Electronic technology has evolved enormously over the past century, but in the most fundamental way it has not changed at all. From the earliest vacuum tube amplifiers to today's billion-transistor processors, all electronic devices work by moving electrical charges around. The countless discoveries and innovations that made the digital age what it is today were all made possible by our ever-improving mastery over electrons.
But those electrons are now beginning to rebel.
As we build transistors and other components with nanoscale dimensions, processors and memories are becoming so dense that even their infinitesimal individual currents are combining to produce scorching heat. Furthermore, quantum effects that were negligible before are now so pronounced that they're threatening to render circuits inoperable. The upshot is that we're fast approaching the point when moving charge is not going to be enough to keep Moore's Law chugging along.
In anticipation of that day, researchers all over the world are already working on a promising alternative. We have set our sights on a different property of electrons, which we hope to exploit for storing and processing data. This property is spin.
Spin is a fundamental yet elusive quantum attribute of electrons and other subatomic particles. It is often regarded as a bizarre form of nanoworld angular momentum, and it underlies permanent magnetism. What makes spin interesting for electronics is that it can assume one of two states relative to a magnetic field, typically referred to as up and down, and you can use these two states to represent the two values of binary logic—to store a bit, in other words.
The development of spin-based electronics, or spintronics, promises to open up remarkable possibilities. In principle, manipulating spin is faster and requires far less energy than pushing charge around, and it can take place at smaller scales. The holy grail in the field is a spin transistor. Chips built out of spin transistors would be faster and more powerful than traditional ones and, farther down the road, may feature such new and remarkable properties as the ability to change their logic functions on the fly.
We're still decades away from being able to build such a thing. But chips that exploit spin in a more modest way are already available. At least one company, Everspin Technologies, of Chandler, Ariz., is now selling magnetoresistive random access memory, or MRAM, a kind of spintronic memory. And many others—including Freescale, Honeywell, IBM, Infineon, Micron, and Toshiba, as well as start-ups and university research groups—are busy investigating MRAM technology.
The reason for all this interest is clear. Today's computers often use four kinds of storage. Dynamic random access memory, or DRAM, has high density but needs to be constantly refreshed and consumes lots of power. Static random access memory, or SRAM, used in caches, is fast to read and write but takes up considerable space on a chip. Flash, unlike SRAM and DRAM, is nonvolatile but is quite slow to write to. And then there are hard disk drives; these have high density but rely on moving parts, which impose size and speed limitations. MRAM is attractive because it could, in principle, replace all other kinds of memory.
Rather than representing a bit as charge in a capacitor or as the state of an interconnected set of transistors, MRAM stores data using the spin of electrons in a ferromagnetic substance, which is to say it stores data by creating a magnetic alignment in one direction or the other. In a tiny region of that material, spin up means 0, and spin down means 1. Proponents say that as MRAM improves, it could combine all the advantages of SRAM, DRAM, flash, and hard disks—with none of their shortcomings. It would be a compact, speedy, low-power, and nonvolatile "universal memory." With MRAM, a computer wouldn't have to juggle data between main memory, cache, and disk; instead, it could load all data into its working memory. This capability would make possible instant-on systems and maybe even change the way we think about computer architecture.
At the moment, however, MRAM suffers from two problems: The density of bits is low, and the cost of chips is high. The early MRAM designs needed lots of current to change a 1 to a 0 or vice versa. This requirement prevented their further miniaturization. Improved designs might overcome that hurdle using novel techniques and materials, but they would operate at only liquid-nitrogen temperatures. This is not going to work for your iPod.
This problem—the need for cryogenic temperatures to reduce the write current of MRAM—has been the focus of our work at North Carolina State University. It's a major challenge, but we've recently made a significant breakthrough: We demonstrated a device that shows potential as an MRAM memory cell. It can be written to using conventional voltage levels and almost no current at all. The key is a material called gallium manganese nitride, a semiconductor whose magnetic properties we can manipulate electrically. And here's the best part: It works at room temperature.