"Colossal Magnetic Effect" Could Lead To Another Breakthrough In Storage Tech

Bryant writes "Scientists with the Carnegie Institution for Science have discovered what could bring yet another massive advance in memory and storage. The discovery, a magnetoresistence literally 'up to 1000 times more powerful' than the Giant Magnetoresistence Effect discovered roughly 20 years ago, which led to one of the major breakthroughs in memory, seems to be a result of high-pressure interactions between Manganites. Manganites aren't new to this game; MRAM uses Manganite layers to achieve the Magnetic Tunnel Effect needed to keep the state of memory stable. Applying significant amounts of pressure to known tech-useful materials isn't a new trick; you might recall the recent breakthrough with Europium superconductivity thanks to similar high-pressure antics."

Comments

And if we can predict anything...

[JoshuaZ]

This discovery seems to still be in the very preliminary stages. It is premature to conclude that this will lead to substantial improvements. Putting things under high pressure is difficult and keeping them under high pressure is really hard (although from my minimal physics understanding it looks like this could be used to assist in low pressure situations also).

One thing is certain. If this does lead to improvement in memory we'll have a few months of people asking whatever they could do with all that memory. And then a few years after they'll complain that it isn't enough.

Re:And if we can predict anything...

[Gibbs-Duhem]

This is not new, nor truly preliminary technology; I researched this back in 2004 and there was already a huge amount in the literature. It's just an incremental improvement and uses by and large existing thin film technologies pushed to their limit.

Most people didn't even notice the transition from regular magnetorestrictive heads to giant magnetoresistive heads, they were just incorporated naturally so that hard drive densities could further increase. This technology is the obvious and natural extension from giant magnetoresistive heads, and the increased signal to noise ratio will allow for denser drives with no doubt -- although we're already at the point where a "bit" is only made up of a few dozen magnetic domains. But in any case, this type of technology is a prerequisite for using more highly nanocrystalline magnetic materials with smaller domains...

Storage....

[Darkness404]

The problem isn't storage its speed. Really with 1TB of HD space there isn't anything you can't have a lot of. On the other hand I/O, especially magnetic I/O is the main bottleneck. Storage isn't a problem.

Re:Storage....

[A beautiful mind]

It's a combination of persistence, random I/O and storage actually.

SSDs are good at the first two, but still have catching up to do on the latter (and price...), but as soon as a reasonably priced 1TB version comes out, that'll be a great boon...

Re:Storage....

[castironpigeon]

The problem isn't storage its speed. Really with 1TB of HD space there isn't anything you can't have a lot of. On the other hand I/O, especially magnetic I/O is the main bottleneck. Storage isn't a problem.

Are you saying that 1TB of space should be enough for anyone?

Re:Storage....

[mea37]

If you're thinking home computers, maybe.

For a lot of businesses, 1TB isn't that much. We have systems with well over 1TB of data, to which over 5GB of new data are added every day, with an accelerating rate of new data coming in (as the systems model more fo the business, in more detail, etc.).

Historically these scales have only increased over time, and nothing is evident that would show that slowing down any time soon.

Now, do you want all that storage in one HDD? Probably not; there are pros and cons. But, there are absolutely applications where the desired amount of storage on a device exceeds what you could get today. It's not all about how many movies you can torrent.

Re:Storage....

[slyn]

It's a combination of persistence, random I/O and storage actually.

SSDs are good at the first two, but still have catching up to do on the latter (and price...), but as soon as a reasonably priced 1TB version comes out, that'll be a great boon...

Though I do agree that SSD's are definitely the next big thing when it comes to computer performance, there are a lot of things that need to happen before they become the definitive standard in storage. As you mention, the price/GB ratio needs to come down, but in addition to that:

- SATA 3 needs to come out. Though most SSD's don't exceed SATA 2 bus speeds, higher end SSD's like the OCZ Vertex or Intels X-25m hit 250MB/s sustained speeds. ONFi (open nand flash alliance? something like that) recently announced what is essentially the NAND 2.0 standard which doubles the speed of NAND modules, meaning next generation SSD's could easily hit sustained speeds of 500MB/s without any special tricks like internal raid. SSD's are already faster, but for better futureproof-ness and the ability to get the full potential out of SSD's, bus speeds need to increase quite a bit.

