Over the last decade, with the introduction of increasingly complex artificial intelligence (AI) technologies, the demand for computing power has increased exponentially. New energy-efficient hardware designs could help meet this demand while reducing the power consumption of computers, supporting faster processing and enabling AI training within the device itself.
In my opinion, we’ve already moved from the Internet age to the age of artificial intelligence, says Shan Wang, Leland T. Edwards Professor in the School of Engineering at Stanford University. We want to enable AI on edge training locally on your home computer, phone or smartwatch for things like heart attack detection or speech recognition. To do this, you need very fast, non-volatile memory.
Wang and his colleagues recently found a material that could bring a new type of memory closer to commercialization. In a new article published in Materials of nature, the researchers showed that a thin layer of a metal compound called manganese palladium three had the properties needed to facilitate a form of working memory that stores data in electron spin directions. This method of memory storage, known as spin-orbit magnetoresistive random access memory or SOT-MRAM, has the potential to store data faster and more efficiently than current methods, which store data using electricity and require continuous power input to maintain that data.
We have provided a building block for future energy-efficient storage elements, says Wang. It’s very basic, but it’s a breakthrough.
Take advantage of the spin of the electron
SOT-MRAM relies on an intrinsic property of electrons called spin. To understand the spin, imagine an electron as a spinning basketball balanced on the tip of a professional athlete’s finger. Since electrons are charged particles, the spin turns the electron into a tiny magnet, polarized along its axis (in this case, a line extending from the finger that balances the sphere). If the electron changes direction of rotation, the north-south poles of the magnet change. Researchers can use the up or down direction of that magnetism known as a magnetic dipole moment to represent the ones and zeros that make up the bits and bytes of computer data.
In SOT-MRAM, a current flowing through a material (the SOT layer) generates specific directions of rotation. The motion of those electrons, together with their spin directions, creates a torque that can change the spin directions and associated magnetic dipole moments of electrons in an adjacent magnetic material. With the right materials, storing magnetic data is as simple as changing the direction of an electric current in the SOT layer.
But finding the right SOT materials is not easy. Because of the way the hardware is designed, data can be stored more densely when the electron spin directions are oriented up or down in the z-direction. (If you imagine a sandwich on a plate, the x and y directions follow the edges of the bread, and the z direction is the toothpick stuck in the middle.) Unfortunately, most materials polarize electron spins in the y direction if current flows in the direction x.
Conventional materials only generate spin in the y direction, which means we would need an external magnetic field to make the switching happen in the z direction, which requires more energy and space, says Fen Xue, a postdoctoral researcher in the lab of Wang. In order to lower power and have higher memory density, we want to be able to accomplish this switching without an external magnetic field.
Researchers have found that manganese palladium three has the properties they need. The material is capable of generating spin in any orientation because its internal structure lacks the kind of crystalline symmetry that would force all electrons into any particular orientation. Using manganese palladium three, the researchers were able to demonstrate the switching of magnetization in both y and z directions without the need for an external magnetic field. Although not demonstrated in the manuscript, the x-direction magnetization can also be switched in the absence of an external magnetic field.
We have the same input current as other conventional materials, but now have three different directions of rotation, says Mahendra DC, who led the work as a postdoctoral researcher at Stanford and is the paper’s first author. Depending on the application, we can control the magnetization in any direction we want.
DC and Wang credit the multidisciplinary and multi-institutional collaboration that has enabled these advances. Evgeny Tsymbals’s lab at the University of Nebraska led the calculations to predict unexpected rotational directions and motion, and Julie Borcherss’s lab at the National Institute of Standards and Technology led the measurements and modeling efforts to reveal the intricate microstructures within manganese palladium three, Wang says. It really takes a village.
Possibility of production
In addition to its symmetry breaking structure, manganese palladium three has several other properties that make it a prime candidate for SOT-MRAM applications. It can, for example, survive and maintain its properties through the post-annealing process that electronics go through.
Post-annealing requires electronics to be at 400 degrees Celsius for 30 minutes, DC says. This is one of the challenges for new materials in these devices and manganese palladium three can handle it.
Additionally, the manganese palladium layer three is created using a process called magnetron sputtering, a technique already used in other aspects of memory storage hardware.
No new tools or new techniques are needed for this type of material, says Xue. We don’t need a structured substrate or special conditions to deposit it.
The result is a material that not only has new properties that could help meet our growing computing requirements, but can be seamlessly adapted to current manufacturing techniques. Researchers are already working on SOT-MRAM prototypes using manganese palladium three that will integrate into real devices.
We’re hitting a wall with current technology, DC says. So we need to figure out what other options we have.
Wang is professor of Materials Science and Engineering, and jointly of Electrical Engineering, a member of the Geballe Laboratory of Advanced Materials, Stanford Bio-X and the Wu Tsai Neurosciences Institute; and an affiliate of the Precourt Institute for Energy and the Stanford Woods Institute for the Environment.
Other Stanford co-authors of this research include senior research scientist Arturas Vailionis, adjunct professor Wilman Tsai, research consultant Chong Bi, research associate Xiang Li, and graduate student Yong Deng. Other co-authors are from the University of Nebraska, Taiwan Semiconductor Manufacturing Company, Kaunas University of Technology, National Institute of Standards and Technology, University of Arizona, Colorado School of Mines, National Yang Ming Chiao Tung University, Seikei University.
This work was funded by Semiconductor Research Corporation (SRC); Defense Advanced Research Projects Agency; the National Science Foundation; the National Research Council; Center for Semiconductor Technology Research; the National Council of Science and Technology, Taiwan; JSPS KAKENHI; Heiwa Nakajima Foundation; PMAC for the Fund for the Promotion of Scientific Research; and JST-FOREST. Some of this work was done at the Stanford Nano Shared Facilities (SNSF)/Stanford Nanofabrication Facility (SNF).
To read all of Stanford’s science stories, subscribe to the biweeklyStanford Science Digest.
#Material #Opens #Door #Energy #Efficient #Computing #Stanford #News