주메뉴 바로가기 본문 바로가기
KAIST AI-PIM PIM반도체연구센터

IP 검색

CMSS (CXL Memory Sub-System) ECC IP

Unity ECC encoder and decoder for DDR5 x4 memory in CXL memory subsystems.

Integrated rank-level protection against symbol and bit errors using shared parity.

Feature
· 1. Single-symbol correction or double-bit correction per codeword.
· 2. Correction of one failed x4 chip per sub-channel.
· 3. 80-bit codeword containing 64 data bits and 16 parity bits.
· 4. Ten 8-bit symbols with GF(2^8) arithmetic.
· 5. Shared syndrome generator with separate symbol and double-bit correction paths.
· 6. One-cycle detection of error presence.
Application
· 1. CXL memory subsystems.
· 2. DDR5 memory controllers.
· 3. Reliable memory systems for AI accelerators
Business Area
Semiconductor IP, memory controllers, and AI accelerator hardware.
Category

Memory Controller & PHY


Memory Controller & PHY > DDR


Interface Controller & PHY


Comumnication > Wireline > Error Correction/Detection


Tech Specs
  • IP Name :

    CMSS (CXL Memory Sub-System) ECC IP

  • Provider :

    Scalable Architecture Lab, Sungkyunkwan University (SKKU)

  • FPGA Device :

    Intel Agilex 7 M-Series AGMA039R47A1E2VC, BittWare IA-860m

  • Foundry :

    N/A

  • Technology :

    FPGA

Deliverables
· ECC encoder implementation for generating an 80-bit codeword from 64 data bits and 16 parity bits.
·
· ECC decoder implementation comprising a shared syndrome generator, a single-symbol corrector, a double-bit corrector, and an output decision block.
·
· C++ reference model implementing finite-field syndrome calculation, error-pattern identification, correction, and decoding-result classification.
·
· Fault-injection simulator supporting single-bit errors, double-bit errors, chip failures, and configurable combinations of faults.
·
· Build and execution scripts for selecting fault scenarios and on-die and rank-level ECC configurations.
·
· IP datasheet describing codeword organization, correction capability, functional blocks, logical data interfaces, and encoding and decoding flows.
Validation Status
· Hardware functional simulation confirms correction of injected double-bit errors through the syndrome generation, correction, and output selection paths.
·
· The C++ fault-injection model evaluates error behavior using the configured DDR5 x4 chip organization and codeword mapping.
·
· Decoded data are compared with the original data to classify correctable errors, detectable uncorrectable errors, and silent data corruption.
·
· Single-symbol correction covers errors confined to one 8-bit symbol; double-bit correction covers two erroneous bits, including errors in different symbols.
·
· External verification confirmed two-chip correction capability, corresponding to one failed x4 chip in each of two independently protected sub-channels.
Availability
Available for evaluation and integration into CXL memory subsystems.
Functional Diagram
Benefits
· Provides single-symbol correction or double-bit correction within each codeword, covering both chip-related faults and scattered bit errors.
·
· Recovers data from one failed x4 DRAM chip per sub-channel by organizing two beats from each chip into one 8-bit symbol.
·
· Extends bit-error correction capability by assigning otherwise unused Reed–Solomon syndromes to double-bit error patterns, without increasing rank-level parity.
·
· Shares syndrome generation between the symbol and double-bit correction paths, allowing both paths to use the same codeword checks.
·
· Provides bit-error protection at rank level without requiring on-die ECC to perform that protection function.
·
· Detects the presence of an error in one clock cycle, independently of the complete correction latency.
List