Application Optimized Adaptive ECC with Advanced LDPCs to Resolve Trade-Off among Reliability, Performance, and Cost of Solid-State Drives

Abstract

The performance of NAND flash based solid-state drives (SSDs) is highly dependent on the application's read and write characteristics [3], where intensity is defined as ratio of read:write requests, and hot/cold considers the write frequency. Moreover, NAND flash memory's reliability degrades with write/erase (W/E) cycling. To optimize performance and reliability, conventional error-correcting code (ECC) scheme switches from fast conventional Bose-Chaudhuri- Hocquenghem (BCH) to slower conventional Low Density Parity Check (LDPC), when the page error rate exceeds BCH's decoding capability. However, advanced LDPCs have been reported, called Quick-LDPC [8] and Error- Prediction (EP-) LDPC without (w/o) upper/lower cells [8], which have (i) higher error correction capability compared to conventional BCH and (ii) shorter decoding time than conventional soft-decoding LDPC. Therefore, this paper proposes an application optimized adaptive (AOA-) ECC for Multi-Level-Cell (MLC) NAND flash-based enterprise SSDs. AOA-ECC includes a new algorithm to efficiently combine the two advanced LDPCs, considering the application's characteristics and memory's W/E cycles. A firmware in the proposed SSD system chooses the optimal advanced LDPC, based on whether the application is read/write-intensive and/or write- hot/cold. Using the proposed AOA-ECC SSD with MLC NAND flash, performance improves by up to 3-times, the reliability improves by 57% and the ECC decoder area decreases by 25%.