Slashdot Mirror
Motorola Mocks-up MRAM
zakath writes "EETimes.com has an article on Motorola's successful presentation of 256-kilobit MRAM at ISSCC this week (Instant-on PCs anyone?). While they're still far from commercial production (2004 is their target) its nice to see some progress being made. Please tell me RAMBUS has no patents for this tech..."
Because people like to express their opinions, and if their opinions are longer than 1 sentence long, everyone starts hollering "SHUT UP JONKATZ PLZTHX!!!!!1". What would be ironic would be for me to add a snide remark about /. readers and the attention span of the average gnat. But I won't.
--
TO BUY A NEW CAR WOULD MAKE YOU SEXUALLY ATTRACTIVE.
I see a lot of posters here saying that "If they double their capacity every x months, that's only 8MB in 4 years" or something...
When working on a new technology, having a 256kbit chip is just as useful as a 256mbit chip, when all you're doing is proving the technology out. They have no reason to be trying to produce massive dies yet, because all that will do is reduce their yields... and when they're making one-off runs of these to test them, that's a killer.
I'm certain that by the time they go into production, the capacity will be impressive.
However, I don't think one would need to save the entire state of system memory for this to be useful. Aside from embedded systems, consider the possibility of an OS that uses a proper swap system and just makes sure the entire physical memory is written to swap when the system suspends. If the MRAM is sufficient to keep the core kernel state and information about the swap pages, your restart would just need to flag all the physical memory as free pages, with everything currently swapped out. Sure it wouldn't be "instant on", but it would be far closer than I've seen to date.
On the flip side, you could probably do something similar with Linux by putting in a boot patch that checks the swap space for a signature indicating it should reload the memory from the swap space instead of going through a full boot process. You'd still have the usual issues of needing to reset hardware, but it would probably be faster than a full boot.
I do not fail; I succeed at finding out what does not work.
What type of actual speed and latency can we see from these things? I mean, in three or four years, will they be able to compete with .5 ns access times and 100's of gigabytes of multiple-pumped bandwidth?
For memory, speed has always been the number one concern. The very definition of memory is volatile, and any non-volatile storage devices created recently or as proof of concepts still haven't been fast enough, otherwise we wouldn't need memory at all. A good memory system's job isn't permanent storage -- it's getting data to the system as fast as possible.
Being able to have an "instant on" PC would be great, but creating non-volatile memory isn't the way. Convergence amoung permanent storage and memory is also great, but the very definition of semiconductor devices in any method demand fast switching, volatile devices to deliver the speed neccessary. Static or programmable elements have physical constraints which don't follow the progession of increased performance over x months like some people here are counting on.
Ultimatly, you can have your "instant on" PC, as long as you don't mind slow memory. Volatile methods will always (for the most part) be faster.
RAMBUS most likely does not have ANY valid patents. As I remember the gist of lawsuits against them by Siemens and others, RAMBUS was one of 12 partners in a tech pool which took their notes and unilaterally filed patents. I have every hope that they will be found to be among the world's worst weasels and the poster boy for the dotcom stock.
if this is supposed to be a new economy, how come they still want my old fashioned money?
Solid state, eh? I can FINALLY get rid of that vacuum tube ram... man my power bill will go so far down.
-
I've had enough abrasive sigs. Kittens are cute and fuzzy.
Fact is that when you make a hole, and then secure it by hiding it, somebody will stumble across it. Once that happens, you're out of luck because it becomes hard as hell to fix it because you can't just give users something easy to patch the system with.
I seriously suspect that you are shooting for low karma or work for M$. Either way, you qualify as a troll. One down-mod coming up!
The problem with capped Karma is it only goes down...
