Pages 549-571 in Molecular
Electronic Devices II, Forrest Carter (ed.), Marcel Dekker, New
York (1987)
Probably from the 2nd International Workshop on Molecular Electronic
Devices, 13-15 April 1983,
Naval Reseaerch Laboratory, Chemistry Division, Washington DC.
Like ordinary electronics and machinery, molecular electronics and molecular machinery promise to be closely linked in developmentand use. They will rest on a common technology base, the ability to assemble structures to complex, atomic specifications. Molecular machines can help assemble molecular electronics, molecular electronics can help direct molecular machinery, and both can be used in hybrid systems. They thus seem likely to emerge together, grow together, and influence each other strongly.
Modern microelectronic systems control macroscopic machines, despite mismatches in scale and power level. Macroscopic machines make microelectronic systems, despite the mismatch of scales and the need for great precision. Microelectronic technology has even brought a limited micromechanical technology (1), though current techniques cannot fabricate the range of moving parts found in ordinary machines.
Molecular machinery, in contrast, will bring a full range of mechanical functions, and will match molecular electronics in both power and scale, encouraging closer interfacing. Accordingly, molecular machinery seems important to the future of molecular electronics. Though molecular electronics can make impressive progress through chemistry, microtechnology, and biological-style self-assembly, advanced molecular electronics will likely be built by molecular machinery, because it will eventually provide the best tools for the job.
One class of protein machines could hold a molecular workpiece and bring
molecular tools to bear, assembling reactive groups in a site–specific
manner. Such machines could be programmed by molecular tapes, like early
numerically controlled machine tools (programmed with punched tapes) or
like the ribosome (programmed with mRNA). With a variety of tools under
programmed control, such a machine could assemble a wide range ot molecular
structures to atomic specifications.
These structures could themselves serve as machine parts having properties (strength, thermal stability, etc.) not found in the coiled, hydrated polypeptide chains of protein. Engineers could use such components in second–generation machines, including machines for assembling reactive groups to workpieces; with proper design, they could assemble atoms into virtually any stable pattern (2). Machines with this capability may be called assemblers.
In 1959, Richard Feynman outlined an alternative path to this capability, based on the use of larger machines to build successive generations of smaller machines (7). He stated, "... It would be possible (I think) for a physicist to synthesize any chemical substance that the chemist writes down.... How? Put the atoms down where the chemist says, and so you make the substance."
Where technology based on bulk processes is often limited by fabricated difficulties, technology using assemblers will be limited chiefly by physics and by our ability to design things.
Assemblers will naturally find use as a direct and versatile means for building molecular devices and systems. Regardless of the assembly techniques used in early MED, construction of molecular machinery to aid future assembly will rapidly become attractive. For the present, the prospect of assemblers able to build almost any stable configuration of atoms should encourage theoretical work on MED to consider a broader range of structures than might otherwise be of interest.
Discussion of the future of molecular technology can, however, avoid scientific speculation. Where a technology has close chemical, biochemical, or macro-mechanical analogies, or where motions can be calculated by Newtonian mechanics, possibilities can be firmly established. Certain statements may then be described, not as scientific speculation, but as engineering projection. The price of limiting discussion to projection, of avoiding speculation, is that one must omit many interesting ideas, some of which may well prove true. The reward, however, is that by using solid, dull bricks one can build a firm structure of greater size. Where speculation piled on speculation rapidly weakens, firm projections can often be combined to form the basis of further projections.
An historical example illustrates this. In the 1930’s (and before), principles known to rocket designers showed the possibility of flight into space. Indeed, knowledge of the strength of materials and the energy content of liquid propellants guaranteed its essential feasibility. Knowing that orbit could be achieved, it was clear that space stations could be built and ships assembled to reach the Moon and beyond. These projections of possibility were correct, though the historical path to the Moon differed from anyone’s detailed speculations.
In a similar fashion, the principles of chemistry and molecular biology now show the possibility of certain classes of molecular machinery. Assemblers form one such class. Their mechanisms would be analogous to those of ribosomes, enzymes, ordinary chemical reactions, and existing molecular machinery. They, too, will open a new frontier: they will make possible a technology limited not by fabrication problems, hut instead by the laws of physics and by our ability to conceive and design.
