Imitation Learning-based Equalizer Design Optimization Method on PCIe 6.0

In this paper, we proposed imitation learning-based equalizer design optimization method on PCIe 6.0. With each update of PCIe, the equalizer scheme has evolved to compensate for channel loss. However, due to the increasing number of equalizer parameters and their inter-coupling, co-optimizing these parameters becomes essential. While various methods have been researched to optimize equalizer parameters, they often require re-optimization for different conditions or they inefficiently train the DNN policy used for optimization. In contrast, our method utilizes imitation learning scheme. We collected training data offline through data exploration using expert policies like Bayesian optimization (BO) and random search (RS), and subsequently trained the DNN policy to mimic collected data. For validation, we compared the PAM-4 eye opening of our proposed method with that of BO and RS. As a result, the performance of our proposed method, with just a single simulation, surpassed that of both RS and BO, each simulated 100 times.