Deep reinforcement learning-based through silicon via (TSV) array design optimization method for high bandwidth-memory (HBM) considering signal integrity (SI)

Abstract

In this paper, we propose the through silicon via (TSV) array design optimization method using deep reinforcement learning (DRL) framework. The agent trained through the proposed method can provide an optimal TSV array that maximizes the eye height of TSV channels in one single step. We define the state, action, and reward that are parameters of the Markov decision process (MDP) for optimizing the TSV array considering signal integrity (SI) and train a deep q network (DQN) agent. The proposed method was compared with the entire search algorithm in a 3x3 TSV array in terms of time required and performance for verification. The verified method was applied to the optimal design of a 4x4 TSV array including a 1-byte signal channel of HBM.