A publication of the Foresight Institute
Since it opened in 1989, the Beckman Institute for
Advanced Science and Technology at the University of Illinois,
Urbana-Champaign, has taken great pride in its
interdisciplinary approach to basic research. When Hungarian Jiri Jonas
became the second director of the Institute, he brought research
efforts into focus in three "main research themes."
They include biological intelligence, human-computer intelligent
interaction, and-of note to Foresight members-molecular and
Several recent publications have high-lighted nanoscale efforts underway at Beckman Institute. In a November 11, 1994 lead editorial, Science lauded Beckman's multidisciplinary approach and described its nanostructure projects. "One is a study of self-organizing structures formed from inorganic substances as well as from protein and other molecules of interest to life scientists."
Science also notes "significant results from research using the scanning tunneling microscope (STM) in fabricating semiconductor nanostructures. A prerequisite was magnificent equipment that facilitates conduct of STM manipulations under a very high vacuum. In one experiment, a clean crystalline silicon surface was exposed to atomic hydrogen with resultant coverage of exposed surface silicon bonds. Later it was possible to selectively remove a narrow band (0.001 micrometer) of the hydrogen. The narrow band could react with other chemicals while the hydrogen-covered silicon remained inert. The STM experiment could be a step on the road to new devices."
Chemical & Engineering News wrote extensively about Beckman's efforts in its March 6, 1995, issue, writing that "nanostructure research merges mechanics, reaction dynamics, optics and electronics...with the merger the disciplines will lose their conventional meaning. Moveover, overlapping interests in chemistry, biochemistry, physics and electrical engineering have aided in refocusing the group's research into fabrication and self-assembly tools, visualization and dynamic probe techniques, and numerical modeling and analysis."
"Experiments at the Beckman Institute seek to marry recent advances in research with scanning probe microscopes for imaging and pattern delineation, genetic engineering, complex synthetic routes to molecules approaching mesoscopic dimensions, and chemical characterizations capabilities that have now evolved to a level permitting the examination of a single molecule."
Research topics include biomolecular electronics, theoretical modeling of nanostructure devices, and protein engineering.
The Beckman Institute brochure, annual activities reports for some research groups, the biannual newsletter, faculty research profiles, and Institute technical reports may be obtained by contacting: Office for External Relations, Beckman Institute, University of Illinois at Urbana-Champaign, 405 N. Mathews Ave., Urbana, IL 61801 USA; tel 217-244-5582, fax 217-244-8371, email email@example.com.
The Arnold O. and Mabel M. Beckman Institute for Science & Technology at the University of Illinois, Urbana-Champaign, houses multidisciplinary work contributing to nanotechnology development. The 300,000-square-foot building, which won the 1990 R&D Magazine Laboratory of the Year Award, was designed by architects Smith, Hinchman & Gryllis Associates and built with a $40 million donation from the Beckmans. (The company founded by Arnold Beckman, Beckman Instruments, is a consistent sponsor of the Foresight conferences on nanotechnology.)
[Editor's Note: This page has been optimized for Netscape 2 and later. If you are using a browser, such as Netscape 1.1, that does not support the html tag for superscripts, please be aware that an number like "2x109" is meant to be scientific notation for "2 times ten raised to the 9th power," and that "e2" means "e squared," etc.]
For the first time: catalysis using an AFM tip
Scanning probe techniques-STM (scanning tunneling microscope) and AFM (atomic force microscope)-have been used to modify individual molecules. Their main advantage is that they let the experimenter make a change at a precisely chosen location. In contrast, chemical reactions in solution will, roughly speaking, react in the same way at all chemically equivalent sites, allowing much less geometrical control. The main disadvantages of probe techniques are that they allow the building of only one molecule at a time and that the set of reactions that can be performed is currently much narrower than the set of reactions in solution-phase chemistry. The research described below has widened the set of reactions the scanning probe techniques can cause.
A scanning probe fabrication advancement came from W.T. Muller, D.L. Klein, T. Lee, J. Clarke, P.L. McEuen, and P.G. Schultz, writing in [Science 268: 272-273 14Apr95]. They coated an AFM tip with platinum, submerged it in hydrogen-saturated isopropanol, and scanned it over a surface coated with a monolayer of an azide (-N3) compound. This resulted "in the catalytic conversion of the azide groups to amino groups...The amino groups formed by this process can be selectively modified with a variety of reagents in a second step to generate more complex structures." Currently, the resolution of this technique has been measured by derivatizing the amine with 40 nm latex beads, and the sharpness of the pattern was limited by the bead diameter. "The free amino groups generated in the catalytic reaction can be derivatized in high yields by a variety of molecules, including acids, aldehydes, and metal complexes." If it should prove possible to modify a surface with atomic precision via this technique, the flexible chemistry should permit derivatization with reagents that introduce new azide groups into additional layers, allowing construction of 3D covalent structures.
