Is Your NVMe Drive Dead?
An NVMe drive that vanishes from BIOS or reports 0 bytes is almost always suffering from firmware panic or PCIe link failure, not dead NAND. Consumer recovery software cannot access a controller that never completed PCIe enumeration. Power the drive off and do not run chkdsk or disk utilities.
- ●Drive not detected in BIOS/UEFI at all
- ●Windows Disk Management shows 0GB, 2MB, or 20MB capacity
- ●Device Manager shows a controller alias like SM2263XT or SATAFIRM S11 instead of the drive model name
- ●"Inaccessible Boot Device" BSOD on a previously working system
- ●Drive enumerates in BIOS but hangs or times out when accessed
Software like Disk Drill, EaseUS, or chkdsk operates through the OS storage driver stack. If the controller is panicked, the OS never sees a valid block device. Running chkdsk on a degraded NVMe drive forces read retries that stress marginal NAND cells and can trigger TRIM operations that permanently erase recoverable data. The only safe action is to power the drive off and send it for professional evaluation.
NVMe PCIe Architecture and Failure Points
NVMe drives connect directly to the CPU via PCIe lanes, bypassing the SATA host bus adapter. This eliminates latency but introduces new failure modes: link-training stalls in Detect or Polling states, reference clock loss, and firmware corruption that passes link training yet fails NVMe initialization.
The PCIe Link Training and Status State Machine moves through a fixed sequence on every cold boot. Each stall location maps to a specific physical fault that determines whether recovery starts with board repair or firmware reconstruction.
- Detect.Quiet / Detect.Active
- The host never sees receiver termination. The drive's PCIe PHY is not powered or its bond wires are open. The cycle repeats at roughly 12 ms intervals indefinitely. Cause: PMIC failure, controller power-rail collapse, or a destroyed PHY analog block. Recovery path: board-level repair first.
- Polling
- Receiver detected but the drive cannot lock onto the host clock or achieve symbol-lock on TS1/TS2 ordered sets. Cause: damaged PCIe lanes, cracked solder under the controller BGA, or a degraded reference clock crystal. Recovery path: PC-3000 forces stable Gen1 connection to bypass the lock failure.
- Configuration
- Bit-lock achieved at Gen1 (2.5 GT/s) but the link cannot negotiate full width or speed. The drive enumerates at x1 Gen1 instead of x4 Gen4. PC-3000 forces a stable Gen1 connection for imaging instead of letting the host retry at higher speeds.
- Recovery.Equalization
- Common stall point on Gen4 and Gen5 drives. Equalization tuning at 16 GT/s or 32 GT/s fails because the controller silicon has thermally degraded or the M.2 edge connector contacts are oxidized. Forcing the link down to Gen3 or Gen2 usually clears it long enough to image.
- L0 reached
- The physical link is healthy. Whatever is wrong with the drive lives above the PHY layer. Recovery shifts to firmware diagnostics: Identify Controller, Get Log Page SMART/Health, and namespace verification.
DRAM vs. Host Memory Buffer: Where the FTL Lives
| Feature | DRAM-equipped NVMe | DRAMless HMB NVMe |
|---|---|---|
| FTL cache location | Onboard DDR4 DRAM chip | Borrowed from host system RAM via PCIe |
| Power loss behavior | FTL in onboard DRAM persists briefly; enterprise drives flush to NAND via capacitors | FTL in host RAM vanishes instantly when the system loses power |
| FTL corruption risk | Lower; periodic NAND checkpoints supplement DRAM copy | Higher; relies entirely on periodic NAND checkpoints |
| Recovery complexity | FTL usually reconstructable from NAND checkpoints | FTL reconstruction may require scanning all NAND pages for metadata fragments |
| Common drives | Samsung 980 Pro/990 Pro, WD Black SN850X, Corsair MP600 | Samsung 980 (non-Pro), WD SN580, Kingston NV2, most budget NVMe |
DRAMless HMB drives trade FTL resilience for lower cost and power consumption. The trade-off becomes critical only during unplanned power loss. If you use a DRAMless NVMe drive for work that cannot be recreated, a UPS is the single most effective protection against this failure class.
The PC-3000 Portable III NVMe Recovery Workflow
The PC-3000 Portable III with the PCIe NVMe adapter on Port 0 acts as its own Root Complex, bypassing the host BIOS and OS entirely. The workflow moves from electrical diagnosis to pin-shorting Techno Mode entry, SRAM loader injection, Virtual Translator construction, and sector-by-sector data extraction.
