| Atomic force microscopes (AFMs) are most often used for high-resolution imaging and detailed surface characterization, but soon after their invention it was recognized that they could also be used to change, interact with, and control nanoscale matter. A well-known early example of this was the IBM logo written with Xenon atoms by Don Eigler‘s group at the IBM Almadén Research Center. Lars Samuelson’s group at the University of Lund suggested it would be possible to build nano-objects with larger, molecular-sized building blocks and assemble them with an AFM in ambient conditions. A group at the University of Southern California’s Laboratory for Molecular Robotics (LMR) headed by Aristides Requicha and Bruce Koel has been investigating this approach for several years. Their research focused on the development of high-level systems for programming an AFM as a sensory robot, and the application of these systems to challenging nanomanipulation problems, such as building prototypes for nanosystems. AFM are designed to work as an imaging tool based on feedback control. A special software solution was needed to minimize the interaction between the tip and the substrate, thereby allowing AFM manipulation. The LMR researchers developed NanoMove manipulation software based on the Application Programming Interface (API) for Veeco’s AutoProbe CP-Research AFM. NanoMove allows AFM manipulation using a variety of protocols and the acquisition of various signals. This article reviews the research conducted by the LMR group. Their work shows that nanomanipulation offers great advantages for computer scientists studying nanorobotics issues, as well as by chemists and physicists investigating nanostructures. Principles of Nanomanipulation Nanorobotics An AFM tip can be used via different mechanisms to modify surfaces with nanometer resolution. Tasks such as pushing and pulling or cutting and indenting can be performed, and nanoscale objects can be mechanically moved by the AFM probe tip. The AFM tip can serve as a robotic hand to precisely position nano-objects and assemble them under computer control. AFM manipulation does pose an interesting problem in robotics, however. It can be likened to a mobile robot (e.g., a helicopter) mapping a terrain and navigating over it by using only altitude radar and dead reckoning in the presence of large spatial uncertainties. The nanorobotic system includes substrates that serve as nano-workbenches on which to place the objects to be manipulated (analogous to fixtures in the macrorobotics world); tips, probes, and molecules that serve to grasp others, and function as grippers or end-effectors; chemical and physical nanoassembly processes; primitive nanoassembly operations (analogous to macroassembly operations like peg-in-hole insertions); methods for exploiting self-assembly to combat spatial uncertainty (analogous to mechanical compliance in the macroworld); hardware primitives for building nanostructures; and software for sensory interpretation, motion planning, and acting (i.e., driving the AFM). NanoMove Software The LMR researchers developed methods for positioning colloidal nanoparticles (typically gold colloids with diameters 5–30nm) accurately and reliably on mica and silicon substrates, in ambient air or liquid environments. The experiments were conducted with a CP-Research AFM in TappingMode using triangularly shaped silicon cantilevers with a spring constant of approximately 13.0 N/m-1 and a resonance frequency around 340 kHz. These relatively stiff cantilevers showed the best results for mechanical pushing. Manipulation of the gold nanoparticles was performed by utilizing the LMR group’s NanoMove software. The software adds several unique features to the instrument to enable manipulation. NanoMove gives the user the ability to 1) control the feedback operation, 2) perform one-line scans in any arbitrary direction in the X-Y plan; and 3) acquire various signals simultaneously with the manipulation. Figure 1 shows the NanoMove user interface. The upper right menu shows the main control and operation window for imaging and manipulation. 
Figure 1. The NanoMove software user interface. After recording a topography image, the user can draw an arrow on the image (see the red arrow in the upper left window) to determine the manipulation trajectory. The arrow dictates the direction and length of the scan line, and can be moved by the operator in the X and Y direction until the displayed topography indicates that its path is centered over the particle. Two bars are positioned along the scan line, showing the range of alternative operating conditions for the AFM, where the ”start” and ”end” points of the manipulation can be selected. The feedback is turned off just before the tip scans across the particle, and is switched back on after reaching the desired lateral position (see Figure 2). 
