A 28nm 68MOPS 0.18\mu\mathrm{J}/\text{Op}$ Paillier Homomorphic Encryption Processor with Bit-Serial Sparse Ciphertext Computing

Cloud computing currently lies at the heart of tremendous emerging information applications, providing various reliable and high-performance services based on vast amounts of individual and organizational data. Paillier homomorphic encryption (PHE) [6] is one of the critical privacy computing techniques that enables the ciphertext to enjoy equivalent utility as the plaintext. In the PHE scheme shown on the upper left of Fig. 15.4.1, the client encrypts its plaintext into ciphertext and then sends it to the server, which performs the homomorphic evaluation and returns the encrypted results to the client. However, this process avoids the delivery of plaintext at the cost of several computing burdens, and we list three dominant challenges in the rest of Fig. 15.4.1. First, the ciphertext domain calculation needs expensive large integer modular arithmetic operations (e.g., modular multiplication (ModMul), modular inversion (Modlnv), and modular exponentiation (ModExp)) with several orders of magnitude higher energy and latency. Second, the client performs encryption and decryption featuring independent vector operations, while the server is expected to perform evaluations with multiply-and-accumulation (MAC) operations. Third, the diversity of tasks requires computing scalability to meet the latency and throughput demands in the cloud.