We already wrote about an SSD that vanishes from BIOS, about a firmware panic that looks like a fake brand, and about NVMe hiding in a camera slot. This is the same class of drive in a different chassis: a 30 mm NVMe that lives under a Steam Deck plate, an Ally back cover, or a Legion Go shield, and that people assume is somehow not a normal SSD because Valve or ASUS or Lenovo put it there.
It is a 2230 NVMe, not a Deck-only widget
M.2 2230 means the module is 22 mm wide and 30 mm long. The protocol is NVMe over PCIe. The parts are the same ones we already see in ultrabooks and small-board PCs: a controller, NAND, a PMIC, firmware, and a flash translation layer. The sticker might say Western Digital, Kioxia, Micron, Kingston, or a no-name aftermarket upgrade. The host is SteamOS, Windows, or a dual-boot. None of that makes the silicon a "Steam Deck SSD" in the recovery sense.
The one real exception is the original 64 GB LCD Deck. That unit used soldered eMMC, not a removable 2230. If that board is dead, you are not pulling an M.2. Everyone else in this class — 256/512 LCD Decks, OLED Decks, aftermarket upgrades, ROG Ally, Legion Go, and the other 2230 handhelds — is holding a short NVMe.
This is also not an Intel Optane H10. That stick is two devices on one M.2. A Deck 2230 is one namespace. Different geometry. Different mistake.
Why sleep / resume is a common way these die
A desktop NVMe spends most of its life in a high power state. A handheld does the opposite. The APU and the SSD share a small battery and a cramped thermal budget. The drive is asked to drop into deep PCIe / NVMe power states — APST, L1.2, suspend-to-RAM, sometimes hibernate — dozens of times a day, then come back in a second when you open the lid or hit a button.
That is a lot of firmware transitions on a DRAM-less or near-DRAM-less 2230 that is already running warm against a shield. Three things go wrong in that loop, and they look the same from the boot menu:
- A write that never finished. Sleep, a battery cut, or a hard power-off can land in the middle of a mapping-table or garbage-collection write. The NAND still has the files. The translator that turns those pages into LBAs may not.
- A power-state the firmware does not resume from. Some 2230 firmwares — stock and aftermarket — are sloppy about coming out of the lowest NVMe states. The host sees no device, a generic NVMe, or a stick that enumerates once and then drops. That is a controller / firmware problem, not SteamOS "forgetting" the disk.
- Thermal and power budget. A 2230 in a handheld sits next to a hot APU with a thin pad and a shared rail. PMIC sag, a controller that will not complete its boot, and NAND that has been run hard on long sessions are the same SSD failure modes we already describe — just in a smaller, warmer box.
What you see: SteamOS or Windows will not boot. The Deck boot menu / BIOS no longer lists the SSD. SMART tools fail or show 0 bytes. The drive is gone from the OS disk list. Sometimes it comes back for one resume and then dies for good. Intermittent is still a dying NVMe. It is not a cue to keep cycling it.
Power it down and get the 2230 evaluated. We will tell you whether this looks like an NVMe the controller will still talk on, or a different failure. Free evaluation, no data, no fee.
What not to do
Do not keep force-rebooting. Holding the power button until the logo comes back, again, is the handheld version of power-cycling a vanished SSD. Two or three attempts tell you it is not coming back. After that you are giving a panicked controller more chances to scribble. Same rule as SSD not detected.
Do not reimage over a dying drive. SteamOS recovery images, Windows installer, "just reinstall and my games will redownload" — that is a full-disk write. It assumes the NVMe is healthy enough to take a new partition table and a new OS. A dying 2230 will often accept some of those writes and then drop. What it accepted is gone. Cloud saves and Steam re-downloads do not bring back local screenshots, recordings, non-Steam libraries, or anything that only lived on that stick.
Do not run destructive disk tools. GParted, diskpart clean / convert, nvme-cli format, vendor "SSD repair" or firmware updaters, TestDisk / Recuva pointed at the live module, chkdsk / btrfs check --repair on the original. Those tools are built for a disk that is already talking. On a firmware-locked or dropping NVMe they write to the only copy of the map. Forum "2230 Deck fix" flashers are the same class of mistake as the SATAFIRM factory-reset tools: they provision empty drives. They do not preserve yours.
