DocumentCode
3259337
Title
Constraint-based voltage island partitioning
Author
Sengupta, Dipanjan ; Saleh, Resve
Author_Institution
Dept. of ECE, UBC, Vancouver, BC
fYear
2007
fDate
5-8 Aug. 2007
Firstpage
1050
Lastpage
1053
Abstract
Among the different methods of reducing power, the voltage island technique is gaining in popularity for core- based SoC design. Assigning cores to the different supply voltages and early voltage island analysis are two key steps in the overall design process. In previous work, each iteration involved a heuristic assignment of cores to voltage islands followed by an expensive floorplanning step. Multiple iterations were required to obtain a solution. However, as the number of cores increases, the floorplanning stage becomes the bottleneck. In this paper, a constraint-based approach is used to identify the best candidates for floorplanning, thus reducing the overall runtime. Starting from a voltage assignment table, we propose a methodology to quickly find an ordered set of feasible voltage assignments. The solutions that satisfy the power constraints are identified and analyzed using a multi- VDd power grid representation to find the optimal amount of decoupling capacitance required to satisfy the supply noise constraint. Next, the solutions that satisfy a total area constraint are selected. Finally, the solutions that satisfy power, area and supply noise constraints are fed to the floorplanner in sorted order until the first acceptable solution is found.
Keywords
constraint handling; logic design; system-on-chip; SoC design; constraint based; decoupling capacitance; ordered set; power constraints; power grid representation; supply noise constraint; voltage assignment table; voltage island partitioning; Capacitance; Cost function; Genetic algorithms; Power grids; Power supplies; Process design; Runtime; Simulated annealing; System-on-a-chip; Voltage fluctuations;
fLanguage
English
Publisher
ieee
Conference_Titel
Circuits and Systems, 2007. NEWCAS 2007. IEEE Northeast Workshop on
Conference_Location
Montreal, Que
Print_ISBN
978-1-4244-1163-4
Electronic_ISBN
978-1-4244-1164-1
Type
conf
DOI
10.1109/NEWCAS.2007.4488000
Filename
4488000
Link To Document