In conjunction with Foresight Update 43
Recent Progress: Highlights from the Eighth Foresight Conference On Nanotechnology
By Jeffrey Soreff
The conference opened with a keynote address by Raymond Kurzweil on the overall long term trends in technology. He displayed exponential (or, in some cases, accelerating exponential) growth in a wide variety of areas, from feature sizes in integrated circuits, to decreasing costs for sequencing genes, to rates of increase in life expectancy.
He emphasized that, while Moore’s law for integrated circuit density looks likely to hit a wall in 10-20 years, it is really the 5th paradigm in a sequence of computing technologies that in aggregate show smooth exponential growth in the power of computers from the 1890’s to the present day. Kurzweil expects molecular computers to add a 6th paradigm to this curve.
Kurzweil viewed nanotechnology not really as a separate technique, but as an extension of longstanding and widely distributed miniaturization efforts in thousands of projects spread around the world. He emphasized that this broad, distributed trend is not one that could be stopped by Bill Joy’s call for relinquishing nanotechnology.
He noted that today, typical machines are a million times less complex than humans, but that this difference would evaporate in roughly 20 years. He commented that it is very difficult for people to internalize the implications of an exponential trend. We tend to awaken to effects when they are in their final doubling or two or three before saturating, when they are visible on a linear scale, even if they could have been anticipated from well before then. The Internet was disregarded by most until 1995 or so, even though its history goes back to 1966.
One enormous attractive application area that Kurzweil envisions is direct functional connections to the human brain, of sufficiently fine granularity and versatility that by 2100 there will essentially be a smooth continuum between humans and machines. His experience has been that when an area of the brain can be probed in sufficiently fine detail, there is a rapid progression from raw data to scientific models to reverse engineering to duplicating the function.
He illustrated this with examples from sound perception, where duplicating two subsystems (of the total of roughly 20 involved) has proved sufficient to recognize speech matching human performance, but using 100-1000 times less computation to do it.
He projects that by 2030 we will have the raw data to reverse engineer the full human brain, and that we functionally duplicate the hundreds of regions in it. This opens up a whole spectrum of possibilities. If we truly understand the architecture of the brain, rather than merely understanding individual neurons but viewing their interconnections as crucial but cryptic, then we have a much better chance of being able to transfer information between individuals. He suggests that we’d be able to transmit not merely primary sense perceptions, but a great deal of subjective experience as well. He projects that we will be able to obtain the same advantages that we now have in electronic memory: backups, copying, density, reliability, and speed, but integrate them into our own brains.
There were several good papers on experimental work in molecular electronics at the conference.
Charles Lieber described work with carbon nanotubes and III-V semiconductor nanowires, solid crystalline cylinders. He described the use of nanotubes as nanometer-scale tips for scanning probe microscopy, including CVD techniques for growing these tubes directly on AFM tips, bypassing the need to mount separately grown tubes on a tip. Nanotubes can be etched at an angle to the tube axis, yielding a single derivatizable benzene ring at the tip. Lieber showed images of IgG structure taken with nanotube tips showing regions and even binding sites in the protein.
Lieber described the construction of mechanically bistable storage elements built from perpendicular nanotubes. They have two states, one where the tubes are close enough Van der Waals attraction holds them together, bending the tubes to keep them close, and the other where the tubes are further apart and close to straight. These devices could be switched from one state to the other electrically, and their states detected electrically as well. Ideally, one wants these crossed devices to form a diode array, which would let their bits be written and read independently.
This is difficult with carbon nanotubes because we still don’t have reliable control over nanotubes’ chirality, which determines which are metallic and which semiconducting. Lieber has circumvented this problem by building these nanoscale mechanical flip-flops from semiconducting InP nanowires. These nanowires can be doped during synthesis to reliably produce n-type and p-type wires, with a 105-fold range in conductivity. By putting one nanowire polarity in the upper layer and the opposite one in the lower layer, all of the necessary diodes are automatically built into the memory array. Lieber showed a 2 x 2 array of these memory devices, which was addressable and functional. He also anticipates being able to make large arrays of them by using fluidics to align the nanowires (or nanotubes).
James Tour described experimental work on his poly-phenylene wires, describing the doubling strategy that allows constructing truly monodisperse linear 16 nm wires, including heterostructures. He described the synthesis of molecules with negative differential resistance, molecular tunnel diodes, including the nitro/amino substituted structure with a 1000:1 peak/valley current ratio.
Tour described the strategy they will be following to interconnect these elements. Rather than deterministically fabricating nanoscale wiring, they will be building “nanocells” with around 1800 switches in a square micron area. A small number of leads, perhaps 20, will connect to each nanocell. The cell will be programmed to realize a desired function by setting switches within it. Because a very complex logic function can be accessed through the leads to the cell, the number of effective gates supplied by the cell can be much larger than the number of leads.
James Ellenbogen described design work and theoretical analysis work on Tour-wire-based logic circuits. He described how a quantum mechanical interaction in an exclusive OR circuit actually makes switching sharper than the classical model would predict. He described work on molSpice, a simulator for molecular circuitry which incorporates quantum mechanical effects.
