Abstract

Continuing advances in the miniaturization of electron devices have made possible the fabrication of nanoelectronic devices with feature sizes in the 1 - lOOnm range. However, it is widely believed that conventional integrated circuit design techniques will become impractical, due to the small size and the low current carrying capacity of nanostructured devices. Proposals which envision novel ways to circumvent this problem have begun to appear in earnest. The majority of these proposals use globally coherent quantum systems to generate computational abilities. We differ from these proposals in that we use semiclassical global models, as the basis from which to conceive nanoelectronic functional devices. An additional point of departure, is our assumption that currently perceived limitations to realizing interconnects amongst the nanostructured devices, will in time be overcome. Finally, we restrict attention to niche applications, in which the collective activity of a Large number of nanostructured devices give rise to useful computational functions. The special pui-pose functional device concept adopted here can be contrasted with other approaches which envision the design of general purpose computers on the basis of quanturn mechanical logic gates. We adopt a research methodology in which computational tasks which are naturally suited to a collective solution strategy are first identified, and then mapped to nanoelectronic physical systems. In making these associations we make well-defined assumptions concerning the properties of interconnection networks. The justification for this approach comes from the extensive experimental activity on novel wiring technologies tailored specifically for nanoelectronics. An aspect of these technologieiss the fact that electronic transport along these wires can themselves introduce nonlinearities which can influence the global behavior of networks. The work discussed in this report has been unified under a particular technology based on the creation of arrays of nanometer-sized metallic islands. We then consider different types of network mechanisms for the transfer of electrons between islands. Depending on the types of transport nonlinearities permitted by the network links, we show that it is possible to generate different kinds of global activity in these networks. We show, in addition that it is possible to impart a computational interpretation to these global activities. In particular, we show that within a classical circuit theoretic model, non-monotone nonlinearities in the local transport can yield global associative memory effects. We then show that this interpretation will remain valid even when single-electron effects come into play, provided that the effective capacitance of the nanometallic islands is not too small. We then investigate networks of islands in which the sole nonlinearity arises from single-electronics. These networks are shown to be capable of associative memory effects, as well as yield approximate solutions to certain NP-complete optimization problems, provided that there is suificient flexibility in the choice of inter-island capacitances.

Date of this Version

July 1994

Share

COinS