In the late 1950s visionary physicist Richard Feynman issued a public challenge by offering $1000 to the first person to create an electrical motor "smaller than 1/64th of an inch". Much to Feynman's consternation the young man who met this challenge, William McLellan, did so by investing many tedious and painstaking hours building the device by hand using tweezers and a microscope (figure 1).
McLellan's motor now sits in a display case at the California Institute of Technology and has long since ceased to spin. Meanwhile, in the field that Feynman hoped to incite, the wheels are turning - both figuratively and literally - in many university and corporate laboratories, and even on industrial production lines. Indeed, the field of microelectromechanical systems (MEMS), which became firmly established in the mid-1980s, has now matured to the point where we can be rather blasé about the mass production of diminutive motors that are hundreds of times smaller than McLellan's original. Along the way the MEMS community has developed some truly intriguing products - from digital projectors that contain millions of electrically driven micromirrors to microscale motion sensors that sit in cars ready to deploy airbags (figure 2).
A whole new realm of interconnected microsensors and instruments is emerging from the minds and laboratories of the scientists and engineers engaged in this research. Their devices are being assigned to a myriad of remote outposts, from the depths of the sea and the Earth's crust, to the far-flung regions of space and distant planets. Moreover, the robustness and low cost of such remote microsensors are helping to provide an avalanche of information about our physical surroundings.
MEMS represent the marriage of semiconductor processing to mechanical engineering - at a very small scale. And it is a field that has grown enormously during the past decade. Numerous companies - from the semiconductor giants to fledgling start-ups - are all now scrambling for a piece of the action at the microscale. Yet very little has been done with MEMS at sizes below one micrometre. This stands in striking contrast to the recent developments in mainstream microelectronics where chips are now mass-produced with features as small as 0.18 microns. Indeed, SEMATECH - a think-tank for a consortia of semiconductor companies in the US - predicts that the minimum feature size will shrink to 70 nm by 2010.
In the face of these achievements, and the advances they are expected to bring to mainstream electronics, the time is ripe for a concerted exploration of nanoelectromechanical systems (NEMS) - i.e. machines, sensors, computers and electronics that are on the nanoscale. Such efforts are under way in my group at Caltech, and in several others around the world. The potential payoffs are likely to be enormous and could benefit a diverse range of fields, from medicine and biotechnology to the foundations of quantum mechanics. In this article I highlight a few of the most exciting promises of NEMS, and the challenges that must be faced to attain them.
What is an electromechanical system?
One of the earliest reported electromechanical devices was built in 1785 by Charles-Augustin de Coulomb to measure electrical charge. His electrical torsion balance consisted of two spherical metal balls - one of which was fixed, the other attached to a moving rod - that acted as capacitor plates, converting a difference in charge between them to an attractive force. The device illustrates the two principal components common to most electromechanical systems irrespective of scale: a mechanical element and transducers.
The mechanical element either deflects or vibrates in response to an applied force. To measure quasi-static forces, the element typically has a weak spring constant so that a small force can deflect it by a large amount. Time-varying forces are best measured using low-loss mechanical resonators that have a large response to oscillating signals with small amplitudes.
Many different types of mechanical elements can be used to sense static or time-varying forces. These include the torsion balance (used by Coulomb), the cantilever (now ubiquitous in scanning probe microscopy) and the "doubly clamped" beam, which is fixed at both ends. In pursuit of ultrahigh sensitivity, even more intricate devices are used, such as compound resonant structures that possess complicated transverse, torsional or longitudinal modes of vibration. These complicated modes can be used to minimize vibrational losses, in much the same way that the handle of a tuning fork is positioned carefully to reduce losses.
The transducers in MEMS and NEMS convert mechanical energy into electrical or optical signals and vice versa. However, in some cases the input transducer simply keeps the mechanical element vibrating steadily while its characteristics are monitored as the system is perturbed. In this case such perturbations, rather than the input signal itself, are precisely the signals we wish to measure. They might include pressure variations that affect the mechanical damping of the device, the presence of chemical adsorbates that alter the mass of the nanoscale resonator, or temperature changes that can modify its elasticity or internal strain. In these last two cases, the net effect is to change the frequency of vibration.
In general, the output of an electromechanical device is the movement of the mechanical element. There are two main types of response: the element can simply deflect under the applied force or its amplitude of oscillation can change (figure 3). Detecting either type of response requires an output or readout transducer, which is often distinct from the input one. In Coulomb's case, the readout transducer was "optical" - he simply used his eyes to record a deflection. Today mechanical devices contain transducers that are based on a host of physical mechanisms involving piezoelectric and magnetomotive effects, nanomagnets and electron tunnelling, as well as electrostatics and optics.
