© 1999 Robert A. Freitas Jr. All Rights Reserved.

Robert A. Freitas Jr., Nanomedicine, Volume I: Basic Capabilities, Landes Bioscience, Georgetown, TX, 1999

2.4.2 Molecular Assemblers

Drexler⁸¹⁰ has proposed the molecular assembler,* a device resembling an industrial robot which would be capable of holding and positioning reactive moieties in order to control the precise location at which chemical reactions take place. This general approach would allow the construction of large atomically precise objects by a sequence of precisely controlled chemical reactions. Much like the ribosome in biology (Section 2.3.1), an assembler would build various classes of useful molecular structures following a sequence of instructions. During this process, the assembler would provide three-dimensional positional and full orientational control over the molecular component (analogous to the individual amino acid in the ribosome model) that is being added to a growing complex molecular structure (analogous to the growing polypeptide in the ribosomal model). In one approach, a molecular assembler may be capable of forming any one of several different kinds of chemical bonds (e.g., by changing tool tips), not just a single kind such as the peptide bond that the ribosome makes. In bonding atoms or molecules to one another, the assembler would provide any needed energy (especially if the reaction happens not to be energetically favored) through physical force, thus performing mechanosynthesis (as opposed to the traditional means of chemical synthesis in solution). In another approach, a molecular assembler might be capable only of noncovalent assembly operations, wherein nanoparts are fabricated by other means and then presented to the assembler, which assembles the nanoparts into working nanomachines.

* But not a "universal" assembler: "Though assemblers will be powerful (and could even be directed to expand their own toolkits by assembling new tools), they will not be able to build everything that could exist" (Drexler⁸ at page 246). The term "universal assembler" appears only in a section heading in Drexler's popular work Engines of Creation⁸ but not in the text; the term appears nowhere in Drexler's technical work Nanosystems.¹⁰

The first simple molecular assembler will almost certainly be a macroscale device, perhaps a modified SPM system as was being pursued by Zyvex²⁷⁹⁴ in 1998. Multiple SPM heads could be equipped with a small number of nanoscale tool tips. In one scenario, nanoparts fabricated using bulk chemistry techniques would be inspected and selected by the SPM, then assembled one by one into working nanomachines (e.g., the desired useful nanoscale products). Such assembly operations will be very slow, because the placement of each new component may require simultaneous rotations and translations of large macroscale SPM components. Assembly time scales roughly linearly with assembler size¹⁰ because smaller assembler components moving at a given velocity need to travel less distance to accomplish a given physical operation, hence consuming less time and energy per physical operation. An important early developmental goal thus will be to design and fabricate nanoscale molecular assemblers.

At its most basic level, the simplest possible nanoscale molecular assembler may be comprised of one or more nanoscale manipulators (Section 9.3). For example, the diamondoid telescoping manipulator arm described in Section 9.3.1.4 (Figs. 9.8 and 9.9) has about 4 million atoms excluding the base and control and power structures; doubling the size to account for support structures gives a molecular weight per arm of ~100 megadaltons and a total molecular volume of ~140,000 nm³. Designed for high strength and stiffness, this robot arm should be able to hold a molecular fragment stiffly enough to make it react with a chosen end of a carbon-carbon double bond with an error rate of only 10^-15.¹⁰ The robot arm could guide chemical reactions with high reliability at room temperature in vacuo, with little need for sensing the positions of the molecules with which it is working.²⁷⁹ Alternatively, with appropriate tool tips, such an arm could grasp and manipulate individual prefabricated nanoparts in solution phase. Assembly of nanoparts without fabrication may require only a very small tool set.

The manipulator arm must be driven by a detailed sequence of control signals, just as the ribosome needs mRNA to guide its actions. However, such detailed control signals can be provided by external acoustic, electrical, or chemical signals that are received by the robot arm via an onboard sensor or power transducer, using a simple "broadcast architecture"^10,280,2872 (Chapter 12), a technique which can also be used to import power (e.g., Section 6.3.3). Such transducers may be extremely small, on the order of (~10 nm)³ each.¹⁰ Interestingly, the biological cell may be regarded as an example of a broadcast architecture:²⁸⁷² the nucleus, located external to the cytoplasm, broadcasts mRNA chemical signals to millions of spatially diverse cytoplasmic ribosomes, thereby remotely controlling the construction of cellular proteins.

