Benchmarking RISC-V Post-Quantum
summary
Benchmarks target Kyber-768 and Dilithium implementations (reference/PQClean and BoringSSL) and show concrete hotspots: Kyber handshakes spend ~2/3 cycles in SHA3/SHAKE, BoringSSL does a 11,000-instruction NTT helper where reference does ~4,000, and LLVM18 auto-vectorizes Barrett reductions into ~5–6 vector ops. Proposals cut that to a single-instruction modular multiply for the fixed 12/23-bit primes and a multi-round 1600-bit Keccak permutation as a vector-group Keccak instruction, promising large instruction-count and latency reductions in real code.
AI-written summary · about
chapters (6)
- 0:00Introduction and motivation
- 3:56Post-Quantum crypto standards
- 8:56Benchmarking Kyber implementations
- 12:51Benchmarking Dilithium implementations
- 14:37Proposed RISC-V instructions
- 18:35Summary and future work
related talks
19mPQCP Support for RISC-V Vector, Future Keccak Extensions
50mProgress in Standardizing Cryptography Extensions for RISC-V Processors
31mPost-Quantum Crypto: Challenges for Embedded Applications
45mPost‑Quantum Cryptography: The Fundamentals of Tomorrow's Digital World
18mMaking the Case for a Keccak Instruction
29mCrunching the Numbers Post Quantum Algorithm Performance
30mInvestigating Post-Quantum Cryptography: building a PQC decision tree for developers
40mImplementation