Since assemblers could build virtually arbitrary molecular structures, systems based on assemblers could, if properly supplied and programmed, build copies of themselves. Like ribosomes in bacteria, assemblers could form the core of replicators; also like ribosomes, they could be made cheaply in bulk through replication. With inexpensive assemblers, many products could be made for energy costs comparable to those of biological materials (2) such as bacteria and wood.
This makes non-electronic products of interest. Diamond, for example, consists of properly arranged carbon atoms and should be inexpensive to manufacture with molecular machinery; with a tensile strength of about 50 giganewtons per square meter, it has roughly 50 times the strength to mass ratio of the aluminum used to build the space shuttle. In an atomically tailored fibrous composite, diamond could yield an exceedingly strong, tough engineering material.
Unlike passive structural materials, muscles can contract by means of molecular machines (the actin/myosin system) that pull fibers past each other. Analogous artificial molecular machines presumably could draw energy from various electrical or chemical sources and could exceed the strength and operating temperature of muscle proteins; the fibers they pull could have great strength. Strong, rugged muscle-like materials thus seem possible.
Assemblers could also make optical fibers with extremely low absorption, scattering, and dispersion. Detailed control of atomic patterns eliminate absorption caused by impurities, and use of a crystalline structure could eliminate scattering caused
by the inhomogeneities inherent in glassy materials. Finally, free use of concentric layers of materials with different dispersion relationships could permit the dispersion curve to be more accurately flattened in a particular frequency band.
These examples describe but a few of the materials properties and device characteristics that could be improved through use of assemblers (novel devices will be made possible as well). Indeed, it is difficult to think of a form of hardware that could not be improved by removing fabrication constraints, that is, by making possible more nearly optimal arrangements of atoms. Assembler technology will bring a fundamental revolution reaching far beyond the field of molecular electronics; the related technology of molecular machinery, however, will be integrated with MED in important ways.
As currently used, "molecular electronics" covers both essentially electronic devices, such as switches based on tunnelling (9), and devices involving substantial mechanical degrees of freedom (10). Since effects such as tautomeric conformational changes and soliton motion (11) involve mechanical degrees of freedom, and since assembly of molecular devices is an inherently mechanical process, proposals for molecular electronics already involve a measure of molecular mechanics.
Molecular machinery can help build purely electronic devices. Assemblers can make precise connections between conducting one–dimensional polymers and switching groups, providing a general solution to the problem of connecting molecular electronic devices to form molecular electronic circuits. Further, they can improve electronic devices which remain stylistically similar to the conductor/aemiconductor/ insulator structures now used.
Examples of the latter might involve very thin, crystalline wires with smooth surfaces, replacing surface scattering with specular reflection, eliminating one source of increased resistance in thin wires. In semiconductors made with similar precision, doping atoms could be placed in regular arrays to tailor scattering properties. Further, use of materials with large bandgaps (such as sodium chloride, with a bandgap over 7.5 eV) should permit use of thinner insulation while keeping tunnelling currents within acceptable bounds––the use of such materials in stable devices will be aided by the ability, implicit in the abilities of assemblers, to hermetically seal circuits. Finally, circuits could be arrayed in three dimensions (to shorten signal transmission times) and could be interwoven with a branched system of tubes, circulating cooling fluid in a pattern based on the principles of heat and mass transfer underlying the artery-capillary-vein system in the mammalian circulatory system.
MED and MMD will share similar scales and similar characteristic energies. The energy required to move an arm on a molecular assembler by a nanometer in a nanosecond, for example, would be about an electron volt–also a reasonable energy for signal processing in molecular electronics. Electronic changes could be coupled to mechanical motions either through direct application of electrostatic forces, or through rearranging bonds and changing molecular conformations. Like electronic devices, molecular machines lack wear mechanisms (though damage mechanisms remain); further, they can operate as swiftly as today’s microcircuits (if not as swiftly as tomorrow’s molecular circuits).
These facts make molecular machinery and electronics extraordinarily compatible. Although differences of scale and energy preclude direct and intimate interfaces between, say, macroscopic robot arms and microcircuits, ordinary electronics and machinery are nonetheless combined; greater compatibility may encourage more intimate hybrids among the molecular analogs of macroscopic technologies.
