Teams of researchers around the world are making steady progress toward the development of computers one million times more efficient than those in use today. These computers would use logic circuits and memory built from nanoscale ferromagnetic particles, not the silicon-based transistors that are at the heart of today’s computers.
Although the results so far are promising, widespread commercialization of magnetic computers lies years or even decades in the future, these researchers say.
The effort aims to solve what in recent years has become one of the biggest challenges facing computer designers. For decades the holy grail in processor research and development has been to boost speed by increasing the density of transistors on a chip. That worked spectacularly well until the density became so great that power consumption, and the resulting heat, became unacceptable. Moore’s Law seemed near its end.
Conventional CMOS microprocessors work by the flow of electrical current, which inherently wastes energy as heat even when they are not computing. But nanomagnets can be combined into logic circuits that require no moving electrons. Researchers at the University of California, Berkeley recently showed that such logic devices could be made to operate at the maximum level of efficiency allowed by the second law of thermodynamics. That bound, called the Landauer limit, was calculated 50 years ago by physicist Rolf Landauer.
Magnetic computers have another desirable property. Because the magnets retain their state even when power is off, no time or energy is wasted in booting up. This nonvolatility property and high efficiency holds great appeal for the U.S. Defense Advanced Research Projects Agency, which in late 2010 awarded contracts to a number of university-industry teams to develop nonvolatile logic, including $9.9 million to a team led by the University of Notre Dame and $8.4 million to one led by the University of California, Los Angeles (UCLA).
Revisiting an Old Idea
Wolfgang Porod, a professor of electrical engineering at Notre Dame and leader of that team’s efforts, is building on work he started 15 years ago in Quantumdot Cellular Automata (QCA). A quantum dot contained a single electron, and information could be encoded by the arrangement and manipulation of electrons in these dots in cellular arrays. While promising on theoretical grounds, the idea proved impractical because the positions of individual electrons were too vulnerable to various types of disturbances, except at extremely low temperatures.
But now Porod is applying the same concept with dots replaced by nano-magnets, which are far more stable and robust. “We have demonstrated that you can, in fact, do logic with different arrangements of magnets, and the corresponding different physical interactions give you different logic functionalities,” Porod says. The simplest of these nanomagnetic devices uses five magnets to build a “majority” logic gate, with three inputs, one output, and one magnet in the center. The device returns a value of zero when two or more of the inputs are zero and one otherwise.
Berkeley researchers are also building logic gates out of nanomagnets. Brian Lambson, a graduate student in electrical engineering and computer science, says extensive simulations have shown that the 100nm x 200nm devices could perform the most basic one-bit operation using 18 milli-electron volts at room temperature, or one million times less than would be required by a conventional microprocessor. Lambson says his group has built such devices, but that it is not yet possible to measure their power consumption.
While nanomagnets can be combined in various ways to perform Boolean operations, individually they can act as one-bit memory cells. “One of the beautiful things about magnetic computing is that the logic devices would be intrinsically memory devices,” says Jeffrey Bokor, a Berkeley professor of electrical engineering and computer science. “You would probably want to have memory more distributed, as opposed to having a main memory with big bandwidth in and out of the logic.”
Such interconnects are presently a big part of the power problem. “With today’s CMOS,” Bokor explains, “the amount of energy spent in interconnect is a significant amount of the total energy used in a large chip like a microprocessor.” The problem is that the transistors require much higher voltage to switch than would be required solely for reliable, noise-free communication across the interconnects. So while magnetic computers might use more or less conventional electrical interconnects, the processors would require far less power than those in use today. And since power consumption in the interconnects varies inversely with the square of the voltage, reducing voltage a hundred-fold would reduce power consumption by a factor of 10,000 in the interconnects. “The problem would be solved, and the interconnects would work just fine,” Bokor says.
The Input/Output Question
A computer with nanomagnetic memory and processors and conventional electrical interconnects would still require a mechanism for switching the magnets and reading out their states, and the input/output question is a subject of intense research. The Berkeley magnets at present are switched by the application of electrical currents to generate external magnetic fields around the nanomagnets, and that requires a lot of current.
While nanomagnets can be combined in various ways to perform Boolean operations, individually they can act as one-bit memory cells.
But Bokor’s team is experimenting with exotic new materials, called multiferroics, developed by Ramamoorthy Ramesh, a materials scientist at Berkeley. The multiferroics act as a dielectric between a nanomagnet and a normal metal layer. When voltage is applied between the magnet and the metal, as in a capacitor, the magnet is made to switch. “No current flows, and this can be done at very low voltages,” Bokor says.
