On use of deep learning for side channel evaluation of black box hardware AES engine

Abstract

With the increasing demand for security and privacy, there has been an increasing availability of cryptographic acclerators out of the box in modern microcontrollers, These accelerators are optimised and often black box. Thus, proper evaluation against vulnerabilities like side-channel attacks is a challenge in absence of architecture information and thus leakage model. In this paper, we show the use of deep learning based side-channel attack can overcome this challenge, allowing evaluation of black box AES hardware engine on a secure microcontroller, without the knowledge of precise leakage model information. Our results report full key recovery with only 3,000 traces under a profiling setting.