A new chemistry for atomically precise surface modification has been introduced by T.-C. Shen, C. Wang, G.C. Abeln, J.R. Tucker, J.W. Lyding, Ph. Avouris, and R.E. Walkup. They have used an STM to desorb hydrogen from hydrogen-terminated silicon (100) surfaces [Science 268: 1590-1592 16Jun95] [Microelec. Eng. 27:23-26 1995]. They observed two modes for this desorption. At high biases, >6.5 volts, the STM operates in a field emission regime. In this regime, the lines drawn are roughly 5 unit cells wide. At lower voltages, desorption becomes less efficient but more selective. Individual rows of silicon dimers could be dehydrogenated with an STM bias of 4.5 volts. In addition, the lines can be reacted with O2 and NH3. "The patterned linewidth appears to be unchanged after oxygen exposure." The authors attribute the low bias desorption process to vibrational excitation of Si-H bonds by tunneling electrons. The bonds accumulate energy through a series of excitations, so unfortunately the mechanism is limited to materials with long vibrational lifetimes. The authors give the vibrational lifetime for Si-H on Si (100) as ~10-8 sec, while on metals it is only ~10-12 sec. Nonetheless, the ability to draw reactive lines two atoms wide is promising, extending the ability to fabricate atomically precise structures.
These structures look promising for building stiff system
In contrast to scanning probe techniques where molecules are built one at a time, large numbers of large, atomically precise structures are routinely built in parallel by living organisms. Enzymes, for instance, consist of one or more protein chains (often containing other groups as well) which typically fold into atomically precise 3D structures. These folds are produced by fairly weak bonds, typically hydrogen bonds and adhesion of hydrophobic (oily) parts of the protein molecule to each other in water. The advantage of self-assembled structures is that large numbers of them can be built at one time. This makes early applications easier than for structures built with scanning probes. Their disadvantages are that they are difficult to design, because a 1D polymer can fold into many different 3D shapes, so it it difficult to ensure that the desired one is formed, and that they are much less stiff than 3D networks of strong, covalent bonds. This makes it difficult to build self-assembled structures which will move as precisely as covalent ones. The research described below ameliorates some of these factors and clarifies others.
R.S. Lokey and B.L. Iverson have synthesized molecules that "fold in water into a pleated structure, as a result of interactions between alternating electron-rich donor groups and electron-deficient acceptor groups." [Nature 375: 303-305 25May95] This structure holds the promise of a class of structures as general as peptides, but with a fixed, predictable secondary structure. The authors appear to have provided us with a structural motif with real backbone. The donor and acceptor groups are rigid, aromatic ring systems, "1,5-dialkoxynaphthalene and 1,4,5,8-naphthalenetetracarboxylic diimide," so undesirable flexibility within these groups will be minimal. The structures allow any (alpha)-amino acid to be placed between the donor and acceptor groups. "They can be prepared by an easy, modular synthetic route. As such, they define a new, and possibly general approach to the construction of large synthetic macromolecules with well-defined higher-order structure."
J. Fredericks, J. Yang, S. J. Geib, and A. D. Hamilton describe a variety of self-assembled structures in [Proc. Indian Acad Sci. (Chem Sci) 106: 923-935 Oct94]. These include a wide variety of structures which self-assemble via hydrogen bonds. Many of their structures resembled the hydrogen-bonded base pairs in DNA. Most of the component molecules contained aromatic cores, usually heterocycles which contained at least some of the hydrogen-bonding groups within the cores. These structures look promising as a design strategy for building stiff components for molecular systems. They have far fewer freely-rotating bonds than do peptides with a similar number of atoms. This reduces the analysis effort needed to predict the structures that can self-assemble. For example, one of the structures described is a cyclic trimer of a pyridoquinoline, which itself consists of three fused six-membered rings. The advantages of such a structure for nanotechnology are that it is inherently quite stiff, with no freely rotating bonds, and that it has 6 positions per monomer where substituents might be attached. The disadvantage of the structure is that it is a symmetrical trimer, so there is currently no way to select which modified monomer would go into which position. This symmetry is typical of the structures described in this paper. Further work to synthesize similar structures, with unique positions for each constituent monomer, would be helpful in extending the design freedom of structures like these.
