We occasionally throw around the '3-bit MLC' (Multi Level Cell) term in place of 'TLC' (Triple Level Cell) when talking about flash memory. Those terms are interchangeable, but some feel it is misleading as the former still contains the term MLC. At Toshiba's keynote today, they showed us why the former is important:
Photo source: Sam Chen of Custom PC Review
That's right – QLC (Quadruple Level Cell), which is also 4-bit MLC, has been mentioned by Toshiba. As you can see at the right of that slide, storing four bits in a single flash cell means there are *sixteen* very narrow voltage ranges representing the stored data. That is a very hard thing to do, and even harder to do with high performance (programming/writing would take a relatively long time as the circuitry nudges the voltages to such a precise level). This is why Toshiba pitched this flash as a low cost solution for archival purposes. You wouldn't want to use this type of flash in a device that was written constantly, since the channel materials wearing out would have a much more significant effect on endurance. Suiting this flash to be written only a few times would keep it in a 'newer' state that would be effective for solid state data archiving.
The 1x / 0.5x / 6x figures appearing in the slide are meant to compare relative endurance to Toshiba's own planar 15nm flash. The figures suggest that Toshiba's BiCS 3D flash is efficient enough to go to QLC (4-bit) levels and still maintain a higher margin than their current MLC (2-bit) 2D flash.
More to follow as we continue our Flash Memory Summit coverage!
Since Toshiba’s 15nm MLC NAND
Since Toshiba’s 15nm MLC NAND has shown up in consumer SSDs already (Plextor M6V), it is likely we will see QLC as well, positioned as “value” SSDs.
Maybe as a quick emergency
Maybe as a quick emergency backup medium, in the last hours before the Hurricane/Typhoon storm surge is about to flood the data center, or more likely the SMB main offices, and things need to be done in a hurry! Just enough time for the IT folks to get out of town to a quickly set up satellite/temporary headquarters location where a proper Tape/Spinning Rust backup could be made and the temporary headquarters rental computers could be re-imaged with the company’s mission critical management software and latest working data sets. But for archival storage, no way in damnation! Spinning Rust, Tape, or that long term archival DVD/whatever disks, but not Never-NAND land, no just no! Quad level Cells for archival storage, that’s hexadecimal madness!
Oh man this just sounds like
Oh man this just sounds like a really bad idea. TLC already scares the hell out of me. Over the course of 20 years data is just something I DO NOT want to take chances with or have to be concerned about. I guess doing proper backups would help ease the mind, but this just sounds like they’re trying to stretch a technology that is already nearing its max potential.
Even 3D stacking will not go on for more than 10 years before MRAM, STRAM, PCM, or something else must come along.
In other news Toshiba has
In other news Toshiba has announced a new line of long term water storage containers each with its very own unique placement of holes along the bottom.
By calling it archival, they
By calling it archival, they seem to mostly be implying that it isn’t going to be fast enough for general storage. So how slow is it going to be? I would assume it is still going to be faster than a hard drive. I wouldn’t trust it for actual archival storage though. Hard drives are actually not that good for long term storage either. If you leave a drive on a shelf for a few years, it may not start back up when you plug it in. Tapes have the advantage that they do not have as many failure points as a hard drive since the storage media is separate from the reading device. The long term optical storage tech may be the best way to go, if it actually works, although the capacity isn’t that high.
If we have some really fast, intermediate, non-volatile memory, like Intel’s xpoint tech, then the next level out doesn’t need to be as fast. This may mean that flash will be moving into more of a mass storage market segment. If Intel’s xpoint works as advertised, then I don’t see how flash can compete with it. It will be more expensive than flash, but it will still probably be cheap enough to use the way most flash devices are used. Most people have a relatively small flash device and then a large hard drive for mass storage. If the small flash drive is replaced by new technology, then a large part of the current market for flash goes away. Flash will need to move into the mass storage market, so it needs to be a lot cheaper and a lot higher density. This tech accomplishes that but I would be concerned about 16 voltage levels. This seems a bit extreme but they should be able to guarantee the performance by doing a lot of error correction.
