Mask Leakage Independence: A Collision-Based Profiled Approach for Side-Channel Analysis on Cryptosystems

As cryptographic hardware implementations become increasingly widespread, the security of their embedded cryptographic modules faces growing threats from side-channel analysis (SCA). Deep learning-based profiled side-channel analysis has been shown to effectively break masking countermeasures, which are designed to protect sensitive intermediate computations from SCA. However, these techniques typically rely on mask leakage, which severely limits their effectiveness when such leakage is unknown, obscured, or temporally distant, as the profiling model cannot effectively exploit it. In this work, we combine the principles of collision attacks and propose a mask-independent collision-based profiled approach named MICA. By reformulating key recovery as a collision detection problem, MICA can identify identical intermediate values within cryptographic computations from SCA leakage, even when the corresponding leakage characteristics differ due to implementation variations across processing units. Moreover, MICA can recover key differences without relying on mask leakage by exploiting repeated computations and potential mask reuse. Extensive experiments on various AES hardware implementations demonstrate that MICA consistently outperforms other theoretically sound profiled SCAs as well as state-of-the-art collision-based attacks.