2013/06/22 by Daniel Miller, Daniel L. Miller, Miller, Daniel L.
Computer Science · Mathematics · #Advanced Data Storage Technologies #Cellular Automata and Applications #Distributed systems and fault tolerance #Error Correcting Code Techniques #FOS: Computer and information sciences #Information Theory (cs.IT) #cs.IT #math.IT
paper · pdf · doi:10.48550/arxiv.1306.5350
4 pages, Submitted to IEEE Transactions on Information Theory
arxiv created 2013/06/22 · openalex publication_date 2013/06/22 · arxiv updated 2013/06/25 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
The scaling of high density NOR Flash memory devices with multi level cell (MLC) hits the reliability break wall because of relatively high intrinsic bit error rate (IBER). The chip maker companies offer two solutions to meet the output bit error rate (OBER) specification: either partial coverage with error correction code (ECC) or data storage in single level cell (SLC) with significant increase of the die cost. The NOR flash memory allows to write information in small portions, therefore the full error protection becomes costly due to high required redundancy, e.g. ∼50%. This is very different from the NAND flash memory writing at once large chunks of information; NAND ECC requires just ∼10% redundancy. This paper gives an analysis of a novel error protection scheme applicable to NOR storage of one byte. The method does not require any redundant cells, but assumes 5th program level. The information is mapped to states in the 4-dimensional space separated by the minimal Manhattan distance equal 2. This code preserves the information capacity: one byte occupies four memory cells. We demonstrate the OBER ∼ IBER3/2 scaling law, where IBER is calculated for the 4-level MLC memory. As an example, the 4-level MLC with IBER ∼10-9, which is unacceptable for high density products, can be converted to OBER ∼10-12. We assume that the IBER is determined by the exponentially distributed read noise. This is the case for NOR Flash memory devices, since the exponential tails are typical for the random telegraph signal (RTS) noise and for most of the charge loss, charge gain, and charge sharing data losses.