(A) deep Q-learning based optimal decoupling capacitor design method for simultaneous switching noise (SSN) suppression in 2.5-D/3-D IC

Abstract

In this paper, we first propose a deep reinforcement learning based optimal decoupling capacitor design method. Based on the deep reinforcement learning, design automation is implemented with only the structure of the power distribution network and the capacitance per unit cell. The proposed method aims to suppress the simultaneous switching noise of 2.5-D / 3-D ICs, and is the first design method that considers the multiple level power distribution networks of 2.5-D / 3-D IC simultaneously. In order to verify the proposed methodology, the results of applying the proposed method to the test hierarchical power distribution network and the results of the conventional simulation methods are compared. In addition, the proposed method is applied to high bandwidth memory for optimal decoupling capacitor design, and the suppression of power distribution network impedance and simultaneous switching noise are simulated and analyzed.