backtop


Print 30 comment(s) - last by Klober.. on Jan 25 at 2:31 PM


Researchers have invented a dual-mode device that can act as both volatile and nonvolatile memory.  (Source: NCSU via IEEE)

They actually have made a rough working prototype of the device.  (Source: NCSU via IEEE)
Volatile storage, meet nonvolatile storage

Scientists have created a new device that can offer both dynamic and long-term storage and could revolutionize the field of electronics.  Where current devices have separate memory and storage chips, future devices could rely on a single bank of homogenous chips.

Traditionally, (volatile) memory and storage are separate entities on modern computers.  The volatile memory involves chips with very fast access times that must be powered to retain information.  Often the term "memory" is used synonymously with the longer phrase volatile memory.  It's typically used to store details about the running programs.  By contrast storage (nonvolatile memory) offers slower access times, but can hold information even after a circuit is powered off.

If you've ever looked at a NAND flash storage cell (nonvolatile) and a DRAM cell (volatile), you'll realize that long-term and short-term storage don't necessarily have to be dramatically different in structure.

Perhaps that was the inspiration of North Carolina State University, which has invented a new device dubbed the dual floating gate field effect transistor cell (DFG-FET), which packs both devices into a single cell.  The DFG-FET consists of the two FET gates separated from each other by an insulating layer of silicon oxide.  Above the top FET rests a control gate, which can change the state of transistors by applying different voltages.

As a RAM cell, the two-transistor device has a read time of 0.31-2.18 ns (similar to SDRAM and faster than DRAM), a 50 ns write time for the charged state ('1'), and a 10 µs write time for the uncharged ('0') state.  Non-volatile writes range from 10 to 30 µs.

The device can tell what mode the current command is, based on the individual voltages provided to the gate transistors via the word line (WL) and select line (SL).

Dr. Paul Franzon, a professor of electrical and computer engineering at NC State and co-author of the paper on the work, states [press release], "We've invented a new device that may revolutionize computer memory."

Possible applications in the short term include unified memory for server farms.  Currently, server farms can't turn off specific servers to save power, as they would lose the information stored in their volatile memory.  As the new device operates faster, it could quickly cache the current volatile memory state in free non-volatile memory and then power off, quickly retrieving it when powered back on.  

Aside from server power savings, the paper shows off an FPGA circuit that could offer both space and power savings over current designs.  FPGAs are reprogrammable hardware chips (think a CPU which could be switched from an ARM architecture to a x86 architecture by flashing the chip), which are increasingly seeing use in a variety of devices.

And the researchers propose that the device could be used to create "instant on" computers and portable electronic devices, which didn't require lengthy boot times (which have even crept into the smartphone space).

The paper on the exciting new device is published [abstract] in the IEEE journal Computer.  The study was funded by a National Science Foundation grant.  Co-authors of the paper included former NC State Ph.D. student Daniel Schinke; former NC State master's student Mihir Shiveshwarkar; and Dr. Neil Di Spigna, a research assistant professor at NC State.



Comments     Threshold


This article is over a month old, voting and posting comments is disabled

Memristor?
By Gungel on 1/24/2011 11:17:30 AM , Rating: 2
Isn't that very similar to HP's Memristor? I think the memristor also offers much higher performance than this DFG-FET.




RE: Memristor?
By Helbore on 1/24/2011 11:55:52 AM , Rating: 2
Plus it has a cooler name.


RE: Memristor?
By vol7ron on 1/24/2011 10:26:12 PM , Rating: 2
What would you use it for? It seems like it'd be best for sleep mode or state saving for quick startups.


RE: Memristor?
By NicodemusMM on 1/25/2011 3:44:41 AM , Rating: 3
I may completely misunderstand this, but my first thought was an SSD with throughput faster than conventional DRAM (bus excluded). Also RAM with higher throughput, but without the loss of data on power down giving instant on capability as you stated. Using this as RAM would depend entirely on whether there is cell degradation as NAND currently suffers. Maybe also for embedded applications... either way it's a win for consumers.


RE: Memristor?
By dani31 on 1/24/2011 12:00:05 PM , Rating: 2
The Memristor varies resistance based on current.

This switches between volatile and non-volatile.

Which could mean instant power-off and resume, with no power needed as in sleep mode.


RE: Memristor?
By Drag0nFire on 1/24/2011 1:18:23 PM , Rating: 1
quote:
Which could mean instant power-off and resume, with no power needed as in sleep mode.


I would think this would be more useful in a laptop application. Correct me if I'm wrong, but keeping memory powered in a "sleeping" server should only cost the price few watts, far less than replacing all the RAM with "Solid State RAM".