- TRIM needs to get at more OS's and SSD's support. SSD's write performance degrades with use due to a combination of the mechanics of NAND flash itself and common wear leveling algorithms. Essentially what happens is that when reading the flash blocks, all the SSD has to do is pass over and read the data. When writing though, if the block was previously written to the SSD has to erase the entire block clean and *then* write it. This is further exacerbated on MLC SSD's, where the individual transistors each store 2 bits, which on average doubles the write time with the benefit of double the space for the same price (instead of 0 or 1 like a SLC SSD, each one stores either 00, 01, 10, or 11). TRIM effectively eliminates a step from the write process on a previously used SSD by erasing blocks marked as free by the OS during an idle period, which means that write speeds degrade less over time.

- Manufacturing processes need to mature, as well as firmwares, wear-leveling algorithms, and filesystems. Unlike platter hard drives SSD's don't have decades of optimization and experience, which means higher than acceptable failure rates, extra consumer knowledge required to properly install and maintain, OS tweaks needed to fully exploit the current capabilities of SSD's, and certain technologies just not being available yet (a recent ext4 v btrfs SSD comparison on phoronix showed that btrfs was much much slower than ext4 despite the potential for btrfs to be better optimized for SSD's).

My personal belief is that by the time SSD's are halfway done with all of the above (including price/GB), they will overtake traditional HDD's in the market. The advantages of SSD's are already here and apparent, they are just expensive and a relatively young technology with a few growing pains. By the time the growing pains are half resolved SSD's will be much superior in just about every way possible, and then they will really really take off.

Re:Storage....

[metaforest]

With the current market focused on exploiting economies of scale to produce the current generation of high density FLASH in 65nm processes. So far this has been VERY lucrative for the companies that bet the farm on building dedicated 65nm fabs for FLASH.

IMO: MRAM has unfortunately come a little late to the party. In it's current role it is kind of the Beta Max of the solid-state storage device market.

Currently only Freescale produces MRAM in any quantity, and they currently only produce 4Mbit parts on a couple of 180nm fabs.

I don't think there is any question that MRAM has the potential to meet or exceed FLASH on the density/price curve. What might keep it out of the high-density market is one of it's great inherent strengths; it doesn't wear out. Who wants to sell storage modules that don't EOL themselves? Try and sell that to your pointy-haired boss' pointy-haired boss...
----
General Electric wasn't interested in mass manufactured light bulbs until Edison figured out how to get them to FAIL reliably.

For them that doesn't know: Edison's original design was a carbon impregnated cotton filament, in as near a total vacuum as could be produced in those days. It didn't suffer from filament migration, and other effects that cause modern bulbs to fail. If the original design had been refined as it was, General Electric could have saturated the market for light bulbs in a relatively short time, and been driven out of the market due to the Edison bulb having no predictable EOL.

Speed and latency matters

[TheLink]

I don't think I'd be complaining much about huge amounts of cheap storage.

However I'd complain about low bandwidth and high latency.

Imagine if you have 100TB drives but they only do sequential transfers at 200MB/sec and are still stuck at about 10milliseconds access time (7200rpm).

What that means: it'll take 6 days to transfer 100TB at 200MB/sec, and random transfer speeds will be about as crap as now (1-2MB/sec).

Re:Speed and latency matters

[donaldm]

Besides, who cares how long your site backup takes or how long it takes to fill up your DVR? That's where spinning disks are going.

Any business that want to remain in business cares significantly about backup times. Basically you want to backup as quickly as possible (ie. not during business hours) and if required recover just as quickly. Even with disk to disk backups (great and relatively cheap for home use) you are always going to have a latency problem. Unfortunately the more elaborate a backup and recovery strategy is the more expensive it becomes.

As for a DVR this is normally up to the household although it can be quite funny or stressful when you want backup up you favourite program and you have no more space on the disk. Mass panic to clean up normally happens. Anyone in the IT industry has seen this on a regular basis.

It must be remembered that backups are not about just recovering data to existing systems if required it is about recovering from disasters as well. It is quite scary that many companies have a half hearted approach to backup and recovery and many don't even go through the exercise of testing a disaster recovery scenario since they think it is going to be disruptive or is going to cost too much. Of course these companies are basically a disaster waiting to happen however it is very difficult for IT to explain to management that they need to test their disaster recovery processes when management can't understand that their own PC's need backing up (at least their user data) as well.

Not that new

[booch]

When I did a presentation on hard drives 3 years ago, I had already read some things saying that the Colossal Magnetorsestive Effect was the next step in read-write head technology. The Wikipedia page says the effect was discovered in 1993. This new discovery might make it more feasible, but hard drive technology developers already knew that CMR would be a part of the technology going forward.