Q&A: As prices fall, flash memory is eating the world

SanDisk's top strategy guru talks about the expanding role of flash in the enterprise and how emerging storage techs like 3D NAND will change the future

1 2 3 Page 2
Page 2 of 3

What continues to hold enterprise back with flash? Just the price perception issue?

First of all, think about the EMCs and NetApps of the world as the primary suppliers of storage solutions. A lot of their customers are looking to these companies and they have been adopting flash at varying rates. EMC has had some faster adoption of certain flash solutions.

NetApp has been slower. As you know, they recently bought SolidFire and got rid of their internal product line, but certainly the pace of flash adoption has been uneven amongst the large systems companies.

We wanted to come out with a solution that allowed customers to have a lot of choice. We have created a platform that allows companies to look at different types of open-source software married with our hardware as well as different types of software stacks that we are making available. We are coming out with CEPH we are coming out with GPFS from IBM, we are coming out with Tegile's software stack, we are coming out with Nexenta's software stack. Tegile and Nexenta have broad deployments in multiple workloads.

We tend to be much faster moving than some of these traditional companies. It is a very disruptive time in this environment and customers are trying to assess these different solutions. We are getting closer to the point where more and more of these solutions are going into production in the months ahead.

You've written about something called storage-class memory. Can you define that?

You can think of storage-class memory as a new tier in the overall hierarchy. If you think about cache in the processor and then you think about DRAM, and today you have flash and hard drive. Most applications program to this hierarchy.

Storage-class memory is nonvolatile memory fitting into the hierarchy between DRAM and flash. It is somewhat slower than DRAM, but it is nonvolatile. It is much faster than flash and, in price point, it is expected to be in between the [cost] per gigabyte you expect out of DRAM and flash products. Of course, a lot of changes would have to happen in order for that kind of capability, when introduced in volume, to be leveraged by a computer system and the associated programming model. But I think it is going to happen over the next several years.

You've got a partnership with Toshiba to develop 3D NAND. Talk about what you see as the upper limits of that, say in the next five years. What are we going to see there?

We have been scaling 2D NAND for many years, and over many generations we have the cost points down to very compelling levels that have enabled all these different applications around the world that you see with flash. Now we are at the cusp of the industry transitioning to 3D NAND.

We expect our 3D NAND to go into volume production later this year. We expect the industry to also make that transition as well, and 3D technology is going to have fairly long legs. We are going to use this technology to continue down the cost-reduction curve that we have set for ourselves over the past few years and to [deliver] higher-capacity memory at lower and lower prices.

I don't see anything that is going to stop that trend anytime in the next five to seven years. Certainly, we need to make more breakthroughs in technology to continue our roadmap but as far as the eye can see, there is nothing I would lose sleep over. The continued cost reduction and higher-density memory will enable a whole bunch of new products and applications as well as continue the growth of flash, including, for example, having things like laptops go to 100% flash from where they are today.

Do you see any technology on the horizon that would threaten NAND flash as the dominant player in the nonvolatile memory market? In specific, I was wondering your perspective on this 3D XPoint technology that was announced by Intel and a partner.

These are very early days for technologies like 3D XPoint. Intel announced that technology with Micron as the manufacturing partner, and as you know, we have announced a partnership with HP to bring our own nonvolatile storage-class memory to market. We are very excited to have that partnership and a route to market for our technology. They, of course, did a comprehensive level of due diligence on our technology before we went into partnership with them so they feel good about our technology.

We think that when 3D XPoint is introduced it will initially be focused on applications that will draw away demand from DRAM. That's the primary area that is going to get impacted. If you think about how things like in-memory compute, in-memory databases like SAP HANA and others are run, they require massive amounts of DRAM, which tends to get very, very expensive. The best way to really reduce the cost of that system and yet have fairly small impact in the overall performance is to adopt technologies like 3D XPoint or ReRAM, basically the storage-class memory technologies that we were talking about. These memory technologies, when coupled with the right type of changes to the software stack, can allow for in-memory compute performance at a fraction of the cost.

Most of that change will reduce the demand for DRAM in these systems and not impact flash all that much. Of course, over time, as that technology becomes cheaper, it is possible that the high end of the NAND flash business sees some impact in enterprise applications. But that will take a while and it is dependent on the cost of those technologies coming down quite a bit and having enough software deployed that can natively take advantage of this type of storage-class memory technology. A lot of things have to happen before flash starts to get impacted in any material way.

I also wanted to ask you about what's happening in the cloud market. What changes are you seeing there and how are you responding to those changes?

The cloud computing business is growing very rapidly. This growth is going to continue at a very robust pace because the scale benefits that the cloud computing infrastructure can bring to customers is just immense. It is not easy for every company to replicate that type of capability. Cloud computing companies are focused on driving very resilient infrastructures. They are trying to create hardware infrastructures that are very flexible and very high performing. [They] require a minimal level of maintenance and high reliability, as well as high levels of automation because these data centers are becoming very, very large, very unwieldy, they are consuming massive amounts of power.

The big trends in the data center relate to increasing the level of virtualization in applications, increasing the level of automation, reducing the power consumption through deployment of technologies like flash and increasing the use of open source. There is experimentation that will happen in the future with some data centers trying to experiment with ARM-based processors over the next couple of years to see how costs and power consumption can be further reduced using the power efficiency of the ARM core, which is better than the Intel core. I think there are some interesting things that can happen there, but a lot more work has to be done on the software stack in order to make that a reality as well.

1 2 3 Page 2
Page 2 of 3
How to choose a low-code development platform