When comparing conventional computing architectures to the architectures of biological neural systems, there are several striking differences. Conventional computers use a low number of high performance computing elements that are programmed with algorithms to perform tasks in a time sequenced way; they are very successful in administrative applications, in scientific simulations, and in certain signal processing applications. However, the biological systems still significantly outperform conventional computers in perception tasks, sensory data processing and motory control. Biological systems use a completely different computing paradigm: a massive network of simple processors that are (adaptively) interconnected and operate in parallel. This massively parallel processing seems the key aspect to their success. The advent of VLSI technologies has provided us with the technological means to implement very complicated systems on a silicon die. Analog VLSI circuits in standard digital technologies open the way for the implementation of massively parallel analog signal processing systems for sensory signal processing applications and for perception tasks. This text describes in depth the theory and practice of designing these massively parallel analog signal processing systems. Techniques are described which overcome the device and transistor mismatches which limit performance of the system. The book also shows how such a system can be efficiently implemented using cellular Neural Networks. 4