SSDs Have Bleak Future, Says Researchers

Page 5 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Toggle sidebar Toggle sidebar
Status
Not open for further replies.
The paper's title is "The Bleak Future of NAND Flash Memory?". Clearly it is commenting *only* on NAND technology and the current trends and implications. Even if NAND is doomed and the anticipated flaws cannot be overcome, there are loads of other technologies being developed, such as memristor, PCM, ReRAM, STT-RAM. The SSD isn't doomed.

The link for the paper (The Bleak Future of NAND Flash Memory?) is given in the article:
• http://cseweb.ucsd.edu/users/swanson/papers/FAST2012BleakFlash.pdf

And, as for those bleating that this is only a "grad-student", she also has 2 co-authors. A 2 second Hoogle search found out the facts. John D. Davis is a Microsoft researcher with a PhD and a long string of peer reviewed publications. He isn't going to put his name on junk research (http://research.microsoft.com/en-us/people/joda/). Steve Swanson is an Assistant Professor at University of California, with a PhD and a long string of peer reviewed publications (http://cseweb.ucsd.edu/~swanson/).

There are so many IDIOTs commenting on articles around here, spouting off without reading the article or making any attempt to understand what has been written. What is really tragic is that it is not that people are so incredibly stupid or that they spout out moronic comments, it is that they are too damned lazy to read more than the article's title, and too damned lazy to lift a finger and do the slightest shred of research, before spouting out their non-sense.

Dr Stabissmuss
 
I dont' get it. So in 2024 my SSD will be only 10x fasted than HDD instead of 20x ? Oh no, I'm going to kill myself.
 
These people are supposed to be intelligent. I have been in the business for 30 years and have watched technology. From tape drives to floppy to hard disk...various technologies came and went. The difference with SSD is that it is very possible to advance that technology and as assembly lines grow, the manufacturers will be purchasing larger and larger volumes of raw materials that go into their production. As their purchasing increases, technology will be invented to produce the materials they need in larger quantities and lower prices. As this process continues, refining will become better. As with any technology, look at MSI motherboards. Worst boards I have ever tried. More returns then any other boards I have built systems from...same with SSD technology, there will be high quality and low quality brands.
 
These people are supposed to be intelligent. I have been in the business for 30 years and have watched technology. From tape drives to floppy to hard disk...various technologies came and went. The difference with SSD is that it is very possible to advance that technology and as assembly lines grow, the manufacturers will be purchasing larger and larger volumes of raw materials that go into their production. As their purchasing increases, technology will be invented to produce the materials they need in larger quantities and lower prices. As this process continues, refining will become better. As with any technology, look at MSI motherboards. Worst boards I have ever tried. More returns then any other boards I have built systems from...same with SSD technology, there will be high quality and low quality brands.
 
[citation][nom]freggo[/nom]1984... bought my first hard drive. 5 1/4 inch, full height ( that's like 2 regular size CD-ROM drives on top of each other).Cost... 2 weeks pay.Capacity unformated : 5MBTape drive manufacturers made it clear that this technology will never replace the massive amounts of storage provided by tapes.Anyone have a tape drive in their computer ?[/citation]

tape is still used in massive backup situations. those things have a massive amount of storage and have less fail points than a hdd, making a more reliable backup... problem is they are single use almost.

[citation][nom]guruofchem[/nom]And those that can't do either one spout moronic platitudes...[/citation]

i thought those that cant do either teach gym.

[citation][nom]stingstang[/nom]You left out that tape drives from back in the 1990s still have more capacity than most HDD's today. If a company wanted to, it could have kept innovating that type of data storage, and we'd all have tapes in our computers. The world just went a different way..[/citation]

no, we would NEVER be using that today as a primary storage, or even as a storage device in active use, it would be strictly backup... however if the evolution of the tape drive was as researched as it is for hdd, ssd, and such, we may all have a dvd drive size tape drive, for backups. and tape is FAR FAR more reliable than disc based, and ssd based storage.

 
As far as that goes, back in the early 90s we didn't have the Internet, we had BBSing (Bulletin Boards). That technology would have killed the Internet if the smallest amount of development went into it. We could run software remotely, group chat...most everything the Internet can do now...and more in those days then the Internet can even dream of at this point. Streaming video and audio was easy except for the limited bandwidth since we did not have broadband internet, only dialup and for the longest time most of us were waiting for the 56k modems being stuck with 300-36000 baud modems. I had a 2400 baud modem for a long time...
 
This is ridiculous... Yes, NAND will undoubtedly not be viable in it's current forms at some point in the future, 2024 doesn't seem like a bad estimate. However, we have so many possible alternatives, many of which are supposed to be in the consumer market within that time frame, some of which are expected to arrive in the consumer market by 2020 or even earlier.

