Soft NMR: Exploiting Statistics for Energy-Efficiency

Eric Kim,  Rami Abdallah,  Naresh Shanbhag
University of Illinois at Urbana Champaign


Achieving energy-efficiency in nanoscale CMOS process technologies is made challenging due to the presence of process, temperature and voltage variations. In this paper, we present soft N-modular redundancy (soft NMR) that consciously exploits statistics of errors due to these nanoscale artifacts in order to design robust and energy-efficient systems. In contrast to conventional NMR, soft NMR employs estimation and detection techniques in the voter. We compare NMR and soft NMR in the design of an energy-efficient and robust discrete cosine transform (DCT) image coder. Simulations in a commercial 45nm, 1.2V, CMOS process show that soft triple-MR (TMR) provides 10X improvement in robustness and 13% power savings over TMR at a peak signal-to-noise ratio (PSNR) of 20dB. In addition, soft dual-MR (DMR) provides 2X improvement in robustness and 35% power savings over TMR at a PSNR of 20dB.