The Quest for the Spin Transistor
The mystical property of electron spin is revolutionizing the memory business. If it can do the same with logic, electronics will become "spintronics"
ILLUSTRATION: ROB MAGIERI
From the earliest batteries through vacuum tubes, solid state, and integrated circuits, electronics has staved off stagnation. Engineers and scientists have remade it repeatedly, vaulting it over one hurdle after another to keep alive a record of innovation unmatched in industrial history.
It is a spectacular and diverse account through which runs a common theme. When a galvanic pile twitches a frog's leg, when a triode amplifies a signal, or when a microprocessor stores a bit in a random access memory, the same agent is at work: the movement of electric charge. Engineers are far from exhausting the possibilities of this magnificent mechanism. But even if a dead end is not yet visible, the foreseeable hurdles are high enough to set some searching for the physics that will carry electronics on to its next stage. In so doing, it could help up the ante in the semiconductor stakes, ushering in such marvels as nonvolatile memories with enormous capacity, ultrafast logic devices that can change function on the fly, and maybe even processors powerful enough to begin to rival biological brains.v A growing band of experimenters think they have seen the future of electronics, and it is spin. This fundamental yet elusive property of electrons and other subatomic particles underlies permanent magnetism, and is often regarded as a strange form of nano-world angular momentum.
Microelectronics researchers have been investigating spin for at least 20 years. Indeed, their discoveries revolutionized hard-disk drives, which since 1998 have used a spin-based phenomenon to cram more bits than ever on to their disks. Within three years, Motorola Inc. and IBM Corp. are expected to take the next step, introducing the first commercial semiconductor chips to exploit spin--a new form of random access memory called M (for magnetic) RAM. Fast, rugged, and nonvolatile, MRAMs are expected to carve out a niche from the US $10.6-billion-a-year flash memory market. If engineers can bring the costs down enough, MRAMs may eventually start digging into the $35 billion RAM market as well.
The sultans of spin say memory will be just the beginning. They have set their sights on logic, emboldened by experimental results over the past two or three years that have shown the budding technologies of spin to be surprisingly compatible with the materials and methods of plain old charge-based semiconductor electronics. In February 2000, the Defense Advance Research Projects Agency announced a $15-million-a-year, five-year program to focus on new kinds of semiconductor materials and devices that exploit spin. It was the same Arlington, Va., agency's largesse of $60 million or so over the past five years that helped move MRAMs from the blackboard to the verge of commercial production.
Now proponents envision an entirely new form of electronics, called spintronics. It would be based on devices that used the spin of electrons to control the movement of charge. Farther down the road (maybe a lot farther), researchers might even succeed in making devices that used spin itself to store and process data, without any need to move charge at all. Spintronics would use much less power than conventional electronics, because the energy needed to change a spin is a minute fraction of what is needed to push charge around.
Other advantages of spintronics include nonvolatility: spins don't change when the power is turned off. And the peculiar nature of spin--and the quantum theory that describes it--points to other weird, wonderful possibilities, such as: logic gates whose function--AND, OR, NOR, and so on--could be changed a billion times a second; electronic devices that would work directly with beams of polarized light as well as voltages; and memory elements that could be in two different states at the same time. "It offers completely different types of functionality" from today's electronics, said David D. Awschalom, who leads the Center for Spintronics and Quantum Computation at the University of California at Santa Barbara. "The most exciting possibilities are the ones we're not thinking about."
Much of the research is still preliminary, Awschalom cautions. A lot of experiments are still performed at cryogenic temperatures. And no one has even managed to demonstrate a useful semiconductor transistor or transistor-like device based on spin, let alone a complex logic circuit. Nevertheless, researchers at dozens of organizations are racing to make spin-based transistors and logic, and encouraging results from groups led by Awschalom and others have given ground for a sense that major breakthroughs are imminent.
