A comparison of fitness function evaluation schedules for multi-objective univariate marginal distribution optimization of mixed analog-digital signal circuits

An increasing complexity of mixed analog-digital signal circuits requires optimization at higher hierarchical level. However, evolutionary optimization of mixed analog-digital signal circuits at the system level results in huge computational costs. A key to manage these computational complexities of evolutionary circuit design is an application of flexible fitness functions evaluation schedules. In this paper we compare the static, dynamic, and co-evolution fitness function evaluation schedules for multi-objective optimization of mixed analog-digital signal circuits at the system level on the base of the univariate marginal distribution algorithm. Experiments for our symmetry recognition circuit benchmark chosen indicate that the dynamic fitness function schedule is a good compromise between computational costs and optimization efficiency.