Paul Weiss described research on self-assembled monolayers, including probes of alkanethiols and Tour wire systems. Some of the techniques that he described for preparing controlled structures were remarkably subtle. For instance, domain boundaries were constructed between normally miscible monolayers by
There are a lot of parameters that can be varied in a SAM:
SAM experiments can probe the behavior of small numbers of parallel wires. Other experiments are often ambiguous about whether individual molecular wires can conduct vs. requiring several parallel wires to conduct. SAM experiments can look at randomly formed small sets of 1, 2, 3 etc. wires, identify how many are present, look to see if they conduct, and even look to see if their conductivity varies over time. A 16 hour movie was made of one nitro-substituted Tour wire switching on and off. It could be deliberately turned off by applying a voltage ramp.
Tour described work on molecular caltropes, tetrahedral structures which are intended to bind to an Au surface with 3 legs and project the last leg upwards. One planned experiment will put a dipole moment on the top leg, so that it can be manipulated with an electric field.
James Heath described a different approach to molecular electronic devices based on catenanes. In these structures, part of a molecule is a redox-active center, which sits near a charged group when the redox group is in one oxidation state, but moves away from it when in the other oxidation state. Sequences of electrochemical oxidation or reduction and isomerization open and close molecular switches in these compounds.
As with the Tour wires, the existing experiments have been on two-terminal connections to these devices. Electrochemistry gives a shared third terminal. Heath suggested that the chemistry of these devices would permit building a crystalline PLA diode matrix.
Heath suggested using multiple voltage levels perform multiple functions by having redox-active groups with sharp “analytical voltages”, voltages at which they changed oxidation state. He showed an example of data storage by electrochemical hysteresis, where well-separated reduction and oxidation potentials permitted two stable states between them.
George Skidmore and Ralph Merkle, both at Zyvex, gave back to back presentations on one top-down approach towards a type of exponential growth. They call this approach “exponential assembly”.
This approach uses one fully assembled structure to grasp a subassembly from a MEMS wafer and rotate it into the proper position for insertion into another subassembly. This action completes the final construction stage for a fully assembled structure. Two parallel wafer surfaces are used, with each stage assembling structures on the opposite wafer.
Each fully assembled structure is a small manipulator with two rotational degrees of freedom (DOF) and an electrothermal gripper. Each assembly operation leaves the manipulator attached to the wafer, so x, y, and z displacements can be applied in common to all of the manipulators by moving the wafer as a whole.
These displacements are used as part of the sequence of actions that let each group of 2n manipulators assemble the next group of 2n. This strategy will exhibit externally controlled parallel assembly of a large number of parts. It is MEMS-based, with micron-scale tolerances rather than atomic precision, but it is scalable with the process technology.
This example of exponential growth uses very highly patterned parts, each part being essentially half of the final assembly. It uses the broadcast architecture in two ways. For the x, y, and z displacements, no independent actuation mechanism exists, so the control of these degrees of freedom is not merely externally controlled, but also intrinsically lock-stepped across the full array of manipulators.
The two rotational degrees of freedom provided by the manipulators themselves are also externally controlled, but, since they are controlled electrically, they can potentially be set independently for each manipulator by bringing out enough independent leads from the manipulator array or by fabricating control circuitry along with the manipulator parts array on the MEMS wafers. In the assembly sequence itself, all of the rotational DOFs are operated in lockstep, so a simpler wiring pattern will suffice.
Merkle’s talk emphasized the distinctions between this approach and the default assumptions that are often carried over from biological systems to any self-replicating system, that the systems are:
None of these are true of MEMS exponential assembly. It is perhaps one of the most extremely safe examples of (partial) self-replication.
Merkle’s talk touched on the wide variety of replicating systems, from bacterial self-replication to the Nanosystems architecture with on board control, to various broadcast architecture options (including his hydrocarbon assembler and this MEMS strategy). He called for more systematic research into the space of replicating systems, noting that only a tiny fraction of the possibly useful architectures have been examined. It has been noted elsewhere that the exploitation of biological systems already creates structures which are capable of replicating in the wild. All of the other architectural options are intrinsically safer than biological work which is already in progress, so we should not unthinkingly exclude replicating strategies from consideration.
Richard Superfine described experimental work primarily on rolling carbon nanotubes across a graphite substrate. This is important because it demonstrates a type of gear, a nonbonded interaction that maintains exact molecular registry, even though only a single layer of atoms is involved. Some gears have been designed with groups such as benzene rings protruding from nanotube surfaces, but this work shows that even much more compact designs work.
Pushing on a nanotube on a graphite surface makes it roll. This can be distinguished from sliding because:
In contrast, nanotubes do slide on other substrates, such as mica.
Rolling two different tubes across the same graphite region generally shows different lock-in angles. This shows that the angle depends on both the lattice of the substrate and on the nanotube, which differ for tubes of different chirality. This result excludes the possibility that the lock-in is purely due to deep wells on the graphite surface, rather than registry between the two surfaces. The force required to roll a tube over the surface can be large, larger than the normal sliding force. Tubes can facet, putting a fraction of their total surface area into contact with a graphite substrate. Rolling requires breaking all of this contact adhesion.
Whether a nanotube is in or out of registry with the surface also dramatically affects the resistance between it and the surface. When it is in registry, part of the Fermi surface in the tube has both energy and momentum matched to the graphite below, so conduction can occur efficiently. Effective resistance can be 40-fold beneath the peak resistance of an out-of-registry tube.
Jeffrey Soreff is an IMM Research Associate.