{
  "slug": "leaftl",
  "term": "LeaFTL",
  "aliases": [
    "Learning-based flash translation layer"
  ],
  "definition": "LeaFTL is a research flash translation layer that compresses logical-to-physical page mappings into learned linear segments.",
  "note": "Approximate segments have a bounded address error. The predicted page's out-of-band metadata identifies neighboring logical pages, allowing correction with one additional flash access in the described design. This is address verification, not permission to return an approximate payload. [lea2023]\n\nThe published evaluation includes both a simulator and an open-channel SSD prototype. The latter runs the FTL on the host; it is not a survey of retail SSD firmware. [lea2023]\n\nThe public artifact documents simulation experiments using WiscSim, PyPy2 and external traces, with substantial CPU and memory requirements. [artifact]\n\nWithout protected DRAM, the described recovery scans flash, rebuilds block-validity counters and relearns recently allocated blocks. The prototype's reported 101.3 ms mean segment-reconstruction time is only one component of its 15.8-minute mean recovery, measured after intentional system restarts. These are author-reported prototype results, not a retail-SSD recovery bound. [lea2023]",
  "status": "research",
  "advantages": [
    {
      "text": "Compressing mappings can leave more controller memory available for data caching.",
      "citationIds": [
        "lea2023"
      ]
    }
  ],
  "limitations": [
    {
      "text": "Lookup correction relies on the bounded-error and metadata design; compression alone does not establish crash consistency or a universal latency bound.",
      "citationIds": [
        "lea2023"
      ]
    }
  ],
  "related": [
    "ftl",
    "mapping-cache",
    "translation-page",
    "page-level-mapping",
    "reverse-mapping-ftl"
  ],
  "supersedes": [],
  "supersededBy": [],
  "firstSeen": null,
  "lastSeen": null,
  "updated": "2026-11-09",
  "citations": [
    {
      "id": "lea2023",
      "title": "Sun et al. LeaFTL: A Learning-Based Flash Translation Layer for Solid-State Drives. ASPLOS 2023, sections 3-5; DOI 10.1145/3575693.3575744.",
      "url": "https://jianh.web.engr.illinois.edu/papers/leaftl-asplos2023.pdf",
      "accessed": "2026-11-09"
    },
    {
      "id": "artifact",
      "title": "LeaFTL author artifact: simulation setup, workloads and reproduction commands.",
      "url": "https://github.com/platformxlab/LeaFTL",
      "accessed": "2026-11-09"
    },
    {
      "id": "analysis",
      "title": "Hesela: SSD mapping caches, learned lookup validation and evaluation boundaries.",
      "url": "https://hesela.com/analyses/ssd-mapping-cache-hmb/#research",
      "accessed": "2026-11-09"
    }
  ],
  "provenance": [
    "Primary-source synthesis. No benchmark replication, power-cut experiment or independent peer review is claimed. LeaFTL (ASPLOS 2023) and LearnedFTL (HPCA 2024) are distinct designs."
  ]
}