- 01
Pre-power electrical diagnostics
The M.2 PCB is inspected with a multimeter in diode mode before any power is applied. A shorted 3.3V rail to ground indicates a blown PMIC or shorted multilayer ceramic capacitor. FLIR thermal imaging localizes the heat signature. Powering a shorted drive forces current through the fault and risks destroying the controller die. Board-level repair must happen before any logical recovery is attempted.
- 02
Techno Mode entry via pin shorting
The controller boot sequence starts from an internal Mask ROM before loading firmware from NAND. By shorting the diagnostic test point (such as ROM_CS or UART_TX pads) to ground during power-on, the BootROM halts in a minimal diagnostic state. This bypasses the corrupted firmware loop and exposes the controller to PC-3000. The specific test point varies by controller family; PC-3000 documentation identifies the correct pad for each supported silicon revision.
- 03
SRAM loader injection
PC-3000 connects via the M.2 NVMe adapter on Port 0 and detects the ROM-mode controller. It sends a vendor-specific volatile microcode loader (LDR) directly into the controller's on-die SRAM. The loader is specific to the controller silicon revision, NAND chip ID, and firmware version. It executes from SRAM without writing to NAND, disabling wear leveling, garbage collection, and TRIM during imaging.
- 04
Virtual Translator construction
The loader provides raw NAND access. PC-3000 scans physical NAND pages and reads the Out-of-Band spare-area metadata: LBA stamps and chronological sequence numbers. A virtual Logical-to-Physical (L2P) table is built in workstation RAM. Where multiple physical pages claim the same LBA, the utility picks the one with the highest sequence number. This reconstructs the FTL mapping without writing anything back to the patient drive.
- 05
Targeted data extraction
PC-3000 Data Extractor issues logical reads against the virtual translator and routes them as physical-page fetches through the LDR. The entire drive is imaged sector-by-sector to a known-good destination before the drive is powered down. Files are verified against directory structure and transferred to your return media.
The PC-3000 Portable III hardware acts as a PCIe Root Complex, managing memory mapping and doorbell signaling to communicate with NVMe controllers that have entered a fault state. It supports vendor-specific diagnostic modes for select Phison, Silicon Motion, and Marvell NVMe controllers. Each controller family requires its own utility module, loader payload, and FTL reconstruction algorithm. Using the wrong module renders the recovery attempt useless.
Why Chip-Off Fails on Modern NVMe SSDs
Modern NVMe controllers implement AES-256 encryption with the media encryption key fused into the controller's one-time programmable (OTP) memory. The key never leaves the controller die. Removing NAND chips for chip-off produces only ciphertext that cannot be decrypted without the original silicon.
Phison PS5012-E12, PS5016-E16, and PS5021-E21 controllers implement hardware AES-256 encryption. Silicon Motion SM2262EN and SM2269XT families also bind encryption keys to controller fuses. The NAND contents are encrypted at the hardware level before being written to flash. If the controller is electrically dead, chip-off yields only ciphertext. There is no offline path to decrypt it.
Board repair is not a separate service from data recovery on encrypted NVMe drives; it is the recovery path. We locate the failed component using FLIR thermal imaging, replace the shorted PMIC or voltage regulator with a Hakko FM-2032 on an FM-203 base station, and bring the original controller back to life. When the controller boots, the encryption keys are intact and your data is accessible. Board repair IS data recovery for encrypted NVMe drives.
Chip-off remains a viable fallback only on non-encrypted drives with destroyed controllers, where the NAND holds plaintext. PC-3000 Flash reads raw NAND pages and reconstructs the FTL mapping table. For encrypted drives, this produces unreadable data. We determine encryption status during the free evaluation and inform you before any paid work begins.
Controller-Specific NVMe Recovery Notes
Each NVMe controller family requires a different PC-3000 utility module, diagnostic mode entry sequence, and FTL reconstruction algorithm. ACELab PC-3000 SSD v3.8.10 supports Phison, Silicon Motion, and Marvell NVMe controllers directly.
