Compiled by Chris Phoenix, Director of Research
Center for Responsible Nanotechnology

Fact: Mechanical systems can do precisely positioned, covalent chemistry in vacuum.
Several different covalent reactions have been demonstrated. Iron has been bonded to carbon monoxide. A single silicon atom has been removed from a silicon crystal surface and then put back in the same place. Richard Terra has written a good overview of work circa 1999. Such work demonstrates that precise positional chemistry can be achieved for a variety of reactions.

Theory: Nanoscale mechanical systems can do the same.
Most scanning probe microscopes use large piezoelectric ceramic actuators. MEMS microscopes have been built using electrostatic actuators. Nanoscale mechanical elements should also be able to function as scanning probes. Stiffness in the face of thermal noise is a significant engineering issue, but calculations indicate that diamond-like materials should be suitable even at room temperature. A variety of actuation methods should be feasible, including stepping drives controlled by a small number of relatively large actuators. Nanoscale machinery should be able to construct probes with six or more degrees of freedom.

Theory: A small set of reactions can construct 3D covalent solids, a few atoms at a time, from simple feedstock of small molecules.
The basic theory is developed in Nanosystems (Drexler, 1992). Several simulations of carbon depositions have been done such as (Merkle and Freitas, 2003). For covalent surfaces that are not prone to reconstruction at suitable temperatures (which should include at least some diamond surfaces), the same deposition reaction should work at numerous points on the surface. A small, fixed number of reactions should be sufficient to build a variety of parts with a large number of atoms.

Theory: Such 3D covalent solids can implement nanoscale mechanical systems.
Just as a rapid prototyping system can deposit dots or beads of material in layers to build 3D shapes, a mechanochemical system depositing a few atoms at a time should be able to build up a covalent surface to create 3D shapes. Different reactions may be required for edges, corners, curves, valleys, and so on. It is not yet known what atomic-scale features will be easily accessible to mechanochemistry, but it seems unlikely that building pixellated shapes will turn out to be impossible. Useful questions are: How small can the pixels be? What other features (e.g. single-bond bearings, springs, hinges) can be achieved with a compact and reliable chemistry? How compactly can general-purpose NEMS be manufactured with this technology?

Friction and efficiency in stiffly-built NEMS are questions of particular interest. Smooth, stiff surfaces should be able to slide past each other with low friction at reasonable speeds, because there will not be many mechanisms to transfer energy or force between the surfaces. If the atoms are spaced differently on each surface, energy barriers to motion should also be low, and may be made low enough that thermal noise can cause the machinery to "float" between states as biomolecules do. See "A Proof About Molecular Bearings" for discussion of such surfaces. Nested carbon nanotubes have been observed to experience extremely low friction, e.g. (Zettl, 2000). Given the range of conditions under which they can be grown, it is likely that buckytubes could be fabricated by carbon mechanochemistry.

Theory: Mechanical chemistry can be extremely reliable, with extremely high yields.
Although many conditions can be found in which mechanically guided chemistry will produce unreliable results, other conditions should produce reliable reactions. Mechanical chemistry proposals involve stiff covalent surfaces and high vacuum. Many critics of the concept are incorrectly extending knowledge about chemistry in solvents or with floppy molecules (i.e. biochemistry) to a very different domain. In the absence of anything like a complete study of useful (e.g. diamond-forming) reactions, we must depend on theory.

There are three issues to consider: How fast will the reaction happen (energy barrier)? How often will it go to completion (equilibrium constant)? Will other reactions happen instead (side reactions)? These issues are considered in detail in Chapter 8 of Nanosystems; the summary below is extremely incomplete.

