Hardware Trojan Attacks on the Reconfigurable Interconnections of Convolutional Neural Networks Accelerators

Convolutional neural networks (CNNs) have demonstrated significant superiority in modern artificial intelligence (AI) applications. To accelerate the inference process of CNNs, reconfigurable CNN accelerators that support diverse networks are widely employed for AI systems. Due to ubiquitous deployment of these AI systems, a strong incentive rises for adversaries to attack CNN accelerators via hardware Trojan, which is one of the most important attack models in hardware security domain. This paper proposed a hardware Trojan that attacks the crucial component in CNN accelerators, i.e., reconfigurable interconnection network. This hardware Trojan changes the data paths under activation, resulting in incorrect connection of the arithmetic circuit, thereby causing wrong convolutional computation. Experimental results show that with increasing only 0.27% hardware overhead to the accelerator, the proposed hardware Trojan can be activated to cause a degradation of inference accuracy by 8.93% 86.20%.