ALCod: Adaptive LDPC Coding for 3D NAND Flash Memory Using Inter-Layer RBER Variation
Meng Zhang,Xiaoyi Zhang,Fei Wu,Kai Tao,Feng Zhu,Shu Li,Yahui Zhao,Changsheng Xie
DOI: https://doi.org/10.1109/tce.2023.3319638
2023-01-01
IEEE Transactions on Consumer Electronics
Abstract:Three-dimensional (3D) NAND flash memory has been frequently utilized in consumer electronics as a popular storage device. However, data reliability has become an important problem to be solved. Low-density parity-check (LDPC) codes with superior error correction capability are commonly used in 3D NAND flash memory to ensure data reliability. Unfortunately, high raw bit error rate (RBER) induced by retention time and program/erase (P/E) cycles leads to increased the number of decoding iterations, failing to correct bit errors. Consequently, data reliability cannot be well ensured by using conventional LDPC coding. Moreover, the number of decoding iterations between layers fluctuates greatly due to process variation, which leads to a large difference in decoding latency. To reduce and shorten the gap of inter-layer decoding iterations, this paper proposes ALCod: an adaptive LDPC coding scheme for 3D triple-level cell (TLC) NAND flash memory by exploiting the inter-layer RBER variation. Specifically, this paper first conducts a preliminary experiment, which shows that RBER and the number of decoding iterations between layers and pages have a great difference. And, high RBER induces decreased decoding performance, thus introducing more decoding iterations and time consumption. Then, inspired by these findings, ALCod adaptively performs LDPC coding operations according to the RBER variation induced by retention time and P/E cycles. For pages and layers with higher RBER, by using ALCod, the original bit sequence is split into two parts on average and encoded separately. During decoding, known bits of 0 information are used to improve the initial decoding information (i.e., log-likelihood ratio (LLR) information per bit). To help with the decoding of the codeword’s unknown information, the LLR amplitude of known bits of 0 is increased. The known information and unknown information participate in the same check equation, which, according to the LDPC decoding principle, can provide useful LLR information to speed up decoding update for unknown information. ALCod can improve LDPC decoding performance and effectively eliminate soft decision decoding at higher RBER. Simulation results show ALCod can significantly reduce the uncorrectable bit error rate (UBER), decoding iterations, and time consumption.