In one class of hybrid system, molecular machinery could be used to reconfigure circuitry that would then compute by means of faster, purely electronic effects. If one identifies mechanical systems with ionic degrees of freedom, then a simple example of this would be any use of a conformation change to reconfigure a molecular circuit that then relies on electronic switching to perform actual computation.
Software has long been used to generate software, through compilers; recently, software has been used to generate hardware designs through so–called silicon compilers. Future systems might incorporate both molecular–scale manipulators and building–block circuit elements, using a "circuit compiler" to direct assembly of custom logic circuits to relieve bottlenecks that might appear in a repetitive computation with more general–purpose circuits. Full customization might involve assemblers using reactive molecules as building blocks; swifter construction (but slower computation) would result from use of prefabricated gates, wires, and so forth. The small scale of molecular technology would allow such compilation to hardware inside a user’s machine, in response to a user’s programs.
This sort of integration of MMD and MED can also ease problems of device failure resulting from radiation. As has been noted, redundancy and error correction will be needed to ensure system reliability (12); repairs cannot substitute for this, since they can neither compensate for circuit noise problems nor negate the effects of damage in mid–computation. Repairs will he useful, however, to prevent accumulation of damage. ibis might
be achieved by means similar to those proposed above for circuit assembly and reconfiguration; the ability to assemble and disassemble a system, combined with knowledge of the structure desired, implies the ability to perform repairs.
A simple calculation provides a lower bound on information storage density. Given abilities less general than those of assemblers, a device could add and remove R-groups to a polyethylene molecule. Assume, for the sake of concreteness, that the side groups are hydrogen and fluorine; in a properly-oriented, partially-fluorinated polyethylene molecule, R-groups can then store two bits per carbon atom. Such molecules could serve as tape memory, analogous ta punched paper tapes or RNA; bits could be read by sensing the size of the R-group with a mechanical probe. Approximating the bulk density of this polymer as the average of those of polyethylene and polytetrafluoroethylene, and allowing a factor of three for the volume of reel, drive, and read/write apparatus, tape memories could store over 15 bits per cubic nanometer.
Access times may also be estimated. Tape speed could be comparable to that in macroscopic devices; to first order, energy densities and dynamic tape stresses remain constant regardless of scale. If a reasonable tape length is a micron and a reasonable speed is a meter per second, then maximum access times would be about a microsecond. At room temperature, the kinetic energy per atom in the moving tape would be on the order of 1/10,000 kT. Such a tape drive would store about 16K bits, and be roughly 10 nanometers across.
In connection with rotating parts, an upper bound can also be set on the energy dissipated in a sigma–bond bearing. Consider hindered rotation: in the unrealistic worst–case, the energy dissipated per rotation would be the amplitude of the potential function times the number of peaks (in practice, of course, energy would be recovered in traveling from peak to valley). For ethane, the rotational barrier is about 12.0 kJ/mole (13), and the molecule has threefold symmetry; the resulting upper bound on energy dissipation is about 0.4 eV/rotation. This is substantial, and even with a more realistic model for energy dissipation, the associated torque may be too large. Fortunately, separating the two methyl groups in ethane with a pair of triple-bonded carbon atoms (to make dimethylacetylene) reduces the rotational barrier to a listed value of 0.0 kJ/mole (13); even taking this as 0.05 kJ/mole, both energies and torques would be low, resulting in good bearing properties. Some carbon–carbon sigma bonds also have rotational barriers listed as 0.0 kJ/mole.
Operation of the memory depends on rod motions; bits are encoded by slider positions. The selector rod determines which row the read/write rods respond to: if it is in the position shown by 10, the read/write rods can move regardless of slider positions; if it is in the position shown by 9, then a forward slider (such as 1) suffices to clamp the read/write rod, blocking motion of the knob past the resulting constriction. The gap between 9 and 2, however, remains wide enough to allow the read/write rod to move upward. Thus, the response of a set of read/write rods to upward tugs depends on the row selected and on the positions of the sliders along that row.
