ECE 556 Project Presentation SAGA-The Unified Algorithm Phillip Heyrman, Matt Kumerow, & Todd Tisch
What is SAGA? Combination of Simulated Annealing (SA) and Genetic Algorithm (GA) Begins as GA and gradually becomes SA The speed of the conversion is adaptive because it is determined by progress Parameters can be setup to run the algorithm in pure SA or GA mode
Why SAGA? Has a large number of modifiable parameters that can be tuned for program performance Past test results have shown that SAGA performs equal to or better than most genetic algorithms Application independent
The SAGA Algorithm….
SAGA Objectives No pairs of cells overlap each other The rectangle B encloses all cells and has approximate horizontal length W The area within B which is not occupied by cells in sufficiently large to contain all routing needed to implement the required interconnections
SAGA’s expected inputs Set of rectangular cells each with a number of terminals at fixed positions along the edges of the cell Netlist specifying the interconnections of all terminals Approximate horizontal length W of the chip
Input File Format Line 1: <number of cells> <number of wires> Line 2 to number of cells + 1: <height> <length> <pins> <x pin location> <y pin location> <connects to>
SAGA’s Computations Absolute position of each cell Orientation and possible reflection(s) of each cell Rectangle B defining the shape of the chip
Output Data Matlab file for graphing of final cell placement Output file that contains total wire length, final position and orientation of cells, total CPU time, final bounding box