The benefits of nanomachines
Nanomechanical devices promise to revolutionize measurements of extremely small displacements and extremely weak forces, particularly at the molecular scale. Indeed with surface and bulk nanomachining techniques, NEMS can now be built with masses approaching a few attograms (10-18 g) and with cross-sections of about 10 nm (figure 4). The small mass and size of NEMS gives them a number of unique attributes that offer immense potential for new applications and fundamental measurements.
Mechanical systems vibrate at a natural angular frequency, w0, that can be approximated by w0 = (keff/meff)1/2, where keff is an effective spring constant and meff is an effective mass. (Underlying these simplified "effective" terms is a complex set of elasticity equations that govern the mechanical response of these objects.) If we reduce the size of the mechanical device while preserving its overall shape, then the fundamental frequency, w0, increases as the linear dimension, l, decreases. Underlying this behaviour is the fact that the effective mass is proportional to l3, while the effective spring constant is proportional to l. This is important because a high response frequency translates directly to a fast response time to applied forces. It also means that a fast response can be achieved without the expense of making stiff structures.
Resonators with fundamental frequencies above 10 GHz (1010 Hz) can now be built using surface nanomachining processes involving state-of-the-art nanolithography at the 10 nm scale (see box). Such high-frequency mechanical devices are unprecedented and open up many new and exciting possibilities. Among these are ultralow-power mechanical signal processing at microwave frequencies and new types of fast scanning probe microscopes that could be used in fundamental research or perhaps even as the basis of new forms of mechanical computers.
A second important attribute of NEMS is that they dissipate very little energy, a feature that is characterized by the high quality or Q factor of resonance. As a result, NEMS are extremely sensitive to external damping mechanisms, which is crucial for building many types of sensors. In addition, the thermomechanical noise, which is analogous to Johnson noise in electrical resistors, is inversely proportional to Q. High Q values are therefore an important attribute for both resonant and deflection sensors, suppressing random mechanical fluctuations and thus making these devices highly sensitive to applied forces. Indeed, this sensitivity appears destined to reach the quantum limit.
Typically, high-frequency electrical resonators have Q values less than several hundred, but even the first high-frequency mechanical device built in 1994 by Andrew Cleland at Caltech was 100 times better. Such high quality factors are significant for potential applications in signal processing.
The small effective mass of the vibrating part of the device - or the small moment of inertia for torsional devices - has another important consequence. It gives NEMS an astoundingly high sensitivity to additional masses - clearly a valuable attribute for a wide range of sensing applications. Recent work by Kamil Ekinci at Caltech supports the prediction that the most sensitive devices we can currently fabricate are measurably affected by small numbers of atoms being adsorbed on the surface of the device.
Meanwhile, the small size of NEMS also implies that they have a highly localized spatial response. Moreover, the geometry of a NEMS device can be tailored so that the vibrating element reacts only to external forces in a specific direction. This flexibility is extremely useful for designing new types of scanning probe microscopes.
NEMS are also intrinsically ultralow-power devices. Their fundamental power scale is defined by the thermal energy divided by the response time, set by Q/wo. At 300 K, NEMS are only overwhelmed by thermal fluctuations when they are operated at the attowatt (10-18 W) level. Thus driving a NEMS device at the picowatt (10-12 W) scale provides signal-to-noise ratios of up to 106. Even if a million such devices were operated simultaneously in a NEMS signal processor, the total power dissipated by the entire system would still only be about a microwatt. This is three or four orders of magnitude less than the power consumed by conventional electronic processors that operate by shuttling packets of electronic charge rather than relying on mechanical elements.
Another advantage of MEMS and NEMS is that they can be fabricated from silicon, gallium arsenide and indium arsenide - the cornerstones of the electronics industry - or other compatible materials. As a result, any auxiliary electronic components, such as transducers and transistors, can be fabricated on the same chip as the mechanical elements. Patterning NEMS so that all the main internal components are on the same chip means that the circuits can be immensely complex. It also completely circumvents the insurmountable problem of aligning different components at the nanometre scale.
Challenges for NEMS
Processes such as electron-beam lithography and nanomachining now enable semiconductor nanostructures to be fabricated below 10 nm. It would appear that the technology exists to build NEMS. So what is holding up applications? It turns out that there are three principal challenges that must be addressed before the full potential of NEMS can be realized: communicating signals from the nanoscale to the macroscopic world; understanding and controlling mesoscopic mechanics; and developing methods for reproducible and routine nanofabrication.