To change slider positions in a row (to write), the corresponding write-protect rod (such as 5 or 6) is moved away from the position shown, placing its knobs in the spaces between the sliders and permitting them to move. If the sliders are biased to move forward unless pushed back (perhaps by the pressure of a few trapped gas molecules), all will then move to the "clamping" position, and could be locked there by returning the write-protect rod to its illustrated position. If the row-selector rod is in the activated position (like 9), however, and the write-protect rod is in the unlocked position (unlike 5), then lifting a read/write rod (like 8) will push the corresponding slider back; it can then be locked in this position. Thus, manipulation of the three sets of rods can both change and determine the position of any given slider, permitting the system to function as a random-access memory.
This scheme is abstract; it could be implemented with rods of tensioned steel cable linking cast-iron blocks. A plausible molecular implementation, however, would use tensioned carbyne chains as the rods. Carbyne, a pure carbon polymer with alternating single and triple bonds, seems the best material to use for thin, longitudinally-stiff rods. Ihe knobs on the write–protect and row–selector rods could then be phenyl groups, and those on the read/write rods bicyclooctane groups; the sliders might be halogenated 9,10–dihydroanthracene molecules. Indeed, in all cases certain hydrogen atoms would be replaced with bulker groups (such as fluorine, chlorine, or methyl moieties) to provide the proper steric hindrances. This approach results in a RAM cell some 1.4 by 1.4 by 2.5 nanometers on a side, or roughly 5 cubic nanometers: it is thus some 75 times less dense than the tape memory described above, but offers faster access times (see below).
Memory cell volumes can also be estimated through comparison to present memory chips. As a crude approximation based on round numbers, take a typical line width as three microns and the corresponding molecular rod diameter as three angstroms. Since the molecular rods are as thick as they are wide, consider the memory chip’s active circuits to be three microns thick, for comparative purposes. Allowing a chip area of one square centimeter, the total equivalent active volume of the chip is 300,000,000 cubic microns; this scales to 300,000,000 cubic angstroms for the molecular mechanical memory. Assuming that the chip stores 64K bits, the volume per bit is 4,700 cubic angstroms, or 4.7 cubic nanometers. This result, based on the first set of numbers that came to mind, is closer to 5 cubic nanometers (the result of the design exercise above) than one has any reason to expect.
This method, scaling from integrated circuits, can be used to estimate the volume of mechanical random–logic "circuits". If anything, the move from two to three dimensions should give a relative advantage in "wiring" molecular machines, so this method may tend to overestimate the volume of molecular random logic. Since simple CPU’s have for years occupied a single chip, this rough scaling relation (10,000-to-one in linear dimension, one–trillion–to–one in volume) suggests that the molecular mechanical equivalent of a simple Cpu should fit in roughly 0.0003 cubic microns, a volume less than 0.07 microns on a side.
This result may be surprising, but contradicts no calculations of which I am aware; nevertheless, it deserves a closer look. In particular, can estimates be made of the effects of thermal fluctuations on the computational elements, and can estimates be made of computational speed?
To limit errors arising from thermal fluctuations and mechanical noise, a substantial energy must be required to deform a rod excessively. For a suitably designed system, "excessive" deformation (the deformation required to move a bulky group out of the way of another moving part) might be about one nanometer; if the elastic energy for this process is 1 eV or more, thermal errors (at least) will be rare. A carbyne rod can approach 0.5 microns in length before the elastic energy required to stretch it by a nanometer falls below 1 eV. This gives an estimate of maximum rod length; reliable signal transmission over greater distances could be accomplished through use of relays or thicker rods. In RAN arrays of the sort described above, however, a variety of factors (such as friction) may limit the size of a block to less than 0.5 microns.
Signal transmission over these distances can he relatively swift. The speed of sound in carbyne is roughly 30 km/a; an acoustic signal would thus cross 0.5 microns in less than 20 picoseconds. Application of a maximal accelerating force (10 nanonewtons) to a 0.5 micron rod would move it a nanometer in about 4 picoseconds if it could respond as a rigid body.
