Spec-Graph Contrastive Learning for Early Detection of Hardware Trojans in Open-Source RTL Designs

DOI

10.2991/978-94-6239-693-7_112How to use a DOI?

Keywords

Hardware Trojan; Open-Source Hardware; RISC-V; Verilog; RTL Design; Static Code Analysis; Machine Learning; NLP Embeddings; Hybrid Embedding; Gradient Boosting; LLM-Guided Trojan Injection; Hardware Security

Abstract

The increasing popularity of open-source hardware has, unfortunately, also made it easier to sneak in malicious changes – specifically, hardware Trojans that can hide really well from normal checks. The usual ways of finding these Trojans often depend on simple structural rules, basic code analysis, or pre-trained classifiers. The problem is, these methods don’t always work when facing new types of Trojans or really complicated designs.

That’s why we’ve developed something new: the Spec–Graph Contrastive Trojan Detector (SGCTD). Think of it as a three-pronged approach that looks at the design from multiple angles: the written specifications, the actual RTL code, and the overall design structure as a graph. Our method uses a technique called contrastive learning to match what the specifications *say* the design should do with how it’s *actually* built. We also have a special tool that only needs to see clean designs to learn what’s normal, so it can spot anything that looks out of place. And to make sure even the sneakiest Trojans don’t get by, we’ve added a module that figures out how easily rare events can set off critical parts of the circuit.

We put SGCTD to the test on some standard benchmarks, as well as RISC-V and cryptographic accelerators that we “infected” with synthetic Trojans. The results? SGCTD outperformed existing methods, especially when it came to finding new, unseen Trojans. Plus, it pointed directly to the suspicious parts of the design, which makes it easier to understand *why* it thinks something is wrong. We think this specification-aware, graph-driven approach is a big step forward in keeping open-source hardware safe and secure.

Copyright

© 2026 The Author(s)

Open Access

Open Access This chapter is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International License (http://creativecommons.org/licenses/by-nc/4.0/), which permits any noncommercial use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.

Download article (PDF)

TY  - CONF
AU  - B. Venkata Shivaiah
AU  - N. Siva
AU  - Shaik Arshiya Anjum
AU  - Greeshma Pothu
AU  - V. Rohith
AU  - T. Chaitanya
PY  - 2026
DA  - 2026/06/16
TI  - Spec-Graph Contrastive Learning for Early Detection of Hardware Trojans in Open-Source RTL Designs
BT  - Proceedings of the International Conference on Intelligent Systems for a Sustainable Future (ISSF 2026)
PB  - Atlantis Press
SP  - 1166
EP  - 1175
SN  - 2589-4919
UR  - https://doi.org/10.2991/978-94-6239-693-7_112
DO  - 10.2991/978-94-6239-693-7_112
ID  - Shivaiah2026
ER  -