ZT: Is the SSD honeymoon over in 2010? Sort of...

步从容

The explosion of data in everything from mobile phones to the enterprise has created an explosion of another sort—low-cost NAND memory. NAND memory started small enough, but over time, it has absorbed just about everything that has been thrown at it. It has grown so much that it is beginning to create new markets and is in fact being used in the most demanding applications—as solid-state drives (SSDs) in big iron enterprise and server farms. We all read about how SSDs started to take hold in 2009. Companies of all sizes were signing up en masse filled with the excitement over the prospect of lowering their energy bills, among other things. And who wouldn't? The concept of using SSDs instead of HDDs to store persistent data is very appealing. So what aren't they telling us? We'll likely start to find out in 2010. Don't get me wrong. There is absolute performance and power value in transitioning from HDDs to SSDs. We believe the transition will happen; the issue is that there may not be enough people who are taking a hard look at the technology behind SSDs and asking the tough questions. There is an enormous amount of system level work going into deploying SSDs, but are we all sure the designs will apply beyond a generation or two? Reliability concerns
One of the main causes for concern in using SSDs in the enterprise—where the endurance requirements are the highest—comes down to scaling, or how long the technology will reliably perform at the levels and cost you need today and in the future. Remember with any non-volatile floating gate memory device, the more you cycle the device the more failures you tend to observe, and the less data retention you get. Does the memory device retain the data after hundreds, thousands, or even millions of write cycles? Most of us own multiple devices such as MP3 players and USB memory devices based on NAND flash technology. In normal use, these applications require writing to memory 10, 50, or even 100 times, but never thousands of times. This is the use model that has shaped the technology definition to date. Higher-end applications like SSDs also use NAND flash, but these applications impose quite different usage conditions and requirements. Standards committees are currently debating testing requirements for enterprise, and whether the use model is 100's of GB's written per day or 1,000's of GB's per day. Others argue that the speeds to storage are increasing so fast that even at ~5% bandwidth utilization you can easily get to 10,000's of GB's per day. This is a far cry from MP3 players. A technology that may have been perfect for an MP3 player is going to impose significant limitations in an enterprise-class server. So far, the cost of NAND flash technology has been driven to extremely low levels through the use of leading edge lithography, but it remains to be seen how much lower it can go. That is why many are looking for alternatives. One alternative that holds tremendous promise is phase change memory (PCM). Much has been said and written about PCM, but what is interesting about PCM is that retention is decoupled from endurance. This means that if we cycle a PCM memory device a million times or just one time, the data retention will be identical. To give you some numbers, while NAND endurance capability is reportedly on a decreasing trend, PCM endurance capability is on an increasing trend, with capability in the range of millions to 100's of millions of cycles. That characteristic provides some relief to the most strict-use models. Another phenomenon that makes it easier to use PCM in system-level designs: when we see a failure, it always occurs during a write. So if we are writing data to the device and a write-verify shows that the data is not there, the data can be immediately written again to another location. PCM is not plagued by intrinsic read disturb mechanisms, as NAND has been. Because of the exceptional reliability attributes of PCM, we will see this technology adopted first in applications with the highest most critical requirements. In addition, the PCM performance characteristics are far faster than NAND as well. At a few orders of magnitude faster latency, system designers can focus their attention on using PCM to extend on the core performance value proposition of SSDs in the enterprise. Also, since PCM has no erase, sectors can be directly overwritten, thereby enabling consistency in seek latency across a much wider range of use conditions. The good news is that there is hope and a very promising alternative that should allow the industry to continue the transition from HDDs to SSDs with confidence. However, we see many of these issues facing NAND technology in SSDs coming to light in 2010 as designers begin to plan for 2011 and beyond. - Ed Doller
  Chief Technology Officer
  Numonyx