A bio-inspired CNN layer with image processing capabilities

The image processing capabilities of an implemented CNN layer are presented, where the network employs bio-inspired MOS cells that exhibit cubic I-V characteristics. The network architecture and an approximate analysis of its DC characteristics are given. Certain image processing capabilities of the CNN layer are elicited by observing the inherent cooperative behavior among the cells when they are formed as a 2D array. The fabricated CMOS test chip includes a 9/spl times/10 array of connected cells. Applications such as image smoothing, horizontal or vertical line detection, and characteristic noise tolerance have been demonstrated using the test chip. The cell circuit and the resulting network are suitable for large scale applications, since they are very compact and easily implementable.