Implementation of analog ICs based on neural networks

Concepts found in the study of neural networks can be used to aid the analysis and increase the understanding of conventional analog ICs, as well as suggest new circuits and applications. The development of a series of analog ICs based on crossbar neural networks is presented. There are two main problems in their implementation: the choice of the correct weights to ensure that stable states correspond to solutions to the problem addressed by the network and the stability of the network which depends critically on its integrated circuit construction. The transfer curve of an analog IC implementation of an A/D converter illustrates the problems of choosing the weights in such networks correctly in order to avoid incorrect solutions.<>