The promise of MRAM

The promise of MRAM
Magnetoresistive random-access memory, better known as MRAM, was a topic of much discussion at the 2002 Symposium on VLSI Circuits, held in Honolulu in mid-June. MRAM offers nonvolatility, and much faster program times than flash. That much seems assured. The main challenge is getting the cell size down to something affordable for embedded applications such as cell phones.

Cutting the power consumption during the write mode is another hurdle. Manufacturing issues remain, too, such as how to avoid damaging the magnetic layers during high-temperature steps in the back end of the line.

Motorola presented a 1-Mbit MRAM test chip at the symposium, made with a 0.6-micron process, with a cell size of 7.2 square microns. On an 0.18-micron process, Saied Tehrani, Motorola's director of MRAM technology, projects the cell size will shrink to about 0.7 micron2, small enough to be competitive as a "universal" memory, replacing many memory types.

Also in Honolulu, Sony introduced an MRAM test chip, made with a 0.35-micron process, which had a cell size of 5.8 microns2. "If we can reduce the power consumption during the write mode, and can get good yields at high densities, MRAM is a good candidate as an embedded memory," said Makoto Motoyoshi of Sony's MRAM team.

In the memory game, cell size roughly translates into cost per bit. IBM and Infineon, partners in MRAM research, have been quiet on the MRAM front lately. One source said the two companies looked at the cell size issue, asked customers for feedback on cost premiums for nonvolatility and decided to scale back their R&D efforts. Bijan Davari, vice president of technologies and emerging products at IBM's Microelectronics Division, disagreed, saying IBM maintains a small research team, spread out over three research labs in Almaden, Calif., and Yorktown Heights and Fishkill, N.Y.

"MRAM is a disruptive technology that could change the world, but it remains unclear whether the cell size can be fundamentally scaled," said Davari. "We asked some of our people to look at the scaling question and the answer that came back was a yes."

If MRAM can get established in niche markets that need a universal memory, then it may go on to replace flash, which faces its own set of scaling issues. With attributes of SRAM, flash and DRAM, there is no question that MRAM offers great promise, with one big question: Will the cost per bit ever get in the ballpark?

What are your thoughts? Please send feedback to dlammers@cmp.com.