Bidirectional Error Correcting Codes and Interference Mitigation for Flash Memories

Abstract

Recently, NAND multi-level cell (MLC) flash memories are now widely used due to low cost and high capacity. However, when the number of cell levels increases, cell-to-cell interference (C2CI) which shifts threshold voltage may degrades the error rate in reading process. There are several approaches to alleviate the errors caused by the threshold voltage shift and we discuss error correcting codes and message encoding schemes.

First, we propose error correcting codes that are effective for multi-level cell flash memory and non-binary WOM (write once memory) codes. In particular, we focus on bidirectional error correction codes. The errors in MLC flash memories tend to be directional and limited-magnitude. Many related works focus on asymmetric errors, but bidirectional errors also occur because of the bidirectional interference and the adjustment of the hard-decision reference voltages. The code treats both upward and downward errors when the error magnitude in each direction differs. The maximum magnitudes of the upward error and downward error are lu and ld, respectively. One of proposed codes extends the technique of the distinct sum sets to the bidirectional error correction codes. The other code is bidirectional limited magnitude error correction codes based on modulo operation and uses non-binary conventional error correction codes. These proposed codes can reduce the parity size, and have better error correction performance than the conventional error correction codes when the code rate is equal. Furthermore, error correcting schemes for non-binary WOM codes are discussed. WOM codes is a coding scheme that allows information to be written in a memory cell multiple times without erasure, and conventional error correction codes cannot be directly applied to WOM codes. The advantages of the proposed methods are that these are practical and systematic codes, and the complexity of encoding and decoding processes are low. We also introduce effective error locating limited-magnitude parity check error correction codes for the MLC flash memory error with lower complexity.

Second, we introduce coding schemes to lower the generated interferences by cell to cell interference. It is known that C2CI is caused by the threshold voltage change of neighbor cells in writing operation. The amount of threshold voltage change is proportional to the magnitude. To minimize the generated interference, the average magnitude needs to be decreased. We propose two new C2CI reduction coding schemes that adjust the average magnitude to reduce C2CI. The proposed coding scheme deals with q-ary message codes, and generates fixed length codes. Message codewords are divided into several blocks, and are modified by modulo addition with proper values to minimize the average magnitude. We also propose low energy Huffman codes based on entropy coding when the frequency of symbols is not distributed uniformly. This scheme produces variable-length codes without redundancy. We modified Huffman codes to minimize average number of high bits ('1' bits). We show that proposed codes generate optimal codewords which have minimum high bits with minimum average codeword length.