Noise Recycling Based Multi-Level Flash Memory

We propose a novel low-complexity Noise-Recyclebased Decoder (NRD) for Multi-Level Cells (MLC) to obtain high storage rates. Our proposed scheme utilizes Block Partition (BP) mapping in multi-level flash memory. Based on multi-stage decoding, NRD method decodes layers sequentially, starting from the MSB (layer 1) to improve noise robustness. Specifically, a digital noise realization is estimated utilizing already decoded layers. This estimated noise is then recycled by subtraction in the subsequent layers pre-decoding to improve Bit Error Rate (BER). Noise Recycling (NR) approach assumes simultaneous reading of an entire MLC, ensuring a fixed correlated noise realization for decoding all layers within a cell. For noise shifts across multiple representation levels, we establish a reliability bound and show via simulations that the proposed NRD solution outperforms Independent Decoding (ID) with both BP and Gray mappings without NR. For a single-level noise shift, we analytically and through simulations demonstrate that the proposed scheme outperforms the baseline ID scheme with BP mapping and no NR, while achieving equal performance to ID with Gray mapping and no NR. We introduce new capacity and reliability bounds for MLC NAND flash memory using BP mapping under single-level noise shifts.