Högskolan i Skövde

This thesis examines the energy efficiency of Advanced Encryption Standard (AES) encryption across various modes of operation (ECB, CBC, CFB, OFB, CTR, GCM, and CCM) on ARM Cortex-A53, Cortex-A72, and Cortex-A76 processors, using Raspberry Pi models 3, 4, and 5 as the experimental platforms. The study primarily investigates the impact of key lengths (128, 192, and 256 bits) and data sizes on energy consumption during encryption tasks. Using an experimental setup with the Raspberry Pi single-board computers, energy consumption was measured and analyzed through repeated encryption operations and data collection via a power meter interfaced with a database. The results reveal only modest increases in energy consumption with larger key lengths across all tested modes and data sizes, suggesting that while key length incrementally affects energy usage, the impact remains relatively minor, thus not significantly compromising energy efficiency for enhanced security. The analysis further shows that ECB mode consistently exhibits the lowest energy consumption, with CTR and CBC not far behind, followed by OFB and then CFB being the least effective among the traditional modes, with AEAD modes like GCM and CCM demanding substantially higher energy, reflecting their more complex processing requirements. Additionally, the study highlights the influence of data size on energy efficiency, showing a decrease in energy consumption per kilobyte with increasing file size up to a certain point, beyond which the benefits diminish. This thesis contributes to a deeper understanding of the trade-offs between security features and energy efficiency in AES encryption on ARM processors, offering insights into scenarios where energy consumption is a critical concern. The findings underscore the importance of selecting appropriate encryption modes and configurations based on the specific requirements and constraints of hardware environments aimed at optimizing energy efficiency in cryptographic operations. Future research could expand on a broader array of ARM-based devices to improve the biases from the Raspberry Pi boards and enhance the reliability of the conclusions drawn from the data.