Review of CMOS implementations of the CNN universal machine-type visual microprocessors

Abstract

While in most application areas digital processors can solve problems initially, in some fields their capabilities are very limited. A typical example is vision. Simple animals outperform super-computers in the realization of basic vision tasks. In order to overcome the limitations of these conventional systems, a fundamentally different array architecture is needed. This architecture is based on the new paradigm of analogic cellular (CNN) computing whose most advanced implementation is the so-called CNN universal machine (CNN-UM). Its main components are: a) parallel architecture consisting of an array of locally-connected analog processors; b) a means of storing, locally, pixel-by-pixel, the intermediate computation results, and c) stored on-chip programmability. When implemented as a mixed-signal VLSI chip, the CNN-UM is capable of image processing at rates of trillions of operations per second with very small size and low power consumption. On the other hand, when integrating the adaptive multi-sensor array in the CNN-UM, the resulting sensor+computer array offers unprecedented capabilities. This paper reviews the latest results on CMN-UM chips and systems, and outlines the envisaged roadmap for these computers.