Robust analog/RF circuit design via Cycle-Consistent Generative Adversarial Networks

In this paper, we propose a novel method based on Cycle-Consistent Generative Adversarial Networks (Cycle-GAN) to efficiently synthesize robust analog/RF circuits. The key idea is to learn a mathematical mapping between nominal and robust designs by using a Cycle-GAN, which can be used to convert a given nominal design to its robust version with great computational efficiency. The proposed Cycle-GAN is learned from a large number of nominal designs synthesized by EDA tools and a small number of robust circuits manually designed by human experts. Hence, it is expected to appropriately incorporate the human design knowledge that is often difficult to capture by other state-of-the-art methods. Two circuit examples demonstrate that the proposed approach can accurately synthesize robust analog/RF circuits with low computational cost.