Figure 2. Imaging takes place with the feedback control on. Turning off the feedback allows the mechanical pushing of the nanoparticle. A contact mode setup is recommended in case the particles and structures one wants to manipulate are strongly attached to the underlying surface. There are several different manipulation protocols in the NanoMove software. One can select either a feedback-off protocol, with or without additional direct movement of the scanner, or a feedback-on protocol, with indirect movement of the scanner. In these experiments, the TappingMode setup was chosen because the nanoparticles cannot be imaged accurately in contact mode, and are replaced during imaging due to the presence of lateral shear forces. Figure 3 shows the manipulation of a 30nm diameter nanoparticle by using the appropriate operation parameters, the particles can even be pushed up a 10nm step on the surface. The step height and the particle size are of the same order, and thus the experiment is a first step towards mechanical construction of three-dimensional structures. 
Figure 3. A 30nm gold particle (a) before and (b) after being pushed over a 10nm high step along the direction indicated by the arrow. Image sizes are both 1μm x 0.5μm. The Manipulation Mechanism When going down to the nanometer scale the physical forces that are dominant in the macroscale become negligible. AFMs provide the ability to study the mechanics in nanoscale. By looking at various tip signals (e.g., the amplitude and the deflection signals) during the manipulation operation and analyzing the changes, it is possible to study the mechanism of the manipulation. In a series of papers the LMR group studied the AFM manipulation phenomena involved in pushing a nanoparticle. They observed that when the tip is oscillating relatively far from the surface, the amplitude decreases as the tip approaches the particle but the particle does not move. When the tip is sufficiently close to the surface, the vibration amplitude goes to zero as the particle is approached. The DC cantilever deflection becomes non-zero, and the particle moves, as long as the deflection is above a certain threshold dependent on the cantilever and various other characteristics of the setup. The changes in vibration amplitude and cantilever DC deflection can be used to monitor the manipulation in real time, and without further imaging verify with a high degree of confidence that it is successful. The studies showed that the manipulation of nanoparticles takes place by sliding and not by a rolling mechanism. Fabrication of 2D and 3D Nanostructures Nanoparticle Patterns Nanoparticles are attractive building blocks for nanostructures because 1) there are many known methods for synthesizing nanoparticles with a variety of characteristics (e.g., metallic, semiconducting, or magnetic) and the state of the art is steadily improving; 2) the particles have more uniform sizes (i.e., are more monodisperse) than structures of comparable sizes made by competing techniques such as electron-beam lithography; and 3) arbitrary planar patterns of nanoparticles can be built by nanomanipulation using the protocols discussed above. Moving and Manipulating Nanoparticles AFM manipulation can be a tool for the fabrication of nanoparticles patterns. Figure 4 shows an example of manipulation of randomly deposited gold particles on a mica substrate. The 15nm diameter gold particles were pushed from an initial random position to form the USC logo. 
Figure 4. A random pattern of 15nm gold balls that was converted into the "USC" logo pattern by a sequence of pushing commands. The LMR researchers also studied the possibility of using nanomanipulation for data storage. Figure 5 shows the construction of a pattern that encodes ASCII characters in horizontal rows of nanoparticles on a surface. The presence of a particle at a node of a regular 2D grid is interpreted as a "1," and its absence as a "0." The pattern, read from top to bottom encodes "LMR." The particles have diameters of 15nm, and the grid nodes are spaced with a 100nm pitch. The real density is on the order of 60 Gb/in2 and it should be possible to increase this density by over an order of magnitude using smaller particles and tighter spacing. This would give entities approaching the Tb/in2. This digital storage technique is a candidate for an editable NanoCD. However, there are obstacles that must be overcome for it to be practical. 
Figure 5. a) The characters "LMR" ASCII encoded in rows of nanobits, and b) the trace obtained by reading the second row with an AFM. The manipulation approach can be extended to 3D fabrication. The LMR group demonstrated the construction of a 3D structure by controlled manipulation of single nanoparticles (see Figure 6). 
