Sweating the small stuff Home

Microelectronic devices that function by using the spin of the electron are a nascent multibillion-dollar industry--and may lead to quantum microchips

By David D. Awschalom, Michael E. Flattˇ and Nitin Samarth

As rapid progress in the miniaturization of semiconductor electronic devices leads toward chip features smaller than 100 nanometers in size, device engineers and physicists are inevitably faced with the looming presence of quantum mechanics--that counterintuitive and sometimes mysterious realm of physics wherein wavelike properties dominate the behavior of electrons. Pragmatists in the semiconductor device world are busy conjuring up ingenious ways to avoid the quantum world by redesigning the semiconductor chip within the context of "classical" electronics [see "A Vertical Leap for Microchips," by Thomas H. Lee; Scientific American , January]. Yet some of us believe that we are being offered an unprecedented opportunity to define a radically new class of device that would exploit the idiosyncrasies of the quantum world to provide unique advantages over existing information technologies.

One such idiosyncrasy is a quantum property of the electron known as spin, which is closely related to magnetism. Devices that rely on an electron's spin to perform their functions form the foundation of spintronics (short for spin-based electronics), also known as magnetoelectronics. Information-processing technology has thus far relied on purely charge-based devices--ranging from the now quaint vacuum tube to today's million-transistor microchips. Those conventional electronic devices move electric charges around, ignoring the spin that tags along for the ride on each electron.

Possible Solutions

Magnetism (and hence electron spin) has nonetheless always been important for information storage. For instance, even the earliest computer hard drives used magnetoresistance--a change in electrical resistance caused by a magnetic field--to read data stored in magnetic domains. It is no surprise that the information storage industry has provided the initial successes in spintronics technology. Most laptop computers now come fitted with high-capacity hard drives that pack an unprecedented amount of data into each square millimeter. The drives rely on a spintronic effect, giant magnetoresistance (GMR), to read such dense data.

More sophisticated storage technologies based on spintronics are already at an advanced stage: in the next few years, MRAM (magnetic random-access memory), a new type of computer memory, will go on the market. MRAMs would retain their state even when the power was turned off, but unlike present forms of nonvolatile memory, they would have switching rates and rewritability challenging those of conventional RAM.

In today's read heads and MRAMs, key features are made of ferromagnetic metallic alloys. Such metal-based devices make up the first--and most mature--of three categories of spintronics. In the second category, the spin-polarized currents flow in semiconductors instead of metals. Achieving practical spintronics in semiconductors would allow a wealth of existing microelectronics techniques to be co-opted and would also unleash many more types of devices made possible by semiconductors' high-quality optical properties and their ability to amplify both optical and electrical signals. Examples include ultrafast switches and fully programmable all-spintronics microprocessors. This avenue of research may lead to a new class of multifunctional electronics that combine logic, storage and communications on a single chip.

Researchers must answer several major questions before the second category of devices can take off as a viable industry: Can we devise economic ways to combine ferromagnetic metals and semiconductors in integrated circuits? Can we make semiconductors that are ferromagnetic at room temperature? What is an efficient way to inject spin-polarized currents, or spin currents, into a semiconductor? What happens to spin currents at boundaries between different semiconductors? How long can a spin current retain its polarization in a semiconductor?

Our own research groups are working on these questions but are keeping one eye also on the more distant and speculative quarry that is the third category of devices: ones that manipulate the quantum spin states of individual electrons. This category includes spintronic quantum logic gates that would enable construction of large-scale quantum computers, which would extravagantly surpass standard computers for certain tasks. A diverse assortment of exotic technologies is aimed toward that goal: ions in magnetic traps, "frozen" light, ultracold quantum gases called Bose-Einstein condensates and nuclear magnetic resonance of molecules in liquids--there are many ways to skin a quantum cat.

We believe it makes sense instead to build on the extensive foundations of conventional electronic semiconductor technology. Indeed, a recent series of unexpected discoveries appears to support our hunch that semiconductor spintronics provides a feasible path for developing quantum computers and other quantum information machines. Whether one looks at the near term for tomorrow's consumer electronics or at the more distant prospect of quantum computing, spintronics promises to be revolutionary.

Exploiting Spin Currents

An intuitive notion of how an electron's spin works is suggested by the name itself. Imagine a small electrically charged sphere that is spinning rapidly. The circulating charges on the sphere amount to tiny loops of electric current, which create a magnetic field similar to the earth's magnetic field. Scientists traditionally depict the rotation by a vector, or arrow, that points along the sphere's axis of rotation. Immersing the spinning sphere in an external magnetic field changes its total energy according to how its spin vector is aligned with the field.