1. If a sharp reactive "tip" is pushed hard enough into a receptive surface, energy barriers can be reduced to zero. Barriers less than 33 zJ (4.7 kcal/mol) allow physically constrained exoergic reactions to equilibrate in 0.1 microsecond at room temperature.
2. With suitable choice of tip atoms, it should be possible to obtain energy differences between unreacted and reacted states, 145 zJ or 21 kcal/mol, corresponding to equilibrium constants >10¹⁵: this is less than the difference in bond energy between carbon and silicon. For some reactions, the physical trajectory of the tip can be adjusted to break bonds by shear or torque. Missed reactions can also be sensed sterically and retried.
3. In vacuum with full positional control, most side reactions can be avoided. Reconstructions of the tip must be avoided by design, but the design space is huge. Surface reconstructions must be considered for each material being built; diamond surface reconstruction is considered in Section 8.6.3. Positional error is an engineering problem; a stiffness of 20 N/m allows reliable (10^-15 error rate) avoidance of side reactions that would be accessible with only 1.35 angstroms of jitter. Reconstruction of the reaction complex will usually require the breaking of one or more covalent bonds, which will not be energetically feasible if the bonds are more or less unstrained. Since applied mechanical and bond forces will be distributed among multiple bonds away from the reaction site, only the atoms closest to the reaction site need to be considered in designing the reaction.

Of course, this does not absolutely prove that a sufficient set of diamond-building reactions can be found. But given the huge number of options for tool tip design and the stability of diamond surfaces at room temperature, it is likely that at least a basic diamond-building capability can be designed.

Theory: An extremely reliable and repeatable manufacturing system can be based on positional mechanical chemistry.
With the ability to select a sequence of reactions from a predesigned set and specify a position for each reaction, large and complicated covalent shapes could be built a few atoms at a time. The expected reliability rates would allow billion-atom structures (~200-nm cube of diamond) to be built with low probability of even a single error. Scaling laws and reaction rates suggest that a billion-atom structure could be built by a single, relatively slow mechanochemical manipulator in a few hours.

Most products would consist of multiple pieces. Pieces could be fabricated separately and then assembled by direct manipulation. The shape of a stiff piece could be calculated within a fraction of an atomic diameter, allowing gripping without feedback. The soft nature of atomic electron clouds would reduce the need for precise alignment. Van der Waals forces have traditionally been viewed as a challenge, but may also be beneficial in reducing the mechanical complexity required of grippers. (Note that the grippers are applied to large molecules, not individual atoms.)

Theory: Such a manufacturing system could be completely automated.
A molecular manufacturing system would use a small set of simple and well-characterized operations with extremely low error rates for both fabrication and assembly of precise parts. A manufacturing program, designed and tested in simulation, could be expected to work reliably and produce a large number of billion-atom products without error.

Theory: With good engineering, the advantages of molecular manufacturing can outweigh its limitations.
Molecular manufacturing will have to compete against other technologies and methods. As currently understood, it appears to have substantial advantages over 3D printing, lithography, and biomimetic manufacturing. All these technologies will take substantial time and effort to develop, but molecular manufacturing can probably beat the alternatives: its basic capabilities, which might be developed in a decade, appear better than the advanced capabilities or even the ultimate limits of competing technologies.

3D printing manufactures parts by depositing or fusing small amounts of material in a raster-scanned pattern. Current 3D printing is limited to one or a few materials. Currently, these materials are no better than those available with other manufacturing technologies, and often worse. Printing whole products in assembled form would be quite difficult. Rates of scanning or deposition are slow; a printer might take months to manufacture its own mass of product. No technology has been proposed that can fabricate 1-nm features, much less with atomic precision.

Lithography is a proven and valuable technology. The feature size is shrinking steadily. However, 1-nm features are still decades away, materials are very limited, and products are essentially two-dimensional. Lithography is also a very expensive technology, suitable for making only small devices.

Biomimetic manufacturing would build products out of biomolecules such as protein and DNA, and possibly biosystems such as chemically driven motors. As Richard Smalley noted recently, "Biology ... can't make a crystal of silicon, or steel, ... or virtually any of the key materials on which modern technology is built." It appears that, although biomimetic engineering can make small precise devices and machines, their material properties will be sharply limited by the chemistry involved. Design is another problem: biomimetic engineering depends on the folding and interaction of solvated linear polymers, and this is a difficult process to predict or to engineer.