RE: Memristor?
By SPOOFE on 1/24/2011 11:56:09 PM , Rating: 2
quote:
Correct me if I'm wrong, but keeping memory powered in a "sleeping" server should only cost the price few watts

I'm sure a lot of it depends on the needs of the server/farm. If you're talking a system that sees very little idle time or opportunity to "sleep", you're certainly right. But if you have a system that sees regular periods of inactivity (say, the office is closed for the weekend and there's next to no need to run the servers but it's a pain in the ass to start up cold on Monday morning) it could potentially pay for itself. Depending on the price of the product, when/if it hits the market.


RE: Memristor?
By bug77 on 1/25/2011 8:34:54 AM , Rating: 3
Think beyond power savings. Think of a HDD that doubles as RAM. Traditionally you get 2-3 orders of magnitude less RAM than you get for storage. With this thing, both would be on par. And instead of a memory controller (on-chip) and a SATA controller (southbridge), you get to fuse them together.

And don't think servers only. These sort of enhancements can allow packing a lot more goodies into a smartphone, too.

Of course, none of this will happen in the next 10 years. And I think that's optimistic.


RE: Memristor?
By Targon on 1/25/2011 9:26:53 AM , Rating: 3
You are looking at a best case sort of situation here. When you use the hard drive for memory and the system crashes, what then? You really do want to keep longterm storage and "RAM" apart, but for virtual memory, "sleep", and "hibernate", then this could clearly be a good thing.

Years ago, back before Vista was going to be released, Microsoft was pushing for having a certain amount of NV memory on hard drives to store things like the operating system that does not get updated all that often. This would allow for much faster startup times since it would be like a SSD combined with a normal hard drive. This new technology sounds like it would be perfect for that purpose as well.


RE: Memristor?
By Micronite on 1/24/2011 1:50:13 PM , Rating: 2
Unfortunately, the Memristor has very slow access times (relatively) which don't make it a very good replacement for DRAM.

In the article, it also mentions that this technology's write times are like 10us, that puts it an order of magnitude slower than DRAM write times.

I keep waiting for one of these memory things to actually pan out and be just as fast as DRAM, but also non-volatile. But nothing so far except perhaps the nanotube memory: http://www.nantero.com/nram.html


RE: Memristor?
By Gungel on 1/24/2011 9:49:34 PM , Rating: 2
The memristor is not that slow. The latest prototypes now also called RRAM or ReRam is showing <0.3 ns switching time and currents less than 30 microamps.


RE: Memristor?
By Klober on 1/25/2011 2:31:40 PM , Rating: 2
You may also want to check out IBM's racetrack memory. Racetrack memory looks to be approximately the performance of current SDRAM with the density of mechanical HDDs - all while being non-volatile and lacking the wear issue of flash memory.

http://www.almaden.ibm.com/spinaps/research/sd/?ra...


How is this revolutionary?
By DanNeely on 1/24/2011 12:02:21 PM , Rating: 2
quote:
Possible applications in the short term include unified memory for server farms. Currently, server farms can't turn off specific servers to save power, as they would lose the information stored in their volatile memory. As the new device operates faster, it could quickly cache the current volatile memory state in free non-volatile memory and then power off, quickly retrieving it when powered back on.


Isn't this possible now by hibernating to an SSD?




RE: How is this revolutionary?
By monitorjbl on 1/24/2011 12:41:47 PM , Rating: 3
You can hibernate to any hard drive, but that isn't the point. What's impressive about this new tech is that you don't need to do anything to preserve the data, its persistent in the cell regardless of the power state.

The entire reason we have RAM now is because hard drives, SSD and fixed disk alike, are slow. Fixed disks have access times on the order of a few milliseconds (1,000,000ns) and SSDs are on the order of ~1 microsecond (1,000ns) or so (and can only weather 5-10k read/write cycles). RAM latency is around 7ns and is very durable, so we use it to cache data for the processor.

Basically if these get read/write latencies down to 10ns and make them as durable as DRAM, we wouldn't need an extra cache for the processor. The CPU could just access the hard drive directly, eliminating the need for RAM entirely in future operating systems. Granted, it's all a long way off and entirely uncertain, but this is still really, really cool.


RE: How is this revolutionary?
By name99 on 1/24/2011 10:55:05 PM , Rating: 1
quote:
You can hibernate to any hard drive, but that isn't the point. What's impressive about this new tech is that you don't need to do anything to preserve the data, its persistent in the cell regardless of the power state.


