BACKGROUND IMAGE: iSTOCK/GETTY IMAGES
3D NAND flash will probably stick around for 20 years, despite continued higher cost-per-GB of NAND flash vs. hard disk drives, says semiconductor analyst Jim Handy.
Two-dimensional NAND flash will fade away within five years, and three-dimensional NAND technology could stick around for at least 20 years.
That's the prediction for the future of NAND flash technology from Jim Handy, general director and semiconductor analyst at Objective Analysis in Los Gatos, California.
"3D is the only way to get costs out of NAND flash in the future," Handy said.
Handy said the price-per-GB cost gap between NAND flash and hard disk drives (HDDs) could be 10 to 1 for a long time. But he forecasted that 3D NAND flash will probably be around for 20 years or more because it doesn't appear to suffer from the scaling limitations once thought to be a barrier to the NAND technology.
"You can build a 3D NAND stack and finish it up, and then build another 3D NAND stack on top of that and finish that one up. And you could continue to do that for dozens, hundreds, maybe even thousands of layers," Handy said.
That was the upshot of a panel discussion at last year's Flash Memory Summit in Santa Clara, California, according to Handy. He chaired a session featuring advanced researchers from leading NAND flash manufacturers.
In contrast to 3D NAND flash, two-dimensional or planar NAND technology has encountered scaling difficulties. Handy said manufacturers have tried to pack more transistors onto a die to reduce the cost, but ultimately that has meant they can store fewer electrons in a smaller area.
Manufacturers also have tried to store more bits per cell. They have progressed from one bit per cell (single-level cell, or SLC) to two bits per cell (multilevel cell, or MLC) and now three bits per cell (triple-level cell, or TLC). But with each step, "the number of electrons between one bit and the next shrinks," Handy said.
"You get down to a point where there are just not enough electrons," said Handy. "And that's the real problem that's keeping NAND from shrinking much below 15 nanometers."
Below is a transcription of TechTarget's interview with Handy on the future of NAND technology.
Transcript - Analyst: 3D NAND technology could stick around for 20 years
Do you think flash drives will ever become cheaper than hard disk drives on a cost per GB basis?
Jim Handy: We've been talking for a long time about that at Objective Analysis. It seems that historic trends have proven that there's always going to be a 10-to-1 gap between the two of them. And, although a lot of people have projected out trend charts that show them crossing over, when you put them on a semi-[logarithmic] scale, which is what we tend to do, where it shows steady growth as a straight line, it shows two parallel lines -- one for NAND flash and one for hard disk drive and what their cost per GB is.
There has been some talk about how hard drives have reached some kind of a plateau and they can't cram any more bits in. But there are a number of technologies, some that are being introduced today, like helium and shingled write, and some that will be introduced in the not-too-distant future, especially what's called heat-assisted magnetic recording -- HAMR. Those will continue to push hard drive price per GB down probably as rapidly if not more rapidly than NAND flash. So, we're expecting for that 10-to 1 price ratio to last for a long time.
Is the future of NAND flash for enterprise use two-dimensional or three-dimensional?
Handy: Enterprise really tends to follow what happens in the client space. So, we saw enterprise drives move to MLC flash only a couple of years after the first MLC client drives came out. We're expecting to see 3D be used in the enterprise maybe a little bit later than planar NAND. But to tell the truth, it seems like the enterprise is catching up and is actually adopting technologies nearly as fast as they are in the client area.
3D is the only way to get costs out of NAND flash in the future. So, if you were to check back with me in two years, we probably would see very serious 3D penetration in both enterprise and client drives, and probably by five years, there will be no such thing as a planar solid-state drive.
What are the pros and cons of 3D NAND flash?
Handy: The only con to 3D NAND flash today is that it is very difficult to master. And this is something that the manufacturers will get over. They always do. They find some new process that when they first start making it is just atrociously difficult. And then at some point or another, they have an 'aha' moment where they learn how to make the process work. And after that, then they just go into mass production. And it's never a problem again. We need that 'aha' moment for 3D NAND because even though Samsung is making a lot of 3D NAND right now, they're not really making it cost effectively or efficiently. Other manufacturers also need to get to that point.
So, that's the only detractor. The benefits are great. It was designed originally to get the costs out. So, costs will reduce, but also one of those strange little side benefits of 3D NAND is that most of them use a charge trap, and charge traps take less time to program and take less energy to program. And so 3D NAND flash is actually faster than planar NAND flash, and it consumes less power to do writes.
What are the main challenges associated with 2D, or planar, NAND moving forward?