NEMS are clearly very small devices that can deflect or vibrate within an even smaller range during operation. For example, the deflection of a doubly clamped beam varies linearly with an applied force only if it is displaced by an amount that typically corresponds to a few per cent of its thickness. For a beam 10 nm in diameter, this translates to displacements that are only a fraction of a nanometre. Building transducers that are sensitive enough to allow information to be transferred accurately at this scale requires reading out positions with a far greater precision. A further difficulty is that the natural frequency of this motion increases with decreasing size. So the ideal NEMS transducer must ultimately be capable of resolving displacements in the 10-15-10-12 m range and be able to do so up to frequencies of a few gigahertz. These two requirements are truly daunting, and much more challenging than those faced by the MEMS community so far.
To compound the problem, some of the transducers that are mainstays of the micromechanical realm are not applicable in the nanoworld. Electrostatic transduction, the staple of MEMS, does not scale well into the domain of NEMS. Nanoscale electrodes have capacitances of about 10-18 farad and less. As a result, the many other, unavoidable parasitic impedances tend to dominate the "dynamic" capacitance that is altered by the device motion.
Meanwhile optical methods, such as simple beam-deflection schemes or more sophisticated optical and fibre-optic interferometry - both commonly used in scanning probe microscopy to detect the deflection of the probe - generally fail beyond the so-called diffraction limit. In other words, these methods cannot easily be applied to objects with cross-sections much smaller than the wavelength of light. For fibre-optic interferometry, this breakdown can occur even earlier, when devices are shrunk to a fraction of the diameter of the fibre.
Conventional approaches thus appear to hold little promise for high-efficiency transduction with the smallest of NEMS devices. Nonetheless, there are a host of intriguing new concepts in the pipeline. These include techniques that are based on integrated near-field optics, nanoscale magnets, high-electron-mobility transistors, superconducting quantum interference devices and single-electron transistors - to name just a few. Discussion of these topics is, unfortunately, beyond the scope of this article (see Roukes in further reading).
The role of surface physics
One of the keys to realizing the potential of NEMS is to achieve ultrahigh quality factors. This overarching theme underlies most areas of research, with the possible exception of non-resonant applications. However, both intrinsic and extrinsic properties limit the quality factor in real devices. Defects in the bulk material and interfaces, fabrication-induced surface damage and adsorbates on the surfaces are among the intrinsic features that can dampen the motion of a resonator.
Fortunately, many of these effects can be suppressed through a careful choice of materials, processing and device geometry. Extrinsic effects - such as air resistance, clamping losses at the supports and electrical losses mediated through the transducers - can all be reduced by careful engineering. However, certain loss mechanisms are fundamental and ultimately limit the maximum attainable quality factors. These processes include thermoelastic damping that arises from inelastic losses in the material.
One aspect in particular looms large: as we shrink MEMS towards the domain of NEMS, the device physics becomes increasingly dominated by the surfaces. We would expect that extremely small mechanical devices made from single crystals and ultrahigh-purity heterostructures would contain very few defects, so that energy losses in the bulk are suppressed and high quality factors should be possible.
For example, Robert Pohl's group at Cornell University, and others, has shown that centimetre-scale semiconductor MEMS can have Q factors as high as 100 million at cryogenic temperatures. But my group at Caltech has shown repeatedly over the past seven years that this value decreases significantly - by a factor of between 1000 and 10 000 - as the devices are shrunk to the nanometre scale. The reasons for this decrease are not clear at present. However, the greatly increased surface-to-volume ratio in NEMS, together with the non-optimized surface properties, is the most likely explanation.
This can be illustrated by considering a NEMS device fabricated using state-of-the-art electron-beam lithography. A silicon beam 100 nm long, 10 nm wide and 10 nm thick contains only about 5 x 105 atoms, with some 3 x 104 of these atoms residing at the surface. In other words, more than 10% of the constituents are surface or near-surface atoms. It is clear that these surface atoms play a central role, but understanding exactly how will take considerable effort. My group and others - at IBM's Almaden Research Center, Stanford University, the University of California at Santa Barbara and Cornell University, all in the US, together with Ludwig-Maximilians University in Munich, Germany - are currently exploring this crucial issue (figure 5).
Ultimately, as devices become ever smaller, macroscopic mechanics will break down and atomistic behaviour will emerge. Indeed, molecular dynamics simulations, such as those performed by Robert Rudd and Jeremy Broughton at the Naval Research Laboratory in Washington DC on idealized structures just a few tens of atoms thick, would appear to support this idea.