In 2017, the Oak Ridge Leadership Computing Facility celebrated 25 years of leadership in high-performance computing. This article is part of a series summarizing a dozen significant contributions to science enabled by OLCF resources. The full report is available here.
At the turn of the 21st century, materials scientists at ORNL were working on one of the biggest technical challenges of the new century: information storage. Since the OLCF was founded in 1992, the amount of digital information stored around the world each year had increased a thousandfold, from billions to trillions of gigabytes.
The ability to increase storage capacity is not only important for preserving information but also for developing more powerful and efficient computers. That’s why a research team led by ORNL’s William Butler used OLCF resources, including an IBM Power system, to model the material properties of layered magnetic films—to expand the limits of computing and storage capacity.
When a computer “reads” data in binary code, expressed as a series of ones and zeroes, the read-out head on the disk drive is sensing the magnetic fields from the magnetic films on a hard disk and converting them into electrical signals.
In 1988, physicists Albert Fert and Peter Grünberg independently discovered that electrical resistance may decrease when an external magnetic field aligns the magnetization of two layers of film, enabling the more efficient writing and erasing of data—a discovery that later won Fert and Grünberg the 2007 Nobel Prize in Physics. This effect, known as giant magnetoresistance (GMR), became a popular subject of materials research for computing and devices.
A decade later at ORNL, Butler’s team was modeling the GMR effect in thin film systems composed of transition metals, which can have useful electromagnetic properties for writing and storing digital information. ORNL researchers with expertise in x-ray diffraction, microscopy, high-temperature materials, and HPC were trying to identify the best transition metals for creating a high magnetoresistance effect to increase storage density. The team was also interested in materials that could withstand power interruptions without losing data, thus increasing reliability.
The thin films like those used on disk drives include two types of magnetic materials: ferromagnetic (in which the magnetism of the atoms aligns in the same direction) and antiferromagnetic (in which magnetism aligns in different directions).
Butler’s team was using computationally intensive, first-principles calculations on ORNL supercomputers to model a phenomenon related to GMR known as tunneling magnetoresistance (TMR)—in which a magnetic tunnel junction is created by “sandwiching” an antiferromagnetic (insulating) barrier between two ferromagnetic (conductive) layers.
In classical physics theory, an electron cannot cross the insulating barrier. However, the team was demonstrating through the laws of quantum mechanics that, with the right materials, a magnetic tunnel junction could be created, allowing electrons to “tunnel” from layer to layer when their magnetism was aligned in the same direction. At a high TMR ratio, or contrast, a material exhibits a high probability of tunneling when magnetism is aligned across the layers, and a low probability for tunneling when magnetism is anti-aligned across the layers. A high TMR contrast means the size of information could shrink with less distortion or loss. At first, in laboratory observations of some common transition metals, the ORNL team and others were recording about a 50 to 70 percent TMR ratio.
The team used a first-principles electronic structure code to calculate the magnetoresistance of an Fe|MgO|Fe “sandwich,” which used three layers of two ferromagnetic layers (iron) on either side of an insulating, antiferromagnetic layer of magnesium oxide (MgO). TMR had been experimentally demonstrated for iron and MgO but had not been calculated. What the ORNL team found was a staggering contrast: about 500 percent.
“The discovery of high-contrast tunneling magnetoresistance by Bill Butler’s team was a clear example of the predictive power of computational science. From there, two courageous experimental groups, one in the US and one in Japan, were able to confirm these computational results.” —Jack Wells, Oak Ridge National Laboratory
Such a high contrast enables the magnetic read-out head of a disk drive to capture more sensitive signals sent by smaller magnetic regions, thereby increasing storage capacity without losing quality. First discovered in computer modeling at ORNL, the tunnel junction between the Fe and MgO layers was confirmed by experimentalists a few years later; and even greater contrasts were observed between crystalline MgO insulating layers and ferromagnetic materials other than iron. Today, most devices in magnetic read-out heads rely on MgO-based tunnel junctions, as does a type of resilient memory known as magnetoresistive random-access memory (MRAM) that has long been of interest to the computing community for its potential use as fast, reliable, and energy-efficient permanent memory.
Related Publication: Butler, W. H., Zhang, X.-G., and Schulthess, T. C. (2001), Spin-dependent Tunneling Conductance of Fe|MgO|Fe Sandwiches, Physical Review B, Volume: 63.