A general survey of self-assembled systems appeared in [Nature 375: 101-102 11May95]. "A covalent network might have the robustness of diamond, but mistakes in connectivity are frozen in. A complex system of any sort gains tremendously from the capacity for error correction." The survey goes on to cite equilibration of ligands and reversibility of certain organomercury bondings as examples. In order to form atomically precise systems, the final geometries of structures must be stable. What can be useful is for rigid pieces such as heterocyclic bases to be interconnected with multiple weak bonds, such as hydrogen bonds, so that partial alignment of the pieces does not result in irreversible misalignments, even though the final bond between the pieces as wholes may be essentially irreversible. The advantage of multiple weak bonds is essentially the specificity that it provides, most obviously in DNA sequence binding, as in N. Seemans's polyhedral DNA structures.
In [Nature 374: 495-496 6Apr95], J.S. Moore reports on some work, primarily by J. Wuerst and co-workers, on organic nanoporous solids. These structures are useful because they provide structured environments for reactions. Wuerst is building them out of organic compounds that hydrogen bond to each other. "Although the topology of a single network can reasonably be predicted from the geometry of its constituents, it is nearly impossible to predict the fine details of the interpenetrated structure." As with proteins, the noncovalent self-assembly lets us create structures which are larger than our largest rigid covalent structures, but the control of the 3D structure needs to be improved.
The structures formed by weak bonds can be imperfect. J.A.Ernst, R.T.Clubb, H.-X.Zhou, A.M. Gronenborn, and G.M. Clore have shown that there is positionally disordered water present in a protein cavity [Science 267: 1813-1817 24Mar95]. They bounded the residence time to be between 1 nanosecond and 100 microseconds. Unfortunately, this range of time scales includes time scales which are desirable for movement of nanostructures. If the water equilibrates much faster than a structure operates, it will just act as an additional potential term for the structures's degrees of freedom. If it equilibrates much slower than operation times, then it will act like an ice member in the structure. If it equilibrates on a time scale comparable to movement, however, it introduces another energy dissipation mechanism into the structure. In the experiments described in the paper, the water is present in a hydrophobic cavity, so designing a structure to eliminate it may require systematically filling even these cavities.
In [Science 267: 1619-1620 17Mar95] P.G. Wolynes, J.N. Onuchic, and D. Thirumalai survey some recent work on the kinetics of protein folding. The essential point is that, for a protein to fold in a reasonable time, the folding temperature must be well above the glass transition temperature which "is equivalent to maximizing the 'stability gap' between the native state and disordered collapsed structures measured in units of the ruggedness [of the potential surface for the disordered structures]." E.I. Shakhnovich investigated the stability gap in simulated 80-mer model proteins [Phys.Rev.Let. 72: 3907-3910 13Jun94] and found that there was a stability gap for proteins designed with a simulated annealing process when 20 residue types were used, but no stability gap was found when only 2 residue types were used. This shows that simplifying the model of protein residues to hydrophobic/hydrophilic is inadequate. It also shows that in order to use nonprotein 1D polymer systems to build 3D structures, more than 2 residue types will be needed.
While most of the research described above extends the ability
to build molecules, it is also important and difficult to choose
which molecules to attempt to build in order to fulfill some
purpose, and to analyze proposed molecules to see if they really
do what the designer wants them to do. The research described
below extends these abilities.
Y.S. Smetanich, Y.B. Kazanovich, and V.V. Kornilov wrote a paper [Discrete Applied Mathematics 57: 45-65 Feb95] which analyzes which sets of components and self-assembly rules give unique structures. They make a number of assumptions which make their results valuable primarily for stiff structures. They model self-assembly as the formation of structures from a set of subunits, where each bond between a pair of subunits follows a specific assembly rule. They model each specific assembly rule as a precise relative position and orientation of the two subunits. In essence, their default is to treat each bonding site as being specific to one other subunit, with the bonding site sufficiently rigid to hold the new subunit in a single position and orientation. I view this as a reasonable approximation to bonds between large and irregular surfaces, as between pairs of proteins. If the bonds are not stiff enough, there are problems with self-assembly that are not covered by this approximation. Most notably, assembly steps that should form cycles may fail to close them, forming extended polymers instead [R. Merkle, private communication]. The authors' approximation is clearly not good where a bond between subunits allows free rotation, for instance where a bond between subunits consists of a single hydrogen bond. It is also not a reasonable approximation in a case like a bearing, where a bond must have a degree of freedom in order to function. The authors treat the subunits as rigid bodies, using information about their shape to eliminate structures that would require intersection or distortion of subunits. Overall, I see this approach as potentially helpful in assisting with the high-level design of a self-assembling structure, essentially at the stage where we wish to try a variety of ways to partition the desired structure into subunits. This approach requires the ability to design subunits so that their bonds are fairly rigid, and do not bend or rotate enough to seriously distort a structure at room temperature.
