The Flash Memory Summit at the end of August highlighted just how far flash-based memory has come compared with the spinning hard drive. We saw announcements of 60 TB and 100 Tb solid state drive (SSD) products, likely to be delivered in the first part of 2017, while performance levels of 10 gigabytes per second on a single drive will be available soon. Hard disk drives (HDD) have been left in the dust!
One analyst asked me, “Why, with all the advantages of SSD, do we not see a total replacement of hard drives?” That's an excellent question. One might argue that supply and demand drive that. However, it's only part of the story.
The traditional large OEMs for servers and storage systems mark up SSDs disproportionately higher than hard drives, leading to the appearance of a large price gap. They are able to do this because of proprietary hooks in their systems that require drives to be purchased from the system maker. This lock-in has really slowed SSD demand, but the inexorable growth in SSD's already large feature lead is now overcoming this.
We are now seeing a migration from locked-in systems to more open hardware, where the drive purchase model will move more towards distribution. At the same time, data integrity and system uptime are moving rapidly away from the traditional redundant array of independent disks (RAID) approach with “enterprise” drives and dual-port interfaces, etc. to use of consumer class drives.
This means a fall in SSD pricing at the end-user level, in part due to the changing drive model mix, in the server and storage system space, which is the largest consumer of NAND flash. This will lead to a demand increase for server/storage products. At the same time, hard drives in PCs are being phased down in favor of flash, with the (flash-based) tablet and smartphone business eroding the PC Market, so adding to flash die demand.
Net this out and demand is increasing rapidly for flash product. More and more, then, the ability of the industry to deliver flash die will be balanced against pricing acting as a break on total replacement. While the HDD makers are deprecating the idea that their spinning drives will go away, and even have talked about demand rebounds, the question now for the industry is not if, but when HDDs are effectively replaced in new sales.
From a supply-chain perspective, the question is focused on the supply of flash die, and the capacity each die stores. Almost all of the action in the die foundry space is in China. The PRC has a strategic goal to achieve chip self-sufficiency, both for its own growing demand and for systems and phones sold to foreign markets. They have a quite a way to go. Today, only 15% of the chips they use are homegrown.
China is determined to change that situation. They aim for 40% of chips to be fabbed in China by 2020. Incentive plans and liberal funding, such as China' National Semiconductor Industry Investment Fund (known as the Big Fund), to build fabs have encouraged a spurt of growth. XMC, for example, has broken ground on one of the largest fabs, due to come on-stream in early 2018 and targeting 3D NAND production with 300mm wafers.
Samsung will be on stream with its expansion at its Xian 3D NAND facility in early 2017, while, not to be outdone, Intel is retooling its 300mm Dalian facility for 3D NAND by 2016 year end. There are numerous other projects in China targeting the same market. The expected move to 3D NAND by Toshiba/Sandisk and SK Hynix (Both of whom are running roughly a quarter behind Samsung and Intel on 3D product release and production ramp) will further boost 3D NAND wafer production.
Expertise may however delay ramp up by China's home-grown fabs. Tsinghua's attempts to buy into Sandisk and Micron, as a way to obtain know-how, have fallen flat in the face of the US Government.
DRAMeXchange estimates that 3D NAND will reach 20% of total NAND production by year end, with the implication that the shift to high capacity 3D NAND die will significantly increase exabytes shipped. Add in the new capacity and supply is going to see a large boost through 2017.
Wafer production isn't flash's only attack in the war on HDD. 3D NAND technology, while complex, increases die capacity dramatically. The approach also backs off the standard NAND curve for capacity versus node size (the size of transistors on the die) and allows higher die yield with existing die-making equipment.
One reason for the slower than expected start of 3D production ramping was the difficulty of adding die layers beyond 32 to 48 layers. The problem's root cause was that current production gear is unable to maintain alignment on high layer counts.
A solution for the alignment issue has been a “coarse/fine” layering approach, where 48-layer die structures are stacked on each other using the same sort of through-silicon via approach as with the fine 48-layer structure.
Another factor is cell durability. Using larger node sizes means a much better wear life for flash cells. This leads the industry to replacing single-level cell (SLC) products with much cheaper two and three level solutions, leading to another large increase in capacity shipped.
With DRAM investment being moved to 3D NAND and Micron converting DRAM production facilities to 3D NAND, financial analysts are pointing to something of a flash oversupply situation in 2017. This is a difficult prediction to make, though, since pundits tend to be less accurate when major inflexion points in demand or technology are involved.
The very high capacity drives mentioned at the start of this article won't arrive until 2017. We can expect the volume ramp to begin some months later, which is where the new foundry capacity will be kicking in. With coarse/fine 3D NAND technology mastered by then, flash capacity measured in exabytes will be up dramatically. This suggests that flash supply will match demand growth, while cost per terabyte will continue to drop.