Vulnerability Analysis of a Soft Core Processor through Fine-grain Power Profiling

Embedded microprocessors are an important component of reconfigurable architectures. Fine-grain (e.g., cycle-accurate) power analysis of such processors has been used to improve power and energy efficiency, and detect implementation vulnerabilities, in embedded applications. However, such analysis is difficult to conduct; it requires either specialized and often expensive equipment, or construction of test architectures using disparate acquisition and analysis tools. In this research, we expand the Flexible Open-source workBench fOr Side-channel analysis (FOBOS) to facilitate exact time-domain correlation of clock cycle and device state to power measurements, and to perform power analysis on a soft core processor. We first validate the fine-grain power analysis capabilities of FOBOS through cycle-accurate analysis of power consumption of AES encryption running on a soft core processor in the Spartan-6 FPGA. We then analyze the results in the context of Simple Power Analysis side-channel attacks, and confirm power correlation of certain instructions with Hamming Weight or Hamming Distance of secret key bytes. Finally, we show that an assumption of a pure Hamming Distance power model for load-to-register instructions is not sufficient for this embedded processor architecture, and that power models using both Hamming Distance and Hamming Weight should be considered for Differential Power Analysis.