Flash memory has been one of my topics of predilection lately. Some 3D XPoint coverage here, a TECHunplugged research paper on the Solid State Industry Memory in 2019 (with an accompanying blog post here) and one earlier TECHunplugged research paper on Computational Storage are a testimony to that. Perhaps the central piece has really been the research paper on the Solid State Memory Industry, where among multiple matters I covered movements in the industry. That specific section of the paper briefly covers the whereabouts of the joint-venture between Western Digital and Toshiba Memory, which makes a perfect introduction to today’s blog article.
As you may know, I was invited in February to Storage Field Day 18, where one of the presenting companies was Western Digital (WD).
From Disk to Flash
It’s interesting to see how some companies endure and transform, WD is certainly one of those. In the heyday of personal computers in the early 90’s, Western Digital was a recognized hard drive manufacturer; there was a certain sense of pride and bragging when one could say that their PC had a Western Digital Caviar drive. I still fondly remember the day when we went to the spare parts shop to upgrade our venerable 80286 AT computer and it’s 40 MB IDE drive to move on to a 486 SX 25 (which then turned out to be an SX 33, and which I promptly clocked at 40 MHz) and… a beautiful, brand new Western Digital Caviar 850 MB drive. Talking about hard drives, there might be a follow up post on WD’s latest innovation in hard drives, but for now let’s focus on flash.
Moving fast forward to 2019: through the years, Western Digital has acquired multiple companies and formed several joint partnerships and is now present in all of the storage media segments. From consumer to enterprise, and from disk all the way up to silicon, there’s probably a Western Digital product in that segment. So let’s have a look at what Western Digital have been up to in the the flash memory segment.
Innovation in 3D NAND
From Planar NAND to 3D NAND
During their presentation, WD made the case for their innovations in the 3D NAND field. The move from flood gate-based Planar (2D) NAND to CTF (Charge Trap Flash) 3D NAND and the use of cylindrical-shaped memory cells allowed to achieve much greater precision (by reducing the proximity effect of 2D NAND) and densities in the production of flash memory.
Cylindrical cells allow more charge per state, and this allows to use the same cell design for SLC, MLC, TLC and QLC devices. There is always a physical limit to what is achievable, and even if solutions are technically achievable it doesn’t means that they are profitable, but one could envision that WD (and likely other silicon manufacturers) may be looking at increasing the density of their flash memory by allowing the cylindrical 3D NAND memory cells to store even more electrons.
Design & Manufacturing Challenges
In fact, the move to 3D NAND is producing different challenges. The lithography size is no longer the main challenge. 3D NAND consists of stacked memory cells which resemble very much the tallest state-of-the-art skyscrapers being built these days.
When building in three dimensions, you want to make sure not only that your foundations are solid, but also that your structure is perfectly aligned from ground to top. This is one of the major challenges of building 3D NAND, since the stacked layers of memory cells need to be all connected to the same electrical circuits (to make it simple).
You’ll agree that putting together 96 floors (at the time of writing 96-layer is the highest available density -from a layer perspective, not talking about cell charge/states- of 3D NAND) and drilling a hole from top to ground may be a bit challenging. Now imagine doing this at the incredibly small scale of memory chips and you’ll get and idea of the tremendous amount of engineering needed (I always think of the size of a micro-SD card to put things in perspective).
Interestingly, WD circumvents the problem by using a modular approach where “blocks” of layers are superposed one on top of another to achieve a greater precision in build & assembly. Nevertheless, I apologize for using very simplified concepts to explain how 3D NAND is built.
The process is extremely complex at this nanometer scale (I wonder if there’s an Imperial unit equivalent – the nanofoot?) and R&D is not only about making the entire memory chip more dense, but also to increase the yield of chips per wafer, while keeping costs in control.
Beyond 96-Layer 3D NAND?
Interestingly enough, WD and Toshiba Memory announced in early March 2019 that they are readying 128-later 3D NAND flash (likely TLC), with a goal to hit the market in 2020-2021. This brings the question of how many layers will we hit before the industry comes up with the “next big thing” and decides to focus innovation on other areas.
It would however be a hazardous bet to ditch 3D NAND any soon. The death of the hard disk has been announced here and there over the years, and even if it is not mainstream media anymore, we always see more innovation hitting the market, increased densities & capacities, so I certainly wouldn’t throw 3D NAND overboard any soon for the same reason.