SIG: HUP
My PC already has instant on (or "OnNow"), all ATX power supplies produce an ancillary current even when the PC is off (soft-off). I have a Gigabyte motherboard, they use this current to keep the DIMM banks powered when the box is off, and therefore the data is retained (this is different from 'standby' btw, the machine is 'off'). When I power back on, it takes about 2-3 seconds then I'm right back into the OS. (This is all in the ACPI spec)
Instant on PC's is not the purpose of MRAM, basically because it's too costly to use as system memory and unnecessary especially when ATX can just keep cheap SDRAM powered. It's great for handheld computers though and devices that have low power constraints and need fast data access, writing to flash memory is relatively low (compared to SDRAM) and it requires a decent amount of power, it's also expensive to produce, MRAM hopes to solve these problems.
You're right, it won't really find its way into PC's, not for a very long-time anyway, the market focus at the moment is handheld/portable devices that have stringent power requirements.
There's little point using it in PC's because the access times are slower than SDRAM, it's expensive to produce (especially so when you're talking about 256meg +), and it's already possible to have instant-on PC's already using Suspend-to-RAM with cheap SDRAM and an ATX power supply.
MRAM is a good technology and has a good niche, it's superior to flash memory (which it's destined to replace), however people seem to think it's a general purpose memory that will replace all existing standards, this is not quite the case.
If you just suspend the PC when you turn it off, then it never gets a reboot. This would require an OS that can sustain long uptimes.
Since by the time this technology is released, Linux and friends won't be among the OSes that are legal to use and officially sanctioned, this technology make PC's worse, increasing the number of crashes, due to the lack of regular reboots that naturally occur when the user turns off the PC.
I'll see your senator, and I'll raise you two judges.
I guess I should have explicitly excluded systems, like most servers, where availability is important.
But for most desktops, availability is usually quite secondary to data protection, so UPSes might be obviated.
-- Mike Greaves
http://www.eet.com/story/OEG20010110S1009
They also said they saw no reason why it could have essentially the same interface as conventional memory. Motherboard architectures would be basically the same.
A beginners' guide to Portland, OR?
IIRC, IBM holds the patent on MRAM, or at least they came up with the stuff. Then again, IBM and Motorola have been known to participate in the past. Still, as one of the last companies to do pure research, IBM is a safe bet.
I wonder who holds the patent on bubble core?
--
ALL YOUR KARMA ARE BELONG TO US
"You're right," Fisheye says. "I should have set it on 'whip' or 'chop.'"
No, that's wrong. The chip is organized *internally* in a 16-kbit-by-16 array. The reason for that is you usually don't want to read one bit at a time, but at least something like 4, 8 or 16. Implementing 32-bit memory using 1-bit chips would take 32 chips and a hefty amount of control logic in the memory controller, unless you had serialized access, which in turn would be very slow. So putting things internally in an array is a standard practise for almost any kind of memory technology.
I agree with your comment about saving the core kernel state, but honestly, you wouldn't notice any much difference between storing it on a hard disk or to MRAM. If the state would be like 4 MB, seek and transfer combined would probably be less than 0.5 seconds.
I think the most interesting use for this chip at this point (if it was available) would be as a transaction log buffer for databases, journaling filesystems and such.
--
You need 32 of these in order to get just 1 megabyte (256kb x 32 = 8192 kb = 1024 kB). For 8 MB, you need 256 (8192 * 8 / 256kb) of these.
--
320MB isn't that big anymore, in a world where 256MB costs about $100.
(And I use Win2k. It's not too bad. Better than Win98 for sure).
A beginners' guide to Portland, OR?
By the way both Linux and Windoze 2000 can suspend and restart, but without crash-recovery and there are some issues with X.(I have not tried it myself though)
"We mustn't be caught by surprise by our own advancing technology" -- Aldous Huxley
8 megs of non-volatile RAM?
That's plenty for a kernal.
Take a look at EROS (www.eros-os.org)
I could see hitting the power key, and immediately getting a system that was up (i.e. the thread scheduler running, and processes getting their time slices), but in effect had everything paged out.
What I'm curious about, is how much power this can save and what the memory cycle times look like.
-jcr
The only title of honor that a tyrant can grant is "Enemy of the State."
goat
If you're interested you can read about it here...
IBM Research News
Ruger
256kb = 32kB
4Mb = 512kB
Buying a Dell computer is equivalent to dropping the soap in a prison shower.
