Full Program
Summary:
Analog-to-digital converters are necessary to interpret the continuous world in digitized format, and side-channel attacks on ADCs can reveal such digital data. Single-slope ADCs constitute an important class of devices which require protection, and this work proposes two techniques to resist classification of power traces by Convolutional Neural Networks. The first technique is duplication of toggle detection circuitry together with a dummy register array to inject false samples into the power trace. The second technique is to randomize the activation time of the comparator based on a secondary ramp signal with a dynamically varying slope. Both techniques reduce classification accuracy of all digital bits to near random guessing, with the first performing slightly better with lower area overhead.Author(s):
Kareem Ahmad
Georgia Institute of Technology
United States
Ece Öztürk
Boğaziçi University
Turkey
Ceyda Körpe
Boğaziçi University
Turkey
Hyunsoo Yang
Georgia Institute of Technology
United States
Junbin Yang
Georgia Institute of Technology
United States
Kanishk Tihaiya
Georgia Institute of Technology
United States
Ryanh Tran
Georgia Institute of Technology
United States
Günhan Dündar
Boğaziçi University
Turkey
Vincent Mooney
Georgia Institute of Technology
United States
Kemal Ozanoglu
Boğaziçi University
Turkey