Thus a two-armed mechanical molecular assembler that receives power and instructions from some external agency may have a total molecular volume of ~300,000 nm³ (a cube ~67 nm on an edge), containing ~16 million atoms with a molecular weight of ~200 megadaltons -- very roughly the mass and scale of a medium-size virus particle, such as an adenovirus. For comparison, the average enzyme (a biochemical crimping tool) weighs about ~0.1 megadalton, while a typical ribosome (a primitive "protein assembler") weighs ~4.2 megadaltons. A simple mechanical assembler in the 10-100 megadalton range cannot be ruled out. However, such a device might have a very small set of tool tips and an extremely limited manufacturing repertoire.

Multiple nanoscale assemblers each capable of independent simultaneous actuation may require control signals more conveniently provided by an onboard nanocomputer (Section 10.2). This programmable nanocomputer must be able to accept stored instructions which are sequentially executed to direct the manipulator arm to place the correct moiety or nanopart in the desired position and orientation, thus giving precise control over the timing and locations of chemical reactions or assembly operations. The mechanical nanocomputer analyzed by Drexler¹⁰ requires ~16 nm³ per logic gate and ~40 nm³ per data register. If such components could be used to construct the equivalent of the most primitive 4-bit Intel 4004 microprocessor, then the nanocomputer could process ~10⁵ bits/sec (~25,000 ops/sec) at a ~1 KHz clock speed within a mechanism volume of ~36,000 nm³ (Section 10.2.1), again neglecting power supply, I/O linkages, and the like. An additional ~160,000 nm³ of rod logic registers (Section 10.2.1) adds ~1 kilobyte of onboard RAM memory or ~40 kilobytes of internal tape memory (conservatively assuming a tape storage density comparable to linear DNA). (A memory tape containing all bits needed for complete self-description (Table 2.1) may be considerably longer, even allowing for substantial data compression.) Doubling the total volume to account for support structure and other overhead gives a minimum mechanical nanocomputer molecular volume of ~400,000 nm³, roughly 70 million atoms with a molecular weight of ~800 megadaltons. Thus the smallest nanocomputer-driven nanoscale molecular assembler with two manipulator arms may have a total molecular volume of ~700,000 nm³ (a cube ~88 nm on an edge), containing ~86 million atoms with a molecular weight of ~1 gigadalton.

By 1998, only a small amount of research targeted at actual assembler design had begun. Following Drexler's original discussions,⁸¹⁰ during 1991-1998 R. Merkle authored or coauthored a continuing series of papers discussing various operational aspects and specific components of assembler design, including mechanosynthetic positional control,^2761,2762 general design considerations for assemblers,^280,2868 the broadcast architecture,²⁸⁷² convergent assembly,²⁸⁶⁹ binding sites,¹¹⁹⁹ positioning devices,¹²³⁹ mechanosynthetic path sets,²⁶⁰² designs for a neon pump²⁸⁵⁸ and a fine motion controller,²⁸⁵⁹ and possible assembler casings.²²⁸¹ J.S. Hall has considered high-level designs for a nanoscale parts-fabrication and parts-assembly nanorobot.²⁸⁷⁰ Zyvex,²⁷⁹⁴ founded in 1996 by James von Ehr, has set itself the task of building the first programmable nanoassembler in a 5-10 year time frame (e.g., by ~2006). W. Goddard and colleagues²⁸⁵³ have proposed a series of molecular dynamics simulations of simple assemblers, although by 1998 these studies evidently had not yet begun. According to Goddard's original proposal:

"Ultimately we need a programmable synthetic system to make a real device. Even though we may not have tools for all the chemical steps and may not have designs for all the pumps, engines, and transmissions needed, we propose to study the dynamics of simplified prototype assemblers. In these studies we anticipate having (1) a reservoir or supplies area for providing the various building units (atoms and fragments) required, (2) a work area in which we construct the nanomachine device (initially we will consider assembling the structure on top of a diamond surface), and (3) a molecular scale nanohand which will extract the atoms from (1) and carry them to (2). We will then use extensions of our massive molecular dynamics program to operate the system: moving the tip from reservoir to work area, moving it to contact the appropriate surface site, moving it to regenerate the active tip, and then moving it back to add new atoms and molecules. This will include proper temperature effects, molecular vibrations, energy release upon the various chemical steps, etc. The ground rules here are that a realistic force field be used and that all pieces be treated at the atomic level (but some might be semi-rigid). This will use the force field developed for nanosynthesis. The purpose of these simulations is to examine issues of vibration caused by chemical forces as the tool picks up and delivers atoms to the growing surface. Also we want to consider the effect of energy release in the chemical steps on the thermal fluctuations in these systems (which may cause displacements and vibrations)."

