Paper Title: Intelligent Read Threshold Tracking to Mitigate Read Retry Requests

Paper Abstract: Read Retry (recursive rereads from the NAND with a preset sequence of adjusted read thresholds) is a popular and widely used option to recover from error correction code (ECC) failure. Although undoubtedly a powerful technique, it does undesirably increase read latency since (often several) additional read operations are required. This presentation addresses how we can do away with Read Retry by using artificial intelligence (AI) techniques that learn the best read thresholds to use ahead of time. This talk will present a simple heuristic learning technique that aims to continually optimize the read thresholds throughout the lifetime of the SSD device. In particular the algorithm does not require any data to be reread during the learning process, so no latency penalty is incurred. Using the optimized read thresholds, I will show results demonstrating the significant reduction in mean raw bit error rate from the NAND flash compared with using default preset read thresholds. This makes ECC failure less likely to occur, reducing or eradicating the need for Read Retry, thereby reducing and stabilizing read latency. To end, I will discuss how further AI techniques can be incorporated.

Paper Author: Oliver Hambrey, Research Engineer, Siglead

Author Bio: Oliver Hambrey is a research engineer at Siglead Europe Limited which he joined in 2010. He has a Master of Mathematics and MSc in Complexity Science degrees from the University of Warwick, UK which he obtained in 2007 and 2009 respectively. At Siglead Europe, he has worked on a variety of projects including the development of low complexity constrained coding algorithms, HDD magnetic read head quality estimation, detection techniques for two-dimensional magnetic recording, NAND flash characterization and the design of the spatially coupled BCH ECC engine used in Siglead’s SATAIII SSD controller.