Benchmarking Homomorphic Encryption on Low-Power Devices: Trade-offs Between PHE and FHE

Homomorphic Encryption (HE) allows computation on ciphertexts, ensuring strong privacy for applications like smart metering, healthcare, and financial analytics. However, instantiating HE on power-constrained embedded devices is challenging as the computation and memory footprints are excessively high—particularly for Fully Homomorphic Encryption (FHE) schemes like BFV and CKKS. Partly Homomorphic Encryption (PHE) like Paillier is lightweight but less functional. This work evaluates the trade-offs among PHE and FHE with an investigation of Paillier, BFV, and CKKS on three representative platforms: ESP32, Raspberry Pi 4, and Arduino Uno. Performance measures like encryption and decryption time, ciphertext size, memory, and energy are compared. Experimentation demonstrates FHE to be impractical on 8-bit microcontrollers but efficient on 32- and 64-bit platforms. Of interest on Raspberry Pi 4, BFV and CKKS demonstrate sub-10 ms encryption times and consume below 5 J per 100 operations, both outpacing Paillier on speed and energy efficiency. Our work refutes the argument on the impracticability of FHE on embedded devices and provides practical advice on selecting among HE schemes according to platform capability. Our research fills the gap between theoretical cryptography and realistic deployment and promotes the use of HE as an enabling solution to trusted edge computing.