IBM and TDK on Monday launched a research and development program to build their own high-capacity MRAM chip, which could someday replace less reliable technology used in disk storage systems and other areas of computing.

The joint R&D project looks to raise the storage capacity of magnetoresistive random access memory from its current limits in megabits to gigabits. The IBM-TDK collaboration places the effort against MRAM development projects at Toshiba, Freescale Semiconductor, and others. Freescale released the first commercialized MRAM chip last July, but at $25 for 4 Mbits, experts did not consider the chip much of a bargain compared with alternatives.

The major advantages of MRAM over flash memory, the most commonly used nonvolatile memory, is MRAM's ability to read and write data quicker, and its seemingly limitless endurance. MRAM's ability to store information doesn't deteriorate over time.

Where flash uses transistors in storing data, MRAM combines magnetic technology with standard silicon-based microelectronics. In place of transistors, MRAM uses so-called magnetic tunnel junctions that involve the switching of magnetic polarizations in storing bits. Because flash involves movement among electrons or atoms, the transistors wear out in time. MRAM has no such movement, so there is no deterioration. Because both technologies are nonvolatile, MRAM and flash hold data even when the power is off.

MRAM could have many applications. For example, it could be used in printers and computers to log data so the machines can be restored to a previous configuration if they're unplugged or lose power. IBM, however, is more interested in developing the technology for use in the data center or in equipment found in telecommunications, manufacturing, and other industries, Bill Gallagher, senior manager of exploratory nonvolatile memory, told InformationWeek.

MRAM, for example, could work well as buffer memory in a disk storage system. Information flowing to a data center is typically stored temporarily before it's written to a disk. The reason is to avoid having the person on the office computer wait while the data waits its turn to get stored.

Currently, buffer memory is in DRAM or SRAM, typical transistor memory that requires a battery to make sure there's always electricity available, so information won't be lost. "People don't like the battery approach because they have to worry about reliability," Gallagher said. If the battery runs down before IT staff has a chance to change it, then the system's buffer memory goes down.

IBM and TDK, a manufacturer of magnetic materials, hope to be the first to go to market with high-performance MRAM capable of storing gigabits of data. "We are planning to work together for about four years to really get this to a level of maturity where we can demonstrate the technology," Gallagher said. After that, IBM could begin designing products.

While MRAM's potential is huge, given its advantages over flash, the technology has been difficult to bring to market at a competitive price and performance level. Freescale, however, has been moving quickly in developing the technology. In launching MRAM last year, Freescale had more than 40 customers testing 9,000 devices. The primary use of the samples was in battery-packed SRAM replacement and high-density storage cache applications.

Freescale has also licensed the technology to Honeywell, which is developing MRAM for military and space applications.