LUT oriented Hardware Trojan for FPGA based AI Module

Abstract

An edge AI module is expected in the fields of automotive system, robot system, and so on. On the other hand, security issues for AI modules are pointed out in promoting social implementation of AI. In the field of hardware security, the hardware Trojan (HT) has been reported as one of the security threats. The HT is embedded into LSI system unnoticed by designer, and it performs the malicious operations under the specific condition. To promote social implementation of AI, the trusted AI is required, that is, it needs countermeasures against malicious attacks for AI. At this time, it is important to reveal the vulnerability of AI modules, in order to establish countermeasure. Therefore, this study proposes a new HT for field programmable gate array (FPGA) based AI module. The proposed HT causes erroneous inference of AI module by focusing on lookup table (LUT) structure. Experiments using an FPGA verify the validity of the proposed HT.