Machine learning based side-channel-attack countermeasure with hamming-distance redistribution and its application on advanced encryption standard

摘要

Side channel analysis (SCA) is effective to reveal the key of crypto devices by applying statistical analysis to a number of power traces, thus hardware countermeasure is necessary to protect the crypto circuits. A SCA-resistance methodology by machine learning trained power compensation module is proposed to compensate the probability of hamming distance (HD) of the intermediate data directly, to make it unable to be distinguished from correct and incorrect sub-key, thus providing resistance to SCA. The machine learning algorithm is used to find out the best HD redistribution mapping by using neural dynamic programming. Implemented on an AES-128 encryption algorithm circuit on a Xilinx Spartan-6 FPGA mounted on a SAKURA-G board, experimental SCA results show that it can provide more than 200 × measures to disclosure and still has no sign to reveal the advanced encryption standard (AES) sub-key. In addition, it has low power and area overhead and zero frequency overhead, thus is appropriate for hardware implementation of SCA countermeasure.