The pioneering work of Eigler, Avouris, Gimzewski, and others has shown that it is possible to precisely position atoms and molecules on a surface by using a Scanning Probe Microscope (SPM). These are remarkable achievements. However, manipulation of particles with dimensions of a few to a few tens of nanometers is likely to have a greater impact in nanometer-scale science and technology in the near future. Patterns of colloidal nanoparticles can be constructed by SPM manipulation, and have several potential applications that are worth investigating. They can be used to efficiently store digital information (Figure 1), to build single-electron transistors, or as templates for building nanostructures that can function as components of nanoelectromechanical systems (NEMS). For example, Au nanoparticles can be linked by dithiols, and the resulting structures can (potentially) be used to construct more complex objects in a bottom-up fashion.

Reliable and accurate manipulation of nanoparticles has been difficult to achieve in the past. This is due largely to a lack of understanding of the underlying phenomena, and to a lack of suitable control software. Detailed experimental studies of tip/particle/sample interactions during manipulation are very few. Typical nanomanipulations are done blindly, i.e. without real-time information about the operation being performed. Being blind during the actual manipulation makes any attempt to understand the pushing process difficult, and no model for it has been published until now. Instrument errors such as creep, hysteresis and thermal drift, lead to a manipulation environment with high spatial uncertainty, especially in ambient air and at room temperature. These errors must be physically eliminated (e.g. by operating at low temperatures), which leads to elaborate and costly procedures and equipment, or the control software must compensate for them, which is the approach taken in our work.

This paper presents experimental results that provide new insights into nanomanipulation phenomena. It also shows that nanoparticle manipulation operations can be executed reliably with an AFM in ambient conditions, and can be monitored in real time, by using the strategies and special-purpose control software we have developed.

Figure 2 shows the data gathered in a typical pushing event. Curves A and B show the topography signal before and during manipulation. The feedback is turned off during pushing (curve B) and thus no topography information is available. We observe that after the feedback is turned on again, a portion of the particle is imaged (curve B). During the pushing process, we acquire the non-contact vibration amplitude (NC-amplitude) (curve C), and the d.c. tip deflection signal (curve D), by using our own probe control software. An additional line scan after the operation proves that the particle has indeed moved (curve E).

Our experiments show that gold nanoparticles manipulated with an AFM on a mica surface are mechanically pushed by the repulsive forces between tip and particle. The pushing operations are performed by moving the AFM tip in non-contact mode against a nanoparticle with the feedback turned off. The cantilever vibration amplitude decreases as the particle is approached, and becomes essentially zero during pushing. The tip first moves upward, in contact with the particle, until the cantilever has flexed enough to exert the force necessary to move the particle. Then pushing begins, and continues until the feedback is turned back on. This breaks the tip/particle contact and restores the vibration amplitude. Monitoring simultaneously the non-contact amplitude and the cantilever deflection provides real-time feedback on the progress of the operation. This information, together with our new understanding of the phenomena and spatial reasoning techniques from the robotics field provide us with the tools necessary for programming the AFM at a high level, so as to automatically construct patterns with large numbers of particles, which are needed in many of the potential applications.

We demonstrate that patterns of colloidal Au nanoparticles can be accurately and reliably positioned by using our pushing protocols. Mechanical pushing is a versatile and nearly universal process, not restricted to unique environments, specific materials or substrates. Colloidal nanoparticles with a variety of characteristics (e.g. magnetic or semiconducting) are readily available. Automatic construction of patterns with these particles will open new avenues of research on nanostructures that exhibit interesting new behaviors, and have a wide range of applications, from nanoelectronics to biology.