Private Information Retrieval (PIR) is a cryptographic protocol enabling a client to retrieve a record from a server’s database without revealing which record was requested. PIR is essential for privacy-preserving applications. Among PIR schemes, single-server PIR—often based on homomorphic encryption—has gained attention for its ease of deployment and reduced trust assumptions. However, single-server PIR remains impractical due to its high computational and memory bandwidth demands. In particular, accessing large databases from SSDs severely limits PIR acceleration in practical, requiring up to 36 seconds to retrieve 288GB database even with custom hardware.

To address this, we propose IVE, an accelerator for single-server PIR, along with a systematic extension that enables practical retrieval of large databases using DRAM. Recent advances in DRAM capacity now allow PIR to be served entirely from DRAM, removing dependence on SSDs even for large databases. Although memory bandwidth still remains a bottleneck, multi-client batching effectively amortizes database access across concurrent requests and improves throughput. However, it fails to address the bandwidth demands of client-specific data, ultimately limiting the degree of improvement. IVE overcomes this by employing a large on-chip scratchpad with an operation scheduling algorithm that maximizes data reuse, further boosting throughput. Additionally, we introduce sysNTTU, a versatile functional unit that enhances area efficiency without sacrificing performance. We also propose a heterogeneous memory–based deployment system architecture that enables linear scaling of database size without throughput degradation. As a result, IVE achieves up to 1,275$\times$ higher throughput than prior hardware solutions for HE-based PIR on practical workloads.