Versatile Genetic Algorithm-Bayesian Optimization(GA-BO) Bi-Level Optimization for Decoupling Capacitor Placement

This paper proposes a versatile genetic algorithm-Bayesian optimization(GA-BO) bi-level optimization method, in which BO determines the optimal hyperparameters for GA that optimizes the decoupling capacitor (decap) placement. Through optimizing the GA hyperparameters, the proposed method ensures effective optimization that leads to an optimal PDN design that meets the target impedance with the minimum number of decaps. The proposed method was applied to the hierarchical power distribution network (PDN) of high bandwidth memory (HBM) and verified the performance of two objective functions for GA and BO, respectively, and its stability. Furthermore, to verify its versatility, the proposed method was also applied to a different type of PDN and outperformed the random search method by successfully placing 18 decaps that satisfy the target impedance.