Molecular manufacturing: perspectives on the ultimate limits of fabrication

Conventional fabrication technologies resist overall analysis not only because of their diversity and complexity, but because of their sensitivity to features of molecular potential energy functions on the order of 10^-21 J per molecule. These differences can determine, for example, the stability of solid phases in materials science and the yield of reactions in organic chemistry, yet they are hard to predict using available computational techniques. Anticipated molecular manufacturing technologies (Drexler 1981, 1992), in contrast, will exploit direct positional control to guide sequences of discrete, reliable molecular transformations in a manner that can be comparatively insensitive to small differences in potential energy functions. The relative simplicity and generality of this approach, together with its reduced sensitivity to molecular unknowns. combine to facilitate understanding of the limits of fabrication and therefore of the limits of technology.

Section 3 examines how the concept of ultimate limits can be applied to simple and complex domains, and to domains in which scientific knowledge is known to be incomplete. Section 4 surveys conventional fabrication processes and their limitations, and § 5 examines molecular manufacturing as a fundamental alternative with fundamental advantages. Taking molecular manufacturing processes as a basis, §§ 6 and 7 revisit, the implications of physical principles for limitations on fabrication. Widespread (though perhaps unexamined) opinion in the scientific community, however, suggests that a fundamental preliminary point (§ 2) must be first discussed.
 

Nonetheless, it is sometimes suggested that studies of the future of technology must by their nature be fruitless. This view seems rooted in a confusion between efforts to distinguish the feasible from the infeasible on physical grounds, and efforts to predict the detailed course of technological development in the marketplace. Since the latter would entail predicting the detailed course of human events, it is presumably unachievable. Thus paper confines itself to the more modest, less useful, but more practicable goal of exploring a few aspects of what physical law can tell us about the shape of technological possibilities. Because these possibilities depend on physical law and not on human activity or history, they have nothing to do with time or prediction except in one quite limited sense: we are reasonably sure that the ultimate limits of technology have not been approximated in the past or present, and so these possibilities – if they achieve physical reality at all – can do so only in the future. To the extent that one is willing to speculate that past. trends toward greater technological capabilities will continue, the prediction that recognized physical possibilities will be achieved or exceeded may perhaps have some plausibility.

To study fabrication processes, however, requires an approach that takes account of the complexities of condensed-matter structures and their transformations. In pursuing these studies, one typically adopts a model that embraces only ordinary matter under accessible conditions, implicitly excluding hypothetical processes involving exotic matter, monopoles, or pressures and densities found inside neutron stars (it is convenient that almost any conclusion regarding matter applies equally to antimatter, save for practical difficulties involving raw materials arid tools).

In domains that build directly on fundamental physical principles, results regarding ultimate limits are often clear and simple: the product of the uncertainties in the simultaneously measured position amid momentum of a particle must exceed h/4p, the speed of light cannot be exceeded; a cycle that writes and erases a bit must dissipate at least ln(2)kT of free energy (Landauer 1961, 1982), and so forth. In domains involving materials and fabrication, however, many results will have a different character – not stating a limit precisely, but instead identifying upper or lower bounds on the value of an imprecisely known ultimate limit. For example, the tensile strength of diamond sets a firm lower bound to the ultimate limit of tensile strength. A series of upper bounds to the ultimate limit of tensile strength can be derived from quantum mechanical principles, with the easiest-to-derive upper bounds being quite large compared to the actual physical limits. Finally, the most accurate available picture of the ultimate limits in a complex field may include qualitative uncertainties. The best available picture of a set of complex limits may consist of fragmentary survey of what sorts of capabilities seem likely and unlikely to be possible.

In macroscopic objects made by conventional fabrication, most nearest-neighbor atomic contacts (which. taken together, define the structure) result not from the operations used to shape the object, but from the operations used to prepare the materials from which the object is made. These processing operations typically control the time-history of macroscopic variables such as composition, temperature, stress, and the like. The trajectories and final locations of the constituent atoms result chiefly from thermal vibration and interatomic, potentials in a disordered, diffusive system. The final configurations accordingly depend on the thermodynamics and kinetics of spontaneously formed local patterns of atoms.

