Deep Reinforcement Learning (DRL) Based Signal Integrity (SI) Performance Optimization Method for Low-Power Double Data Rate (LPDDR) Memory

In this paper, we propose deep reinforcement learning (DRL) based LPDDR memory package optimization method for maximize SI performance. A Markov Decision Process (MDP) is designed to reflect the characteristics of the LPDDR package, employing SI metrics to capture the influence of SoCs with unknown design variables. Furthermore, Agent is designed using multi-stage multi-layer perceptron to handle multiple interrelated design parameters, and is trained on randomly generated SoC packages to address various SoC scenarios. The proposed DRL based optimization method is verified in 6 coupled channels. The optimization performance of the proposed method is confirmed in various SoC channel. The superb optimality performance is also confirmed through comparison with conventional optimization method, and other DRL method.