In some ways, an electron is just like such a spinning sphere of charge--an electron has a quantity of angular momentum (its "spin") and an associated magnetism, and in an ambient magnetic field its energy depends on how its spin vector is oriented. But there the similarities end and the quantum peculiarities begin. Electrons seem to be ideal dimensionless points, not little spheres, so the simple picture of their spin arising from actual rotation doesn't work. In addition, every electron has exactly the same amount of spin, equal to one half the fundamental quantum unit of angular momentum. That property is hardwired into the mathematics that describes all the elementary particles of matter, a result whose significance and meaning are another story entirely. The bottom line is that the spin, along with a mass and a charge, is a defining characteristic of an electron.

In an ordinary electric current, the spins point at random and play no role in determining the resistance of a wire or the amplification of a transistor circuit. Spintronic devices, in contrast, rely on differences in the transport of "spin up" and "spin down" electrons. In a ferromagnet, such as iron or cobalt, the spins of certain electrons on neighboring atoms tend to line up. In a strongly magnetized piece of iron, this alignment extends throughout much of the metal. When a current passes through the ferromagnet, electrons of one spin direction tend to be obstructed. The result is a spin-polarized current in which all the electron spins point in the other direction.

A ferromagnet can even affect the flow of a current in a nearby nonmagnetic metal. For example, present-day read heads in computer hard drives use a device dubbed a spin valve, wherein a layer of a nonmagnetic metal is sandwiched between two ferromagnetic metallic layers. The magnetization of the first layer is fixed, or pinned, but the second ferromagnetic layer is not. As the read head travels along a track of data on a computer disk, the small magnetic fields of the recorded 1's and 0's change the second layer's magnetization back and forth, parallel or antiparallel to the magnetization of the pinned layer. In the parallel case, only electrons that are oriented in the favored direction flow through the conductor easily. In the antiparallel case, all electrons are impeded. The resulting changes in the current allow GMR read heads to detect weaker fields than their predecessors, so that data can be stored using more tightly packed magnetized spots on a disk, increasing storage densities by a factor of three.

Another three-layered device, the magnetic tunnel junction, has a thin insulating layer between two metallic ferromagnets. Current flows through the device by the process of quantum tunneling: a small number of electrons manage to jump through the barrier even though they are forbidden to be in the insulator. The tunneling current is obstructed when the two ferromagnetic layers have opposite orientations and is allowed when their orientations are the same.

Magnetic tunnel junctions form the basis of the MRAM chips mentioned earlier. Each junction can store one bit of data in the orientation of its unpinned ferromagnetic layer. That layer retains its magnetic state whether the power is on or off, at least until it is deliberately rewritten again.

Whereas the metallic spintronic devices just described provide new ways to store information, semiconductor spintronics may offer even more interesting possibilities. Because conventional semiconductors are not ferromagnetic, one might wonder how semiconductor spintronic devices can work at all. One solution employs a ferromagnetic metal to inject a spin-polarized current into a semiconductor.

In 1990 Supriyo Datta and Biswajit A. Das, then at Purdue University, proposed a design for a spin-polarized field-effect transistor, or spin FET. In a conventional FET, a narrow semiconductor channel runs between two electrodes named the source and the drain. When voltage is applied to the gate electrode, which is above the channel, the resulting electric field drives electrons out of the channel (for instance), turning the channel into an insulator. The Datta-Das spin FET has a ferromagnetic source and drain so that the current flowing into the channel is spin-polarized. When a voltage is applied to the gate, the spins rotate as they pass through the channel and the drain rejects these antialigned electrons.

A spin FET would have several advantages over a conventional FET. Flipping an electron's spin takes much less energy and can be done much faster than pushing an electron out of the channel. One can also imagine changing the orientation of the source or drain with a magnetic field, introducing an additional type of control that is not possible with a conventional FET: logic gates whose functions can be changed on the fly.

As yet, however, no one has succeeded in making a working prototype of the Datta-Das spin FET because of difficulties in efficiently injecting spin currents from a ferromagnetic metal into a semiconductor. Although this remains a controversial subject, recent optical experiments carried out at various laboratories around the world indicate that efficient spin injection into semiconductors can indeed be achieved by using unconventional materials, called magnetic semiconductors, that incorporate magnetism by doping the semiconductor crystals with atoms such as manganese.

Some magnetic semiconductors have been engineered to show ferromagnetism, providing a spintronic component called a gateable ferromagnet, which may one day play an important role in spin transistors. In this device, a small voltage would switch the semiconductor between nonmagnetic and ferromagnetic states. A gateable ferromagnet could in turn be used as a spin filter--a device that, when switched on, passes one spin state but impedes the other.