Most microlithographic processes are similar to macroscopic processes in this regard. Although many patterning technologies can produce submicron structures, the variables controlled are usually macroscopic relative to the atomic size scale, and most processes involve either volumetric or surface diffusion. Even the atomically precise layered structures produced by molecular beam epitaxy result from control of fluxes distributed over a macroscopic area. and accordingly give no precise control of transverse dimensions.
 

Organic synthesis and supramolecular chemistry face several basic difficulties. Because they rely on diffusion through solution to bring substructures together at least. one component in each reaction must be soluble. Further. each substructure must be chosen (or designed) to exhibit strong reactivity with only a single partner and with a single result; tendencies for a substructure to combine with copies of itself, or with inappropriate parts of other substructures, usually reduce the yield of the desired product. In organic synthesis, a yield of 90% is ordinarily considered high, yet a series of 200 sequential steps with this yield would convert. 10³ kg of reagents into less than 0.001 g of product. One thousand sequential steps would quite reliably yield no product at all. These difficulties impose severe constraints on organic synthesis and significantly limit the kinds of structures that can be built, by supramolecular assembly. Despite these limits, however, it appears that. chemical techniques can produce structures that can assemble to form devices that. can perform the operations necessary for molecular manufacturing (Drexler 1994).

Most of these structures are irregular and asymmetrical, and hence cannot be made by techniques that produce crystals. Since 10¹⁴⁸ is greater than the number of particles in the observable universe, random generation processes would be ineffective. to say nothing of inefficient. The technology most nearly suited for this task today is organic synthesis, yet no irregular, prespecified diamondoid structure approaching 100 atoms has ever been made by these means. Accordingly, the probability that modern technology can fabricate a particular randomly picked structure from the set defined above is effectively zero. Present synthetic techniques can make almost any specified structure, provided that it is stable and contains no more than a few atoms. Out of the total set of stable objects containing 100 or more atoms, however, present technology can make almost nothing.
 

Practical molecular manufacturing processes must use machines with components of microscopic scale, preferably built with atomic precision. (Such machines are themselves natural candidates for production using molecular manufacturing.) Design and modeling exercises indicate that machines of substantial complexity (e.g. six-axis robotic positioning mechanisms) can be built on a 100 nm scale (Drexler 1992). The physical possibility of durable, nanoscale moving parts (despite intermolecular forces) is demonstrated by the durable mobility of solvent molecules in solution, by the low frictional forces between misaligned graphitic planes, and by the observed motions of molecular machines in biological systems (e.g. the bacterial flagellar motor).

In conventional manufacturing, a standard method for creating complex structures is to grasp parts and place them where the designer has directed. (Other techniques, such as machining and moulding, have a comparable directness in the relationship between tool geometries and product structure.) Molecular manufacturing will apply this elementary principle to the molecular domain, replacing diffusive molecular motions with mechanically guided motions. The spelling of ‘IBM' using 35 xenon atoms positioned by the tip of a scanning tunnelling microscope (Eigler 1990) provided the first clear laboratory demonstration of this principle. Although the specific technique demonstrated in this work was unable to produce stable structures of practical interest. Feynman’s vision of 'maneuvering things atom by atom' had clearly been realized.

In conventional manufacturing, no two products are identical. In molecular manufacturing. as in digital logic, processes can be precise: the products consist of a precise number of parts (atoms, bits) of distinct kinds (elements, logic states) in distinct arrangements (patterns of bonding, sequences in memory). Accordingly, two products can be identical, and the distinction between a correct structure and an incorrect structure can be unambiguous. The following section discusses the physical principles that determine the reliability of the potentially precise processes of molecular manufacturing, together with the broader issue of physical limits to fabrication processes.

A less well-defined but more significant set of limits circumscribes the set of objects that can be made with high reliability given a reasonable amount of time and material. These limits appear to be more constraining than mere physical stability. Although no formal proof has been given, it seems reasonable to assume that structures can be designed such that they would be stable in their final configuration, yet unstable on all trajectories through phase space leading to that final configuration. Certain structures with large internal stresses may be members of this class. More generally, structures of low stability (e.g. liquids, gels) do not lend themselves either to precise structural definition or to step-wise construction using reliable assembly operations. The upper bounds on fabrication capabilities in such areas resist. clear definition.

