LUT-Optimized Decoding of QC-LDPC Codes for Flash Memory with EoL Layers Variation Model
Yury A. Zamaraev, Vasiliy S. Usatyuk, Sergey I. Egorov · 2025
NAND flash memory is a pivotal technology in contemporary data storage, but its reliability diminishes due to retention noise and wear-out effects. This paper proposes a novel decoding optimization for Quasi-Cyclic Low-Density Parity-Check (QC-LDPC) codes tailored for NAND flash memory. The methodology incorporates Lookup Table (LUT)-based reconstruction and quantization, complemented by adaptive read threshold adjustments at the End-of-Life (EoL) stage. Key contributions include a Reconstruction Computation Quantization (RCQ) decoder, which achieves reduced computational and routing complexity through a two-stage optimization process, and a dynamic read threshold adjustment mechanism addressing layer-to-layer variation to improve error correction under evolving threshold voltage distributions. Simulation results demonstrate enhanced reliability and performance, with improvements in Raw Bit Error Rate (RBER) of over 0.004 and a throughput-to-area ratio increase of up to 16%. Additionally, the proposed approach extends the endurance of TLC 3D NAND memory by over 30% at a fixed error probability compared to single-layer model and 12% compared to the Neural Network model. This work offers a step forward in improving SSD reliability at the EoL stage.