Deep Neural Network-Based Surrogate-Assisted Inverse Optimization for High-Speed Interconnects

Abstract

Deep neural networks (DNNs) have been broadly adopted in efficiently modeling and optimizing the signal integrity of high-speed interconnects. However, using DNNs could cause inaccuracies in modeling and inverse optimization of multidimensional design parameters. In this article, we propose a novel method to enhance the accuracy of modeling and optimization using ensemble learning and a surrogate-assisted optimization approach. First, an ensemble of DNN models instead of a single DNN is trained to enhance the modeling accuracy. Based on the trained DNN ensemble, inverse optimization of interconnects’ multiple design parameters can be efficiently achieved. To address possible inaccuracies and finetune the inverse optimization results, a surrogate-assisted local optimization (SALO) approach is proposed. Based on the SALO method, more accurate optimization results can be achieved using only a few extra simulations, enabling highly efficient and accurate optimization of high-dimensional design parameters for high-speed interconnects.