© 2004 Robert A. Freitas Jr. and Ralph C. Merkle. All Rights Reserved.

Robert A. Freitas Jr., Ralph C. Merkle, Kinematic Self-Replicating Machines, Landes Bioscience, Georgetown, TX, 2004.

6.4.2 Exemplify a Simple Design

Another important design objective is simplicity. The term “assembler” does not describe a specific design, but a large family of designs. Our objective would not be to describe the best, or the most efficient, or the most flexible, or the fastest possible design. Instead, our objective would be to describe a simple design that satisfies our basic goals of safety and efficacy, while illustrating the use of programmable positional control to achieve flexible manufacturing capabilities and to achieve self-replication or massively parallel operation. It is hoped that the design would serve, among other things, as a theoretical demonstration of concept – a demonstration which must necessarily precede the experimental proof of concept in the laboratory.

(1) Offloading Complexity. Creating a simple design that is capable of achieving the main objectives requires a general design philosophy which we call “offloading complexity.” In this design philosophy (explicitly adopted here), whenever practicable, structures and functions required to be performed by the device are made as simple as possible by transferring as much as possible of the structure or function to outside of the device [1287], or up to the macroscale [208]. In the Merkle-Freitas example (Section 4.11.3), for instance, rather than using a more complex onboard computer with stored instructions or even a somewhat simpler signal demultiplexor to control the assembler, a far simpler direct-drive control chain is employed onboard the device and the train of control signals is provided sequentially from without. Rather than a more complex indigenous onboard power supply, power is supplied via a simple pressure-driven actuator and source power is generated and transmitted from the outside; and so forth.

We seek to offload functional complexity so that the operations required to be performed by the device and the analyses necessary to rigorously estimate and validate device performance can be made as simple as possible, thus maximizing the probability that the proposed device will work as claimed. We seek to offload structural complexity so that the device is as close as possible to current or foreseeable “manual” molecular construction capabilities, an important step toward addressing the well-known “chicken-or-egg” problem. The principle of offloading complexity is somewhat analogous to the “end-to-end principle” in network design – the concept that intelligence should be placed at the edge of a network to keep the network itself simple, making the technological evolution of the network much more flexible by requiring only a minimum amount of coordination among network owners and users [3073].

(2) Restriction to Hydrocarbon Materials. An important design tradeoff results from the conflict between the desire for a manufacturing system able to arrange atoms in most of the ways consistent with physical law, and the desire for onboard structural and functional simplicity. A good compromise seems to be to restrict the system objectives to the synthesis of a diverse range of stiff hydrocarbons. A complete analysis of the reactions by which an assembler converts incoming raw materials into reactive tools used to synthesize molecular structures is greatly simplified if we restrict ourselves to the elements hydrogen and carbon, and further restrict our attention to structures that are relatively stiff (excluding, for example, floppy polymers). The stiff hydrocarbons include a wide enough class of materials to be a very attractive goal – diamond, graphite, and structurally related materials are found in this class. Essentially all mechanical structures can be made from stiff hydrocarbons including struts, bearings, gears, levers, etc. This can be most readily seen by noting that the strength-to-weight ratio of diamond is over 50 times that of steel or aluminum alloys – a single part made of metal could be functionally replaced by a similarly shaped stiff hydrocarbon part. The resulting part would be lighter and stronger than the part it replaced, improving overall performance. The class of stiff hydrocarbons also includes molecular computers which, by today’s standards, would be extraordinarily powerful [208].

A more general assembler, able to manufacture structures which incorporate most of the elements of the periodic table, would be substantially more difficult to analyze. One approach to breaking down the task of building a relatively large and complex structure would be to consider a series of small incremental changes to an exposed surface, the cumulative effect of which would be to manufacture the whole. This implies we must analyze small changes to the exposed surface, presumably by considering small clusters of atoms on that surface. A very minimal cluster might be a single atom and the atoms to which it is bonded. If one atom is bonded to (say) three neighbors, and all four atoms can be any one of about 100 possibilities, then this gives us 100⁴ or ~100,000,000 possible clusters. This analysis is crude and likely too small because (a) atoms are often bonded to more than three other atoms and (b) understanding an incremental change to a small cluster often requires examination of atoms farther away than one bond length. Despite its shortcomings, this crude model tells us that we would need to analyze many types of incremental surface modifications before we could reasonably hope to synthesize the full range of structures accessible using this approach.