Handy: The big problem is the number of transistors that are available to store a bit of information. Two ways that the price has been worked out of planar NAND is to go the standard semiconductor shrinking way, which is what Moore's Law actually was based on. It's getting more transistors onto a certain size die so you can get the cost down, just by getting more for the same cost. And that reduces the size of the floating gate on the NAND flash bits. I won't go into what that all is, but it's just, if you've got a smaller area, you can store fewer electrons.
The other way that costs have been worked out is by storing increasing numbers of bits per cell. So we've gone from single bit or single-level cell, SLC, to multi-level cell, which is today taken to mean just two bits on the cell, and then to what technologists call three-bit, multi-level cell, but most people call TLC, triple-level cell. And that's storing three bits on a cell. Basically, you can get your bits for one-third the cost if you're storing three of them on a cell than if you're just storing one on a cell. But each time you do that then the number of electrons between one bit and the next shrinks. So with TLC, if you can only store a few thousand electrons on a cell, you could use those few thousand electrons in SLC to show a one or a zero. You could use one-quarter that much for MLC, the two-bit MLC, because you'd be storing four levels on the cell to indicate two bits. And with TLC, you're actually storing eight levels on the cell. And so you only get one-eighth as many electrons to indicate whether it's a one or a zero.
You get down to a point where there are just not enough electrons. And so that's the real problem that's keeping NAND flash from shrinking much below 15 nanometers. And that's what's driving 3D to become the next generation of cost reductions in NAND flash.
Should enterprise users concern themselves with the type of NAND flash when they buy storage systems?
Handy: Enterprise users do tend to focus on what's in there, and they look askance at new technologies. But what tends to happen is that the controller technology for any SSD develops to the point where it masks any issues or differences between the new technology and the old technology. I'm fully expecting to see that happen with 3D NAND.
But what's interesting is, as I mentioned before, 3D NAND technically is an easier technology to deal with. It also has larger gates, which is another benefit that I didn't highlight before. That means that the error correction is not as difficult of a task in 3D NAND flash as it is in planar NAND. Because of that, the controllers for 3D NAND actually don't have to work as hard as controllers for, say, TLC planar NAND. Because of that, I would expect enterprise controllers to very rapidly develop to take advantage of 3D NAND and the enterprise users will never really have to worry about making the change from planar NAND to 3D NAND.
What are your long-term predictions for NAND flash? How long do you think it's going to be around?
Handy: I've always been an advocate from the demand side of NAND flash playing a much larger role in computing. We see reasons to believe that computers in the future will combine DRAM, NAND flash and hard disk drives. And with the advent of technologies like the Intel-Micron 3D XPoint memory, there will probably be another memory layer in there so that it could go DRAM to 3D XPoint to NAND flash to hard drives.
But whatever happens, NAND flash will end up in all computers just because of the fact that it gives a better cost-performance tradeoff than DRAM does. It's actually more of a threat to DRAM than it is to hard disk drives because of the fact that it improves performance without having to add DRAM.
Now SSDs are not the optimal way to add flash to a computer. They actually slow the flash down considerably. When accessing the flash the way that you access the SSD, that software also needs to be optimized to work. So I would expect to see computing architecture move to a model where NAND flash is not put behind an SSD interface but it actually could be on the PCI bus. But I would expect it to be maybe even on a faster bus like the [Open NAND Flash interface] ONFi bus that Intel tried to put onto PCs back in 2009. That will probably come of age, and when it does, then that will become a very large use of NAND flash. It's just in general computing applications.
As far as the [NAND] technology and manufacturing it, it used to be that people thought that NAND flash would go through three generations of three dimensions, and then nobody would be able to make anything beyond that because the trenches that would have to be dug into the multiple layers would end up being too narrow and too deep to manage in a fab.
At the last Flash Memory Summit, I was chairing a panel of advanced 3D NAND researchers from NAND flash manufacturers, Hynix and Micron, from Applied Materials. And what we ended up hearing on that panel from the gentleman from Hynix was that 3D NAND can be made a whole lot higher in stacks than this limitation that was previously thought to be a barrier to the technology, because what you can do is you can build a 3D NAND stack and finish it up and then build another 3D NAND stack on top of that and finish that one up. And you could continue to do that for dozens, hundreds, maybe even thousands of layers. So, that indicates that 3D NAND flash is probably going to be with us for 20 years or longer.
Flash technology will get even hotter in 2016
How flash storage technology tops HDDs
Phase-change memory promises advantages over NAND