Original URL: http://www.theregister.co.uk/2010/12/24/ibm_racetrack_memory/

IBM races bits along a nanowire track

Precisely-placed bit pitstops

By Chris Mellor

Posted in Storage, 24th December 2010 12:51 GMT

IBM scientists have found out how to move bits precisely at an atomic scale along a nanowire, bringing racetrack memory closer to reality. It's glamming up this project by saying a portable racetrack memory box could store a year's worth of movies.

Racetrack memory involves the setting of the magnetisation direction of magnetic domain walls in a magnetic material, a permalloy nanowire, by the application of a spin-polarised current. The walls are the boundaries between the magnetic domains in the material. Each domain wall has a north pole or head and south pole or tail. IBM says successive domain walls alternate between head-to-head and tail-to-tail configurations with the spacing between them controlled by so-called pinning sites fabricated along the nanowire; pitstops for the bits as it were.

Racetrack memory

U-shaped nanowire racetrack memory.

Each permalloy nanowire contains several domains and and is sited by several but fewer read/write heads or sensing devices. As the current is applied it causes a spin transfer torque to impinge on the domain wall causing it to move. The domains move along the nanowire analogously to racing cars around a race track, although the nanowires are not closed loops but U-shaped or straight. What the IBM Almaden-based boffins, led by Dr Stuart Parkin, have done is to discover how to move the domain walls precisely.

Dr Stuart Parkin

IBM's Dr. Stuart Parkin

They didn't know if the domain walls were mass-less, in which case they would accelerate instantly when a pulsed current was applied and instantly stop when it was pulsed off, or whether they had mass. In that case an electrical pulse would cause acceleration at some rate over time and deceleration at some rate when it was switched off.

Parkin's team found that domain walls had mass and that the acceleration and deceleration could be controlled by timing an electrical pulse such that the domain's placement on the nanowire was precise down to the atomic level. What does that give us?

It means that a nanowire containing several domains (bits) could have them over from known place - a "bitstop" - to known place at nanoscale precision and so enable what IBM calls magnetic shift registers. The cycle time for the writing and shifting of the domain walls is a few tens of nanoseconds. Each sensing device could read 10 to 100 bits, implying that racetrack memory device could have very, very much faster access to data than a hard disk drive, but not as fast as DRAM. (There is more general information on racetrack memory here.)

The prospect IBM is raising is that its racetrack memory could provide a solid state, non-volatile storage tier between DRAM and flash, being denser than flash, bit-addressable, and not suffering the lighted endurance that's endemic to flash. It is not known how much power such racetrack memory chips would draw and what their performance would be. This would vary with nanowire length and domain number and with the number of sensors.

What's the capacity of a portable racetrack memory box? IBM saying it's a year's worth of movies isn't very helpful. It points us to the Answers.com website to find out how many movies that is and the pesky site doesn't provide a clear answer. At an approximation we can say 4,000 movies are produced each year. How much space would they take up on a storage device? It depends upon the definition and bit rate and length and … enough already. Let's just say that they'd take up 4TB if each movie was a gigabyte in size and 10TB if the average movie size was 2.5GB. So IBM is saying it can foresee a 4TB - 10TB portable storage device.

IBM is not alone in working on spintronics-type memory technology. Grandis is active in the STT-RAM (Spin-Transfer Torque RAM) area and Hynix has licensed its IP. We understand Fujitsu, Hitachi and Renasas are also researching the topic. We don't imagine any working devices will appear before 2016 or later.

There are competing technologies being researched as well, such as Phase-Change Memory and HP's Memristor. These might appear before IBM's racetrack. In fact, it's all a bit of a race... ®