Cellular nonlinear network (CNN) refers to a general scheme for realizing image processing in arrays of coupled nonlinear elements. Chua's "classic" CNN is based on a 19-parameter template that allows programmable coupling to neighboring cells and inputs for realizing various image processing
functions in real time. Image processing in "classic" CNN circuitry has
already been demonstrated experimentally in CMOS technology. The scalability of "classic" CNN to larger networks is limited, however, both by cell complexity and transistor dimensions, even considering the predicted downsizing of CMOS technology. With the goal of surpassing these limits, we are investigating the potential of highly scalable forms of CNN based on
extremely simple cells and minimal interconnection. In this talk I will
discuss theoretical results for a nanoscale CNN based on a two-dimensional array of resistively loaded tunnel junctions that are capacitively coupled
to their nearest neighbors. Coulomb blockade in the ultra-small tunnel
junctions produces neuron-like behavior with waveform spikes that phase-lock
to an ac reference signal in the array. The evolution of the spike phase
in response to a dc bias image is shown to generate complex patterns that exhibit basic image processing functions such as edge detection, binary image enhancement, and rudimentary segmentation. Exploiting this behavior for realizing image processing functions and more general information processing functions will require an interdisciplinary effort, including advanced mathematical analysis of large nonlinear systems.
Copyright. All Rights Reserved.