Identify the vendor from the prefix
The first two or three characters of a memory part number are effectively a manufacturer signature. Learn eight of them and you can place almost any chip you will meet on a phone board, even with the logo worn off.
| Manufacturer | Prefix | Parts here |
|---|---|---|
| Samsung | KLM · KM · K | 7,796 |
| SK hynix | H9 · H2 | 5,252 |
| Micron | MT · JW / JY / JZ … | 2,013 |
| YMTC | YM | 1,103 |
| SanDisk | SDIN · SDAD | 120 |
| Kioxia | TH · TC | 83 |
| Kingston | EMMC · KE | 91 |
| FORESEE | FE · NCE | 85 |
What each scheme actually encodes
Samsung — KLM, KM, K
KLM marks storage-only eMMC. KM marks a stacked MCP with LPDDR in the package. The four characters after the dash identify the revision — two parts differing only in that suffix are usually the same capacity on different silicon.
SK hynix — H9, H2
H9 is used broadly across stacked eMCP and uMCP parts; H2 appears on storage-only packages. The markings are dense and small, and this is the vendor whose codes get misread most often at the bench.
Micron — MT, JW / JY / JZ …
Marked two ways. MT is the full part number. The short five-character FBGA code is an index into Micron’s catalogue, not a specification — both forms resolve here.
YMTC — YM
Chinese vendor, increasingly common on recent budget and mid-range boards sold across Asia. The code describes die generation and density fairly directly.
SanDisk — SDIN, SDAD
SDIN on eMMC, SDAD on stacked packages. Capacity often appears near the end of the string, which makes these among the easier codes to read at a glance.
Kioxia — TH, TC
Toshiba Memory until the 2019 rename. Older stock still carries Toshiba branding on otherwise identical silicon, so both names appear in the catalogue.
Kingston — EMMC, KE
Frequently second-source parts, so the same capacity can appear under several closely related part numbers.
FORESEE — FE, NCE
Longsys sub-brand. Turns up on low-cost hardware and in the aftermarket replacement trade.
Markings that are not part numbers
Three things on a memory package look like identifiers and are not:
- The vendor logo block. Samsung's
SECblock with a short digit group beside it is a manufacturing marking. Searching it returns nothing because it is not a part number. - The lot code. Second line, changes every production run, same character set as the part number. This is the single most common reason a lookup fails.
- The date and origin code. Third line, typically a country abbreviation and a four-digit year-week.
The part number is the top line, on every vendor, without exception.
Why you cannot just decode the capacity yourself
It is a reasonable question — the capacity is in there, so why not read it off directly? Three reasons it does not work in practice:
- Every vendor uses a different position and encoding. There is no shared convention to learn.
- The schemes change between generations. A rule that holds for one Samsung series does not hold for the next.
- Stacked parts encode two capacities. On eMCP and uMCP the storage and the LPDDR are both in there, interleaved with package and speed-grade information.
Which is why 16,776 part numbers sit in a lookup table rather than in anyone's head. Read the prefix to know the vendor, then check the full code for the specification.
Second-source and rebranded parts
Not every chip on a board came from a tier-one vendor. Smaller houses buy die and package them under their own marking, and aftermarket replacement stock in particular skews toward these. They are not automatically bad parts, but their documentation is thinner and their codes are less consistent. Where a lookup returns a vendor you do not recognise, treat the capacity as indicative and verify before you commit.
See also: eMMC vs UFS vs eMCP for what the package type changes, and the mobile IC code list for common part numbers grouped by capacity.