Evolutionary Design of Large Approximate Adders Optimized for Various Error Criteria

Approximate computing, genetic algorithm, equivalence checking

As a promising approach to the design of energy efficient circuits, approximate circuits and approximate circuit design methodologies have attracted a significant attention of researchers as well as industry. Compared to the traditional design methods, it has been demonstrated that evolutionary approaches are able to discover approximate circuits exhibiting a good trade-off between the energy consumption and circuit quality. In this work, evolutionary design of large approximate adders is addressed. In order to improve scalability, the quality of the candidate solutions is analysed using a formal approach based on Binary Decision Diagrams. Compared to the common approach based on a parallel circuit simulator, the proposed method is able to evaluate 2-3 orders of magnitude more generations.