Slashdot Mirror
New Technique Promises Much Faster Hard Drive Write Speeds
MrSeb writes "Hold onto your hats: Scientists at the University of York, England have completely rewritten the rules of magnetic storage (abstract; full paper paywalled). Instead of switching a magnetic region using a magnetic field (like a hard drive head), the researchers have managed to switch a ferrimagnetic nanoisland using a 60-femtosecond laser. Storing magnetic data using lasers is up to 1,000 times faster than writing to a conventional hard drive (we're talking about gigabytes or terabytes per second) — and the ferrimagnetic nanoislands that store the data are capable of storage densities that are some 15 times greater than existing hard drive platters. Unfortunately the York scientists only detailed writing data with lasers; there's no word on how to read it."
Who needs to read data back anyway?
A future-proof storage medium.
frickin hard drives with laser beams!
Considering how often I back stuff up, but how rarely I ever use those backups, I'll gladly take 1,000 times faster backups even if it means slower read speeds than we have now. Really, I'ld take that trade-off in a heartbeat.
If they can read it at least as fast as today's technologies, the power required to read/write data is roughly the same as today's drives and the manufacturing cost is also about the same, this is good news for everyone:
1. On the consumer side, cheaper drives per terabyte meaning cheaper home media servers
2. On the commercial side, a lot less energy required, i.e. no need for ultra-fast 15k RPM drives in servers, need up to 15 times fewer drives in server farms. This is BIG.
There is only one problem.
It's stored in the same way as a normal hard disk - in ferromagnetic domains on a platter. You can still read it back using the same techniques as current drives (i.e. put a coil over it and see which way the induced current flows), but you then have a drive that you can write to orders of magnitude than you can read from it. I can think of a few places where this might be useful. The most obvious is the underlying storage for something like ZFS. For reliability, you want to flush everything to the backing store as quickly as possible, and with copy-on-write and snapshotting you may never erase it, but most of your reads are satisfied from flash or DRAM caches. A drive using this technology would let you dump data there as quickly as you wanted and would let you read it back for data recovery if you needed to, while in normal operation you wouldn't care about the read speed because reading from the disk is comparatively rare. It would also be useful for a number of scientific applications. I did some work a few years ago with someone building a solar observatory. A single one of their cameras generated 10GB/s of data, and they had 8 cameras in a typical setup. They run these for the entire time that the sun is visible. A single drive that can handle a sustained write speed of 1GB/s would be very useful for them (although they'd fill up several per hour...).
For consumer devices, random read speed is still the most important factor, and mechanical drives suck at that.
I am TheRaven on Soylent News
"Unfortunately the York scientists only detailed writing data with lasers; there's no word on how to read it." A bit of a paradox don't you think? How did they know it was written without being able to read it?
I would think that you would still have to read the location of the cluster before writing to it. Sure you can flip magnetic particles N > S or S > N at bazillions per second speed but if you don't know what you're flipping that's not good.
Why? If they can write TB/s and store data at 15X of current capacity, and SSDs can't, why move to SSDs?
The read problem is easily resolved by having multiple read heads that can read independently.
Nope, that was a few years ago. Now they use TMR http://en.wikipedia.org/wiki/Tunnel_magnetoresistance#Applications .
http://compsoc.man.ac.uk/~shep/
This solves a major problem with mag recording. Readback head have always been way smaller than write head. You can read back with just a tiny permalloy head but to write you need large currents and loops of wire. So miniaturization has been limited by the write head size not the read head. This solves the write-head size problem but may have created a new read head problem. But that's very promising.
Some drink at the fountain of knowledge. Others just gargle.
This is the data equivalent of freezing Walt Disney and assuming that someday we'll figure out how to thaw and revive him. Write now, read someday.
Some drink at the fountain of knowledge. Others just gargle.
It's time to dust off the old concept of hard sectored discs ;) Realistically, of course, it's a bit more complex than that.
First of all, modern hard drives have a servo track that's used to maintain radial position of the head servo. Instead of each hard drive having a very accurate (and expensive) radial and axial head position sensor, you pay for it once, install it in the factory, use it to accurately guide a hard drive to write the servo track. Its cost is amortized over thousands of drives made. This is probably the reason for a covered up radial slot in many hard drive enclosures: I guess it's used for the sensor to couple with the head system while the drive writes the servo track. Or perhaps the servo platter is prewritten outside the disc? Someone familiar with how it's made please chip in!
The servo track can be also used to provide angular position feedback. A rough estimate of angular position of the spindle is available first from the Hall sensors in the spindle motor. A somewhat more accurate estimate can be had from back-EMF from the spindle motor windings. This still is methinks a couple orders of magnitude away from what's needed to pack sectors tightly on the drive -- thus the feedback can come from the servo track. Not having to read the data tracks helps with packing the sectors: there's no read-write switchover overhead (if it were significant -- perhaps it isn't nowadays). The servo head is always reading, and the data heads can be kept in write/erase standby. It'd be nondestructive, but read amplifiers are disconnected to prevent saturating them -- amplifier overload recovery is slow. Heck, if you want an amp that recovers from overloads quickly, you have to split it into more stages, and you need fast clamps between each stage. There are other similar approaches to this problem, too, and perhaps modern read amps are designed to deal with overloads gracefully -- I never tested a recent one. Stuff from a decade ago was painfully slow on overloads (tried to reuse a head amp from a hard drive for a non-drive-related project).
Alas, this ultra-fast-writing drive would unfortunately need very accurate position sensors -- both angular and radial. It's an engineering issue to make those affordable, as is the design of the optochip with femtosecond laser and its driver and serializer. The latter would probably take a couple serial lanes and multiplex them -- I presume it's not all that easy to push 10gbit/s data between external chips and the laser driver/laser combo. I think that to make it all practical you need an on-chip serializer, write precompensation, driver, and the diode. Perhaps the diode would be "tacked on" later to a substrate that has everything else. I only imagine that bond wire parasitics, even over a couple mm, become kinda important when the laser waveform has a 100GHz bandwidth...
A successful API design takes a mixture of software design and pedagogy.
Goatse Drive technology?
I swear to God...I swear to God! That is NOT how you treat your human!