6T SRAM

The stability of SRAM cells in high density ICs during read cycle is an extremely critical performance metric for data retention. A frame work to identify the tradeoff between Yield and Static Noise Margin (SNM) in the region of high sigma tail end distributions is presented in this paper. For developing the framework, the sensitivity of SNM is observed for different device variations using Design of Experiments (DoE) method whereas estimation of yield for different targeted SNMs is done using Bivariate Linear (BL) model and Bivariate Nonlinear Quadratic (BNLQ) models. The developed framework demonstrates that for achieving a yield of 99% there is a need to design a memory cell at a cost of 150mV SNM using BL model post silicon. On the other hand at the same cost of 150mV SNM, the yield can be enhanced to 99.5% if BNLQ model is adopted post silicon.