A Novel Tiny Machine Learning Framework for Side Channel Attack Detection in Vehicular Networks

Side channel attacks (SCA) are considered the major threats that predominantly affect the vehicular networks by recovering the sensitive information by using power consumption, electromagnetic variations, and other fault outputs generated when encrypted algorithms are executed in vehicular devices. Though the ML and DL algorithms are employed for the detection of SCA, resource‐constrained devices and a limited number of datasets remain a major bottleneck for the existing models to identify the growing SCA attacks in vehicular systems. In this regard, the research article proposes a lightweight hardware‐centric learning framework based on TinyML, based on few‐shot learning to achieve accurate detection of SCA with reduced latency. The proposed framework consists of five components: (a) Intelligent Data Collection from the power consumption traces captured from vehicular devices while running the strong encryption schemes, (b) Data preparation for effective training and testing the module, (c) Model design using the FSL‐Siamese Feed Forward networks, (d) Model validation and evaluation, (e) Hardware Deployment. The comprehensive experimentation is conducted using the Raspberry PICO microcontrollers in which 2500 data traces were collected, and performance metrics such as accuracy, precision, recall, and F1‐score are calculated. The empirical results were analyzed and compared with the other existing state‐of‐the‐art detection systems. The experimental outcomes demonstrated that the proposed approach has achieved the 99.5% accuracy, 99.4% precision, 99.2% recall, and 99.7% F1‐score, which shows a promising direction for involving this hybrid approach for mitigating the SCA in the ITS systems.