Dr. Dobb's Special Report December 2000
Conventional lithography-based VLSI technology is fast approaching the limits of its capabilities. The reasons for this range from ultra-thin gate oxides, short channel effects, and doping fluctuations, to increasingly difficult and expensive lithography. To surmount problems such as these, nanoscale quantum devices and circuits have been proposed for some time, and demonstrations of many of these technologies have been accomplished. Still, the question remains whether these will be viable technology alternatives for the postVLSI era.
Many of these proposals do not critically address the major limiting factors of a 2D lithography-based technology: accessible parallel fabrication, interconnection density, and alignment. Any successful new technology must solve the interconnect problem, use self-aligned fabrication, and operate at room temperature and at the atomic level. The last point is important because scaling any technology to the 10-nM level may not be cost effective, as the performance increase is marginal compared to the development costs.
Recently, molecular electronics-based computation has attracted attention because it addresses the ultimate in a dimensionally scaled system -- ultra-dense and molecular-scale. An additional driving factor is the potential to utilize thermodynamically driven directed self-assembly of components, such as chemically synthesized interconnects, active devices, and circuits. This approach for spontaneously assembling atomic scale electronics attacks the interconnection and critical dimension control problems in one step, and is implicitly atomic scale. Concurrently, the approach utilizes inherently self-aligned batch processing techniques that address the fabrication limitations of conventional VLSI.
The world of synthetic chemistry lets you create complex structure and function in a batch parallel fabrication method. And unlike the operation of semiconductor devices that depends critically on lithography-defined processes (and thus inherently has variation), every molecular device is precisely the same. The nagging problem of having to make hundreds of millions of integrated silicon devices, each having characteristics identical to one another, becomes difficult (if not impossible) to achieve as the device is downscaled. This problem does not exist in the molecular device realm, where all the molecules must inherently be the same.
The methods used to produce these molecules are the same as those in the pharmaceutical industry. For example, a molecular device is created by starting with a compound, then gradually transforming it by adding prescribed reagents whose molecules are known to bond to others at certain specific sites, and creating the desired orbital structure. The procedure may take many steps, but gradually the pieces come together to form a new potential molecular device.
To achieve these results and demonstrate electrical function, researchers needed a way to make physical and electrical contacts to something the size of a single molecule -- a challenge that has stymied them for many years. How do you contact the molecular world? Once a molecular species is made, how is it turned into a device that can be plugged into a conventional electronic circuit?
For molecular-scale electronics to come of age, fabrication and measurement techniques had to reach the atomic scale. The advent of atomic imaging techniques, such as the scanning tunneling microscope (STM) and the atomic force microscope, have given us an atomic view of molecular placement, fabrication, and self-assembly.
Manipulating molecules, one at a time, has become possible with the invention of the scanning tunneling microscope. However, STM, or single manipulation, loses its utility for large-scale electronics -- what you want is a method to manipulate molecules to where you want (that is, fabricate the device) in a manner that is analogous to the parallel processing strategy of lithography.
The key is a technique called "self-assembly." First developed by Dave Allara at Bell Labs, it was discovered that certain molecular endgroups could have an extremely high affinity to binding onto metal surfaces. One of the most common and well studied is the sulfur endgroup gold system called "thiol." A variety of other endgroup-contact systems are available as well. You now only need to attach the sticky endgroups to the molecular device of interest, and they spontaneously assemble onto the metallic contacts, forming the device. If you can assemble truly large-scale functional electronic components, the technique is not only analogous to lithography, it is better -- no lithography steps need to be performed in the device fabrication.
In 1996, the first measurements of a self-assembled molecule were performed by Paul Weiss's group at Penn State University using molecules synthesized by James Tour's group at the University of South Carolina. Using STM, they qualitatively observed conductivity of a designed molecular wire, self-assembled onto a substrate. Similar results were also obtained in a group at Purdue University. At the same time, the group at Yale University was performing the first quantitative electrical measurements of a single molecule, fabricated by self-assembly; see Figure 1. Using the simplest possible construct to demonstrate the principles -- a single benzene ring for mobile, delocalized electrons, with sticky thiol endgroups on both ends to contact the metal leads -- the experiment for the first time supplied real electrical characteristics to compare to theory, and demonstrated that self-assembly was a viable technique.
The first observations of conduction in molecules quickly led to simple device demonstrations. The simplest electronic device you can imagine is a diode. In 1997, only a few months after the first measurements, diodes were realized by two separate groups -- one by Robert Metzger's group at the University of Alabama, and another by Chong-Wu Zhou at Yale. The Alabama results were based on a molecule that had an internal energetic line-up or orbitals, different depending on the polarity of voltage applied. The Yale result used a different principle, the difference in the line-up of the energies of the external metal contacts to the internal molecular orbitals, to create large rectification. These first basic device demonstrations set the stage for the design of truly useful and interesting molecular devices and circuits.
