Keywords

netlist reverse engineering, graph neural networks, fpga, ip protection, design assurance

Abstract

Netlist reverse engineering enables many applications, including detecting IP theft, verifying CAD tool correctness, and detecting hardware trojans. However, reconstructing high-level information and circuit structure from a flat, nameless netlist is challenging. In this work we focus on the problem of locating known IP cores in an FPGA netlist, which is especially challenging due to the prevalence of highly configurable IP cores. We present Neurocore: a graph neural network-based approach to classifying nodes in a netlist as instances of known IP cores, and present and evaluate different models for different use cases. We have created a large open-source dataset of thousands of FPGA designs with known IP cores, and demonstrate that our approach can obtain 80–98% accuracy, depending on the model and the use case.

Document Type

Peer-Reviewed Article

Publication Date

2026-02-03

Language

English

College

Ira A. Fulton College of Engineering

Department

Electrical and Computer Engineering

University Standing at Time of Publication

Graduate Student

Share

COinS