Meanwhile, Porod’s team is working with IBM on an I/O technique originally developed for nonvolatile magnetic RAM (MRAM). The nanomagnets in his logic devices sit on wires and are switched by the application of current pulses, very much as in one of the original implementations of MRAM. The currents are able to switch single magnets or groups of magnets, thus leveraging their energy. Other ways of switching the magnets also are under investigation, including spin torque transfer at Notre Dame and multiferroics at Berkeley.
At UCLA, researchers have developed spin-wave buses, or magnetic nanoscale wires, which form interconnects that do not require the flow of current. The buses propagate waves of magnetization, which are waves of electronic spins, through nanomagnetic lattices. The waves can control the switching of nonvolatile nanomagnetic memory bits that are combined to perform logical operations. The waves are generated by the application of voltages, but no current is conducted.
The initial goal in the UCLA project, says Kang Wang, director of the university’s Device Research Laboratory, is to produce two-bit adder devices able to operate at 10ns per operation and 100 attojoule per operation (about 600eV), or 100 times more efficient than CMOS. Unlike the approaches at Notre Dame and Berkeley, which uses discrete magnets, the UCLA method uses wave propagation through a continuous magnetic medium. Both can be integrated with CMOS, but spin-wave propagation is likely to be more scalable at very small dimensions, Wang says.
Efficiency at a Price
Dozens of universities, research consortia, and companies are doing research in magnetic computing. “What they have in common is elements that are nanoscale magnetic dots fabricated by patterning methods out of magnetic thin-film materials,” Bokor says. “They are coming up with different schemes for putting them together, but there is not yet an agreed-upon best approach.”
While nanomagnets score a home run for efficiency, the technology now comes at a price in terms of speed and reliability. “More than the energy efficiency, these are preventing us from achieving any sort of commercial product,” Lambson says. As there is a trade-off among efficiency, speed, and reliability, a 10,000- or even 1,000-fold improvement in efficiency—rather than the theoretically possible one million times improvement—would be a huge advance and might be possible with good speed and reliability, he says.
Nanomagnets score a home run for efficiency, but the technology now comes at a price in terms of speed and reliability.
That type of compromise is the likely path to commercialization, according to Lambson. Still, he says “we are thinking well beyond five years, maybe 10 to 20 years.”
So might the ultra-efficient nano-magnet one day replace the power-hungry transistor? Bokor won’t predict that, but says “we are working toward a vision in which magnetic computers could become mainstream.” There are many emerging applications, such as sensor and body-area networks, where individual nodes may be relatively slow but not 100% reliable.
For example, nanomagnetic processors could be used to build wireless networks of autonomous sensors of environmental conditions or for the detection of sound or vibration in law enforcement or national security applications. Such a network could consist of thousands of processing nodes, each attached to several sensors and each with its own two-way communication ability, and the super-efficiency of the processors would enable them to work without batteries or other bulky power mechanisms.
“Maybe they operate on energy they scrounge out of the environment,” Bokor says. “For applications like that, a completely nonvolatile logic technology has a major advantage—the fact that it doesn’t have to reboot to wake up and do some function before going back to sleep.”
Further Reading
Carlton, D., Emley, N.C., Tuchfeld, E. and Bokor, J.
Simulation studies of nanomagnet-based logic architecture, Nano Letters 8, 12, Dec. 2008.
Imre, A., Csaba, G., Ji, L., Orlov, A., Bernstein, G.H. and Porod, W.
Majority logic gate for magnetic quantum-dot cellular automata, Science 311, 5758, Jan. 13, 2006.
Lambson, B., Carlton, D. and Bokor, J.
Exploring the thermodynamic limits in integrated systems: magnet memory, nanomagnetic logic and the Landauer limit, Physical Review Letters 107, 1, July 1, 2011.
Wang, K.L. and Ovchinnikov, I.V.
Nanoelectronics and nanospintronics: Fundamentals and materials perspective, Materials Science Forum, 608, Dec. 2008.
Wu, Y., Bao, M., Khitun, A., Kim, J.-Y., Hong, A., Wang, K.L.
A three-terminal spin-wave device for logic applications, Journal of Nanoelectronics and Optoelectronics 4, 3, Dec. 2009.
Join the Discussion (0)
Become a Member or Sign In to Post a Comment