Deep Reinforcement Learning-based Power Distribution Network Structure Design Optimization Method for High Bandwidth Memory Interposer

Abstract

In this paper, we propose a deep reinforcement learning (DRL)-based power distribution network (PDN) structure design optimization method for high-bandwidth memory (HBM) interposer. Due to the usage of multi-voltage PDNs and various components in limited PDN space, the design optimization of interposer PDN is required. The proposed method provides an optimal PDN shape and area that can satisfy target impedance. For the verification of the proposed method, the initial PDN and optimized PDN using the proposed method are compared in terms of PDN impedance and PDN shape. We successfully optimize the PDN shape and area while satisfyings target impedance. By applying the proposed method to test interposer PDN, about 23% of the area was saved for design constraint such as the expansion of other PDNs and placement of various components.