Read Voltage Optimization in MLC NAND Flash Memory via the Density Evolution

The multi-level-cell (MLC) NAND flash memory can typically use multiple reads for obtaining soft information for the LDPC decoder. The multiple reads give the soft information which is quantized to a certain level. The major challenge is that the read voltages must be precisely selected to provide better soft information. In the previous work, the read voltages are selected so that the soft information has the maximum mutual information (MMI). However, the error-correction capability of LDPC decoder is not considered. Therefore, in this work, we analyze the performance of LDPC decoder by density evolution whereby the soft information is quantized. Then the optimal read voltages for given LDPC codes are obtained. As a result, for a regular LDPC code with the read voltages optimized by density evolution can provide the lower BER performance compared with the MMI technique.