Routers, print servers, calculators, game consoles, cars, traffic signal equipment, space shuttle, ISS, cellular phones, and I'm sure there are other uses.
Withdrawal before climax is very ineffective and those who try this are usually called "parents."
One word: EPROMs.
I'm sure that problem could be dealt with in software. Not, however, without creating new problems. Which means there will be new revisions which will create new problems, and so on. *Sigh* Such are the tradeoffs that come when everything is centered around money.
Withdrawal before climax is very ineffective and those who try this are usually called "parents."
Uhmm, what about having Windows on these chips... I'm not sure I trust MS having Windows install itself to one of these... you'd never get it off, so you'd have to make sure the chips were removable/replacable or whatever so you'd be free to use anything else
well that's what they would do! they would cram 64 of these onto a momory stick and inlcude them WITH regular DRAM sticks. this way the OS could use the MRAM to boot and for the OS kernal while the rest of the computer used the DRAM
It looks like someone else has a sense of humor too:
Your right to not believe: Americans United for Separation of Church and
I think the point is that existing RAM is already solid-state.
Your right to not believe: Americans United for Separation of Church and
OK, how'd you do that? That seems much tougher than getting FP.
Your right to not believe: Americans United for Separation of Church and
How much juice does it take to keep your standar PCs memory refreshed? If you turn down the refresh rate to absolute minimum and perhaps stagger the refreshes to keep current constant, how long could an embedded battery keep say 128MB alive?
Ihave no idea... but someone has to have an [informed] opinion.
And just what wonderful position was open at the illustrious RAMBUS? Why, they want to hire Patent Attorneys!!!
Anyone else find this funny? [yet sad?] I actually laughed out loud as I passed their booth when I saw that. :)
Ender
Nothing to see here
There was alot of talk about this sort of stuff 2 years ago, but I haven't heard anyone mention it since. I'm really glad to see research continuing. Wonder what solid state memory would do for 3d cards. 4-Mbit MRAM by 2003... well it sure as hell couldn't hurt. Just in time for Doom 3.
And I thought I had a handle on memory technologies... The article speculates that MRAMs may eventually replace DRAMs, etc for main memory, but I wonder how feasible this is. It sounds like the manufacturing control has to be a lot tighter for these than for standard memories, so given equally stringent process won't DRAM yields be better? What other memory technologies are being developed right now?
Derek
Don't Panic...
256 kbits is 64 kilobytes. Even their 4Mbit chip in 2003 will be still only 500 kbytes.
Eight of those suckers could provide "instant-on" capabilities to my old 486.
RAMBUS has no patents for this tech.
Actualy I have no clue, but you asked for someone to tell it.
"You saved 1968." - Ms. Valerie Pringle to the crew of Apollo 8
This mean that a cold reboot wont fix memory leaks anymore ?
Even presuming that they can double the density every 9 months, they will be making 512kB chips in 2004, their estimated release date. Figure 16 chips per stick and you have 8MB. No where near enough for instant-on PCs (I mean full on OS's with GUIs, not LoaF or PBSD).
Mark Duell
90th POST!
there will be much larger capacities available once they get these to the fabrication facilities. However, the articles does not mention cost, so how exactly are we supposed to know if these aren't going to cost $1.50/MB like flash does? And how about making these chips in modules that can be relocated easily? For example, say your MRAM in your old computer will work in your new one? There's got to be a way to have them packaged as little clip-on chips that can be clipped onto a board like wireless PC cards can be clipped into a PCI board... It would save costs big-time. I have plenty of old 72-bit RAM lying around that could be used in newer computers...
I wish I had some mod points right now; you deserve +1, Funny.
God invented whiskey so the Irish would not rule the world.
er... ah... um... nope... morons and idiots and sheep and yes, lemmings all use some variant of windoze OH ESS produktz... it is a widely-known and generally accepted fact. you COULD look it up.
Just in case you're not a troll...
The fact that it's solid state implies it's not as likely to wear out as a hard disk. In fact, they're projecting a ten year lifespan for these things.
