| Summary: | 碩士 === 國立成功大學 === 電機工程學系碩博士班 === 95 === In the thesis, we present two fast algorithms for simultaneous scheduling and assignment of multiple supply voltages and multi-threshold voltages in high level synthesis. We consider not only dynamic energy consumption but also static energy consumption for data-dominant applications. The first algorithm uses Energy-Sensitive Move-Based (ENSEMBA) scheduling method to minimize the energy consumption of functional units simultaneously under constraints. The second algorithm, Aggressive Energy-Sensitive Move-Based (AENSEMBA) scheduling, is an enhanced version of ENSEMBA. It can handle energy consumption related to not only functional resources but also registers as well as level converters. The two algorithms can handle the scheduling of operations, the assignment of supply voltages and threshold voltages under user-define timing and area constraints. ENSEMBA is the core base of AENSEMBA, ENSEMBA can give users enough information to schedule operations and assign voltages and it can also prove that our design concept is efficient. AENSEMBA can handle more issues than ENSEMBA and it can give more information to back-end EDA tools to speed up the IC design time. ENSEMBA algorithm can save energy consumption from 20% to 41% for DSP test cases in technology process 180nm and from 67% to 96% for the same DSP test cases in technology process 32nm. AENSEMBA algorithm can save energy consumption from 16% to 40% for 180nm and from 65% to 95% for 32nm.
|
|---|
