Mark A. Glover, Alister Hamilton, Leslie S. Smith, Dept of Electrical Engineering, Univ of Edinburgh, Kings Buildings, Edinburgh EH9 3JL, and Dept of Computer Science, Univ of Stirling, Stirling FK9 4LA
An analogue VLSI implementation of an integrate and fire neural network chip is presented. It is to be used to cluster onset and offset signals in a sound segmentation system. The principle challenge in the implementation is the realisation of analogue circuits capable of operating with time constants in the millisecond range. Two solutions to this problem are detailed. The chip fabricated contains 8 neurons and is cascadable, allowing construction of larger networks. Initial results indicate the network successfully clusters incoming tonic signals.