How do SSDs work?
- User
How do SSDs work?
- Assistant
Solid-State Drives (SSDs) in a nutshell
An SSD is basically a tiny computer whose job is to remember your data when the power is off and hand it back very, very quickly.
Under the hood it has four main parts:Part What it does NAND Flash Chips Non-volatile memory cells that store the actual bits (0s/1s). Controller A dedicated CPU that orchestrates reads/writes, wear-leveling, error correction, encryption, etc. DRAM / SRAM Cache Fast working memory for the controller’s mapping tables and write buffers. Interface SATA, PCIe/NVMe, USB, etc.—how the drive talks to the rest of the computer.
1. NAND Flash Memory
-
Cells
• Each cell holds a charge; the amount of charge represents 1, 2, 3, or 4 bits (SLC, MLC, TLC, QLC).
• Cells are arranged into pages (e.g., 16 KB) and blocks (hundreds of pages). -
Program / Erase
• You can write (“program”) only to a fresh page.
• To change existing data you must erase an entire block first (slow, ~2 ms vs. ~100 µs to read).
• Every erase slowly degrades a cell (limited to ~1k–100k cycles depending on type).
2. Flash Translation Layer (FTL)
Because the OS expects a disk that can overwrite 512-byte sectors freely (like HDDs), the controller runs firmware called the FTL that translates:
Logical Block Address ⇒ Physical Page + offsetKey tricks:
• Wear-leveling – spreads writes so no block wears out prematurely.
• Garbage collection – in the background, moves still-valid pages out of partially “dirty” blocks, erases those blocks, and recycles them.
• TRIM – when the OS deletes a file, it can inform the SSD; the FTL then treats those pages as free, reducing future GC work.
• Over-provisioning – extra hidden capacity (e.g., a 1 TB SSD may have 1.1 TB of flash) gives the FTL breathing room for GC and wear-leveling.
3. Error Handling
As cells shrink, raw bit error rates rise, so the controller:
• Stores ECC (error-correcting codes) alongside your data.
• Uses RAID-like parity across multiple dies (e.g., “RAID-5 inside the SSD”).
• Refreshes weak data during idle time.
4. Parallelism = Speed
A single flash die is only ~40 MB/s. SSDs stack:
• Multiple channels (wires) from the controller.
• Multiple dies per channel.
With 8 channels × 4 dies, reads/writes occur in parallel, giving hundreds or thousands of MB/s.
5. Interfaces
• SATA (AHCI) – legacy HDD protocol, tops out ~550 MB/s.
• PCIe/NVMe – modern, many lanes, queues, and commands; >7000 MB/s on Gen4 x4, even faster on Gen5.
• M.2 / U.2 / E1.S – just physical form factors that carry SATA or PCIe signals.
Life of a Write (simplified)
- OS issues “write 4 KB to LBA 12345.”
- Controller puts data into DRAM, assigns a fresh physical page, updates its mapping.
- Data is written to NAND.
- When idle, controller cleans up old pages, erasing blocks and recycling them.
Result: Near-instant random access, no moving parts, silent operation, and much better shock resistance than HDDs—though with finite write endurance.
⚡ That’s the core of “how SSDs work”!
-
Other models, same or similar prompt
Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.