VLSI designs are typically data-independent and as such, they must produce the correct result even for the worst-case inputs. Adders in particular assume that addition must be completed within prescribed number of clock cycles, independently of the operands. While the longest carry propagation of an n-bit adder is n bits, its expected length is only O(log_2 n) bits. We present a novel dual-mode adder architecture that reduces the average energy consumption in up to 50%. In normal mode the adder targets the O(log_2 n)-bit average worst-case carry propagation chains, while in extended mode it accommodates the less frequent O(n)-bit chain. We prove that minimum energy is achieved when the adder is designed for O(log_2 n) carry propagation, and present a circuit implementation. Dual-mode adders enable voltage scaling of the entire system, potentially supporting further overall energy reduction. The energy-time tradeoff obtained when incorporating such adders in ordinary microprocessor's pipeline and other architectures is discussed.
展开▼
