IBM Is One Step Closer to High Speed, Low Power Racetrack Memory
January 24, 2011 Timothy Prickett Morgan
Tape storage, the oldest commercially viable magnetic medium, works by moving the magnetic tape that encodes the 1s and 0s of data over a stationary head. The genius of disk storage was to move both the disk and the head together to radically speed up the retrieval of data. Now, IBM researchers are one step closer to coming up with a completely new way of storing and reading data.
The new technology that the techies at IBM Research are working on is called racetrack memory, and no, it is not about picking the ponies and remembering to tell the truth.
IBM’s researchers have just recently published a paper in Science that details a new discovery that could possibly lead to the commercialization of this technology, allowing for a factor of 100 times better data density that is possible with disk technology while consuming lots less energy and being more affordable than flash memory.
IBM started the racetrack memory project at the Almaden Labs six years ago. With this approach to data storage and retrieval, magnetically encoded data is stored on a bent nanowire. The racetrack memory uses a physical phenomenon called spintronics, and it allows for magnetic regions on the nanowire to be pushed along the wire by electrical pulses at very high speeds without degrading the regions. A sensor for encoding the magnetic regions and another transmitter for creating them are at the bend in the wire, as the illustration shows above, and depending on which end of the wire you push the electrical pulses from, you can make the data (the magnetic regions) zip back and forth along the wire. It is, in essence, making a tape drive where the tape doesn’t move, the magnetic data does. (See an animated version of this graphic at this page.)
The researchers at Almaden have figured out how to precisely control the movement of these magnetic regions on the nanowire–called domains–and that is what the Science paper is all about. The researchers figured out that these domains take exactly as long to slow down as they do to speed up as they move along the wire, which means data access on racetrack memory can be precisely controlled. The initial tests not only show that racetrack memory could have 100 times the density of current storage, but that the domains can move down the tiny wires at hundreds of miles an hour without deforming or breaking down and can be stopped in precise locations, ” allowing massive amounts of stored information to be accessed in less than a billionth of a second,” as IBM put it in its announcement of the research breakthrough.
I love flash memory and have, in fact, just bought my first flash-based solid state drive as the main OS drive for my new Windows 7 workstation. (Soon to be a hybrid machine with a virtualization hypervisor and lots of different operating systems). Switching to flash from disk for the OS gives a huge response in the system, as I have been telling you it would do for servers. Flash-based SSDs have a random access time on the order of 100 nanoseconds, with disk drives ranging from between 5 to 10 milliseconds. DRAM main memory is on the order of between 10 and 70 nanoseconds, depending on the memory technology used and if it has SRAM caching. IBM says that racetrack memory, which is also non-volatile storage like flash and hard disks (meaning that it does not need to be powered on to retain its data, and distinct from main and cache memories), will in theory allow for data to be accessed in less than a billionth of a second as it passes over read heads. A fraction of a nanosecond. IBM’s initial tests show racetrack data can be moved in tens of nanoseconds over micrometer wire distances in early devices.
Imagine how this might speed up a PC, server, or handheld device, and radically cut its power consumption.