Reproduce “Enabling Low-Cost Secure Computing on Untrusted In-Memory Architectures” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Sahar Ghoflsaz Ghinani, Jingyao Zhang, Elaheh Sadredini
Researchers from the University of California, Riverside, developed a Multi-Party Computation (MPC) framework for secure processing on untrusted in-memory architectures, supporting both linear and nonlinear computations while mitigating CPU bottlenecks. The approach demonstrated up to 14.66x speedup over secure CPU methods on real UPMEM hardware, achieving robust security with minimal performance overhead.
Key insights
- Robust security for PIM can be achieved with minimal overhead (as low as 4% compared to insecure PIM) on untrusted hardware without requiring TCB extension.
- Combining arithmetic secret sharing with Yao's Garbled Circuits allows secure processing of both linear and nonlinear operations, significantly expanding PIM's applicability.
- Strategic precomputation effectively mitigates CPU bottlenecks inherent in MPC schemes, optimizing overall secure PIM performance for memory-intensive workloads.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.