On the other hand, this looks suspiciously like it's solving a non-problem.
What exactly is so bad about having a bank of RAM chips and a separate bank of flash chips? Especially if the compromises necessary to manufacture this thing means that it is substantially slower than RAM, and substantially less dense than flash?

Sure, if we could make a chip that was better in every way than existing RAM and flash that would be great. It would also be great if we could make a chip that were better in every way than an existing CPU or an existing GPU. Looking at something and saying "it would be nice if it were twice as fast, and used half the power" does not make one a visionary.


RE: How is this revolutionary?
By SPOOFE on 1/25/2011 12:04:58 AM , Rating: 2
quote:
What exactly is so bad about having a bank of RAM chips and a separate bank of flash chips

More points of failure. More cost to the device. Potential of filling up your RAM and really slowing down your system as it writes to disk.

quote:
Especially if the compromises necessary to manufacture this thing means that it is substantially slower than RAM, and substantially less dense than flash?

"If" is a funny thing that way.

quote:
Looking at something and saying "it would be nice if it were twice as fast, and used half the power" does not make one a visionary.

Yeah, it's like people only get excited when people actually go out and make their idea a reality, kinda like the fellas mentioned in the article up there. Oh, didn't read the article? Huh.


RE: How is this revolutionary?
By Strunf on 1/25/2011 8:15:32 AM , Rating: 2
quote:
More points of failure. More cost to the device. Potential of filling up your RAM and really slowing down your system as it writes to disk.

Not really if RAM and flash are both more efficient (since more specialized) than this new type of memory, the result may not be better, also since both RAM and flash are less complex they scale much better.

The reading speed when on RAM like memory is faster but it doesn't to seem to be the case when writing.


RE: How is this revolutionary?
By monitorjbl on 1/25/2011 1:32:30 PM , Rating: 2
Hibernating is slow and inefficient. You have to essentially serialize the data in volatile memory onto the hard drive before its safe to cut power to memory. Also, this isn't just a "twice as fast, half the power" upgrade to existing tech; if this were to take off, operating system design would need to change drastically.


nice
By lonechicken on 1/24/2011 11:27:06 AM , Rating: 5
This could be nice to have in the electronics of my personal jet pack or hovercar.




RE: nice
By EricMartello on 1/24/2011 12:07:38 PM , Rating: 2
Better call them up on your Video Phone and order one!


RE: nice
By gfxBill on 1/24/2011 2:46:53 PM , Rating: 2
I laughed at both comments but to be fair, most of us carry a video phone in our pockets these days ;)


Yawn
By Jammrock on 1/24/2011 2:12:25 PM , Rating: 2
I'll add it to the list of other "game changers" that will probably never be.

MRAM
PRAM
CBRAM
SONOS
RRAM
Racetrack memory
NRAM
Millipede
Memsistor
DFG-FET <-




RE: Yawn
By Gungel on 1/24/2011 9:43:19 PM , Rating: 2
The memristor is coming in a couple of years. Actually the prodcut is now called ReRam from HP and Hynix.


RE: Yawn
By SPOOFE on 1/25/2011 12:07:15 AM , Rating: 2
There will always be more failures than successes. Babe Ruth had a great home run record, but he also had a crapload of strikeouts.


"could"
By kattanna on 1/24/2011 11:13:13 AM , Rating: 2
could.. is the key word here.

while i do still hold out hope that soon we will have such a tech in mainstream use, pardon me for being skeptic as i have heard this one before many many times.





RE: "could"
By zodiacfml on 1/24/2011 9:03:03 PM , Rating: 2
could..yes it could, the question is where.

i'm sure this will never get into desktop or any system that is any faster, as the feature is not so valuable versus performance and cost.

i see this, being very nice on devices such as phones, tablets and notebooks as it will integrate better, lower power consumption and will appear as instant-on on device from a no power state.


Jamrock, my man...
By mosu on 1/24/2011 5:22:34 PM , Rating: 2
I sincerely hope you're wrong. I wonder if they tried GaAs.




RE: Jamrock, my man...
By melgross on 1/24/2011 10:59:08 PM , Rating: 2
Only if they ate beans first.


By Tanclearas on 1/24/2011 1:39:13 PM , Rating: 2
So the funding was provided by an NSF grant? :)




Missed pun opportunity
By zozzlhandler on 1/24/2011 4:37:35 PM , Rating: 2
dram-atically different...




"I f***ing cannot play Halo 2 multiplayer. I cannot do it." -- Bungie Technical Lead Chris Butcher














botimage
Copyright 2014 DailyTech LLC. - RSS Feed | Advertise | About Us | Ethics | FAQ | Terms, Conditions & Privacy Information | Kristopher Kubicki