Homomorphic Encryption on GPU

Homomorphic encryption (HE) is a cryptosystem that allows the secure processing of encrypted data. One most popular HE schemes Brakerski-Fan-Vercauteren (BFV), which supports somewhat (SWHE) and fully homomorphic (FHE). Since overly involved arithmetic operations are amenable to concurrent computation, GPU devices can be instrumental in facilitating practical use real world applications thanks their superior parallel capacity. This paper presents an optimized highly parallelized library accelerate BFV scheme. includes state-of-the-art implementations Number Theoretic Transform (NTT) inverse NTT minimize kernel function calls. It makes efficient memory hierarchy computes 128 for ring dimension 2 14 only 176.1 μs on RTX 3060Ti GPU. To best our knowledge, this fastest implementation literature. The also improves performance Although independently used, it integrated with Microsoft SEAL library, well-known implements For one ciphertext multiplication, modulus bit size 438, offers 63.4 times speedup over running high-end CPU. compares favorably other operations. Finally, we implement privacy-preserving application classifies genome data tumor types achieves speedups 42.98 5.7 CPU using single 16 threads, respectively. Our results indicate facilitate deployment cryptographic libraries real-world applications.