PORTLAND, Ore. IBM Corp. is launching a joint research and development project with TDK Corp. to create high-density magnetic random access memories (MRAMs) in four years. The new multiyear program will aim for a 20-fold increase in the memory density of MRAMs by switching to a writing mechanism, called spin-momentum transfer, that draws less power and uses smaller bit cells.
"We have had a few research breakthroughs in tunnel junction MRAMs in the last few years, but none of our designs have made it into commercial chips yet," said Bill Gallagher, Senior Manager, Exploratory Nonvolatile Memories at IBM's T.J. Watson Research Center (Yorktown Heights, N.Y.). "We think this new program using the spin-momentum transfer has a better chance. We think we can lower power requirements, shrink the bit cell and get unlimited longevity using this new writing method called the spin-momentum transfer effect."
Spin-momentum uses the magnetization of the spinning electron to change the magnetization of the MRAM bit cell. Just by passing a spin-polarized write-current through the magnetic layer of an MRAM bit cell, the new approach can flip it from a one to a zero, and vice versa.
|Artist's rendering of the layers of IBM's MRAM prior to its collaboration with TDK.|
Magnetic tunnel junction MRAM bit cells are composed of a transistor and two magnetic layers: a fixed-orientation magnetic layer and a free magnetic layer separated by a tunnel barrier. Data is usually written by changing the magnetic orientation of the free layer--accomplished by running current through the two bit lines to generate a magnetic field over the right bit cell. Reading is accomplished, without using as much power, by sensing a change in resistance.
The only commercially available MRAMs today, such as those from Freescale, are based on the old magnetic-field data-writing, but the new method being explored by IBM and TDK uses less power and smaller bit cells. Because spin-momentum transfer MRAMs write data by flowing a stream of spin-polarized electrons through the free layer, they eliminate the need to run current through the bit lines. Reading is still accomplished by sensing a change in resistance.
IBM previously partnered with Infineon to craft magnetically writable MRAMs at Altis (Corbeil Essones, France) using the old writing method in an effort that expired without meeting its goal. IBM has since partnered with Infineon spinout Qimonda AG and Taiwan's Macronix International Co. Ltd. on phase-change RAM, or PRAM--hedging its bets that PRAM will be a viable alternative technology to MRAM. PRAM is also being promised by Intel Corp.
"Of course there is phase-change [PRAM] and other memories, but we see MRAM as ultimately a major player because it has such a good combination of properties," said Gallagher.
IBM has pioneered the magnetic tunnel junction (MTJ) since its inception, but TDK has also gained significant experience in magnetic tunnel junction design, especially its manufacturability, by virtue of its hard disk head designs, which use MTJs. By adapting its MTJ expertise to designing MRAMs, TDK also hedges its bets that hard disks will someday be displaced by nonvolatile such memories as MRAMs, PRAMs and ferroelectric random access memories (FRAMs). Luckily, the nanoscale magnetic tunnel junctions used in TDK hard disk write heads is the same technology used in magnetic tunnel junction MRAMs.
"TDK is expert is manufacturing exceedingly small magnetic tunnel junctions for hard disk heads, but magnetic tunnel junctions are also the key element of our MRAMS," said Gallagher. "Their magnetic tunnel junction R&D and their manufacturing reliability experience will be very important in helping us to meet our design goals."
The goals for the four-year project will be increasing the density of IBM's current 16-Mbit MRAM design, by shrinking it from 180 to 65 nanometers and by designing smaller bit cells that do not need the magnetic programming overhead, because writing will instead be done with spin-polarized electrons. Shrinking gives about a 9-fold increase in density, from 16 kbits to 144 kbits. The rest of the way to the goal of a 20-fold increase would come from shrinking the bit cell size--something that TDK is expected to be helpful in solving because they have done it for hard disk heads.
The research work will be conducted at IBM's Watson Research Center along with its Almaden Research Center (San Jose, Calif.), ASIC Design Center (Burlington, VT) and TDK's subsidiary Research and Development Center (Milpitas, Calif.).