A Siamese deep learning framework for efficient hardware Trojan detection using power side-channel data.

Area of Science:

• Computer Engineering
• Cybersecurity
• Artificial Intelligence

Background:

• Hardware Trojans (HTs) pose significant security risks to integrated circuits (ICs), enabling unauthorized access and data theft.
• Effective detection of HTs is crucial for maintaining the integrity and security of ICs.
• Existing detection methods often require a 'golden model', limiting their applicability.

Purpose of the Study:

• To propose a novel, non-destructive Hardware Trojan Detection (HTD) framework using Siamese neural networks (SNNs).
• To evaluate the performance of SNNs integrated with different neural network architectures (CNN, GRU, LSTM) for HTD.
• To detect HTs by analyzing power side-channel signals without relying on a golden model.

Main Methods:

• Developed a Siamese neural network (SNN) framework for Hardware Trojan Detection (HTD).
• Integrated Convolutional Neural Network (CNN), Gated Recurrent Unit (GRU), and Long Short-Term Memory (LSTM) models with the SNN framework.
• Trained the models using features extracted from the Trojan Power & EM Side-Channel dataset.

Main Results:

• The Siamese LSTM model achieved the highest detection accuracy at 86.78%.
• The Siamese GRU model demonstrated strong performance with 83.59% accuracy.
• The Siamese CNN model reached 73.54% accuracy, indicating varying effectiveness of different architectures.

Conclusions:

• The proposed Siamese LSTM framework is a highly promising approach for non-destructive Hardware Trojan Detection (HTD).
• This method effectively detects HTs using power side-channel signals, eliminating the need for a golden model.
• The SNN-based approach outperforms existing state-of-the-art methods in HTD.