To make these devices, an approach adopted by the Yale group (an adaptation of a structure made first by Kristan Ralls and Bob Buhrman at Cornell University) used an extremely minute device, called a "nanopore," in which a self-assembled monolayer of a small number of the candidate species formed the active region of the device. In a small 30-nanometer hole, a small number -- approximately a thousand -- of the molecules self-assemble to form the device. Self-assembly is a technique used to create a perfect single layer (or monolayer) of molecules in a rigid structure. Typically, you start with a beaker containing a solution of the molecules you want to attach to the active region of the device, which has an exposed metal surface. By merely dipping the wafer with these exposed areas into the beaker, the molecules attach themselves to the surface in a regular, typically well-ordered array resembling a bed of nails. Evaporating a metal contact onto the top of the self-assembled monolayer (SAM) completes this device.
The first molecular diodes were made using devices like these -- but a diode is not sufficient. To make a general-purpose computer, you need at minimum a controllable switch. Even better, you need that switch to have amplification (gain) to achieve large scales of integration. The silicon transistor is the heart of integrated circuits for just that reason. Although the molecular equivalent of a transistor with gain is yet to be discovered, researchers have taken the first steps along the path by realizing switches.
The first observations of switching behavior, caused by a voltage-dependent overlap of orbitals through a molecule, exhibited a very large switching behavior; see Figure 2. This switching effect, first seen by Jia Chen at Yale, has an on/off ratio greater than 1000 (compared to around 100 seen in the solid state analogous device called a "resonant tunneling diode," or RTD). Similar behavior was observed in experiments jointly conducted by the University of California at Los Angeles and Hewlett-Packard. Although these elements do not yet exhibit amplification, they still have a range of potential circuit applications when combined with conventional electronics.
As an example of the ease and flexibility in which molecular-scale devices can be redesigned, a simple change to the internal electrically active unit made the molecule sensitive to the charge state of the internal group. Absence or presence of charge in the internal node would modify the orbital overlap -- and it would retain, or remember, this charge for a long time. Thus, using molecules, you have created a memory. In combination with switches, complex circuits can be realized.
Given the potential fabrication and density advantages of molecular devices, why don't we scrap silicon research and proceed wholeheartedly to molecular-based systems? A sober view identifies a number of difficult obstacles, both fundamental and technological, which must be surmounted before the potential benefits can be realized.
Foremost is to make molecular devices that operate analogously, like a transistor, with a third control terminal. With this achievement, the chemical synthesis of tremendously efficient and complex circuits will be possible. Even before then, combinations of molecular systems with conventional electronics will probably be used in places where the advantages of self-assembly are natural.
Utilizing the impressive difference in densities may not be as simple as it appears. The sheer thermal aspects of molecules as electronic devices, run in a conventional mode, is staggering. Present state-of-the-art microprocessors with 10 million transistors and a clock cycle of half a gigahertz (half a billion cycles per second) emits almost 100 Watts -- similar in radiant heat to a range-top cooking surface in the home -- and is close to the thermal limitations of the technology. Knowing the minimum amount of heat that a single molecular device gives puts a limit on the number of devices we can have on a chip (and this underestimates the total amount of heat), because there is heat generated in the interconnections between devices. This fundamental limit of a molecule, operating at room temperature and at today's speeds, is about 50 picoWatts (a millionth millionth of a Watt), which means a number of devices no greater than 100,000 times more than what we can do today. Although an appreciable advance, it is far below the density one can achieve (in area, or perhaps in three dimensions) by simple volume considerations.
Assumed here is the conventional notion that every device is addressable -- that you can find a specific device from among the countless millions through the interconnections, such as distinguishable street addresses. This kind of addressing (random access) would be required, for example, to retrieve the contents of a particular memory location. At present, there is no notion of how you can create such an interconnect structure on the molecular level. Straightforward extensions of the present techniques we employ to create complex microelectronics are not useful for molecular scale electronics. Is the ability to address every device -- the common architecture we use today -- necessary or efficient at the type of molecular-scale densities we can achieve? Are the logic gate and circuit approaches we have been using in semiconductor systems for the last few decades the right approach? What will large-scale circuits of this technology look like?
In the future, radical departures from present computing-design approaches, for which we only have limited ideas and insight, will probably be needed to fully exploit molecular computing systems if we want to extend microelectronics significantly beyond the Moore's Law limit. The ability to construct complex molecular devices, with a shopping list of how what can stick to whom, and in what configurations, opens up an entirely different way to think about computer construction.
DDJ