Multi-Objective Evolutionary Algorithms for Hardware Design Space Exploration

This work investigates the use of Multi-Objective Evolutionary Algorithms (MOEAs) in the Hardware Design Space Exploration context. Through the use of a heterogeneous data set that spanned different hardware configurations and performance measurements, four widely known MOEAs namely NSGA-II, SPEA2 IBEA MOPSO have been systematically assessed. Experiments have been aimed at major indicators, including convergence, diversity, and hypervolume maximization. It has been shown that NSGA-II and SPEA2 have been largely well-balanced in terms of convergence versus diversity, a feature that makes them suitable options for the optimization task. IBEA has been best in hypervolume maximization, which indicates that it is a powerful method when applied to scenarios where extensive coverage of the design space needs to be achieved. MOPSO, using the ideas from particle swarm optimization, gave a solid competitor that featured excellent convergence and diversity. This work also does a comparative study with related works, which shows the generality of MOEAs in different domains such as facility location optimization, system design and control theory robotics bioinformatics computer vision, and networked systems. The research provides relevant inputs to help researchers and practitioners choose the best algorithms for hardware design optimization, which includes considering specific goals and constraints. Consequently, the research shows us how this way and a method of creation.