Reads would still be just as
Reads would still be just as fast as other flash, but writes would require extra care as you need at least double the precision as compared to MLC. When flash pages are written, cells are incrementally nudged towards the target voltage. If you go too far, you have to erase the block and start over (or give up on that page and retry the write with another fresh page), which is why the extra care and time is needed.
Hopefully they can design the
Hopefully they can design the controller such that doing a significant amount of error correction does not slow the device down. Even with the larger amount of charge trapped on a 3D cell, the amount of degradation that would cause an error will be very small with 16 levels. I wonder if they will have to implement something where they detect if a block has not been written in a long time, and refresh it if so. As long as the drive is powered, then such a strategy should work fine. We all depend on DRAM with charge stored in capacitors that need to be refreshed ever few milliseconds or something.
There would probably be a lot of use for such a device. I don’t know what a company like YouTube does for storage, but there are a lot of videos that might go unwatched for a very long time, but we still expect it to come up almost instantly. Something like this type of flash storage would work quite well for that and may be lower power than keeping arrays of disk spinning for data that is unlikely to be accessed. For real mass storage, they still use tape systems some places. They have a robot that goes and finds the desired tape and moves it to a reader. This is quite a bit higher latency, so I don’t think it would work for something like YouTube or any VOD services unless you don’t mind waiting for a while for your video to start. .
Like many I’m totally freaked
Like many I’m totally freaked out by TLC. It’s not good enough. Period. However this one sounds interesting. Archival style of drive without blazing fast transfers. It may be the thing I need. Have few arrays with rarely written data. Let’s face fact archival drives don’t need millions of MB/s R/W. Looking forward to see drives with like 250 MB/s R/W, 5TB of space and 250$ price tag. Time to replace clunky HDDs.
TLC has been more than proven
TLC has been more than proven so long as there is sufficient cell volume to make it reliable. 850 EVOs on VNAND have been selling like hotcakes and working fine even with folks beating on them for months straight (just as if they were MLC).
Forget about 4-Bit MLC.
Forget about 4-Bit MLC. Everyone knows that it has higher capacity but downside is that it reduces write endurance, reliability, and may cause inconsistent random read/write performance. Thus, 4-Bit Storage Technology is so cheap. So I recommend stick with 2-Bit MLC for balanced performance and price point.
NOW we’re talkin’.
NOW we’re talkin’.
While TLC has a shorter
While TLC has a shorter lifespan than MLC, the dangers of TLC are greatly exaggerated. A typical desktop (gaming, browsing, Word docs, photo editing) user will not wear out any SSD, TLC or MLC.
http://techreport.com/review/27909/the-ssd-endurance-experiment-theyre-all-dead
http://www.anandtech.com/show/6459/samsung-ssd-840-testing-the-endurance-of-tlc-nand
NAND cells with three leave
NAND cells with three leave it be!
The less cell states to manage the faster the access. And Crossbar memory is incoming, by by NAND! Spinnng Rust will not go away, nor will Tape, backups need longevity.
True, TLC will most likely
True, TLC will most likely last long enough before a person replaces the drive. Unfortunately, some people are cheap and I’ve seen quite a few horror stories of people being “cheap” and buying consumer SSD’s (TLC especially) using them in server workloads and within months the drive giving out all kinds of errors. Turns out those certain server workloads were just shredding through those drives like nothing.
People will be cheap, but because we’re so used to using consumer gear that’s “good enough” they’re often used in serious workload environments. It isn’t hard to vouch for people when it comes to other hardware being used in a server environment even though they aren’t designed for it, but SSD’s are a completely different beast. Sadly most people aren’t aware of how wear-leveling works and that it’s a serious impairment to use an SSD outside its intended environment. Literally throwing money into a black hole.
Great, now your SSD can die
Great, now your SSD can die from just having a text file saved to it.
Toshiba has announced that
Toshiba has announced that resulting in capacities of up to 128TB by 2018. Which means the QLC SSD may have some special use such as library.
Those consumers may need a cheap and large disk to store their data. They won’t write too much, all they want to do is to read.
If the QLC SSD can offer a cheap enough solution, it can be a good idea.