Extending the Life of NAND

Envisioning a "football field's worth of space and your own personal electrical power plant" is every CIOs worst nightmare, according to IBM technologist Geoff Burr. Yet, 10 years from now, CIOs may well be facing that scenario if they don't have a game plan to migrate their data centers to solid-state technology.

The evidence to support that proposition rests in past solutions to storage performance and capacity shortfalls, where it has been convenient to simply add conventional hard disk drives when more performance or capacity was needed. Now, a 2008 IBM research paper that Geoff Burr authored points out that "by 2020 this trend [to add storage] will call for millions of HDDs in large server installations. In such a situation, these drives not only consume most of the available space and power budget, but the logistics of dealing with failures during recovery from a drive failure become extremely difficult. These issues can no longer be managed by adding more drives, no matter how low their costs... The goal is to develop a non-volatile, low-cost, high-performance solid-state memory that could extend beyond Flash memory," according to the paper in the IBM Journal of Research and Development (July-September, 2008).

Whether it is through replacing or extending Flash, new techniques and/or alternative technologies will have to produce superior results via some combination of factors such as further scalability, lower cost per bit, and better performance. If the cost per bit could be driven low enough through ultra-high memory density, ultimately such storage-class memory devices could replace magnetic hard disks in enterprise storage server systems.

It comes as no surprise, then, that miniaturization is a constant focus with solid-state NAND, as it has been with hard drives throughout their history. Performance and capacity are improved, at the same time that the cost per bit of storage is drastically reduced. All of this is achieved by increasing the areal (or bit) density, which is the amount of data that can be packed into a given amount of real estate on a storage medium.

NAND manufacturers are keenly aware of this. "It is important to recognize that there are ten times more engineers working on extending Flash to such small dimensions than on all other new technologies combined," Burr says. "So there is always the possibility that with so many smart people working on it, that new breakthroughs will occur, which will make it easier for Flash to scale below 22 nanometers. However, there are some particularly large roadblocks ahead."One of these roadblocks concerns the scaling of NAND memory. "It's becoming more challenging to scale NAND memory, especially horizontally," says Greg Wong, analyst for research firm Forward Insights. "One option to getting around this challenge is stacking."

Three-dimensional, or 3-D, stacking of NAND has an advantage over total NAND replacement technologies like phase-change memory because you do not have to research and prove out new materials and processes as you must with any new technology. Instead, you take the existing NAND process and materials and extend the lifespan of the product.

NAND stacking occurs in one of three ways: horizontal stacking, in which memory cells are three-dimensionally stacked on top of each other horizontally; vertical stacking, in which the memory cells are three-dimensionally stacked on top of each other vertically; and cross-point arrays, which also employ three-dimensional memory cell stacking -- but add to it a redistributing storage mechanism that uses electrical current to change the resistivity of the materials.

"We have compared 3-D NAND memory with new memory technologies like phase-change memory, and have tried to determine what makes the most sense to extend the life of NAND," says Josef Willer, a Munich-based engineer and analyst for Forward Insights. "We think that 3-D NAND has immediate potential over the next four years from cost and implementation standpoints. If you are a manufacturer, you have to consider that it is going to take a longer time to perfect new solid-state memory technologies beyond Flash -- and you can also see that techniques like 3-D stacking can provide immediate capacity and performance improvements that can expand the life of NAND. For this reason, we have looked at new technologies like phase-change memory, but we have also reviewed other options that involve stacking NAND for better performance and capacity."

NAND manufacturers like Toshiba and Samsung began adopting 3-D NAND stacking strategies in the last few years, and there will be significant expansion in the 3-D NAND market in 2009 and 2010. "If you are a NAND manufacturer, when you move to three-dimensional stacking, your processing costs go up but you also achieve much greater capacities," Willer says. "Ultimately, this works out economically because while your manufacturing costs might increase by 20 percent, you are able to double your capacity. In effect, you are reducing your cost per bit because of the higher density of bits."IBM's Geoff Burr still sees phase-change memory as having the brightest future in enterprise solid-state disks, and Willer does not disagree. "What we are most likely to see in the market is that, even as new memory technologies enter the market, there will continue to be market niches for older technologies that coexist with these," Willer says. This has been true for tape drives and hard drives. It will also hold true for NAND, even if technologies like phase-change memory begin making an impact in tomorrow's enterprise storage.

"In our opinion, because of the tremendous costs of developing any semiconductor technology that wants to be -- or worse yet, wants to define -- the bleeding edge of Moore's Law, these technologies will not become commercially available without an obvious and large market," says IBM's Burr. "For storage-class memory, this could happen one of two ways: either as an evolution to higher performance markets after adoption as a Flash 'replacement' in its traditional markets; or as an answer to inadequate performance of NAND Flash when it is used as a solid-state disk in enterprise-level computing."

InformationWeek Analytics has published an independent analysis of the challenges around enterprise storage. Download the report here (registration required).