Building mega-atom or giga-atom nanoproducts one at a time would be incredibly expensive and time consuming. For example, imagine that we wish to construct a simple medical nanorobot such as the 1-micron spherical respirocyte¹⁴⁰⁰ described in Chapter 22, which consists of ~18 billion atoms (dry structure). A factory employing a coordinated team of 100 macroscale SPM assemblers, each able to place ~1 atom/sec-SPM on a convergently-assembled workpiece achieves a pitiful manufacturing throughput rate of ~2 respirocytes per decade. Nanoscale assemblers with appendages ~10⁶ times smaller than macroscale SPM assemblers might plausibly achieve net assembly rates of ~10⁶ atoms/sec-nanoassembler, but even a production line employing ~300 such nanoassemblers (which the abovementioned 100-SPM factory team could build in ~1 year, assuming ~10⁷ atoms/nanoassembler) can only manufacture ~1 respirocyte nanorobot per minute. At that rate, it would take ~2 million years to build the first ~1 cm³ therapeutic dosage containing ~10¹² respirocyte nanorobots.

The necessary solution to this mass-production bottleneck is to employ any of several massively parallel approaches to manufacturing, including such techniques as self-assembly (Sections 2.3.1 and 2.3.2), convergent assembly,^10,2869 or, most usefully, self-replication (Chapter 14). The basic advantage of self-replication is readily illustrated. Consider a single ~10⁸atom "seed" nanoassembler (having an onboard nanocomputer) that has been painstakingly built in ~10⁶ sec using the abovementioned 100-SPM macroscale assembler team, working at ~1 atom/sec-SPM. The seed nanoassembler is first programmed to build a copy of itself, which it accomplishes (working at ~10⁶ atoms/sec-nanoassembler) in ~100 sec. These two nanoassemblers then each build a copy of themselves in another ~100 sec; now there are four nanoassemblers. After ~48 generations, requiring a total of ~80 minutes to complete, there are ~3 x 10¹⁴ nanoassemblers.* These nanoassemblers are then reprogrammed for the manufacture of 18-billion-atom respirocytes and are fed the appropriate, presumably different, feedstock. This vastly expanded nanoassembler manufacturing system can now produce ~10¹² respirocytes (~1 cm³ therapeutic dose) per minute.

* Numerous complexities and limitations have been ignored here. For example, exponential growth cannot continue indefinitely in a finite environment with limited materials transport speeds -- e.g. a 2-nm wide nanopart suspended in 310 K water diffuses only ~26 nm in 10^-6 sec (Eqn. 3.1). Mechanical replicating systems designed purely for molecular manufacturing may be inflexible and brittle, employing limited energy resources and materials feedstocks not found outside of the immediate manufacturing environment.

The design of machines able to make copies of themselves was first described by von Neumann.¹⁹⁸⁵ Many variations on this theme have been reviewed by Freitas and Gilbreath¹¹⁵ and Sipper,²⁸⁷¹ and design for self-replication in the context of nanoscale assemblers has been considered by Drexler,⁸¹⁰ Merkle,^{116,2868,2872,2873} and Hall.²⁸⁷⁰ As Drexler¹⁰ notes: "It may seem somehow paradoxical that a machine can contain all the instructions needed to make a copy of itself, including those selfsame complex instructions, but this is easily resolved. In the simplest approach, the machine reads the instructions twice: first as commands to be obeyed, and then as data to be copied. Adding more data does not increase the complexity of the data-copying process, hence the set of instructions can be made as complex as is necessary to specify the rest of the system. By the same token, the instructions transmitted in a replication cycle can specify the construction of an indefinitely large number of other artifacts."