As in biomimetics, the feature size of molecular manufacturing products would naturally be atomic-scale. The ability to build nanoscale mechanical parts using nanoscale mechanical systems would allow a fabricator to produce its own weight in a few hours or less. Carbon lattice--diamond and buckytubes--appears to be an obvious and relatively easy material to fabricate with this technology.

It is often claimed that molecular manufacturing systems would be inefficient compared with biological methods. There are two answers to this. The first is that the products of molecular manufacturing would include extremely useful devices that biological methods simply cannot build. Some inefficiency in manufacturing and even in operation would be acceptable. The second, and stronger, answer is that the efficiency of biology depends on its use of physics phenomena (such as using thermal noise to cross low energy barriers), not on its use of particular chemical or material properties. Nanoscale machinery can use these same phenomena and achieve the same efficiencies.

Theory: We could build incredibly complex hardware with reliable programmable chemical operations.
Design of mechanical systems can be approached by using levels of abstraction to divide design issues and hide most of the details. A small set of well-characterized chemical operations would be recombined to build any part. The shapes built by molecular manufacturing would be specified directly by the sequence of operations used to fabricate them. Once a suitable set of reactions was known, design of new shapes would be straightforward. Likewise, once it was known how to produce shapes reliably, these shapes could be combined into a variety of parts. Standardized parts could be combined into a large variety of machines. Standardized machines could be combined into a large variety of systems, and so on. At each step, the design would be amenable to direct human understanding, with simple repetition sufficient to produce large numbers of identically placed atoms (i.e. a crystal) or large numbers of parallel machines (e.g. a computer from a few repeated logic elements).

Design at each level would be nearly independent of designs at higher and lower levels. Competence in each level could be developed in parallel, enabling rapid increase in engineering ability. Design skills and practices could be transferred from today's engineering disciplines. Isolating adjacent systems to allow independent treatment is an engineering problem, not a fundamental limitation; engineering today deals with effects such as heat and vibration, and these will be as easy if not easier to deal with on the nanoscale.

With the shape of each part and the deviation caused by thermal noise known precisely, each step of operation (including compound operations) could be tested in simulation for reliability and repeatability. Products, as well as the process to manufacture them, could be verified before they were ever built.

Theory: The range of hardware could include a system capable of copying its structure.
A mechanical system capable of doing programmed positional chemical operations could be very small. The manipulator is likely to require only a few degrees of freedom; a Stewart platform or similar stiff 6DOF manipulator should be adequate to do a wide range of reactions. Diamond is stiff enough to do room-temperature diamond-building mechanochemistry. With a palette of shapes and parts to choose from, a six-actuator mechanical system could be built. The "tips"--the small reactive molecules that do the deposition reactions--may require some combination of solution chemistry and mechanochemistry, but the rest of the structure should be buildable entirely with mechanochemistry. A billion atoms and a few hundred nanometers should be sufficient to encompass a simple mechanochemical fabrication system.

Theory: Automated production of molecular machine parts from straightforward design appears possible.
In a mechanochemical system building covalent solids, atoms will stay where they are placed. This means that the shape of the part can be predicted directly from the sequence of assembly operations. If the chemical operations are as reliable as expected, large numbers of billion-atom systems can be produced without a single error and without complicated error detection or correction mechanisms. A system that can exactly duplicate its structure in a few hours and makes errors only rarely does not need to be repaired; it can simply be thrown away after the first error.

Theory: Systems and products, including macroscopic products, can be produced from arrays of nanoscale chemical fabricators and larger assembly robotics.
As discussed in this paper, molecular manufacturing can be the basis for manufacturing large products. Given a mechanochemical fabricator (just the manipulator, not the computer or power source) that fits within 200 nm and can undertake all the necessary diamond-forming operations, it appears fairly straightforward to combine large numbers of such fabricators with control and power supplies into a nanofactory. The billion-atom products of the fabricators can be fastened together into much larger products. A nanofactory should be able to fabricate another nanofactory. Design of the nanofactory and products should be straightforward.