R-INN: An Efficient Reversible Design Model for Microwave Circuit Design

作者：Yang Zhang, Yi Pin Xu, Lingrui Shen, Zhijin Chen, Hongcai Chen · 发表于：IEEE Transactions on Microwave Theory and Techniques · 年份：2025 · DOI：10.1109/tmtt.2025.3581947 · 被引用次数：3 · 研究领域：Microwave Engineering and Waveguides、Radio Frequency Integrated Circuit Design

To address the increasing complexity and cost challenges in reversible circuit design, we propose an enhanced invertible neural network (INN) model called R-INN, which integrates the real nonvolume preserving (Real NVP) transformation. By precisely deriving the posterior distribution, this model not only improves the accuracy of forward predictions but also provides a range of probability density functions for inverse circuit design, thus achieving a reversible design paradigm. The model demonstrates exceptional efficiency and accuracy when handling circuit design parameter spaces with dimensions as high as 200. The superiority of the R-INN is validated through different circuit designs, including an active mixer, a high-speed link, and a dual-band microwave filter. The comparison results have confirmed the model’s superior generalization capabilities and practical value in multivariate reverse design problems. Additionally, the reversible nature of the model offers multiple solutions for the same output specifications, providing designers with more flexibility in design exploration.