The estimated information content of self-replicating systems -- the length of the instruction tape -- can be surprisingly small (Table 2.1). Von Neumann's original analysis¹⁹⁸⁵ concluded that perhaps twelve different kinds of units of unknown complexity could be required as building materials, and Haldane²⁸⁸⁴ inferred that as many as ~10⁵ individual parts might be needed to make a replicator. This inference was refuted just three years later with the arrival of the first in a stream of very simple but ingenious designs for mechanical self-replicating machines that were assembled and operated in the late 1950s. In the first example,²⁸⁷⁹ a pair of joined ratchet-like blocks,²⁸⁸⁶ when placed in a "sea" of left and right blocks and then physically agitated, replicated itself from this well-ordered input substrate, making more block-pairs until all the single blocks were used up, a process similar to chemical autocatalysis. More complex congeries of blocks, including clever four-block, eight-block, and 12-block replicators that could replicate themselves in a sea of blocks were presented by Penrose.^2880,2881 Jacobson²⁸⁸² demonstrated a 3-unit replicator consisting of toy train engines circulating on HO model railroad tracks, and Morowitz²⁸⁸³ designed a simple two-unit device with one unit comprised of about a dozen components including switches, batteries and electromagnets, that could assemble copies of itself from parts floating on the water surface of a bathtub. In 1998, Lohn and colleagues²⁸⁸⁵ gave two designs for simple self-replicating systems that could be constructed out of wood, batteries and electromagnets, with explicit analogies to nanoscale fabrication. Several important conclusions may be drawn from these examples:

First, replication is fundamentally so simple a task that machines capable of displaying this behavior predate most of the modern electronic computer era.

Second, assembly is an inherently simpler operation than fabrication. A complex part may embody hundreds or thousands of prior fabrication and assembly operations, yet may be installed within an assemblage in a single step. Hence nanopart assembly may be an easier candidate for early implementation in first-generation self-replicating molecular manufacturing systems than is molecular fabrication.

Third, and most important, the simplest nanoreplicator may require the ability to assemble, but not to (atomically) fabricate, in order to replicate itself. A nanomachine capable of assembly alone can replicate itself only from a very limited set of well-ordered input materials of relatively high complexity, but such a machine can be extremely simple both in structure and function. Certainly a nanomachine capable of both assembly and fabrication can replicate itself from a more diverse and disordered set of more elementary input materials but only at the expense of far greater internal structural and functional complexity.

At least two distinct models of replication have been identified in replicating systems design.¹¹⁵ The first model may be called the "unit replication" or organismic model (Fig. 2.34), in which the replicator is an independent unit which employs the surrounding substrate to directly produce an identical copy of itself; both the original and the copy remain fertile and may replicate again, thus exponentiating their numbers. The second model may be called the "unit growth" or factory model (Fig. 2.35), in which a population of specialist devices, each one individually incapable of self-replication, can collectively fabricate and assemble all necessary components comprising all specialist devices within the system, hence the factory is capable of expanding its size (or of manufacturing duplicate factory systems) indefinitely in an appropriate environment. The factory model is sometimes called "bootstrapping," which is any production of more productive capacity, from less. T. McKendree [personal communication, 1999] likens the situation to an ant colony: "There necessarily is the line of reproducing queens, making a sufficient number of ants feasible; [however,] most of the queen's products are 'workers' that perform useful functions, but themselves are incapable of reproduction."

It cannot be emphasized too strongly that mechanical medical nanodevices will not self-replicate inside the human body, nor will they have any need for self-replication themselves (Section 1.3.3). Machines that perform medical tasks are fundamentally different from machines that manufacture other machines. As R. Merkle explains:²⁸⁷³ "While self-replicating systems are the key to low cost [manufacture], there is no need (and little desire) to have such systems function in the outside world. Instead, in an artificial and controlled environment they can manufacture simpler and more rugged systems that can then be transferred to their final destination. Medical devices designed to operate in the human body don't have to self-replicate: we can manufacture them in a controlled environment and then inject them into the patient as needed. The resulting medical device will be simpler, smaller, more efficient and more precisely designed for the task at hand than a device designed to perform the same function and self-replicate. This conclusion should hold generally: optimize [product] device design for the desired function, manufacture the [product] device in an environment optimized for manufacturing, then transport the [product] device from the manufacturing environment to the environment for which it was designed. A single device able to do everything would be harder to design and less efficient."