As your vacuum tube example shows, the absense of moving parts is not the only important attribute. Capacity and speed are also valuable attributes in memory. MRAM would also have those going for it as well.
of course, it's MOTO... makers of the oh-so-tasty CISC killers... heh heh
Yes, I agree that Windows would be much better-suited to MRAM than any of the free OSes -- instant-on technology just isn't that interesting when you never have to reboot! :-)
Why get excited over tech this new and not that interesting... It will be replaced by something better before it is mature.
Republicans are Nazis. LetsRiot!
Windows is the single largest sitting duck OS out there... its weaknesses as regards vulnerability to virii attacks are legion and legendary... if you want security through obscurity, use MacOS... though of course MacOS X is a UNIX variant and... but both are eminently stable, OS X being a great deal more stable than co-called classic MacOS and EXPONENTIALLY more stable than any windoze variant. no comparison there, any windoze variant needs a reboot far far far far far (etc) more often than does MacOS X.
typical... another misinformed PPC-basher. Moto never tried to get out of the PPC biz... the same RISC PPC chips you find in the best quality desktops (Apple's) are also used in embedded systems (where Moto makes most of their money on PPC chips), in no small measure due to their low heat generation. Further, 450MHz is the lowend chip now... they've been bumped to 733MHz, pal, and they kill a 1+GHz (you dope: "1000+ GHz"? sorry, no such beast) in many applications, viz the famous Photoshop comparison in which the Mac wins hands down, still. for more FUD from dopes like this guy, read anoncow posts in /. more often!
nazis... they are all that and so much more, coming soon to bust a door down in YOUR neighborhood! off-topic? so what. flamebait? who cares. have a nice day.
But traditionally, non-volatile storage has not been solid state.
there will be much larger capacities available once they get these to the fabrication facilities. However, the articles does not mention cost, so how exactly are we supposed to know if these aren't going to cost $1.50/MB like flash does? And how about making these chips in modules that can be relocated easily? For example, say your MRAM in your old computer will work in your new one? There's got to be a way to have them packaged as little clip-on chips that can be clipped onto a board like wireless PC cards can be clipped into a PCI board... It would save costs big-time. I have plenty of old 72-bit RAM lying around that could be used in newer computers...
But remember, one of Motorola's main markets is embedded systems - they're looking at memory for mobile phones etc initially, so the 2004 date is perhaps not so unrealistic.
Now, if you use a decent OS, like Windows, the kernel can have suitable protection mechanisms, such as security through obscurity. In Linux, a virus knows exactly where to put itself, but in Windows, the kernel is like a maze of twisty little functions, all alike. As you can see, if you're gonna use instant-on technology, you should use a suitable OS.
Also, don't think about MRAM replacing your RAM, think about it replacing your hard drive.
return 0;
showing your monumental ignorance... my, how pointed your head is! Moto and PPC are not going away, get used to THAT. PERHAPS, you blithering idiot, you have noticed that the whole damn industry is in a donwturn? no? i thought not. Oh, and IBM happens to make PPC chips as well... they happen to have the largest capacity chip foundries in the world, so much manufacturing capacity in fact that they can and do make chips for other 'manufacturers'.
Instant-on PCs anyone
You guys turn your PCs off?
If MRAM replaces DRAM (not just flash), then UPSes will no longer be necessary. Since a UPS is currently advisable for any system doing serious work, this has big cost implications.
On many new systems with a UPS, the UPS costs more than the DRAM in the box. So it follows that MRAM could be more than twice as expensive as DRAM, and still displace it for many serious applications.
-- Mike Greaves
Um. Why is this news? What possible use could a 256 meg MRAM module have? I can see this being news once it reaches 4 megs in 2003. As it is, 256 KB would only really be useful in something like a BIOS or (very) small data quantities. Even if you had 64 of these, you'd still only have 8 megs of solid-state storage. That would be enough to store a few documents, but you could save those to a hard drive instead. The only applications where the speed of solid-state storage aids them are too large to fit in these modules, unless you were to use more of these than would be cost-effective. It's a great advancement, but it's not newsworty. Yet.