LightPIR: Single-Server PIR via FHE without Gaussian Noise

We introduce the LightPIR family, a new series of single-server PIR protocols with reduced overhead in terms of storage, communication, and computation. The protocols rely on a new GSW-like homomorphic cryptosystem based on the Ring Learning with Rounding (RLWR) problem and several ciphertext conversion algorithms for matrix encoding. Our RLWR-based techniques offer considerable advantages across all evaluation metrics compared to the previous state-of-the-art, the Spiral family (S&P 2022), under various database configurations. On the server, the LightPIR family simultaneously achieves a 1.2−1.8× increase in the throughput and up to a 1.4× increase in the rate while reducing the storage requirements of public parameters by 1.6−5×. For the client, the LightPIR family eliminates the Gaussian noise sampling, resulting in an average 36% reduction in query encryption time and resilience against side-channel attacks. Meanwhile, communication efficiency between the client and server is also enhanced by a reduction of up to 3.8× in query size and up to 1.8× in response size. All the optimizations indicate that our constructions are more lightweight for both the client and the server, highlighting a new practical scenario where RLWR-based schemes can showcase their advantages.