Side Channel Cryptanalysis Using Machine Learning Using an SVM to recover DES keys from a smart card .

Cryptographic devices use power as they operate. Variations in the amount of power consumed during an operation may leak information about data values being processed. Methods such as differential power analysis (DPA) [1] have been used to recover keys by exploiting such leaks. But DPA is tougher to apply to leakage that depends only on the key. For this project we attacked leakage from an 8-bit smart card performing DES decryptions with a 56-bit key. We explored using an SVM to exploit the side channel leakage. Using an SVM classifier we reached near 100% classification accuracy using features drawn from measurements during the DES key schedule, after training with at least 1000 samples out of our set of 11000. This is the first publication in which an SVM has been used to recover an entire key.