Here's a list of non-volatile memories being researched that I recall: Racetrack, Millipede, MRAM, PRAM, SONOS, PMC, FRAM, NRAM, nvSRAM, and RRAM. There are more and if you count volatile RAMs being researched, there are even more. Furthermore, many of the non-volatile memories I listed have multiple variants being made, further increasing the number of possible replacements for NAND Flash memory.

An SSD does not need to be made of NAND Flash, so no... SSDs are not doomed. An SSD is just a storage drive that has no moving parts, unless I'm mistaken. We will simply have SSDs made of one or more of the above technologies or another one that I'm not aware of.
 
[citation][nom]danwat1234[/nom]No, by 2024...I think a 12TB SSD in my computer will be plenty[/citation]

By that time we are prob. switching to 4K video and your 12TB are not that much anymore 🙂
A movie can eat up in excess of 1TB quite easily today !
 
I don't see why so many are harsh on the researcher because she is a graduate student. Lots of excellent work gets done in university, some real breakthroughs that have guided our lives.

Secondly, just because someone publishes research that illustrates a potential issue were we to continue on the way we are going, does not mean she thinks it isn't possible to overcome. Technological progress does not "just happen", it is the result of a lot of people's hard work. The first step is isolating a potential barrier, the second step is overcoming the barrier.

Read the research paper, it is only an exploration of a potential barrier that must be overcome. I would care to bet that everyone involved would agree that these limitations will be overcome. It may involve flash, it may involve a different solid state memory type.
 
For me personally, I am not criticizing the grad student but the big names in the industry that put their reputation on this research. Anyone with a bit of industry knowlege should know that many technologies that have issues but are major breakthroughs have room for advancement. To say ss days are numbered is rather glib...
 
12 years from now? 12 years ago we were using zip disks that held 100 megs and were 100 times the size of an SD card.
 
SSD sales are being held back by their cost per GB. Going from $1/GB to $0.10/GB or less will do more to open the market than a 2.5x bump in r/w latencies. I'm more disappointed by the projected limit of 16TB per drive than the increased latencies. This is the best poised technology I've found to overtake spinning platters in storage densities and $$/GB, but it looks like it may be the end of the road about the time flash catches traditional hard drives.

If latencies get worse as the process size shrinks and this is a problem we cannot address, then why has the performance of flash SSD's gotten considerably better? Any latency problems should be addressable by a combination of hardware and software.

Reconfiguring the RAID type technology used across the chips and adding capacitor backed disk cache should fix most of it. If you add in some load balancing software to avoid hammering a single region of the flash, larger block sizes, push for more streaming reads, and Dell stops overcharging people for RAM to cache data, then that will fix another large chunk. Imagine Windows 10 pre-buffering parts of your SSD to unused RAM while the drive idles after boot.

Most services that would be hurt by a doubling of flash latencies are probably using a layer RAM cache or RAM SSD's which makes the problem a lot more meaningless. At 8-32x the flash capacity, several times more copies of the data can be stored in a cluster for the same price, which in the end will improve performance, even if individual devices are slower. If nothing else "fast" flash can be produced using a larger process, or people will just migrate to another SSD technology.
 
[citation][nom]rtoskevich[/nom]12 years from now? 12 years ago we were using zip disks that held 100 megs and were 100 times the size of an SD card.[/citation]

Sorry I thumbed you down, mis-click. Don't forget how often we still used floppy drives that only had 1.44MB or so. Micro SD cards aren't even as big as the "mini" IDE connector on my old 2.5" laptop hard drives, yet some micro SD cards have higher capacities than the hard drive itself. That was a decent hard drive 8 years ago too, 30GB drive. Now we have these tiny 32 and 64 GB cards. At least the old drive is still faster than most SD cards :)

Now if only new hard drives lasted as long as all of my older hard drives, then I'd pay the prices they want for them right now.

@jbo5112


Why does it need to be a RAM SSD to have worse latencies? System RAM latency has been improving ever since it came out and still is. For example, an old 400MHz DDR module with a CAS of 3 has a latency of 15 nanoseconds and a 1600MHz DDR3 module with a CAS of 9 has a latency of 11.25 nanoseconds and that's a decent improvement, so it looks like RAM latency is improving over time. Just because the timings are going up doesn't mean the latency isn't still going down anyway, the two are not identical. A timing is how many clock cycles it takes for a certain thing. However, clock cycles improve faster than timings go up, so latency improves. For example, DDR 400MHz CL3 to 1600MHz CL9 shows a 300% improvement in clock frequency and a 200% increase in timings, the clock frequency is high enough to offset the timings increase. Even if you use 400MHz CL2.5 DDR memory, the 1600MHz CL9 still shows lower latency. Furthermore, 1600MHz DDR3 can go a lot lower than CL9 if you buy more expensive modules.

Flash isn't so lucky. Flash increases latency, probably because it's clock frequency isn't increasing. It's not the exact same reasoning, but it's pretty similar. When you get a chip that has x amount of columns, and going to a higher density increases the number of data rows per column, then it takes more time to cycle through the rows completely than it did before because the increase in number offset the process shrink.

