Noise Recycling Based Multi-level Flash Memory

In this seminar, I will present a new decoding approach for multi-level cell (MLC) flash memory that improves storage efficiency while keeping computational complexity low. The method, called the Noise-Recycle-based Decoder (NRD), leverages a block-partition (BP) mapping strategy and a sequential, multi-stage decoding process. The key idea is to estimate the noise from already decoded layers and then recycle it—through subtraction—to improve the decoding of subsequent layers.
The NR approach assumes simultaneous reading of an entire MLC, which provides a fixed, correlated noise realization across all layers in a cell. This NRD scheme significantly enhances robustness to noise, reducing both bit and block error rates under challenging noise conditions.
I will also discuss the theoretical foundations of the approach, including reliability bounds under noise shifts and a new capacity bound for NAND flash memory with BP mapping. Finally, I will share simulation results demonstrating that our NRD scheme consistently outperforms conventional independent decoding methods, especially under large noise shifts. The seminar will highlight both the practical performance improvements and the theoretical contributions of this work to modern flash-memory systems.

M.Sc. student under the supervision of Assistant Prof. Alejandro Cohen.