A filter design for blind deconvolution to decouple unknown RDF/RTN factors from complexly coupled SRAM margin variations

摘要

This paper demonstrates a blind deconvolution technique for decoupling the two variation factors caused by the Random Telegraph Noise (RTN) and the Random Dopant Fluctuation (RDF). Unlike the non-blind deconvolution, the blind deconvolution has to seek both of the two unknown factors for RTN and RDF simultaneously, given only the information about the overall SRAM margin distribution. This paper proposes a new filter design technique for the Richardson-Lucy (R-L) blind deconvolutions. This allows to enjoy the benefits of the R-L algorithm while avoiding the inherent pitfall or ringing errors even in the blind deconvolution. The relative errors of the blind-deconvolution for RDF and RTN are reduced to less than 1% within only 300-iteration cycles. This is 400-times shorter than the conventional one.