Figure 6. 3D image projection of a pyramid like structure. The preparation took place by pushing a 30nm nanoparticle up between two others. However, the manipulation of asymmetrically shaped features is more complicated. Gold nanorods, 100nm in length and 10nm in diameter, were used by the LMR researchers to study the manipulation of elongated nanoobjects (see Figure 7). 
Figure 7. A sequence of SFM images (500nm x 500nm scan size) displaying the manipulation of four gold nanorods. The arrows in each image show the manipulation direction that results in the rod configuration in the next image: (a) Initial arrangement of the rods; (b) result of translational manipulations of "1" along it s longitudinal axis and "2" acrossthis axis; (c) results of rotational manipulation operations of all four rods by 45°, relative to their original orientation. The height scale from black to white is 10nm. Translational manipulation of a nanorod without rotation takes place whenever the tip hits the nanorod at its center. In these experiments, it was found that it was easier to translate the rods when the pushing direction was along the longitudinal axis than when pushing was done transverse to this axis. This is because it is easy to locate the highest point across the width of the rod (which is the center of the rod). Therefore, the result of longitudinal manipulation was often perfect translation while transverse manipulation often caused a combination of translation and rotation. This information on rod manipulation is important for assembling a functional nanostructure. It will be somewhat complicated to build such structures with nanorods, because after the rods are roughly positioned; subsequent movements may have to specify both position and angle of the nanorods. Cutting and Bending Materials With An AFM Tip As mentioned above, an AFM tip can also be used to cut or bend soft materials, such as polymers, DNA, and nanotubes. Figure 8 shows the cutting operation of DNA plasmid. It can be seen that the cutting result is too coarse. A better approach would be to use enzymes to cut the biological sample in combination with an AFM probe to select the exact site for the modification. Figure 9 shows an AFM tip that is under NanoMove control bending a nanotube. 
Figure 8. Using an AFM tip to cut plasmid. 
Figure 9. Using an AFM tip to bend nanotubes. Using Nanomanipulation for Prototyping Devices Manipulation of nanoparticles can also be used to build prototypes of electronic and optoelectronic devices. In fact, many of the existing nanoelectronic devices have either relied on chance to place an element in the desired relationship with others or have used AFM manipulation. For example, placing a nanoparticle at tunneling distances between two electrodes (source and drain) can be used to make a single-electron transistor (SET). Figure 10 shows AFM images that were taken during the manipulation of two particles into a gap of a SET structure. 
Figure 10. Steps in the manipulation of two gold particles into a single electron transistor (SET) junction. A similar approach was used for another prototype system. The LMR group and the Atwater group at Caltech collaborated on the fabrication of a “plasmonic” waveguide by placing colloidal 30nm diameter gold nanoparticles at equal distances from each other in a chain, with a 100nm fluorescent latex particle at the end of the chain. Energy at a wavelength in the visible range is injected into the gold particle at one end of the chain, and propagates through the chain by exploiting near-field effects. The propagation is detected by observing the fluorescence of the latex ball. The waveguide can also be constructed by using e-beam lithography to fabricate gold nanostructures, but the AFM tip is still needed to manipulate the latex fluorescent bead to the end of the structure (see Figure 11). Therefore, the usage of the AFM manipulation is crucial for the construction of the prototype. This nanowaveguide is unique because it has transverse dimensions much smaller than the diffraction limit for the wavelengths (hundreds of nm) that are being studied. It may also serve to feed light to individual molecular machines without exciting other machines in the same neighborhood. 
