This book provides comprehensive coverage of the state-of-the-art in Convolutional Neural Network (CNN) hardware accelerator design, security, and its applications in hardware security. The first part gives a foundational understanding of CNN architectures, emphasizing their computational demands and the necessity for specialized hardware solutions. It also proposes an emulation method with open-source code to mimic CNN hardware accelerator behavior. The second part presents security applications of CNN models, featuring a case study in Network-on-Chip security. It covers threat modeling, countermeasures, and the use of alternative machine learning models to CNNs. The third part explains security threats throughout the AI model production lifecycle, including software vulnerabilities and hardware risks, and explores techniques to enhance the robustness of CNN hardware accelerators, focusing on preventing hardware Trojan and backdoor attacks and analyzing the vulnerability levels of different CNN layers.
- Provides a tutorial on designing a Convolutional Neural Network (CNN) hardware accelerator using System Verilog.
- Demonstrates the emulation of a CNN hardware accelerator with open-source code.
- Offers a guide on integrating approximate computing technologies to optimize CNN hardware accelerators.
