Impact of system specifications on analogue CMOS implementations of continuously programmable cellular neural networks
Peter R. Kinget, Michiel S. J. Steyaert · 2002
The analogue VLSI implementation of cellular neural networks that are fully programmable with continuous valued weights and operate in continuous time, is treated. The impact of dynamic range of the weights, the connection pattern of the cells and the necessary accuracy of the system on the hardware requirements is discussed. A practical example of the first fully programmable 4/spl times/4 CNN chip is presented. The successful operation of the chip is illustrated with measurement results. This chip is a programmable image processor and the templates can be considered as its instructions.>