Do we really need to keep on making circuits smaller? The miniaturization of silicon microelectronics seems so inexorable that the question seldom comes up--except maybe when we buy a new computer, only to find that it becomes obsolete by the time we leave the store. A state-of-the-art microprocessor today has on the order of 500 million transistors; by 2015 it could have nearly five billion. Yet within the next two decades this dramatic march forward will run up against scientific, technical and economic limits. A first reaction might be, So what? Arent five billion transistors enough already?
Yet when actually confronted with those limits, people will no doubt want to go beyond them. Those of us who work to keep computer power growing are motivated in part by the sheer challenge of discovering and conquering unknown territory. But we also see the potential for a revolution in medicine and so many other fields, as extreme miniaturization and new ways of building electronics enable people and machines to interact in ways that are not possible with existing technology.
As the word suggests, microelectronics involves components that measure roughly one micron on a side (although lately the components have shrunk to a size of about 50 nanometers). Going beyond microelectronics means more than simply shrinking components by a factor of 10 to 1,000. It also involves a paradigm shift for how we think about putting everything together.
Microelectronics and nanoelectronics both entail three levels of organization. The basic building block is usually the transistor or its nanoequivalent--a switch that can turn an electric current on or off as well as amplify signals. In microelectronics, transistors are made out of chunks of semiconductor--a material, such as impure silicon, that can be manipulated to flip between conducting and nonconducting states. In nanoelectronics, transistors might be organic molecules or nanoscale inorganic structures.
The next level of organization is the interconnection--the wires that link transistors together in order to perform arithmetic or logical operations. In microelectronics, wires are metal lines typically hundreds of nanometers to tens of microns in width deposited onto the silicon; in nanoelectronics, they are nanotubes or other wires as narrow as one nanometer.
At the top level is what engineers call architecture--the overall way the transistors are interconnected, so that the circuit can plug into a computer or other system and operate independently of the lower-level details. Nanoelectronics researchers have not quite gotten to the point of testing different architectures, but we do know what abilities they will be able to exploit and what weaknesses they will need to compensate for.
In other ways, however, microelectronics and nanoelectronics could not be more different. To go from one to the other, many believe, will require a shift from top-down manufacturing to a bottom-up approach. To build a silicon chip today, fabrication plants start with a silicon crystal, lay down a pattern using a photographic technique known as lithography, and etch away the unwanted material using acid or plasma. That procedure simply does not have the precision for devices that are mere nanometers in width. Instead researchers use the methods of synthetic chemistry to produce building blocks by the mole (6 10 pieces) and assemble a portion of them into progressively larger structures. Thus far the progress has been impressive. But if this research is a climb up Mount Everest, we have barely reached the base camp.
THE USE OF MOLECULES for electronic devices was suggested more than three decades ago in a seminal paper by Arieh Aviram of IBM and Mark A. Ratner, now at Northwestern University. By tailoring the atomic structures of organic molecules, they proposed, it should be possible to concoct a transistorlike device. But their ideas remained largely theoretical until a recent confluence of advances in chemistry, physics and engineering
Several groups have worked to evaluate Aviram and Ratners ideas, including teams at the University of California, Berkeley, the California Institute of Technology, Hewlett-Packard, Yale University and Rice University. They have demonstrated that sandwich structures containing thousands of molecules clustered together can carry electrons from one metal electrode to another. Each molecule is about 0.5 nanometer wide and one or more nanometers long. The research groups have shown that the clusters can behave as on/off switches and might thus be usable in computer memory; once on, they will stay on for 10 minutes or so. That may not sound like a long time, but computer memory typically loses its information instantly when the power is turned off; even when the power is on, the stored information leaks away and must be refreshed every 0.1 second or so.
The switching mechanism for the molecules has been the subject of considerable debate. Some researchers believe it involves oxidation reduction to induce conduction, whereas others have presented strong evidence for conduction through metal filaments that form reversibly between metal contacts separated by the molecules. This latter idea is a well-known phenomenon being investigated for nonvolatile memory in conventional microelectronics.
In the on position, the clusters of molecules may conduct electricity as much as 1,000 times better than in the off position. That ratio is actually rather low compared with that of typical semiconductor transistors, whose conductivity varies a millionfold. Researchers are now working to understand the switching process itself in order to improve the observed characteristics.
