It's a Small, Small, Small, Small World
By Ralph C. Merkle

With the tools of the nanotechnology trade
becoming better defined, the ability to create
new materials and devices by placing
every atom and molecule in the right place
is moving closer to reality.

This article was published in Technology Review, Feb/Mar 1997 page 25.  A more detailed version (with links to further information) is available.

The properties of materials depend on how their atoms are arranged. Rearrange the atoms in coal and you get diamonds. Rearrange the atoms in soil, water, and air, and you have grass. And since humans first made stone tools and flint knives, we have been manipulating atoms in great thundering statistical herds by casting, milling, grinding, and chipping materials. We rearrange the atoms in sand, for example, add a pinch of impurities, and we produce computer chips. We have gotten better and better at it, and can make more things at lower cost and with greater precision than ever before.

Even in our most precise work, we move atoms around in massive heaps and untidy piles--millions or billions of them at a time. Theoretical analyses make it clear, however, that we should be able to rearrange atoms and molecules one by one--with every atom in just the right place--much as we might arrange Lego blocks to create a model building or simple machine. This technology, often called nanotechnology or molecular manufacturing, will allow us to make most products lighter, stronger, smarter, cheaper, cleaner, and more precise.

The consequences would be great. We could, for starters, continue the revolution in computer hardware right down to molecular-sized switches and wires. The ability to build things molecule by molecule would also let us make a new class of structural materials that would be more than 50 times stronger than steel of the same weight: a Cadillac might weigh 100 pounds; a full-size sofa could be picked up with one hand. The ability to build molecule by molecule could also give us surgical instruments of such precision and deftness that they could operate on the cells and even molecules from which we are made.

The ability to make such products probably lies a few decades away. But theoretical and computational models provide assurances that the molecular manufacturing systems needed for the task are possible--that they do not violate existing physical law. These models also give us a feel for what a molecular manufacturing system might look like. This is an important foundation: after all, the basic idea of an electrical relay was known in the 1820s, and the concept of a mechanical computer that operated off a stored set of instructions--a program--was understood a few years later. But computers using relays were not built till much later because no good theoretical comprehension of "computation" existed. Today, scientists are devising numerous tools and techniques that will be needed to transform nanotechnology from computer models into reality. While most remain in the realm of theory, there appears to be no fundamental barrier to their development.

A Nano Tool Chest

Imagine putting some wires, transistors, and other electronic components into a bag, shaking it, and pulling out a radio--fully assembled and ready to work. Although this sounds fanciful, such remarkable "self-assembly" is, in essence, what chemists do whenever they synthesize materials. Mixing solutions in a beaker, a chemist lets the intrinsic attractions and repulsions of certain molecules and atoms take over. An art and science has evolved to arrange conditions so that atoms spontaneously assemble into particular molecular structures.

Similarly, we are surrounded and inspired by products that are marvelously complex and yet very inexpensive. Potatoes, for example, consist of tens of thousands of genes and proteins and intricate molecular machinery; yet we think nothing of eating this miracle of biology, mashed with a little butter. Potatoes, along with many other agricultural products, cost less than a dollar a pound. The key reason: if provided with a little soil, water, air, and sunlight, a potato can make more potatoes. Likewise, if we could make a general-purpose programmable manufacturing device that was able to make copies of itself--what nanotechnology researchers call an assembler--then the manufacturing costs for both the device and anything it made could be kept low.

A basic principle in self-assembly is selective "stickiness." If two molecular parts have complementary shapes and charge patterns--that is, one has a hollow where the other has a bump, or one has a positive charge where the other has a negative charge--then they will tend to stick together in a particular way to form a bigger part. This bigger part can combine in the same way with other parts so that a complex whole emerges from molecular pieces.

Self-assembly is not by itself sufficient, however, to make the wide range of products that nanotechnology promises. If the parts are indiscriminately sticky, for example, then stirring them together would yield messy blobs instead of precise molecular machines. We can solve this problem by holding the molecular parts in the proper position and orientation so that when they touch they will join together the way we want them to. At the macroscopic scale, the idea that we can hold parts in our hands and assemble them by properly positioning them with respect to each other goes back to prehistory: we celebrate ourselves as the tool-using species. But the idea of holding and positioning molecules is new and almost shocking. Nanoscale equivalents of "arms" and "hands" must be developed.