In an effort to stimulate scientific and engineering interest in constructing the first nanoassembler, in November 1995 the Foresight Institute created the $250,000 Feynman Grand Prize,²⁸⁶⁰ with funding contributed by Zyvex founder James von Ehr and St. Louis venture capitalist Marc Arnold. The prize will be awarded to the individual or group that first achieves both of two significant nanotechnology breakthroughs -- first, the design and construction of a functional nanometer-scale robotic arm, and second, the design and construction of a functional 8-bit adder computing nanodevice. According to Grand Prize rules, the robotic nanomanipulator must fit entirely inside a 100-nm cube, carry out actions directed by input signals of specified types, be able to move to a directed sequence of positions anywhere within a 50-nm cube, complete all directed actions with a positioning accuracy of 0.1 nanometer or better, and perform at least 1,000 accurate, nanometer-scale positioning motions per second for at least 60 consecutive seconds. The adder must fit entirely within a 50-nm cube, and must be capable of adding accurately any pair of 8-bit binary numbers, discarding overflow, accepting input signals of specified types, and producing its output as a pattern of raised nanometer-scale bumps on anatomically precise and level surface. (J.S. Hall notes that a conventional 8-bit adder may be constructed using a total of 94 AND, OR, and NOT gates; using XOR gates, the total may be reduced to 37 gates.) Both devices may accept inputs from acoustic, electrical, optical, diffusive chemical, or mechanical means, although any mechanical driving mechanism used for input must be limited to a single linkage that either slides or rotates on a single axis. To demonstrate the capacity for mass production, at least 32 copies of each device must be provided for analysis and destructive testing by judges.

Once nanoassemblers are available, the design of far more complex nanomachines each containing tens or hundreds of billions of precisely arranged atoms will require new molecular CAD (Section 2.4.1) tools and techniques, including automated hierarchical design decomposition of objects (that must be built up from an array of nanoparts) and a shape description language.¹⁰ Nanosystem design compilers, conceptually similar to silicon compilers that generate a complex pattern of transistors and conductors from an abstract specification of the properties of a digital circuit, will also be required for efficient nanorobot design.¹⁰ Given a fully specified design that meets all constraints necessary to physically permit assembly, all operations of the assembly process must be specified using an assembly process compiler. Explains Drexler:¹⁰ "The design of structures and assembly procedures by hierarchical decomposition directly generates a tree of assembly steps. If this tree has been chosen to generate parts of the appropriate sizes and numbers, then it can be mapped onto a manufacturing layout. The vast number of manipulator motions to be specified at the finest, earliest, and most-dispersed levels of the assembly process can be planned in parallel by identical software systems running almost independently on separate processors. The result of this parallel assembly process compilation is a set of instructions that, when executed by the transportation system and manipulator controllers of a manufacturing mechanism, will result in the assembly of an object corresponding to the initial design." Scale-up to larger systems, including the control, coordination, and programming of large aggregates of cooperating nanorobots, is discussed in Volume II.

Since molecular manufacturing systems can be used to make more molecular manufacturing systems, it is believed that the capital costs of production can be quite low. Drexler⁷³ claims that an analysis of inputs, outputs, and productivity suggests the total cost of production can be in the range familiar in agriculture and in the production of industrial chemicals -- on the order of tens of cents per pound. Merkle²⁶² has made similar assertions, for example: "Eventually (after amortization of possibly quite high development costs), the price of assemblers (and of the objects they build) should be no higher than the price of other complex structures made by self-replicating systems. Potatoes -- which have a staggering design complexity involving tens of thousands of different genes and different proteins directed by many megabits of genetic information -- cost well under a dollar per pound." Such analogies from conventional agriculture or industrial chemistry are not strictly applicable to the medical products and pharmaceutical areas, where structural precision, product repeatability, and intensive quality control are of supreme importance, and where most products are subject to rigorous government regulation -- unlike potatoes, which vary widely in size, shape, and molecular constitution, and generally do not require stringent product liability insurance, multiphase clinical trials, or FDA approval.

