Research has shown that when a microcontroller (MCU) is powered up, the emitted electromagnetic radiation (EMR) patterns are different depending on the executed instructions. This becomes a security concern for embedded systems or the Internet of Things. Currently, the accuracy of EMR pattern recognition is low. Thus, a better understanding of such issues should be conducted. In this paper, a new platform is proposed to improve EMR measurement and pattern recognition. The improvements include more seamless hardware and software interaction, higher automation control, higher sampling rate, and fewer positional displacement alignments. This new platform improves the performance of previously proposed architecture and methodology and only focuses on the platform part improvements, while the other parts remain the same. The new platform can measure EMR patterns for neural network (NN) analysis. It also improves the measurement flexibility from simple MCUs to field programmable gate array intellectual properties (FPGA-IPs). In this paper, two DUTs (one MCU and one FPGA-MCU-IP) are tested. Under the same data acquisition and data processing procedures with similar NN architectures, the top1 EMR identification accuracy of MCU is improved. The EMR identification of FPGA-IP is the first to be identified to the authors' knowledge. Thus, the proposed method can be applied to different embedded system architectures for system-level security verification. This study can improve the knowledge of the relationships between EMR pattern recognitions and embedded system security issues.
Keywords: EMR measurement platform; EMR side-channel effects; NN model dataset building.