Robust CMOS CNN implementation with respect to manufacturing inaccuracies

A new idea of cellular neural network VLSI implementation based on a high gain sigmoid function is shown. It has been proved in this paper that this method is robust to the parameter inaccuracies. Conclusions lead to some suggestions and restrictions for template tolerances. An example of VLSI implementation-the test chip consisting of 16/spl times/16 cells in 1.2 /spl mu/m CMOS with 19 fully programmable coefficients has been fabricated. Measurements indicate good efficiency of this implementation.

[1]  Kari Halonen,et al.  About the robustness of CNN linear templates with bipolar images , 1996, 1996 IEEE International Symposium on Circuits and Systems. Circuits and Systems Connecting the World. ISCAS 96.

[2]  Á. Rodríguez-Vázquez,et al.  Current-mode techniques for the implementation of continuous- and discrete-time cellular neural networks , 1993 .

[3]  Kari Halonen,et al.  Digitally controllable weights in current mode cellular neural networks , 1994, Proceedings of the Third IEEE International Workshop on Cellular Neural Networks and their Applications (CNNA-94).

[4]  Leon O. Chua,et al.  CNN cloning template: connected component detector , 1990 .

[5]  Lin-Bao Yang,et al.  Cellular neural networks: theory , 1988 .

[6]  Michiel Steyaert,et al.  Evaluation of CNN template robustness towards VLSI implementation , 1994, Proceedings of the Third IEEE International Workshop on Cellular Neural Networks and their Applications (CNNA-94).