My own research group at Harvard University is one of several that have focused not on organic molecules but on long, thin, inorganic wires. One example is the carbon nanotube, which is typically about 1.4 nanometers in diameter. Not only can these nanoscale wires carry much more current, atom for atom, than ordinary metal wires, they also can act as tiny transistors. By functioning both as interconnections and as components, nanowires kill two birds with one stone. Another advantage is that they can exploit the same basic physics as standard silicon microelectronics, which makes them easier to understand and manipulate.
In 1997 Cees Dekkers group at the Delft University of Technology in the Netherlands and Paul L. McEuens group, then at the University of California, Berkeley, independently reported highly sensitive transistors made from metallic carbon nanotubes. These devices could be turned on and off by a single electron but required very low temperatures to operate.
More recent efforts have focused on semiconductor carbon nanotubes, which can function as field-effect transistors, as first shown by Dekker and his co-workers. In addition, Hongjie Dai of Stanford University, Ali Javey, now at U.C. Berkeley, and Phaedon Avouris of the IBM Thomas J. Watson Research Center have shown that nanotube transistors can exhibit extremely high performance--exceeding that of conventional transistors--and can be configured into basic circuits such as logic gates and ring oscillators. Finally, my group has demonstrated a very different type of switch, a nanoscale electromechanical relay made from carbon nanotubes.
A MAJOR PROBLEM with nanotubes is that they are difficult to make uniform. Because a slight variation in diameter or helicity can spell the difference between a conductor and a semiconductor, a large batch of nanotubes may contain only a few working devices. In April 2001 Avouris and his co-workers started with a mixture of conducting and semiconducting nanotubes and, by either applying a current between metal electrodes or reacting with gaseous etchant, selectively removed the conducting nanotubes until just semiconducting ones were left. The solution is only partial, however, because it leaves behind open space (thus reducing device density) where metallic nanotubes once were.
My group has also been working on a different type of nanoscale wire, which we term the semiconductor nanowire. It is about the same size as a carbon nanotube, but its composition is easier to control precisely. To synthesize these wires, we start with a metal catalyst, which defines the diameter of the growing wire and serves as the site where molecules of the desired material tend to collect. As the nanowires grow, we incorporate chemical dopants (impurities that add or remove electrons), thereby controlling whether the nanowires are n-type (having extra electrons) or p-type (having a shortage of electrons or, equivalently, a surfeit of positively charged holes).
The availability of n- and p-type materials, which are the essential ingredients of transistors, diodes and other electronic devices, has opened up a new world for us. We have assembled a wide range of devices, including both major types of transistors (field-effect and bipolar); inverters, which transform a 0 signal to a 1; and light-emitting diodes, which pave the way for optical interconnections. Our bipolar transistors were the first molecular-scale devices ever to amplify a current. Xiangfeng Duan, a former member of my lab, was the first to assemble memory from crisscrossing n- and p-type nanowires. The memory can store information for 10 minutes or longer by trapping charge at the interface between the crossing nanowires [see box below].
Breaking the Logjam
BUILDING UP AN ARSENAL of molecular and nanoscale devices is just the first step. Interconnecting and integrating these devices is perhaps the much greater challenge. First, the nanodevices must be connected to molecular-scale wires. To date, organic-molecule devices have been hooked up to conventional metal wires created by lithography. It will not be easy to substitute nanowires, because we do not know how to make a good electrical connection without ruining these tiny wires in the process. Using nanowires and nanotubes both for the devices and for the interconnections has, however, been shown to solve that problem.
Second, once the components are attached to nanowires, the wires themselves must be organized into, for example, a two-dimensional array. Duan and another member of my team, Yu Huang, made the first significant breakthrough: they assembled nanocircuits by means of fluid flows. Just as sticks and logs can flow down a river, nanoscale wires can be drawn into parallel lines using fluids.
More recent work by members of my lab has expanded these basic ideas in several very significant directions that bode well for large-scale integration and manufacturing. First, Song Jin and Dongmok Whang showed that the Langmuir-Blodgett technique could be used to organize nanowires en masse on the surface of water and then transfer them at controlled density and orientation to centimeter-scale substrates. Since then, Javey and SungWoo Nam have shown that nanowires can be directly printed onto moderate-scale wafers with controlled orientation and density, and Guihua Yu in my lab and Anyuan Cao of the University of Hawaii at Manoa have blown polymer bubble nanowire films that can be transferred to commercial-scale wafers and even very large flexible substrates.
