'Magnetic memory' chip unveiled (BBC) {'Hard drive on a chip'}

[eppur_se_muova]

A microchip which can store information like a hard drive has been unveiled by US company Freescale.

The chip, called magnetoresistive random-access memory (Mram), maintains data by relying on magnetic properties rather than an electrical charge.

One analyst told the Associated Press news agency that the chip was the most significant development in computer memory for a decade.
***
The benefit of Mram chips is that they will hold information after power has been switched off.
***
A number of chip makers have been pursuing the technology for a decade or more, including IBM, but Freescale is the first company to offer a chip with practical usage for many of today's electronic devices.
***
more: http://news.bbc.co.uk/2/hi/technology/5164110.stm

Darn, someone beat me to it.:)

Don't think of this as a magnetically-based memory chip -- think of it as a hard drive with nanosecond access times, gigabit/sec read-write speeds, and zero latency.

on edit: Freescale Semiconductor (http://www.freescale.com/) is a spin-off of Motorola. They still make PowerPC chips and CodeWarrior IDE's, but don't seem to support MacOS anymore -- it's Windows, Solaris, and Linux now.

[wakeme2008]

1. most ppl do not know that

every bit of their main memory has to be read and re-written once a second or it forgets.

[mainegreen]

2. But do you want your computer memory to store data after power down?

:P
Seriously though, it will be nice to see NAND Flash die out. If what they're touting is true, these will make very nice portable storage indeed!

[eppur_se_muova]

3. This is a replacement for hard disk, not RAM.

You will still want volatile RAM to store the 'active' copy of your system software, apps, docs. etc. Then when it crashes, you can reboot from the nonvolatile MRAM in a second or two.

[mainegreen]

4. I wont hold my breath on that one.

I suspect it will be a while before MRAM is ready to replace the hard disk! It would be nice though, to get rid of that whirring noise!

[rman]

5. If it's faster than RAM, why not?

It will give a whole new meaning to "instant-on".
Also i'm sure that the data it holds can be erased if so required.

[Tesha]

6. No, it's RAM.

35 ns to read, 35 ns to write, random access, non-volatile memory.
So it's about as fast as any of the current DRAM technologies plus
non-volatile plus no power wasted for refresh.

Tesha

[eppur_se_muova]

8. It is RAM (truly random access) -- but functions as HD replacement.

That's MORE exciting than a new (volatile) RAM technology! It means disk access will not be the bottleneck on performance that it currently is (for big, data-intensive jobs, anyway).

It's non-volatile storage -- maybe we should start calling it 'hard RAM' or something like that, to get that point across.

[Tesha]

9. No, it *DOESN'T* function as a hard-drive replacement.

> It is RAM (truly random access) -- but functions as HD replacement.

No, it *DOESN'T* function as a hard-drive replacement; it's far
too small today (at 0.5MB, about 1/50,000 the size of a typical
cheap 250GB desktop drive these days).

BBC got the story wrong. Where it *IS* being proposed for use
is in the caching system of modern hard drives. At the moment,
most hard drives cache recently-read data in dynamic RAM so
if you read it again soon, it comes from the very fast RAM
rather than the very slow spinning magnetic platters. But
they can't cache writes because, if power failed, the data
that was allegedly already-written (but was really only in
RAM) would be lost forever. But replace the ordinary RAM with
this stuff, and the data would survive power failures and
could be written to the real disks later when power was restored.

High-performance disk systems do this today with battery-
backed-up CMOS RAM, but this MRAM could do it better and,
eventually, cheaper and in a much wider variety of drives
without the need for an unreliable battery.

Tesha

[eppur_se_muova]

10. Well, I'm anticipating a bit (as is BBC). Yes, it's too small now, but ...

the goal is eventual replacement of HD's altogether. This is what the researchers in this area have been aiming for all along. Now we have to wait for Moore's Law to kick in and bring the price/MB of these down to where they can compete with HDs (which continue to drop in price/MB themselves). Considering how astonishingly fast flash RAM has dropped since its introduction, Moore's Law redux may be even more impressive than the original. The first commercial device to be introduced is only the thin end of the wedge -- wait three years, then see what you think of the potential of MRAM.

