Seeds of SEED: H-CRAM: In-memory Homomorphic Search Accelerator using Spintronic Computational RAM
Hüsrev Cılasun, Salonik Resch, Zamshed I. Chowdhury, Masoud Zabihi, Zhengyang Zhao, Thomas Peterson, Jian‐Ping Wang, Sachin S. Sapatnekar, Ulya R. Karpuzcu · 2021
Growing use of cloud raises privacy concerns for search in sensitive databases such as bioinformatics, where homomorphic encryption can help through direct computation –hence search/pattern matching–on the encrypted data. The recently proposed homomorphic secure content-addressable memory (SCAM) exploits this principle. Data dependencies in SCAM, however, lead to a memory bottleneck which has been addressed by various near-memory computing solutions. In this paper we demonstrate a more efficient alternative based on the true in-memory computing substrate spintronic Computational RAM (CRAM). The resulting SCAM accelerator, H-CRAM, achieves significant speedup and energy reduction.