I'm not a Flash expert so I can't say for certain every reason for increased latency at smaller process nodes, but I'm pretty sure that this isn't helping it. Adding more chips instead of denser chips improves performance when the performance per chip isn't improving. It improves bandwidth because there are more devices that all have equal bandwidth. Fr example, four 50MB/s chips is better than two 50MB/s chips (theoretical numbers to prove the point). Having more chips can improve latency or at least average latency because of interleaving and similar techniques.

Denser chips can also have intra-chip interleaving between different parts of the chip, but that would require redesigning a chip every time you increase it;s capacity and you might need to redesign the controller/interface between the chip and controller every time as well.

If Flash is becoming a latency bottleneck, then how about SRAM caching it instead of DRAM or supplementing DRAM caches with SRAM? Sure, its more expensive than DRAM, but if you do a process shrink on the flash and DRAM and then add in the SRAM, the costs shouldn't be too different because the flash and DRAM is then cheaper than before. Have a prediction algorithm like how CPUs predict data they will need before they need it instead of the algorithms that carry the most often used data to the Flash and you should be good to go with small SRAM caches.

SRAM is a lot faster than even DRAM, which is faster than Flash in reading and a lot faster than Flash in writing, which is a lot faster than hard drives, it looks like a decent hierarchy storage system could be used here to fix most of the problems that Flash is facing, all without even considering the many alternatives to Flash that are being worked on.

However, more chips means more space and more power usage too. With electronics, there are so many variables that there are always trade-offs somewhere in everything, although that can apply to many aspects of life similarly well.
 
[citation][nom]divhon[/nom]truly yours, THAILAND HDD manufacturers[/citation]

Truly yours, your hard earned American tax dollars at work, through the pig and trough "research" system that predicted humanity's extinction was to come to pass two years ago after the runaway AGW global warming. Thankfor all the fun and keep the $$$ coming ! :)
 
[citation][nom]silicondoc_85[/nom]Truly yours, your hard earned American tax dollars at work, through the pig and trough "research" system that predicted humanity's extinction was to come to pass two years ago after the runaway AGW global warming. Thankfor all the fun and keep the $$$ coming ![/citation]

No, I wasn't kidding, not one single bit... thank you land of fruits and nuts, California U...

" at the 10th Usenix Conference on File and Storage Technologies here this week, Laura Grupp, a graduate student at the University of California, San Diego, said that as NAND flash densities..."
--
Now all of you who down voted me above, you have some shame to cover up. Do it again.
 
Writing an article about what a retarded student wrote. All that's going to happen, even if there are no technological breakthroughs, is to use a multi-controller type system such as with the Revodrive's from OCZ, which currently run about 2GB/s [Revodrive 3 X2 Max Iops 960gb].

This also reminds me of "omg...cpu's have reached a peak...you can't hit over 5ghz without burning up!" Well ok, dumbass. It's true that even 10 years ago you could get a 3ghz intel processor. And now you have 12 core processors and even more in commercial applications. Which is the same road SSD's will go down once they've reached their supposed linear peak. Not to mention that a single core of a new i7 ivy bridge processor, clocked at the same speed as the old pentium 3ghz chips, would blow it out of the water.

Absolutely stupid comments by, I'm sorry if this offends you, a women claiming she knows anything about technology.
 
[citation][nom]HyperMatrix[/nom]Writing an article about what a retarded student wrote. All that's going to happen, even if there are no technological breakthroughs, is to use a multi-controller type system such as with the Revodrive's from OCZ, which currently run about 2GB/s [Revodrive 3 X2 Max Iops 960gb].This also reminds me of "omg...cpu's have reached a peak...you can't hit over 5ghz without burning up!" Well ok, dumbass. It's true that even 10 years ago you could get a 3ghz intel processor. And now you have 12 core processors and even more in commercial applications. Which is the same road SSD's will go down once they've reached their supposed linear peak. Not to mention that a single core of a new i7 ivy bridge processor, clocked at the same speed as the old pentium 3ghz chips, would blow it out of the water.Absolutely stupid comments by, I'm sorry if this offends you, a women claiming she knows anything about technology.[/citation]

None of that matters if NAND can't keep scaling down to lower process nodes (we already know it can't). Netburst and Sandy Bridge are completely different architectures, NAND is a single architecture (with slightly variations of course) that hasn't been changing because to change it would need a completely new set of controllers too. Sooner or later, NAND will need to be replaced by a different technology just as Netburst was replaced by other technologies such as Core 2, Nehalem, Sandy Bridge, and soon enough, Ivy Bridge.

What should have been said instead of "SSDs are doomed" is "NAND based SSDs are doomed". SSD does not mean Flash (as you seem to know already anyway), but simply means a drive that has no moving parts. NAND will fall and b replaced some day.
 
Status
Not open for further replies.

TRENDING THREADS

Latest posts

Moderators online

Share this page

COMPANY
RESOURCES
FOLLOW
Top Bottom