A programmable logic gate that uses changes in magnetic fields to alter its behavior could provide a low-energy alternative to circuits based on the transistors that are used in practically all computers today.
Rather than take the approach of pure 'spintronic' computing devices [see "Computing with Magnets," Communications of the ACM, March 2012], the device developed at the Spin Convergence Center in Seoul, South Korea, uses magnetism to enhance the operation of semiconductor-based electronics.
Jinki Hong, professor at the Korea Institute of Science and Technology, who led the team at the Spin Convergence Center, says the aim of the work there is to develop 'chameleon processors.' "These are devices that function as both programmable logic and nonvolatile memoryretaining information even when not powered," says Hong.
By incorporating nonvolatile memory into their structure, devices like those developed in Seoul could slash the amount of energy that computers need to process data. Although it uses current to pass information between gates, switching the polarity of magnetic fields around the device performs logic functions. Two of these devices wired in series provide the same kinds of logic operation for which 10 or more transistors are needed today.
Sayeef Salahuddin, assistant professor at the University of California, Berkeley, says the motivation for developing spintronic and hybrid devices such as Hong's "is the realization that transistors are becoming excessively power-hungry. A more energy-efficient solution, even if low-performance, could make a difference."
The problem with the transistors used in all computers today is that they leak even when turned off. Planar transistors made on a 22nm processwhich is currently the most advanced in productionhave a ratio of on-state to off-state current flow of just two orders of magnitude. A decade ago, this ratio was close to a million. Designing ICs to support low leakage levels calls for devices that operate with a low on-state current, which generally results in low performance.
In 2002, Kailash Gopalakrishnan and colleagues at Stanford University proposed the impact-ionization switch as an alternative to the conventional transistor that could boost the current ratio to 10 million. The heart of the impact-ionization switch is a diode, rather than a transistor.
Diodes pass current in one direction, but almost completely block it in the reverse direction. A common application for this is in power supplies, where they are used to help convert alternating current (AC) into direct current (DC). However, there is a limit to how effectively they can block current.
Even when reverse-biased, a diode will pass very low levels of current that are largely the result of electrons absorbing enough thermal energy to break free from the bond to a nucleus and then being swept through the device in the direction of the applied voltage. These carriers are often involved in collisions with other electrons; if energetic enough, these electrons, in turn, can break loose.
A high voltage can generate an electric field that will accelerate electrons to high speeds, enough to free any electrons they strike. In turn, these cause even more impact ionization. The result is an avalanche that causes the diode to start passing high levels of current in the wrong direction.
"Impact ionization is widely used in semiconductor devices, and many kinds of devices based on this technology have been already commercialized. For example, the avalanche photodiode is used for optical communication," says Hong. "Our device can be considered to be a magnetic-field version of an electrical diode."
The key to the operation of the Seoul device lies in its application of magnetoresistance, a property that revolutionized hard-drive technology 20 years ago. Magnetoresistance is the tendency for the resistance to electrical current to increase with changes in the strength and direction of a magnetic field. Traditionally, the effect was found strongly only in magnetic materials.
Five years ago, Bert Koopmans and colleagues at the Eindhoven University of Technology stumbled across a large magnetoresistive effect in the non-magnetic material silicon. "We saw very large effects that resulted from impact ionization."
The Seoul team used the non-magnetic material indium antimonide (InSb)which has been proposed as a possible high-speed successor to silicon for future transistorsbecause its properties "lead to easy control of impact ionization by magnetic fields," Hong says.
The experiment, reported in the science journal Nature earlier this year, used externally applied magnetic fields to control the motion of electrons along a channel of InSb. Underneath this layer was a thin strip of InSb doped to have a low electron concentrationif electrons moved into this region, they would have a high probability of forming bonds with nuclei there and drop out of the free-carrier pool, preventing them from ionizing any more electrons. A magnetic field applied in one direction provided the necessary force, with the result of turning the device off. When the field reversed, the electrons were able to move from one end of the device to the other more freely and take part in impact ionization.
Building magnetic-field control into the device itself is a major stumbling block, but Hong envisages tiny ferro-magnets being deposited alongside the InSb diode channels, pointing to work performed in the lab five years ago. "We demonstrated the fabrication of 1m magnets," he says, noting that the techniques used in magnetic memories that are just beginning to move into production could be used to switch field directions as needed.
