Prospects for improving the reliability of flash memory for SSD

Solid-state drive (SSD) flash storage devices accelerate notebook and server performance. As organizations embark on a digital transformation journey and leverage technologies such as artificial intelligence, there is a demand not only for higher-capacity storage systems but also for ever-faster access to data to better understand the data they collect.

For example, Total Gas & Power deployed flash storage to accelerate data center applications that need to integrate with cloud services. The corporation has used a Nutanix hyper-converged infrastructure and is replacing NetApp storage devices with a flash-based storage array.

Enterprise storage systems have become synonymous with huge arrays of hard drives spinning at thousands of revolutions per minute, with magnetic heads hovering just a few micrometers above them. If the head comes into contact with the disc, data may be lost. Flash memory on SSDs using Nand chips gets rid of spinning disks, so it should be more reliable, right?

In 2016, researchers studying the use of flash in data centers warned that on days when flash drives are used, in most cases there will be at least one fixable error.

True, the researchers reported that other types of transparent errors that a disk can mask from the user rarely compare in number with nontransparent errors.

The results of a study of the reliability of flash storage in data centers were presented in February 2016 by Associate Professor at the University of Toronto Bianca Schroeder and Google engineers Raghav Lagisetti and Arif Merchant. They have been evaluating the reliability of flash storage in Google data centers for six years. The authors noted that flash memory is replaced much less frequently than traditional hard drives. However, flash memory has a much higher percentage of uncorrectable errors.

The study found that up to 80% of drives created damaged data blocks, and 2-7% had damaged Nand memory chips during the first four years of its existence. These drives, which do not have mechanisms for tagging defective chips, require repair or return to the manufacturer.

SSDs are undoubtedly more reliable than traditional enterprise disk arrays, which are also expensive to maintain. The SSD firmware continuously monitors write cycles and detects failures to flag blocks of memory that are causing too many errors. However, in an IT automation landscape, Tribe admits, operating systems are not yet fully utilizing the data provided by SSD firmware. Reliability is measured by the number of I / O failures - read and write to disk. But in the future, the SSD firmware will be used for preventive maintenance, in which the system itself will monitor the wear of the SSD.