Silicon implementation of computational intelligence for mechatronics

Digital implementations of neurocomputers are presently quite expensive, they require excessive power, they suffer from a number of issues that cause performance characteristics to differ from the theoretical model of the system, and they are relatively intolerant of fault conditions. The inherent advantages of the massively parallel structure of these systems are also lost in the common practice of executing algorithms sequentially on a conventional computer. The paper presents nonlinear analog signal methodology where for nonlinear processing nonlinear characteristics of MOS transistors are used.