Still, it is likely that molecular manufacturing will eventually allow reasonably inexpensive production of large batches of designed medical nanorobots. As a very crude analogy, even in 1995 a typical mail-order bioscience vendor²⁸⁸⁷ could manufacture 15 micromoles (~9 x 10¹⁸ product items) of customized oligonucleotides for $18/base up to at least 110 bases with a convenient 48-hour turnaround on orders; other online vendors offer similar or slightly higher prices up to 200-mer sequences with additional charges for necessary purifications.²⁸⁸⁸ If a billion-base structure could be induced to self-assemble (Section 2.3.1) from 10 million different sets of 100-mer custom-built oligonucleotides (total ~20 billion atoms), forming a desired ~20-billion-atom nanomachine, then, optimistically assuming a ~100% yield, the cost would be ~$2 x 10^-9 per nanomachine or ~$2000 per ~1 cm³ of product volume which would include 10¹² finished nanomachines. In 1998, many specialty biotechnology-related drug treatments were comparably expensive. For example, treatments for multiple sclerosis using two closely related ~22.5 kilodalton interferons were administered in dosages of ~1 milligram/week at a treatment cost of ~$10,000/yr,²⁹⁵⁰ which is equivalent to a treatment cost of ~$2 x 10^-9 (2 "nanodollars") per giga-atom. The famous HeLa (Henrietta Lacks) cell line²⁸⁹⁰ is an example of a once tiny population of unique cancerous cells that were purposely replicated in vitro (~24-hour replication time) over many decades and are now widely available worldwide at very low cost (e.g., $216 per 1-cm³ ampule of the ATCC CCL-2 HeLa cell line²⁸⁹¹). Swallowable therapeutic pills containing bacteria (i.e., natural biological nanomachines) are already widely available over the counter for gastrointestinal refloration, as for example Salivarex which "contains a minimum of ~2.9 billion beneficial bacteria per capsule",³⁰⁴⁸ and Alkadophilus which "contains 1.5 billion organisms per capsule",³⁰⁴⁹ both at a 1998 price of ~$0.2 x 10^-9 per bacterium.

Another useful analogic approach to cost estimation is suggested by the plunging price of computer power during the 20th century. Figure 2.36 shows the cost, in constant 1998 U.S. dollars, of a computer processing capability equivalent to ~1 TeraFLOP (10¹² floating-point operations per second, ~10¹⁴ bits/sec assuming 64-bit words). The plotted data is from a compilation of 67 historical computers by Moravec¹ through 1989, with an additional 27 data points for other computers spanning 1973-1999 added by the author. The chart shows that the cost of 1 TFLOP has fallen from $2 x 10²⁰ in 1908 (the mechanical Hollerith Tabulator) to just $6 x 10⁶ in 1998 (the SGI/Cray T3E-1200E), a 90-year price decline of almost 14 orders of magnitude -- on average, a halving every ~2 years, a century-long example of Moore's Law. This trend may give us some confidence that even if the first individual 1 micron³ nanorobot costs ~$100,000 to assemble (much like the first Nippondenso microcar; Section 2.2), eventually the unit price may fall at least 14 orders of magnitude to below ~$10^-9 per nanorobot or <$1000 per ~1 cm³ dosage (10¹² nanorobots). Recycling or remanufacturing each nanomachine ~1000 times before final discard would then imply a net ~$1/cm³ treatment cost. Of course, in 1998 all such estimates were mere guesswork, but the ultimate expectation of relatively low-cost nanomedical treatments was certainly rational, if not provable.

This Chapter has demonstrated that there are many different pathways leading toward molecular nanotechnology, each one of which provides incremental benefits to motivate further travel down the paths (Fig. 2.37). The likelihood of all the paths being blocked is low, even if we have little confidence that we can predict which approach will succeed first.^8,9 Progress along the many pathways will provide precursor products, including some early medical applications, but the ultimate achievement of molecular manufacturing will finally make nanomedicine feasible.

As Drexler concluded Nanosystems:¹⁰ "Each step along [these pathways] will present great practical challenges, but each step will also bring valuable new capabilities. The long-term rewards, measured in terms of scientific and technological capabilities, appear large." This author agrees. For the remainder of the present work, we shall assume that a molecular manufacturing technology will someday exist that can economically manufacture macroscopic batches of microscopic machines comprised of atomically-precise nanoscale components. The medical applications and implications of such a technological capability are the principle subject of this three-volume book.

Last updated on 27 August 2011