80 likes | 229 Views
Genetic Approach to Standard Cell Placement (GASP). Using Meta-Genetic Parameter Optimization. What is a Genetic Algorithm?. Maintains a pool of solutions called the population Generates new solutions by combining genes from two parents. Parents are selected according to fitness
E N D
Genetic Approach to Standard Cell Placement (GASP) Using Meta-Genetic Parameter Optimization
What is a Genetic Algorithm? • Maintains a pool of solutions called the population • Generates new solutions by combining genes from two parents. Parents are selected according to fitness • Fitness of a solution is 1/cost • Tries to mimic biological evolution
Important Terminology • Crossover & Crossover rate (PMX) – Crossover is the operation that generates new solutions from parents. • Mutation & Mutation rate – Mutation introduces new genes or tries lost genes in a new context • Inversion & Inversion rate – Only Changes Representation of Solution.
Implementation Details • Data Structures: Each solution is 3 arrays, connectivity list, span structures • Crossover: PMX • Cost function: Semi-perimeter • Population is sorted according to cost (not fitness!)– Many benefits to sorting! • Np = 24, RC = .34, RM=.005, RI = .15, weighted random parent selection.
What is meta-genetic parameter optimization? • We want to find the best crossover rate, mutation rate, population size, inversion rate, parent selection technique, etc • Run a genetic algorithm on the GASP algorithm to find the best parameters. • Best results: NP= 24, RC= .33, RI= .15, RM = .005, Crossover = Cycle
Performance of GASP • Compared to TW ~ similar run time • Same cost function • GASP returns results which are 5%-20% better than TimberWolf • GASP IS impressive!