"A year and a half ago, when I was giving a talk [and] said something about magnetic logic, before I went on with the rest of my talk I'd preface my statement with, '...and now, let's return to the planet Earth,'" said Samuel D. Bader, a group leader in the materials science division at Argonne National Laboratory, in Illinois. "I can drop that line now," he added.
Quantum mechanical mystery
Spin remains an unplumbed mystery. "It has a reputation of not being really fathomable," said Jeff M. Byers, a leading spin theorist at the Naval Research Laboratory (NRL), in Washington, D.C. "And it's somewhat deserved."
Physicists know that spin is the root cause of magnetism, and that, like charge or mass, it is an intrinsic property of the two great classes of subatomic particles: fermions, such as electrons, protons, and neutrons; and bosons, including photons, pions, and more. What distinguishes them, by the way, is that a boson's spin is measurable as an integer number (0, 1, 2...) of units, whereas fermions have a spin of 1/2, 3/2, 5/2.... units.
Much of spin's elusiveness stems from the fact that it goes right to the heart of quantum theory, the foundation of modern physics. Devised in the early decades of the 20th century, quantum theory is an elaborate conceptual framework, based on the notion that the exchange of energy at the subatomic level is constrained to certain levels, or quantities--in a word, quantized.
Paul Dirac, an electrical engineering graduate of Bristol University, in England, turned Cambridge mathematician, postulated the existence of spin in the late 1920s. In work that won him a Nobel prize, he reconciled equations for energy and momentum from quantum theory with those of Einstein's special theory of relativity.
Spin is hard to grasp because it lacks an exact analog in the macroscopic world we inhabit. It is named after its closest real-world counterpart: the angular momentum of a spinning body. But whereas the ordinary angular momentum of a whirling planet, say, or curve ball vanishes the moment the object stops spinning and hence is extrinsic, spin is a kind of intrinsic angular momentum that a particle cannot gain or lose.
"Imagine an electronics technology founded on such a bizarre property of the universe," said Byers.
Of course, the analogy between angular momentum and spin only goes so far. Particle spin does not arise out of rotation as we know it, nor does the electron have physical dimensions, such as a radius. So the idea of the electron having angular momentum in the classical meaning of the term doesn't make sense. Confused? "Welcome to the club," Byers said, with a laugh.
The smallest magnets
Fortunately, a deep grasp of spin is not necessary to understand the promise of the recent advances. The usual imperfect analogies that somehow manage to render the quantum world meaningful for mortal minds turn out to be rather useful--as is spin's role in magnetism, a macroscopic manifestation of spin.
Start with the fact that spin is the characteristic that makes the electron a tiny magnet, complete with north and south poles. The orientation of the tiny magnet's north-south axis depends on the particle's axis of spin. In the atoms of an ordinary material, some of these spin axes point "up" (with respect to, say, an ambient magnetic field) and an equal number point "down." The particle's spin is associated with a magnetic moment, which may be thought of as the handle that lets a magnetic field torque the electron's axis of spin. Thus in an ordinary material, the up moments cancel the down ones, so no surplus moment piles up that could hold a picture to a refrigerator.
For that, you need a ferromagnetic material, such as iron, nickel, or cobalt. These have tiny regions called domains in which an excess of electrons have spins with axes pointing either up or down--at least, until heat destroys their magnetism, above the metal's Curie temperature. The many domains are ordinarily randomly scattered and evenly divided between majority-up and majority-down. But an externally applied magnetic field will move the walls between the domains and line up all the domains in the direction of the field, so that they point in the same direction. The result is a permanent magnet.
Ferromagnetic materials are central to many spintronics devices. Use a voltage to push a current of electrons through a ferromagnetic material, and it acts like a spin polarizer, aligning the spin axes of the transiting electrons so that they are up or down. One of the most basic and important spintronic devices, the magnetic tunnel junction, is just two layers of ferromagnetic material separated by an extremely thin, nonconductive barrier [see figure, "How a Magnetic Tunnel Junction Works" ]. The device was first demonstrated by the French physicist M. Jullière in the mid-1970s.
