Due to the increasing role of leakage power in CMOS circuit's total power dissipation, leakage reduction has attracted a lot of attention recently. Input vector control (IVC) takes advantage of the transistor stack effect to apply the minimum leakage vector (MLV) to the primary inputs of the circuit during the standby mode. However, IVC techniques become less effective for circuits of large logic depth because theMLV at primary inputs has little impact on internal gates at high logic level.

In this paper, we propose a technique to overcome this limitation by directly controlling the inputs to the internal gates that are in their worst leakage states. Specifically, we propose a gate replacement technique that replaces such gates by other library gates while maintaining the circuit's correct functionality at the active mode. This modification of the circuit does not require changes of the design flow, but it opens the door for further leakage reduction, when the MLV is not effective. We then describe a divideand- conquer approach that combines the gate replacement and input vector control techniques. It integrates an algorithm that finds the optimal MLV for tree circuits, a fast gate replacement heuristic, and a genetic algorithm that connects the tree circuits.

We have conducted experiments on all the MCNC91 benchmark circuits. The results reveal that 1) the gate replacement technique itself can provide 10% more leakage current reduction over the best known IVC methods with no delay penalty and little area increase; 2) the divide-and-conquer approach outperforms the best pure IVC method by 24% and the existing control point insertion method by 12%; 3) when we obtain the optimal MLV for small circuits from exhaustive search, the proposed gate replacement alone can still reduce leakage current by 13% while the divide-and-conquer approach reduces 17%.