What are the 10 most important things designers need to know about flash memory today? Participants in the closing panel of the Flash Memory Summit offered different lists, each coming from different perspectives, and each including insights and predictions that some may find to be surprising or controversial.
I went to learn more about flash memory, which has become a key part of the Cadence design IP strategy following last year's acquisition of Denali Software. In combination with the opening keynotes, which I blogged about previously, this closing panel offered an excellent snapshot of what's going on with flash memory from both a business and technology perspective, and what's coming up in the near future.
Panelists provided three different "top ten" lists as follows:
The panel was chaired by Andy Marken of Marken Communications. Here are the lists, with supporting comments made by the panelists.
Troy Winslow's Top Ten List
1. NAND solutions have a clear value proposition today. Client computing, data centers, and applications in the cloud are drivers.
2. That value proposition will only increase over time. Flash has already demonstrated a 20X performance improvement over hard disk drives (HDDs) and more improvement is coming.
3. Not all flash solutions are created equal. "This means we can differentiate from one another. On the flip side, it means we all have to be aware about quality issues."
4. Standard plug-and-play solutions and interfaces are key. Only solid state drives (SSDs) can take full advantage of SATA 6 Gb/s or PCIe, but standardizing around these interfaces is vital.
5. Process lithography advances are a fact, but the industry benefits. "As manufacturers make advances in lithography, all are going to benefit."
6. Lithography advances are a challenge that not all will overcome. But Intel will. "We are continuing to push on that [scaling] wall, whether it's 3D NAND or PCM [phase change memory] or some other technology, we are going to deliver it."
7. The right partnerships add value. Always true, but becoming more relevant as NAND gets more complex.
8. There will be SSD consolidation. This is the one previous statement that did NOT become true. "The top ten providers supply 90% of the volume, and 200 others fight for the other 10%. It's a little crazy. I don't understand it."
9. Flash memory is the storage of the future. "I'm not saying we're replacing 100% of HDDs, but there's no question that nonvolatile memory solutions are the future of storage."
10. Get in the game, try an SSD! Try it, and you won't go back to rotating media.
Radoslav Danilak's Top Ten (12, actually) List
1. Cost and density improvements are continuing. 1x nm flash and 128Gb die is just around the corner. Oversupply will probably result in reduced prices early next year.
2. Endurance continues to deteriorate. This is not a big problem for laptops (as was originally feared), but existing approaches at 1x nm may only write 1,000 times, not good enough for enterprise applications.
3. The flash industry is fundamentally changing. Consumer flash products are migrating to managed flash, and the PC market is moving from aftermarket to OEM SSD sales.
The next three points are for consumer NAND flash:
4. Managed flash is becoming more and more complex. "The technology around the controller needs to improve. We need flash that requires less design effort to use."
5. Managed flash is becoming mainstream. Basic controller functionality and error correction (ECC) are being incorporated with the flash.
6. Raw NAND is for very high volume applications only. Unmanaged flash works for a limited number of vendors due to increased IP requirements.
The next three points are for client NAND flash:
7. The flash SSD market is maturing and changing. Two years ago, most sales were from channel marketing; this year most SSD sales will be through OEMs.
8. Consolidation is ending the Wild West of SSDs era. (Note: This is what Troy Winslow predicted two years ago, and it didn't become true then...we'll see what happens now).
9. Flash is becoming the performance limiter, not the controller. Today, performance is limited by the controller; in the future it will be limited by the flash.
The final three points are for enterprise NAND flash:
10. MLC (multi-level cell) flash cost is required for mainstream enterprise adoption. Innovation is needed to improve reliability and endurance, and thus reduce costs.
11. Flash requires vertical integration at enterprise level. Further endurance gains can't be realized at the controller or SSD level.
12. The next 10x improvements in endurance have already been found. StorCloud has developed technical solutions for 10x improvement in endurance, enabling 1x nm MLC in enterprise storage systems.
Jim Handy's Top Ten List
1. Enterprise SSDs will be used in all data centers. Some data centers use SSDs today, others are cautious; eventually they will all embrace SSDs.
2. There is still a lot of growth in NAND. "A lot of applications we can't dream of are going to happen with NAND flash."
3. Controllers will get more sophisticated. "We're not seeing incremental error correction, but much larger chunks going to 8 bits, or 24 bits -- and somebody mentioned 100 bits."
4. System software will be designed for NAND first. "You're going to start seeing systems where software was first designed with SSDs in mind and then later on it was adapted backwards so it can work in systems with HDDs."
5. Tablet PCs will morph into newer devices. Today's tablets aren't exactly what the market needs, and by 2014 they will have much more capability.
6. Not everyone can be a successful SSD supplier. "Right now anybody with a good soldering iron who can buy an SSD controller, a PC board, and some NAND flash can cobble together an SSD...that whole market is going to be weeded out."
7. NOR flash has a long future in code storage. The chips are cheap and will continue to be used inside PCs.
8. NAND in PCs is a threat to DRAM, not to HDDs. "The PC of the future is going to have a complement of an HDD, NAND flash, and DRAM that will give you hard disk drive storage and SSD-like performance."
9. The death of flash is not imminent. Scaling breakthroughs have already taken us much farther than people thought a few years ago. Until NAND flash really hits the scaling wall, no alternative memory technology will take over.
10. SSDs in PCs will lose out to NAND plus HDD. "If you can get a hard drive for $50 and put $10 in NAND flash around it, it's going to meet the needs of most PC users."
Both Handy and Danilak emphasized the importance of flash memory controllers and flash management, as well as the complexity that flash management involves. This points to an increasing need for third-party memory and storage controller IP, which, as described in a recent blog post, is a cornerstone of the Cadence design IP strategy.
Another important point is the need to design with an awareness of flash. Handy predicted a time when system software is designed for flash from the beginning. Danilak said that "if you just build an SSD and plug it into a system, you cannot get full performance benefits. But if you build solutions around flash from the ground up, you can increase endurance way more."
So in the end, the NAND flash story is not just about NAND flash silicon -- it's also about management and control, and effective use by system software. The EDA360 vision describes the development of systems in which hardware and software are co-optimized to serve end-user applications. This concept applies to memory just as much as logic.
Other Flash Memory Summit blog posts:
Flash Memory Summit: New Insights Into the Future of NAND Flash
3D Thursday: A look at some genuine 3D NAND cells, courtesy of Micron
Apps define storage task, create application-specific SSD
Intel's Knut Grimsrud explains how to get another 20x improvement in SSD performance