Michael Delmo, a postdoctoral fellow at Osaka University who has studied transport mechanisms that lead to magnetically controlled impact ionization, says incorporating magnetic elements into what is primarily an electronic device is a realistic proposition. "But this technology is in its infancy, and many things about it are still unknown," he adds.
In addition, the magnetically controlled impact-ionization device is large compared with today's logic transistors, and its current ratio is less than one order of magnitude. The question is whether the device will scale down easily and fulfill the promise of impact ionization devices to provide significant energy advantages over conventional transistors.
The magnetoresistive device may lend itself to plastic electronics, in which organic polymers are printed onto a surface to form circuits.
Koopmans says when the Eindhoven team found silicon could display large magnetoresistance, "It was a new and very interesting effect. But, at the time, I was not sure whether it would be easy to handle in miniaturized nanodevices. We had been working on rather large devices."
Chiara Ciccarelli, a researcher at the University of Cambridge who has investigated magnetoresistance in silicon, says as devices get smaller, "although there is no fundamental limit to room-temperature magnetoresistance, its magnitude decreases."
Ciccarelli points out that high-mobility materials, of which InSb is an example, show stronger magnetoresistance than silicon, which should help in smaller devices. Scaling could compensate in other ways. "As the length of the channel decreases, the electric field increases," she says. "That suggests miniaturization could play a positive role in the magnetoresistance of these devices. However, this is not what has been observed in the devices studied so far."
Even if magnetoresistance can be maintained at a high-enough level, there is a limit to how small the device can be made before impact ionization itself ceases.
"The minimum length of the channel is determined by a 'dead space,' the distance a carrier travels before acquiring enough energy from the electric field to participate in impact ionization," says Hong. This dead space may not shrink past 20nmthe length of the channel in today's most advanced transistors.
Even if it cannot be made as small as a logic transistor, the magnetically controlled device could still have a future. The nonvolatile nature of spintronic devices also means that for systems that are active only intermittentlysuch as smart sensorsthe energy savings achieved by only having to power a circuit while it is processing could be immense, and outweigh silicon's likely advantage in size.
The magnetoresistive device may lend itself to plastic electronics, in which organic polymers are printed onto a surface to form circuits. These processes can be performed at very low temperatures and with much cheaper equipment than that used in conventional semiconductor fabs.
Delmo says the electron dynamics are different in organic semiconductors from those in silicon or InSb: "If these mechanisms do not depend on the device size, organic semiconductors could be strong candidates for magnetoresistive-based semiconductor logic technology."
Joo S., Kim, T., Shin, S.H., Lim J.Y., Hong, J., Song J.D., Chang J., Lee, H.W., Rhie, K., Han, S.H., Shin, K.H. and Johnson, M.
Magnetic-field-controlled reconfigurable semiconductor logic, Nature 494, 72, Feb. 7, 2013
Schoonus, J.J.H., Bloom, F.L., Wagemans, W., Swagten, H.J.M., Koopmans, B.
Extremely large magnetoresistance in boron-doped silicon. Physics Review Letters 100, 127202, Mar. 27, 2008
Ciccarelli, C., Park, B.G., Ogawa, S., Ferguson, A.J., and Wunderlich, J.
Gate-controlled magnetoresistance in a silicon metal-oxide-semiconductor field-effect transistor. Applied Physics Letters 97, 082106, Aug. 25, 2010.
Datta, S., Behin-Aein B., and Salahuddin, S.
Non-volatile spin switch for Boolean and non-Boolean Logic, Applied Physics Letters 101, 252411, Dec. 20, 2012. Applied Physics Letters 95, 132106, Sep. 30, 2009.
Figure. When magnetic fields are aligned to turn on the programmable logic gate, the flow of electrons results in an avalanche. In the off-state, conduction-band electrons are pulled toward the lower p-type layer and recombine with nuclei, moving them out of the conduction band.
©2013 ACM 0001-0782/13/09
Permission to make digital or hard copies of part or all of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and full citation on the first page. Copyright for components of this work owned by others than ACM must be honored. Abstracting with credit is permitted. To copy otherwise, to republish, to post on servers, or to redistribute to lists, requires prior specific permission and/or fee. Request permission to publish from email@example.com or fax (212) 869-0481.
The Digital Library is published by the Association for Computing Machinery. Copyright © 2013 ACM, Inc.
No entries found