Scholarlink

<p>Since, the initial proposition of Hardware Trojans' potential existence in 2007, side-channel technology has been extensively employed in research concerning Hardware Trojan detection. Concurrently, the ongoing advancements in integrated circuit technology have facilitated the widespread adoption of deep learning methodologies in the Hardware Trojan detection domain. Notably, in 2020, Google’s introduction of the vision transformer ignited renewed interest in leveraging Deep Learning for hardware security. In 2021, Apple introduced the MobileViT network, which is a lightweight, versatile, and low latency network. But, it is crucial to acknowledge that as the complexity of Deep Learning model architectures deepens, there is a corresponding rise in the number of model parameters. Moreover, MobileViT faces challenges in extracting significant features and demonstrates limited sensitivity towards channel and spatial information during image feature extraction. Hence, we propose the TA-MobileViT lightweight model, which combines the almost parameterless attention mechanism—triplet attention (TA) and the lightweight universal MobileViT. Without expanding the model's parameter count, it acquires cross-channel interaction capability, improves its focus on critical information, and enhances both classification accuracy and generalization ability. By combining convolutional blocks and transformer blocks, the model has both the ability to extract local features and the ability to extract global features, thereby improving the model's expressiveness and performance, and achieving higher accuracy in detection. The experimental results show that TA-MobileViT can quickly and efficiently detect Hardware Trojans. When detecting a single Trojan, the recognition accuracy of three types reaches 100%, while the detection accuracy of AES-600 reaches the highest of 72.2%. When detecting multiple types of Hardware Trojans. Our model’s mean accuracy reached the highest of 97.04%. Compared with other deep learning methods, this network model has fewer parameters and more competitive classification accuracy. The code is available at <a href="https://github.com/GaunX/TA-MobileViT/tree/master">https://github.com/GaunX/TA-MobileViT/tree/master</a>.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A lightweight Hardware Trojan detection approach in the waveform diagram based on MobileViT and attention mechanism

  • 期刊论文

  • 发表日期：

    2025年3月5日

查看全文
  • Shouhong Chen,
  • Guanxiang Qin,
  • Ying Lu,
  • Tao Wang,
  • Xingna Hou

摘要

Since, the initial proposition of Hardware Trojans' potential existence in 2007, side-channel technology has been extensively employed in research concerning Hardware Trojan detection. Concurrently, the ongoing advancements in integrated circuit technology have facilitated the widespread adoption of deep learning methodologies in the Hardware Trojan detection domain. Notably, in 2020, Google’s introduction of the vision transformer ignited renewed interest in leveraging Deep Learning for hardware security. In 2021, Apple introduced the MobileViT network, which is a lightweight, versatile, and low latency network. But, it is crucial to acknowledge that as the complexity of Deep Learning model architectures deepens, there is a corresponding rise in the number of model parameters. Moreover, MobileViT faces challenges in extracting significant features and demonstrates limited sensitivity towards channel and spatial information during image feature extraction. Hence, we propose the TA-MobileViT lightweight model, which combines the almost parameterless attention mechanism—triplet attention (TA) and the lightweight universal MobileViT. Without expanding the model's parameter count, it acquires cross-channel interaction capability, improves its focus on critical information, and enhances both classification accuracy and generalization ability. By combining convolutional blocks and transformer blocks, the model has both the ability to extract local features and the ability to extract global features, thereby improving the model's expressiveness and performance, and achieving higher accuracy in detection. The experimental results show that TA-MobileViT can quickly and efficiently detect Hardware Trojans. When detecting a single Trojan, the recognition accuracy of three types reaches 100%, while the detection accuracy of AES-600 reaches the highest of 72.2%. When detecting multiple types of Hardware Trojans. Our model’s mean accuracy reached the highest of 97.04%. Compared with other deep learning methods, this network model has fewer parameters and more competitive classification accuracy. The code is available at https://github.com/GaunX/TA-MobileViT/tree/master.