Key Takeaways
- EPC memory structure
- User memory strategy
- Kill password security
01EPC memory structure
The EPC memory bank of a UHF tag is a 96-496-bit writable region that holds the unique identifier. GS1 recommends SGTIN-96 encoding: Header (8) + Filter (3) + Partition (3) + Company Prefix (20-40) + Item Reference (4-24) + Serial Number (38). This structure guarantees global uniqueness and is parseable by any GS1-compliant reader.
- Header (8 bits): encoding scheme, SGTIN-96 = 0x30
- Company Prefix: GS1-assigned, variable length
- Serial Number (38 bits): up to 2^38 = 27.4B serials
- Reserve bits for future migration to SGTIN-198
02User memory strategy
User Memory is the extension bank beyond EPC, ranging from 0 to 512 bits. It suits data that "travels with the tag and must be readable offline", such as lot number, production date, and QC status. Caveat: User Memory has no standardized layout — each chip vendor may arrange bytes differently, hurting cross-vendor interoperability.
We recommend the principle "EPC is the source of truth, User Memory is cache". Critical data must live in the backend database; User Memory is only a fallback for offline reads. When writing User Memory, use ASCII with a 2-byte CRC for cross-reader verification. Never store unrecoverable data (e.g. the unique serial) only in User Memory — once the chip dies, the data is gone forever.
03Kill password security
The Kill Password is a 32-bit access credential that permanently "kills" a tag — once executed, the tag is never readable again. This is common in retail after checkout to prevent privacy tracking. But Kill is irreversible and the cost of a mistake is high. We strongly recommend Locking the tag (requiring a password to read) before Killing — this protects privacy while keeping recoverability.
- Kill is irreversible: tag is permanently unreadable
- Lock is recoverable: locked tags can be unlocked with password
- Access Password and Kill Password should be managed separately
- Store passwords encrypted at the gateway; never transmit in plaintext
04Encoding workflow
The recommended encoding workflow has four steps: 1) generate SKU+lot+serial in MES/ERP; 2) dispatch via middleware to an encoding printer or handheld; 3) write EPC then immediately read back to verify; 4) map EPC to business data in the backend. The whole flow should complete within 2 seconds, or it becomes a production-line bottleneck.
The verification step is the most often skipped yet the most critical. We recommend "read-after-write + dual check": immediately read back the EPC and compare to the target, and verify the User Memory CRC at the same time. If either fails, flag the tag as defective and rewrite. A robust encoding station should see a defect rate below 0.05%.
