Power optimization in data-path scheduling and binding with multiple supply voltages and threshold voltages by simulated annealing

We present a scheme to optimize power consumption in data-path scheduling and binding with resources operating at multiple supply voltages and threshold voltages by simulated annealing. The proposed scheme considers both scheduling and binding simultaneously such that we have a wider solution space to explore and a solution with much lower power can be obtained. Besides, we give a more accurate high-level power model so that not only the dynamic and leakage power of function units but also the sharing of function units is taken into account. Experimental results on a number of high-level benchmark circuits using three supply voltage and threshold voltage levels show that an average power saving of about 68% can be obtained compared to using a single supply and threshold voltage level (with a time constraint of 1.2 times the critical path delay and a resource constraint of two function units of each type).