RF-Informed Graph Neural Networks for Accurate and Data-Efficient Circuit Performance Prediction

TL;DR

This paper introduces a graph neural network framework tailored for RF circuit performance prediction, achieving high accuracy with minimal data and strong generalization across different circuit topologies.

Contribution

The work presents a novel topology-aware GNN model that efficiently predicts RF circuit metrics, enabling scalable and accurate RF design automation with minimal training data.

Findings

Achieves an average MRE of 3.45% in RF performance prediction.

Improves class-level generalization by approximately 161 times.

Outperforms state-of-the-art methods by 9.2 times in accuracy.

Abstract

Accurately predicting the performance of active radio frequency (RF) circuits is essential for modern wireless systems but remains challenging due to highly nonlinear, layout-sensitive behavior and the high computational cost of traditional simulation tools. Existing machine learning (ML) surrogates often require large datasets to generalize across various topologies or are not accurate on unseen circuits. This work presents a lightweight, data-efficient, and topology-aware graph neural network (GNN) framework for predicting key performance metrics of active RF circuit classes, such as low-noise amplifiers (LNAs), mixers, voltage-controlled oscillators (VCOs), and power amplifiers (PAs). The proposed framework employs RFIC domain-informed feature indexing to enable cross-topology adaptability by cheap encoding of functional device semantics (e.g., differential pair and varactor transistors) and efficient knowledge transfer. The surrogate model represents circuits using device-terminal graph abstractions to preserve fine-grained connectivity and transistor-level symmetry. The final model is generalized to a wide variety of classes by being trained in parallel. Experimental results demonstrate accurate modeling of multimodal and heavy-tailed RF performance distributions, achieving an average mean relative error (MRE) of 3.45%, an improvement of 9.2x compared to state-of-the-art. Furthermore, the method improves class-level generalization performance by ~161x compared to prior art, demonstrating its effectiveness for scalable and deployment-ready RF design automation.