Explainable automated data estimation in Logic State Imaging of embedded SRAM

Semi-invasive logic state imaging of SRAM cells via optical analysis has been demonstrated to be a powerful attack vector on advanced ICs. Reported applications tend to focus on the recovery of small size secrets such as encryption keys. However, the emergence of new cryptographic algorithms with encryption keys or other assets of larger size makes manual approach challenging. In this manuscript, a framework for the automated estimation of bitcell data is developed. It focuses on understanding signal generation and detection processes to assess viability of well-established classification and clustering algorithms. A main advantage of such an approach is that it makes optical attacks for large scale recovery more feasible since limited to no training data is required. The approach is evaluated on a dataset generated under Thermal Laser Stimulation on a commercial SoC-FPGA manufactured in 28 nm technology.