It works like this: suppose the spins of the electrons in the ferromagnetic layers on either side of the barrier are oriented in the same direction. Then applying a voltage across the three-layer device is quite likely to cause electrons to tunnel through the thin barrier, resulting in high current flow. But flipping the spins in one of the two ferromagnetic layers, so that the two layers have opposite alignment, restricts the flow of current through the barrier [bottom]. Tunnel junctions are the basis of the MRAMs developed by IBM and Motorola, one per memory cell.
Any memory device can also be used to build logic circuits, in theory at least, and spin devices such as tunnel junctions are no exception. The idea has been explored by Mark Johnson, a leading spin researcher at the Naval Research Laboratory, and others. Lately, work in this area has shifted to a newly formed program at Honeywell Inc., Minneapolis, Minn. The challenges to the devices' use for programmable logic are formidable. To quote William Black, principal engineer at the Rocket Chips subsidiary of Xilinx, a leading maker of programmable logic in San Jose, Calif., "The basic device doesn't have gain and the switching threshold typically is not very well controlled." To call that "the biggest technical impediment," as he does, sounds like an understatement.
Already on the drawing board are spin-based devices that would act something like conventional transistors--and that might even produce gain. There are several competing ideas. The most enduring one is known as the spin field-effect transistor (FET). A more recent proposal puts a new spin, so to speak, on an almost mythical device physicists have pursued for decades: the resonant tunneling transistor.
In an ordinary FET, a metal gate controls the flow of current from a source to a drain through the underlying semiconductor. A voltage applied to the gate sets up an electric field, and that field in turn varies the amount of current that can flow between source and drain. More voltage produces more current.
In 1990 Supriyo Datta and Biswajit A. Das, then both at Purdue University, in West Lafayette, Ind., proposed a spin FET in a seminal article published in the journal Applied Physics Letters. The two theorized about an FET in which the source and drain were both ferromagnetic metals, with the same alignment of electron spins. Electrons would be injected into the source, which would align the spins so that their axes were oriented the same way as those in the source and drain. These spin-polarized electrons would shoot through the source and travel at 1 percent or so of the speed of light toward the drain.
This speed is important, because electrons moving at so-called relativistic speeds are subject to certain significant effects. One is that an applied electric field acts as though it were a magnetic field. So a voltage applied to the gate would torque the spin-polarized electrons racing from source to drain and flip their direction of spin. Thus electron spins would become polarized in the opposite direction to the drain, and could not enter it so easily. The current going from the source to the drain would plummet.
Note that the application of the voltage would cut off current, rather than turn it on, as in a conventional FET. Otherwise, the basic operation would be rather similar--but with a couple of advantages. To turn the current on or off would require only the flipping of spins, which takes very little energy. Also, the polarization of the source and drain could be flipped independently, offering intriguing possibilities unlike anything that can be done with a conventional FET. For example, Johnson patented the idea of using an external circuit to flip the polarization of the drain, turning the single-transistor device into a nonvolatile memory cell.
A recent German breakthrough will "revolutionize" a major
spintronics subfield, one expert declared
Alas, 11 years after the paper by Datta and Das, no one has managed to make a working spin FET. Major efforts have been led by top researchers, such as Johnson at the NRL, Michael Flatté at the University of Iowa, Michael L. Roukes at the California Institute of Technology, Hideo Ohno of Tohoku University in Japan, Laurens W. Molenkamp, then at the University of Aachen in Germany, and Anthony Bland at the University of Cambridge in England. The main problem has been maintaining the polarization of the spins: the ferromagnetic source does in fact align the spins of electrons injected into it, but the polarization does not survive as the electrons shoot out of the source and into the semiconductor between the source and drain.
