Advances in deep learning algorithms has shown that neural network can outperform existing approaches in Side Channel Attacks (SCAs). To improve upon our previously validated neural network for SCAs in Phase I, the Scope of Work (SOW) will focus on: 1) the meanings of the activation functions and weights correspondent to the keys and architectures under SCAs; 2) features or a set of features that are correspondent to the different components in one system architecture; and 3) a refined/assembled neural network once it is trained with one system architecture. To best protect embedded systems against SCAs, we will build upon prior experimental results and knowledge to leverage the most highly successful SCAs Neural Network models, a requirement so as to effectively defend against the most sophisticated SCA an adversary might leverage against US embedded systems. Lidar, Radar, Telecommunication antennas, long-range cameras sensors are widely used in DoD systems across all branches of the services for: ground sensors, Ground to Air radio systems, weaponsâ systems, drones, and for various medical applications as well. We will provide components and architecture design for these systems to optimize measurements (e.g. signal spectrums, power traces) and to provide a Deep Neural Network model thatâs optimized to defend against the most effective SCA methods which will be further refined to ensure the defensive measures have optimized resiliency and efficiency. The trained neural network will be tested again on the Lidar, Radar, telecommunication antennas, long-range camera systems and Black Fur high speed drone. Based on the test results, the Black Fur team will identify resilient architectures (both hardware and firmware) for these components/systems. This optimization process and architecture is expected to be cost effective, and highly repeatable for other embedded systems, drones, radar, lidar, long-range camera systems and dron