The filtering effect could be amplified by placing the ferromagnet in a resonant tunnel diode. Conventional resonant tunnel diodes allow currents to flow at a specific voltage, one at which the electrons have an energy that is resonant with the tunneling barrier. The version incorporating a ferromagnet would have a barrier with different resonant voltages for up and down spins.

The most exciting developments in semiconductor spintronics will probably be devices we have not imagined yet. A key research question for this second category of spintronics is how well electrons can maintain a specific spin state when traveling through a semiconductor or crossing from one material to another. For instance, a spin FET will not work unless the electrons remain polarized on entering the channel and after traveling to its far end.

The question of how fast spin polarization decays becomes all the more acute if one is to build a quantum computer based on electron spins. That application requires control over a property known as quantum coherence, in essence the pure quantum nature of all the computer's data-carrying components. Quantum data in semiconductors based on the charges of electrons tend to lose coherence, or dissipate, in mere picoseconds, even at cryogenic temperatures. Quantum data based on spin should be intrinsically more sturdy. Curiously enough, our research groups stumbled on significant basic results regarding coherent electron spins while doing experiments aimed at developing practical magnetic semiconductors.

A Pleasant Surprise

In 1997 at the University of California at Santa Barbara, we were experimenting on zinc selenide (ZnSe), a long-studied conventional semiconductor. The ZnSe was meant to be a control for an ongoing project studying magnetic semiconductors. In our experiment we used pulses of circularly polarized light to excite pools of electrons in the ZnSe into identical spin states. In a circularly polarized light wave, instead of oscillating in intensity, the electric and magnetic fields rotate in a circle, transverse to the direction of the light.

We sent the ultrashort (100-femtosecond) pulses horizontally through the semiconductor, exciting electrons into horizontal spin states, initially aligned with the light beam. In a vertical ambient magnetic field the electron spins precess--the direction of each electron's spin vector rotates in the horizontal plane, similar to how a tilted gyroscope precesses in the earth's gravitational field. The precession enables us to monitor how long these states remain coherent, but the horizontal spin state has another, more important property.

For a baseball, say, horizontal spinning is nothing special and is quite distinct from the two vertical modes of spinning. For electrons, however, the horizontal quantum spin states are actually coherent superpositions of the spin-up and spin-down states. In effect, such electrons are in both the up and the down state at the same time. This is precisely the kind of coherent superposition of states employed by quantum computers.

Each electron spin can represent a bit; for instance, a 1 for spin up and a 0 for spin down. With conventional computers, engineers go to great lengths to ensure that bits remain in stable, well-defined states. A quantum computer, in contrast, relies on encoding information within quantum bits, or qubits, each of which can exist in a superposition of 0 and 1. By having a large number of qubits in superpositions of alternative states, a quantum computer intrinsically contains a massive parallelism so that quantum algorithms can operate on many different numbers simultaneously.

Unfortunately, in most physical systems, interactions with the surrounding environment rapidly disrupt these superposition states. A typical disruption would effectively change a superposition of 0 and 1 randomly into either a 0 or a 1, a process called decoherence. State-of-the-art qubits based on the charge of electrons in a semiconductor remain coherent for a few picoseconds at best--and only at temperatures too low for practical applications. The rapid decoherence occurs because the electric force between charges is strong and long range. In traditional semiconductor devices, this strong interaction is beneficial, permitting delicate control of current flow with small electric fields. To quantum coherent devices, however, it is anathema.

Electron-spin qubits interact only weakly with the environment surrounding them, principally through magnetic fields that are nonuniform in space or changing in time. Such fields can be effectively shielded. The goal of our experiment was to create some of these coherent spin states in a semiconductor to see how long they could survive. The results are also useful for understanding how to design devices such as spin transistors that do not depend on maintaining and detecting the quantum coherence of an individual electron's spin.

Our experiment measured the decoherence rate by monitoring the precession of the spins. Each electron would continue precessing as long as its superposition remained coherent. We used weak pulses of light to monitor the precession, in effect obtaining stroboscopic images of the spin dynamics. As the electrons precessed, the measured signal oscillated, in magnitude; as coherence was lost, the amplitude of the oscillations fell to zero.

Much to our surprise, the optically excited spin states in ZnSe remained coherent for several nanoseconds at low temperatures--1,000 times as long as charge-based qubits. The states even survived for a few nanoseconds at room temperature. Subsequent studies of electrons in gallium arsenide (GaAs, a high-quality semiconductor commonly used in everyday applications such as cellul