Blind deconvolution technique for extracting unknown two factors of RTN and truncated RDF from given target for overall SRAM margin variations

This paper proposes for the first time a blind deconvolution technique for extracting the unknown two variation factors of the Random Telegraph Noise (RTN) and the truncated Random Dopant Fluctuation (RDF) solely from the given target for SRAM margin variations. Unlike the non-blind deconvolution, the blind deconvolution has to extract the both of the two unknown factors of RTN and truncated RDF simultaneously, that can be sort of ill-posed problem. The proposed algorithm features a sequentially-dual iteration loop and an adaptively segmented forward-problem based blind deconvolution (DIAS-BDCV) process. This allows a free of convergence error in the optimization process. This effectiveness has been demonstrated for the first time with applying to a real SRAM design analysis. It has been shown that the proposed DIAS-BDCV technique allows: (1) a free of convergence-error and local-minimum-error in blind deconvolution even if the total number of parameters to be sought in the optimization problem exceeds 20, and (2) a low enough blind deconvolution errors of the RTN and RDF comparable to the level (<; 10^-13) of the non-blind one.