Current proposals for molecular-scale positional devices resemble normal-sized robotic devices, but they are about one ten-millionth as big. A molecular robotic arm could sweep systematically back and forth, adding and withdrawing atoms from a surface to build any structure that the computer instructed it to. Such an arm, composed of a few million atoms, might be 100 nanometers long and 30 nanometers around. Although it would have roughly 100 moving parts, it would use no lubricants--at this scale, a lubricant molecule is more like a piece of grit. Such ultraminiature tools should be able to position their tips to within a small fraction of an atomic diameter. Trillions of such devices would occupy little more than a few cubic millimeters (a speck slightly larger than a pinhead).

Molecular arms would be buffeted by something we don't worry about at the macroscopic scale: thermal noise. Atoms and molecules are in a constant state of wiggle and jiggle; the higher the temperature, the more vigorous the motion. To maintain its position, therefore, a nanoscale arm must be extremely stiff.

The stiffest material around is diamond. The strength and lightness of a material depends on the number and strength of the bonds that hold its atoms together, and on the lightness of the atoms. The element that best fits both criteria is carbon, which is lightweight and forms stronger bonds than any other atom. The carbon-carbon bond is especially strong; each carbon atom can bond to four neighboring atoms. In diamond, then, a dense network of strong bonds creates a strong, light, and stiff material. Indeed, just as we named the Stone Age, the Bronze Age, and the Steel Age after the materials that humans could make, we might call the new technological epoch we are entering the Diamond Age.

How can a diamond device of this scale be produced? One answer comes from looking at how we grow diamond today. In a process somewhat reminiscent of spray painting, we build up layer after layer of diamond by holding a surface in a cloud of reactive hydrogen atoms and hydrocarbon molecules. When these molecules bump into the surface they change it, either by adding, removing, or rearranging atoms. By carefully controlling the pressure, temperature, and the exact composition of the gas in this process, called chemical vapor deposition (CVD), we can create conditions that favor the growth of diamond on the surface.

But randomly bombarding a surface with reactive molecules does not offer fine control over the growth process; it is akin to trying to build a wristwatch using a sand blaster. We want the chemical reactions to occur at precisely the places on the surface that we specify. A second problem is how to make the diamond surface reactive at the particular spots where we want to add another atom or molecule. A diamond surface is normally covered with a layer of hydrogen atoms. Without this layer, the raw diamond surface would be highly reactive because it would be studded with the carbon atoms' unused (or "dangling") bonds. While hydrogenation prevents unwanted reactions, it also renders the entire surface inert, making it difficult to add carbon (or anything else) to it.

To overcome this problem, we could use a set of molecular-scale tools that would, in a series of steps, prepare the surface and create structures on the layer of diamond, atom by atom and molecule by molecule. The first step in the process would be to remove a hydrogen atom from a specific spot on the diamond surface, leaving behind a reactive dangling bond. This can be done with a "hydrogen abstraction tool"--a molecular structure that has a high chemical affinity for hydrogen at one end but is elsewhere inert. The tool's unreactive region serves as a kind of handle. The tool would be held by a molecular positional device, such as the molecular robotic arm discussed earlier, and moved directly over particular hydrogen atoms on the surface we wish to abstract.

This creates a chicken-and-egg problem: we need a molecular robotic arm to build another molecular robotic arm. To solve this problem, we must at some point build a molecular robotic arm with something other than a molecular robotic arm. We could, for example, use a macroscopic positional device--such as an improved version of an existing atomic-force microscope--to make our first molecular robotic arm. Alternatively, we could self-assemble a simplified molecular positional device. These first crude positional devices could then be used to make better ones.

One suitable molecule for a hydrogen abstraction tool is the acetylene radical--two carbon atoms triple bonded together. One carbon would be the handle, and would link to a nanoscale positioning tool. The other carbon has a dangling bond where a hydrogen atom would be in ordinary acetylene. The environment around the tool would be inert (typical proposals involve the use of either vacuum or a noble gas, such as neon).

Once this tool has created a reactive spot by selectively removing hydrogen atoms from the diamond surface, it becomes possible to deposit carbon atoms at the desired sites. In this way a diamond structure is built, molecule by molecule, according to plan. One proposal for this function is the dimer deposition tool. A dimer is a molecule consisting of two of the same atoms or molecules stuck together. In this case, the dimer would be C2--two carbon atoms connected by a triple bond. In the deposition tool, each carbon in the dimer would be connected to a larger molecule by single bonds with oxygen atoms.

The hydrogen abstraction tool and dimer deposition tool would work together. First, the abstraction tool would remove two adjacent hydrogen atoms from the diamond surface. The two dangling bonds would react with the ends of the carbon dimer. This reaction would break the carbon-oxygen bonds and then transfer the carbon dimer from the tool to the surface. Because the energy released during the reaction is much larger than thermal noise, the dimer will "snap" onto the surface and stay there.