Phison E12 / E16 / E21
Phison PS5012-E12 (Gen3), PS5016-E16 (Gen4), and PS5021-E21 (Gen4) controllers enter ROM mode when the NAND-resident firmware fails validation. PC-3000 detects ROM mode and injects the matching Phison loader into SRAM. Early E12 firmware has a documented L1.2 power-state re-initialization bug: the PHY fails to retrain after sleep or hibernate. PC-3000 forces a cold PERST# reset to restore link training. The PS5016-E16 runs hotter than later Gen4 designs and is prone to thermal FTL corruption. Recovery requires the PC-3000 Portable III with the Phison NVMe utility.
Silicon Motion SM2262EN / SM2263XT / SM2267XT / SM2269XT
Silicon Motion NVMe controllers use a NANDXtend ECC engine. When the FTL collapses, the drive enumerates but advertises an anomalous capacity (0 GB, 1 GB, or 1023 MB) and surfaces the generic controller string instead of the OEM model name. Recovery follows the ROM-mode workflow: short the diagnostic test point on the PCB, inject the volatile loader into SRAM, and rebuild the translator from OOB spare-area metadata. The SM2267XT in the Kingston NV2 and the SM2269XT in newer budget Gen4 drives both follow this same safe-mode and microcode injection workflow.
Marvell 88SS1320
The Marvell 88SS1320 series is a PCIe Gen4 NVMe 1.4 design with a triple-core ARM Cortex-R5 processor and a four-channel NAND interface at 1200 MT/s. It is rare in consumer intake and shows up almost exclusively in OEM designs. Recovery is strictly limited to board-level repair to preserve the original controller. Because the 88SS1320 is currently absent from the ACELab PC-3000 SSD supported list, firmware-level FTL reconstruction or SRAM loader injection is not possible if the controller itself has panicked.
Samsung Phoenix / Elpis / Pascal
Samsung's in-house controllers run the 970 EVO/Pro (Phoenix), 980 Pro (Elpis), and 990 Pro (Pascal) product lines. These controllers implement always-on AES-256 encryption with keys stored in the controller's secure area. Rossmann does not currently offer in-lab recovery for modern Samsung in-house controllers such as Phoenix, Elpis, or Pascal. Board-level repair to preserve the original controller is the only potential path, but full FTL reconstruction through PC-3000 is not available for these families.
Realtek
Realtek RTS5762 and RTS5763DL controllers are absent from the ACELab PC-3000 SSD supported list. Rossmann does not currently offer in-lab recovery for Realtek controllers. Realtek implements proprietary XOR data scrambling tied to the controller's internal key material. Chip-off without the original Realtek silicon yields pseudo-random noise, not usable data.
Innogrit
Innogrit controllers are absent from the ACELab PC-3000 SSD supported list. Rossmann does not currently offer in-lab recovery for Innogrit controllers.
Maxio MAP1602 / MAP1602A
The Maxio MAP1602/MAP1602A is a 4-channel DRAMless controller common in 2024-2025 budget NVMe SSDs. It is explicitly absent from ACELab PC-3000 SSD v3.8.10. Rossmann does not currently offer in-lab recovery for Maxio MAP1602.
NVMe Data Recovery Pricing
NVMe recovery ranges from $200–$2,500 depending on the failure type. Every case starts with a free evaluation and a firm quote before any paid work. If we recover nothing, you pay nothing. No attempt fees. +$100 rush fee to move to the front of the queue.
| Tier | When It Applies | Price |
|---|---|---|
| Simple Copy | Your NVMe drive works, you just need the data moved off it | $200 |
| File System Recovery | Your NVMe drive isn't showing up, but it's not physically damaged | From $250 |
| Circuit Board Repair | Your NVMe drive won't power on or has shorted components | $600–$900 |
| Firmware Recovery | Your NVMe drive is detected but shows the wrong name, wrong size, or no data | $900–$1,200 |
| PCB / NAND Swap | Your NVMe drive's circuit board is severely damaged and requires NAND chip transplant to a donor PCB | $1,200–$2,500 |
A donor drive is a matching SSD used for its circuit board. Typical donor cost: $40–$100 for common models, $150–$300 for discontinued or rare controllers.
NVMe firmware recovery runs $900–$1,200. Cases requiring board-level repair to revive the controller before firmware work fall in the $600–$900 circuit board tier plus the firmware tier. Large labs typically quote $1,600 to $2,100 for NVMe firmware work. We publish our pricing because you should know what you are paying before you ship anything.
Frequently Asked Questions
Why is my M.2 NVMe SSD not detected in BIOS?
