Flash memory’s ‘midlife crisis’ boosts Magnetic RAM

A new form of memory based on magnetic principles is nearly ready for mass production, offering benefits for increasingly complex mobile devices.

IBM and Infineon will jointly present a paper this week that demonstrates how Magnetic Random Access Memory, one of the leading candidates to replace flash memory in mobile phones, could be ready for commercial production by 2005.

Magnetic Random Access Memory (MRAM) combines technological principles from both the magnetic world - the basis for the hard-drive industry - and silicon manufacturing.

In MRAM, a tiny magnetic field is created inside a memory cell on a chip. The computer then measures the electrical resistance exhibited by the magnetic field at any given moment to determine whether the cell should be read as a "1" or a "0", the binary building blocks of data.

Conventional flash memory, the mainstay for storing data on phones, also works by exhibiting different levels of electrical resistance, but it requires a considerable amount of electricity to switch between the "1" and "0" state. Ideally, MRAM will use less power and capture data faster than current flash memory.

While IBM has shown off MRAM circuits before, the data being released at the Very Large Scale Integration (VLSI) Symposium taking place this week in Kyoto, Japan, shows that MRAM chips are amenable to mass manufacturing, said Randy Isaac, vice president of strategic alliances at IBM Technology. The chip described in the paper was made on the 180-nanometre process, which has been used in mass manufacturing since 1999, and holds 128Kb of data.

"The big advantage here is the write time. In flash, the write process can take milliseconds. MRAM could put it in the nanosecond range," Isaac said.

Intel, AMD and others will present papers at the conference this week, one of the major, annual semiconductor research conferences.

Although humans can't detect the difference between a millisecond (one thousandth of a second) and a nanosecond (one billionth), the gap could make a lot of difference to mobile phone manufacturers in the future. With the advent of picture and video mobile phones, capturing and storing large amounts of data is becoming more crucial, Isaac said.

Flash memory, which retains data even after the host computer is turned off, is approaching a midlife crisis. The technology is a necessary component of PDAs (personal digital assistants), telecom equipment, most small MP3 players and mobile phones. Semiconductor manufacturers, however, say they are having increasing difficulty shrinking the size of these chips, a factor that could begin to reduce the profitability of making flash.

As a result, chipmakers have been tinkering with alternatives such as Ovonics Unified Memory, memory made out of the same material as CD discs; Silicon Nanocrystals, which replaces a solid layer inside chips with a crystal lattice; and MRAM.

MRAM "shows an awful lot of promise", said Jim Handy, a memory analyst at Semico Research. "They have a lot of experience in combining magnetic technology with silicon technology."

Still, Handy pointed out, MRAM involves a process that makes producing these chips fundamentally different from standard silicon chips, and anytime a manufacturer veers off the typical silicon road map, it becomes questionable whether the new technology will ultimately be more profitable than making chips using the old methods.

Critics have said that the resistance levels in MRAM are too subtle; in other words, a "1" looks too much like a "0". Changes between the "1" and "0" state are accomplished by altering the spin of electrons on one of the magnetic poles. In flash, electrons are shoved through a glass barrier, a process that researchers call "violent".

Isaac acknowledged that subtlety was a problem but has improved. There is now a 30 percent to 40 percent difference in resistance between the two basic states.

"Magnetic fields have been used to store data since time immemorial," he joked. "We were using it in the early 1960s and 1950s."

IBM is aiming to produce samples that could be issued to device makers by 2004. Potentially, MRAM chips could be mass manufactured by 2005, but IBM won't commit to any plans until demand for these chips begins to appear.

IBM and Infineon have worked jointly on various memory projects for more than a decade.