A third proposed tool for making nanostructures is the carbene insertion tool. Carbenes--highly reactive carbon atoms with two dangling bonds--will react with (and add a carbon atom to) many molecular structures. Carbenes will readily insert into double or triple bonds, like the bond in the carbon-carbon dimer described above. A positionally controlled carbene could be attached almost anywhere on a growing molecular workpiece, leading to the construction of virtually any desired shape.

A fourth proposal is for a hydrogen deposition tool. Where the hydrogen abstraction tool is intended to make an inert structure reactive by creating a dangling bond, the hydrogen deposition tool would do the opposite: make a reactive structure inert by terminating dangling bonds. Such a tool would let us stabilize reactive surfaces and prevent the surface atoms from rearranging in unexpected and undesired ways. The key requirement for such a tool is that it include a weakly attached hydrogen atom. While many molecules fit that description, the bond between hydrogen and tin is especially weak; thus, a tin-based hydrogen deposition tool should be effective.

These four molecular tools should enable us to make a wide range of stiff structures--but only those that are composed of hydrogen and carbon. This is a much less ambitious goal than attempting to use all 100 or so elements in the periodic table. But in exchange for confining ourselves to this more limited class of structures, we make it much easier to analyze those that can be fabricated and the synthetic reactions needed to make them. In any case, this narrower proposal can be more readily and more thoroughly investigated than full nanotechnology. And diamond and its shatterproof variants fall within this category, as do the "fullerenes"--sheets of carbon atoms rolled into spheres, tubes, and other shapes. These materials can compose all the parts needed for basic mechanical devices such as struts, bearings, gears, and robotic arms.

Ultimately we'd like to add other elements--to create diamond electronic devices, for example, or add some nitrogen to the internal surface of a bearing in order to relieve strain (the carbon-nitrogen bond is shorter than the carbon-carbon bond). Such structures, composed primarily of carbon and hydrogen in combination with nitrogen, oxygen, fluorine, silicon, phosphorous, sulfur, or chlorine, constitute what we call the class of "diamondoid" materials.

The Diamond Age

Natural diamond is expensive, we can't make it in the shapes we want, and it shatters. Nanotechnology will let us inexpensively make shatterproof diamond (with a structure that might resemble diamond fibers) in exactly the shapes we want. This would let us make a Boeing 747 that would weigh one fiftieth of today's versions without any sacrifice in strength. The benefit to space travel would also be dramatic. The strength-to-weight ratio and the cost of components are critical to the performance and economy of space ships: nanotechnology could improve both of these parameters by about two orders of magnitude.

Nanotechnology could also radically alter the economics of energy production. The sun could provide orders of magnitude more power than people now use--and do so more cleanly and less expensively than fossil fuels and nuclear reactors--if only we could make low-cost solar cells and batteries. We already know how to make efficient solar cells: nanotechnology could cut their costs, finally making solar power economical. In this application we need not make new or technically superior devices; just by making inexpensively what we already know how to make expensively we would move solar power into the mainstream.

The manufacture of computer chips could undergo a profound change. There seem to be fundamental limits in how much further we can improve lithography, the process by which chips are now made. In lithography (literally, "stone writing"), we draw fine lines on a silicon wafer using methods borrowed from photography. A light-sensitive film--called a "resist"--is spread over the silicon wafer. The resist is exposed to a complex pattern of light and dark, like a negative in a camera, and developed. By repeating this process, an intricate set of interlocking patterns can be made that defines the complex logic elements of a computer chip.

But arranging atoms by throwing photons (or other particles) at a surface from a distance doesn't seem like the best approach, especially if we want to use three dimensions instead of just two; imagine building a car by throwing tools at it from more than a mile away. Thus if improvements to computer hardware are to continue at the current pace, in a decade or so we'll have to move beyond lithography to some new manufacturing technology. Designs for computer logic elements composed of fewer than 1,000 atoms have already been suggested--but each atom in such a small device has to be in exactly the right place. And spraying chemicals around simply can't arrange atoms with the needed precision.

Fortunately, diamond is an excellent electronic material. It outperforms silicon in several key respects. For one thing, electrons move faster in diamond than in silicon. Diamond can also work better than silicon at high temperatures. This is important because as chips get faster and faster, their performance is limited by the need to dissipate the heat that builds up in the circuitry.