Just a reminder: There was a time when a computer's entire OS and one or two applications could be stored on a 770K floppy, and the owners of the first personal computers often did not buy a HD until after using the computer for months, or even a couple of years. So while 1/2MB may not replace the HD on your current computer, it's not as ridiculous as it first sounds.

[Tesha]

11. Even Moore's Law will take a while to overcome a factor of 125,000.

Right now, that hard drive I spoke about (250 GB for about
$100) is about 125,000 times cheaper than the MRAM chip (at
0.5MB for about $25).

*ASSUMING HARD DRIVE MANUFACTURERS STOOD STILL*, it would
still take about 17 iterations of Moore's Law to allow
that MRAM chip to become competitive. And do you suppose
hard drive vednors are standing still?

Tesha

[Tesha]

13. I mis-spoke here. MRAM = 1/500,000 the size of a 250GB disk. (NT)

[eppur_se_muova]

16. And about 500,000 times faster, no? If you're paying for speed ...

it's already competitive.

Of course, what you're really paying for is a combination of speed and reliable nonvolatility which can't be provided by other means. If you set up a system to take maximum advantage of each of the available components, there are definite advantages -- and 'first adopters', especially performance-dependent institutions, will find it worthwhile to start mixing in some MRAM within the next couple of years.

Picture a 'hybrid' or (better) 'tiered' system: Sufficient MRAM to hold your core OS and scratch data to which you need fast access (say ~~1 GB, or even smaller, if you strip the OS down to its basics). In the event of failure of either the OS or process, you can recover much faster than from a HD. But for long-term storage (you know, all those files that you never throw out, even though you hardly ever use them) use a conventional HD. This is actually very similar to the idea of using HD for online storage, and tape (or other removable media) for off-line storage (as most large installations do), but much faster. I would expect this to be the type of system where MRAM first migrates to desktop CPU's, probably starting out in high-performance workstations or mainframes. I say that because that's exactly the route that solid-state disks originally took on the way to cheap flash RAM. The real competition for MRAM (at least initially) won't be with HDs -- it will be with other SSDs, like flash.

http://www.storagesearch.com/bitmicro-art3.html claims

(Editor's notes:- at the low end of the capacity range SSDs actually cost less than HDDs - see the Semico article which includes a timeline for this. At the high capacity end SSDs are often cheaper than the CPU server upgrades needed to deliver the same speedup benefits.)

The big qualifier: all this will matter to high-end users much more than to your typical business laptop user. For those of us who do scientific computing, the temptation is to run out and invest in the tech as soon as we can. For people who use their machines only for email and Web surfing, there's no reason to pay much attention for several more years, when the costs of mfgr will have dropped enough to make MRAM a common option.

/edit to change para order

[Tesha]

17. Depends on what you mean by "speed"

Access time? MRAM is lots faster. 35nS versus, say
35mS, so 1,000,000:1 faster.

Bandwidth? MRAM is probably "in the same ballpark".

Disk subsystems obviously run at some ratio between
their potential bandwidth and their access rates, and
it's highly dependent on the workload and the amount
of smarts in the controller.

Tesha

[Tesha]

7. Here's a detailed article on the MRAM chip...

http://www.eetimes.com/news/latest/showArticle.jhtml?articleID=190301113

Tesha

[eppur_se_muova]

12. Interesting article, thanks.

Freescale, Tehrani said, "has no intention of becoming a commodity MRAM supplier."

I had to smile when I read that. Sure, they're not planning on it. Oh, but they might license the technology to someone who will.

[Tesha]

14. Those are codewords.

What Freescale is saying there is that they don't intend to
compete in the hyper-competitive classical MOS DRAM market.


Of course, if they just happen to create a new market segment
with their MRAM offering, then...

Tesha

[eppur_se_muova]

15. That's how I read it. nt