Increasing Redundancy in Reduced Precision Redundancy by Approximation (5RPA) Systems in the Nanoscale

Nanoscale computing systems require redundancy for reliable processing. For operation in the presence of many errors/faults, an increase in redundancy is almost unavoidable; this paper proposes a new design for 5-input Reduced Precision redundancy by Approximation (RPAs). Two schemes (one-exact four-approximate data words RPA (1E4A) and four-exact one-approximate data types (4E1A)) are proposed. As per accuracy, the probabilities of getting the exact data word (P) at the output for each scheme under the scenarios of two errors on the most (U) and least significant parts (L) of a data word are analytically found. An excellent agreement between theoretical and simulated results is found; circuit-based metrics such as delay, power dissipation and area are compared to previous RPA-based schemes found in technical literature. A comparative analysis based on the redundancy in these schemes (and the increase in redundancy for the proposed 5RPA) is pursued; it shows that in the presence of multiple errors the proposed schemes attain substantial improvements in accuracy at the output even when the same redundancy is introduced. This conclusion suggests that the redundancy required at nano scales can be adjusted using RPA schemes to satisfy the reliable delivery of accurate computation results.