Western Digital and Persistent Memory
Talking about innovation and the “next big thing”, what had me interested is whether WD is doing anything yet in the Persistent Memory space. Currently, it seems that only Intel and Micron have been working into building an entirely new market around 3D XPoint memory.
I concede that other persistent memory media types are pre-dating 3D XPoint, but crosspoint seems to be a turning point in the industry where we can expect a massive adoption curve over the next years as cutting-edge applications become more and more mainstream.
Where’s Western Digital at?
Low Latency Flash NAND
Western Digital announced Low Latency Flash (LLF) NAND at Storage Field Day 18 as their technology which will fit between DRAM and 3D NAND. It will compete in the same segment as 3D XPoint (currently Optane).
LLF NAND will use a “Charge Trap Device” architecture (similar to Charge Trap Flash, presumably?) and will leverage 1 or 2 bits per cell, with access times in the microsecond range. Price-wise, we can expect it to be priced similarly to Optane DC Persistent Memory i.e. more expensive than regular NVMe Flash capacity, but still more affordable (WD says 10x cheaper) than DRAM.
WD hasn’t been very talkative however about the details surrounding their LLF NAND technology, besides saying that it fits into the BiCS4 3D NAND architecture, which makes me believe that it can just be the same production process and similar CTF cells, but with different building tweaks here and there.
Let’s remember that because they will use only 1 or 2 bits per cell, and because persistent memory devices are usually less dense from a capacity standpoint than regular 3D NAND TLC or QLC devices, there may be architectural changes about how the cells are built and disposed, or how the die / vertical stacking may be done. Since I’m not an expert in the area, this could very well be pure speculation.
Is LLF NAND Persistent Memory?
Note however that we need to take my mentions about “persistent memory” with caution. All that we know so far is that LLF NAND will be faster than regular 3D NAND and thus competing in the Optane courtyard from a performance perspective. Certainly, 3D NAND and its derivates are persistent storage.
The term of Persistent Memory refers however to persistent storage that is memory addressable. It is not clear year whether LLF NAND will be memory addressable.
The fact that WD announced in 2016 that they are working on a 3D ReRAM-based persistent memory technology may make the reader ponder as to whether LLF NAND will be used at all for SCM use cases, or if it will just be confined as a faster 3D NAND tier until (and if even at all) it eventually supplants 3D NAND.
Max’s Opinion
Besides bittersweet comments about the time that passes and disappears forever, leaving memories of then-awesome and now-outdated technologies, the main takeaway from this session about flash memory was the introduction of Low Latency Flash NAND.
The insights about Western Digital innovation and commitment to building Flash memory using 3D NAND technology were truly remarkable and interesting, however I was a bit disappointed that not so many details about Low Latency Flash NAND were introduced.
What is however interesting and exciting is that Western Digital is looking at this large gap that exists between DRAM and 3D NAND. This is exciting for the industry and also for customers and use cases.
We are far away yet from adopting persistent memory as a standard storage method, and as we discussed recently in a TECHunplugged Podcast episode, nearly all of the existing applications on the market are built to use memory as a transient storage space, while disk and flash are used as the persistent storage media.
A new era of opportunities is opening itself with the introduction of persistent memory. New applications will be built, or existing applications completely redesigned to take profit from the innovations introduced, which also means programmers need to become acquainted with new application design models.
The circle of innovation sometimes completes a full revolution and it is interesting to see how we get back to one previously known technology decades later. Innovations with batteries and miniaturization made the electric car dream possible again; innovations with memory architectures from a performance, miniaturization and cost perspective are making persistent memory architectures that existed 40-50 years ago sexy again.
We’re just at the dawn of this brave new world and I’m quite excited to see what technologies will be mainstream in 10 years from now.
Disclosure
This post is a part of my Storage Field Day 18 post series. I am invited to the event by Gestalt IT. Gestalt IT will cover expenses related to the events travel, accommodation and food during the event duration. I will not receive any compensation for participation in this event, and I am also not obliged to blog or produce any kind of content. Any tweets, blog articles or any other form of content I may produce are the exclusive product of my interest in technology and my will to share information with my peers. I will commit to share only my own point of view and analysis of the products and technologies I will be seeing/listening about during this event.
Additional disclosure: Western Digital offered a WD Black SN750 1 TB NVMe drive in M.2 format to each delegate. This was an unsolicited gift which is not influencing the writing of this article.