These processes create interconnections in the direction of alignment, thus yielding parallel nanowire arrays. To add wires in other directions, we repeat the process, building up additional layers of nanowires. For instance, to produce a right-angle grid, we first lay down a series of parallel nanowires, then rotate the direction by 90 degrees and lay down another series. By using wires of different compositions for each layer, we can rapidly assemble an array of functional nanodevices using equipment not much more sophisticated than that in a high school chemistry lab. A grid of diodes, for example, consists of a layer of conducting nanotubes above a layer of semiconducting nanotubes, or a layer of n-type nanowires atop a layer of p-type nanowires. In both cases, each junction serves as a diode.
Intimately linked to all these efforts is the development of architectures that best exploit the unique features of nanoscale devices and the capabilities of bottom-up assembly. Although we can make unfathomable numbers of dirt-cheap nanostructures, the devices are much less reliable than their microelectronic counterparts, and our capacity for assembly and organization is still quite primitive.
In collaboration with Andr DeHon of the University of Pennsylvania, my group has been working on highly regular architectures based on crossed-nanowire arrays that can be generalized for universal computing machines. For memory, the architecture starts with a two-dimensional array of crossed nanowires or suspended electromechanical switches in which one can store information at each cross point. The same basic architecture is being pursued by researchers at Caltech and Hewlett-Packard, and it resembles the magnetic-core memory that was common in computers of the 1950s and 1960s.
Law of Large Numbers
TO OVERCOME the unreliability of individual nanodevices, we may rely on sheer numbers--the gizmos are so cheap that plenty of spares are always available. Researchers who work on defect tolerance have shown that computing is possible even if many of the components fail, although identifying and mapping the defects can be slow and time-consuming. Ultimately we hope to partition the enormous arrays into subarrays whose reliability can be easily monitored. The optimum size of these subarrays will depend on the defect levels typically present in molecular and nanoscale devices.
Another significant hurdle faced by nanoelectronics is bootstrapping. How do engineers get the circuit to do what they want it to? In microelectronics, circuit designers work like architects: they prepare a blueprint of a circuit, and a fabrication plant builds it. In nanoelectronics, designers will have to work like computer programmers. A fabrication plant will create a raw nanocircuit--billions on billions of devices and wires whose functioning is rather limited. From the outside, it will look like a lump of material with a handful of wires sticking out. Using those few wires, engineers will somehow have to configure those billions of devices. Such challenges are what keeps me tremendously excited about the field as a whole.
Even before we solve these problems, nanodevices may have useful applications. For example, Gengfeng Zheng in my group has used semiconductor nanowires as ultrasensitive detectors. This technology has even been used to detect single virus particles and single pieces of DNA and, with the assembly of many sensors, could sequence the entire human genome on a single chip. The technology could also serve in minimally invasive medical devices and, as Fernando Patolsky, now at Tel Aviv University, and Brian Timko in my group have shown, could be used to build artificial synapses or two-way interfaces to live neurons.
Although substantial work remains before nanoelectronics makes its way into computers, this goal now seems less hazy than it was even a year ago. As we gain confidence, we will learn not just to shrink digital microelectronics but also to go where no digital circuit has gone before. We might assemble and interconnect multiple layers of unique functional building blocks to enable truly 3-D computational engines and nanoelectronic systems. Nanoscale devices that exhibit quantum phenomena, for example, could be exploited in quantum encryption and quantum computing. And building nanoelectronic devices on biocompatible polymers could usher in a totally new form of smart tissue or hybrid bionanoelectronic brains. The richness of the nanoworld will change the macroworld.
CHARLES M. LIEBER spent much of his childhood building--and breaking--stereos, cars and model airplanes. He is now the Mark Hyman Professor of Chemistry at Harvard University, where he directs a group of 25 undergraduate, graduate and postdoctoral researchers who focus on nanoscale science and technology. Lieber founded NanoSys, Inc., with Larry Bock of CW Ventures and Hongkun Park of Harvard in 2001. Work from Liebers laboratory also helped to form another nanotech company, Nantero, focused on nonvolatile memory. Lieber is currently working to start a new company focused on nanoelectronics for personalized medicine.