Recent work in Berlin, Germany, may change all that. In a result published last July in Physical Review Letters, Klaus H. Ploog and his colleagues at the Paul Drude Institute disclosed that they had used a film of iron, grown on gallium arsenide, to polarize spins of electrons injected into the GaAs. Not only was the experiment carried out at room temperature, but the efficiency of the injection, at 2 percent, was high in comparison with similar experiments. The work was "extremely important," said the Naval Research Laboratory's Johnson. "It will revolutionize this subfield. A year from now many spin-FET researchers will be working with iron."
The other kind of proposed spin transistor would exploit a quantum phenomenon called resonant tunneling. The device would be an extension of the resonant tunneling diode. At the heart of this device is an infinitesimal region, known as a quantum well, in which electrons can be confined. However, at a specific, resonant voltage that corresponds to the quantum energy of the well, the electrons tend to slip--the technical term is "tunnel"--freely through the barriers enclosing the well.
Generally, the spin state of the electron is irrelevant to the tunneling, because the up and down electrons have the same amount of energy. But by various means, researchers can design a device in which the spin-up and spin-down energy levels are different, so that there are two different tunneling pathways. The two tunnels would be accessed with different voltages; each voltage would correspond to one or the other of the two spin states. At one voltage, a certain level of spin-down current would flow. At some other voltage, a different level of spin-up current would go through the quantum well's barriers.
One way of splitting the energy levels is to make the two barriers of different materials, so that the potential energy that confines the electrons within the quantum well is different on either side of the well. That difference in the confining potentials translates, for a moving electron, into two regions within the quantum well, which have magnetic fields that are different from each other. Those asymmetric fields in turn give rise to the different resonant energy levels for the up and down spin states. A device based on these principles is the goal of a team led by Thomas McGill at the California Institute of Technology, with members at HRL Laboratories LLC, Jet Propulsion Laboratory, Los Alamos National Laboratory, and the University of Iowa.
Another method of splitting the energy levels is to simply put them in a magnetic field. This approach is being taken by a collaborative effort of nine institutions, led by Bruce D. McCombe at the University at Buffalo, New York.
Neither team has managed to build a working device, but the promise of such a device has kept interest high. A specific voltage would produce a certain current of, say, spin-up electrons. Using a tiny current to flip the spins would enable a larger current of spin-down electrons to flow at the same voltage. Thus a small current could, in theory anyway, be amplified.
Ray of hope
As these researchers refine the resonant and ballistic devices, they are looking over their shoulders at colleagues who are forging a whole new class of experimental device. This surging competition is based on devices that create or detect spin-polarized electrons in semiconductors, rather than in ferromagnetic metals. In these experiments, researchers use lasers to get around the difficulties of injecting polarized spin into semiconductors. By shining beams of polarized laser light onto ordinary semiconductors, such as gallium arsenide and zinc selenide, they create pools of SPIN-POLARIZED ELECTRONS.
Some observers lament the dependence on laser beams. They find it hard to imagine how the devices could ever be miniaturized to the extent necessary to compete with conventional electronics, let alone work smoothly with them on the same integrated circuit. Also, in some semiconductors, such as GaAs, the spin polarization persists only at cryogenic temperatures.
In an early experiment, Michael Oestreich, then at Philips University in Marburg, Germany, showed that electric fields could push pools of spin-polarized electrons through nonmagnetic semiconductors such as GaAs. The experiment was reported in the September 1998 Applied Physics Letters.
Then over the past three years, a breathtaking series of findings has turned the field into a thriving subdiscipline. Several key results were achieved in Awschalom's laboratory at Santa Barbara. He and his co-workers demonstrated that pools of spin-coherent electrons could retain their polarization for an unexpectedly long time--hundreds of nanoseconds. Working separately, Awschalom, Oestreich, and others also created pools of spin-polarized electrons and moved them across semiconductor boundaries without the electrons' losing their polarization.
If not for these capabilities, spin would have no future in electronics. Recall that a practical device will be operated by altering its orientation of spin. That means that the spin coherence has to last, at a minimum, longer than it takes to alter the orientation of that spin polarization. Also, spintronic devices, like conventional ones, will be built with multiple layers of semiconductors, so moving spin-polarized pools across junctions between layers without losing the coherence will be essential.
