Maximizing Mutual Information in Flash Memories
summary
Argues that placing read thresholds and write levels by maximizing mutual information yields near-optimal LDPC FER for real flash, using a greedy chain-rule decomposition and hierarchically-dynamic quantization to keep hardware reads fast. Shows concrete tradeoffs: optimize for the initial hard read or for the slower enhanced reads and get different level placements; optimizing for enhanced reads is a practical compromise. Uses 60k-page LDPC sims, non‑equally‑spaced levels to equalize per‑page mutual information, and warns that learning the time‑varying channel wears out cells.
AI-written summary · about
chapters (5)
- 0:00Flash memory and mutual info
- 7:06Optimizing read thresholds
- 18:22Balancing mutual info across pages
- 27:54Practical conclusions
- 30:01Question and answer session
related talks
15mLESS is More for I/O-Efficient Repairs in Erasure-Coded Storage
11mExploiting Combined Locality for Wide-Stripe Erasure Coding in Distributed Storage
15mSelectiveEC: Selective Reconstruction in Erasure-coded Storage Systems
18mWho's Afraid of Uncorrectable Bit Errors? Online Recovery of Flash Errors with Distributed Redundancy
18mMEV Minimization on L2: Solution Space and Design Constraints
46mA Complete Re-Think for Memory Config...
31mMulti-version Coding and Shared Memory Emulation
18mDRBoost: Boosting Degraded Read Performance in MSR-Coded Storage Clusters