Future of storage: Micron bets chips on 3D NAND flash – but NOT YET
NAND planarity has a few good years still
As far as the world's number two DRAM chip-maker is concerned, 2D or planar NAND can go through another lithography cycle before 3D techniques may become necessary because a two-year-old NAND design change has enabled Micron to scale its lithography down to 16nm and maybe beyond.
El Reg interviewed Glen Hawk, Micron's VP for its NAND Solutions Group, who briefed us on the changes. He started out by identifying two fundamentals about Micron in the flash business.
First off, he said, it is a flash foundry operator and "flash manufacturers have an inherent advantage with SSDs." Essentially, they have supply certainty – not knowing where your next chip is coming from has sunk many a storage firm – and the integration of flash design, controller functionality and firmware were two of the advantages of this. Secondly, Micron is focused most on the enterprise.
So having sorted that out we asked the busy storage veep a set of questions. His answers have been edited slightly for readability and brevity.
El Reg: What's Micron's attitude to flash software?
Glen Hawk: Software is clearly the next thing. Our best strategy is to work with our customers who have expertise in that area. So far it is not needed for our growth.
El Reg: What's Micron's attitude to NVMe?
Glen Hawk: That's clearly something worth pursuing. There are rival standards and we'll support whatever our customer base is interested in.
El Reg: What does Micron think of three-bits-per-cell (TLC) flash?
Glen Hawk: We do have a robust portfolio of TLC chips. Our strategy, with each lithography, we have SLC, MLC and TLC products. During different market cycles we'll run different products.
It's a tight market now and we reserve capacity for customers wanting SLC and MLC. We are capacity-constrained and can barely supply enough SLC and MLC SSDs. We don't anticipate offering any TLC SSDs in the foreseeable future.
El Reg: Talk about 128Gbit product please.
Glen Hawk: We currently have 20nm 128Gbit product in high-volume production. We're sampling 128Gbit product using 16nm lithography – the same product but shrunk. Micron is now aligning SSD development with the lithography cycle. We'll have a 16nm SSD in the first half of 2014.
Two years ago we changed our fundamental planar cell architecture to use a high K dielectric and metal gate structure. Both these things make it easier for us to continue to scale and so the 16nm lithography became possible.
El Reg: Will there be another planar lithography shrink or will you go 3D?
Glen Hawk: We're pursuing both paths for now. When we reach the cost crossover point of 3D technology depends partly on how many flash cells you can stack on one another.
(Note: Samsung, the number one NAND supplier, has just announced its volume production of vertically stacked 3D V-NAND using 128Gbit parts and three layers, eventually progressing to 24 layers, interconnected by Through-Silicon Vias (TSVs). Samsung does not say what the lithography is in its chip.)
El Reg: Who will integrate PCIe flash with servers, the sever manufacturers or somebody else?
Glen Hawk: We think the server manufacturers will integrate flash PCIe cards. HP, Dell, IBM, etc. service enterprise customers. But there is the existing retrofit market. We have to be multi-lingual.
I think it is inevitable that Fusion-io will focus more on software. Some of the OEMs will evolve too and the value chain will re-establish itself.
El Reg: What do you think about the acquisitions taking place in the flash technology area?
Glen Hawk: Micron is an exception to the wild frenzy of acquisitions right now. We have been systematically building our capabilities from the ground up. We've certainly been in the market, having bought Virtensys in the UK, but some of the valuations have been too high - that reflects the breadth of the gap in competencies [between the acquirer and the acquiree.]
Also, acquirers could be seen to be competing with their customers.
El Reg: It seems clear that Micron could build larger capacity parts using its 16nm technology. The 16nm chips must be smaller than 20nm ones and this will enable Micron and its OEMs to build higher capacity flash devices in the same footprint as their existing devices.
For example, Micron's M500 client SSD uses 20nm NAND and has capacities between 120GB and 960GB. We'd assess that this could simply double with 16nm parts, meaning a 240GB to 1.92TB range.
We could be about to see substantial jumps in flash device and array capacities as a result of this 20nm to 16nm NAND lithography shrink. ®