Hardware Trojan Detection: CNN-LSTM Model with Character and Word Embedding-based NLP Techniques

• Zhiqiang Wang,
• Ying Liu,
• Zhenlong Dong,
• Hang Wu,
• Lei Ju,

Vol. 20, No. 3, March 31, 2026
10.3837/tiis.2026.03.018, Download Paper (Free):

• Hardware Trojan detection
• Convolutional Neural Network
• Long short-term memory
• NLTK word segmentation
• Natural Language Processing

Abstract

In recent years, the widespread use of integrated circuits (IC) has heightened the risk of supply chain vulnerabilities being exploited by Hardware Trojans. Detecting hardware Trojans is challenging due to the lack of unified standards, lengthy detection times, and high resource consumption. This paper proposes a hardware Trojan detection model combining Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) networks, along with a vectorization algorithm integrating character embedding and word embedding. The dataset was obtained from a public hardware security platform. Trojan samples were preprocessed using the NLTK tokenization tool, while normal samples were generated through a Generative Adversarial Network (GAN), thereby constructing a comprehensive dataset. Dimensionality reduction is performed via vector embeddings, transforming each sample into a two-dimensional feature matrix. The experimental results demonstrate that the proposed model achieves a detection accuracy of 82.108%, representing an improvement of 2.451% to 13.235% compared to existing models applied in hardware Trojan detection.

Statistics

Cite this article