Figure 11. Plasmonic waveguide: (a) Schematic of a plasmonic waveguide, (b) SEM micrographs of e-beam lithography fabricated gold nanostructures, and (c) AFM image of a latex bead (marked by yellow arrow) that was manipulated to the end of the gold nanostructures matrix. Solid Nanostructures and Layered Fabrication Linking Nanoparticles Though patterns of unlinked nanoparticles can be useful, many applications require “solid” nanostructures of specific shapes. These can be approximated by groups of suitably positioned and linked nanoparticles. The LMR group investigated several approaches to linking. The first uses covalent bonding to a linker. For example, gold particles can be connected with dithiols (organic molecules with sulfur at both ends). The dithiols self-assemble to the gold and serve as chemical glue. Two variants of this approach were demonstrated: 1) depositing the particles, positioning them, and then immersing the sample in the dithiol solution to link them; or 2) depositing the particles, applying the dithiols, and then manipulating the particles into linked contact. It was found that it is indeed possible to push a group of nanoparticles linked by dithiols as a whole. These results demonstrate hierarchical assembly at the nanoscale (i.e., the construction of assemblies of components, which are themselves subassemblies of other components or of primitive building blocks). The second approach to linking also uses selective self-assembly. Additional material is deposited on the particles until they become connected. The material and experimental conditions must be selected to ensure that the material assembles to the particles but not to the remainder of the sample. For example, a pattern of gold nanoparticles can be used as a template for the electroless deposition of additional gold (see Figure 12). 
Table 12. SFM images (1μm x 1μm scan size) displaying 8nm gold colloidal particles on SiO2 as randomly deposited (left), after manipulation of thirteen particles to form a wire nanotemplate (center), and after 5 minutes in the seeding solution (right). Gold wires of arbitrary geometry can be built by first manipulating the particles into the desired geometry and then linking them by immersion of the sample in the electroless solution with a specific set of parameters, such as immersion time, concentration, and so forth. A third approach discovered very recently uses sintering to connect fluorescent latex nanoparticles. The particles are first manipulated to form a desired template. The template is then heated, melting the particles together into a single nanostructure (see Figure 13). 
Figure 13. Sequence of AFM images showing the construction of a 3D nanostructure: a) randomly deposited particles, b) after manipulation, c) after sintering at 160 ± 2ºC for 10 minutes, d) after a single particle is ‘pushed’ on top of the island. Embedding of Nanostructures For certain applications it is necessary to ensure that nanocomponents are fixed on the substrate. This can also be done by selective self-assembly. A material that assembles to the substrate but not the particles is used, thus embedding the particles in a thin layer. The LMR group demonstrated particle embedding in a silicon oxide layer by first depositing particles and manipulating them, then depositing a monolayer of a silane (an organic molecule containing silicon atoms that attaches only to the substrate), and finally oxidizing the silane layer. Successive layers were used to embed particles for a proposed new rapid prototyping technique at the nanoscale, called layered nanofabrication or LNF (see Figure 14). Three-dimensional objects were fabricated by nanoparticle manipulation, and each layer was planarized by adding a molecular sacrificial layer whose top surface served as support for the next processing step. The sacrificial layers were removed in a final step. Thus, the researchers demonstrated that it is possible to build sacrificial layers and to manipulate gold nanoparticles on top of them (see Figure 15). 
Figure 14. Schematic view of the embedding procedure of nanoparticles in a SiO2 matrix. 
Figure 15. AFM images and corresponding line scans displaying the successful construction of a two-particle, upright column by pushing particle ”1” on top of particle ”2”. The scan size is 600nm x 600nm, and the height scale is 6nm from black to white. Conclusion AFM provide effective tools for fabricating nanodevice and nanosystem prototypes and products in small quantities. By using the NanoMove software developed by the Laboratory for Molecular Robotics (commercially available from Veeco), it is possible to use an AFM for manipulation. AFM manipulation can be used to accurately and reliably position molecular-sized components. Unlike its macroscopic counterparts, which are primarily governed by classical mechanics, nanomanipulation phenomena fall mostly in the realm of chemistry. The linking and assembling of nanoscale objects can be accomplished via chemical and physical means using such techniques as ”gluing” with suitable compounds, chemical deposition, or simple heating. Demonstrations that may lead to useful applications of nanoassembly are beginning to appear. However, increased levels of automation in nanomanipulation are needed to prototype more complex and useful devices and systems. Pick-and-place operations and the construction of three-dimensional nanostructures are still very primitive and need further development. Clearly, AFM will have a crucial role in the further investigation of these processes. |