Portland, Ore. For all their promise, magnetic random-access memories have barely made it out of the lab due to problems plaguing their scaling to smaller sizes namely, the need for lower drive current and thinner metallization. Now researchers at Sandia National Laboratories claim to have patented a method for solving MRAM and other such metal-on-insulator problems.
"Ordinarily, putting metal on an insulator is like putting water on a waxed car," said Dwight Jennison, a theoretical physicist at Sandia National Laboratories (Albuquerque, N.M.). "What we are offering here, to anyone who is trying to mix insulators and metals, is the ability to make a strong interface between them, resulting in more reliable devices that are less likely to develop cracks in everything from thin films to macroscale metal-clad ceramics."
MRAMs promise to be low-power, nonvolatile, radiation-hardened memories capable of storing the entire operating system of a personal computer. "IBM, Motorola and Hewlett-Packard have all promised to deliver MRAMs within a year," Jennison said. "But we believe our invention will allow their magnetic layers to be flipped with much less current, thereby enabling smaller sizes with much less heating."
The method discovered by Jennison with chemist Scott Chambers at the Pacific Northwest National Laboratory and Jeffrey Kelber, a professor of chemistry at the University of North Texas could have implications well beyond MRAMs. It could also revolutionize every macroscale industrial process that involves putting metal on insulators, such as metal-clad ceramics that today require extensive brazing.
The deceptively simple process enables the top metal electrode on MRAMs to be slimmed down from microns of thickness to just six atomic layers. When incorporated into the "Model-T" MRAM processes being authored at semiconductor labs today, said Jennison, Sandia's "hot-rod" process should turbo-charge production.
"Our process should be able to catapult the Model-T designs of today up to something like a 1930 Hudson," he said. "We think all MRAM makers will adopt it for their next-generation devices."
MRAMs require two magnetic layers per bit. On the bottom, a few atomic layers of crystalline aluminum metal are placed over a fixed magnetic layer, thereby fixing the aluminum layer's magnetic direction. Then the metallic aluminum is oxidized, leaving a nanoscale layer on top of the aluminum oxide that acts as an insulator. Next, metal is put on top of the oxide but wired so its magnetic direction can be flipped, resulting in a 15 percent difference in resistance through the device if the magnetism lines up, compared with when the two magnetic layers point in opposite directions.
But, as is always the case with metal on insulators, when you deposit the metal, it balls up like water on a waxed car, and the balls are not even magnetic. MRAM makers have to keep depositing metal until there are so many balls that they suddenly coalesce into a magnetic, but thick, top layer.
"With our method, you need only six atomic layers of metal to make it crystalline and, therefore, magnetic," said Jennison. "Our process can greatly thin the top metal layer of aluminum oxide on the next generation of nonvolatile magnetic random-access memories, because without our process, the aluminum top layer does not wet the aluminum oxide beneath it."
How does it work?
The Sandia scheme has both high-temperature and low-temperature variations in its methodology, but the common denominator is water vapor. By introducing water vapor into the reaction chamber over the aluminum oxide, the oxide becomes hydoxolated that is, hydoxile (OH) groups attach to the surface, thereby wetting the metal-deposition step and enabling only six atomic layers to attain crystallinity and thus be magnetic.
"By exposing the aluminum oxide to water vapor, you create a whole hydoxile layer on top of the film," said Jennison. "And now, when you start to deposit that second metal layer which is usually done by evaporating the metal onto the surface about one-tenth of a monolayer of the atoms will react, becoming oxidized with a metal atom reacting with two neighboring OHs, releasing molecular hydrogen gas [H2 from the two OHs], leaving the oxidized metal atoms as anchors. Without the electrons previously provided by the hydrogens, the oxygens bond to a single metal atom, with the metal giving up two of its electrons to the oxygen atoms and thereby becoming doubly oxidized."
Jennison discovered the high-temperature process first, in cooperation with Kelber at the University of North Texas. It's suitable to substitute for brazing in traditional methods of putting metal on ceramics, he said. In that application, the ceramic is exposed to water vapor and then heated before the metal is deposited. The heating evaporates most of the OH, leaving about one-third of the surface to bond strongly with the metal.
"My friend Jeffrey Kelber called and asked if copper should wet sapphire, and I said no. And he said, 'Well, it's wetting.' So I asked him if hydoxyl groups were present," said Jennison. "He hung up and looked at the photo-emission spectra, which can detect OH, and called back to say, 'Guess what? I have a one-third monolayer of OH.' " Jennison said he knew that was the cause of the wetting, "because I had already done the calculation that showed this great increase in bonding that occurs if you have an OH on the surface and then put a copper atom right next to it." Jennison jointly holds the high-temperature patent with Kelber.
The low-temperature variation, suitable for use in fabricating MRAMs, was the result of an experiment Jennison subsequently suggested that Pacific Northwest's Chambers perform. The two hold the low-temperature patent jointly. "We found that even if there is up to a 7 percent mismatch in the interatomic spacing between the metal and the oxide, you still achieve crystallinity in the metal with only about six atomic layers," said Jennison.