Do not treat a USB 2230 enclosure as a diagnosis. If the stick enumerates at full capacity, clone off it once to a different drive and stop using the original. If it shows 0 bytes, a generic device, or nothing, the enclosure did not "unlock" it. You are looking at the same controller state we already cover on Phison NVMe 0 MB / generic identifies.
Do not assume chip-off will bypass it. Modern NVMe is usually encrypted to the original controller. Pulling the NAND packages off a 2230 and reading them raw often yields ciphertext. We will say which case you have after we look. We will not promise chip-off up front.
What recovery actually is
We do not run a Valve salvage utility. We do not reimage SteamOS in the lab to "see if it boots." We do not treat the chassis as the patient.
The job, when it is possible, is standard NVMe work:
- If the controller still enumerates — even as a short or generic identify — we talk to that controller with tools that are trying to preserve the translator, then image to a copy. That is the same path as a laptop M.2 that dropped after sleep.
- If it is a firmware lock / ROM / safe mode — 0 bytes, busy, generic NVMe — the work is controller-family work, not "Deck work." Phison, Silicon Motion, and the other parts that ship in 2230s are the same families we already see on 2280s.
- If nothing enumerates — often PMIC, a rail, or a controller that will not come up — that is board-level, the disappears from BIOS case. The NAND may still hold the data. The original controller usually has to come back, because of encryption.
Worn NAND is still worn NAND. A 2230 that was the only disk for a game library plus recordings has been written hard in a hot chassis. That is the degradation problem, not a Valve problem. We will tell you if the flash itself is the limit. We don't advertise a success rate.
There is no Steam Deck, Ally, Legion Go, or 2230 handheld case in our public case log. We will not invent one. The published SSD write-ups on that log and the blog are other devices — an Optane H10, Phison SATA ROM mode, a Sabrent Rocket NVMe. Different boards. Same rule: we do not fabricate a handheld serial to make this post look like a bench report.
What to do right now
Do not reimage. Do not format. Do not keep holding the power button until it "takes."
- Power it down. Leave it off. One more resume attempt is not a diagnostic after the SSD has already vanished from the boot list.
- Pull the 2230 if it is removable — once, with the unit cold, using the shield/screw the manufacturer documented. Do not reseat it ten times to see if it enumerates. Label it. If this is the soldered 64 GB eMMC Deck, do not start prying chips. That is a board job, not a weekend teardown.
- Note what you have: Deck LCD vs OLED vs aftermarket 2230, Ally, Legion Go, or other handheld; SteamOS, Windows, or dual-boot; whether it died on resume, after a battery death, or after an upgrade image. Photograph the boot menu / BIOS screen. Photos are not a diagnosis.
- Get the stick evaluated. Send the 2230, not a USB clone of whatever 0-byte identify an enclosure showed. The technician on your case will tell you whether this looks like an NVMe the controller will still talk on, a firmware lock, or a board that never comes up. Free evaluation, no data, no fee.
If you are mailing it in, pack the module like any other failed SSD — how to ship a failed drive safely. A padded envelope is how 2230s arrive bent. For the work itself, see SSD & NVMe recovery. If the drive vanished from every host, start with SSD not detected. If a tool already printed 0 MB or a generic NVMe, the closer firmware write-ups are SATAFIRM S11 and Sabrent / Phison NVMe — same class of panic, different sticker. If the only problem is that NVMe is what everyone is shipping now, that is the SATA-to-NVMe post.
Steam Deck / 2230 handheld NVMe — FAQ
Is the Steam Deck SSD a special Valve part?
It died right after sleep or resume. Is that a different failure than a laptop NVMe?
Should I keep force-rebooting or reimage SteamOS to "fix" it?
Can I clone it in a USB 2230 enclosure?
Is recovery some Valve-specific salvage?
Is there a Steam Deck or 2230 handheld case in the public log?
What should I do right now?
The handheld will not boot and the short SSD is gone from the menu. That is still an NVMe. Start with a free evaluation before anyone reimages it.
Request free evaluation →Free evaluation · No data, no fee · Talk directly with a technician.