Analog VLSI implementation of neural networks

This chapter introduces the motivation for doing signal processing by means of analog VLSI, before discussing the peculiarities and implementation constraints of this approach. The possible modes of operation and the model of the MOS transistor are then recalled before identifying the properties of individual transistors and of their basic combinations to be exploited opportunistically in analog circuits. Some implementations of local and collective operators relevant to neural networks are then discussed. The difficult problems of analog storage of synaptic weights and of communication between cells are addressed in the last part.