| Summary: | 碩士 === 國立臺灣大學 === 電子工程學研究所 === 95 === Power optimization is a crucial concern for modern circuit designs, and multiple supply voltages (MSV’s) provide an effective technique for the optimization. This thesis addresses a voltage partitioning problem arising in MSV design during high-level synthesis. We point out a theoretical mistake in a recent publication and prove that the partitioning problem is NP-hard. Despite its NP-hardness, we propose an efficient α2-approximation algorithm for the problem, where α is the constant ratio
of the maximum to the minimum voltages. Compared with the previous work that runs in O(dn2) time, the time complexity of our algorithm is only O(dkn), where d is the number of voltages employed in the final designs (i.e., voltage domains), k is the number of available supply voltages in the technology library, and n is the number of functional units. Note that both d and k can be considered as small constants for practical applications. Experimental results show that our algorithm (with empirical O(n0.87) and O(n0.93) time) can achieve very significant run-time speedups than the recent work (with empirical O(n2.02) and O(n2.08) time), with the same power reductions of about 60% by using dual-Vdd’s and 67% by using triple-Vdd’s.
|
|---|
