One standout in terms of technology development news is NVMDurance, a Limerick, Ireland startup headed-up by CEO Pearce Coyle, and the original University of Limerick discovery team Professor Conor Ryan, PhD. CTO Software and Joe Sullivan BEng. PhD, CTO Hardware. Ryan and Sullivan have long experience in Computer Science in Machine Learning.
Ryan and Sullivan realized early in the drive toward solid-state-drives that the NAND-Flash memory devices undergo predictable changes over their lifetime. The team discovered that by carefully profiling a particular manufacturer’s device type they could formulate a controller framework that when applied to SSDs extended the durability of the device by over 20 times.  
A phenomenon called “Charged Trapping” is responsible for limiting flash durability – the number of times that the first cell to fail can be programmed/erased. There’s about 100 Million Volts (per centimeter) stress placed on the oxide whenever the cell is programmed or erased. Continued usage expands results in a conductive “trapping volume” between the drain and floating gate – which is exacerbated each time the cell is programmed or erased. At some point the charge trap volume expands to the point that the cell becomes incapable of storing bits and fails the block.
     
 

I became aware of Ryan’s and Sullivan’s efforts during client work in 2008.
I’ll let their documentation tell their story:

“The company’s core IP is a set of NAND flash optimization techniques that have now been implemented in software. The team has been doing parameter discovery and memory characterization since 2000, initially with SLC NOR, later with SLC NAND, then MLC and laterally, TLC. This has all been done with silicon fabricators and major data storage manufacturers.
Originally applied manually in flash optimization work, NVMdurance has moved to a fully autonomous on-controller system, NVMdurance Navigator. Navigator constantly monitors the condition of the SSD and automatically adjusts the control parameters in real time. Prior to deployment of the flash memory, offline, a great deal of compute-intensive pre-processing is done by NVMdurance Pathfinder, a custom-built suite of machine learning techniques.
HOW DOES IT WORK?
Before the memory product (e.g. an SSD) goes into production NVMdurance Pathfinder, a custom built suite of machine learning techniques, determines multiple viable sets of flash register values. Then on the controller of the live memory product NVMdurance Navigator chooses which of these predetermined sets to use for each stage of life to ensure that the flash lasts as long as possible.
No factory determined operating parameters are ideally suited to your application nor are they tracking the condition of the flash as it degrades through use.
The power behind NVMdurance is the use of off-line machine learning software that automatically learns the optimal parameter settings for the NAND device. It provides either:
•    A static set of parameters, which are set at the start of life for the device, or for optimal endurance
•    A dynamic set, which will periodically optimize the parameters over time as the device ages
The parameters provided are all relative to the manufacturers’ specified parameters so they work even though individual chips/batches have slight variations in their absolute performance.
The actual extension of endurance depends on the particular use-case and the information available to the NVMdurance Navigator autonomic system.”
 
Note that the terms “particular use-case” and “information available to the NVMdurance Navigator autonomic system” – imply that clever people can find malicious ways to negate the company’s methods. What’s more, the company claims that they’ve now proven the claim of “more than 20 times more endurance” on 1x nm chips from multiple manufacturers.
It will get even more interesting when the SSD manufacturers adopt the technology for their drives – reportedly this will be very soon.
The NVMdurance booth was extremely crowded with customers – not bad for a company on its second tranche…,

Tell us what you think in our Forum