Diamond has this advantage for two reasons. First, diamond has greater thermal conductivity than silicon, which lets heat move out of a diamond transistor more quickly. Second, diamond has a larger "bandgap" than silicon--5.5 electron volts, as opposed to 1.1 electron volts in silicon. The bandgap is the minimum amount of energy required to boost an electron from its relatively immobile state into the semiconductor's conduction band, where the electron moves freely under the influence of a voltage. As the temperature increases, more electrons gain the energy needed to jump into the conduction band. When too many electrons do this, the device changes from a semiconductor into a conductor; the transistor shorts out and stops working. Diamond's higher bandgap means it shorts out at a higher temperature.

With nanotechnology, we should be able to build mass storage devices that can store more than 100 billion billion bytes in a volume the size of a sugar cube, and massively parallel computers of the same size that can deliver a billion billion instructions per second--a billion times more than today's desktop computers.

The availability of nanoscale devices could radically redefine surgery, too. There is today a fundamental mismatch between what's needed to treat injuries and the capabilities of our tools. The cellular and molecular machinery in our tissue is small and precise, yet today's scalpels are, as seen by a cell, crude scythes that rip through tissue, leaving dead and maimed cells in their wake. The only reason that modern surgery works is the remarkable ability of cells to regroup, bury their dead, and heal over the wound.

Surgical tools that are molecular in both size and precision should let us directly heal, at the molecular and cellular level, the injuries that cause disease. A molecular robotic arm less than 100 nanometers long, for example, would easily fit into the circulatory system (a single red blood cell is about 8,000 nanometers in diameter) and would even be able to squeeze inside individual cells.

One application would be in cancer therapy. We could design a small device able to identify and kill cancer cells. The device, which would incorporate a nanoscale computer and several binding sites that are shaped to fit specific molecules, would circulate freely throughout the body, periodically sampling its environment by determining whether its binding sites were occupied. The more frequently a site was occupied, the higher the concentration of the molecule for which that site was designed. A nanodevice with a dozen different types of binding sites could in this way monitor the concentrations of a dozen different types of molecules that occur normally in the body but whose concentrations relative to one another change when cancer is present. The computer could determine if the profile of concentrations fit a preprogrammed profile and would, when a cancerous profile was encountered, release a poison that selectively kills the cancer cells.

Each device could incorporate a nanoscale pressure sensor that would allow the cancer killer to receive instructions through ultrasonic signals in the megahertz range. By "listening" to several macroscopic acoustic signal sources, the device could determine its location within the body much as a radio receiver on earth can use the transmissions from several satellites to determine its position. Awareness of its own location within the body would help the device decide whether it was near the cancer. In the absence of location information, it might sometimes mistakenly release poison in a cell that seemed to be a cancer cell. If the objective was to kill a colon cancer, for example, a cancer killer in the big toe would not release its poison no matter what its cancer sensors told it.

How Can We Get There?

The wondrous capabilities described here are, for the most part, theoretical. How can they be made real? How can we build a general-purpose, programmable manufacturing system using highly reactive, positionally controlled tools that could inexpensively manufacture most diamondoid structures?

The magnitude of this challenge should not be underestimated. Present proposals for an assembler able to fabricate diamondoid structures involve hundreds of millions or billions of atoms--with no atom out of place. Even a simple robot arm, which might be composed of only a few million atoms, would have to be accompanied by other components. The robotic arms would work in a vacuum, for instance, dictating the need for a shell around the arm to maintain that vacuum. Other ancillary gadgets that will be needed include acoustic receivers, computers, pressure-actuated ratchets, and binding sites. If each operation, such as hydrogen abstraction or carbene deposition, typically handles one or a few atoms, then the error rate must be fewer than one in a billion.

Although such perfection is theoretically attainable, today's technology is not up to the task. A chemical synthesis process that chemists view as very good converts 99 percent of the reactants to the desired product. Yet that 99 percent yield represents an error rate of one in 100, which is ten million times less perfect than we desire for a mature nanotechnology. The synthesis of proteins from amino acids by ribosomes has an error rate of perhaps one in 10,000. DNA, by relying on extensive error detection and correction along with built-in redundancy (the molecule has two complementary strands), achieves an error rate of roughly one base in a billion when replicating itself.

No existing technology can approach this level of performance. One technique that can position individual atoms, for example, is the scanning probe microscope (SPM), in which a sharp tip is brought down to the surface of a sample so that a signal is generated that lets us map out the surface being probed, like a blind person tapping with a cane to sense the path ahead. Some SPMs literally push on the surface and note how hard the surface pushes back. Others connect the surface and probe to a voltage source, and measure the current flow when the probe gets close to the surface. A host of other probe-surface interactions can be measured, and are used to make different types of SPMs.

