TOKYO Japan's semiconductor industry is rushing headlong into the development of magnetic RAMs, spurred by the market prospects for advanced nonvolatile technologies and by an acknowledged need to catch up to the West in the fledgling field. The involvement of the Japanese could accelerate the technology's expansion beyond the defense applications targeted by the U.S. military.
MRAM proponents say the technology scales better to low voltages than DRAM, SRAM and flash memory and that it requires fewer mask steps than DRAM. Observers and researchers here said Japan's MRAM researchers are about three years behind their colleagues in the United States, where the Defense Advanced Research Projects Agency (Darpa; Arlington, Va.) spearheaded a $50 million MRAM-development effort in 1994 and commissioned several major U.S. semiconductor makers for the task. A handful of companies are pursuing the technology outside the Darpa effort, most notably Hewlett-Packard Co., which said in February that it expects to field MRAMs in two years that will be targeted as next-generation replacements for hard-drive storage.
Darpa raised eyebrows here last fall when Stuart Wolf, program manager for the Defense Sciences Office, visited Japanese chip manufacturers to explain the advantages of MRAM, including unlimited read/writes, full military temperature ranges, immunity from radiation, fast access times, high cell densities, high write speeds, low power consumption and long-term storage. Stuart told the vendors that a radiation-hardened 16-kbit device with an access time of under 100 ns had been fabricated and that a 1-Mbit prototype could be ready late this year.
European companies and research institutes are also taking an interest in MRAM. Japanese sources said Philips and Siemens are involved in MRAM research, as are European universities. The German government is interested in MRAM as a national technology project, observers said.
"The Americans and Europeans are very excited about MRAM research," said Kazuo Kobayashi, research fellow at Fujitsu Laboratories Ltd. (Atsugi, Japan). "We cannot ignore this activity."
The Advanced Technology Information Program (ATIP), a non-profit U.S. organization that studies major areas of electronics research, will sponsor an international workshop on spin electronics in Tokyo at the end of November. Spin electronics concerns the manipulation of electrons as they rotate around a nucleus and is the science behind disk-drive giant magnetoresistive (GMR) heads as well as MRAM technology.
Many of Japan's leading electronics companies are well-schooled in spin electronics because of their work in GMR. But they lag in applying the concepts to other technologies.
That situation could soon change, though. Those invited to attend the ATIP-sponsored conference include Darpa's Wolf; Gary Prinz, the Naval Research Laboratory's resident expert on magneto-electronics; senior engineers at Toshiba, Fujitsu, Hitachi and NTT; and, significantly, Koji Ando, who oversees the Electrotechnical Laboratory at Japan's powerful Ministry of International Trade and Industry. With the Asian flu having afflicted corporate R&D budgets, MITI funding is an important missing link for MRAM development in Japan.
Sources here said Ando's organization is doing much of the preliminary work to kick off a domestic research project for MRAMs, perhaps by next year or 2001. Among the companies involved in drawing up the plans are Toshiba, Fujitsu, Hitachi, NEC and Matsushita.
While progress has been made in determining the structure and materials needed for MRAM development, there are still many hurdles to jump before MRAMs can be made production-worthy. That's particularly true in Japan, where no market exists for a specialized technology that to date has been defense-centric.
Among the issues to tackle are architectures, materials development, submicron manufacturing, wiring and the feasibility of integrating MRAM with logic, said Koichiro Inomata, chief fellow at Toshiba Corp.'s Research and Development Center (Kawasaki, Japan), which started researching MRAM about two years ago.
"We need big money to set up a memory line, do basic research and fund the technology," Inomata said. "Japanese companies have hesitated to put money into this, so we'll need a national project."
While Darpa's Wolf has be-come the leading evangelist or MRAM technology in the United States, "in Japan one person cannot take the initiative," said Kazuhiro Goto of ATIP. "The direction of research is usually decided by the majority of engineers. It's democratic and safe, but it can be dangerous."
Recently, however, MRAM research activity has picked up in Japan, and Japanese researchers are confident they can catch up to the West once funding and a national plan are available. Several companies here have already done initial investigations and say they have nailed down the basic structure and materials needed for MRAM.
While much of the early U.S. work in MRAM focused on GMR technology, many researchers say GMR is not viable for commercial applications. Rather, a technology known as tunneling magnetic resistance (TMR) is expected to be the basis of future MRAM. Darpa contractor Honeywell Inc. is close to proceeding with a GMR production line; contractors IBM Corp. and Motorola Inc. are concentrating on tunneling approaches.
