Towards hardware design of realistic neurobiological systems

Some general issues relating to the design of artificial neural systems based upon known neurobiological structures and functions are discussed, and then an example of a reconfigurable hybrid analog-digital neural processor array designed along these lines is presented. Our approach is to make use of a universal neurocomputational element which can be used to build up neural network topologies inspired particularly by those of the cerebral and cerebellar cortex. The processor element is based upon consideration of key neurobiological mechanisms which should give rise to biologically realistic global behaviours, rather than selecting circuits based upon their expectation to perform functions described by specific artificial neural network paradigms