A new compact programmable νBJT cellular neural network structure with adjustable neighborhood layers for image processing

A new compact structure called the neuron-bipolar cellular neural network (νBJT) CNN is proposed and analyzed. In the new structure, the νBJT and lambda BJT realized by parasitic pnp BJTs in the CMOS process are used to implement the neuron whereas the coupling MOS resistor is used to realize the symmetric synapse weights among various neurons. It has the advantages of compact structure and small chip size. In order to further realize the asymmetric template coefficients with negative values, the coupling MOS resistor is replaced by a fully programmable stage containing two lateral PNP bipolar junction transistors and two MOS switches. The proposed symmetric νBJT CNN and fully programmable νBJT CNN with single neighborhood layer r=1 have been successfully applied to noise removal, hole filling, connected component detection, and shadowing creation. Moreover, the proposed νBJT CNN can be easily extended to realize the multi-layer neighborhood structure. Thus the proposed νBJT CNN has great potential in the VLSI implementation of neural networks. A test chip consisting of 16×16 νBJT CNN has been designed in 0.6 μm CMOS technology