TMR may be the Japanese chip vendors' foot in the door to commercial MRAMs. "In my opinion, materials for TMR are very important for realizing MRAM technology, and much of that material has not been developed yet," Inomata said. "In that regard, U.S. and Japanese companies are in nearly the same position."
What's more, TMR technology can be applied to recording heads and encoders.
TMR-based MRAMs sandwich a layer of insulating material between two electrodes of magnetic material, such as ion nickel. Current runs perpendicularly, "tunneling" through the insulator that separates it from a sheath of copper. At the base of one of the electrodes is a fixed anti-ferromagnetic layer that creates a strong coupling field.
When a magnetic field is applied, electrons flow from one electrode to another, creating 0 and 1 states. With GMR, the current flows horizontally rather than perpendicularly and does not use an insulator layer.
Initial research indicates that GMR technology has a much lower magnetic resistance (MR) ratio than TMR. GMR's ratio is about 7 percent and has the potential to increase to about 15 percent. That limits is potential performance compared with TMR, which has the potential to hit 30 to 40 percent, according to Fujitsu's Kobayashi.
One downside to TMR is its higher resistance, which slows the flow of the electrons. Toshiba, however, says it has developed an alumina oxide insulator that improves TMR's resistivity while maintaining its superior magnetic resistance ratio.
Toshiba has also come up with a double-junction structure that it says keeps the MR ratio from dropping too low when the bias voltage is applied. Normally the MR ratio falls 50 percent, but there's only a 10 percent drop using the double-junction structure, Inomata said.
Increasing the overall MR ratio, initially to 30 percent, is the next step. "Theoretically, TMR has the potential of going to 50 percent," Inomata said.
Even if researchers can get to that point, however, they face a tougher challenge in finding the right process recipe to make manufacturable ICs. "Semiconductor technology and magnetic materials are hard to blend; magnetic material is like an impurity," Inomata said.
The current talk of MRAM's potential as a nonvolatile memory is reminiscent of the praise heaped on ferroelectric RAM (FRAM) more than a decade ago by Japanese vendors. While several companies produce FRAMs here today, many remain stymied by process-technology difficulties.
"FRAM has very superior qualities but is very difficult to integrate on semiconductors because it needs high temperature," said Inomata, adding that FRAM's capacity is limited to 10 Mbits as a result.
MRAM, on the other hand, has the potential to scale to more than 1 Gbit. Other advantages include power consumption of less than 10 mW, 30- to 40-ns access times (for TMR) and infinite read/write cycles, Inomata said.
In the United States, Darpa launched a research program in magnetic materials and devices in '94, sharing the costs with Honeywell, IBM and Motorola. Several other U.S. companies have developed products based on GMR technology, and last winter Hewlett-Packard said it intends to join IBM and Motorola in the TMR market.
Darpa's Wolf, who moved to the agency from the Naval Research Lab to spearhead the U.S. MRAM effort, said that "major decisions will be made in the next year or two" on U.S. production of MRAM devices.
The U.S. military is funding spin-electronics research in hopes of exploiting nonvolatile-memory capabilities in embedded systems for use in satellites, strategic missiles, avionics and other mission-critical applications. For instance, coding information for satellites could be loaded on rad-hard MRAM devices that would ensure that satellites remain on station.
Rebooting aircraft computers using traditional storage technologies delays operations on the flight line. Nonvolatile MRAM technology "would change the way the military operates," Wolf said.
But the subtle mechanics of micromagnetic technology continue to challenge U.S. researchers and potential manufacturers. What needs to be better understood, according to Darpa's Wolf, is how magnetic domains switch. Researchers at Carnegie-Mellon University (Pittsburgh) and at the University of Arizona have developed simulation codes designed to help engineers better understand the phenomenon.
Beyond military apps, government researchers envision MRAM technology showing up within five years in embedded applications, such as cell phones and digital cameras. The dawn of the MRAM-based laptop, which will eliminate boot-up delays, will take longer, they said.
"The payoff is going to be in all the mobile applications," the Naval Research Laboratory's Prinz said, particularly when Gbit MRAMs can be integrated into cell phones to dump data onto hard drives.
Among the Darpa contractors, Honeywell is the closest to realizing production of GMR memory cells. Honeywell has also licensed its technology to smaller companies.
IBM and Motorola are focusing on spin-dependent tunneling approaches. IBM is applying previous work on Josephson junction technology to its MRAM effort.
HP, meanwhile, will pit its MRAMs against more-expensive flash memories. HP also plans to combine MRAMs with atomic-resolution storage technology to replace hard drives. Mike Matson, general manager of HP's Information Storage Group, said he expects the combined technology to grab half the traditional hard-drive market over time.
George Leopold contributed to this story