Modern workstations and laptops widely use PCIe/NVMe SSDs, primarily via the M.2 SSD interface. Servers and storage systems, meanwhile, rely on enterprise-class PCIe/NVMe SSDs with U.2 and E3.S interfaces. However, many users see an NVMe SSD after booting up without understanding the complex journey from power-on to visibility. Today, we’ll dissect the step-by-step process of how a PCIe/NVMe SSD transitions from a powered-off state to a fully operational “read/write-ready” status. This includes power delivery, circuit initialization, protocol link establishment, controller firmware execution, and file system recognition—all working in intricate harmony.

Background Setup
- SSD Type: PCIe Gen5 x4 interface, supporting NVMe 1.4 or latest 2.0/2.1 protocols
- Controller Platform: ARM Cortex-R series + NAND controller + DRAM + PMIC
- Motherboard BIOS: Supports NVMe boot
Stage 1: Hardware Power-On + Physical Layer Startup (0–100ms)
1.1 Power Delivery Path Activation
- At power-on, the motherboard supplies the SSD via the PCIe slot:
- 12V main power (stepped down internally to power SoC, NAND, DRAM)
- 3.3V auxiliary power (controller/PMIC initialization)
- The SSD’s PMIC (Power Management IC) sequentially enables voltage rails:
- Vcore (1.0V) → Controller CPU/PLL
- Vddq (1.2–1.8V) → DRAM
- Vcc (3.3V) → NAND
*Click to enlarge* a SerialTek PCIe 5.0 protocol analyzer capture showing power-up timing (12V, 3.3V, PERST#, CLKREQ#, REFCLK).
1.2 Controller Boot ROM Execution
- The SSD controller launches from its internal BootROM:
- Loads firmware from SPI NOR Flash or ROM-resident code
- Initializes DRAM/NAND interfaces (DDR PHY, ONFI/NV-DDR)
- Checks NAND BBT (Bad Block Table) and FTL mapping structures
1.3 PCIe PHY Link Training
- The controller’s PCIe PHY initiates Link Training and Status State Machine (LTSSM):
- LTSSM state transitions:
- Detect → Polling → Configuration → L0 (Active)
- Once the PCIe link is established:
- The PCIe switch/root port detects the SSD as an endpoint, assigns Bus/Device/Function (BDF) addresses
- Full PCIe (eGen5 x4, 32GT/s per lane) is operational
Stage 2: Enumeration & NVMe Initialization (100–500ms)
2.1 BIOS/UEFI PCIe Device Enumeration
- BIOS scans PCIe buses, reads the SSD’s Configuration Space
- Identifies device as `Class Code = 01 08 02` (Mass Storage, NVMe)
- Loads UEFI NVMe driver or Option ROM if NVMe boot is supported
2.2 NVMe Admin Queue Setup
- The system sends Admin Queue initialization commands:
- Creates Submission/Completion Queues (SQ/CQ)
- Retrieves Controller Identify Data (capacity, queue limits, command sets)
- The controller allocates DRAM addresses and initializes Admin Queue metadata
2.3 Namespace Creation & Activation
- The system requests namespace creation (a logical disk in NVMe):
- The controller:
- Scans NAND block metadata, builds L2P (Logical-to-Physical) mapping tables
- Initializes FTL (Flash Translation Layer)
- Configures SLC Cache (if applicable)
- Returns `Namespace ID + capacity` once ready
Stage 3: OS Loading + File System Mounting (500ms–2s)
3.1 OS NVMe Driver Initialization
- Windows/Linux loads `nvme` driver modules
- Driver maps the controller’s MMIO register space via BAR0
3.2 OS I/O Queue Configuration
- OS creates multiple I/O queues (e.g., 4/8/16 queues matching CPU cores)
- Controller allocates DRAM buffers, maps host memory addresses
- Enables MSI-X interrupts for per-queue low-latency responses
3.3 Partition Scanning + File System Recognition
- OS reads initial LBAs to detect MBR/GPT partition tables
- Recognizes file systems (ext4/NTFS/APFS) and mounts them
- Visible as `/dev/nvme0n1` (Linux) or `C:` (Windows)
Stage 4: Operational Data Path (Read/Write)
4.1 Write Workflow (e.g., NVMe Write Command)
- CPU submits `nvme_write()` → SQ entry
- Controller fetches SQ entry → DMAs host data to DRAM
- Writes to NAND (via SLC cache if enabled) → Updates L2P/FTL
- Generates CQ entry → Host polls CQ or triggers interrupt
4.2 Read Workflow (e.g., NVMe Read Command)
- CPU submits `nvme_read()` → SQ entry
- Controller checks L2P → Reads NAND → ECC decoding
- DMAs data to host → Updates CQ
Key Firmware Modules
| Module | Function |
| FTL | Manages LBA↔PBA mapping, garbage collection (GC) |
| Wear Leveling | Balances NAND erase cycles |
| Bad Block Management | Skips faulty NAND blocks |
| Thermal Throttling | Prevents overheating |
| SLC Cache / pSLC | Boosts write speed us MLC/TLC |
Timeline Overview
| Time | Action |
| 0–20ms | Power stabilization → PMIC + reset |
| 20–100ms | Boot ROM → NAND/DRAM init → PCIe link training |
| 100–300ms | BIOS PCIe enum → Option ROM load |
| 300–500ms | Admin queues → Namespace ready |
| 500ms – 2s | OS driver → I/O queues → File system mount |
| 2s+ | Read/write operations |
Summary
A PCIe/NVMe SSD transitions from power-on to usability through:
Power init → BootROM → PCIe link training → NVMe enum → Queue setup → FTL mapping → OS driver → User access
This orchestration spans hardware (PHY, PMIC), protocols (PCIe/NVMe), firmware (FTL, wear leveling), and OS layers—delivering a “ready-to-use” SSD in seconds.






