To protect data, smartphones, PCs, networks, and other computer devices utilize block encryption algorithms, among which Advanced Encryption Standard (AES) is a representative encryption algorithm. AES is evaluated to be the safest one of the encryption algorithms at present and is universally applied. However, the AES encryption algorithm began to be threatened by Side-Channel Attack (SCA) since 1996. Many studies have been done and many more will probably be conducted, but so far, the results are showing some significant correlation between a variety of leaked information and success rate in side-channel attacks. Many different side-channel attacks are using a variety of leaked information (e.g., power consumption, electromagnetic waves, fault-injection results), among which Correlation Power Analysis (CPA) is the technique of analyzing power consumption with the Pearson correlation coefficient and it can be presenting a high success rate of attacking most hardware devices. If the correlation power analysis is applied and measured by an oscilloscope, it is possible to obtain secret keys of the low-power computer-based equipment, such as IoT or wearable devices. Also, recently, deep learning-based analytic studies on side-channel attacks are actively conducted, therefore, the need for research on attack methods and defense techniques becomes more urgent. In this paper, we upload AES’s AddRoundKey function to Arduino UNO, which has the capability that can be used for IoT, measure power consumption with an oscilloscope for all keys, check the correlation power analysis success rate, and confirm the need for defense techniques. For the experiment, based on the AddRoundKey function of AES, 256 secret keys were measured with a 1GS/S oscilloscope. At this time, these secret keys were bundled in tens per case, and therefore a total of ten cases were measured. In the comparison of power consumption, the Pearson correlation coefficient, and guess the key, correlations with secret keys were inferred. According to the simple comparison of power consumption and the Pearson correlation coefficient, the accuracy rate was about 65%. In this case, when the Pearson correlation coefficient was compared for the position of the power consumption waveform, the accuracy rate was about 80%. Additionally, with the use of the point that the positions of guess keys were mostly constant, it was possible to achieve the success rate of about 93.75%.
Read full abstract