A. Roitberg, R. B. Gerber, R. Elber, and M. A. Ratner wrote a paper [Science 268: 1319-1322 2 June95] which can assist with calibrating molecular mechanics for proteins and similar molecules. One way to calibrate the potential energy surfaces for these molecules is to calculate the vibrational spectrum and compare it with experiment. The authors showed that anharmonic terms in the potential have quite a dramatic effect on the predicted spectrum for transitions from the vibrational ground state. They also showed that the effect of these terms is very largely confined to anharmonic corrections within modes, with very little net shift in the energies of modes due to interactions with other modes. This is different from the results seen in small molecules. The authors attribute the difference to the interaction between any given mode and all of the other modes averaging to zero for large molecules. This work can help make the calibration of potential energy models from vibrational measurements more accurate, while keeping the calibration computationally efficient.
J.F.Y. Brookfield writes in [Nature 375: 449 8Jun95] on the application of "forced evolution" (generating and selecting large number of candidate structures by combinational techniques) to solving "problems in the design and optimization of new molecules." The key problem to using this for nanotechnology is that "The experimenter still has to test alternative proteins, however, and retain the genes encoding the best." This appears to be very natural when maximizing the stability of an interface between an existing protein and a new one. One can simply use the binding of a fluorescent derivative of the old protein to assay the new one. It is less clear, however, how to assay for a specific shape of the new protein, and that will often be equally important in building nanostructures.
DNA computation doesn't seem to extend our 3D control of
There are two proposals in progress that could very loosely be considered early applications of nanotechnology to data processing. One applies DNA binding to computation, and the other applies internal movements in a protein called bacteriorhodopsin to data storage.
There has been a flurry of interest in DNA-based computation recently, based on L.M. Adleman's demonstration of a computation of a Hamiltonian path. R. J. Lipton has described a way to extend this work to finding solutions of the satisfaction problem. [Science 268: 542-545 28Apr95]
In general, I find proposals such as Lipton's, where the total amount of external information that must be translated into DNA is small, to be more plausible than proposals like E. B. Baum's [Science 268: 583-585 28Apr95], which suggest using DNA as a readout mechanism from a database. While the matching process in Baum's proposal is indeed fast, the construction of the database from external data must go through DNA synthesis. Changes to the database would have to be exceedingly rare in order for Baum's proposal to be attractive.
The core of the capability that DNA computation provides is a highly parallel matching capability, evaluating ~1020 alternatives against a constraint in ~102 seconds. By comparison, specialized electronic chips now have around 106 gates, and operate at ~108 Hertz. If ~103 gates are needed for each matching engine, then a chip could perform ~1011 matches per second. A DNA matching engine would approximately equal the speed of 107 current day chips. Is this sufficient to be worthwhile? Further work may tell.
I find it difficult to assess the implications of a practical DNA computing application on nanotechnology in general. On the positive side, such a development would create a routine use of atomically precisely encoded data.
It is unclear whether it would create an incentive to extend DNA synthesis. Current capabilities seem adequate to make the limiting factor in DNA computation be the volume of material required rather than the number of bases that can be strung together. This work doesn't seem to extend our three-dimensional control of synthesis. The matching in it is essentially one-dimensional.
There have been a number of articles on bacteriorhodopsin optical storage elements for computers recently (M. Freemantle [C&EN 24-26 22May95] R.R. Birge [Sci. Am. 272: 90-95 Mar95]). The basic idea is that this protein has a number of metastable states, notably a "Q" state "which is stable for extended periods, even up to several years." Bacteriorhodopsin can be switched from state to state by irradiating it with light of the proper frequency. Some of the transitions require absorbtion of several different photons, so they can be used to limit transitions to the intersection of several beams, allowing 3D memories. The implications for nanotechnology are somewhat hazy. As it stands, an ensemble of molecules in roughly a cubic micron volume is being used to store each bit. From a molecular viewpoint, this isn't extremely different from magneto-optical storage, for instance. On the other hand, bacteriorhodopsin is a protein, so it could be tailored at many sites for various desirable properties: lifetimes of states, absorbtion spectra, and so on. This would amount to use of the states of a complex molecule for information storage. Over a longer term, if variations on the protein could be designed to store multiple bits, perhaps acting as nanometer scale shift registers, they would provide an incentive to design structures with a number of moving parts. However, bacteriorhodopsin's "unique photophysical properties were discovered in the early 1970s." Waiting for nanotechnology to be driven by multi-bit versions of this protein could leave nanotechnologists looking for light at the end of a long tunnel.