The drive either failed PCIe link training or the controller firmware panicked before completing NVMe initialization. Link training stalls happen when the PHY cannot lock onto the reference clock or negotiate lane width. Firmware panics happen when the Flash Translation Layer corrupts and the controller drops to ROM mode. Both prevent the BIOS from seeing a valid NVMe device. Consumer recovery software cannot fix either problem because the OS storage stack never sees the drive. Recovery requires the PC-3000 Portable III to act as an independent PCIe Root Complex and force communication with the controller. NVMe recovery starts at $200.
Does running CHKDSK on a corrupted SSD destroy data?
Yes, if the controller is already in a degraded state. CHKDSK issues read and write commands through the OS driver. A panicked controller may misinterpret those commands, trigger TRIM on still-recoverable blocks, or overwrite surviving FTL metadata. Each power cycle and access attempt stresses marginal NAND cells. If your NVMe drive shows 0 bytes, the wrong capacity, or is not detected in BIOS, power it off and do not run any software utility.
Can data be recovered from a dead NVMe SSD after a power surge?
Yes, if the NAND is intact and the controller can be revived. A power surge typically destroys the PMIC or a voltage regulator on the PCB, not the controller die itself. FLIR thermal imaging locates the shorted component. The Hakko FM-2032 replaces the failed part while the controller remains soldered in place, preserving the AES-256 encryption keys inside the silicon. Once power delivery is restored, PC-3000 Portable III forces link training and accesses the NAND through the revived controller. Cases requiring board repair before firmware work fall in the $600–$900 circuit board tier plus the firmware tier.
What does PCIe link training failure mean?
PCIe link training is the electrical handshake between the host and the NVMe drive. The Link Training and Status State Machine moves through Detect, Polling, Configuration, Recovery, and L0. A failure means the drive never reached L0, so no NVMe commands can be sent. Common causes include a lost 100 MHz REFCLK, cracked BGA solder joints under the controller, or a degraded PCIe PHY from thermal cycling. PC-3000 Portable III bypasses the host's link training and forces negotiation at Gen1 or Gen2 speeds to establish a stable connection.
Can an NVMe SSD be repaired to working condition?
Data recovery and functional repair are different goals. We recover your data by reviving the controller long enough to image the NAND contents. Full functional repair of a consumer NVMe SSD is usually not economical; the goal is extraction, not returning the drive to daily use. After recovery, we recommend replacing the drive with a new unit. Our focus is getting your files back, not fixing the hardware for reuse.
What is ROM Mode on an NVMe SSD?
ROM Mode is a minimal bootstrap state where the controller boots from its internal Mask ROM instead of the NAND-resident firmware. It happens when the firmware service area is corrupted beyond the controller's self-repair capability. In ROM Mode, the drive responds to a limited command set and reports a generic controller alias with 0 bytes capacity. PC-3000 detects ROM Mode and injects a volatile loader into controller SRAM to bypass the corrupted firmware and access the NAND directly.
How does TRIM affect deleted file recovery on NVMe?
TRIM makes deleted file recovery virtually impossible on modern NVMe SSDs once garbage collection executes. The operating system sends Dataset Management Deallocate commands to the controller, which marks logical blocks as invalid. The controller's background garbage collector then physically erases those NAND blocks. After a physical erase, the data is gone at the transistor level. No software and no lab can reverse a hardware-level erase. Recovery of deleted files is only possible if TRIM did not execute: the drive was pulled immediately after deletion, TRIM was disabled, or the file system does not support TRIM.
What is the difference between firmware corruption and NAND degradation?
Firmware corruption damages the Flash Translation Layer, bad-block tables, or service-area code while the NAND cells themselves still hold readable charge. PC-3000 can rebuild the translator in workstation RAM and extract user data. NAND degradation is physical cell wear-out: oxide layer breakdown, raised uncorrectable error rates, and unreliable reads from the silicon itself. A firmware fault is logically reversible. Cell-level wear is not, though chip-off into PC-3000 Flash can sometimes salvage what remains on unencrypted drives.
How much does NVMe firmware recovery cost?
NVMe firmware recovery costs $900–$1,200. Cases that also need board-level repair to revive the controller before firmware work fall in the $600–$900 circuit board tier plus the firmware tier. Every case starts with a free evaluation and a firm quote before any paid work begins. If we recover nothing, you pay nothing. +$100 rush fee to move to the front of the queue.
NVMe SSD not responding?
Free evaluation. From $200. No data, no fee.