Awschalom and his co-workers used a pulsed, polarized laser to establish pools of spin-coherent electrons. The underlying physics revolves around the so-called selection rules. These are quantum-theoretical laws describing whether or not an electron can change energy levels by absorbing or emitting a photon of light. According to those selection rules, light that is circularly polarized will excite only electrons of one spin orientation or the other. Conversely, when spin-coherent electrons combine with holes, the result is photons of circularly polarized light.
In his most recent work, Awschalom and his graduate student, Irina Malajovich, collaborated with Nitin Samarth of Pennsylvania State University in University Park and his graduate student, Joseph Berry. As he has in the past, Awschalom performed the experiment on pools of electrons that were not only spin polarized but were also precessing. Precession occurs when a pool of spin-polarized electrons is put in a magnetic field: the field causes their spin axes to rotate in a wobbly way around that field. The frequency and direction of rotation depend on the strength of the magnetic field and on characteristics of the material in which the precession is taking place.
The Santa BarbaraPenn State team used circularly polarized light pulses to create a pool of spin-coherent electrons in GaAs. They applied a magnetic field to make the electrons precess, and then used a voltage to drag the precessing electrons across a junction into another semiconductor, ZnSe. The researchers found that if they used a low voltage to drag the electrons into the ZnSe, the electrons took on the precession characteristics of the ZnSe as soon as they got past the junction. However, if they used a higher voltage, the electrons kept on precessing, as though they were still in the GaAs [see illustration, "Precessional Mystery" ].
"You can tune the whole behavior of the current, depending on the electric field," Awschalom said in an interview. "That's what was so surprising to us." The group reported its results in the 14 June issue of Nature, prompting theorists around the world to wear out their pencils trying to explain the findings.
Other results from the collaboration were even more intriguing. The Santa Barbara and Penn State researchers performed a similar experiment, except with p-type GaAs and n-type ZnSe. N-type materials rely on electrons to carry current; p-type, on holes. Because the materials were of two different charge-carrier types, an electric field formed around their junction. That field, the experimenters found, was strong enough to pull a pool of spin-coherent electrons from the GaAs immediately into the ZnSe, where the coherence persisted for hundreds of nanoseconds.
The result was encouraging for two reasons. As Awschalom put it, "It showed that you can build n-type and p-type materials and spin can get through the interfaces between them just fine." Equally important, it demonstrated that the spin can be moved from one kind of semiconductor into another without the need for external electric fields, which wouldn't be practical in a commercial device.
"The next big opportunity is to make a spin transistor," Awschalom added. "These results show, in principle, that there is no obvious reason why it won't work well."
Such a device is at least several years away. But even if researchers were on the verge of getting a spin transistor to work in the laboratory, more breakthroughs would be necessary before the device could be practical. For example, the fact that the device would need pulses of circularly polarized laser light would seem an inconvenience, although Awschalom sees a bright side. The gist is that the photons would be used for communications among chips, the magnetic elements for memory, and the spin-based devices for fast, low-power logic.
It's far-fetched now--but no more so than the idea of 1GB DRAMs would have seemed in the days when triodes ruled.
To Probe Further
Hot off the presses is Semiconductor Spintronics and Quantum Computation, edited by David D. Awschalom, Nitin Samarth, and Daniel Loss. The 250-page book was released last October by Springer Verlag, Berlin/Heidelberg; ISBN: 3540421769.
The November/December issue of American Scientist, published by the scientific research society Sigma Xi, included an eight-page overview titled "Spintronics" by Sankar Das Sarma. See Vol. 89, pp. 516523.
Honeywell Inc.'s Romney R. Katti and Theodore Zhu described the company's magnetic RAM technology in "Attractive Magnetic Memories," IEEE Circuits & Devices, Vol. 17, March 2001, pp. 2634.