AI Hardware Oriented Trojan Detection Architecture
Shu Takemoto, Yoshiya Ikezaki, Yusuke Nozaki, Masaya Yoshikawa · 2022
In recent years, AI edge computing has expanded to include the implementation of AI models on edge devices to achieve real-time inference. Also, high-level synthesis is important for optimizing the implementation of AI models and peripheral circuits in hardware. On the other hand, the threat of hardware Trojan has been reported in hardware implementations. Therefore, several detection methods focusing on netlists in hardware description languages have been proposed. However, the Trojan generated by high-level synthesis have not been evaluated. For this background, this study proposes a hardware Trojan detection method oriented to high-level synthesis of AI models. Also, evaluation experiments reveal the detection difficulties of conventional methods and the effectiveness of the proposed method.