The SPM can not only map a surface but can change it--depositing individual atoms and molecules in a desired pattern, for example. In a well-publicized case, scientists arranged 35 xenon atoms on a nickel surface to form the letters identifying their employer: IBM. But this SPM manipulation required cooling to 4 degrees above absolute zero--not exactly ideal conditions for large-scale manufacturing. More recently, IBM scientists have precisely arranged molecules at room temperature on a copper surface. However, SPMs have error rates high enough that they must use relatively sophisticated error detection and correction methods. And while these systems can move around a few atoms or molecules, they can't manufacture large amounts of precisely structured diamond of the kind that might be used to build a car or a plane.

Finally, today's SPMs are much too slow. In nature, ribosomes take tens of milliseconds to add a single amino acid to a growing protein. But if an assembler is to manufacture a copy of itself in about a day, and if this takes a few hundred million operations, then each operation must take place in a fraction of a millisecond. An SPM, by contrast, takes hours to arrange a few atoms or molecules. Rather than attempting to solve all these problems in a single giant leap, we might approach them more incrementally--developing a series of intermediate systems. One approach, for example, would be to eliminate the requirement that the assembler be made from diamondoid structures. Diamondoid is attractive, as we've seen, because of its strength, stiffness, and electrical properties. But an intermediate system need only be able to make a more advanced system, and perhaps products that are impressive in comparison with today's products. It doesn't have to be diamondoid itself.

This suggests what might be called building blockÜbased nanotechnology. Rather than building diamond, we'll build some other material from relatively large molecular units consisting of tens, hundreds, or even thousands of atoms. Such large building blocks reduce the number of assembly steps, so fewer unit operations are required, and they need not be as reliable. Soluble building blocks that stick only to other building blocks, not to the solvent or low concentrations of contaminants, eliminate the need for working in a vacuum.

In selecting such building blocks, we have many choices: any of the many molecules that chemists have synthesized, or could reasonably synthesize, with the desired properties. Each molecular building block should have at least three sites where it can link to other building blocks. Units with two bonding sites suggest the polymers ubiquitous in biological systems, such as DNA, RNA, and proteins. Building blocks that have three bonding sites make the design of stiff three-dimensional structures much easier.

Such building blocks could be linked to each other using any one of a variety of well-understood chemical reactions. A particularly attractive possibility is the Diels-Alder reaction, in which a diene (a hydrocarbon with two carbon-carbon double bonds) can be made to react with a specific molecule.

Answering the Doubters

Despite the plausibility of developing nanotechnology, there are skeptics. Their criticisms, however, are poorly informed. For example, chemist David Jones, a Nature columnist, was quoted in Scientific American that the construction of a molecular assembler was doomed because individual atoms are "amazingly mobile and reactive. They will combine instantly with ambient air, water, each other, the fluid supporting the assemblers, or the assemblers themselves."

Proposals involving reactive molecular tools, however, specify that the environment should be inert--either vacuum or a noble gas; there would be no "ambient air" to react with. And because the molecular tools are positionally controlled, they will not react with each other or the assembler itself--for the same reason that a hot soldering iron does not react with the skin of the person who wields it.

I am commonly asked how long it will be before we can make molecular computers, before inexpensive photovoltaic cells bring cheap, clean solar power, before ultralightweight spacecraft dramatically lower the cost of space exploration. The scientifically correct answer is: I don't know. But looking at one technology that nanotechnology can improve--computing--gives one perspective. From electromechanical relays to vacuum tubes to transistors to integrated circuits, we have seen steady declines in the size and cost of logic elements and steady increases in their performance for the last 50 years. Extrapolation of these trends suggests that for the computer hardware revolution to stay "on schedule" will require the development of molecular manufacturing by about 2010 or 2020.

Of course, extrapolating past trends is a philosophically debatable method of technology forecasting. While no fundamental law of nature prevents us from developing nanotechnology on this schedule (or even faster), there is equally no law that says this schedule will not slip. Much worse, though, such trends imply that there is some ordained schedule--that nanotechnology will inevitably appear regardless of what we do or don't do. Nothing could be further from the truth. How long it takes to develop this technology depends very much on what we do. If we pursue it systematically, it will happen sooner. If we ignore it, or simply hope that someone will stumble over it, it will take much longer. Fortunately, by using theoretical, computational, and experimental approaches together, we can reach the goal more quickly and reliably than by using any single approach alone. Just as Boeing can design, "build," and "fly" airplanes in a computer before making them in the real world, we can do the same for molecular manufacturing. We can quickly eliminate most of the false starts and blind alleys and rapidly focus on the best approaches.

Like the first human landing on the moon, the Manhattan project, or the development of the modern computer, the advent of molecular manufacturing will require the coordinated efforts of many people for many years. How long will it take? A lot depends on when we start.