Low-Power Filtering via Minimum Power Soft Error Cancellation (MP-SEC)

In this paper, an energy-efficient estimation and detection problem is formulated for low-power digital filtering. Building on the soft digital signal processing technique proposed by Hegde et al., that combines algorithmic noise tolerance (ANT) and voltage scaling to reduce power, the proposed minimum power soft error cancellation (MP-SEC) technique detects, estimates and corrects transient errors that arise from voltage over-scaling. These timing violation-induced errors, called soft errors, can be detected and corrected by exploiting the correlation structure induced by the filtering operation being protected, together with a reduced-precision replica of the protected operation. By exploiting a spacing property of soft errors in certain architectures, MP-SEC can achieve up to 30% power savings with no SNR loss and up to 55% power savings with less 1 dB SNR loss, for an example 25-tap frequency-selective filter. Index Terms Low power, digital filter, supply voltage scaling, overscaling, soft error, algorithmic noise tolerance