from Proceedings of the Third International Conference on Supercomputing, International Supercomputing Institute (1988)
Abstract. The advance of technology in chemistry, biochemistry, and micromanipulation is carrying us toward the ability to build complex molecular structures, including molecular machines and molecular electronic devices. Although molecular electronic computers will surely be faster, molecular mechanical computers are easier to design and analyze. Their capabilities set lower bounds to the performance of supercomputing systems in the coming era of nanotechnology.
Nanotechnology will be based on programmable machine tools—termed assemblers—which will enable the construction of a wide range of molecular structures. Living systems demonstrate that molecular machines can build molecular machines. Some enzymes assemble small reactive molecules to build larger molecules, and ribosomes act as genetically-programmed machine tools, assembling small reactive molecules in complex patterns to form large molecular machines.
Ribosomes build machines only from protein, but a molecular-scale robot
arm, working with a wide range of reactive molecules, should be able to
build molecular machines with almost any chemically-reasonable structure
(some limits are described in ). Synthetic organic chemists make a wide
range of molecular structures by mixing reactive molecules in solution.
Characteristically, however, they cannot make very complex structures (say,
a molecular object with the complexity of a computer), owing to the difficulty
of controlling the site of a reaction on the surface of a large molecule.
Liquid-phase diffusion bumps molecules together in all possible positions
and orientations; wherever reactions are chemically feasible, they occur.
Assemblers will sidestep this limit by eliminating diffusion: they will
position reactive molecules mechanically, making reactions occur only at
the sites selected by the designer.
Figure 1: A block of diamond with a volume slightly under one cubic nanometer (block surface planes fall between atomic planes). Contains 176 atoms.
First-generation assemblers may be developed through protein engineering; biochemical analogies indicate that protein engineering (when sufficiently advanced) will enable the design and fabrication of complex, self-assembling molecular machines [1] (relevant work is discussed in [4,5,6,7]). Likewise, first-generation assemblers may be developed through the synthesis of self-assembling sets of non-protein molecules (relevant work is discussed in [8,9,10]). Alternatively, advances in micromanipulation may enable the construction of first-generation assemblers through mechanically-directed molecular assembly; reports of atomic rearrangement through field-induced evaporation from scanning tunneling microscope (STM) tips [11] and of highly localized chemical reactions induced by currents at an STM tip [12] are suggestive in this regard. In practice, development may well involve a combination of chemical, biochemical, and micromechanical techniques. But however assemblers may first be built, later assemblers will be built using assemblers. The nature of nanotechnology and its capabilities will then be independent of the nature of proteins, conventional chemistry, and initial micromanipulation technologies.
Since multiple paths lead to molecular machines and nanotechnology,
no one development problem is likely to block advance in this direction.
With multiple paths, multiple research groups, and multiple chains of short-term
rewards along each path, it is (in a competitive world) hard to imagine
that nanotechnology will not eventually be realized. This adds to its interest
as a object of study.
Molecular machines will operate on the same scale of sizes and energies as molecular electronics [13], and are easier to design and analyze: Newtonian mechanics is adequate. What is more, they can be used as a basis for computers—for molecular Babbage machines, albeit with more modem architectures. This makes molecular machines of interest in understanding the future of computing.
A well-established nanotechnology is unlikely to use biomolecules as
machines. Just as engineers prefer to work with light, rigid materials
on a large scale, so nanoengineers will prefer such materials on a small
scale. Thus, parts will typically contain patterns of atoms like those
found in engineering plastics, ceramics, graphite, and diamond. To visualize
molecular mechanical devices, it is important to recognize that molecules
are objects, with size, shape, mass, strength, stiffness, and so forth.
Large machines are made of parts with many atoms; nanomachines will be
made of parts with few. The molecular machines envisioned for a well-established
nanotechnology are surprisingly conventional—they include electric motors
(electrostatic, rather than electromagnetic [2]),
gears and bearings [15], and a range of other
moving parts.
The scheme analyzed parallels CMOS logic, with movable rods analogous
to conduction pathways, rod motions analogous to currents, rod displacements
analogous to voltages, and mechanical locks analogous to transistors. Proposed
rods are of carbyne [17], a linear carbon
structure with alternating single and triple bonds. Knobs (Figure
2) enable rods to interact: rods slide in channels, and a gate knob
can block or unblock the motion of a probe knob (Figure
3). A rod that is free to move when pulled is in a 1 state; a locked
rod is in a 0 state. The speed of sound in carbyne rods (when weighted
by knobs) is some 17,000 meters per second, allowing a signal to propagate
0.1 micron in about 6 picoseconds. The chief breakdown of the analogy with
CMOS circuits is that rods must be reset to their original positions in
a distinct series of operations.
Figure 2: Probe knob (left) and gate knob (right), with carbyne rod
segments
Figure 3: Configuration of knobs in a lock with (right) and without (left) the associated matrix and channels. Probe knobs are above, gate knobs below.
Work remains to be done in the design of several elements. These
include the non-moving matrix that surrounds the channels, the drive system
that links an electrostatic motor to the oscillating rods, springs to establish
restoring tensions and essential degrees of compliance in rod systems,
and a carry chain for the ALU (systems analogous to the Manchester carry
chain seem feasible). Even in the absence of these, rough estimates of
many system parameters can be made with some assurance [16].
The delay associated with the motion of a single rod (having many inputs and up to 16 outputs) is about 50 picoseconds. Based on this, a plausible cycle time for a CPU is about one nanosecond. Power dissipation is difficult to estimate accurately, but appears to be less than kT per gate per operation in reversable combinational logic (dissipation lessens at lower speeds); at room temperature, kT is about 4x10-21 joules. The rate of soft errors (resulting from thermal noise) can apparently be kept below one in 1012 logic operations.
The volume of a CPU can be estimated by scaling from semiconductor technology. Roughly speaking, a logic rod is equivalent to a conduction path, but its 0.3 nm diameter is some 10-4 the width of a 3-micron line. Locks and transistors scale comparably. Applying a 10-4 scale factor to the active volume of a CPU chip with 3-micron line width (and pretending that lines, like rods, are as thick as they are wide) leads to the conclusion that an equivalent nanomechanical device could fit in a few thousandths of a cubic micron.
Design concepts for RAM cells suggest RAM storage densities of roughly 1020 bits per cubic centimeter. Design concepts for molecular tape memory suggest tape storage densities of roughly 1022 bits per cubic centimeter. With tape speeds on the order of a meter per second, seek times for tapes of reasonable size can be fractions of a millisecond.
Compared to hypothetical electronic nanocomputers, these mechanical devices would be at a disadvantage in speed, but might have advantages in compactness and energy efficiency. Their only clear advantage is in ease of design and analysis—a crucial factor, when attempting to estimate the capabilities of technologies that are still far from implementation.
will surely use electronic systems for their time-critical operations. Still, mechanical nanocomputers set a lower bound on performance for massively parallel system elements.
Consider the computational capacity that can be packaged in a cubic centimeter. Allowing half the volume for cooling channels and communications, the remainder holds room for half a trillion complex CPUs, each with several hundred megabytes of fast tape, a half-dozen megabytes of RAM, and a potential clock speed of about a gigahertz. Passing cooling water through the system at a rate of three liters per minute with a temperature rise of 20 kelvins provides 4 kW of cooling capacity. Given a power estimate of kT per gate per operation, this cooling rate can support 1024 gate-operations/second. Such a system would contain more computational capacity than all the computers in the world today.
