Is information leakage spilling?
Résumé
Correlation Power Analysis (CPA) is one of the best known approaches for performing power analysis on cryptographic algorithms.
If the implementation is unprotected, a reasonable number of power traces is required to validate the optimal key hypotheses and
probably retrieve some part of a secret key. This is achieved by computing a correlation matrix with as many columns as the total
number of key guesses and as many rows as the number of samples in the power traces. Then, the correct hypothesis is chosen as
the guess with the largest correlation coefficient (or the largest absolute, depending on who you ask). But this ignores the second
dimension of the matrix since we assume that a single sample will exhibit the greatest correlation against the model. However, what
happens when the correlation “spills” over multiple samples? In this talk we show that in such a case it is possible to use an area
estimator to improve the accuracy of selecting the correct hypotheses. We also discuss our impressions on how is this spill can be
achieved.