Flash is fast becoming the de facto standard for new storage deployments in the datacentre.
HDD-based storage arrays are considered legacy, with focus for primary storage now directed to hybrid and all-flash systems.
At the same time, hyper-converged and software-defined storage products are also pulling more flash directly into servers.
Demand for storage capacity is insatiable, so storage media suppliers have had to look for new ways to increase the capacity of their flash products. 3D-Nand is one technology that increases the capacity of today’s flash devices.
At its introduction, Nand technology looked like any solid-state device, with millions of transistors arranged on a 2D chip, connected together to form memory cells that store data by retaining an electrical charge.
So-called “planar” devices – because they only scale in the horizontal plane – have increased capacity through two methods; by reducing cell size and increasing the number of bits per cell that can be stored.
The latest production processes for planar Nand flash have reached the 15nm level – that’s nanometres, the distance possible between repeated features on the chip – with further reductions proving difficult because of leakage of electrons between cells.
Cell bit-density has been increased from single-level cell (SLC) to double (multi-level cell – MLC) and triple (TLC) capacities, with QLC (four bits) on the horizon in the future.
However, increasing the cell bit-count, which is achieved by storing and measuring multiple voltage levels per cell, has scalability issues. Each increase in cell count significantly reduces the endurance of the cell and the subsequent number of times the cell can be re-written. Each step change reduces the endurance by an order of magnitude, making QLC devices only suitable as a Worm (write once, read many) technology.
What is 3D-Nand?
3D-Nand technology takes a different approach to increasing the density of Nand chips by layering cells on top of each other and creating a 3D structure of cells that allows scaling to occur in the third dimension.
With 3D-Nand technology, the process of laying down the individual cells is repeated across many layers, allowing capacity to be vastly increased, with a negligible increase in size of the Nand chip itself.
In order to deliver reliable products, Nand flash manufacturers had to increase the production process size to accommodate multiple layers, initially starting at around 40nm. But, over time, the production process size has been reduced, with current products currently at around 21nm.
The specific capacities of chips and their architectures vary by supplier. Samsung has typically led the 3D market with its V-Nand products and was first to market with 3D-Nand. Today, the company offers products based on 48-layer technology (256Gb), but 64-layer devices were announced in mid-2016 with capacities of up to 512Gb per die.
Western Digital has also announced 64-layer products, due in mid-2017, based on a partnership with Toshiba. Capacities are expected to be 256Gb, 384Gb and 512Gb per die. SanDisk announced a 32-layer 256Gb product in August 2015, while Intel and Micron have been developing products around 32-layer TLC and 256Gb MLC chip dies.
3D capacity flash
The continued increase in capacity of Nand flash dies means higher-capacity Nand packages (chips) and so larger capacity flash devices. Products are available that use both MLC and TLC Nand.
Samsung already has a 16GB SSD (the PM1633a), which is supported in NetApp and HPE 3PAR arrays. This is set to be replaced by the PM1643 (announced at the Flash Memory Summit in August 2016) with 32TB of capacity, based on 512Gb V-Nand (Samsung’s name for its 3D-Nand technology).
Excluding Seagate’s 60TB proof-of-concept flash drive, the remaining SSD suppliers offer 3D-Nand products that top out in the 2TB to 4TB range.
These include Intel’s Enterprise DC S3520 and P3250 models, which offer up to 1.6TB (SATA SSD) or 2TB (PCIe SSD) of capacity, based on 32-layer MLC.
Micron’s new 5100 series offers up to 2TB of capacity in a range of price or performance models, including the 5100 ECO, 5100 PRO and 5100 MAX. The ECO and PRO models are available in either SSD (2.5in) or M.2 format, whereas the MAX is available only as an SSD.
Obviously, 3D-Nand products still have the same issues of endurance as traditional planar Nand.
But, 3D-Nand presents additional challenges in terms of managing the “disturb” effect – where setting the value of one cell can affect another – and which now has a third dimensional component.
How a chip maker designs its flash to store the charge in a cell can potentially affect reliability. The two main methods used by suppliers are charge trap and floating gate.
Samsung uses charge trap technology, which is seen as potentially more scalable for 3D than floating gate.
However, floating gate is potentially faster on reads and is cheaper to manufacture. Micron has announced technology for 3D-Nand based on Floating Gate, but this hasn’t been confirmed as the basis of any future products yet.
In terms of product reliability, SSD suppliers have been able to make 3D-Nand products as reliable as planar devices.
Micron 5100 devices, for example, have standard two million mean time between failures (MTBF) figures with up to five DWPD (device writes per day), depending on the drive model. This has been achieved by using increasingly sophisticated algorithms to manage the Nand and with over-provisioning of capacity to provide additional wear capacity.
3D-Nand in storage arrays
The continual push towards higher capacity, lower cost storage arrays means 3D-Nand technology offers significant benefits for all-flash and hybrid flash array suppliers.
3D-based products can be offered as a capacity variant by suppliers in their offerings, making all-flash and hybrid options more attractive. The technology has seen quick adoption from suppliers, with many having announced or supported 3D-Nand over the past 12 months.
These include (not an exhaustive list):
- Kaminario, an all-flash array supplier, has been using TLC 3D-Nand drives since August 2015 with the release of version 5.5 of its K2 platform. The use of these devices has allowed the company to quote usable capacity prices of less than $1/GB.
- Pure Storage announced support for 3D-Nand technology in its FlashArray//m platform at the end of 2015. Current products now support 4TB and 8TB 3D-Nand modules, using either TLC or MLC technology.
- HPE 3PAR has supported 3D-Nand technology since November 2015, which at the time reduced effective capacity pricing to $1.50/GB. HPE has since announced support for the Samsung PM1633e 15TB drives.
- In July 2015, Dell claimed to be first with TLC 3D-Nand in a standard storage array, adding the capability to use 3.8GB drives its SC4020 all-flash and SC8000 systems. This allowed usable all-flash capacity prices to be reduced to under $2/GB at the time.
- In September 2016, NetApp launched the AFF A700 system. This provided support for Samsung PM1633a 15TB drives.
3D-Nand in the future
Flash suppliers will move the majority of their manufacturing to 3D-Nand, either as MLC or TLC products, because they offer the best price/capacity for solid-state storage.
We’ve seen flash gradually replace 15,000rpm hard disk drives (HDDs), and the increase in capacity offered by 3D TLC also looks to be edging out 10,000rpm drives. With a range of media from SLC to QLC and the option of 3D technology, hybrid all-flash systems look like they will replicate the tiering previously seen with HDD-based systems. We are edging closer to the all-flash datacentre.
Source: Tech News