What Do You See? Transforming Fault Injection Target Characterizations

Abstract

In fault injection attacks, the first step is to evaluate the target behavior for various fault injection parameters. Showing the results of such a characterization (commonly known as target cartography) is informative and allows researchers to assess the target’s behavior better. Additionally, it helps understand the performance of new search methods or attacks. Thus, publishing obtained results is essential to provide relevant information for reproducibility and benchmarking, improving state-of-the-art results and general security. Unfortunately, publishing the results also allows malicious parties to reverse engineer the information and potentially mount an attack easier. This work discusses how various transformations can be used to occlude sensitive information but, at the same time, still be useful for interested researchers. Our results show that even simple 2D transformations, such as rotation, scaling, and shifting, significantly increase the effort required to reverse engineer the transformed data but maintain the interesting data distribution. Consequently, this work provides a method to allow publishers to share more data in a confidential setting.