Stochastic Mixed-PR: A Stochastically-Tunable Low-Error Adder

Approximate computing is a promising paradigm in low-power design to trade off power efficiency by accuracy. However, its usability in real applications is restricted due to the lack of a dynamic configuration of the error characteristics. Most of the existing approximate adders have a fixed level of accuracy. On the other hand, the approximate adders which are configurable can only switch between an exact mode and a fixed level of approximation. This brief presents the mathematical stochastic error analysis of the adders, for the first time. Furthermore, based on the analysis, we propose to use a mixed adder as a reconfigurable architecture. This adder outperforms the state-of-the-art configurable adder. Moreover, it reduces the energy-delay product considerably in comparison with its conventional counterpart.