Abstract
Recent research has indicated that Neural Networks may offer powerful
and effective solutions to certain classes of problems which von
Neumann machines do not handle effectively. These problems include
the areas of visual and speech recognition.
This thesis describes the basic principles behind neural networks and
neural network implementation.
The main aim of the thesis is to investigate the implementation of
neural networks in silicon integrated technology. To this end the design
of a CMOS VLSI test neural network is described and the problems
of VLSI implementation discussed. The neural network design utilizes
switched\capacitor techniques to implement the basic McCulloch-Pitts
neuron. A test chip has been fabricated using this circuit technique
and tested successfully. Comparisons between the switched capacitor
technique and other neural network implementations are also discussed.