A Cost-Driven Method for Deep-Learning-Based Hardware Trojan Detection

Area of Science:

• Computer Science
• Electrical Engineering
• Cybersecurity

Background:

• Cyber-physical systems and the Metaverse face increasing hardware security threats, particularly hardware Trojans in integrated circuits.
• Existing hardware Trojan detection methods struggle with large-scale integration due to limitations like golden chips and high computational demands.
• Traditional machine learning methods for hardware Trojan detection are often unstable due to difficulties in manual feature extraction.

Purpose of the Study:

• To propose a novel deep learning-based multiscale detection model, MHTtext, for automatic hardware Trojan feature extraction and identification.
• To develop strategies within MHTtext that balance detection accuracy with computational efficiency for practical applications.
• To introduce a new evaluation metric, the stabilization efficiency index (SEI), for assessing the model's performance and stability.

Main Methods:

• The MHTtext model employs deep learning for automatic feature extraction from netlist data.
• Two distinct strategies (global and local) are implemented to cater to different accuracy and computational requirements.
• TextCNN is utilized for hardware Trojan identification, with mechanisms to ensure non-repeated component information for enhanced stability.

Main Results:

• The global strategy of MHTtext achieved an average accuracy of 99.26% in detecting hardware Trojans on benchmark netlists.
• The MHTtext model demonstrated high stability and flexibility, with one strategy ranking first in SEI among comparison classifiers.
• The local strategy also yielded excellent results, demonstrating the model's effectiveness across different operational modes.

Conclusions:

• The proposed MHTtext model offers a stable, flexible, and accurate solution for hardware Trojan detection in large-scale integrated circuits.
• The dual-strategy approach allows for adaptable performance based on specific application needs, addressing limitations of traditional methods.
• MHTtext contributes to enhancing the security of critical hardware components within evolving digital environments like the Metaverse.