Allowing 0.2 nanoseconds for the motion (thus simplifying dynamics by allowing acceleration over 10 times the acoustic signal time) would require an accelerating force of only 4 piconewtons (about 0.00025 of the breaking stress) and a peak speed of 10 m/s (implying a kinetic energy of about kT at room temperature). Thus, a properly designed mechanical system could apparently yield subnanosecond gate delays.
These estimates omit much mechanical detail, including supply of power, triggering of one action by the next, resetting of spring–loaded parts, and so forth. It seems clear, however, that a sliding framework of rods could transmit mechanical power and synchronization into the volume of a computational device. Even if the volume of such a power transmission system substantially exceeded that of the computational elements, this would make little practical difference; a factor of ten expansion in volume would place active elements less than a factor of 2.2 farther apart; a tenfold delay in signal propagation would require a thousandfold increase in volume. Most conclusions one might draw are insensitive to assumptions regarding such specifics.
The carbyne–rod based approach to building a RAM array (and random logic) may be considered somewhat speculative. Though a variety of calculations support its plausibility, it is a specific enough concept that some specific problem could conceivably invalidate it. Like most molecular mechanical schemes, however, it is simply a molecular–scale implementation of a concept that can clearly work on a macroscopic scale. Thus, by making thick enough rods, using large enough forces, and so forth, one could modify the design to overcome any problems that might emerge. The plausibility of this approach at the scale described, however, suggests that such modifications need not change the scale dramatically, and that they may well be unnecessary.
This illustrates one of the strengths of engineering projection compared to scientific speculation: engineers can specify conservative lower bounds on performance and begin with tentative conceptual designs. As they firm up a design, engineers can correct their mistakes and produce a working system that does what they projected to be possible. Science, in contrast, generally demands falsifiable statements about specific systems and hence cannot use conservatism and redesign to make speculations come true. Thus, statements in the realms of science and engineering can differ fundamentally.
In the most straightforward approach to computation, a certain free energy would be expended per step––perhaps a certain multiple of kT per bit per gate, chosen to ensure a certain reliability against reversal by thermodynamic fluctuations. This would make the energy cost of a computation proportional to its complexity.
It is possible, however, to devise constraints such that a mechanical system will always represent a state in a computation (15); such a computation could be reversible in the way that diffusion of a gas molecule from one end of a tube to the other is reversible. In a system of this sort, a substantial fluctuation would be required to disrupt the computation, but otherwise the process could move forward and backwards by random walk. The energy coat of a computation would thus depend on both its complexity and speed; free energy would be expended only to bias the movement of the computation in the desired direction, to force the computation forward at the desired speed.
Finally, it has been demonstrated that, with Newtonian mechanics, perfect initial conditions, and complete decoupling of thermal and mechanical modes, a computation of arbitrary complexity can proceed at a constant speed with a net input of free energy depending only on the number of bits in the input and output (16). Though this behavior almost certainly cannot be achieved in practice, its consistency with the laws of thermodynamics suggests that energy dissipation in computation could be far less than a naive look at NAND gates (which are not reversible elements) might suggest. Until practical implementations of reversible computation are suggested, however, conservative estimates should probably assume the dissipation of energies in excess of kT per gate per operation; in the scheme outlined above, for example, substantial amounts of energy will be dissipated in friction.
Shrinkage of computers from the room-sized and cabinet sized machines of the 1950’s and 60’s opened an era of personal and pocket computers. Shrinkage of computers with a gigabyte of secondary memory to less than a cubic micron (as indicated by the above calculations) will bring still greater changes––it would be absurd to think of their applications solely in terms of devices in cabinets with keyboards, or even in terms of packages bolted to cars. An obvious area of application for molecular electronics (and perhaps molecular mechanical computers) will be In providing local control of complex processes on a small scale, such as control of molecular machinery interacting with biological systems. This seems an appropriate result; the path outlined above begins from biotechnology, which itself has been funded largely because of its application to medicine.
The last section used calculations based on Newtonian mechanics to explore the abilities of molecular machinery; biological applications lend themselves to another line of argument, to the use of existing biological systems as feasibility proofs. Projections in this area are of interest, both as an exercise in reasoning about the limits of future technology, and for what they may show about the incentives for developing molecular technology.