Research in organic synthesis has developed techniques for making a wide range of small covalent structures. Most of these result from a series of molecular collisions, rearrangements, and fragmentation reactions that occur spontaneously in solution, driven by thermal energy and directed by intermolecular potentials. Molecular manufacturing will exploit these amid other interactions, driven by both thermal and mechanical energy and guided by stiff positioning mechanisms. In light of the wide range of operations and products demonstrated by organic synthesis, the chief physical question to be answered regarding molecular manufacturing is whether nanomechanical positioning mechanisms can in fact guide these operations with high reliability, so as to enable the construction of objects with many distinct molecular features.

The probability density function along such a coordinate is Gaussian. The reaction rate between a positioned molecular species A and a potentially reactive structure B will be proportional to the probability density of A at the position of B, all else being equal. The error rate of a mechanosynthetic process will accordingly depend on the ratio of probability densities at the target reaction site and at the nearest sites that can undergo a misreaction. This, in turn will depend on the stiffness of the positioning mechanism k_s, the temperature T, and the distance between the target site amid the potential error sites.

The separation of lattice sites on a diamond (111) surface, 0.25 nm, can be taken as a typical distance in the above calculation. At room temperature, 300 K, the probability of a reaction directed to one site instead occurring at an adjacent site will be less than 10^-15 provided that the stiffness k_s > 5 N m^-1. Inasmuch as the shear stiffness of a cubic-nanometre block of diamond is about 500 N m^-1 (and stiffness increases in proportion to size), while the bending stiffness of a single carbon–carbon bond with respect to an sp³ site is about 30 N m^-1 it should not be surprising that positioning mechanisms with stiffnesses greater than 10 N m^-1 are feasible (Drexler 1992). Error rates for molecular assembly of less than 10^-15 permit the construction of nanoscale systems of substantial complexity.

It might be thought that this physical possibility is of little practical interest because any process that. handles matter in such small particles will of necessity he slow and expensive. Surprisingly, this expectation appears to be incorrect. As suggested by the scaling laws illustrated in figure 1, the high frequencies characteristic of small machines enable them to be extraordinarily productive, both in terms of parts handled per second per kilogram of mechanism amid in terms of kilograms of product per second per kilogram of mechanism. Moreover, detailed studies of molecular manufacturing processes (Drexler 1992) have not identified a requirement for grossly wasteful energy dissipation at any stage. The free energy input required to organize matter into a precise solid structure, starting with raw materials in solution, is typically dominated by the change in the molecular potential energy of the system. The entropic cost of the transformation is comparable to that of solidifying a liquid to a crystalline solid. Accordingly, the present evidence suggests that molecular manufacturing processes can be precise, highly productive, and reasonably efficient.
 
 

It seems likely that some physically stable structures will prove impossible to make with any significant probability of success. Members of a broad class of diamondoid structures, however, are both stable and accessible through a series of stable intermediate structures. Mechanical devices can guide reactive molecules with atomic precision and high reliability, enabling the construction of covalent solids having complex, atomically specified structures. This capability will greatly expand the range of structures that can be made, and will greatly increase the performance of devices that are constrained by strength of materials, stiffness, defect densities, geometrical precision, or any of several other parameters. Because molecular manufacturing also promises to be productive and reasonably efficient, it appears that developments in this area could open a broad new domain of science and technology.

• Burkert, U. & Allinger, N. L. 1982 Molecular mechanics, ACS Monograph 177. Washington. DC: American Chemical Society.
• Clark, T. 1985 A handbook of computational chemistry. New York: Wiley.
• Drexler, K. E. 1981 Molecular engineering: an approach to the development of general capabilities for molecular manipulation. Proc. Natn. Acad. Sci. (USA)78, 5275–5278.
• Drexler, K. E. 1992 Nanosystems: molecular machinery, manufacturing, and computation. New York Wiley.
• Drexler, K. E. 1994 Molecular nanomachines: physical principles and implementation strategies. Rev. Biophys. Biomolec. Struct. 23, 377–405.
• Eigler, D. M. &. Schweizer, E. K. 1990 Positioning single atoms with a scanning tunnelling microscope. Nature 344, 524–526.
• Feynman, R. 1960 There’s plenty of room at the bottom. Engng Sci. 23, 22–36.
• Landauer, R. 1961 Irreversibility and heat generation in the computing process. IBM J. Res. Devel. 3, 183–191.
• Landauer, R. 1982 Uncertainty principle and minimal energy dissipation in the computer. Int. Theor. Phys. 21, 283–297.