SSD controller
In a NAND-based SSD, the controller connects the host interface to flash and coordinates address translation, media operations and error correction.
Functions: A conventional NAND SSD controller coordinates host requests, logical-to-physical mapping, bad-block handling, garbage collection and wear leveling. Its hardware and firmware also provide error correction and schedule work across flash resources. [4] [3]
Interfaces: Host-link width and NAND-channel count describe different paths. One 2020 controller supports PCIe Gen 4 x4 or dual x2, eight 8-bit NAND channels with 16 chip enables each, and hard/soft LDPC decoding. These are one design's capabilities, not universal SSD parameters. [3]
Power loss: A 2011 drive design combines power-fail detection, hold-up capacitors and firmware to move temporary-buffer data to NAND. This requires the complete drive implementation, not the controller chip alone. [5]
Recovery: Flushing volatile user data and rebuilding FTL metadata solve different problems. Recoverable mappings alone do not establish survival of every acknowledged cached write. Check the finished drive's persistence contract and qualification conditions. [6]
Completion: Volatile write-back caches can acknowledge I/O before persistence. In Linux 6.12, REQ_PREFLUSH persists prior completed writes before the associated I/O starts; REQ_FUA persists that request before completion. These are block-layer contracts, not a power-cut qualification. [7]
Relations
- related
- Solid-state drive (SSD), Flash translation layer (FTL), NAND flash, Bad block management, NAND channel, PCIe lanes (storage), Planar NAND, LDPC (NAND ECC), Wear leveling, Power loss protection, Over-provisioning, Sprinkler
Advantages
- Interleaving operations on independent flash resources can hide media-operation latency when the workload exposes parallelism. [4]
- Programmable controller firmware allows SSD builders to tune media management for a particular design. [3]
Limitations
- NAND capacity alone does not establish host throughput: shared buses, allocation and cleaning compete for resources. [4]
- Power-fault coverage depends on the finished drive's hardware, firmware and documented write-completion behavior. [5, 6]
Citations
- Cornwell. Anatomy of a Solid-State Drive. Communications of the ACM / ACM Queue, December 2012.
- Microchip. Flashtec NVMe3108 product brief, DS00003534A, July 2020, pp. 1-2.
- Agrawal et al. Design Tradeoffs for SSD Performance. USENIX ATC 2008, sections 2.2 and 3.1-3.4.
- Intel. Enhanced power-loss data protection in the SSD 320 Series. March 2011, 325207-001US, p. 1.
- Kingston. SSD Power Loss Protection: hardware hold-up energy and firmware mapping recovery. April 2026.
- Linux 6.12. Explicit volatile write back cache control: completion, cache flushes and FUA.
Further reading
Sources & caveats
- Agrawal et al. describe design choices and trace-driven simulation in 2008, not measurements of current controllers. [4]
- The 2020 Microchip product brief advertises IOPS and lossless abrupt shutdown without a reproducible qualification method or protected write-completion boundary. These are product claims, not general controller guarantees. [3]
- The 2011 Intel SSD 320 brief describes one drive's power-loss design, not a requirement that every SSD use capacitors. [5]
- The acknowledged-write boundary is an engineering inference from the distinction between volatile user-data protection and FTL recovery, not a new power-cut test. [6]
- Cornwell's 2012 article is retained as historical further reading. Its full text returned HTTP 403 during this review; its numeric examples are no longer presented as current general specifications. [2]