Consider biochemical and biological analysis. The ability of antibodies to distinguish among proteins, and of cellular molecular machinery to distinguish among all functionally distinct molecular species (a virtual tautology), shows that molecular machinery can sense molecular structures. Further, if it can report its own position during sensing, it can report the spatial distribution of different molecules. This line of reasoning leads to the conclusion that suitably designed molecular machinery, if interfaced to molecular computers and data storage devices, could characterize the molecular structure of cells in virtually complete detail. Procedures ranging from molecule-by-molecule disassembly of preserved cells to the insertion of molecular-scale probes into functioning cells could provide a general, direct solution to the data collection problem in cell (and hence tissue and organ) biology.
Consider biochemical and biological synthesis. The ability of molecular machines in cells to build all the molecular components of cells is well established–they do this every time a cell divides. Again, since existing molecular machinery sets a lower bound on the capabilities of artificial molecular machinery, it is clear that the latter could build or rebuild all or part of a cell.
Consider biochemical and biological repair. The ability to sense a structure, to disassemble it, and to reassemble it implies the ability to repair it, provided that the desired structure is known. Examples of sensing and reassembly were described above; disassembly is demonstrated by digestive enzymes (and by harsher chemicals). Knowledge of correct cellular structures can be obtained through study of healthy tissue. Molecular tools are clearly adequate to repair cells, since all of the elementary operations required parallel those observed in nature; the major remaining issue is computation.
The volume of a typical mammalian cell is roughly 1,000 cubic microns. The calculations above indicate that a simple CPU would occupy less than 0.001 cubic microns, and that 0.1% of the volume of a cell could contain both a powerful computer and secondary storage holding more information than the cell’s own DNA. Thus, considerable on–site computational power could be brought to bear. The combination of a computer with a variety of molecular sensors and tools, suitably programmed, could be called a cell repair machine.
Cell repair machines seem a natural outgrowth of the development of molecular machinery, molecular electronics, and improved biological instrumentation. Indeed, pharmaceutical design already provides a driving force behind molecular engineering (17). Cell repair machines will nonetheless bring a profound change in medicine. Neither drug therapy, nor radiation, nor surgery can heal tissue: medicine today can only bring about conditions under which tissue can heal itself. Physicians today lack the tools needed to repair molecular machinery; attempting to repair a cell with a scalpel and an injection would be like trying to repair a watch with an axe and a drop of oil. Accordingly, medicine must now at all costs attempt to preserve tissue function, since without it, healing is impossible. With cell repair machines to perform healing, however, the emphasis in emergencies will shift to preserving structure, particularly in diseases (e.g., stroke) that destroy unique tissue patterns important to the individual. Ultimately, there seem no barriers to repairing tissue so long as characteristic cellular structures remain intact; those that do not could still be replaced. The engineering challenges along the way are, of course, numerous.
By avoiding speculation about unknown scientific facts, one may
project future engineering developments, at least to the extent of setting
some lower bounds on what will be possible. Aided by chemical and biochemical
examples, Newtonian mechanics allows projection of molecular machinery
with a wide range of capabilities, including the ability to handle and
assemble individual atoms and molecules.
As one might expect, this ability will lead to broad synthetic abilities, making possible assembly of virtually any molecular Structure with chemically reasonable bonding. Assemblers will make possible construction of both molecular circuits and novel molecular replicators (as one might expect from the existence of bacteria). Calculations indicate that molecular mechanical computers with subnanosecond gate delays could fit in a fraction of a cubic micron. Biological analogies indicate that molecular devices (perhaps directed by such computers) could be used in cell repair machines able to recognize, disassemble, and rebuild cellular structures, thus effecting repairs.
In considering the future of MED, we must consider a broad range of molecular technologies that will mature with it. Molecular machinery can build MED’s and be directed by them; it can even perform traditionally electronic functions, such as computing. It will, however, bring revolutions of its own: replicators could revolutionize the world economy, or could be agents of destruction; cell repair machines could revolutionize medical care, or could be abused. To serve our common future interests, the emerging molecular technology research community can serve best by examining the possibilities carefully, judging them dispassionately, and transmitting its understanding to the public clearly. By distinguishing speculation from projection, considerable foresight seems possible. It may be necessary (18).
