Extending Moore’s Law with Low-Cost Memory Reliability
summary
Claims a practical path to tolerate DRAM scaling faults and Rowhammer-style attacks using low-cost trackers and controller-side sampling rather than expensive per-row hardware. Demonstrates a one-entry single-entry greedy tracker and a probabilistic MINT probabilistic sampler with analyses and experiments showing thresholds: legacy 140k→~5k activations decline, MINT tolerates ~2,800 (single-sided) or ~1,400 (double-sided), and a PRAC-style single-entry design tolerates ~100–200 with ~1% slowdown. Designs are implemented in model and hardware-aware simulation and proveably reduce attack success at minimal cost.
AI-written summary · about
chapters (10)
- 0:00DRAM reliability and rowhammer
- 4:54Memory systems and DRAM
- 16:42What is rowhammer
- 29:33Rowhammer challenges
- 36:08Deterministic trackers limits
- 46:36Low-cost rowhammer mitigations
- 56:35Probabilistic solutions insights
- 1:02:20Cloud vendor mitigations
- 1:08:29Industry challenges and future
- 1:16:58Q&A
related talks
41mUsing Microarchitectural Design to Break KASLR and More
13mMemPool Flavors: Between Versatility and Specialization in a RISC-V Manycore Cluster
39mSecure Cache Designs: State of the Art and ...
27mConsistency and Coherence of the NVIDIA Grace-Hopper Superchip
01h 28m“Spectre: Secrets, Side-Channels, Sandboxes, and Security”
45mHackathon Presentations
01h 21mPanel+Closing
24mKEYNOTE: RISC-V: Enabling a New Era of Open Data-Centric Computing Architectures