Rethinking Architectures and Abstraction for a World Where Security Improvements Matter More than Performance Gains
summary
Argues that Spectre as symptom reveals a wide gap between architectural guarantees and what software actually needs, citing side channels, Rowhammer, speculative execution, memory compression and dedup as real-world leaks. Points to concrete mitigations: wiping hardware state between domains (effective when reset clears state), spatial separation and heterogeneous secure cores on-die (trusted small cores plus legacy fast cores), RISC‑V experimentation, and redundancy math—10^7 elements at 10^-4 defect rate turned into triplets raises success probability to ~0.7—promising but still engineering-challenging.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 2:40Reindeer experiment
- 6:51Timing side channels
- 13:24USB driver problem
- 16:34Time-space separation
- 23:13Secure processor architectures
- 25:20Security as probability
- 34:10Industry priorities shift
- 36:55Conclusion
related talks
01h 28m“Spectre: Secrets, Side-Channels, Sandboxes, and Security”
43m“Closing Panel: Spectre”
33mGhosting the Spectre
01h 01mTutorial: Side-Channel Attacks and Transient Execution Vulnerab...
16mMitigating Transient-Execution Attacks with CHERI Compartments
51mKeynote |The Microarchitectures That I Saw And The Ones That I Hope To One Day See
50mPractical Ghostbusting: Meltdown, Spectre and You
23mHardware security for critical national infrastructure