Seonglae ChoFully Homomorphic Encryption
A technology that performs arbitrary computations on encrypted data without decryption, and when decrypted, yields the same results as if the computations were performed on plaintext.
While currently computationally expensive compared to plaintext operations, FHE performance has improved approximately 8-fold annually since 2011, rapidly expanding practical applications. The primary bottleneck is bootstrapping (refreshing noise), and noise management techniques (bootstrapping, relinearization, modulus switching) are central to performance innovations. The fundamental mathematical foundation is based on lattice (LWE/RLWE) hard problems (SVP, CVP), which provides quantum resistance.
Currently, some specialized applications (encrypted analytics, certain ML inference, etc.) have become feasible, and if this trend continues, "transmission-storage-use" full-cycle encryption (Full-Privacy Computing) could become a reality. This has the potential to weaken data collection-based business models, and just as HTTPS became standard, we may see an "FHE by default" era in the future. Paillier (Partially Homomorphic Encryption) provides intuition as an example of partially homomorphic encryption supporting only addition, while full homomorphism is achieved through unlimited addition, multiplication, and bootstrapping. Conclusion: If the performance curve maintains its trajectory, most cloud computations could potentially transition to operating on encrypted data.
FHE Usages