By contrast, if our structures contain only hydrogen and carbon then the problems of design and analysis become much simpler. Hydrogen can only be bonded to one other atom which, if we exclude hydrogen gas, must be carbon. A carbon atom will usually be bonded to two, three, or four neighboring atoms, which can only be hydrogen or carbon. Our previous crude analysis would assign 2⁴ or 16 possible local clusters for carbon. While this can be reduced by considering isomers, it must also be increased to consider interactions that extend beyond a single bond length, e.g., aromatic rings, conjugated systems and the like. In any event, the complexities of analyzing hydrocarbon structures with sufficient accuracy for the purposes discussed here is tractable with present capabilities. We can cut short the combinatorial explosion before it begins.

While this rather drastic pruning makes the problems of designing and analyzing a hydrocarbon assembler more tractable, it does not directly address the feasibility of more general assemblers. Smalley [3041] in particular has argued that a “completely universal” assembler is impossible, though he has also admitted [3089] that “most interesting structures that are at least substantial local minima on a potential energy surface can probably be made one way or another.” Smalley argues that a small set of molecular tools will be unable to catalyze all the reactions needed to synthesize the remarkably wide range of structures that are possible. We agree with these assessments. Success may require the use of a great many custom-made catalytic structures.

Given the remarkable size of the combinatorial space of possible molecular structures, it seems likely that at least some members of this space will resist direct synthesis by an assembler equipped with a relatively modest number of molecular tools. However, even if we assume that a substantial percentage of the space is inaccessible via this route (an assumption as yet lacking any clear support) the remaining “small” fraction would still include structures of enormous economic value. Even the ability to manufacture only the highly restricted range of structures defined by the stiff hydrocarbons would usher in a revolution in manufacturing.

Further research aimed at clarifying the range of structures amenable to synthesis by positionally controlled molecular tools is urgently needed. Ideally, this would include not only the proposal and analysis of particular sets of molecular tools and the range of structures they could reasonably make, but also proposals of structures which could not be synthesized by the use of positionally controlled molecular tools. One example of an “impossible” structure is a cubic meter of flawless diamond – before its manufacture could be finished, background radiation would have introduced flaws. Drexler [199], p. 246, argued that it should be possible to define a structure which would be stable if complete but unstable when almost complete, a sort of molecular stone arch [3090]. However, a specific, relatively small, stiff and stable structure that can reasonably be viewed as “impossible” to synthesize using positionally controlled tools has not yet been proposed. While it seems likely that at least some such structures must exist, our understanding of this issue would be greatly improved by specific examples.

(3) Design for Assembly. Because the unitary construction of interlocked parts can be extremely difficult, it will likely be necessary to fabricate individual parts which must then be assembled into completed structures. For this reason, an important objective is “design for assembly” – the idea that among numerous available design alternatives for a given machine part, the one that is easiest to physically assemble with other parts into desired structures should be preferred over other parts and parts combinations that may be more difficult to assemble. Design for manufacturing/assembly (DFM/A) [3091-3098] and computer-aided assembly planning [3099, 3100] are recognized specialties in conventional manufacturing engineering, and these emerging disciplines will likely play an important future role in molecular manufacturing as well.

(4) Design for Analysis and Validation. As a practical matter, the chosen design should be accessible to analysis by currently available computational tools for molecular modeling and simulation. The design should be mechanically simple, thus readily permitting basic static, kinematic [3101], and dynamic analyses. It should employ the minimum number of physical or chemical processes that have not yet been validated experimentally. In short, the best design proposal for a molecular assembler will be one that can be validated as correct – or likely to be correct – using our present-day analytical knowledge base.

Last updated on 1 August 2005