While a variety of scanning probe techniques now allow
reactions with single molecules, they are not yet at the point
where one can assume that a reaction has succeeded without
checking. The scanning probe microscopes themselves can detect
certain reactions, but it is often helpful to have a variety of
instruments which can confirm whether a reaction took place, and
help an experimenter see if they built the structure that they
intended to build. The research described below extends these
T. Funatsu, Y Harada, M Tokunaga, K. Salto, and T Yanagida have improved the sensitivity of a fluorescent detection technique to the point where they can detect single molecules, and detect whether they are bound to a surface or diffusing in liquid. [Nature 374: 555-559 6Apr95] The sensitivity of this technique may help confirm construction of individual molecules with STMs or AFMs. This is not a technique that images features within a single molecule, but it can confirm that a single molecule has been successfully modified. The most sensitive technique that the authors developed, total internal reflection fluorescence microscopy, confines the exciting laser beam to a thin, 150 nm evanescent layer. The emitted fluorescence is currently detected from the other side of the sample, so this technique requires a transparent sample, and access to both the top and bottom of it. The authors intend for their methods to "be extended to simultaneous measurement of the ATPase reaction and force/movement by single myosin molecules," presumably with AFM techniques. One physical effect that adds both an advantage and a disadvantage to this technique is photobleaching. Under the authors' conditions, the lifetime of their fluorophores is ~15 seconds. On the one hand this limits the maximum observation time for a molecule. On the other hand "The high-rate imaging allowed to see clearly the time course of stepwise photobleaching of fluorescence from two dye molecules, probably bound to a single HMM [a myosin fragment]." One concern that this raises in using the technique to confirm a mechanosynthetic step is that the reactive fragment of a photobleached fluorophore must not react with the workpiece in the AFM in a damaging way.
J. Kohler, A.C.J. Brouwer, E.J.J. Groenen, and J. Schmidt describe a technique that uses the fluorescence of a single molecule to observe the spin of a single 13C nucleus in that molecule. [Science 268: 1457-1460 9Jun95]. The observation mechanism is rather intricate, involving two triplet states with different lifetimes, a microwave-induced transition between them (which has the effect of extending the average triplet lifetime), and the ground and first excited singlet state, which are responsible for the fluorescence. The experiment was done with deuterated pentacene, present in a dilute solution in deuterated p-terphenyl crystals. The spectrum of the microwave resonance depends on the existence and location of 13C nuclei in the pentacene molecule, and on the strength of the externally applied field. It is analogous to measuring spin-spin coupling strengths in NMR, a standard technique used in determining molecular structures. This technique therefore holds the potential for confirming the identity of individual nuclei in selected locations in individual molecules, a capability not directly available from STM or AFM techniques.
A number of papers described extensions to scanning probe techniques. S. Fujisawa, E. Kishi, Y. Sugawara, and S. Morita described observation of atomically resolved friction measurements on a NaF (100) surface [Nanotechnology 6: 8-11 Jan95]. This work used an Si3N4 tip on the NaF substrate, with tip loadings from 4.9-14 nN. The authors chose NaF rather than a layered material in order to avoid generating flakes and dragging them with the scanning tip, which would generate periodic signals which don't directly reflect atomic scale imaging, because they average across the flake. In this paper, both the periodicity and the amplitude of the friction signals are explained by the surface atom positions. A paper on a similar system, an AFM measurement on an LiF (100) surface, by E. Meyer, H. Heinzelmann, D. Brodbeck, G. Overney, L. Howard, H. Hug, T. Jung, H.-R. Hidber, and H.-J. Guntherodt appeared in [J.Vac.Sci.Technol. B 9: 1329-1332 Mar/Apr91]. They also used ~10nN loadings, but the earlier paper imaged the lattice with vertical, rather than lateral, deflection. Both techniques should prove useful in analyzing sufficiently stiff workpieces.
On more flexible substrates, F.A. Schabert, C. Henn, and A. Engel imaged phospholipid/membrane protein crystals with an AFM in liquid [Science 268: 92-94 7Apr95]. Since these crystals are much more flexible than the alkali fluorides described above, the AFM loadings used were much lower, 0.1-0.3 nN, and the lateral resolutions are much coarser, ~1nm. A good deal of this paper's information came from scans which averaged over the three-fold symmetry of the crystal. If complex, asymmetrical, soft nanostructures (perhaps built through self-assembly) are to be examined with AFMs, this sort of averaging will generally not be feasible. In order to confirm the construction of some structures, it may be necessary to design them to be easy to image, perhaps with some multi-nm features present purely for imaging.
Jeffrey Soreff is a researcher at IBM with an interest in nanotechnology.
From Foresight Update 22, originally published 15 October 1995.